Plasma Tie2 is a tumor vascular response biomarker for VEGF inhibitors in metastatic colorectal cancer

Abstract

Oncological use of anti-angiogenic VEGF inhibitors has been limited by the lack of informative biomarkers. Previously we reported circulating Tie2 as a vascular response biomarker for bevacizumab-treated ovarian cancer patients. Using advanced MRI and circulating biomarkers we have extended these findings in metastatic colorectal cancer (n = 70). Bevacizumab (10 mg/kg) was administered to elicit a biomarker response, followed by FOLFOX6-bevacizumab until disease progression. Bevacizumab induced a correlation between Tie2 and the tumor vascular imaging biomarker, Ktrans (R:-0.21 to 0.47) implying that Tie2 originated from the tumor vasculature. Tie2 trajectories were independently associated with pre-treatment tumor vascular characteristics, tumor response, progression free survival (HR for progression = 3.01, p = 0.00014; median PFS 248 vs. 348 days p = 0.0008) and the modeling of progressive disease (p < 0.0001), suggesting that Tie2 should be monitored clinically to optimize VEGF inhibitor use. A vascular response is defined as a 30% reduction in Tie2; vascular progression as a 40% increase in Tie2 above the nadir. Tie2 is the first, validated, tumor vascular response biomarker for VEGFi.

Conflict of interest statement

G.J.M.P. is a director of and shareholder of Bioxydyn, a company spun-out from the University of Manchester with an interest in imaging biomarkers. G.C.J. and C.D. have received grant funding for this project from Roche, which was administered through the University of Manchester. G.C.J. has attended advisory boards for Roche. J.P.B.O’.C. and C.D. declare no competing interests relevant to this manuscript. All remaining authors declare no competing interests.

Figures

Fig. 1

Progression-free and overall survival Kaplan–Meier estimator graphs. a Shows the proportion of the 70 eligible patients who are alive and free from disease-progression; the median PFS was 283 days (9 months; 95% CI 265–351 days or 8.8–11.7 months). One year and 2 year PFS statistics were 31 and 7% respectively. Note; no patient developed PD between days 98 and 182. b Shows the proportion of the 70 eligible patients who are alive; the median OS was 578 days (19.3 months; 95% CI 477–651 days or 15.9–21.4 months). One and 2-year survival statistics were 76 and 29%, respectively. The slightly lower OS length are not statistically shorter than other published series, but likely reflect the eligibility requirement for at least one metastatic lesion with a diameter between 3 and 10 cm that rendered the lesion amenable to serial MRI. Solid black lines; progression-free or overall survival, dotted gray lines; 95% CI, dashed black lines; median values, black crossed; patients whose data was censored. PFS progression-free survival, OS overall survival, CI confidence interval, MRI magnetic resonance imaging

Fig. 2

Correlation networks of circulating and imaging biomarkers. a Pearson’s correlation networks were constructed for circulating and imaging biomarkers measured at baseline. Clusters with median correlation coefficients above 0.5 are shown in thick black lines, while <0.5 but ≥ 0.35 are shown with dotted lines and correlations <0.35 are not displayed. b Compares PFS outcomes in two groups of patients defined by the ratio of pre-treatment VEGFR2:Ktrans, where the cut-off was selected as the 33rd percentile of the ratios. Thus 24 patients were included in the worse prognostic group (dashed black line) and 46 in the better prognostic group (solid black line). The median survival intervals of the two groups were 248 (range 58–423) and 348 (73–1750) days (p = 0.0008, Log rank test), respectively. Prognostic factors such as performance status and tumor volume were controlled for by using Cox proportional hazard analysis. The hazard ratio demonstrated that patients with high VEGFR2:Ktrans ratio have a significantly greater risk of progression than the other cohorts (HR 3.01, p = 0.00014, Wald test). This defines this cohort as the poor prognostic group, we compared the group’s WTV derived from DCE-MRI data with that of the other patient groups. c Pearson’s correlation networks were constructed for circulating and imaging biomarkers measured after two weeks of treatment with bevacizumab; tight correlations between circulating biomarkers were lost and some biomarkers were completely removed from the correlative relationship (arrows). In contrast bevacizumab induced a correlative relationship between Ktrans, Ang2, and Tie2 (gray line). Clusters with median correlation coefficients above 0.5 are shown in thick black lines, while <0.5 but ≥ 0.35 are shown with dotted lines and correlations <0.35 are not displayed. Ang1 and 2, angiopoietin 1 and 2; FGFb, fibroblast growth factor beta; HGF, hepatocyte growth factor, IL6 and 8, interleukins 6 and 8; KGF; keratinocyte growth factor, CK18; cytokeratin 18, PDGFbb; platelet-derived growth factor bb isoform, PlGF; placental growth factor, SDF1b; stromal-derived growth factor beta, VCAM1; vascular cell adhesion molecule 1; VEGFA, C, D, R1 and R2; vascular endothelial growth factor A, C and D and receptors 1 and 2, ADC; apparent diffusion coefficient, EF; ejection fraction, ETV; enhancing tumor volume, WTV; whole tumor volume, IAUC; initial area under the contrast agent concentration curve; Ktrans; endothelial transfer constant, Ve; extracellular extravascular space fractional volume, Vp; plasma fractional volume, DCE-MRI; dynamic contrast-enhanced magnetic resonance imaging

Fig. 3

Changes in Tie2 but not CK18 reflect tumor vascular control (Ktrans). Two cohorts (or clusters) of patients were identified through an unsupervised hierarchical clustering of Tie2 trajectories (a). In one cohort, there was an immediate but transient reduction (dotted line) in Tie2, while in the other cohort there was a later but more sustained reduction (solid line). The two cohorts behaved significantly differently (P = 1.8 × 10−6, Mann–Whitney U test). Similar cohorts of patients were identified with Ktrans values (b), which again behaved significantly differently (P = 0.003, Mann–Whitney U test). The pattern of behavior of CK18 was different (c), and the two cohorts were not significantly different (P = 0.18, Mann–Whitney U test). This indicates that Tie2 reflects the impact of bevacizumab on tumor vasculature. Error bars indicate the standard error. d Shows an example of Ktrans parameter maps, which reflect the tumor behavior of the two cohorts, shown in b. The largest difference in parameter maps can be seen in the two middle position maps, which correspond to the maximum difference between the two cluster-derived curves. CK18; cytokeratin 18, Ktrans; endothelial transfer constant

Fig. 4

Prognostic value of angiogenic biomarkers. The association between a less profound reduction in Tie2 and higher Ktrans:VEGFR2 ratio (Fig. 2b) identified a poor prognostic group. The WTV derived from DCI-MRI of this poor prognostic group (dotted lines) was compared with better prognostic outcome group (solid lines). a Shows the differential impact of single agent bevacizumab (10 mg/kg) on WTV between the two groups up to 15 days after treatment, with effects emerging by day 4 (P = 0.029, Mann–Whitney U test). b Volumetric analysis showed persistence of this difference throughout the course of combined cytotoxic chemotherapy and bevacizumab, with changes in WTV plotted using the same groups against the percentage time that elapsed between randomization and the date of progression/censoring (%PFS; P = 0.012, Mann–Whitney U test). Error bars indicate the standard error. WTV; whole tumor volume, DCE-MRI; dynamic contrast-enhanced magnetic resonance imaging, VEGFR2; vascular endothelial growth factor receptor 2, PFS; progression-free survival

Fig. 5

Predicting disease progression using plasma CK18 and Tie2 levels. The figure shows the potential of Tie2, cytokeratin 18 (CK18), and both biomarkers modeled together to predict disease progression. The prediction made by CK18 alone is shown in light gray, by Tie2 alone is shown in mid-gray and using the combination of CK18 and Tie2 in black. a Shows predictions based on actual data where the combined biomarkers predicted progression in 66% of patients, whereas single biomarker predictions of progressive disease were achieved in 54 and 41%. Interestingly the prediction derived from CK18 occurred later than that from Tie2 with a lead-time of 50 days (14% PFS) as the increase in the Tie2 biomarker predates that of CK18. b Shows the same calculation but using simulations to fill any missing data. Again a significant improvement on the number of predictions was observed when Tie2 and CK18 were used jointly for prediction (p < 0.0001, Mann–Whitney U compared with either single antigen), revealing clear evidence of additivity between the two biomarkers with respect to predicting progressive disease. Error bars indicate the standard error. CK18; cytokeratin 18, PFS; progression-free survival

Fig. 6

The tissue multi-compartment model of cancer treatment. a The biomarker data show that Tie2 is a tumor response biomarker for VEGFi and that the predictive value of vascular biomarker data adds to epithelial biomarker data to improve modeling of progressive disease. Together with our previous data in ovarian cancer, the implication of these findings is that the vascular and epithelial compartments are distinct, valid and useful concepts in the treatment of each recurrence of solid tumors. Examples of individual patient data are shown in b–e; the x-axis represents the number of days that the patient received treatment until they developed RECIST-defined progression (i.e., total tumor burden), Tie2 data are shown as solid black lines and CK18 in dotted black lines. The units for both biomarkers are pg/ml. In b Patient 32 has undergone an epithelial response, vascular response, epithelial progression, and vascular progression. In c, the patient had an epithelial response but not a vascular response thus the patient could have stopped bevacizumab after ~8 weeks of treatment. The patient subsequently had epithelial and vascular progression. In d the patient had an epithelial and vascular biomarker response and then developed vascular progression but not epithelial biomarker progression. In e the patient attained an epithelial and vascular response with subsequent epithelial progression but not vascular progression, suggesting that the vasculature was still controlled at the point of RECIST progression. CK18; cytokeratin 18, VEGFi; vascular endothelial growth factor inhibitor, RECIST; response evaluation criteria in solid tumors