Immature myeloid progenitors promote disease progression in a mouse model of Barrett's-like metaplasia.

Cdx2, an intestine specific transcription factor, is expressed in Barrett's esophagus (BE). We sought to determine if esophageal Cdx2 expression would accelerate the onset of metaplasia in the L2-IL-1β transgenic mouse model for Barrett's-like metaplasia. The K14-Cdx2::L2-IL-1β double transgenic mice had half as many metaplastic nodules as control L2-IL-1β mice. This effect was not due to a reduction in esophageal IL-1β mRNA levels nor diminished systemic inflammation. The diminished metaplasia was due to an increase in apoptosis in the K14-Cdx2::L2-IL-1β mice. Fluorescence activated cell sorting of immune cells infiltrating the metaplasia identified a population of CD11b+Gr-1+ cells that are significantly reduced in K14-Cdx2::L2-IL-1β mice. These cells have features of immature granulocytes and have immune-suppressing capacity. We demonstrate that the apoptosis in K14-Cdx2::L2-IL-1β mice is CD8+ T cell dependent, which CD11b+Gr-1+ cells are known to inhibit. Lastly, we show that key regulators of CD11b+Gr-1+ cell development, IL-17 and S100A9, are significantly diminished in the esophagus of K14-Cdx2::L2-IL-1β double transgenic mice. We conclude that metaplasia development in this mouse model for Barrett's-like metaplasia requires suppression of CD8+ cell dependent apoptosis, likely mediated by immune-suppressing CD11b+Gr-1+ immature myeloid cells.


INTRODUCTION
Esophageal adenocarcinoma (EAC) is an important cause of death in the US, and one of a few cancers whose incidence is on the rise [1]. EAC develops from a metaplastic, premalignant columnar epithelium known as Barrett's esophagus (BE) rather than the squamous epithelium typically present in the esophagus [2][3][4]. BE is frequently observed in the setting of chronic gastric acid and bile reflux that results in injury and inflammation. The prevalence of Barrett's is estimated at 10-15% of patients with gastroesophageal reflux disease (GERD) [5]. Given its prevalence, its association with EAC, and the increasing incidence of EAC, understanding BE pathogenesis is an important clinical and research imperative.
An important focus of our research is the development of novel mouse transgenic models for BE and EAC. Previously we ectopically expressed Cdx2, an intestine-specific transcription factor commonly observed in BE, in the mouse esophagus and forestomach using the murine Keratin 14 gene promoter [6]. Cdx2 expression was associated with altered cell morphology and ultrastructure of the esophageal epithelium. In particular we observed dilated intercellular spaces between the squamous basal cells and a compromised epithelial barrier ( Figure 1A). However, the formation of a true intestinal metaplasia did not occur.
More recently, a physiologically relevant transgenic mouse model for BE and EAC was described by our group [7]. It utilized an Epstein-Barr virus L2 promoter to over-express human IL-1β in the oral cavity, esophagus, and squamous forestomach of mice . These L2-IL-1β mice develop a chronic [8] inflammatory esophagitis by 3 months ( Figure 1A) that is followed subsequently by the development of a columnar metaplasia with intestinal features that later progresses to dysplasia and cancer. The strength of this transgenic mouse model is that in many ways it strongly phenocopies the pathogenesis of the human Barrett's esophagus as it is presently believed to occur [4,9], with a chronic inflammatory esophagitis preceding the onset of metaplasia, followed subsequently by dysplasia and cancer. Moreover, this disease sequence is accelerated in the L2-IL-1β mice by the addition of bile acids, as is hypothesized for the human disease. In addition, based on histologic and molecular criteria, the columnar metaplasia which develops in the L2-IL-1β resembles that of human BE [7]. Lastly, the metaplasia, dysplasia and cancer arise at the squamo-columnar junction (SCJ) much as in the human disease. Together, these observations all suggest the L2-IL-1β mouse is an excellent animal model for human BE and EAC. However, there are limitations of this animal model. Anatomically, mice have a squamous forestomach, and therefore this metaplasia arises at the SCJ in the stomach. In addition, although the production of intestinal mucins is strongly observed and consistent with an intestinalized metaplsia, mature goblet cells are not typically seen unless the animals are treated with Notch signaling inhibitors. For this reason, the metaplasia that develops has been described as "Barrett's-like metaplasia" [7].
Given that Cdx2 is expressed in BE, is required for the intestinal phenotype [10], and that ectopic expression of Cdx2 in the esophagus induces a barrier dysfunction, we hypothesized that the K14-Cdx2 transgene would synergize with the L2-IL-1β transgene and promote a more rapid progression to metaplasia and cancer. Unexpectedly, the double transgenic mice had fewer metaplastic nodules at the SCJ compared to the L2-IL-1β control mice. This was not due to diminished esophagitis or systemic inflammation. The reduction was due to an observed increase in apoptosis in the developing metaplasia at the SCJ of the double-transgenic mice that was not present in the single transgenic controls. Mechanistically, we provide evidence that this apoptosis is immune-mediated and increased due to significant reductions in the levels of an immune-suppressing subpopulation of immature CD11 + Gr-1 + myeloid cells. These CD11 + Gr-1 + cells have been implicated in promoting tumorigenesis in a number of mouse models of cancer [11][12][13]. We conclude this population of immature myeloid cells with immune suppressor function are critical for disease progression in the L2-IL-1β transgenic mouse model for BE and EAC.

Ectopic Cdx2 expression in murine esophageal epithelium does not alter the inflammatory esophagitis induced by transgenic IL-1β expression
To investigate for synergy between ectopic esophageal expression of the intestine-specific transcription factor Cdx2 (K14-Cdx2) and an animal model for Barrett's-like metaplasia, L2-IL-1β transgenic mice, we crossed them to yield doubly transgenic K14-Cdx2::L2-IL-1β mice ( Figure 1A). To enhance the onset of metaplasia, we treated all mice in this study with 0.2% deoxycholic acid treatment (DCA, pH 7.0) in the drinking water beginning at 8 weeks of age and continued this treatment for 12 months total (to age 14 months, Figure 1B), as was done in the initial report describing the IL-1β transgenic model for a Barrett'slike metaplasia [7]. Histologic analysis of the esophagus confirmed Cdx2 protein expression in the basal epithelial cell population of only the K14-Cdx2 and K14-Cdx2::L2-IL-1β transgenic lines but not in wild-type littermates or L2-IL-1β transgenic mice ( Figure 1C). This indicates that prolonged DCA exposure does not itself induce Cdx2 expression in the murine esophagus. In both of the K14-Cdx2 containing lines, Cdx2 mRNA levels induced over WT control were no different (400-fold ± 100 in K14-Cdx2 mice vs 408-fold ± 160 in K14-Cdx2::L2-IL-1β mice; n = 6) and protein expression levels are similarly equivalent by immunohistochemistry and not affected by the L2-IL-1β transgene ( Figure 1C and 1D).
The L2-IL-1β transgene induces a brisk inflammatory infiltrate in the esophagus, oral cavity, and tongue of both L2-IL-1β possessing transgenic lines ( Figure 2A and data not shown). At the SCJ where the metaplasia occurs, the level of inflammation is not significantly different in both IL-1β expressing transgenic lines, based on histopathology scoring by observers blinded to the genotype ( Figure 2B). Levels of IL-1β mRNA in the esophagus were not altered by the K14-Cdx2 transgene (122-fold ± 48 induction over WT controls in IL-1β mice against 116-fold ± 54 induction in K14-Cdx2::L2-IL-1β mice ; n = 6) ( Figure 2C). Lastly, systemic inflammation induced by IL-1β expression is similarly unaffected by the K14-Cdx2 transgene. Serum levels for IL-1β were below our ability to measure reliably in both K14-Cdx2::L2-IL-1β and L2-IL-1β mice (data not shown). However, levels of the pro-inflammatory cytokine IL-6, which is required for the L2-IL-1β metaplasia phenotype [7], were not measurably different in L2-IL-1β (33 ± 11pg/μl; n = 5) and K14-Cdx2::L2-IL-1β mice (36 ± 7pg/μl; n = 6) ( Figure 2D). Together, these findings strongly suggest the local esophageal as well as systemic inflammatory activities are equivalent in L2-IL-1β and K14-Cdx2::L2-IL-1β mice, and that the K14-Cdx2 transgene does not appear to affect the inflammatory response. www.impactjournals.com/oncotarget A. Model of the crossings to generate K14-Cdx2::L2-IL-1β transgenic mice. TEM image is of dilated intracellular spaces in the esophageal epithelium of K14-Cdx2 transgenic mice. H&E image is of inflammation in the esophagus of L2-IL-1β transgenic mice. B. Experimental approach; 8-week old mice were started on 0.2% DCA in their drinking water and maintained on this for 12 months, at which time the mice were examined for disease extent. C. Representative Immunostaining for Cdx2 expression in esophagi of transgenic mice. (X100 magnification; black bar = 50 μm). WT: wild-type; X2: K14-Cdx2; IL-1β: L2-IL-1β; and X2/IL-1β: K14-Cdx2::L2-IL-1β. D. Cdx2 mRNA expression by Real-Time PCR analysis for Cdx2 mRNA levels for each group of mice; (a = significantly differs from WT and IL-1β controls by one-sided ANOVA and Tukey Multiple Comparisons testing, adjusted p < 0.047; n = 6). www.impactjournals.com/oncotarget Score of inflammation at SCJ by observers blinded to genotype at 14 months of age. L2-IL-1β and K14-Cdx2::L2-IL-1β mice develop increased inflammation that is significant different from WT mice but not different from each other. ‡ Not significantly different, (adjusted p = 0.40); † not significantly different from WT control (adjusted p > 0.99); * significantly differs from WT mice (adjusted p ≤ 0.0002) by Kruskal-Wallis ANOVA and Dunn's multiple comparisons testing. C. Determination of esophageal IL-1β mRNA levels by qPCR analysis. Significance determined by one-sided ANOVA and Tukey's Multiple Comparison testing, n = 6 for each line. ( ‡ both significantly differ from WT control adjusted p < 0.01 but no significant difference between L2-IL-1β and K14-CDX2/L2-IL-1β mice). D. ELISA for serum levels of IL-6, n = 4. ( † not significantly different from WT mice by one-sided ANOVA and Tukey Multiple comparisons testing, adjusted p = 0.99; ‡ both significantly differ from WT control adjusted p < 0.03 but no significant difference between each other). www.impactjournals.com/oncotarget Ectopic Cdx2 expression in murine esophageal and forestomach epithelium reduces metaplasia development at the squamo-columnar junction in the IL-1β transgenic mice At the end of the DCA treatment period, the mice were sacrificed, and the squamo-columnar junctions (SCJ) were examined under a dissecting microscope with a dilute methylene blue stain. A prominent, nodular metaplasia was found at the SCJ in nearly 75% of the L2-IL-1β mice (n = 19), in keeping with previously reported observations ( [7] and Figure 3A and 3B). In the double transgenic K14-Cdx2::L2-IL-1β mice, the nodular metaplasia was similarly observed in nearly 70% of mice (n = 17). Unexpectedly, nearly 20% of the single transgenic K14-Cdx2 mice developed small single nodules at the SCJ (n = 11), much smaller than the disease noted in the L2-IL-1β and K14-Cdx2::L2-IL-1β mice. These small nodules were seen only in the K14-Cdx2 mice receiving DCA in their drinking water ( Figure 3B and data not shown).
Despite their similar disease frequencies, K14-Cdx2::L2-IL-1β mice appeared to have a reduced burden of nodular metaplasia compared to the L2-IL-1β mice ( Figure 3A). To quantify this, we counted the numbers of nodules per mouse, as well as determined the volume of this metaplasia using standard tumor volume approaches. Most significantly, the K14-Cdx2::L2-IL-1β mice (2.2 nodules/mouse ± 2.0; n = 17) had half as many metaplastic nodules as the L2-IL-1β littermates (4.6 nodules/mouse ± 2.4; n = 19) ( Figure 3C). There were no significant differences in the average size of the nodules in the L2-IL-1β and K14-Cdx2::L2-IL-1β mice (data not shown). However the nodules in both were significantly larger than those observed in the K14-Cdx2 mice (data not shown). Overall, the nodule burden in the L2-IL-1β mice was greatest, primarily due to the increased numbers of nodules in these mice. In summary, Cdx2 co-expression with IL-1β reduced the number of nodules of Barrett'slike metaplasia observed in the L2-IL-1β mice treated with DCA for 12 months.

Apoptosis is significantly increased in the Barrett'slike metaplasia of K14-Cdx2::L2-IL-1β mice
To understand better what the effects of Cdx2 coexpression are on the L2-IL-1β mouse phenotype, we next examined the metaplasia that develops in the K14-Cdx2::L2-IL-1β mice. The metaplasia which arises in L2-IL-1β mice has been previously-described as Barrett's-like based on several criteria, the most importantly being 1) the induction of intestinal mucin-producing cells (but not classic goblet cells), 2) the disease arising in the setting of chronic inflammation at the SCJ (as occurs in the human disease), and 3) a gene expression pattern which significantly overlaps with the human BE disease [7].
As in this published description, there is an expansion of a glandular, columnar epithelium at the SCJ and displacement of the oxyntic gastric glands distally in both the L2-IL-1β and K14-Cdx2::L2-IL-1β mice ( Figure 4). Histopathologically, this metaplasia was present in both the L2-IL-1β and K14-Cdx2::L2-IL-1β mice but not the WT and K14-Cdx2 littermates. In both L2-IL-1β and K14-Cdx2::L2-IL-1β mice, the metaplastic cells express intestinal mucins, as evidenced by positive staining with Alcian blue and Muc2 ( Figures 5A and 5B). Moreover, consistent with the published report, we can demonstrate increased mRNA expression of the Barrett's esophagus associated genes Cckbr, Tff2, and Krt19 (data not shown). Together these findings confirm the previously published description of the L2-IL-1β mouse and establish that K14-Cdx2 co-expression does not significantly alter the phenotype of metaplasia which develops in K14-Cdx2::L2-IL-1β mice.
We next considered other mechanisms by which Cdx2 co-expression in the squamous epithelium reduced the development of the nodular metaplasia at the SCJ. Cdx2 is a transcription factor with tumor-suppressor activity in the intestine, possibly mediated by inhibitory effects on cell proliferation, but it is also reported to have tumorigenic activity by repressing cell apoptosis [6,14,15]. In the mouse esophagus, IL-1β expression increased cell proliferation equally, as demonstrated by incorporation of the DNA analogue EdU in the Cdx2 non-expressing (0.29 ± 0.06%EdU+nuclei; n = 15 mice) and expressing (0.28 ± 0.07%EdU+nuclei; n = 10 mice) littermates ( Figure 6A). At the SCJ, there was no significant effect of either transgene or DCA treatment on SCJ metaplasia EdU incorporation (data not shown), suggesting the observed reduction in metaplastic development in K14-Cdx2::L2-IL-1β mice was not due to changes in cell proliferation either in the esophagus or at the SCJ.
As an increase in apoptosis in the K14-Cdx2::L2-IL-1β mice could be an explanation, we examined apoptosis rates by Caspase 3 immunohistochemistry (IHC) and TUNEL staining. In the esophagus, there was no evidence that the L2-IL-1β or K14-Cdx2 transgenes alone altered cellular apoptosis by either technique ( Figure 6B and 6C and data not shown). Similarly, there was no significant apoptosis noted in the SCJ metaplasia in L2-IL-1β transgenic mice ( Figure 6D and 6F and F-inset). However, in the K14-Cdx2::L2-IL-1β mice, there was a noticeable increase in apoptosis demonstrated by both approaches (Figure 6E and 6G and G-inset). The TUNEL staining in the K14-Cdx2::L2-IL-1β mice is localized to nuclei. This is most evident in the inset, and includes many nuclei from the metaplastic glandular epithelium ( Figure 6E, 6G, and 6G-inset). In summary, the K14-Cdx2 transgene limits the formation of the BE like metaplasia in L2-IL-1β transgenic mice by increasing apoptosis in the developing SCJ metaplasia.

Myeloid cell associated genes are diminished in the metaplastic nodules of K14-Cdx2::L2-IL-1β mice
In order to determine why cell apoptosis is increased in the developing metaplasia of the double transgenic mice, we performed an Affymetrix microarray analysis of gene expression differences. We compared gene expression in the metaplastic nodules from L2-IL-1β and K14-Cdx2::L2-IL-1β mice. We identified 199 genes whose expression differed by 2-fold or more and had less than a 10% false discovery rate (Table 1 and Supplementary  Table S1). Only 47 of the 199 differentially expressed genes had increased levels in the K14-Cdx2::L2-IL-1β mice. The two most strongly induced were Syncollin (Sync) and Cadherin related family member 5 (Cadhr5), both of which are expressed in the intestine and one, Cadhr5, is a known transcriptional target of Cdx2 [16]. A gene function annotation of these 47 genes using DAVID bioinformatic resources [17] found only a few weak associations (p values between 0.02 and 0.05) with early developmental processes ( Figure 7A).
A gene ontogeny analysis of genes whose expression was diminished in the double transgenic mice revealed a number of defense and immune response pathways significantly associated (p values ≥ 0.0005) with the 152gene list ( Figure 7B). In particular, there were several genes reduced whose products are associated with immature myeloid cells, including IL-17a, IL-17c, IL-23a, S100A8, S100A9, and Csf3r (Colony stimulating factor 3 receptorgranulocytes), as well as reductions in Granzyme B and a number of serine proteases suggestive of cytotoxic T cells (Table 1). We had previously established that the inflammation in the esophagus and systemic IL-6 levels were no different between the two transgenic lines. To explore whether the inflammatory infiltrate at the SCJ was diminished in K14-Cdx2::L2-IL-1β mice, we stained for CD45+   and cell nuclei were stained with DAPI (blue). At least 300 cells per mouse were counted. L2-IL-1β over-expression promotes cell proliferation in esophageal epithelium. * Significantly differs from control WT mice by one-sided ANOVA and Tukey's Multiple comparison test adjusted p < 0.043. ‡ no significant difference between double and single L2-IL-1β TG mice adjusted p = 0.9996; † no significant difference between K14-Cdx2 and WT mice, adjusted p = 0.9932. n = 15 mice WT, n = 8 mice K14-Cdx2, n = 15 mice L2-IL-1β, and n = 10 K14-Cdx2::L2-IL-1β mice. B. and C. Representative imaging of combined nuclear (DAPI-blue), TUNEL (Tetramethylrhodamine -red) and differential interference contrast microscopy (Nomarski) of the SCJ of (B) WT and (C) K14-Cdx2 mice. (x200, red bar = 50 μm). D. and E. Representative imaging TUNEL (Tetramethylrhodamine -red) staining of the SCJ metaplasia of (D) L2-IL-1β and (E) K14-Cdx2::L2-IL-1β mice. There is a highly significant increase in TUNEL labeling of the nuclei of the glandular epithelial cells. (x200, red bar = 50 μm). F. and G. Representative imaging of combined nuclear (DAPI-blue), TUNEL (Tetramethylrhodamine -red) and differential interference contrast microscopy (Nomarski) of the SCJ metaplasia of (F) IL-1β and (G) K14-Cdx2::L2-IL-1β mice. (×200, red bar = 50 μm). Inset: ×400 magnification of SCJ metaplasia with only nuclear (DAPI-blue) and TUNEL (Tetramethylrhodamine -red) staining illustrate the localization of TUNEL staining to nuclei, including the glandular epithelium of the metaplasia.  infiltrating inflammatory cells. We found that both L2-IL-1β and K14-Cdx2::L2-IL-1β mice maintained a brisk CD45+ immune cell infiltration at the SCJ junction as compared to normal WT littermates ( Figure 7C, 7D, and 7E). Therefore, while the microarray results suggest there may be significant alterations in a subset of myeloid cells in the developing metaplasia, the broader increased inflammatory response is intact in the K14-Cdx2::L2-IL-1β mice. There were greater changes in other subpopulations. Cell surface markers for granulocytes (Gr-1) and monocytes (CD11b) appeared to be significantly diminished in the K14-Cdx2::L2-IL-1β mice, by 4 to 8-fold in both the esophagus (p = 0.049 and p = 0.053, respectively n = 5 mice each genotype) and the SCJ metaplasia (p = 0.003 and p = 0.002, respectively, n = 5 mice each genotype) ( Figure 8C), suggesting this effect was systemic and not confined to the SCJ. Both of these markers can also be co-expressed on immature myeloid cell populations [12]. Surprisingly, the majority of CD11b+ and Gr-1+ cells were in fact double positive, consistent with these cells representing an immature myeloid population ( Figure 9). This population was described previously in the L2-IL-1β mice [7], but their importance for the development of the SCJ metaplasia was not evident. We do not know when these CD11b + Gr-1 + cells are first present but we have detected them at 9 months (data not shown) and Quante et. al. reported them in 6 month old mice [7], before the onset of the SCJ metaplasia.

Immature myeloid cells are diminished in the
CD11b + Gr-1 + immature myeloid cells originate in the bone marrow in response to factors secreted by tissues and tumors. They have been implicated in enhancing tumor growth in the breast, colon, and pancreatic cancers [11,12], but have not been described in precancerous conditions previously. These immature myeloid cells can enhance tumorigenesis by several mechanisms, including inhibition of tumor cell immune surveillance (mediated by cytotoxic T-cells), enhancement of angiogenesis, and the overall promotion of tumor cell survival [11,12]. Immature myeloid cells can belong to either the monocytic or granulocytic lineages [18]. To determine which lineage these cells belonged we examined for the expression of Ly-6C and Ly-6G in CD11b+ cells. Monocytic lineages are Ly-6G -Ly-6C Hi whereas granulocytic lineages are Ly-6G + and Ly-6C Lo/+ . We found the CD45 + CD11b + cells from esophagus and SCJ metaplasia of the L2-IL-1β mice were Ly-6G + and Ly-6C + (Figure 10A, 10B), which is consistent with a granulocytic lineage as previously described [12]. Moreover, these cells are largely lost in the double transgenic K14-Cdx2/L2-IL-1β mice (n = 3 mice for each genotype) ( Figure 10A, 10B). As an additional confirmation of their lineage, we performed intracellular staining for myeloperoxidase (MPO), a lysosomal enzyme abundantly expressed in neutrophil granulocytes. CD45 + CD11b + Gr-1 + cells from the SCJ metaplasia strongly express MPO, in contrast to the CD45 + CD11b + Gr-1 − and CD45 + CD11b − Gr-1 − populations from the same tissue ( Figure 10C). Lastly, immature granulocytes are also known as band cells due to their distinct densely staining and unsegmented nucleus [19]. We isolated CD45 + CD11b + Gr-1 + cells by FACS, pelleted, fixed, sectioned and stained them for histologic analysis. Morphologically these cells possess the classic prominent, unsegmented nuclei of band cells ( Figure 10D), further establishing their identity as an immature granulocytic cell population.
Two key properties of the tumor-promoting CD45 + CD11b + Gr-1 + cell population is that they emerge from the bone marrow, and therefore are a systemic response and not tissue-specific, and that they possess immunesuppressor properties (specifically T-cell suppression) [12,18]. To determine if these cells exist beyond the esophagus and SCJ metaplasia, we isolated CD45 + CD11b + Gr-1 + cells from the esophagus and the spleen and again assayed for intracellular MPO levels. We observed abundant MPO protein in the CD45 + CD11b + Gr-1 + cell population from both organs, and little MPO in the other control cell populations ( Figure 11A and 11B), suggesting they are identical cell types. Moreover, the presence of these cells in the spleen indicates this is a systemic response, not one localized to the esophagus and SCJ metaplasia.
To assay for immune suppressor function in our CD11b + Gr-1 + cells, we performed an in-vitro T cell suppression assay. The assay measures the ability of CD45 + CD11b + Gr-1 + cells to suppress CD4 + T-cell proliferation in response to CD3/ CD28 costimulation [20]. As this assay requires significant numbers of CD45 + CD11b + Gr-1 + cells, they were isolated from the spleens of L2-IL-1β mice rather than the esophagus or SCJ metaplasia. As can be observed, increasing the ratio of the CD45 + CD11b + Gr-1 + cells to CD4 + T-cells led to a significant decrease in the proliferative response of the T-cells     to costimulation by CD3/CD28 beads ( Figure 11C). Together, these observations suggest that the CD45 + CD11b + Gr-1 + cells, which are significantly reduced in the K14-Cdx2/L2-IL-1β mice, are immature granulocytes with an immune-suppressor phenotype. Given the association of CD11b + Gr-1 + cells with increased tumor growth in a number of mouse models of cancer, it raises the possibility that these cells may also contribute to metaplasia formation and expansion in the L2-IL-1β mouse model of Barrett's-like metaplasia.

CD8 + cells are required for the increased apoptosis in the SCJ metaplasia of the K14-Cdx2/L2-IL-1β mice
Studies using both mouse cancer models and human subject samples have indicated that increased levels of immature myeloid cells can suppress normal CD8 + cytotoxic T-cell responses and thereby enhance tumor growth [11,12,20]. To determine if CD8 + T-cells are involved in the induction of apoptosis observed in the metaplasia of K14-Cdx2::L2-IL-1b mice, we targeted CD8 + cells using an antibody. 14 month-old K14-Cdx2::L2-IL-1b mice were injected with 200 μg of anti-mouse CD8 antibody or an isotype control antibody on days 1, 3, and 5 and sacrificed on day 6. FACS analysis of immune cell populations in the spleen, blood and esophagus demonstrated a nearly complete absence of CD8 + cells in the mice receiving the anti-CD8 antibody and not those mice receiving the isotype control ( Figure 12). Moreover, CD4 + T cell levels were not affected by either treatment, attesting to the specificity of the ablation.
We then assessed the SCJ metaplasia for histologic changes with the CD8 + cell knockdown. Since the CD8 + cell knockdown was a short duration, we did not observe significant changes in metaplasia abundance or morphology (data not shown). Significantly, however, there was a near complete loss of the apoptosis previously observed at the SCJ in the K14-Cdx2::L2-IL-1b mice ( Figure 13). Under closer examination, it can be observed that the apoptosis is reduced in the squamous forestomach as well as the adjacent SCJ metaplasia (Figure 13 inset). We conclude that the elevated apoptosis in the squamous forestomach and SCJ metaplasia of the K14-Cdx2::L2-IL-1b mice is due to an immune-mediated and CD8+ T-cell dependent mechanism in the L2-IL-1β transgenic mouse model.

DISCUSSION
Our understanding of the mechanisms driving the development of Barrett's esophagus and its progression to esophageal adenocarcinoma has been limited by the paucity of animal models of these conditions. The development of the L2-IL-1β transgenic mouse model for intestinal metaplasia [7] has been a significant advance, and the studies we report here demonstrate the importance of animal models in advancing our understanding of disease mechanisms. Currently, the pathogenesis of Barrett's esophagus is thought to be a response to the chronic inflammation and injury from repeated gastric acid and bile acid reflux into the esophagus [4,28]. The strength of the L2-IL-1β transgenic mouse as a model of BE and progression to EAC is that it is 1) dependent upon chronic esophageal inflammation, 2) that the metaplasia, which mimics BE at the level of morphology and gene expression, occurs at the SCJ as it does in BE, and 3) that bile acids can both accelerate disease progression and enhance intestinalization. Given this strength, our studies which elucidate a novel mechanism driving the onset of the metaplasia in these mice takes on even greater significance, providing new avenues of exploration for the human disease.
It is interesting to note that although conceptual models for the development of BE have for many years emphasized the role of reflux injury and inflammation in the development of the metaplasia, surprisingly few studies have deeply explored the inflammation and inflammatory cells associated with disease onset and progression. Proinflammatory cytokines including IFN-γ, IL-1β, IL-6, and IL-8, are expressed by epithelial cells in response to acid and bile reflux and drive the influx of a variety of proinflammatory cells including neutrophils, eosinophils, mast cells, macrophages as well as the adaptive immunity T and B cells [29,30]. In the esophagus, activation of a T H 1 pro-inflammatory response, characterized by production of interferon (IFN)-γ, is typical for acid reflux esophagitis [31]. Progression to BE is accompanied by a shift in cytokine expression patterns, including increased levels of interleukin IL-4, IL-5, IL-10, and IL-13, which are hallmarks of a T H 2 humoral immune response [31][32][33]. Associated with this shift in cytokine patterns is a change in the infiltrating immune cells, with a reduction in macrophages in favor of T H 2 associated plasma cells [34]. More recently, it was demonstrated that myeloid and plasmacytoid dendritic cells are recruited during the esophageal metaplasia-dysplasia-carcinoma sequence [35]. In the same report, it was demonstrated that cordblood derived myeloid dendritic cells, when co-cultured with Barrett's esophagus and esophageal adenocarcinoma cell lines, displayed an immune tolerogenic behavior. This has direct bearing on our novel and important findings.
In humans, Cdx2 is expressed in BE and can be detected very early in the disease process, in the setting of reflux esophagitis [36,37]. We anticipated that crossing IL-1β mice after 12 months of DCA treatment were subject to three injections with an anti-CD8 antibody or an isotype antibody control. Dot-plots from FACS analysis for CD45 + CD8 + cells in the spleen, blood, and esophagus of A. isotype control and B. anti-CD8 antibody treated K14-Cdx2::L2-IL-1β mice. There was significant reduction in CD45 + CD 8 + cells in all three tissues without an appreciable effect on CD4 + T-cells in those same tissues. n = 3 K14-Cdx2::L2-IL-1β mice for each condition, 6 mice total. www.impactjournals.com/oncotarget  our previously described K14-Cdx2 mice with the L2-IL-1β transgenic mice might yield a more advanced, aggressive intestinal metaplasia. However, the opposite result was obtained. In our exploration of the mechanism for this protection, we eliminated trivial possibilities, such as Cdx2 expression reducing esophageal IL-1β levels and systemic inflammation, or that Cdx2 expression reduced cell proliferation, thereby limiting metaplasia formation. In contrast, we clearly demonstrate a very novel mechanism, that Cdx2 expression is associated with an increase in CD8 + cell-dependent apoptosis in the developing metaplasia. We demonstrate a very significant reduction in a subpopulation of immature myeloid cells with immune suppressor properties, known by the cell surface marking CD45 + CD11b + Gr-1 + . Based on the literature regarding these immature myeloid cells [12,18], their loss or disruption in vivo typically leads to enhanced immunemediated disruption of tumor growth, often via the actions of cytotoxic T-cells [38], consistent with our findings. We therefore speculate that with the loss of the immunesuppressing myeloid cells in the K14-Cdx2::L2-IL-1β mice, there is increased immune mediated surveillance and targeting of the abnormal metaplastic cells, leading to the increased apoptosis we observe in the metaplastic nodules at the SCJ. Evidence for the CD11b+Gr-1+ cells fulfilling this component of the model (Figure 7) is at present correlative but is fully consistent with the published literature on immature immune-suppressing myeloid cells. Formal testing of this mechanism is will be pursued in future experiments, including crossing the L2-IL-1β mice with IL-17 and S100A9 knock-out and S100A9 over-expressing mice.
An additional important question we do not fully answer here is the mechanism by which esophageal Cdx2 expression inhibits the development and recruitment of these immature myeloid cells. We clearly demonstrate that Cdx2 expression reduces levels of proinflammatory S100A8/A9 proteins and also IL-17 in the esophagus, but transgenic IL-1β is entirely unaffected. This is an important distinction, as S100A8/A9 and IL-17 are known from multiple studies to be critical for the induction of immature myeloid cells in cancer [12,18]. Moreover, the S100A8/a9 and IL-17c proteins are expressed by squamous keratinocytes, esophageal epithelium, and Barrett's esophagus epithelium [23][24][25][26][27], and therefore may be directly affected by esophageal Cdx2. Together these findings suggest a Cdx2 expression in the esophagus reduces the production of the proinflammatory S100A8/a9 proteins and IL-17c ( Figure 14).
However, elements of this mechanism remain unknown. Most critically is how the Cdx2 transgene yields this response, and this is a much harder question to address. We have explored Cdx2's role in intestinal biology [39][40][41][42] and observed that generally the transcription factor Cdx2 behaves as a transcriptional activator, not a repressor [10,43]. We performed an in silico analysis of all three genes to determine if Cdx2 is known to bind their promoter regions using the ENCODE and TRANSFAC databases [44,45] but found no evidence for an interaction with these genes by Cdx2. Moreover, we explored published genome-wide analyses of Cdx2 binding sites in intestinal epithelial cells and found no reports that Cdx2 protein bound and regulated IL-17, S100A8/A9 genes [46,47].
It is interesting to note that there are several published reports exploring interactions between proinflammatory cytokines, NF-κB, and Cdx2. In gastric and cholangiocarcinoma cell lines, proinflammatory cytokines including IL-1β, IL-6, and TNF-α have all been reported to induce Cdx2 gene expression [48][49][50]. Several studies have even suggested this may be mediated by NF-κB binding to the Cdx2 promoter [51,52]. In contrast, other studies have suggested Cdx2 may inhibit the NF-κB signaling pathway, possibly by binding the p65 subunit and inhibiting DNA binding of the NF-κB complex [53,54]. We speculate that this latter mechanism may be how Cdx2 limits IL-17 and S100A8/9 expression. Until we clarify this mechanism, we cannot know whether Cdx2 plays a "protective" role, limiting disease progression to dysplasia and EAC by limiting the production of the immature myeloid cells. Therefore, this will remain an important focus for our future research efforts.
In many human cancers, immature myeloid cells have been under intense investigation due to their ability to promote tumor immune evasion and enhance tumorigenesis [12]. They are under investigation not only for the insights they yield into the pathogenesis of cancer, but for the potential therapeutic applications, including enhancement of anticancer immunotherapies [55]. Therefore, our work here, establishing their importance for disease onset in a mouse model for BE and EAC, is extremely significant not only for these new insights into BE pathogenesis our work provides, but also for potentially novel avenues of research and therapeutics which should now be now explored for patients with advanced BE and EAC.

Animal studies
All studies with the mouse models were fully approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Pennsylvania (IACUC#525400), and the animal care and use program conforms to all required standards. Mice were maintained in a specific pathogen-free facility with standard bedding and 12-hour light-dark cycles. The generation and genotyping of K14-Cdx2 [6] and L2-IL-1β [7] transgenic mice have been previously described. K14-Cdx2 transgenic mice were crossed with L2-IL-1β and yielded four genotypes analyzed for this study: K14-Cdx2::L2-IL-1β double transgenic, L2-IL-1β and K14-Cdx2 single transgenic mice and corresponding wild type control as well. Mice were placed on drinking water containing deoxycholic acid (0.2% DCA, pH7.0) at age of 8 weeks. After 12-months of treatment, mice were sacrificed for analysis. The nodules were measured and excised for routine pathological examination. The following nodule indicators were evaluated in each mouse by iVision software: the number of mice with nodules as well as the numbers of and the calculated volumes of the nodules per mouse.

Morphometric analysis
The squamocolumnar junction in HE-stained sections was evaluated for epithelial nodular hyperplasia, percentage of fields with mononuclear inflammatory infiltrates, neutrophilic inflammatory infiltrates, mucosal metaplasia, and mucosal lymphoid follicles. The extent of the histologic changes was assessed by determining the percentage of microscopic fields with positive criteria for lesions. All microscopic fields of each SCJ section were evaluated.

Quantitative real-time PCR analysis
Samples were stored in tissue storage reagent (RNAlater; Ambion, Austin, TX). Total RNA was isolated using RNeasy Mini kit (Qiagen, Valencia, CA). cDNA was prepared from total RNA using the SuperScript ® VILO™ cDNA Synthesis Kit (Invitrogen, Carlsbad, CA). Primers were designed using Primer Express software (Applied Biosystems). Quantitative RT-PCR was performed on an ABI 7000 sequence detection system (Applied Biosystems, Foster City, CA), with SYBR green or Taqman as the fluorescent dye using standard PCR conditions. A dissociation curve was run with each PCR as a control. A ribosomal phosphoprotein, 36B4, was used as the normalization control.
All the sequences of other primers used for real time PCR were described previously [6]. p values were determined by analysis of variance and Tukey rank mean test. ΔCt values were calculated after duplicate PCRs for each sample as described, and statistical analysis was performed. ΔΔCt values were then calculated and used to determine fold-change in expression.
Alcian blue staining, slides were deparaffinized. After application of 3% aqueous acetic acid to the slides, 1% Alcian blue in 3% acetic acid, pH 2.5, was applied. Sections were washed and counterstained with 0.1% nuclear fast red, dehydrated, and mounted. For immunofluorescence detection, the tissue sections were incubated with primary antibody overnight at 4°C and secondary antibodies at 37°C for 30 minutes. After incubation, slides were washed with PBS three times, counterstained with DAPI, and then photographed with a Nikon E600 fluorescent microscope and confocal microscope.

EdU cell proliferation assay
The transgenic and wild-type littermates were injected intraperitoneally with EdU (Life Technologies) 1 hour prior to sacrifice, esophagus and squamo-columnar junction area were harvested and embedded in paraffin. 5 μm thick sections were subjected to the Click-iT and subjected to the Click-iT EdU proliferation Assay (Life Technologies). EdU that had been incorporated into newly synthesized DNA was detected by Alexa Fluor 594 azide (red) and cell nuclei were stained with DAPI (blue) counterstain. All Images were captured at 10× and 20× magnification. Three random 10× fields were taken from each group of litter matched transgenic and wild type mice. The EdU positive proliferating cells were quantified and normalized to the total cell number in each field. The graph is generated from the average ratio (EdU/DAPI) of three 10× fields in each group.

ELISA
The levels of mouse IL-6 in sera of the transgenic mice were determined using an ELISA kit (BD Company, San Diego, CA). Absorbance was measured at 450 nm by a Tecan plate reader, and the samples were analyzed by Magellan 7.1 SP software. www.impactjournals.com/oncotarget TUNEL assay Apoptosis in sections was performed using In situ cell death detection kit with TMR Red according to the manufacturer instructions (Roche, West Sussex, UK) (Roche #12 156 792 910) and stored at 4°C until analysis.

RNA microarray analysis
Microarray analyses were performed on triplicate RNA samples of SCJ metaplasia nodules from mice to identify differentially expressed genes comparing L2-IL-1β forestomach and K14-Cdx2::L2-IL-1β forestomach. The total nodules from mouse SCJ were isolated and snap-frozen and stored at −80°C for RNA preparation. cDNA was transcribed using 5 μg total RNA (Affymetrix) and run on Affymetrix Mouse 1.0ST Affymetrix Arrays. The statistical test significance analysis of Microarrays was applied using a two-class unpaired analysis and differentially expressed genes were identified using a fold change cutoff of ≥1.5 (up or down). Gene expression differences were considered statistically significant if the p-value was less than 0.01. A global test was done as to whether the expression profiles differed between the classes by permuting the labels of which arrays corresponded to which classes. The false discovery rate was estimated to be less than 0.13%. Cluster analysis was performed with Cluster and Treeview software. Microarray files were submitted to the GEO repository; file GSE60320.
For CD8 + T-cell ablation studies, K14-Cdx2::L2-IL1β mice were injected on days one, three and five with 200 μg of anti-mouse CD8a (clone 53-6.72) or isotype control (clone 2A) antibodies from BioXcell. On the sixth day, blood, spleens and esophagi were collected for flow cytometry confirmation of decreased CD8 + cells. SCJ metaplasia sections were fixed for histology and TUNEL assays.

In vitro T cells suppression assay
To evaluate the ability of CD11b + Gr-1+ myeloidderived suppressor cells (MDSC) isolated from the spleen from L2-IL-1β transgenic mice to suppress antigen specific T cell proliferation, we performed a T-cell proliferation assay. A single cell suspension of splenocytes isolated from wild-type mice was made by homogenizing spleen with a 1 ml syringe through a 70 μM filter into a 50 ml conical tube. Red blood cells were lysed using lysis solution and quenched with HBSS. Red blood cell depleted murine splenocytes were resuspended at 1 × 10 6 /ml and labeled with 5 μM green fluorochrome carboxyfluorescein succinimidyl ester (CFSE). Cells were activated with anti-CD3/CD28 coated beads (Gibco, Life technologies) and seeded in triplicate and cultured in RPMI-1640 supplemented with 10% fetal bovine serum. Splenocytes were cultured either alone or in the presence of CD11b + Gr-1 + cells isolated from L2-IL-1β mice at different ratios 0:1, 1:1, 1:2, 1:4, 1:8, 1:16. After 72 h, cells were collected and stained with 7AAD, anti-CD45, CD3, CD4, Gr-1 and CD11b mAbs cocktail. Proliferation was determined by CFSE dilution and flow cytometric analysis on a FACSLSRII cytometer (BD Biosciences) with initial gating on the CD3 + /CD4 + populations.

Statistical analysis
GraphPad Prism version 3.04 was used for all statistical analyses (GraphPad, San Diego, CA, USA). For power analysis GraphPad Statmate or G*power3 was used. All data are represented as mean and standard deviation unless otherwise stated with p-value cutoff of ≤ 0.05 unless otherwise indicated. Replicate numbers and specific statistical tests employed are indicated in the figure legends.
All authors had access to the study data and had reviewed and approved the final manuscript.