Biomarkers Predicting Outcome in Patients with Advanced Renal Cell Carcinoma: Results from Sorafenib Phase III Treatment Approaches in Renal Cancer Global Evaluation Trial

The Von Hippel-Lindau (VHL) gene, commonly mutated in renal cell carcinoma (RCC), is associated with clear cell carcinoma, the most common RCC histology (1–4). Under normoxic conditions, wild-type VHL protein binds to hypoxia-inducible factor, a transcription factor that promotes cell growth and survival under hypoxic conditions, and targets hypoxia-inducible factor for proteosomal degradation (4–6). The loss of VHL protein leads to hypoxia-inducible factor transcriptional activity, including production of carbonic anhydrase IX (CAIX), a tissue hypoxia marker, and growth and angiogenesis factors such as vascular endothelial growth factor (VEGF), transforming growth factor α (TGF-α), erythropoietin, and platelet-derived growth factor (PDGF; ref. 2). The binding of these growth factors to their cognate tyrosine kinase receptors drives cell proliferation and angiogenesis.

Targeted therapies that prevent tyrosine kinase signaling induced by these growth factors have proven efficacious in the treatment of RCC (7–9). Sorafenib is an orally active multikinase inhibitor that targets the intracellular serine/threonine kinase Raf-1 (10) and the receptor tyrosine kinases VEGFR, PDGFR receptor β (PDGFR-β), c-KIT, RET, and FLT-3 (11, 12). Sorafenib is able to directly target both tumor cell proliferation and angiogenesis through the role of these factors in aspects of tumor growth and blood vessel formation (10). Components of these proliferative and proangiogenic signaling pathways, some of which are detectable in plasma, may provide sensitive indicators of tumor activity and response to treatment in RCC.

Multiple preclinical studies have shown the importance of VEGF in driving angiogenic processes through its cognate receptors VEGFR-1, VEGFR-2, and VEGFR-3 (9). VEGFR-2 signaling in particular has been shown to be a major driver of tumor angiogenesis (9, 13). Ras is an intracellular signaling molecule that plays an essential role in transmitting signals downstream of receptor tyrosine kinases (14). Mutational activation of Ras proteins promotes tumorigenesis by disturbing cellular processes, including cell cycle progression, cell proliferation, survival, and migration (15). In RCC, unregulated activation of Raf kinases and receptor tyrosine kinases through mechanisms such as point mutations or overexpression can induce Ras p21 activation and subsequent upregulation of the downstream Raf/MEK/ERK pathway (2, 16), making Ras p21 potentially useful as a prognostic marker in patients.

Other signaling pathways involved in metastasis and tumor microenvironment interactions, including CAIX and tissue inhibitors of metalloproteinases (TIMP), may also have value as prognostic indicators. CAIX regulates intracellular pH in response to hypoxia (17, 18) and has been investigated as a prognostic factor for survival in patients with metastatic RCC (19). CAIX is highly expressed in advanced RCC but is not expressed in normal kidney, suggesting that it may be a useful marker of tumor progression (19). An imbalance of TIMPs and matrix metalloproteinase enzymes leading to an excess of proteolytic degradation of the extracellular matrix and basement membrane has also been linked to tumor invasion (20). Recent evidence suggests that TIMPs play a complex multifunctional role that includes growth-promoting abilities, depending on the tissue microenvironment (21).

In the phase III Treatment Approaches in Renal Cancer Global Evaluation Trial (TARGET), sorafenib doubled progression-free survival (PFS), showed a significant benefit in overall survival (OS) compared with placebo, and had a manageable safety profile (7, 22) in patients with advanced RCC. An additional objective of TARGET was to explore candidate biomarkers for prognosis and sorafenib benefit in patients with advanced RCC. Plasma levels of VEGF, soluble (s)VEGFR-2, CAIX, TIMP-1, and Ras p21 were assayed at baseline and after 3 and 12 weeks of treatment. Tumor VHL mutational status was determined in archival tumor samples. We report the results of these exploratory analyses here.

The TARGET design has been previously described (7). Briefly, eligible patients with advanced RCC were randomized (n = 903) to receive either 400 mg sorafenib twice a day (n = 451) or placebo (n = 452). The majority of patients enrolled (>99%) had RCC with clear-cell histology. The primary end point was OS. Secondary end points included PFS, response rate, and patient-reported outcomes.

Plasma samples were collected at baseline and after 3 and 12 weeks of treatment. Blood samples (10 mL) were drawn at each time point using venipuncture or through a porta-catheter into a Vacutainer (BD) containing potassium EDTA and gently inverted. Blood samples were centrifuged in a refrigerated (4°C) centrifuge for 10 minutes to separate the plasma within 10 to 15 minutes after collection. Plasma samples were frozen upright at ≤−70°C within 20 minutes of centrifugation and kept frozen until ready for shipment to the sponsor. Archival formalin-fixed, paraffin-embedded tumor biopsy samples were collected for VHL mutational analysis.

Commercially available enzyme-linked immunosorbent assay (ELISA) kits for VEGF (detecting VEGF-A; R&D Systems; ref. 23), sVEGFR-2 (R&D Systems; ref. 24), CAIX (detecting the extracellular domain; Siemens Medical Solutions Diagnostics; ref. 25), TIMP-1 (Siemens Medical Solutions Diagnostics; ref. 26), and Ras p21 (detecting H-Ras, K-Ras, and N-Ras; Siemens Medical Solutions Diagnostics; ref. 27) were used per the manufacturers' specifications to measure protein levels in plasma. Laboratory personnel performing the assays and reporting the data were blinded to the trial outcome.

DNA was isolated from formalin-fixed, paraffin-embedded tumor samples using Qiagen's DNeasy Blood & Tissue Kit (28). The three coding exons of the VHL gene were amplified through PCR using the following primers: exon 1, forward—GCGGCGTCCGGCCCGGGTGGTCTGGAT, reverse—TGGGTCGGGCCTAAGCGCCGGGCCCGT; exon 2, forward—GTTTCACCACGTTAGCCAGGA, reverse—GATTGGATACCGTGCCTGACA; exon 3, forward—GTTGGCAAAGCCTCTTGTTCG, reverse—GCCCCTAAACATCACAATGCC. PCR conditions were as follows: 94°C for 10 minutes, followed by 35 to 40 cycles of 94°C for 30 seconds, 70°C for 1 minute, and 72°C for 5 minutes, and lastly 4°C indefinitely. PCR products were purified using Qiagen QIAquick PCR Purification Kit, and sequencing was done using PCR primers on the Applied Biosystems 3700 DNA Analyzer. The resulting sequences were aligned with the wild-type (WT) VHL gene sequence (Genbank accession no. NM_000551) to identify mutations.

The outcomes and clinical data used for the present correlative analyses were taken directly from the TARGET database. OS was defined as the time elapsed from randomization to death from any cause, and PFS was defined as the time from randomization to disease progression based on radiologic or clinical assessment or death (whichever occurred earlier). A preplanned interim PFS analysis showed that treatment with sorafenib doubled PFS compared with placebo, allowing for crossover of placebo patients to sorafenib (7, 22). The investigator-assessed PFS data used for biomarker analyses were collected precrossover (data cutoff May 2005), and the OS data used for biomarker analyses were 5 months postcrossover (data cutoff November 2005). The 5-month postcrossover OS data set was chosen for the biomarker correlative analyses to minimize the number of censored OS data points in the data set, increasing the power of the statistical analyses.

Patients with biomarker data available for at least one time point were included in statistical analyses. Patients missing a biomarker value for a given time point were excluded from analyses of that biomarker at that time point. Evaluation of the prognostic value of the biomarker was done initially using univariable Cox and Kaplan-Meier analyses, followed by multivariable Cox proportional hazard models. The multivariable models included baseline biomarker value(s) and two clinical variables shown to be prognostic in the overall study population [Eastern Cooperative Oncology Group performance status (ECOG PS) and Memorial Sloan-Kettering Cancer Center (MSKCC) score]. MSKCC score for previously treated patients with RCC is based on three risk factors: ECOG PS ≥2 (or Karnofsky performance status <80%), low hemoglobin, and high corrected calcium. Low MSKCC score is defined as having none of these risk factors; medium as having one; and high as having two to three (29). The effect of sorafenib treatment on biomarker levels was determined with one-way ANOVA tests. The relationship between baseline biomarker levels and sorafenib benefit and between change in biomarker levels and patient outcome was examined using proportional hazards models with an interaction term. Because of the exploratory nature of the biomarker analyses, P values were not corrected for multiplicity of testing. P values reported should be interpreted as a strength of evidence measure and not as a declaration of statistical significance.

Valid baseline plasma samples were received, and the resulting evaluable ELISA data were available from a total of 712 patients for VEGF and 713 for sVEGFR-2. CAIX, TIMP-1, and Ras p21 were evaluated as an exploratory analysis of a subset of these patients, with ELISA data generated for 128 patients for CAIX, 123 for TIMP-1, and 125 for Ras p21. Valid tumor samples for VHL mutational status analysis were received, and at least one of the three VHL coding exons was sequenced successfully for 134 patients.Baseline characteristics in the biomarker subpopulations were similar to those in the overall TARGET population (Table 1). Baseline characteristics of the VEGF subpopulation were described previously by Escudier et al. (22) and are also demographically similar to the overall study population. The majority of patients were male (∼72–74%), under 65 years of age (∼70%), and had an ECOG PS of ≤1 (99%); 51% to 55% of patients also had a low MSKCC score (Table 1; ref. 22). In addition, 31.3% of available patients (n = 48; includes only patients from whom all three exons were sequenced successfully) were found to harbor mutations in the VHL gene.

Analysis of baseline plasma biomarker levels as a function of demographic variables (Table 1; ref. 22) indicated that patients with intermediate MSKCC scores had significantly higher baseline VEGF (P < 0.0001), CAIX (P = 0.027), and TIMP-1 (P = 0.0001) than those with low MSKCC scores. Baseline levels of sVEGFR-2, Ras p21, and VHL mutational status did not differ significantly between patients with low and intermediate MSKCC scores. Patients with poor ECOG PS scores had significantly higher baseline VEGF levels (P < 0.0001; ref. 22) and a higher incidence of VHL gene mutations (P = 0.008; Table 2). Baseline levels of sVEGFR-2, CAIX, TIMP-1, and Ras p21 did not significantly correlate with ECOG PS score.

Univariable analysis of the placebo cohort identified baseline plasma levels of VEGF (n = 348; P = 0.0024), CAIX (n = 66; P = 0.034), TIMP-1 (n = 63; P = 0.001), and Ras p21 (n = 65; P = 0.016) as prognostic factors for OS in RCC (Fig. 1). Patients with high (>median) baseline VEGF (>130.8 pg/mL), CAIX (>296.0 pg/mL), TIMP-1 (>332.7 pg/mL), or Ras p21 (>2775.9 pg/mL) had shorter survival times than patients with low (<median) baseline levels of these factors. Similar univariable analyses of PFS revealed that elevated levels of Ras p21 (hazard ratio, 1.84; median PFS, low and high Ras patients, 4.7 and 2.8 months, respectively; P = 0.040) and VEGF (P < 0.01; ref. 22) also correlated with shorter PFS in placebo-treated patients. sVEGFR-2 (n = 350) did not show a prognostic relationship with either OS (Fig. 1) or PFS (data not shown).

The prognostic value of VEGF, VEGFR-2, CAIX, TIMP-1, and Ras p21 was further explored using multivariable analyses. Multivariable analysis of all available patients with baseline VEGF data has been published previously (22) and revealed that VEGF retained independent prognostic value in placebo patients for both OS (P = 0.042) and PFS (P = 0.023). A further multivariable analysis was done in the subset of patients (n = 59; placebo patients only) who had data available for all five plasma biomarkers assayed (Table 3). Inclusion of these biomarkers plus ECOG PS and MSKCC score in the same model revealed that only TIMP-1 remained independently prognostic for survival (P = 0.002).

To investigate the potential relationship between baseline biomarkers and sorafenib benefit, Cox analyses including patients from both the sorafenib- and placebo-treated cohorts were done to examine the biomarker treatment interaction. The effect of baseline VEGF level on sorafenib benefit has been described previously (22); results suggest that patients with elevated VEGF (>75th percentile) may experience greater benefit from sorafenib treatment (in terms of PFS) than those with low VEGF. Similar analyses of sVEGFR-2, CAIX, TIMP-1, Ras p21, and VHL mutational status revealed no relationship between these biomarkers and sorafenib benefit (Table 4).

The changes in biomarker levels from baseline to weeks 3 and 12 of treatment were analyzed to identify biomarkers that are potentially modulated by sorafenib treatment or by the course of disease (Table 5). Baseline VEGF (n = 712) and sVEGFR-2 (n = 713) levels did not differ significantly between the sorafenib and placebo cohorts. Mean VEGF levels increased significantly in sorafenib-treated patients at weeks 3 and 12 (P < 0.0001 for both), but did not change significantly in those treated with placebo. The VEGF response during treatment also differed significantly between the sorafenib and placebo cohorts at both time points assayed (week 3, P = 0.005; week 12, P = 0.007). Plasma sVEGFR-2 levels decreased in sorafenib-treated patients (P < 0.0001 at weeks 3 and 12), and increased marginally at week 3 in patients treated with placebo (P = 0.010). The change in sVEGFR-2 differed significantly between the sorafenib- and placebo-treated cohorts at both weeks 3 and 12 (P < 0.001 for both time points).

Furthermore, whereas baseline levels of CAIX (n = 128) did not differ significantly between the sorafenib and placebo cohorts (P = 0.129), baseline TIMP-1 (n = 123) and Ras p21 (n = 125) differed significantly between treatment arms (P = 0.005 and P = 0.001, respectively), with higher levels of both factors occurring in the sorafenib cohort (Table 5). TIMP-1 decreased significantly during treatment in the sorafenib cohort at weeks 3 (P = 0.002) and 12 (P = 0.006), whereas it increased in the placebo cohort at week 12 (P < 0.001). These changes in TIMP-1 differed significantly between the sorafenib and placebo cohorts at weeks 3 (P = 0.004) and 12 (P < 0.001). Although Ras p21 level did not change significantly in the sorafenib cohort, it increased significantly in the placebo cohort at week 12 (P = 0.048). The Ras response at week 12 differed significantly between the placebo group and the sorafenib group (P = 0.022). Plasma CAIX did not change significantly in either treatment cohort at week 3 or 12.

No correlations were observed between the magnitude of change in biomarker levels from baseline to week 3 or 12 and outcome (OS or PFS) in sorafenib-treated patients (data not shown).

The phase III placebo-controlled TARGET provided an ideal setting in which to examine potential biomarkers of disease prognosis and drug benefit. In the present exploratory analysis, plasma samples from RCC patients were analyzed to examine the potential utility of these biomarkers as prognostic indicators of disease outcome and as predictors of sorafenib benefit.

In a recent publication, we reported that baseline plasma VEGF correlated with MSKCC score and ECOG PS, and was prognostic when analyzed in 348 patients from the placebo group enrolled in TARGET (22). A multivariable analysis that examined VEGF, ECOG PS, and MSKCC score in these placebo patients showed that all three factors were independently prognostic for OS (22). Lastly, exploratory analyses of the relationship between VEGF and sorafenib benefit in 712 TARGET patients suggested that those with levels above the 75th percentile at baseline may benefit more from sorafenib (in terms of PFS) than those with low levels, although sorafenib benefit was apparent in both groups (22).

The current analyses of baseline biomarker levels as a function of demographic variables were conducted to determine if levels of other biomarkers correlated with established prognostic indicators (i.e., MSKCC and ECOG PS scores). Indeed, elevated baseline plasma CAIX, TIMP-1, and the presence of tumor VHL mutations were all associated with poor clinical prognostic scores. Baseline biomarkers were further examined in univariable analyses in patients randomized to placebo to determine whether they also offer prognostic value for the natural course of RCC. These analyses indicated that high baseline plasma CAIX, TIMP-1, and Ras p21 were prognostic factors for reduced OS (as well as PFS for Ras p21) in patients with advanced RCC. Analysis of VEGF, CAIX, TIMP-1, and Ras p21 together in 59 patients who had received placebo in TARGET showed that of these prognostic biomarkers, TIMP-1 remained the sole independently prognostic factor for OS in a multivariable model that also included MSKCC score and ECOG PS.

The proangiogenic factor VEGF is known to have increased tumor activity in RCC (1). High serum VEGF-165 (the predominant isoform of VEGF-A in circulation) levels have also been associated with tumor stage, tumor grade, and poor prognosis in RCC (30). Elevated levels of tumor and/or plasma TIMP-1 are associated with poor survival prognosis in a number of tumor types [colorectal (31, 32), gastric (33), breast (34–36), ovarian (37), and lung (38) cancers]. In a number of these studies, plasma TIMP-1 was an independent prognostic factor upon multivariable analysis. Elevated tumor levels of TIMP-1 have also been linked to tumor grade and prognosis in RCC (39), although plasma levels of TIMP-1 have not been similarly reported in RCC. Thus, the current finding that plasma TIMP-1 levels are independently prognostic for survival in previously treated patients with RCC is novel and supports the tumor TIMP-1 findings. Plasma TIMP-1 warrants further study as a prognostic factor for RCC and may provide value when included in prognostic scoring systems for RCC.

Low levels of tumor CAIX have been linked to poor patient prognosis in RCC (19, 40), although other studies have not shown this relationship (41). The present study examined the plasma-detectable shed extracellular domain of CAIX, and, in contrast to the published results on tumor CAIX, elevated CAIX extracellular domain correlated with poor survival. Although the relationship between tumor and plasma levels of CAIX is not understood, the current observation suggests that the level of shed CAIX extracellular domain may reflect relevant tumor biology or disease state.

Elevated serum Ras p21 correlated with poor survival in one study of hematologic malignancies (42). The present finding that high plasma levels of Ras p21 protein correlated with poor survival suggests that circulating Ras reflects relevant tumor biology or disease state.

Plasma levels of VEGF, sVEGFR-2, CAIX, TIMP-1, and Ras p21 were determined at baseline and after 3 and 12 weeks of treatment. Baseline levels of TIMP-1 and Ras p21 were significantly higher in the sorafenib group compared with the placebo group, indicating a potential imbalance between the treatment populations. Nevertheless, plasma levels of VEGF increased significantly in response to sorafenib but remain unchanged in the placebo group, whereas sVEGFR-2 decreased in response to sorafenib. TIMP-1 plasma levels also decreased in the sorafenib cohort but increased in the placebo group (only at 12 weeks). Increases in plasma Ras protein levels were also observed in placebo-treated patients (only at week 12), whereas no significant changes were observed in patients treated with sorafenib. The magnitude of the changes did not correlate with outcome in sorafenib-treated patients for any of the proteins assayed.

The reciprocal changes observed in VEGF and sVEGFR-2 levels following sorafenib treatment are similar to those observed with other antiangiogenic agents (43–47), and collectively are suggestive of a class effect. The lack of correlation observed in the present study between the change in VEGF or sVEGFR-2 and outcome may be due to the systemic nature of the changes, as opposed to a tumor-specific phenomenon. Increases in TIMP-1 and Ras p21 observed in placebo patients (but not in sorafenib patients) during the course of the trial may reflect the advancement of disease.

The results of the present analyses are hypothesis generating, and interpretation should be tempered by the retrospective nature of the biomarker study and the small sample size available for some of the analyses. Additionally, the results should be interpreted with caution due to the large number of statistical analyses done. Based on this concern, only one bifurcation cutoff point to separate “high” from “low” biomarker values was analyzed for most plasma biomarkers (the median). The one exception to this is the analysis of VEGF versus sorafenib benefit, for which the median VEGF level was used in the initial analysis, but additional bifurcation cutoffs were also examined. To address the increased risk of generating false-positive signals due to multiple testing, for the analyses of the additional bifurcation points (the 25th and 75th percentiles), the VEGF population was split into two random halves for hypothesis generation and hypothesis confirmation.

Additional study would be required to confirm the clinical relevance and utility of the biomarkers assayed here. In analyses designed to detect predictors of sorafenib benefit in RCC, only VEGF showed any predictive ability (22). Based on the strength of the association between TIMP-1 and prognosis, TIMP-1 should be investigated prospectively as a potential biomarker, such as the MSKCC prognostic factor analysis, in both previously treated and untreated RCC patients.

C. Pena, M. Shan, C. Lathia, employment, Bayer HealthCare Pharmaceuticals. B. Escudier, advisory board, Bayer, Pfizer, Roche, Novartis, GSK. R. Bukowski, honoraria from speaker's bureau and advisory board, Pfizer, Novartis, Genentech, Bayer.

We wish to acknowledge the contributions of Ogilvy Healthworld Medical Education, New York, NY, for medical writing/editorial support of this article and Walter Carney, Sheryl Brown-Shimer, and Peter Hamer from the Oncogene Science Group of Siemens Healthcare Diagnostics for technical support. We thank the patients who participated in TARGET and their families, and the TARGET Investigators and their staff for conducting the trial.

Grant Support: Funding for TARGET was provided by Bayer HealthCare Pharmaceuticals and Onyx Pharmaceuticals.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.