|Home | About | Journals | Submit | Contact Us | Français|
No validated biological markers (or biomarkers) currently exist for appropriately selecting patients with cancer for antiangiogenic therapy. Nor are there biomarkers identifying escape pathways that should be targeted after tumors develop resistance to a given antiangiogenic agent. A number of potential systemic, circulating, tissue and imaging biomarkers have emerged from recently completed phase I–III studies. Some of these are measured at baseline (for example VEGF polymorphisms), others are measured during treatment (such as hypertension, MRI-measured Ktrans, circulating angiogenic molecules or collagen IV), and all are mechanistically based. Some of these biomarkers might be pharmacodynamic (for example, increase in circulating VEGF, placental growth factor) while others have potential for predicting clinical benefit or identifying the escape pathways (for example, stromal-cell-derived factor 1α, interleukin-6). Most biomarkers are disease and/or agent specific and all of them need to be validated prospectively. We discuss the current challenges in establishing biomarkers of antiangiogenic therapy, define systemic, circulating, tissue and imaging biomarkers and their advantages and disadvantages, and comment on the future opportunities for validating biomarkers of antiangiogenic therapy.
Tumors acquire blood vessels by co-option of neighboring vessels, from sprouting or intussusceptive microvascular growth and by vasculogenesis from endothelial precursor cells.1 In most solid tumors the newly formed vessels are plagued by structural and functional abnormalities owing to the sustained and excessive exposure to angiogenic factors produced by the growing tumor.2 Despite being abnormal, these new vessels allow tumor expansion at early stages of carcinogenesis and progression from in situ lesions to locally invasive, and eventually to metastatic tumors. The hypothesis that tumor progression can be arrested by antiangiogenesis3 has been confirmed experimentally by a large body of evidence over the past three decades. Enhanced survival of patients has yet to be achieved in phase III clinical trials by antiangiogenic agents that only target VEGF. Nevertheless, the addition of bevacizumab (a VEGF-specific blocking antibody) to standard chemotherapies or to interferon therapy (in metastatic renal cell carcinoma [mRCC]), as well as the use of anti-VEGF receptor tyrosine kinase inhibitors (TKIs) with wide spectra of activity, has proven efficacious in multiple advanced cancers such as metastatic colorectal cancer (mCRC), metastatic non-small-cell lung cancer (mNSCLC), metastatic breast cancer, mRCC, hepatocellular carcinoma (HCC) and gastrointestinal stromal tumors (GISTs).4–12 Moreover, bevacizumab has been recently approved for recurrent glioblastoma based on phase II trial data.
These agents have changed the practice of oncology but stimulated important questions: how do these therapies work in patients? Is their mechanism of action in patients the same as originally envisioned for antiangiogenic agents? Is the mechanism the same as demonstrated in animal models? Could the overall survival benefit be increased beyond a few months? Could we successfully use these agents in the adjuvant setting after surgical resection? Why do some patients develop severe toxicities from antiangiogenic therapy? Why is the benefit from antiangiogenic therapies seen only in some patients? How do we preselect these patients, or the most appropriate therapy? Why do tumors stop responding to antiangiogenic therapy? What new pathways should be targeted to optimize the response and prolong the duration of response and survival without increasing toxic effects? How do we tailor these new therapies to individual patients? How do we schedule them with contemporary and future therapeutics? The answers to these fundamental questions are not fully known for the approved antiangiogenic agents, and will be critical in choosing the appropriate agent(s), and to determine their optimum dose and schedule.13–17 We propose that validated pharmacodynamic, prognostic, predictive and surrogate biomarker studies can address these questions (Box 1), and we outline challenges in identifying and validating biomarkers for response, toxicity and resistance to antiangiogenic therapy, and finally, discuss emerging systemic, circulating, tissue and imaging biomarkers.
An array of antiangiogenic biomarkers have been studied (Box 2), including systemic measurements (for example, changes in systemic blood pressure), genotypic analyses (for example, VEGF or interleukin [IL]-8 polymorphisms), circulating markers (for example, plasma levels of VEGF), tissue markers (tumor microvessel density) and imaging parameters (for example, Ktrans, the volume transfer constants of gadolinium between blood plasma and the extravascular extracellular space measured by MRI; Table 1). Although promising candidates have been identified, important challenges limit their translation into practice.
The first challenge is establishing adequate criteria of response. This issue is especially problematic when using agents that target stroma, such as antiangiogenic agents. Standard lesion size (defined, by RECIST [Response Evaluation Criteria In Solid Tumors] or WHO criteria) evaluations of response might not optimally assess these agents, particularly when used as monotherapy with agents such as sunitinib or sorafenib in RCC or HCC. Anti-VEGF therapy has primarily cytostatic effects, might prune and normalize the tumor vasculature, and can have substantial systemic effects such as modulation of circulating proangiogenic and proinflammatory cytokines and cells.18–23 These effects might not shrink but rather stabilize the tumor size.24
The success in identifying a predictive biomarker for a drug will require elucidation of its mechanism of action (Supplementary Table 1). For example, detection of over-expression or amplification of human epidermal growth factor receptor 2 (HER2/neu) in breast cancer cells and its use as a predictive biomarker is consistent with the mechanism of action of the anti-Her2 antibody trastuzumab or the EGFR-HER2 TKI lapatinib.25,26 Unfortunately, the mechanisms of action of currently approved antiangiogenic agents are not fully understood.27,28
The anti-VEGF antibody bevacizumab has yet to show improved overall survival as monotherapy in a phase III trial. The mechanisms of action of multitargeted TKIs are even less well understood—they probably target both stromal and cancer cells. Indeed, a number of potential mechanisms by which these different agents work have been hypothesized. Of these mechanisms, ‘vascular normalization’ has the most robust clinical evidence,29 and this mechanism alone might contribute to improved survival in some cancers such as glioblastoma and/or potentially as a sensitizer to cytotoxic therapies in others (for example, rectal carcinoma)18,20,22,30 (Box 3 and Table 2). Thus, the determinants of vascular normalization could serve as candidate biomarkers. Conversely, a candidate biomarker that is associated with improved benefit could provide insight into the mechanism of action or resistance of a drug.
The second challenge results from the heterogeneous and dynamic nature of cancer. Ideally, treatment outcome would be predicted from a single measurement in the tumor biopsy sample or from the circulation before treatment initiation. Indeed, VEGF polymorphisms in tumor biopsy sample and baseline VEGF plasma levels are candidate biomarkers for bevacizumab combined with chemotherapy for patients with metastatic breast cancer.38,39 By contrast, baseline VEGF is not associated with survival outcomes for mCRC or mNSCLC.40,41 The dynamics of cancer must be recognized: not only might the biology of the primary tumor be different from its metastases, but it might also change with tumor progression and treatment. Thus, a biopsy sample before first treatment might not reflect the biology before subsequent treatment. Finally, there is regional heterogeneity: one part of a tumor may not have the same vascularity or angiogenesis as another part. Thus there is a need for spatially resolved ‘dynamic biomarkers’.
The third challenge is the inability to perform repeated biopsies (that is, before and after antiangiogenic therapy) to assess ‘dynamic biomarkers’. This challenge can be partially addressed by using novel imaging techniques, which can also provide spatial information. Such an approach has shown promise in identifying potential biomarkers for treatment outcome in patients with glioblastoma. In some cases, different types of biomarkers (for example, imaging and circulating) might need to be combined, yielding a ‘composite biomarker’, to make robust predictions.
The fourth challenge in identify biomarkers resides in the inherent design of clinical trials. Human studies are expensive and require expertise in a wide range of areas, and almost all exploratory biomarkers to date have emerged from single-arm trials.42 It is, therefore, difficult to ascertain whether the biomarker is prognostic or predictive. This is problematic for anti-VEGF therapies where VEGF levels in the circulation or tumor biopsy samples have been shown to be prognostic in a number of cancer types.43,44 Even markers identified from randomized trials have emerged from secondary analyses, and require independent validation.
A fifth challenge is the unpredictability of response or toxicity, and resistance by activation of tumor VEGF-independent angiogenic pathways. Thus, biological (mechanism-driven) marker (biomarker) discovery has become a priority for these costly therapies that can be associated with rare but serious adverse effects.
A sixth challenge is to optimize and standardize various biomarkers assays. For example, different approaches are being used to measure vascular imaging parameters or circulating proteins and cells. Each approach gives a different result, which makes it difficult to compare trial results. This is further confounded by the inability of widely used imaging techniques to distinguish antivascular effects from antitumor effects of antiangiogenic agents.45 Thus, overall response rate and/or progression-free survival outcomes on the basis of contrast-enhanced imaging might not reflect a true antitumor response.
When a biomarker has been identified and validated the question for their clinical implementation will arise. Will they be generic for any anti-VEGF drug, any tumor type or stage or combination regimen? Despite these limitations, a number of candidate biomarkers are emerging for antiangiogenic therapy of cancer. Some of these findings are provocative and raise new questions about the efficacy, safety and cost–benefit ratios of these therapies. We discuss the current understanding and the future directions in establishing candidate ‘predictive biomarkers’, ‘toxicity biomarkers’ and ‘VEGF-resistance biomarkers’, and the steps necessary for their future validation.
Pharmacodynamic biomarkers should reflect modulation of a defined biological target. Whether this biological change translates into a clinical benefit in the patient can only be ascertained if the biomarker correlates with the treatment outcome (that is, it is a predictive biomarker). Although several pharmacodynamic biomarkers have emerged, little progress has been made in the validation of prognostic and predictive biomarkers (Tables 3 and and44).
The most widely used pharmacodynamic biomarker is a systemic effect of most antiangiogenic agents—that is, the increase in blood pressure (hypertension). Hypertension has been observed in patients with cancer treated with anti-VEGF antibodies or TKIs and is clinically manageable in most cases with medication. Two studies have proposed the degree of hypertension as a predictive biomarker of survival in patients with cancer after bevacizumab or axitinib treatment.34,38 This finding should be validated in other large studies.
Several parameters measured in the tumor itself or in the blood circulation of patients with cancer might hold pharmacodynamic or predictive biomarker value. Naturally, the most extensively explored biomarker has been VEGF. Associations between outcomes of antiangiogenic therapy with VEGF levels in the circulation has been reported in some phase II studies. In three randomized trials of vandetanib, baseline plasma VEGF levels were correlated directly with progression-free survival in patients with advanced NSCLC, but only in patients with low pre treatment plasma VEGF levels. Likewise, baseline plasma VEGF levels were correlated with time to progression in patients with metastatic breast cancer in a study of bevacizumab with chemotherapy, and also with progression-free survival in patients with HCC treated with sunitinib.23,39,46 By contrast, a randomized study of sorafenib with or without interferon in patients with mRCC reported an inverse correlation between baseline plasma VEGF levels and progression-free survival (Table 3).47 The reason(s) for this contrasting association are unclear.
Many studies have shown a lack of correlation between VEGF levels at baseline and outcome of antiangiogenic therapy. In the pivotal phase III trial of bevacizumab with chemotherapy in patients with mCRC, VEGF expression in primary tumor tissue has not been predictive of outcome (Table 4).40 Similarly, in a phase II/III trial examining bevacizumab and chemotherapy treatment in patients with mNSCLC, a high baseline circulating plasma VEGF level did not predict progression-free survival or overall survival, despite a correlation with improved overall response rate.41 Phase II studies of sunitinib in RCC, bevacizumab combined with chemoradiation in rectal cancer and cediranib in patients with recurrent glioblastoma showed no correlation between VEGF and outcome of therapy.18,22,48
The inconsistencies in these results emphasize the necessity of evaluating the predictive biomarkers in a dynamic manner, that is, before and soon after commencement of antiangiogenic treatment. Intriguingly, the circulating levels of VEGF seem to be significantly elevated after most antiangiogenic therapies targeting this pathway.32 Circulating plasma VEGF has also been shown to increase after therapy with anti-VEGFR TKIs (Table 5).18,21–23,32,39,48–53 In addition, plasma VEGF levels decrease after antiangiogenic treatment is discontinued, which supports its potential pharmacodynamic biomarker value.18,21,23,49,52
The increased level of VEGF after antiangiogenic treatment raises many questions. In patients treated with bevacizumab, is the VEGF detected by various technologies freely circulating or bound to the drug? Does it emanate from the tumor or host cells? Why does it increase? What does this excess VEGF do? We and others have tried to address these questions. Given that bevacizumab is administered at doses high enough to bind to circulating VEGF, there is a debate whether the circulating VEGF is free or bound to the antibody. Our studies using standard enzyme-linked immunosorbent assays and multiplex arrays suggest that the measured protein is free.22 Moreover, the phenomenon occurs in the context of VEGFR blockade with TKIs. Preclinical data indicate that this increase in VEGF might be induced by hypoxia in tumors as a result of excessive vessel pruning.54 However, in mice lacking tumors, circulating VEGF has also been shown to increase after TKI blockade of VEGF signaling.19 By comparing the expression profiles of cancer and stromal cells in rectal carcinoma biopsy samples taken before and after a single cycle of bevacizumab monotherapy, we have observed that the increased circulating VEGF most likely emanates from stromal cells (L. Xu, unpublished data). Thus, this increase in VEGF is likely to be a host-response to neutralizing such a critical growth factor.
Perhaps a more intriguing question is what does this VEGF do? One study has shown that platelets take up bevacizumab, which neutralizes the VEGF stored within platelet granules.55 Another preclinical study has suggested that VEGF (and other cytokines) released after sunitinib treatment might facilitate the growth of metastases in mice.56 If confirmed in patients, this finding would indicate that treatment discontinuation with these agents should be avoided in patients with tumor progression. However, a recent phase III trial of adjuvant bevacizumab in early-stage CRC failed to meet its efficacy endpoint.57
While the use of circulating VEGF as a biomarker remains unclear, evaluation of the VEGF genotype has emerged as a predictive biomarker candidate from the phase III study of bevacizumab with chemotherapy versus chemotherapy alone in patients with metastatic breast cancer (ECOG 2100 trial). In that study, the VEGF-2578AA genotype was associated with a superior overall survival in the combination arm compared with the alternate genotypes combined,38 and should be tested in future trials of bevacizumab and other anti-VEGF agents. Unfortunately, baseline plasma levels of VEGF were not available in this study, so this polymorphism could not be compared with the circulating levels of VEGF in these patients.
Circulating levels of PIGF (placental growth factor)—another VEGF family member—also increase in response to anti-VEGF treatment. Thus, plasma PIGF dynamics is now being considered as a potential pharmacodynamic biomarker (Table 5).22,23,37,48,50,52,58 In addition, targeting PIGF is being considered as a novel approach to prevent tumor escape from anti-VEGF therapy.54 It is worth noting that in a study of bevacizumab in rectal carcinoma, increased circulating PIGF levels, as well as VEGF levels, emanate from host cells (L. Xu, unpublished data). Moreover, the extent of increase in PIGF levels in plasma was associated with a better outcome in patients with rectal cancer treated with bevacizumab and chemo-radiotherapy and cediranib monotherapy in patients with recurrent glioblastoma.22,37 The limitation of these single-arm phase II studies is that one cannot distinguish between predictive and prognostic biomarkers. The role of PIGF needs to be further explored in large studies, therefore, both as a target after VEGF blockade and as an early pharmacodynamic marker and predictive biomarker for antiangiogenic therapy. Other pharmacodynamic biomarker candidates seem to be agent-specific. For example, circulating levels of soluble VEGFR2 and VEGFR3 proteins are decreased by TKIs that directly target these receptors (Table 5),18,21,23,39,48–50,52,53 but not by bevacizumab.22,58 The mechanisms by which these changes occur, their biological significance, and predictive biomarker value are not understood.
Exploration of biomarkers other than VEGF members is critical given their known involvement in tumor angiogenesis and vessel maturation.1,2 However, in patients with mCRC treated with bevacizumab and chemotherapy, pretreatment evaluation of biomarkers such as microvascular density, tumor tissue expression of TSP2, P53 and KRAS mutations has not been predictive of efficacy (Table 4).40,59 On the other hand, in previously untreated patients with mCRC responses to vatalanib plus chemotherapy in correlated directly with tissue messenger RNA levels of VEGFR1, LDHA (lactate dehydrogenase A) and Glut1 (CONFIRM1 trial) and inversely with hypoxia-inducible factor 1α (in the second line setting—CONFIRM2 trial; Table 3).60 In addition, patients with high baseline serum lactate dehydrogenase levels had longer progression-free survival and overall survival after treatment with vatalanib and chemoradiation.61 Unfortunately, both trials failed to show benefit in the experimental arm containing vatalanib.61 Baseline soluble intracellular adhesion molecule 1 was an independent prognostic factor of overall survival in patients treated with bevacizumab and chemotherapy or with chemotherapy alone in the phase III trial of bevacizumab in mNSCLC.41
Certain inflammatory cytokines might have potent proangiogenic effects (IL-1β, IL-6, IL-8, stromal-cell-derived factor [SDF]-1α, etc.). A phase II study suggested that the IL-8A-251T polymorphism (associated with increased protein expression) might be a molecular predictor of response to bevacizumab-based chemotherapy in ovarian cancer.62 Finally, in phase II studies, the extent of increase in inflammatory cytokines such as IL-6 in the plasma during treatment was associated with an inferior outcome in patients with rectal and ovarian cancer after bevacizumab and chemoradiation treatment, and an inferior outcome in patients with advanced HCC after sunitinib therapy.22,23,58 In line with these findings, preclinical studies have shown that sunitinib can induce elevation of circulating inflammatory cytokines in mice, which might result in more-aggressive recurrent or metastatic tumors.19,56,63,64
VEGF and other pathways targeted by certain TKIs (for example, c-KIT by sunitinib) might be important for the proliferation, survival and/or mobilization of certain cell populations into the blood circulation. Several groups have explored blood-circulating cells as potential biomarkers of antiangiogenic therapy (Tables 3 and and5).5). Indeed, in response to sunitinib, the number of circulating progenitor cells and monocytes can be decreased in patients with HCC and GIST, respectively.21,23 However, TKIs such as cediranib or bevacizumab combined with chemotherapy did not decrease or increase the circulating progenitor cells.65 The reasons for these differing results need to be addressed in preclinical models.
Several noninvasive, reproducible and quantitative radiological methods are emerging as potential pharmacodynamic biomarkers. For example, changes in dynamic MRI and CT-based tissue vascular measures such as blood flow, blood volume, or permeability have been shown to occur after treatment with bevacizumab or anti-VEGFR TKIs in clinical studies (Table 5). Water self-diffusion is also sensitive to changes in tumors after therapy,66 and might be a predictive marker in patients with glioblastoma treated with chemoradiation.67 Magnetic resonance spectroscopy (MRS) also holds promise as it provides chemically specific information;68 however, exploitation of the ability of this technique in predicting response to antiangiogenic agents is still in early stages of development. It remains unclear how and when these measurements should be performed for each agent, and whether these biomarkers have a predictive value.
Our group reported that the extent of decrease in Ktrans at day 1 after a single dose of cediranib (compared with the pretreatment value) as measured by vascular MRI in a patient with recurrent glioblastoma was associated with improved progression-free survival and overall survival.69 Similarly, the extent of drop in Ktrans at day 14 after sunitinib (compared with the pretreatment value) in advanced HCC was significantly associated with progression-free survival.23 Our data might explain, at least in part, the association between the decrease in tumor fluorothymidine uptake on PET assessment and overall survival in patients with recurrent glioblastoma treated with bevacizumab and irinotecan.70
The decrease in tumor vascular permeability and/or flow, as estimated by Ktrans,71 is consistent with vascular normalization, so we have proposed a composite ‘vascular normalization index’ as a biomarker that is associated with improved outcomes after cediranib treatment. This index integrated dynamics of Ktrans, MRI-measured cerebral blood volume and plasma collagen IV after one dosing of cediranib correlated with both progression-free survival and overall survival in patients with recurrent glioblastoma.69
Increased interstitial fluid pressure is a hallmark of solid tumors, and is caused by tumor vascular abnormalities.29 Given the potential of antiangiogenic agents to normalize tumor vasculature, this functional tumor parameter is also being explored as a biomarker in clinical studies. Indeed, in a study of bevacizumab in patients with rectal cancer, blockade of VEGF led to a drop in tumor interstitial fluid pressure (Table 5).22 Our group has also observed a decrease in tumor interstitial fluid pressure after bevacizumab treatment in patients with ovarian and metastatic breast cancer (Y. Boucher and R. K. Jain, unpublished data). Unfortunately, these measurements require insertion of a pressure-sensing needle into tumors and certain tumors are not amenable to this measurement. Thus, even if this biomarker can be validated independently, it would be hard to implement it in the clinic. Finally, another approach for predicting outcome has been the development of predictive models or nomogram. One such nomogram was developed for sunitinib in mRCC and included clinical scores as well as serum levels of alkaline phosphatase and lactate dehydrogenase.72
Collectively, these studies show that vascular permeability and perfusion, and circulating VEGF and PIGF should be further investigated as potential generic pharmacodynamic biomarkers for antiangiogenic therapies. Other candidates, such as soluble VEGFRs or circulating progenitor cells should be further evaluated as potential pharmacodynamic biomarkers for specific antiangiogenic agents or specific tumor types (Supplementary Tables 2–5). Prognostic biomarkers will most likely be disease specific and have the potential to aid the clinical management of cancer. Establishment of a predictive biomarker (Figure 1) remains a challenge, as discovery and validation will have to be tailored to the known mechanisms of action of a certain agent in a certain disease, and will probably necessitate standardization of costly, sophisticated protocols. Nonetheless, the benefit to patients—once these predictive biomarkers are established—is clear.
Avoiding serious toxic effects is critical in oncology, as most regimens contain potent cytotoxic drugs. It has been difficult to identify biomarkers of toxicity, primarily because of the low incidence of serious adverse events (for example, hemorrhage, perforations). Retrospective analysis of data from patients with lung cancer treated with chemotherapy plus bevacizumab in a phase III trial showed that tumor cavitation pretreatment might be a potential biomarker of pulmonary hemorrhage.73 Analysis of the VEGF-634 CC and VEGF-1498 TT genotypes in patients with metastatic breast cancer treated with bevacizumab plus chemotherapy showed significant associations with reduced risk of grade 3 or 4 hypertension.38 The clinical benefit in this study, however, was more prevalent in patients who developed grade 3 or 4 hypertension, which raises important questions as to whether this toxic effect should be a dose-limiting one.38 Collectively, these data show that the quest to establish biomarkers of toxicity will be challenging. Identifying the mechanisms underlying serious toxic effects might enable the discovery of such biomarkers.
Clinical experience has shown that antivascular effects of antiangiogenic agents are transient, and that tumors remain vascularized, probably by using or activating alternative proangiogenic pathways (Figure 2 and Supplementary Table 6). This hypothesis has been tested and proven in several preclinical studies,20,74–76 and clinical experience confirms that recurrent tumors are often highly vascularized after antiangiogenic therapy. Tumor tissues are difficult to obtain at recurrence after therapy, so most of the evidence has been obtained by studying circulating biomarkers. In phase II studies, our group has found that elevated plasma basic fibroblast growth factor and SDF1α in patients with recurrent glioblastoma receiving cediranib, and elevated plasma SDF1α and IL-6 and circulating progenitor cells in patients with advanced HCC treated with sunitinib, were associated with a poor outcome.23,37 Although these proangiogenic and proinflammatory biomarkers of resistance might not directly help in the clinical management of patients, they may aid in the identification of new targets and ideally, in the future, by allowing design of combinatorial schemes for individualized antiangiogenic therapy.
Alternative proangiogenic pathways might have a key role in cancer resistance to antiangiogenic therapy. Fortunately, since the interactions between these pathways can be studied preclinically, and because many of these targets can be inhibited with drugs, there is optimism that combinations of antiangiogenic agents or multitargeted antiangiogenic agents will substantially improve the outcomes of this therapy beyond a few months.
With increasing numbers of antiangiogenic agents being approved, or considered for approval, the need for biomarkers is more critical than ever for efficacy, safety, and cost considerations. Preliminary biomarker data are emerging. These data will have to be tested and validated in large, well-designed, prospective clinical trials. Biomarker selection would be greatly supported if we achieved a better understanding of the mechanism of action of these agents in cancer patients. Finally, once the candidate biomarkers are identified, standardized techniques will be required to measure imaging or circulating biomarkers. Although many challenges remain, future validation of biomarkers and their eventual incorporation into clinical practice holds promise for improved cancer treatment with anti angiogenic agents. Now, a collaborative effort between pharmaceutical companies, governmental agencies and private foundations is needed to realize this goal.
Information on clinical trials of antiangiogenic agents (available from the NIH databases) and the publications related to these studies were retrieved from the NIH website (http://www.clinicaltrials.gov), using the search engine on this site. PubMed was searched for studies of antiangiogenic agents using Entrez for articles published before 24 February 2009, including early-release publications. Search terms included “cancer”, “clinical trial”, “biomarker”, “anti-angiogenesis”, “anti-vascular”, “imaging”, and “tyrosine kinase inhibitor”. Full articles were checked for additional material when appropriate. The results of unpublished data conveyed to the authors by personal communication have also been included.
The authors thank the members of the Steele Lab, especially M. Ancukiewicz, Y. Boucher, E. di Tomaso, and L. Xu and M. Buyse, H. Chen, A. Grothey, C. Hudis, R. Horvitz, and A. Marshall for their helpful comments on this manuscript. The authors’ work is supported by grants from the National Cancer Institute P01-CA80124, P41-RR14075, R01-CA115767, R01-CA85140, R01CA126642, R21-CA99237, R21-CA117079, R01-CA129371, R01CA57683, K24-CA125440, Federal Share/NCI Proton Beam Program Income, M01-RR-01066, Harvard Clinical and Translational Science Center (CTSC) grant; the National Foundation for Cancer Research; the Richard and Nancy Simches Endowment for Brain Tumor Research; the Montesi Family Fund; and MIND Institute.
R. K. Jain declares associations with the following companies: AstraZeneca, Dyax, Millenium, Pfizer, Roche and SynDevRx. C. G. Willett declares associations with the following company: Genentech. A. X. Zhu declares associations with the following companies: Bayer and Genentech. T. T. Batchelor declares associations with the following companies: AstraZeneca, EMD-Serono, Exelixis, Genentech, ImClone Systems, Millenium and Schering-Plough. A. G. Sorensen declares associations with the following companies: AstraZeneca, Exelixis, Genentech, Millenium, Novartis and Schering-Plough. See the article online for full details of the relationships. The other authors declare no competing interests.
Supplementary Information is linked to the online version of the paper at www.nature.com/nrclinonc