Expression of delta-like ligand 4 (Dll4) and markers of hypoxia in colon cancer

Background: Delta-like ligand 4 (Dll4) is a Notch ligand that is upregulated by hypoxia and vascular endothelial growth factor-A (VEGF-A) and is reported to have a role in tumor angiogenesis. Evidence from xenograft studies suggests that inhibiting Dll4–Notch signalling may overcome resistance to anti-VEGF therapy. The aim of this study was to characterise the expression of Dll4 in colon cancer and to assess whether it is associated with markers of hypoxia and prognosis. Method: In all, 177 colon cancers were represented in tissue microarrays. Immunohistochemistry was performed using validated antibodies against Dll4, VEGF, hypoxia-inducible factor (HIF)-1α, HIF-2α, prolyl hydroxylase (PHD)1, PHD2, PHD3 and carbonic anhydrase 9 (CA9). Results: The expression of Dll4 was observed preferentially in the endothelium of 71% (125 out of 175) of colon cancers, but not in the endothelium adjacent to normal mucosa (none out of 107, P<0.0001). The expression of VEGF was significantly associated with HIF-2α (P<0.0001) and Dll4 (P=0.010). Only HIF-2α had a significant multivariate prognostic effect (hazard ratio 1.61, 95% confidence interval 1.01–2.57). Delta-like ligand 4 was also expressed by neoplastic cells, particularly neoplastic goblet cells. Conclusion: Endothelial expression of Dll4 is not a prognostic factor, but is significantly associated with VEGF. Assessing endothelial Dll4 expression may be critical in predicting response to anti-VEGF therapies.

Colorectal cancer is the second most common cause of cancer deaths worldwide, with an estimated 1 023 152 new cases and 528 978 related deaths in 2002 (Parkin et al, 2005). The early growth of colorectal tumours requires angiogenesis (Goodlad et al, 2006;Korsisaari et al, 2007), the consequence of increased expression of pro-angiogenic factors (e.g., vascular endothelial growth factor-A (VEGF-A); Ferrara et al, 1991;Kim et al, 1993;Jubb et al, 2003;Korsisaari et al, 2007). The expression of VEGF in cancer is controlled by both oncogenic signalling (such as Wnt-signalling in colorectal cancer; Zhang et al, 2001) and hypoxia (Mizukami et al, 2004). Although there is redundancy among pro-angiogenic factors in advanced cancer (Relf et al, 1997;Hanrahan et al, 2003), many early cancers (Hanrahan et al, 2003;Jubb et al, 2003) and in vivo cancer models (Bergers et al, 1999;Hanrahan et al, 2003;Joyce et al, 2003;Goodlad et al, 2006;Korsisaari et al, 2007) are VEGF dependent. This observation has been exploited by the addition of an anti-VEGF monoclonal antibody (bevacizumab) to first-line chemotherapy in metastatic colorectal cancer, which prolonged the median overall survival (from 15.6 to 20.3 months, P ¼ 0.001; Hurwitz et al, 2004). Nevertheless, all patients in this trial eventually progressed after 2 years (Jubb et al, 2006b), and there are no valid predictors of the survival benefit afforded by bevacizumab in colorectal cancer (Ince et al, 2005;Jubb et al, 2006a;Grothey et al, 2008a, b).
Additional therapeutic agents that disrupt functional tumour angiogenesis have been developed to target tumours that are inherently resistant to anti-VEGF therapy or become resistant during the course of therapy (Noguera-Troise et al, 2006;Ridgway et al, 2006;Pan et al, 2007;Caunt et al, 2008). Delta-like Ligand 4 (Dll4) is a ligand for Notch 1, 3 and 4 proteins, which is expressed by endothelial cells (Thurston et al, 2007;Indraccolo et al, 2009) and may be induced by VEGF and hypoxia, through hypoxiainducible factor (HIF)-1a (Patel et al, 2005). In the retina, the interaction between Dll4 and Notch regulates endothelial sprouting and branching to form functional vascular beds (Hellstrom et al, 2007). Disruption of Dll4 signalling by overexpression or inhibition of Dll4 may impair tumour angiogenesis (Noguera-Troise et al, 2006;Ridgway et al, 2006), and blockade of Dll4 -Notch signalling results in an increased density of non-functional vasculature and is associated with a reduction in the growth of human xenografts (Noguera-Troise et al, 2006;Ridgway et al, 2006). Indeed, certain xenografts that are resistant to anti-VEGF therapy are reported to be sensitive to anti-Dll4 (Noguera-Troise et al, 2006;Ridgway et al, 2006;Li et al, 2007), and combination treatment with anti-VEGF and anti-Dll4 has additive effects on tumour growth (Ridgway et al, 2006). In addition, human umbilical vein endothelial cells transfected with Dll4 downregulate the VEGF receptor 2 and co-receptor neuropilin-1, and show reduced proliferative and migratory responses to VEGF (Williams et al, 2006). Together, these data suggest that Dll4 may have a role in mediating resistance to anti-VEGF therapies.
The characterisation of Dll4 protein expression in human cancer is important for the rational design of clinical trials to test the safety and activity of anti-Dll4 therapy. Moreover, defining the context of Dll4 expression, in terms of known markers of hypoxia and angiogenesis, may identify subgroups of tumours with distinct clinical behaviour and response to treatment. The aims of this study were to characterise the in situ expression of Dll4 in colon cancer, to assess the association between Dll4 and established markers of hypoxia and angiogenesis and to determine the prognostic significance of these markers.

Cell lines
Recombinant human Dll4 was cloned into a pcDNA3.1 vector (Invitrogen, Carlsbad, CA, USA). Vectors were transfected into human U87 cells with the Fugene 6 transfection reagent (Roche Applied Science, Indianapolis, IN, USA). Human umbilical vein endothelial cells were cultured in vitro in media with and without VEGF as previously described . Cell pellets were then formalin fixed and paraffin embedded. Further details are available from the authors on request.

Patients and tissue samples
Formalin-fixed and paraffin-embedded tissue blocks and corresponding pathology reports were retrospectively obtained for 177 sequential patients with colon adenocarcinomas (surgery was performed from 1997 to 2000 at the John Radcliffe Hospital, Oxford, UK). Sample size was determined by the availability of tissue with full clinicopathological data, survival follow-up and ethical approval for research. No patients received preoperative chemotherapy. Rectal cancers were excluded from this series to avoid the confounding effects that preoperative radiotherapy may have on the expression of hypoxic markers. The mean age at diagnosis was 71 years (range 17 -92 years), 97 patients (55%) were male and 73 cancers (41%) were at or distal to the splenic flexure. In all, 13 cancers were in Dukes' stage A (7%), 88 in stage B (50%), 55 in stage C (31%) and 21 in stage D (12%) at the time of surgery. The mean follow-up time was 5.6 years (range 1 month to 11.9 years). No patients received postoperative radiotherapy. No Dukes' stage A patients, 38% (24 out of 64) of Dukes' stage B patients, 58% (32 out of 55) of Dukes' stage C patients and 52% (11 out of 21) of Dukes' stage D patients received 5-fluorouracil-based chemotherapy according to the local protocols (adjuvant regimen: six 5-day courses of 370 mg m À2 5-fluorouracil and 25 mg L-folinic acid as an intravenous bolus every 4 weeks). Follow-up data were correct as of January 2009, with a median follow-up time of 5 years and 7 months. Tissue microarrays were assembled as described previously (Bubendorf et al, 2001) with three replicate cores for each tumour and one core for each normal mucosal sample. An additional 12 whole sections of colorectal adenomas with adjacent adenocarcinoma were also collected (John Radcliffe Hospital) to further analyse the expression of Dll4 in the adenoma to adenocarcinoma sequence. Approval was obtained for the use of all human tissue from the local research ethics committee (C02.216).

In situ hybridisation
Isotopic in situ hybridisation for Dll4 was performed using previously described methods (Poulsom et al, 1998). The probe used was a 741-bp cloned fragment of human Dll4 (position 1775 to 2516 bp, relative to the adenosine of the start codon ATG; Patel et al, 2006). In situ hybridisation for b-actin was used as a positive control for mRNA integrity.
Immunohistochemistry for Dll4 was performed using an anti-Dll4 monoclonal antibody (the variable regions of this antibody are fully human and the Fc-domain is mouse; clone 242) that binds to the extracellular domain of human Dll4 (the epitope is in EGF-like domains 3-5 and generated in VelocImmune mice (Regeneron Pharmaceuticals, Inc., Tarrytown, NY, USA)). In brief, antigen retrieval was performed in target retrieval solution (Dako, Carpinteria, CA, USA) using a Decloaking Chamber (Biocare Medical, Concord, CA, USA). Sections were incubated for 16 h at 4 o C with the primary antibody at 1 mg ml À1 . Bound antibody was labeled with Novolink polymer (Leica Microsystems, Bannockburn, IL, USA), visualised using 2,3-diaminobenzidine chromogen and counterstained with hematoxylin. Control sections of bladder cancer and normal kidney were included to assess the specificity of the antibody.
Staining for HIF-1a and HIF-2a was scored positive if expression was observed in any epithelial cell nuclei. Cores were scored positive for CA9 if membranous expression was observed in 410% of epithelial cells. Staining for VEGF, PHD1, PHD2 or PHD3 was scored positive if cytoplasmic expression was observed in 410% of epithelial cells. Cores were scored positive for Dll4 if expression was observed in any endothelial cells. In addition, Dll4 was scored in the neoplastic cells within TMA cores in which it was observed at high levels, that is, levels that approximated the intensity of Dll4 expression by endothelial cells. The highest intensity score among replicate cores was used as the score for each patient. Assays and scoring were performed blind to clinical and pathological data (withheld for analysis until all data were complete).

Statistics
The w 2 test was used to evaluate associations between categorical variables. The Student's t-test was used to evaluate associations between continuous and categorical variables. The Benjamini and Hochberg false discovery rate controlling procedure was used to eliminate spurious statistical associations as a result of multiple testing (Benjamini and Hochberg, 1995). All survival analyses refer to overall survival times, in which death from any cause represents an event. Marker values were assessed as ordinal categorical data in survival analyses. Median overall survival times within each subgroup were estimated from Kaplan -Meier curves. Patients were censored in survival analyses according to the date last seen by a doctor. The log-rank (Mantel -Cox) test was used to assess the significance of univariate survival analyses. For multivariate survival models, a Cox regression analysis was used to identify independent prognostic factors. All statistical analyses were carried out using SPSS Statistics (version 16.0, SPSS, Chicago, IL, USA). The two-sided P-values of o0.05 were considered statistically significant. Cases with missing data were omitted from statistical analyses.

Reporting recommendations for tumour marker prognostic studies (REMARK) criteria
The REMARK criteria of the National Cancer Institute were used in the design, analysis and interpretation of this research (McShane et al, 2005).

Frequency and pattern of expression of hypoxia-regulated proteins
Analyses of tumours from the 177 patients yielded informative data on 155 to 177 patients. Results were not available for the remaining cases in certain assays because of the limited amounts of tissue represented in the tissue microarrays.
In all, 39% (60 out of 155) and 53% (85 out of 159) of colon cancers were positive for nuclear HIF-1a and HIF-2a expression, respectively; 21% of cases (31 out of 147) expressed both HIF-1a and HIF-2a and 51% of cases (75 out of 147) expressed either HIF-1a or HIF-2a. In addition, HIF-1a expression was only observed in cell nuclei when present ( Figure 1A), as previously described (Jubb et al, 2004). The expression of HIF-2a by neoplastic cells was predominantly cytoplasmic, with nuclear expression in a proportion of cases ( Figure 1B; Talks et al, 2000). The expression of HIF-2a was also observed in a proportion of infiltrating inflammatory cells as previously reported (Talks et al, 2000), although these data were not specifically recorded. The expression of VEGF was observed in the cytoplasm of neoplastic cells (88%, 149 out of 169), stromal fibroblasts, inflammatory cells, endothelial cells and the extracellular matrix ( Figure 1C), as previously reported (Jubb et al, 2006a).

Validation of the anti-Dll4 antibody
The human monoclonal antibody (clone 242) that specifically recognises the extracellular domain of human Dll4 did not stain U87 cells transiently transfected with empty vector (Supplementary Figure 1A), but showed membranous and cytoplasmic staining in U87 cells transiently transfected with the recombinant human Dll4 full-length gene (Supplementary Figure 1B). Human umbilical vein endothelial cells showed immunoreactivity for Dll4 that was greater in intensity when cells were treated with VEGF ( Supplementary Figures 1C and D). Normal kidney did not express Dll4, but endothelial cells in renal cell carcinoma showed endothelial expression of Dll4, consistent with previous in situ hybridisation results (Supplementary Figures 1E and F) (Patel et al, 2005). Serial sections of bladder cancer showed endothelial colocalisation of Dll4 mRNA using in situ hybridisation (Patel et al, 2006), and protein using immunohistochemistry (Supplementary Figures 1D -F).

Frequency and pattern of expression of Dll4
In 71% (125 out of 175) of colon cancer tissues analysed, Dll4 expression was observed in the cytoplasm of the endothelial cells lining vessels adjacent to cancer ( Figure 2B). Membranous localisation of Dll4 was difficult to differentiate from cytoplasmic localisation because of the slender nature of endothelial cytoplasmic processes. The expression of Dll4 was only rarely observed in the endothelial lining of large vessels around or into which cancer cells had invaded. In the whole sections of colorectal adenomas with adjacent adenocarcinoma, Dll4 was observed at identical frequencies in the endothelium found in both the adenomas and adenocarcinomas (12 out of 12, Figure 2C, Supplementary Figures  2A -C). The endothelium associated with colonic adenomas and adenocarcinomas (i.e., within two low-power fields of the tumour cells) was positive for Dll4 in 60 -100% of vessels (identified by CD34 immunoreactivity in serial sections, data not shown). Similar levels of endothelial Dll4 expression were observed in the centre of the tumour and at the invading edge. In all, 16% (28 out of 177) of colon cancers represented in the tissue microarrays expressed levels of Dll4 in the cytoplasm and membrane of neoplastic cells that approximated the intensity of expression observed in endothelium. In the whole sections, 410% of the malignant epithelial cells of the colorectal adenomas and adjacent adenocarcinomas expressed Dll4 in 8 of 12 cases (median 30%, range 10 -60%). In regions of both colorectal adenomas and adenocarcinomas, high levels of Dll4 expression were observed in neoplastic epithelial cells with goblet cell differentiation ( Figure 2D, Supplementary Figure 2D). (Goblet cells were defined morphologically.) Elsewhere, Dll4 was expressed at a lower level by neoplastic cells without goblet cell morphology ( Figure 2E). Endothelial cells adjacent to normal colon mucosa distant from cancer did not express Dll4 protein (0 out of 107). Epithelial cells lining normal colonic crypts, irrespective of their morphology, did not express Dll4 ( Figure 2F), although occasional cells in the superficial surface mucosa of normal colon stained weakly for Dll4 (data not shown). Specifically, normal colonic goblet cells did not express Dll4. The pattern of Dll4 expression by in situ hybridisation matched the immunohistochemistry results for the 12 colorectal adenomas with adjacent adenocarcinoma (Supplementary Figure 3).

Associations between molecular and pathological variables
Hypoxia-inducible factor-1a and CA9 were not significantly associated with any other immunohistochemical markers in colon cancer (Table 1). In contrast, HIF-2a was significantly associated with VEGF, Dll4, PHD1, PHD2 and PHD3. Vascular endothelial growth factor was significantly associated with endothelial Dll4, PHD1, PHD2 and PHD3. Endothelial Dll4 was also significantly associated with PHD1 and PHD3. Prolyl hydroxylase (PHD)1, PHD2 and PHD3 were significantly associated with each other. In addition, endothelial expression of Dll4 was negatively associated with nuclear HIF-2a expression by tumour cells, but this association was not significant after correction for multiple testing. Delta-like ligand 4 expression by malignant epithelial cells was not associated with any clinicopathological or molecular variables (data not shown). None of the immunohistochemical markers analysed in this study were associated with tumour stage, lymphatic invasion, vascular invasion, grade, sex or site after the false discovery rate controlling procedure was used to exclude spurious results (Table 2). Similarly, no immunohistochemical markers were associated with age using the Student's t-test (data not shown).
Endothelial Dll4 expression was not a statistically significant prognostic factor (Table 3). Although these are exploratory analyses, power calculations for Dll4 showed a statistical power       Dll4 and markers of hypoxia in colon cancer AM Jubb et al physiological and pathological angiogenesis (Duarte et al, 2004;Noguera-Troise et al, 2006;Ridgway et al, 2006). Moreover, Dll4 expression may influence the sensitivity of endothelial cells to anti-VEGF therapy, with potential implications for the use of bevacizumab and novel anti-Dll4 antibodies in colorectal cancer (Noguera-Troise et al, 2006;Ridgway et al, 2006;Williams et al, 2006;Li et al, 2007). This is the first study to perform a detailed evaluation of Dll4 expression in colon cancer, and to relate Dll4 expression to other known histological and prognostic markers.
In this series of colon cancers, Dll4 expression was observed in the cytoplasm of the endothelium lining small vessels of neoplastic, but not normal tissue. This is consistent with data from human xenografts (Noguera-Troise et al, 2006) and in situ hybridisation data for Dll4 in human bladder cancer (Patel et al, 2006) and human renal cell carcinoma (Mailhos et al, 2001;Patel et al, 2005). The lack of expression of Dll4 in the endothelial cells and epithelium of normal colonic crypts is consistent with the lack of VEGF expression by normal colorectal mucosa (Jubb et al, 2003) and the lack of intestinal toxicity and goblet cell hyperplasia observed in anti-Dll4-treated mice (Ridgway et al, 2006). These data suggest that strategies targeting Dll4 -Notch signalling may selectively affect the immature developing endothelium associated with colon cancer, but not endothelium associated with normal colonic mucosa or normal colonic crypt epithelium.
The observation of Dll4 expression by neoplastic crypts and the association with goblet cell morphology in colon adenomas and adenocarcinomas is novel. However, a report by van Es et al (2005) claims that inhibiting Notch signalling in the mouse intestine favours goblet cell differentiation in the proliferating cells of intestinal adenomas. Together with findings presented in this study, this suggests that both forward and reverse Notch signalling interact to define goblet cell differentiation in proliferating intestinal epithelium. The available data do not shed further light on the precise roles of Dll4 and its Notch receptors in the differentiation of goblet cells. Nevertheless, it is conceivable that manipulating Dll4 -Notch signalling may have therapeutic relevance to colon cancer progression.
Activity of the hypoxia-inducible transcription factor isoforms, HIF-1 and HIF-2, is controlled by turnover of their a-subunits, HIF-1a and HIF-2a, respectively (Appelhoff et al, 2004). The data show that both HIF-1a and HIF-2a are expressed at similar frequencies in colon cancer. However, it was HIF-2a that significantly associated with VEGF expression, suggesting that the hypoxic induction of VEGF is predominantly controlled by HIF-2a in colon cancer. This is consistent with data by Mizukami et al (2004) who suggest that VEGF is induced in hypoxic regions of colorectal cancer by an HIF-1a-independent mechanism. Indeed, data from Knowles et al (2003) suggest that the availability of factors, such as oxygen, ascorbate and iron, may regulate which HIF isoform is active, even when both HIF-1 and HIF-2 are expressed. Therefore, it is possible that the microenvironment of colorectal cancer favours the activity of HIF-2 over HIF-1. Carbonic anhydrase 9 is predominantly regulated by HIF-1 (not HIF-2) activity (Sowter et al, 2003), but does not significantly associate with HIF-1a expression in colon cancer. This may be due to the confounding effects of the reported HIF-1independent induction of CA9 by hypoxia (Kaluz et al, 2002). The positive association between endothelial Dll4 and epithelial VEGF in colon cancer, observed in this study, is also consistent with data from model systems in which Dll4 expression by endothelial cells is reported to be regulated by VEGF (Patel et al, 2005). Together, these data suggest that, in colon cancer, VEGF expression may be regulated by HIF-2 activity and the expression of VEGF may induce Dll4 expression in adjacent endothelial cells. This is consistent with data reported by Skuli et al (2009) who show that the deletion of HIF-2a in immortalised endothelial cells results in the downregulation of Dll4 expression.
The activity of HIF-1 and HIF-2 is reported to be regulated by hydroxylation of their a-subunits by the PHD family members (Appelhoff et al, 2004). The hydroxylation is believed to target the HIF-1 and HIF-2 a-subunits for destruction by the Von Hippel -Lindau tumour suppressor protein (Appelhoff et al, 2004). In this study the expression of the PHD proteins was strongly associated with each other, consistent with data in prostate cancer (Boddy et al, 2005), and suggesting that they are regulated by a common mechanism. Expression of the PHD proteins did not significantly associate with HIF-1a, consistent with studies in prostate cancer (Boddy et al, 2005), but were also unexpectedly positively associated with HIF-2a. The reasons for this are not clear, but suggest that the role of the PHD proteins under model physiological conditions may be different to their role in cancer. However, PHD2 and PHD3 are targets of HIFs and are induced by hypoxia (Appelhoff et al, 2004), which may explain the paradoxical association, which limits the HIF effects in a well-recognised feedback loop (D'Angelo et al, 2003). Alternatively, other factors may regulate HIF-2a expression in colon cancer, such as irondeficiency anemia (Sanchez et al, 2007), confounding the data. Further work is needed to analyse the role of the PHD proteins in regulating HIF-1 and HIF-2 activity in human neoplasia.
In survival analyses, the absence of an observed prognostic association for HIF-1a in colon cancer is consistent with previous reports (Yoshimura et al, 2004). Moreover, the lack of a prognostic association for endothelial Dll4, a VEGF target (Patel et al, 2005), is in agreement with the lack of a prognostic association for VEGF in colon cancer in this study and elsewhere (Nanni et al, 2002;Zheng et al, 2003;Jubb et al, 2006a). In this study the expression of CA9 was not prognostic in terms of overall survival, in contrast to a report that expression of CA9 is a poor prognostic factor in colorectal cancer (Cleven et al, 2007). However, the series described by Cleven et al (2007) includes rectal cancers that have received neoadjuvant radiotherapy, which may confound their results as hypoxic cells are relatively resistant to radiation-induced cell death (Wouters and Brown, 1997). Unlike other published studies on hypoxia in colorectal cancer, no patients in this series of colon cancers received neoadjuvant therapy. Therefore, treatment effects do not confound the data presented in this study.
The association between HIF-2a and prognosis in colon cancer is consistent with previously published reports in two series of colorectal cancers (Yoshimura et al, 2004;Cleven et al, 2007), although our paper is the first to report the effect of HIF-2a in a large series of large intestinal cancers that have not received radiotherapy. Mathematical modelling suggests that hypoxic microenvironments provide evolutionary pressure, clonally selecting more aggressive neoplastic cells that are able to survive hypoxic stress (Anderson et al, 2006). Moreover, molecular markers of hypoxia have been associated with radiation treatment failure in rectal cancer (Korkeila et al, 2009). Together, these data suggest that cancers expressing HIF-2a are more aggressive and may warrant more intensive treatment.
This study has two principal limitations. First, it is exploratory, and although the results largely agree with the literature, they have not been internally validated. Limitations of sample size mean that the study is not adequately powered to detect small but statistically significant differences in survival between subsets. However, such differences are arguably too small to be clinically significant. Second, tissue microarrays are not always suitable for assessing markers of hypoxia and angiogenesis. Nevertheless, provided the cohort is large, there is sufficient core redundancy and the prevalence of positive markers is sufficiently high (as is presented in this study), any effects on the data should be minimised.
In summary, Dll4 is expressed in cancer-associated endothelial cells, but not the endothelium adjacent normal colonic mucosa. This study suggests that tumour endothelial expression of Dll4 may not be a significant prognostic factor, but is significantly associated with VEGF expression. In addition, Dll4 is expressed in some neoplastic cells with goblet cell differentiation, suggesting that Notch signalling can have a more direct role in colon cancer oncogenesis. Assessing which patients express Dll4 in their tumour endothelium may be critical in predicting response to new therapies that target Notch or VEGF signalling. Furthermore, the antibody may be useful for imaging the vasculature of tumours before therapy.