Improving gastric cancer preclinical studies using diverse in vitro and in vivo model systems

Abstract

Background: "Biomarker-driven targeted therapy," the practice of tailoring patients' treatment to the expression/activity levels of disease-specific genes/proteins, remains challenging. For example, while the anti-ERBB2 monoclonal antibody, trastuzumab, was first developed using well-characterized, diverse in vitro breast cancer models (and is now a standard adjuvant therapy for ERBB2-positive breast cancer patients), trastuzumab approval for ERBB2-positive gastric cancer was largely based on preclinical studies of a single cell line, NCI-N87. Ensuing clinical trials revealed only modest patient efficacy, and many ERBB2-positive gastric cancer (GC) patients failed to respond at all (i.e., were inherently recalcitrant), or succumbed to acquired resistance.

Method: To assess mechanisms underlying GC insensitivity to ERBB2 therapies, we established a diverse panel of GC cells, differing in ERBB2 expression levels, for comprehensive in vitro and in vivo characterization. For higher throughput assays of ERBB2 DNA and protein levels, we compared the concordance of various laboratory quantification methods, including those of in vitro and in vivo genetic anomalies (FISH and SISH) and xenograft protein expression (Western blot vs. IHC), of both cell and xenograft (tissue-sectioned) microarrays.

Results: The biomarker assessment methods strongly agreed, as did correlation between RNA and protein expression. However, although ERBB2 genomic anomalies showed good in vitro vs. in vivo correlation, we observed striking differences in protein expression between cultured cells and mouse xenografts (even within the same GC cell type). Via our unique pathway analysis, we delineated a signaling network, in addition to specific pathways/biological processes, emanating from the ERBB2 signaling cascade, as a potential useful target of clinical treatment. Integrated analysis of public data from gastric tumors revealed frequent (10 - 20 %) amplification of the genes NFKBIE, PTK2, and PIK3CA, each of which resides in an ERBB2-derived subpathway network.

Conclusion: Our comprehensive bioinformatics analyses of highly heterogeneous cancer cells, combined with tumor "omics" profiles, can optimally characterize the expression patterns and activity of specific tumor biomarkers. Subsequent in vitro and in vivo validation, of specific disease biomarkers (using multiple methodologies), can improve prediction of patient stratification according to drug response or nonresponse.

Fig. 1

ERBB2 gene mutation, amplification, and expression in tumors and cell lines. a ERBB2 mutations were observed in 11 (5 %) out of 220 patients. b Out of 293 GC patients, 38 (13 %) had ERBB2 amplifications, but not deletions. Interestingly, 38 patients did not have MSI-H status. a and b used The Cancer Genome Atlas (TCGA: http://cancergenome.nih.gov/). c Expression of the ERBB2 gene throughout various cancer cell lines. We used the Cancer Cell Line Encyclopedia (CCLE). The X-axis represents origins of the cancer cells, and the numeric followed by the origin indicates the number of cell lines assigned to that origin (www.broadinstitute.org/ccle)

Fig. 2

Assessment of concordance of in vitro ERBB2 protein levels. a ERBB2 protein expression, as assessed by immunochemistry (IHC) of cell lines (red bars) or xenografts (blue bars) b Assessment of ERBB2 copy number by fluorescence in situ hybridization (FISH, green bars) or silver in situ hybridization (SISH, yellow bars) in cell lines (upper panel) or xenografts (lower panel)

Fig. 3

Assessment of concordance of in vivo ERBB2 proteins levels in five distinct gastric cancer cell lines, as determined by H&E (hematoxylin and eosin stain, left panel) and immunohistochemistry (IHC, middle panel) and ERBB2 gene expression levels, as determined by silver in situ hybridization (SISH, right panel), using five separate cell (CMAs) a or xenograft b microarrays (XMAs)

Fig. 4

Signaling network is common to both ERBB2 high- and low-expressing gastric cancer patients in TCGA. a The network was delineated by our established systems biology algorithm, PATHOME [22]. The network consists of subsets of multiple KEGG pathways, as indicated by the numerals. Nodes represent gene symbols, with the depth of red indicating greater upregulation in the ERBB2 low-expressing patient group. The depth of the blue color indicates upregulation in ERBB2 high-expressing patients group. b KEGG information provided according to the numerals in a. c ERBB2 downstream signaling. We extracted ERBB2 downstream from Fig. 4a, revealing 51 downstream genes (including ERBB2 itself)

Fig. 5

Identification of genetic alterations of ERBB2 downstream signaling genes between the high- vs. low-expressing groups. a From the ERBB2 downstream genes (including ERBB2 itself), we identified genetic alterations (e.g., copy number variations, mutations) between the two GC patient groups using cBioPortal (cbioportal.org) at its default setting. The y-axis represents the percentage of patients with the altered gene, in terms of copy number variations and mutations. The red bar indicates the high ERBB2-expressing sample group, and the blue bar indicates the low ERBB2-expressing sample group. The three genes (PIK3CA, PTK2, NFKBIE), indicated by numbers showed the most alteration in the high ERBB2-expressing group, compared to the low ERBB2-expressing group. b The genetic alteration profile for the three genes (PIK3CA, PTK2, and NFKBIE) indicated in the numbers above, is shown for the high ERBB2-expressing GC patients