Expression of HOXB2, a Retinoic Acid Signaling Target in Pancreatic Cancer and Pancreatic Intraepithelial Neoplasia

Pancreatic cancer is the fifth leading cause of cancer death in Western societies with a 5-year survival rate of <10% (1). Pancreatic cancer presents at an advanced stage; thus, only 10% to 20% of patients are suitable for surgical treatment at the time of presentation (1). Clinical management of these patients is complicated by inconsistencies in the influence of conventional clinicopathologic variables on outcome suggesting that some of these variables lack accuracy. In addition, preoperative assessment of some variables such as lymph node metastases is difficult. Whereas in other cancers assessment of aberrations in gene expression that cosegregate with therapeutic response and outcome are being adopted routinely to increase predictive power (e.g., ER and HER-2/neu in breast cancer), there remain no molecular markers of clinical utility in pancreatic cancer. This highlights the need for the identification of novel regulatory pathways important in pancreatic cancer that may also have diagnostic, therapeutic and prognostic utility.

There is now compelling histopathologic and molecular evidence to support the evolution of pancreatic cancer through a series of noninvasive duct lesions called pancreatic intraepithelial neoplasia (PanIN; refs. 2, 3). Early duct lesions designated PanIN-1A and PanIN-1B show minimal cytologic and architectural atypia and are associated with activating K-ras mutations (4), shortened telomeres (5), and overexpress p21^WAF1/CIP1 (6). PanIN-2 lesions exhibit mild to moderate cytologic and architectural atypia and are associated with loss of p16^INK4A expression (7) and cyclin D1 overexpression (6). PanIN-3 exhibits significant cytologic and architectural atypia, manifests p53 mutations (8), and loss of DPC4/Smad4 expression (6). These molecular aberrations increase in frequency with advancing PanIN lesions through to invasive cancer.

During vertebrate development, retinoic acid (RA) signaling is important for the correct patterning of embryonic structures (9). Endodermal expression of pdx-1 (a homeobox-containing transcription factor essential for pancreatic development) is induced by RA (10) and marks a pluripotent population of cells that give rise to all cell types in the pancreas. RA signaling regulates pancreas exocrine lineage selection, and treatment with RA analogues can effect a shift from an acinar to a ductal phenotype through epithelial-mesenchymal interactions (11). Such a shift from an exocrine to a predominantly ductal phenotype is characteristic of mouse models of pancreatic cancer development. In addition, pancreatic stellate cells, which are essential for the development of fibrosis associated with chronic pancreatitis and pancreatic cancer, store retinoids in fat droplets, and in turn can have their function altered with RA analogue treatment in vitro (12). The retinoid signal is transduced by two families of nuclear transcription factors: RA receptors (RAR) and retinoid X receptors, that are members of the nuclear receptor superfamily, which in the presence of ligand heterodimerize to activate the transcription of target genes through RA response elements (13). Although few RA response elements have been identified, one of the mechanisms by which retinoids exert their effects is thought to be through regulation of HOX gene expression (9, 14).

Homeobox genes are transcription factors with established roles in development and cell function. The homeobox is a highly conserved 183-bp DNA sequence coding for a 61-amino-acid domain, the homeodomain (15). This region binds DNA elements, primarily those that contain a TAAT core motif (16). Accordingly, homeodomain containing proteins act as both activators and repressors of transcription. Human class 1 homeobox genes called HOX genes consist of 39 genes arranged in four clusters HOXA, HOXB, HOXC, and HOXD localized on chromosomes 7, 17, 12, and 2, respectively (17). Mammalian development requires a complex interaction of HOX gene networks, with HOX gene expression commencing during gastrulation and collectively controlling the identity of various regions along the body axis from the hindbrain to the tail (18, 19). Aberrant expression of HOX genes has been implicated in the development of solid tumors including renal carcinoma (20), colon cancer (21), ovarian carcinoma (22), and breast carcinoma (23, 24). Given the emerging importance of developmental pathways in pancreatic cancer such as Notch (25) and sonic hedgehog (26), the role of RA signaling in early pancreas development and evidence of RA signaling and homeobox gene network dysregulation in carcinogenesis, we present data suggesting that aberrant RA signaling may be important in pancreatic cancer. Based on these data, we assessed HOXB2, a RA-responsive gene, and show that ectopic expression of HOXB2 occurs in a significant proportion of pancreatic cancer, is detectable in a proportion of PanIN, and is associated with a poor prognosis, supporting a potential role of HOXB2 in the biological behavior of some pancreatic cancer.

RNA preparation and transcript profiling. Ethical approval was obtained from five teaching hospitals (The Princess Alexandra Hospital, Brisbane, Australia, Westmead Hospital, Concord Hospital, Royal Prince Alfred Hospital, and St. Vincent's Hospital Campus in Sydney, Australia) for the acquisition of fresh and archival tissue and recording of clinicopathologic data. Multiple samples of pancreatic tissue of ∼500 mg were excised intraoperatively from 12 patients, undergoing pancreatic resection for pancreatic cancer, immediately snap frozen in liquid nitrogen and stored at −80°C, before RNA extraction. Total RNA was isolated from 12 pancreatic cancer specimens and six macroscopically and microscopically normal appearing pancreas from the same patients (matched). Biotinylated cRNA for Affymetrix Genechip hybridization was prepared through a single round of reverse transcription with Superscript II (Life Technologies, Rockville, MD) followed by second strand synthesis to create double stranded cDNA. After purification the cDNA was transcribed and labeled using a T7 polymerase (Enzo Technologies, New York, NY) and purified (27). Hybridization cocktails were prepared as per the Affymetrix protocol (Affymetrix, Santa Clara, CA) and quality assured on Affymetrix Test3 arrays, before hybridization to HG-U133A and B oligonucleotide microarrays.

Data analysis. A relational database was constructed using FileMaker Pro (FileMaker, Inc., San Francisco, CA) to facilitate multiple queries of gene expression data generated from the above experiments and public domain data available electronically from the Internet. The database incorporated (a) transcript profiles of pancreatic cancer and normal pancreas from the experiments done in this study (absolute values); (b) mathematical algorithms programmed within the database to generate fold change comparisons between the average expression across all samples of pancreatic cancer to the average in normal pancreas; (c) linear statistical analyses generated using the Affymetrix Data Mining Tool Software (MAS 5.0), which included t test and Mann-Whitney U test data for comparisons between normal pancreas and pancreatic cancer and (d) interactive molecular pathway maps were generated using GenMAPP software (Gladstone Institutes UCSF, San Francisco, CA, http://www.GenMAPP.org/default.html), designed to incorporate transcript profile data into maps of known pathways including those involved in carcinogenesis and development. Data files using Swissprot identification numbers were uploaded into the program, and various pathway maps available as part of the package were used to model numerous pathways. An existing RA signaling GenMAPP was modified to include all molecules thought to be regulated by RA signaling and is presented in Fig. 1. Statistical data were generated using the t test and Mann-Whitney U tests to compare the average expression across samples of pancreatic cancer to the average expression of all samples of normal pancreas for the GenMAPP that is presented.

Patient cohort. We identified a cohort of 128 patients with a diagnosis of pancreatic adenocarcinoma that underwent pancreatic resection or biopsy between January 1972 and November 2001 with available archived tissue. This cohort represents a subset of a previously described group of 348 patients (28). Archival formalin-fixed, paraffin-embedded tissue from all 128 pancreata that were resected or biopsied were used to construct seven pancreatic cancer tissue arrays, which contained up to 55 × 1.6 mm cores per slide. Conventional sections of 26 cases of normal pancreas from areas distal to the pancreatic cancer were used to assess gene expression in benign ductal epithelial cells and PanIN lesions.

For this cohort, the average age at diagnosis was 63.8 years (median, 66.5; range, 34-86; Table 1). Of the 128 patients, 76 were from pancreatic resections, 46 intraoperative incision biopsies, and 6 postmortem specimens. Median follow-up for the cohort was 7.6 months (range, 0-117 months). Eight patients were alive at the census date (September 21, 2002). Median disease-specific survival was 7.25 months. For the resected group of 76 patients, 39 (51%) had lymph node metastasis (Table 1). The mean tumor size was 31 mm. Resection margins were microscopically free of tumor in 40 patients (53%). Poorly differentiated tumors occurred in 25 patients (33%). Median follow-up was 11.0 months with a median disease-specific survival of 10.1 months, 1-year survival of 48.6%, and 5-year survival of 11%. The 30-day mortality for resection was 2 (3%).

Immunohistochemistry. Pancreatic tissue microarrays were cut at 4 μm, deparaffinized, and rehydrated before unmasking in target retrieval solution (EDTA and citrate, DAKO Co., Carpinteria, CA) in a microwave for 30 minutes. Using a DAKO autostainer, endogenous peroxidase activity was quenched in 3% hydrogen peroxide in methanol followed by avidin/biotin and serum-free protein blocks (DAKO). Sections were incubated for 30 minutes with 1:200 anti-HOXB2 (P-20) goat polyclonal antibody (Santa Cruz Biotechnology, Santa Cruz, CA). A streptavidin-biotin peroxidase detection system was used according to the manufacturer's instructions (LSAB label + link kit; DAKO) with 3,3′-diaminobenzidine as a substrate. Counterstaining was done with Mayer's hematoxylin. HOXB2-positive breast cancer was used as a positive control (24), whilst ovary was used as a negative control. Antibody specificity was confirmed using blocking peptide sc-17165 P (Santa Cruz Biotechnology), which abrogated nuclear staining for immunohistochemistry and eliminated a specific band on Western blotting using pancreatic cancer cell lines. In addition, mRNA expression in cell lines using reverse transcription-PCR correlated with protein expression on Western blotting.

Immunohistochemical scoring. Up to four separate samples of pancreas were examined per patient. Staining was assessed by two blinded independent observers (D.S. and J.G.K.). Standardization of scoring was achieved by comparison of scores between observers and by conferencing, where any discrepancies were resolved by consensus. Scores were given as the percentage of nuclei staining positive within the representative area of the tissue microarray core and the absolute intensity of nuclear staining on a scale of 0 to 3 (0, no staining; 1, slight/weak heterogenous nuclear staining; 2, strong homogenous nuclear staining; and 3, intense homogenous nuclear staining). The criteria to achieve a positive score were HOX B2 nuclear intensity of >1 in >20% of nuclei.

Statistical analysis. Kaplan-Meier and the Cox proportional hazards model were used for univariate and multivariate analysis using Statview 5.0 Software (Abacus Systems, Berkeley, CA). P < 0.05 was accepted as statistically significant. Those factors that were prognostic on univariate analysis were assessed in a multivariable model to identify factors that were independently prognostic and those that were the result of confounding. This analysis was done sequentially on all patients who had available tissue (n = 128) and on a subgroup of patients who underwent operative resection (n = 76).

Transcript profiling data analysis. Whereas previous transcript profiling studies have been limited to identifying single genes aberrantly expressed in pancreatic cancer (29, 30), we employed a strategy that used GenMAPP software to identify molecular pathways in which a significant proportion of genes showed aberrant expression. Using this approach, we confirmed aberrations in molecular pathways known to be important in pancreatic cancer (transforming growth factor–β signaling, cell cycle regulation, and apoptosis; data not shown). In addition, we identified aberrant expression of a significant number of components of RA signaling (Table 2; Fig. 1). RAR-α and RAR-γ were up-regulated 2.9- and 2.2-fold, respectively, in pancreatic cancer compared with normal pancreas. Expression of a substantial number of known RA-responsive genes was also altered in pancreatic cancer, consistent with dysregulated RA signaling activity, primarily demonstrating up-regulation of genes downstream of RAR-α. A substantial number of genes regulated by RA and known to be highly expressed in pancreatic cancer and PanIN from other studies, were also up-regulated: S100 calcium binding protein P (S100P; ref. 31; 152-fold), MUC4 mucin (ref. 32; 24.6-fold), matrix metalloproteinase 9 (MMP9; ref. 33; 2.0-fold), Id-1 (ref. 34; 2.3-fold), urokinase plasminogen activator receptor (uPAR; ref. 35; 13.5-fold), and heparin-binding epidermal growth factor-like growth factor (HB-EGF; ref. 36; 2.5-fold; Table 2). Other genes, yet to be characterized in pancreatic cancer but thought to be regulated by RA were also aberrantly expressed, including a RA-induced G-protein–coupled receptor (26.3-fold) and RAR responders RARRES 1 (16.5-fold) and RARRES 3 (3.8-fold; Table 2).

Studies of hindbrain development have provided the greatest insights into the mechanism of RA signaling. RA-dependent lineage restriction in rhombomeres 3 and 5 is marked by krox20 and HOXB2 expression. RA, through an as yet unknown mechanism that may involve CEBPβ (37), results in increased krox20 expression, which in turn increases HOXB2 expression by directly binding promoter elements of HOXB2 (38). krox20 also suppresses HOXB1 expression. The expression profile in the present study is consistent with activity of this pathway of HOXB2 regulation (Fig. 1). In addition, the variant promyelocytic leukemia fusion protein PLZF-RARA also regulates HOXB2 expression through a similar mechanism and is thought to be important in promyelocytic leukemia development and resistance to RA therapy (39). For these reasons, HOXB2, a previously uncharacterized gene in pancreatic cancer, which showed a 6.7-fold increase (P < 0.001) compared with normal pancreas, was selected for further study.

HOXB2 expression in pancreatic cancer and pancreatic intraepithelial neoplasia. Representative examples of HOXB2 immunostaining are shown in Fig. 2. Nuclear expression was identified in 48 of 128 cancers (38%). When HOXB2 expression was present within the tumor, >80% of the nuclei stained positively. HOXB2 expression was detected in the histologically normal pancreatic ducts of 2 of 26 (8%) patients, in 1 of 24 (4%) PanIN-1A lesions, 3 of 20 (15%) PanIN-1B, 3 of 10 (30%) PanIN-2, and 1 of 4 (25%) PanIN-3 lesions, showing that HOXB2 expression occurs in PanIN and may play a role in the evolution of PanIN.

HOXB2 expression in the whole cohort was associated with a poor outcome (median survival, 5 versus 9.9 months; log-rank P < 0.0001; Fig. 3A). In addition, operative resection (P < 0.0001), low-stage (P < 0.0001), and non–poorly differentiated tumors (P = 0.0152) were associated with significantly improved survival using Kaplan-Meier analysis. However, multivariate analysis identified resection and stage as the only independent prognostic factors when modeled together with degree of differentiation and HOXB2 status (Table 3A). Whereas HOXB2 expression was identified in 32 of 52 (62%) unresected tumors, it was present in only 16 of 76 (21%) resected pancreatic cancers. Hence, HOXB2 expression was associated with nonresectable tumors (χ²; P < 0.0001) and consequently was not an independent prognostic factor. Operative resection did not benefit those patients whose tumors expressed HOXB2 (log-rank P = 0.37; Fig. 3B) but was beneficial to those patients who did not express HOXB2 (median survival, 14.0 versus 3.7 months; log-rank P < 0.0001; Fig. 3C). Survival for patients with tumors that were HOXB2 negative and who underwent resection was significantly longer than survival in all other groups (14 versus 4.3 months; log-rank P < 0.0001; Fig. 3D). Hence, in this cohort, lack of HOXB2 expression cosegregated with operative resectability, with only those who were HOXB2 negative having a survival advantage from operative resection.

Survival analysis of patients that underwent operative resection identified decreased survival associated with HOXB2 nuclear expression (median survival, 6.75 versus 14.0 months; log-rank P < 0.0001; Fig. 3E). Kaplan-Meier analyses identified clear margin status (P = 0.0014), tumor size of ≤20 mm (P = 0.0375), and no lymph node involvement (P = 0.0235) as being associated with a survival advantage (Fig. 3F-H). Degree of differentiation was not associated with a survival advantage (P = 0.0582). HOXB2 expression and involved surgical margins were independent prognostic factors when modeled against all combinations of HOXB2 expression, involved surgical margin, lymph node involvement, and tumor size in the subgroup of patients who underwent surgical resection (Table 3B-D).

Expression profiling identified differential expression of a significant number of RA signaling pathway components and downstream responders in pancreatic cancer compared with normal pancreas. These included some genes that are known to be associated with pancreatic cancer: MUC4, MMP9, Id-1, uPAR, HB-EGF, and S100P, as well as novel candidates. Although there is substantial evidence implicating aberrant retinoid signaling in carcinogenesis (e.g., acute promyelocytic leukemia; ref. 40), the mechanisms by which retinoid target genes exert these effects remains to be elucidated. Here we present evidence, implicating RA signaling in pancreatic cancer and show that a RA-responsive homeodomain transcription factor, HOXB2, is ectopically expressed in a significant proportion of pancreatic cancer, with a profound association with tumor progression. HOXB2, which is not normally expressed in the pancreas at any stage during development or adult life, was expressed in 38% of pancreatic cancers and seemed to occur during the development of a proportion of PanIN. Ectopic HOXB2 expression was associated with nonresectable tumors and was an independent prognostic factor in resected tumors when modeled with known clinicopathologic prognostic factors. In addition, only those patients that were HOXB2 negative obtained a survival advantage with operative resection.

Numerous lines of evidence from separate studies suggest the importance of individual RA signaling components in pancreas development and pancreatic cancer evolution. RA regulates early instructive signals from lateral plate mesoderm that is essential for specification of endoderm towards a pancreatic fate (41). RARs also regulate exocrine pancreatic development at later stages (42), by modulating lineage selection favoring ductal rather than acinar differentiation primarily through RAR-α (10, 11). Reactivation of developmental pathways, specifically those that regulate exocrine cell lineage, have been implicated previously in the early development of pancreatic cancer (25) and other pathways that determine duct cell versus acinar cell differentiation involving RA signaling, may also be important. In addition, forced expression of cellular retinol binding protein (CRABP1), a mediator of RA signaling, in transgenic mice results in the development of poorly differentiated pancreatic cancer (43), further supporting a role of aberrant RA signaling in pancreatic cancer evolution. The transcript profile data presented here suggests the RA signaling pathway has a role in pancreatic cancer, specifically, a number of RA-responsive genes known to be important in pancreatic cancer and PanIN development were aberrantly expressed in this study: MUC4 mucin is overexpressed in a significant proportion of pancreatic cancer (44) and PanIN (32) and can be induced through RAR-α activation (45); similarly, MMP9 is expressed in pancreatic cancer (33) and is up-regulated by RA treatment (12) as is uPAR, HB-EGF, and p21^WAF1/CIP1 (46). Id-1, which antagonizes basic helix loop helix proteins, inhibits differentiation and can enhance cell proliferation is overexpressed in PanIN lesions (34) and is also RA responsive (47). In the present study, HOXB2 expression was also detected in PanIN lesions. RA seems to exert its effect on exocrine lineage selection towards a ductal phenotype during development through laminin-β1 (11), which was up-regulated in the present study. As was the putative tumor suppressor gene RARRES 3 (48). Stra6, a gene whose function is yet to be determined responds to RA, is up-regulated in colon cancer (49) and was up-regulated in this study as was transglutaminase 2 (TGM2) and tumor necrosis factor α–induced protein 2 (TNFAIP2), both also RA-responsive genes (47). There is clear evidence supporting RA regulation of HOXB2 expression from studies of hindbrain patterning and branchial arch development (38); however, the mechanism by which RA signaling imparts its effects on the HOX network and cellular function is poorly understood. There were, however, some inconsistencies in the data where downstream targets of RA were down-regulated such as HOXD8. Presumably, other mechanisms can also regulate the expression levels of these transcripts other than through RA signaling alone. Validation of these data with manipulation of RA signaling is required to further investigate a putative functional role in pancreatic cancer; however, the strength of data from the literature and evidence presented here makes a strong case for an important role in pancreatic cancer.

Multivariate analysis identified HOXB2 expression as an independent predictor of survival in the subgroup of patients that underwent pancreatic resection. Although HOXB2 expression was not identified as an independent predictor of survival in the whole cohort due to its association with resection, lack of HOXB2 expression combined with surgical resection conferred a significant survival advantage. Because all known prognostic indicators in pancreatic cancer, such as tumor size, resection margins, and lymph node status can only be determined post resection, HOXB2 expression has potential utility as a prognostic indicator in pancreatic cancer, especially because it seems to have a profound independent influence on survival, with the advantage that it can be assessed using biopsy techniques without resection. We have previously identified that loss of DPC4/Smad4 expression is associated with poor outcome in pancreatic cancer (28). However, HOXB2 may have more efficacy as a marker of prognosis in pancreatic cancer, as ectopic expression of a protein is a more reliable indicator than loss of expression. Although pancreatic resection offers the best chance of cure in patients with pancreatic cancer, it is a procedure which carries significant morbidity and mortality. The development of a reliable preoperative assessment of HOXB2 status would be an important addition to a physician's limited diagnostic armamentarium in this disease and may be used, together with current clinicopathologic variables of disease outcome, to determine a patient's suitability for operative resection.

In conclusion, gene expression profiling of pancreatic cancer has suggested that RA signaling is a potentially important regulatory pathway in pancreatic cancer evolution. Ectopic expression of HOXB2 in pancreatic cancer is a frequent occurrence, an event which manifests in the development of PanIN in a proportion of cases, and is possibly a consequence of aberrant RA signaling. Current prognostic factors for pancreatic cancer remain poorly defined, depend upon examination of the resected pancreas, and cannot be accurately determined preoperatively. Assessment of HOXB2 expression may provide an alternative method for determining the suitability for resection and the prognosis of patients with pancreatic cancer. Further study to determine the effects of ectopic HOXB2 expression and other components of the HOX transcriptional network, its relationship to RA signaling, and clinical utility in pancreatic adenocarcinoma is required.