Genomic Profiling in Gastrointestinal Cancer: Are We Ready To Use These Data to Make Treatment Decisions?

Molecular signatures have become more important than histologic features in making treatment choices. The success rate of these therapies depends on many factors but, perhaps most importantly, on patient selection according to the genetic analysis results of their individual tumors.

Abstract

Remarkable strides have been made in the past 10–15 years in identifying the molecular events that drive cancer. With an enormous amount of new data, including those from The Cancer Genome Atlas Project, therapies are increasingly being developed and tested in clinical trials specifically designed to target some of these molecular events. Often, molecular signatures have become more important than the histologic features in making treatment choices. The success rate of these therapies depends on many factors but, perhaps most importantly, on patient selection according to the genetic analysis results of their individual tumors.

Richard M. Goldberg

Gastric Cancer Paving the Way for Targeted Therapies

Selecting Patients by Molecular Signatures in Gastrointestinal Malignancies

Gastric cancer has historically had a poor prognosis, with a 5-year overall survival (OS) measured in months for patients with metastatic disease. A phase II trial combining 5-fluorouracil (5-FU), doxorubicin, and mitomycin-C gained quite a bit of attention in the early 1980s because of a response rate of 42% and a median OS of 5.5 months [1]. With further refinement of cytotoxic multiagent treatment regimens, the median OS for patients with metastatic gastric cancer reached approximately 9 months in a phase III trial comparing cisplatin/5-FU with docetaxel, cisplatin, and 5-FU [2]. Subsequently, OS was extended to 11.2 months with the regimen known as EOX (epirubicin, oxaliplatin, and capecitabine) in the REAL-2 (Randomized ECF for Advanced and Locally Advanced Esophagogastric Cancer 2) trial, and this appeared to be the upper limit of OS using the available combinations of cytotoxic agents [3].

Among the earliest paradigm-changing clinical trials in gastric cancer to meet the goal of using biomarker selection to enhance the outcomes of targeted therapy is the Trastuzumab for Gastric Cancer (ToGA) trial, published in 2010 [4]. HER2 overexpression was used to select patients who would be more likely to respond to the HER2 targeted agent trastuzumab. Nearly 4,000 patients were screened to find 584 patients with tumor HER2 positivity, by either immunohistochemistry (IHC) or fluorescence in situ hybridization (FISH). The patients were subsequently randomized to 5-FU or capecitabine with cisplatin, with or without trastuzumab. The patients on the trastuzumab arm had a significantly better response rate (47.3% vs. 34.5%), longer median progression-free survival (6.7 months vs. 5.5 months; hazard ratio [HR], 0.71; p = .0002), and better median OS (13.8 months vs. 11.1 months; HR, 0.74; p = .0046). This difference was even more pronounced when the criteria for declaring the presence of HER2 amplification was made more stringent, requiring 2+ IHC/FISH positivity or 3+ IHC positivity (median OS, 16.0 vs. 11.8 months; HR, 0.65; 95% confidence interval [CI], 0.51-0.83). This was a remarkable improvement in OS using a targeted treatment added to cytotoxic chemotherapy in an appropriately selected patient group. With this study, the 12-month median OS barrier that had appeared to be the upper limit of efficacy in trials combining cytotoxic agents was finally passed.

Further attempts at incorporating targeted agents have proved daunting. For example, adding panitumumab to EOX resulted in worse outcomes than those observed with EOX alone (median OS, 11.3 months vs. 8.8 months in the panitumumab arm; HR, 1.35; p = .013) [5]. Vascular endothelial growth factor (VEGF) inhibition (bevacizumab) as first-line therapy combined with cisplatin/capecitabine failed to improve the outcomes [6]. However, the antiangiogenic compound ramucirumab has been moderately successful (extending median OS by 2.3 months) as second-line therapy when that agent was combined with paclitaxel. These trials were all done in unselected patients, and a search for predictive markers that could identify subgroups of patients who might preferentially benefit from targeted therapy has continued. However, no proven markers other than HER2 overexpression have been added to standard clinical practice.

The Gastric Cancer Genome Atlas Data—Opening More Doors?

The Cancer Genome Atlas (TCGA) data in gastric cancer was presented in 2014, with some interesting findings [7]. These data included comprehensive molecular characterization of 295 gastric adenocarcinomas. Four molecular subtypes were characterized, each somewhat specific to a certain location within the stomach and histologic type (Fig. 1). Of the tumors, 9% were positive for the Epstein-Barr virus. These tumors commonly harbored PIK3CA mutations, exhibited DNA hypermethylation and PD-L1 and PD-L2 amplification, and preferentially arose in the fundus or body of the stomach. PD-1 inhibitor therapy might benefit this group, given the PD-L1/2 amplification, and this strategy is under investigation. Approximately 20% of the tumors were microsatellite unstable (MSI), most of them via inactivation of the MLH1 gene. These tumors are hypermutated, with an average of 50 mutations per megabase (Mb) compared with ≤5 per Mb in the other tumor subtypes. It is therefore possible that these tumors could be treated with a PD-1 inhibitor based on the encouraging results recently published for pembrolizumab in MSI-high colorectal cancer (CRC) and other MSI-high cancers [8]. A genomically stable group of tumors constituted approximately 20% of the cohort, and a group characterized by chromosomal instability was the largest at 50% of the 295 tumors tested. TCGA data were not coupled with the outcome data, which would be an important step in helping us apply this to clinical practice. However, clearly, certain mutational patterns can be seen, for example PIK3CA mutations are more common in Epstein-Barr virus-related (80% have a PIK3CA mutation) and MSI-high tumors (42% have a PIK3CA mutation). These data will help shed light on tumor drivers and pave the way for developing new therapies aimed at defined targets in gastric cancer.

Figure 1.

The four molecular subtypes described in The Cancer Genome Atlas gastric cancer study, their mutational patterns, and location [5, 7].

Abbreviations: CIN, chromosomal instability; EBV, Epstein-Barr virus; GE, gastroesophageal junction; GS, genomically stable; MSI, microsatellite instability.

In conclusion, trastuzumab was the first targeted therapy in gastrointestinal cancers that added significantly to chemotherapy efficacy by enabling the identification of a patient subgroup whose tumors express a predictive marker. Other targeted therapies in gastric cancer have not been as successful; however, these agents all have in common the lack of predictive markers to guide their use. The recently published TCGA data will hopefully allow for the identification of many more predictive markers to enhance the outcomes via the use of new and existing targeted therapies.

Genomics in Colorectal Cancer

Patterns of Cancer Spread and Tumor Clonal Heterogeneity

In the 1980s, we believed that cancer spread was a rather orderly process and that localized disease had a better prognosis than regional and distant disease. That its anatomy was a more important factor in determining tumor dissemination, trumping tumor biology. Dr. Leonard Gunderson made a seminal observation using data from multiple studies when he determined that T4 tumors without nodal spread actually have a worse prognosis than lower T-stage tumors with nodal involvement, an observation that changed our understanding of disease-related outcomes [9]. The new notion is that tumors are heterogeneous, with multiple subclones that have developed by the time of diagnosis. A disorderly process, rather than orderly progression, prevails [10]. When tumors become more advanced, multiple tumor clones can respond differently to therapy, making eradication of the disease more difficult with advancing disease (Fig. 2).

Figure 2.

The evolution of clonal populations [10].

Abbreviation: MRCA, most recent common ancestor.

Colorectal TCGA Data

TCGA data on colorectal cancer were published in 2012 [11]. TCGA Network analyzed 276 colorectal cancers with whole exome sequencing (and whole genome sequencing in 97 tumors), DNA copy number analysis, promoter methylation, and messenger RNA and microRNA expression. Several critical observations were made. Hypermutated tumors (defined as >12 per 10⁶ mutations) were found in 16% of the patients, 75% of which were MSI high (Fig. 3). In contrast to the gastric TCGA data, hypermutation driven by mismatch repair defects appears to be less common in colorectal cancer (12%) than in gastric cancer (∼20%). Another important finding from this work was that the colon and rectal cancers tested had the same genomic spectrum of fingerprints.

Figure 3.

Hypermutated versus nonhypermutated colorectal cancers, their MSI, CIMP, and MLH1 silencing status [11].

Additionally, 24 genes were found to be recurrently mutated; most of these were expected, such as APC, TP53, SMAD4, PIK3CA, and KRAS, but a few were surprising, such as ARID1A, SOX9, FAM123B, ERBB2, and IGF-2. ERBB2 and IGF2 could be potential new targets for drug therapy. Distinct differences were seen between the hypermutated and nonhypermutated groups. The hypermutated group was mostly driven by MLH1 silencing by hypermethylation. The remaining hypermutated tumors lacked MSI-high, CPG island methylation phenotype (CIMP), or MLH1 hypermethylation but displayed mutations in one or more mismatch repair genes or POLE mutations. The mutational patterns were different between the two types. For example, APC was mutated in 50% of hypermutated tumors compared with 81% of the nonhypermutated tumors. When considering pathway signaling, WNT signaling, TGFβ signaling, PI3K signaling, RTK-RAS signaling, and p53 signaling appeared to be common denominators among the tumors [11]. Patient outcome data were not available for that study, limiting the clinical application of the findings.

Harold Varmus, National Cancer Institute director, and now National Cancer Institute Director Emeritus and Nobel Prize winner, said of the TCGA: “While it may take years to translate this foundational genetic data on colorectal cancers into new therapeutic strategies and surveillance methods, this genetic information unquestionably will be the springboard for determining what will be useful clinically against colorectal cancers.” In some ways, this could serve a critical decoding function in our understanding of cancer, much like that served by the Rosetta Stone in archeology. These findings will help us understand and therapeutically exploit targets identified as relevant in this endeavor.

Linking Genomic Data to Patients With Known Outcomes

The European Pan-European Trials in Adjuvant Colon Cancer (PETACC) group has performed gene expression profiling studies using clinical trial data. These investigators used tumor tissue and data derived from the PETACC-3 cohort for which the patient outcomes were known. With the data from 1,113 patients included in the discovery data set, they also validated their findings in an independent set of 720 patients. They found five colon cancer subtypes, the surface crypt-like, lower crypt-like, CIMP-high-like, mesenchymal, and mixed subtypes. These each had differences in prognosis and morphological features. Also, MSI, BRAF mutation status, tumor site, mucinous histologic features, and p53 expression were significantly different in their prevalence among the various subtypes [12].

Two U.S. cooperative groups (Alliance and NRG) are developing a study of genotype tumors in patients with stage II-III colon cancer (Alliance A151304). The aim is to identify the genomic markers associated with recurrence and poor outcomes. Approximately 1,400 tumors will be investigated from patients enrolled in the Alliance N0147 study, which randomized patients to FOLFOX (folinic acid, 5-FU, oxaliplatin) with or without minus cetuximab [13]; the National Surgical Adjuvant Breast and Bowel Project CO7 study, which randomized patients to FOLFOX plus or minus bevacizumab [14]; and an earlier study of 5-FU/leucovorin versus FLOX (5-FU, leucovorin [folinic acid], oxaliplatin) [15]. Whole genome sequencing, RNA sequencing, and miRNA analyses will be performed to characterize the role of germline and somatic genetic variability with respect to prognosis, mainly in stage III colorectal cancer. This data set can also be used for pharmacogenomics analysis to study the polymorphisms associated with treatment toxicities.

Applying Genomics to Colorectal Cancer Management

Colorectal cancers can be divided into two major pathways, the chromosomal instability pathway (85%) and the microsatellite instability pathway (15%). Almost all tumors that arise through the chromosomal instability pathway are believed to be sporadic, arising from interactions between the host’s genetics and the environment. The MSI pathway is linked to a deficiency in the mismatch repair system (dMMR). The mismatch repair system is a DNA repair mechanism aimed at correcting errors in matching bases during DNA replication. dMMR tumors become MSI-high (because microsatellite regions are particularly prone to these DNA repair defects). dMMR tumors can be caused by germline mutations in the mismatch repair genes, MLH1, MSH2, MSH6, or PMS2 (Lynch syndrome, in 2%–3% of CRC cases) or sporadic inactivation of MLH1 by hypermethylation (in 12%–15% of CRC cases) [16–18].

Mismatch repair deficiency in CRC has been shown to be a prognostic and predictive marker, predicting the lack of benefit from adjuvant 5-FU chemotherapy [19]. In one study, the 5-year OS for patients with untreated stage II CRC was 56% in patients with proficient mismatch repair tumors versus 80% in patients with dMMR tumors (HR, 0.51; p = .009), clearly establishing it as a prognostic marker. When 5-FU-treated patients were compared with untreated patients with dMMR tumors, the 5-year OS was inferior in the treated patients (75% vs. 93%; HR, 3.15; p = .03) [19]. This finding is in line with what had been shown in in vitro studies, in which the MSI-high cell line HCT116 was resistant to 5-FU but sensitive to irinotecan and oxaliplatin [20]. Somatic MSH3 mutations are seen in 40% of hypermutated colorectal cancers, and MSH3-deficient cells are resistant to 5-FU but maintain oxaliplatin and irinotecan sensitivity. MSH3 mutations might therefore mediate 5-FU resistance [21]. Bertagnolli et al. investigated dMMR in patients from a Cancer and Leukemia Group B (CALGB) trial that randomized patients between 5-FU and bolus 5-FU and irinotecan (IFL) in the adjuvant setting. Although that trial showed no benefit of IFL compared with 5-FU, it did show a trend toward better outcomes in dMMR tumors in the cohort that received IFL [22].

The prognostic value of MSI has subsequently been confirmed in several studies, including a large meta-analysis with 12,782 patients [23, 24]. When separating MSI-high tumors further into those associated with Lynch syndrome and sporadic dMMR tumors, it appears that the Lynch syndrome patients might benefit from 5-FU and the sporadic patients would not [25].

In conclusion, the pooled analysis of data from randomized clinical trials has validated dMMR as a prognostic marker. Individual data sets and pooled data provide no suggestion of benefit from 5-FU-based treatment in patients with dMMR tumors, and a significant decrement was observed in OS in patients with stage II disease treated with 5-FU-based regimens. At the Ohio State University, we screen all CRC tumors for dMMR using IHC both to identify patients with Lynch syndrome and to aid in the adjuvant treatment decision. We do not give patients single-agent 5-FU if they have dMMR tumors. If patients present with stage III disease, we would use FOLFOX as the standard of care. That can be particularly important in older patients, because sporadic MLH1 hypermethylation is much more common in older patients. Some physicians would consider treating older patients with single-agent 5-FU owing to data showing no benefit from adding oxaliplatin for patients aged >75 years; however, if the tumor is dMMR, we would recommend against that approach [26].

At the Ohio State University, we screen all CRC tumors for dMMR using IHC both to identify patients with Lynch syndrome and to aid in the adjuvant treatment decision. We do not give patients single-agent 5-FU if they have dMMR tumors.

Understanding the PI3K Pathway and Aspirin Use

Aspirin has multiple effects on endothelial cells, platelets, inflammatory cells, and stromal cells and affects pathways deranged in malignancies such as the PI3K pathway (Fig. 4) [27]. PIK3CA mutations are found in approximately 15% of colorectal cancers. In the CALGB 89803 trial, which compared 5-FU/leucovorin with IFL adjuvant therapy in stage III disease, PIK3CA exon 9 and 20 mutations were neither prognostic nor predictive [28]. PIK3CA mutations, however, did predict a benefit from aspirin in the Women’s Health Study, in which colorectal cancer-specific mortality was reduced by aspirin use in patients with PIK3CA-mutated tumors (CRC-specific survival, HR, 0.18; 95% CI, 0.06-0.61) but not in patients with PIK3CA wild-type tumors (CRC-specific survival, HR, 0.96; 95% CI, 0.69-1.32) [29]. In the Nurses’ Health Study, aspirin use was associated with a lower risk of developing BRAF wild-type CRC (HR, 0.73; 95% CI, 0.64-0.83) but not a lower risk of BRAF-mutated CRC (HR, 1.03; 95% CI, 0.76-1.38) [30].

Figure 4.

The effects of aspirin on cellular pathways and microenvironment [27].

Aspirin appears to benefit patients with mutant PI3K tumors and to prevent the development of BRAF wild-type tumors and COX-2-expressing tumors. In my clinic, I recommend aspirin to patients who have had stage II or III colon cancer and undergoing curative treatment as long as they have no contraindications to aspirin use. For patients with stage II disease who do not need chemotherapy, I generally recommend aspirin to all, along with exercise and vitamin D and calcium supplementation.

Gene Expression Platforms and Adjuvant Therapy

Several companies have developed and marketed gene expression platforms that calculate recurrence scores according to gene expression. These include Genomic Health (Oncotype DX; Redwood City, CA, http://www.genomichealth.com) [31], Almac (ColDx; Craigavon, United Kingdom, http://www.almac-mcdx.com) [32], Agendia (ColoPrint; Amsterdam, The Netherlands, http://www.agendia.com) [33], and DiagnoCure (Previstage; Quebec City, Quebec, Canada, http://www.diagnocure) [34] (Table 1). They examine approximately 12–634 genes or guanylyl cyclase C gene expression in lymph nodes, by either microarray or reverse transcription polymerase chain reaction (all but one use formalin-fixed, paraffin-embedded tissue) in stage II-III disease. The best developed platform is the Genomic Health Oncotype DX assay, which studies 12 genes, including genes involved with the stroma (FAP, INHBA, BGN) and the cell cycle (Ki-67, c-MYC, MYBL2). This was first validated using the QUASAR (Quick and Simple and Reliable) adjuvant trial [31]. In stage II CRC, the test differentiates between a low 9% likelihood of recurrence versus a high 22% likelihood of recurrence using a complicated algorithm that weights each of the genes differently. This can be useful in the clinical setting but has less utility than recurrence scores in breast cancer, because those are both prognostic and predictive. It is not clear whether the difference between 9% and 22% is enough on which to base treatment decisions, because we do not know whether the test is predictive. Hopefully, new and better signatures will be discovered with the ability to discriminate between high and low recurrence risk and predict the value of adjuvant therapy.

Table 1.

Marketed gene expression array platforms

RAS Mutation: A Marker for Drug Resistance

In 2006 [35], investigators discovered a crucial biomarker in CRC, KRAS wild-type status in the tumor, which predicted the potential for an individual patient to benefit from EGFR monoclonal antibody treatment (cetuximab or panitumumab). In 2013, the predictive value of RAS mutations was further elucidated in the Panitumumab Randomized Trial in Combination With Chemotherapy for Metastatic Colorectal Cancer to Determine Efficacy (PRIME) study, which randomized patients to FOLFOX with or without panitumumab. Other RAS mutations (KRAS exon 3-4; NRAS exon 2-4) were found in 17% of patients with wild-type KRAS, and these were also predictive of shorter progression-free survival and OS when patients were given an EGFR monoclonal antibody (median OS, 26.0 months without a RAS mutation and 20.2 months with a RAS mutation; p = .04) [36]. BRAF mutations were not predictive, but were prognostic, with a shorter OS in the cohort harboring that mutation compared with patients with wild-type BRAF. Furthermore, a meta-analysis revealed that mutations in PIK3CA and inactivating mutations in PTEN, as well as KRAS, NRAS, and BRAF mutations, might all predict resistance to EGFR monoclonal antibodies [37].

BRAF mutations occur in 10%–15% of CRC and cause a constitutive activation of the mitogen-activated protein kinase (MAPK) pathway. BRAF mutations are associated with poor survival in microsatellite stable CRC but have less effect on the prognosis in MSI CRC. [38, 39] BRAF inhibitor (BRAFi) treatment has been associated with an impressive, albeit, short-lived response rate in the treatment of malignant melanoma. However, in CRC, BRAFis have not proved efficacious as single agents [40]. Mechanisms of resistance seem to differ by tumor type; in melanoma the resistance mechanism includes upregulation of the MAPK pathway through mutations in NRAS, MEK, and BRAF alternate splicing [41]. In CRC, primary resistance seems to be related to upregulation of the EGF receptor and the MAPK pathway [42]. The combination of BRAFi with MEK inhibitors or EGFR monoclonal antibodies is being tested in clinical trials and preliminary results suggest improved efficacy [43, 44]. Secondary resistance can still develop, due to alterations in the MAPK pathway [45].

BRAF mutations are associated with poor survival in microsatellite stable CRC but have less effect on the prognosis in MSI CRC. BRAF inhibitor treatment has been associated with an impressive, albeit, short-lived response rate in the treatment of malignant melanoma. However, in CRC, BRAF inhibitors have not proved efficacious as single agents.

At the Ohio State University, all patients with metastatic disease undergo tumor pan RAS testing and BRAF mutation testing. We do not use EGFR-targeted monoclonal antibodies if mutations are found. We are not yet testing for epidermal growth factor receptor copy numbers or PI3K or PTEN mutations, although that may come in time.

Integrating Genomic Information Into Prospective Clinical Trial Designs

The question remains of how to develop biomarker-driven randomized trials. The first step is to evaluate the credentials of the biomarker (i.e., to establish how confident we are about its predictive value before initiating a phase III trial). If the biomarker has very strong credentialing, such as BRAF mutations had in the investigation of vemurafenib, the phase III trial could screen patients to enrich for that biomarker and then randomize biomarker-positive patients between experimental therapy and standard therapy (or placebo). If the biomarker has less convincing credentials, the trial design could use a biomarker-stratified design randomizing both biomarker-positive and biomarker-negative patients, and the accrual of the biomarker-negative group could be guided by an interim analysis [46].

The Medical Research Council FOCUS4 study, taking place in the United Kingdom, is an integrative trial of parallel molecularly stratified randomized comparisons for patients with metastatic CRC who are fit for first-line chemotherapy (Fig. 5). The time during first-line chemotherapy is used to perform molecular characterization of the tumor. The investigators screen the tumors for BRAF mutations, PI3K mutations, PTEN loss, and RAS mutations. The trial includes an all wild-type treatment arm and a nonstratified treatment arm for unclassified patients or patients without alternations that fit the stratification. The patients with potential predictive genomic alterations are randomized to different agents or to standard regimens [47].

Figure 5.

Medical Research Council FOCUS 4 clinical trial design [47].

Abbreviations: CRC, colorectal cancer; FFPE, formalin-fixed, paraffin-embedded; IHC, immunohistochemistry; MMR, mismatch repair; OS, overall survival; P, placebo; PFS, progression-free survival; Rx, prescription.

Biomarkers in Other Gastrointestinal malignancies

Biomarkers in several other gastrointestinal (GI) malignancies have been investigated or are currently being pursued (Table 2). In pancreatic cancer, the story of undruggable cancer-driving KRAS mutations has been ongoing for the past decade. A recent review [48] has described the many attempts and challenges and how investigators are now focused on targeting the downstream pathways, including the MAPK pathway and the PI3K pathway.

Table 2.

Targeted pathways with predictive biomarkers and therapies in different cancer types

In cholangiocarcinoma, isocitrate dehydrogenase (IDH-1/2) mutations are found in ∼25% of tumors. Although they do not appear to be prognostic [49], they offer a potential druggable target, and their role in cancer formation and progression has been well characterized in mouse models [50]. Clinical trials are being implemented with IDH inhibitors.

c-MET expression is found in 26%–74% of gastroesophageal junction and gastric cancers and gene amplification in 2%–23% of cases. Although the results from early trials were promising, phase III trials have not shown the same promise. Rilotumumab was tested in a phase II trial that randomized patients to epirubicin/cisplatin/capecitabine with or without rilotumumab (a monoclonal antibody against hepatocyte growth factor [HGF], the only known c-MET receptor ligand) as first-line therapy. The study performed a subgroup analysis of c-MET expression and found that patients with >50% c-MET-expressing tumors had the greatest benefit (median OS, 11.1 months vs. 5.7 months if given placebo; HR, 0.29; p = .012) [51]. The phase III trial was stopped in November 2014 because of safety concerns, with excess deaths in the rilotumumab arm. The c-MET monoclonal antibody onartuzumab, which blocks an HGF-binding domain on the receptor, did not show benefit in a randomized phase II trial that combined it with FOLFOX [52].

Future Directions

Currently, we have clinically validated genomic testing that can identify prognostic and predictive markers. In colorectal cancer, we use RAS and BRAF mutation testing, and in gastric cancer, we have HER2 testing helping us make treatment decisions. We have many more mutational profiles of potential significance that are currently investigational but that will hopefully become relevant in the clinical setting in the near future. We perform universal screening tests for Lynch syndrome (via mismatch repair stains or MSI testing), which can also guide adjuvant therapy for stage II–III disease. We are beginning to create novel and adaptive clinical trial designs that will allow biomarker enrichment, such as were made in the ToGA trial, which used HER2 to select gastric cancer patients for trial enrollment.

With the ability of mass sequencing technologies, we can now characterize individual tumors and the individual patient’s genome to compare the differences and help determine the driving mutations for tumors. The genomic knowledge available is ahead of our ability to therapeutically target the tumors; however, we are now able to offer patients genomic-driven clinical trials. We have just begun to realize the potential of bringing molecular genomics into clinical decision-making for patients with GI tract cancer.

Disclosures

Richard M. Goldberg: Bayer, Lilly, Biothera (C/A), Sanofi-Aventis (H), Bayer, Merrimack, Sanofi-Aventis (RF).

(C/A) Consulting/advisory relationship; (RF) Research funding; (E) Employment; (ET) Expert testimony; (H) Honoraria received; (OI) Ownership interests; (IP) Intellectual property rights/inventor/patent holder; (SAB) Scientific advisory board