The mitogen activated protein kinase (MAPK) pathway is an enriched pathway in oesophageal and gastric adenocarcinomas

HOPACH clustering was performed to visualise groups with similar GSEA enrichment scores in a cohort of 75 largely chemo-naive patients with oesophago-gastric junction (GOJ) malignancies (demographics in table S1)21 for signalling pathways denoted in the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Biocarta databases (figure 1A). A large group (38/75, 50.7%) of samples enriched for overexpression of signalling pathway genes was identified (p<0.001), and was more stable than the non-overenriched groups using non-parametric bootstrap resampling (n=10 000) (figure 1A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

MAPK pathway is the most overenriched pathway in oesophago-gastric cancers. Hierarchical ordered partitioning and collapsing hybrid clustering (A) was applied to the enrichment scores for the KEGG signalling pathway in the cohort of 75 cancer samples. The distance matrix allows visualisation of the similarity of gene expression: red represents close proximity and white a large distance. Clusters with similar expression appear as blocks on the diagonal axis; solid lines indicate a stable cluster of samples enriched for overexpression of signalling pathway genes (n=38/75, indicated in the highlighted box). The signalling pathways on the KEGG database which were statistically enriched in the cancer dataset are shown in order of frequency (B). Pathways signalling through the MAPK pathway (hashed bars), including ErbB2 and Toll-like receptor which the KEGG database annotates separately, represent the most enriched pathways in oesophago-gastric cancers. The dashed line shows those pathways enriched in >25% cases. The six pathways enriched in more than 25% of the samples from the cancer cohort (above dashed line in panel B) were then examined in a progression cohort (C) consisting of samples from normal oesophagus (NE) (n=6), Barrett's oesophagus (BE) (n=21), dysplastic Barrett's (DYS) (n=30) and adenocarcinoma (AC) (n=8).

The signalling pathways from KEGG displaying statistically significant enrichment in the case cohort (p<0.05) are displayed in figure 1B. The MAPK pathway was statistically enriched in the highest number of samples (32/75 samples, 42.7%). Three MAPK related pathways (MAPK, ErbB2 and Toll-like receptor), out of the 21 statistically enriched pathways, were relevant in 80% of samples. The same analysis was performed using pathways from the Biocarta database and six out of the top 20 (30%) statistically enriched pathways (MAPK, EGF, PDGF, p38, Erk and Akt) were found to contribute to the MAPK pathway (data not shown). If oesophageal and gastric cancers were considered separately, the MAPK related pathways remained the most commonly enriched pathway suggesting that tumours from adjacent anatomical sites, including around the GOJ, have similar pathway dependence.

Signalling pathways are not independent and therefore some caution must be exercised in determining the importance of individual pathways from these analyses. However, if pathways are biologically important, the frequency of their enrichment should increase progressively over the pathogenesis of the disease. To examine this, enrichment of KEGG pathways, known to be relevant in more than 25% of cancer cases in the microarray analysis (MAPK, GNRH, ErbB2, insulin, Wnt, phosphatidyl inositol and calcium which are displayed above the dotted line in figure 1B), were assessed in samples representative of the pathogenic stages of oesophageal and GOJ adenocarcinoma (normal squamous oesophagus, Barrett's metaplasia, Barrett's with dysplasia and adenocarcinoma). MAPK and Wnt were the only pathways with progressive enrichment over the progression continuum (p<0.05, figure 1C). While the Wnt pathway is interesting in itself and has a known role in colonic carcinogenesis,32 it is not easily targetable by drugs at present. We therefore focused our efforts on the MAPK pathway.

Phospho-Erk status confirms MAPK as a key pathway

The importance of the activation status of the MAPK pathway, measured by phospho-Erk immunostaining, was confirmed in a large independent cohort of 434 cases of oesophageal and gastric adenocarcinomas (figure 2A, demographics in table S1).21 In all, 148/434 (34.3%) of cases expressed high levels of phosphorylated Erk (2+, 3+) (figure 2B). This is in the same range of the samples displaying MAPK pathway enrichment (42.7%) from our expression profiling data.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

MAPK pathway activation is driven by multiple active receptor tyrosine kinases (RTKs). MAPK activation was confirmed by examining P-Erk status in 434 samples on a tissue microarray (A). Two representative cases, one with P-Erk and one lacking P-Erk are shown. Percentage of the 434 cases with and without P-Erk (B).

Constitutive phosphorylation of Erk may also be a consequence of a single base substitution in codon 12 or 13 of RAS; however, no mutations were identified in a subset of 67 samples from the cancer cohort (data not shown).

Selective inhibition of activated RTKs inhibits MAPK signalling, cell growth and results in a loss of cell viability

We next profiled a panel of 14 oesophageal and gastric cell lines using an RTK array (table S2 and examples of arrays are shown in figures 3⇓–5). To determine whether RTK profiling could be used to infer therapeutic decisions, relevant TKIs were selected based on the RTK profile. Inhibition of the downstream MAPK component Mek was also performed since resistance to RTK inhibition develops clinically and so alternative approaches need to be pursued. A range of RTK profiles were seen in the cell lines and varied from no RTK activation, MKN1 (figure 3A), to one dominant active RTK in OE19, MKN45 and KATOIII (figure 4A–C) and highly complex profiles with multiple active RTKs in HSC39 and OE33 (figure 5A,B).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Receptor tyrosine kinase (RTK) inhibition is ineffective in inducing cell death in a cell line without RTK activation. MKN1 RTK activation profile (A) and full RTK array map (B). Each RTK is represented in duplicate on the array and each dot in the four corners is a loading control. A number of negative controls are included: IgG1, IgG2A and IgG2B are mouse-IgG controls, IgG is a goat control and a PBS control. The proliferation dose–response of MKN1 to tyrosine kinase inhibitors is shown in (C). Data are presented as median (±95% CIs) as a percentage of DMSO treated cells.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Single tyrosine kinase inhibitors (TKIs) or Mek inhibition, identified from receptor tyrosine kinase (RTK) array, are sufficient to induce cell death and downstream targets loss of phosphorylation in OE19, MKN45 and KatoIII. RTK activation profile (A–C), proliferation dose–response to TKIs (D–F), cell cycle profile in response to TKIs (G–I) and western blotting for downstream MAPK components Atk and ERK (J–L) for OE19, MKN45 and KatoIII. Each active RTK is numbered with the corresponding name indicated below the membranes and each pair of dots in the four corners is loading controls. Proliferation and cell cycle profiles were performed following treatment with a variable concentration of TKIs (alone or in combination). Data are presented as median (±95% CIs) as a percentage of DMSO treated cells. Only the single inhibitors and the best combinations are presented; a full summary of all is given in table 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Multiple tyrosine kinase inhibitors (TKIs) or Mek inhibition, identified from receptor tyrosine kinase (RTK) array, are required to induce cell death and downstream targets loss of phosphorylation in HSC39 and OE33. RTK activation profile (A and B), proliferation dose–response to TKIs (C and D), caspase 3/7 activation dose–response to TKIs (E and F) and western blotting for downstream MAPK components Atk and ERK (G and H) for HSC39 and OE33. Each active RTK is numbered with the corresponding name indicated below the membranes and each pair of dots in the four corners are loading controls. Proliferation and cell cycle profiles were performed following treatment with a variable concentration of TKIs (alone or in combination). When combinations of drugs were used, the concentration of the TKI targeting the most active RTK (FGFR2 for HSC39 and Met for OE33) was varied (from 0.01 nM to 100 μM; exact range depending on the cell type) and the concentration used for the other TKIs was kept constant at 1 μM. Data are presented as median (±95% CIs) as a percentage of DMSO treated cells. Only the single inhibitors and the best combinations are presented.

For each cell line, TKIs were used singly or in combination on the basis of the RTK activity profile to establish whether RTK(s) could be predicted to be critical for cell proliferation and survival on a case by case basis. The complete experimental data are shown in table 1.

In MKN1, with no active RTKs, TKIs or Mek inhibition did not inhibit proliferation until the μM range (figure 3C).

A single TKI targeting the most highly phosphorylated RTK (EGFR/ErbB2 for OE19, Met for MKN45 and FGFR for KatoIII; figure 4A–C) in cells with a simple RTK profile inhibited proliferation with IC50 of 100 nM (95% CI 39 to 120) for OE19, 5.6 nm (95% CI 4.0 to 7.7) for MKN45 and 4.8 nM (95% CI 3.4 to 6.7) for KatoIII (figure 4D–F). This was accompanied by a cytotoxic effect with an increased subG0 fraction in OE19 and a cytostatic effect with an increased G0/G1 fraction in MKN45 and KatoIII together with a fourfold increase in caspase 3/7 activity in OE19 and a twofold increase in MKN45 and KatoIII at 24 h (figure 4G–I and table 2). Alternatively, the use of a Mek inhibitor was also sufficient to inhibit proliferation with IC50 values of 95 nM (95% CI 34 to 263) for OE19, 14 nM (95% CI 0.4 to 46) for MKN45 and 5.8 nM (95% CI 3.9 to 8.4) (figure 4D–F). Apoptosis was also induced with a cytotoxic effect in OE19 and cytostatic effect in MKN45 and KatoIII by 24 h (figure 4G–I). These effects correlated with the abrogation of Erk phosphorylation at a TKI or Mek inhibitor concentration comparable with the IC50 (figure 4J–L). However, loss of Akt phosphorylation was only seen for the TKI and but not for Mek inhibition (figure 4J–L). The addition of further TKIs to inhibit RTKs coactivated at a lower level had no cumulative effect; and when used alone had little efficacy (table 1). This finding was in keeping with the failure of these additional TKIs to suppress downstream pathway activity, especially phosphorylation of ERK and Akt (table 1).

HSC39, with constitutive activation of four RTKs (EGFR, ErbB3, FGFR2 and Met), demonstrated a different response to RTK inhibition (figure 5A). First, HSC39 was sensitive to a single TKI targeting the most highly active RTK alone, FGFR2. The FGFR2 inhibitor resulted in a dramatic reduction in proliferation with an IC50 of 5.9 nM (95% CI 4.8–7.3 nM, figure 5C, table 1); a cytotoxic effect (figure 5E) and a fivefold increase in caspase 3/7 activation within 24 h (table 2); and suppressed phosphorylation of ERK and Akt (figure 5G, table 1). A similar effect was seen with Mek inhibition although 72 h incubation was necessary to have the cytotoxic effect (data not shown). Treatment with single TKIs targeting the other coactivated RTKs had negligible effects (figure 5C,E, table 1). However, in contrast to the previous cell lines, a combination of TKIs to target FGFR2, EGFR, ErbB2 and Met, which were all constitutively active in HSC39, was significantly better than FGFR2 inhibition alone, though the absolute benefit was small (figure 5C,E and table 1).

OE33 displayed a complex activation map with six constitutively active RTKs: EGFR, ErbB2, ErbB3, FGFR3, Met and Ret (figure 5B). Treatment with a single TKI against any of these RTKs failed to inhibit proliferation unless doses in the μM range were used (figure 5D) and none fully abrogated ERK or Akt phosphorylation (table 1, figure 5H). However, the combined use of TKIs targeting Met (or with siRNA against Met, figure S1) and the ErbB family or of a Mek inhibitor inhibited OE33 proliferation (IC50 124 nm, 95% CI 80–193 nM for TKI combination and IC50 9.1 nm, 95% CI 3.7–22 nm for Mek inhibition; figure 5D), triggered caspase activation by 48 h with a small but consistent increased subG0 fraction at 24 h and completely suppressed downstream pathway activation (table 2 and figure 5F). In contrast to HSC39, the further addition of inhibitors targeting Ret and/or the FGFR family was not superior to the combined EGFR/ErbB2 and Met inhibition (figure 5D and table 1).

In summary, in 3/5 cases, inhibition of the single most active RTK predictable from the RTK array or using a Mek inhibitor had the maximal therapeutic effect (KATOIII, MKN45, OE19); in two cases with a more complex RTK activation profile, the use of TKI combinations or using a Mek inhibitor had an additional benefit (HSC39) or was essential to achieve the maximal effect (OE33). Furthermore, the choice of TKI would have been arduous and not as predictable without the RTK array and functional data.

RTK activation in a patient cohort

Since using an RTK array in cell lines could help in drug selection to ensure the maximal therapeutic effect, we postulated that a similar approach could be used in patients.

The phosphorylation status of 42 different RTKs in 46 snap frozen or fresh tumour samples (95% chemonaive) was measured using the same membrane array to determine the extent and range of RTK activation potentially in largely chemo-naive (93%) oesophageal and gastric tumours potentially responsible for this MAPK signalling (demographics in table S1 and figure 6). No RTKs were active in 8/46 cases (17.4%); EGFR alone was active in 8/46 cases (17.4%); two RTKs, usually members of the EGFR family, were active in 13/46 cases (28.3%); and multiple RTKs (median of 9) were active in 17/46 cases (37.0%) (figure 6A,B). The range of activated RTKs was extremely diverse with 38 different receptors, representing 13 distinct families, showing phosphorylation in at least one tumour case (table S3). These data confirm that MAPK activation can be driven by a range of upstream receptors and that individual tumours need to be profiled to maximise the therapeutic benefit.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

Receptor tyrosine kinase (RTK) activation in human oesophageal adenocarcinoma specimens. RTK activation in human specimens using membrane arrays (A) was used to classify patients according to the number and the type of active RTKs (B).