ABSTRACT
The non-invasive detection of cancer mutations is a breakthrough in oncology. Here, we applied whole-exome sequencing of matched germline and basal plasma cell-free DNA samples (WES-cfDNA) on a RAS/BRAF/PIK3CA wild-type metastatic colorectal cancer patient with primary resistance to standard treatment regimens including VEGFR inhibitors. Using WES-cfDNA, we could detect 73% (54/74) of the somatic mutations uncovered by WES-tumor including a variety of mutation types: frameshift (indels), missense, noncoding (splicing), and nonsense mutations. Additionally, WES-cfDNA discovered 14 high-confidence somatic mutations not identified by WES-tumor. Importantly, in the absence of the tumor specimen, WES-cfDNA could identify 68 of the 88 (77.3%) total mutations that could be identified by both techniques. Of tumor biology relevance, we identified the novel KDR/VEGFR2 L840F somatic mutation, which we showed was a clonal mutation event in this tumor. Comprehensive in vitro and in vivo functional assays confirmed that L840F causes strong resistance to anti-angiogenic drugs, whereas the KDR/VEGFR2 hot-spot mutant R1032Q confers sensitivity to cabozantinib. Moreover, we found a 1-3% of recurrent KDR somatic mutations across large and non-overlapping cancer sequencing projects, and the majority of these mutations were located in protein residues frequently mutated in other cancer-relevant kinases, such as EGFR, ABL1, and ALK, suggesting a functional role.
In summary, the current study highlights the capability of exomic sequencing of cfDNA from plasma of cancer patients as a powerful platform for somatic landscape analysis and discovery of resistance-associated cancer mutations. Because of its advantage to generate results highly concordant to those of tumor sequencing without the hurdle of conventional tumor biopsies, we anticipate that WES-cfDNA will become frequently used in oncology. Moreover, our study identified for the first-time KDR/VEGFR2 somatic mutations as potential genetic biomarkers of response to anti-angiogenic cancer therapies and will serve as reference for further studies on the topic.
INTRODUCTION
Judah Folkman demonstrated almost half a century ago that neovascularization is a key requirement for solid tumor growth and metastasis1. This discovery motivated the development of anti-angiogenic drugs targeting cancer endothelial cells in order to deprive tumors from their blood supply2. Three decades later, bevacizumab, a recombinant humanized monoclonal antibody blocking vascular endothelial growth factor A (VEGF-A) and consequently, the activation of the VEGF receptor 2 (VEGFR2) on endothelial cells, became the first anti-angiogenic drug to be approved by the Food and Drug Administration (FDA) for cancer treatment3. The safety and efficacy of bevacizumab have been assessed in several randomized controlled clinical trials that showed an extended overall survival of patients with metastatic colorectal cancer (mCRC)3,4. However, the majority of patients (approximately 50% and 80% in first and second lines, respectively) do not benefit from such treatment; biomarkers predictive of the response to bevacizumab, as well as new agents targeting the VEGFR2 pathway, such as regorafenib, are still a fundamental unmet medical necessity5.
The possible involvement of genetic variants of the VEGF-VEGFR1/2 pathway in the outcome of anti-angiogenic treatment has been extensively investigated. Although some studies have suggested the potential association of tumor response with VEGF/VEGFRs germline polymorphisms6, these results could not be confirmed in subsequent assessments7. Recent in vitro studies demonstrated that VEGFR2 plays a functional role not only in endothelial cells, as is usually assumed, but also prominently in cancer cells8. This finding indicates that, similar to other drug target kinases such as the epidermal growth factor receptor (EGFR) and the ABL proto-oncogene 1 (ABL1)9-11, VEGF/VEGFR2 mutations occurring in the tumor cells, rather than inherited polymorphisms, could be regulating drug efficiency.
The conventional approach used for the discovery of the genetic causes of therapy resistance relies on genetic or genomic analyses of the patient’s tumor, which implies surgery or tumor biopsy. Here, we identified a novel somatic mutation in the key player of angiogenesis KDR/VEGFR2 (L840F) in an mCRC patient who was highly refractory to standard and experimental therapies, including angiogenesis inhibitors, despite wild-type (WT) results in targeted liquid biopsy analysis, by following an innovative approach of whole-exome sequencing of matched germline and plasma cfDNA (WES-cfDNA)12. In vitro and in vivo functional analyses demonstrated that this mutation, as well as other recurrent KDR/VEGFR2 mutations reported in large cancer databases, can modulate the efficiency of anti-angiogenic therapies.
RESULTS
CLINICAL
Case report
A 56-year old man was diagnosed in our center with a WT KRAS/NRAS/PIK3CA/BRAF mCRC cancer with liver and lung metastases (cT4N2M1). He received FOLFIRI-cetuximab as frontline treatment but there was evidence of tumor progression after the first tumor evaluation as suggested by the appearance of a new liver lesion. Consequently, he received the following treatments sequentially: FOLFOX-bevacizumab, afatinib with cetuximab, oncolytic adenovirus monotherapy (the last two lines of treatment in the context of phase 1 clinical trials), rechallenge with capecitabine-bevacizumab, and finally, regorafenib; however, there was no response and progressive clinical worsening occurred (Figures 1A and S1). The lack of radiological or clinical benefits in response to any of these treatments was mainly because of the persistent growth of his liver tumor burden, as the single lung metastasis grew less prominently. The patient died within a short time period (14 months) after the initial diagnosis due to complications of his progressive disease. Highly-sensitive BEAMing (beads, emulsions, amplification and magnetics; Sysmex Inostics, USA) digital PCR liquid biopsy assays revealed no alterations in the hot-spots of the KRAS/NRAS/PIK3CA/BRAF genes in the patient’s prior-and on-treatment plasma cell-free DNA (cfDNA) samples. Moreover, none of these mutations were identified in the tumor samples by Cobas (Roche), suggesting that other genes could be responsible for such prominent resistance. Therefore, we sought to design an unbiased strategy to identity and characterize newly implicated genes in cancer therapy resistance.
Samples
The first sample available for molecular analyses was a fragment from the patient’s liver metastasis obtained during the diagnostic biopsy, which was used (and exhausted) on the histopathology and the routine genetic analysis, owing to its small size. Following the diagnosis of KRAS/NRAS/PIK3CA/BRAF WT mCRC, the patient accepted to participate in this exploratory study. Peripheral whole-blood and basal plasma samples were collected and used to test the WES-cfDNA strategy. After progression to first line, a second blood/plasma sample was collected, and another liver metastasis biopsy was performed (Figures 1 and S1). The on-treatment plasma sample was used for validating the KDR mutation identified on basal plasma. The second liver biopsy was divided in two parts: one fraction was used to generate the patient-derived xenograft (PDX) model, herein called Avatar model, and the other fraction for tumor WES (WES-tDNA). The list of somatic mutations identified by WES-cfDNA was compared to that from WES-tumor to assess the mutation detection efficiency of WES-cfDNA.
GENOMIC ANALYSES
Discovery of the clonal, somatic KDR/VEGFR2 L840F mutation by WES-cfDNA
WES-cfDNA confirmed the WT status of the KRAS/NRAS/PIK3CA/BRAF genes (Figures S2-S4), as previously indicated by BEAMing. In addition, WES-cfDNA uncovered two known colorectal cancer driver mutations, APC c.3964G>T E1322X (COSM18702) and TP53 c.659A>C Y220S (COSM43850) (Figures S5-S6), as well as the KDR c.2518C>T mutation leading to the L840F change in the VEGFR2 receptor (Figures 1B, S7). This KDR mutation was predicted to be pathogenic based on the high degree of conservation of L840, and had not been previously reported in cancer or population sequencing projects. We confirmed the KDR c.2518C>T allele in the patient´s basal and on-treatment cfDNA samples (collected after progression with FOLFIRI-cetuximab) by TaqMan genotyping assay, but not in the corresponding gDNA, confirming its somatic status (Figures 1B, S7-S8).
To estimate the clonality of KDR/VEGFR2 L840F, we used mutated allele frequencies (MAFs) data from WES of both tumor and plasma cfDNA (shown in Figure 2A and Table S1). Importantly, the MAFs of KDR/VEGFR2 L840F were similar to the MAFs of trunk CRC mutations, such as those in APC and TP53. The MAF of KDR/VEGFR2 L840F in the tumor sample was approximately 30%, whereas the MAFs for the APC p.E1322EX and TP53 p.Y220SY mutations were both approximately 50%. Importantly, the MAF of KDR/VEGFR2 mutation in plasma (11%) was higher than the MAF of the APC mutation (8%), whereas the MAF of the TP53 mutation was the highest (18%). Our data demonstrate that the KDR/VEGFR2 mutations can occur before and independently of targeted therapy pressure. Moreover, after normalization with the MAFs of APC and TP53, our results show that L840F was likely a clonal mutation event in this tumor (see (supplementary discussion, point I).
High concordance between WES-cfDNA and WES-tumor
Once we gained access to the tDNA after the second liver metastasis, we confirmed the L840F mutation by Sanger sequencing (Figure 1B) and performed WES-tumor for comparison with WES-cfDNA. Applying the same stringent bioinformatics filters (described on the online methods), WES-cfDNA could detect 73% (54/74) of the somatic mutations uncovered by WES-tumor (Figure 2A). WES-cfDNA identified a variety of mutation types: frameshift (including insertions and deletions), missense, noncoding (splicing), and nonsense mutations. Additionally, WES-cfDNA discovered 14 high-confidence somatic mutations not identified by WES-tumor. Importantly, in the absence of the tumor specimen, WES-cfDNA could identify 68 of the 88 (77.3%) total mutations that could be identified by both techniques. The complete list of somatic mutations and all the sequencing parameters and genomic annotation are depicted in Table S1. Our results reveal the high capacity of tumor-free WES-cfDNA for global detection of somatic mutations, which was similar to that of tumor DNA sequencing. There was also high concordance in the identification of copy number variation (CNV) between WES-cfDNA and WES-tumor (Figure 2B).
KDR somatic mutations are recurrent in cancer
After the exclusion of the variants present also in the general noncancerous population, KDR bona-fide cancer mutations were found in 691 (1.6%) of the 33,320 cancer samples in the Catalogue Of Somatic Mutations In Cancer (COSMIC, v80) database. Furthermore, somatic mutations in KDR occurred in 481 (2.6%) of the total samples and in 56 (2.7%) of the 2,102 CRC samples from almost 19,000 cancer samples included in the AACR Project Genomics Evidence Neoplasia Information Exchange (GENIE) project (Figure 3A). We also analyzed KDR mutations from a third non-overlapping cancer genomic project, The Pancancer Analysis of Whole Genomes (PCAWG) study, which includes whole-genome sequencing data from 2,800 samples of 37 distinct types of cancer. From the 35 CRC samples included in the PCAWGS project, 4 (11.5%) harbored KDR mutations considered pathogenic. Although smaller than the other datasets, the PCAWGS database presents the advantage that whole-genome sequencing yields a more homogeneous coverage than whole-exome sequencing, enabling greatly more accurate copy number analysis results. Interestingly, 15 of the 35 (43%) CRC samples in PCAWGS had an amplification of the KDR gene. Moreover, KDR mutations were identified in 2.5% of the recently published Memorial Sloan Kettering Cancer Center (MSK-IMPACT) Clinical Sequencing cohort, including almost 11,000 samples from several cancer types13 (Figure 3A). The KDR somatic mutations found in the cancer genomics databases are shown in Table S2.
Importantly, KDR was statistically classified as a driver of colorectal cancers (p = 2.0 × 10-7) and melanomas (p = 4.3 × 10-5) in the PCAWGS database, and a driver of lung adenocarcinoma and glioblastoma multiforme tumors by the Integrative Onco Genomics (IntOGene) platform14, which applies OncodriveFM, OncodriveCLUST, MutSigCV, and OncodriveROLE cancer driver detection methods.
Mutations structurally analogous to the recurrent KDR mutations occur in other kinases in cancer
In addition to the previously unrecognized occurrence of KDR mutations in cancer, we also found that structurally similar mutations occur in hot-spot residues of other kinases, clinically relevant for human cancers. For example, the analogous mutations of VEGFR2 L1049W (EGFR L858) and of VEGFR2 D1052N/G/H (FLT3 D835, KIT D816, EGFR L861, and PDGFRA D842) can be consider cancer hot-spots, as they are mutated in 58,871 cancer samples from the COSMIC database alone (Figures 3B-C).
Importantly, we found that R1032Q (COSM192176) was the most frequent mutation in three large non-overlapping cancer databases (TCGA, GENIE, PCAWGS), suggesting that R1032 is a possible mutational hot-spot in VEGFR2. Indeed, previous studies have shown that R is the most mutated residue in many kinases in different cancer types owing to the predominant C>T (G>A) transition mutability in CpG sites (four of the six codons for R have a CpG dinucleotide in the first and second positions)15. Moreover, our comparative structural analysis confirmed the mutation predisposition of this specific conserved R residue among several oncogenic kinases: VEGFR2 R1032Q; EGFR R841K/R; KIT R796A/G/P/K; ALK R1253G/T; AXL R676M; CSK R318C; EPHA2 R743G/H/R; EPHA3 R750L/Q/W; EPHA7 R762C; EPHB1 R748K/S; EPHB2 R750C; JAK3 R651Q; MERTK R727Q; NTRK3 R683S; PDGFRA R822F/H; ROR1 R619C/H/S (COSMIC and TCGA databases; Figures 3B and S9). In addition to R1032Q, identified in 20 cancer samples from the cancer sequencing projects analyzed, the R1032X nonsense mutation was detected in seven samples from distinct types of cancers, such as colorectal, lung, prostate, and haematopoietic cancers, as well as in melanomas and glioblastomas.
FUNCTIONAL STUDIES OF THE VEGFR2 CANCER MUTANTS
L840F causes therapy resistance
In agreement with the phenotype we observed in the patient, the patient-derived Avatar model (AvatarVEGFR2:L840F) did not respond to multiple anti-VEGF and VEGFR2 inhibitors, whereas the AvatarVEGFR2:WT model, used as a control, was sensitive to all the treatments (Figure 4A). In order to understand better how VEGFR2 L840F caused such broad in vivo resistance to anti-angiogenic drugs we conducted both in silico and biochemical analyses.
Our 3D structural analysis revealed that L840 is located exactly at the entrance of the ATP-binding pocket of the tyrosine kinase domain of VEGFR2 (aa840-LGXGXXG-846aa), and it forms hydrophobic interactions with many FDA-approved small-molecule kinase inhibitors. Figures 4B and S10-S14 illustrate the direct contacts of VEGFR2 L840 with lenvatinib, sorafenib, axitinib and tivozanib, while the bindings of the VEGFR2 L840 analogous mutations ABL1 L248 with bosutinib, nilotinib, dasatinib, imatinib and axitinib, and EGFR L718 with WZ4002 are shown in Figures S15-S20.
We then modeled the L840F mutation in silico and observed that a clash occurs between the tyrosine kinase inhibitors (TKIs) and the F840 residue (Figure 4C), which would prevent the original mode of TKI binding to the receptor. Molecular dynamics simulations of the L840F mutant VEGFR2 also show that most of the conformations of F840 observed in the simulations are not compatible with inhibitor binding (Figures 4C, S21). Consistent with the computational model, in vitro kinase assays with recombinant VEGFR2 kinase domains showed that WT VEGFR2 has high kinase activity, whereas L840F VEGFR2 has impaired kinase activity, suggesting possible loss of ATP binding (Figure 5A). Consistent with these results, Y1175 phosphorylation of L840F VEGFR2 was significantly reduced compared to WT VEGFR2, both in human embryonic kidney (HEK293) cells transiently transfected with WT or L840F VEGFR2 and in porcine endothelial (PAE) cells stably expressing WT or L840F VEGFR2 (Figure S22). We also explored whether the molecular clash caused by the L840F mutation would prevent TKI inhibition in vitro. For this, we rescued L840F kinase activity by increasing the enzyme concentration a thousand times and evaluated its inhibition by TKIs. Indeed, we found that whereas WT VEGFR2 was sensitive to axitinib, cabozantinib, dovitinib, and levantinib, L840F VEGFR2 was resistant to all these drugs, albeit at distinct levels (Figure 5B).
R1032Q confers sensitivity to strong VEGFR2 inhibitors
In addition to L840F VEGFR2, we characterized the kinase activity of R1032Q VEGFR2 with our in vitro kinase assays, and we found that similar to the L840F, the R1032Q mutation greatly reduced VEGFR2 kinase activity (Figure 5A), although the structural features underlying this phenotype would be distinct. Whereas L840F most likely blocks ATP and TKI entrance to the ATP-binding site leading to kinase inactivation and TKI resistance, R1032Q does not directly interfere with the ATP-binding pocket. Instead, R1032Q affects the kinase catalytic motif DxxxxN (aa1028-DxxxRN-1033aa), which is present in all known protein kinases and is directly involved in the catalytic mechanism of the enzyme16. Notably, in vitro analyses by different research groups have identified the following mutations in R1032-analogous residues of other kinases that greatly impair kinase activity: EGFR R84117-19, c-KIT R79620, ALK R125321, CSK R31822.
Since R1032, contrary to L840, does not directly participate in receptor:TKI binding (Figures S10-20), we asked whether R1032Q VEGFR2 would be inhibited by TKIs. In vitro kinase assays showed increased sensitivity of R1032Q VEGFR2 to TKIs (Figure 5B). Furthermore, proliferation studies with the Colo-320 colorectal cell line, which has a similar mutation profile as that of the patient’s tumor (WT KRAS/NRAS/BRAF/PIK3CA status, and TP53 and APC mutations), showed that stable expression of R1032Q VEGFR2 conferred sensitivity to lenvatinib (growth inhibition (GI50) = 20.8 for the R1032Q compared to 36.4 for WT VEGFR2) and to cabozantinib (GI50 = 2.5 for the R1032Q compared to 7.9 for WT VEGFR2) (Figure 5C). Moreover, cabozantinib treatment of the MDST8 CRC cell line, naturally harboring the KDR R1032Q mutation, led to a prominent decrease in cell growth rate in vitro and diminished the high constitutive ERK phosphorylation levels (Figure S23). Importantly, we found that such downstream inhibition was specific to cabozantinib, a very strong VEGFR2 (0.035 nM) and c-MET (1.3 nM) inhibitor, and occurred in cells treated in the absence or presence of VEGF (Figure S23).
Oncogenic effects of L840F, R1032Q, and other VEGFR2 cancer-related mutants
Based on recent studies that comprehensively showed that Protein Kinase C β (PKCβ) and Mixed-Lineage Kinase 4 (MLK4) loss-of-function mutations play an oncogenic role in CRC23,24, we asked if in addition to modulating the response to TKIs, VEGFR2 cancer mutants could promote tumor growth. Indeed, we showed that Colo-320 cells stably expressing L840F VEGFR2, even when injected in small numbers and without matrigel, could generate tumors that reached the established humane endpoint (Figure 6). In addition, the R1032Q hot-spot mutant, as well as other cancer-related VEGFR2 mutants (D717V, G800D, G800R, G843D, S925F, R1022Q, R1032Q, and S1100F), promoted tumor growth in vivo. Importantly, Colo-320 expressing empty vector (EV) or the kinase inactive dominant-negative K868M VEGFR2 did not generate tumors within 120 days after cell injections. Our data suggest that cancer-associated VEGFR2 mutants might have oncogenic potential (Figure 6).
DISCUSSION
Our study brings three main novelties to the field of cancer translational research. First, we established the potential of WES-cfDNA as a global tumor-free genomic platform to explore genetic causes of primary resistance to cancer therapies. Second, following the identification of the L840F VEGFR2 clonal, somatic mutation in a highly refractory mCRC patient by WES-cfDNA, we performed functional analyses that provide a mechanistic understanding of KDR/VEGFR2 somatic mutations as genetic modulators of the response to anti-angiogenic drugs. Lastly, we explored genomic sequencing databases and showed that recurrent KDR/VEGFR2 somatic mutations, which are structurally analogous to cancer host-spot mutations in other kinases, occur in 1-3% of human cancers.
The high capacity of WES-cfDNA for portraying the somatic mutation and copy number variation landscapes of tumors has immense potential for research in translational oncology12,25,26. During the last decade, cancer genomics projects proved whole-exome sequencing of tumor samples, WES-tumor, to be a powerful and affordable technique that is still considered the gold-standard platform for the unbiased discovery of somatic mutations and the characterization of cancer genome landscapes. A key limitation of WES-tumor is the use of tumor fragment(s) acquired during surgery or biopsy as the source of cancer DNA. The future implementation of WES-cfDNA analysis into genomic projects will enable the investigation of a much larger number of cancer patients, including those for whom only blood/plasma samples are available because of an inoperable tumor or because the tumor localization renders biopsy unsafe. Additionally, WES-cfDNA can uncover somatic mutations not represented by the WES-tumor, therefore complementing the knowledge about the patients’ cancer and allowing the identification of more drug targets and therapeutic options. Following these promising results, we have implemented the WES-cfDNA prospective gene discovery and clonal evolution platform into the NCT02795650 ongoing clinical trial and have identified potentially therapeutic relevant mutations in BRCA2 and ROS1 in advanced pancreatic cancer patients (Toledo RA et al, manuscript in preparation). VEGFR2, EGFR, ABL1, and PDGFRA are major cancer therapeutic targets with several TKIs already approved in the clinic. However, whereas TKI sensitizing and resistant mutations in EGFR, ABL1, and PDGFRA have already been well characterized, much less is known about mutations in VEGFR2 and their clinical implications (supplementary discussion, points I and II). The main reasons for this surprisingly limited amount of information on KDR/VEGFR2 mutations and their pharmacological impact are: a) the concept that VEGFR2 is mostly expressed and playing a role on endothelial but not on tumor cells, and b) the previous notion of paucity or even absence of KDR mutations in cancers, as suggested by small genetic screening studies27. Thus, after the identification and confirmation of the L840F clonal, somatic mutation in a refractory mCRC cancer (supplementary discussion, point III), we aimed to further characterize the frequency and the role of KDR/VEGFR2 mutations across human cancer. To this end, we performed an expanded screening of cancer genome sequencing projects and carried out a series of comprehensive in vitro and in vivo functional experiments.
Our experiments with cell lines, animal models, and biochemical assays showed that the L840F ATP-binding pocket domain mutation causes very strong and broad resistance to anti-VEGF and VEGFR2 inhibitors (supplementary discussion, point V). On the contrary, we showed that the R1032Q kinase hot-spot mutation is sensitive to the strong VEGFR2 inhibitor cabozantinib. These new findings demonstrate that, as occurring in EGFR, ABL1, and PDGFRA, mutations in VEGFR2 can have functional consequences and influence the efficiency of cancer targeted therapies. In addition, taking advantage of publicly available sequencing data from the TCGA, GENIE, PCAWGS, and MSK-IMPACT cancer genomic databases, we found that 1-3% of the 73,389 cancer samples analyzed harbored a potentially pathogenic somatic mutation in KDR/VEGFR2.
In agreement with our findings of a consistent occurrence of KDR mutations in cancers, three brief case reports describing patients with a KDR-mutated tumor became very recently available. Although no accompanying experimental data were provided, the patients were treated with anti-angiogenic drugs based on the expectation that the VEGFR2 mutant could be a potentially sensitive target, as follows. Knepper et al. reported a prolonged complete response to pazopanib in a metastatic basal cellular carcinoma patient carrying the somatic KDR/VEGFR2 R1032Q28. This case represents the clinical confirmation of our preclinical experimental results, showing that R1032Q is a sensitizing mutation for at least some strong VEGFR2 inhibitors. The second case was a mCRC patient treated with 5-fluorouracil-bevacizumab for six cycles with progression of the disease. A panel of 47 genes was analyzed by NGS and in addition to common mutations in APC and KRAS, the KDR R961W mutation was observed in nearly 30% of the reads29. The patient was offered low dose regorafenib (80 mg/day) that promptly needed to be reduced to 40 mg/day owing to secondary effects. After 3 months of treatment, imaging scans reveled remarkable improvement in the hepatic metastases, abdominal and retroperitoneal lymph nodes, and rectosigmoid colon hypermetabolic lesions. The third case was a mCRC patient who progressed to FOLFOX-cetuximab, folinic acid/fluorouracil-cetuximab, and was then treated with successive irinotecan-cetuximab plus ramucirumab30. After two cycles, the patient developed a sporadic expanding angioma that was shown by WES to be clonal and carry the KDR T771R as the only somatic mutation. The authors suggest that KDR T771R may offer a proliferative advantage in the setting of ramucirumab treatment. A fourth KDR-mutated case was recently identified during the generation of tumor organoids in a series of twenty consecutive CRC patients, however no information on therapeutics was provided (supplementary discussion, point VI),
Collectively, the data from these three cases show no evidence of benefit by indirect or direct blocking of the extracellular domain of VEGFR2 mutants (i.e. with bevacizumab and ramucirumab). Interestingly, the two patients with tumors carrying mutations located in the kinase domain (R1032Q and R961W) responded well to VEGFR2 inhibitors (pazopanib and regorafenib), whereas the other two patients with tumors carrying mutations outside of the kinase domain (T771R and L840F) did not respond to TKIs. These latter patients not only did not respond, but it seems that blocking of VEGFR2 had adverse effects. For example, our mCRC patient carrying the L840F ATP-binding domain mutation progressed very fast to regorafenib (Figure 1A) and the mCRC KDR T771R-mutated patient developed an expanding angioma after ramucirumab30.
The classical understanding of kinases as straightforward “activators/phosphorylators” of downstream pathways has been greatly expanded and at least six types of cancer kinase mutations have been discovered and characterized, including genesis/extinction of phosphorylation sites, node activation/inactivation, and downstream/upstream rewiring31. The L840F and R1032Q VEGFR2 mutations impairing kinase activity would best fit the upstream rewiring mutant kinase oncogenic model, whereby the mutant receptor recruits unusual partner(s), which are responsible for MAPK pathway activation (Figure S24). This has been demonstrated for BRAF impaired mutants that recruit CRAF, promoting alternative MAPK pathway activation and leading to BRAF inhibitor therapy resistance. Interestingly, this BRAF:CRAF pathway rewiring confers de novo clinical sensitivity to dasatinib (ABL, SRC, and c-Kit inhibitor)32. In our study, tumor growth of the AvatarVEGFR2:L840F model was not significantly reduced upon treatment with inhibitors of EGFR (afatinib), MET (crizotinib), and MAPK (MEKi) (Figure 4A), and the possible rewiring partner of L840F VEGFR remains unknown (supplementary discussion, point VI). Our findings resonate the difficulties of therapeutically targeting kinases with mutations in the L residue located at the entrance of ATP/TKI binding.
The current study is the first to implicate in a robust manner VEGFR2 as a direct regulator of the efficacy of anti-angiogenic therapies in common cancers and can facilitate answering a long-standing unmet medical question regarding biomarkers of response to anti-angiogenic drugs. A complete functional and clinical characterization of KDR cancer mutations will be necessary to classify phenotypes of cancers carrying VEGFR2 mutants as neutral, sensitizing, or resistant to anti-VEGF and VEGFR2 inhibitor treatments. These data should be taken into consideration for the future design of small population basket clinical trials based on KDR genotypes. Moreover, a retrospective or prospective investigation of KDR gene expression within CRC cohorts treated with anti-angiogenic drugs could also be of interest, since we showed that among more than 1,000 patients analyzed, those with mCRC with high tumor KDR gene-expression did significantly worse (a finding confirmed in four independent CRC cohorts, Figure S25).
In summary, the current study highlights the capability of exomic sequencing of cfDNA from plasma of cancer patients as a powerful platform for somatic landscape analysis and discovery of resistance-associated cancer mutations. Because of its advantage to generate results highly concordant to those of tumor sequencing without the hurdle of conventional tumor biopsies, we anticipate that WES-cfDNA will become frequently used in oncology. Moreover, our study identified for the first-time KDR/VEGFR2 somatic mutations as potential genetic biomarkers of response to anti-angiogenic cancer therapies and will serve as reference for further studies on the topic.
Author contributions
Conceived and oversaw the project: RAT, JLMT, MH; conceived the plasma cfDNA whole-exome sequencing gene discovery strategy: RAT; diagnosed and treated the patient: EG, EV, RA, AC, MH; provided clinical and sample management: FS, SP; designed, performed, interpreted experiments: RAT, MM, JM, MC, TSP, MIA, AO, ADM, DL, CBA, OD, JLMT; performed in vivo studies: NB, YD, VB; performed protein structural analyses: TP, MC, DL; performed genomic and protein database analyses: TP, RAT; wrote the paper: RAT with inputs from JLMT, EG, MM, MH and all authors.
ONLINE METHODS
Study supervision
The study was approved by the institutional Review Boards of Hospital Universitario HM Sanchinarro and conducted in agreement with the Declaration of Helsinki and the International Conference on Harmonization of Good Clinical Practice guidelines. The patient gave written informed consent to participate in the study. Mice used in this research were treated humanely according to the regulations laid down by the Spanish National Cancer Research Centre (CNIO) Bioethics Committee.
DNA extraction
DNA was extracted from leukocytes (gDNA), liver metastasis (tDNA), and basal and on-treatment plasma samples (cfDNA), using commercial kits according to the manufacturer’s instructions (Qiagen, Germany). The DNA amount was quantified with a Qubit™ Fluorometer (Thermofisher, USA) and reported in ng. cfDNA samples were also quantified using a modified version of human LINE-1-based quantitative real-time PCR and reported in genome equivalents (GE; GE being one haploid human genome weighing 3.3 pg). gDNA and tDNA was sheared to 300-bp fragments on a Covaris instrument (Covaris, Woburn, MA) according to standard procedures. The 2100 Bioanalyzer (Agilent, USA) was used to access the quality and size of the pre-processed and post processed samples and libraries.
Routine genetic analysis
The FDA-approved Cobas mutation kit (Roche, Switzerland) was used to analyze the following mutations in the diagnostic biopsy tDNA: KRAS (G12S/R/C/V/A/D, G13D, Q61H, A146T), NRAS (Q61K/R/L/H), BRAF (V600E), and PIK3CA (E542K, E545K/G, Q546K, M1043I, H1047Y/R/L). The presence of the same mutations in the patient’s basal and on-treatment cfDNA samples was assessed by the highly sensitive BEAMing technique, as previously described1.
Whole-exome sequencing
Sequencing libraries of cfDNA (15 ng), and gDNA and tDNA (70-110 ng) samples were prepared using the ThruPLEX Plasma-and DNAseq Kits (Rubicon Genomics Inc, USA), respectively. Barcode indices were added to samples during eight PCR cycles of template preparation, and 550 ng of each sample was processed through the SureSelectXT Target Enrichment System (Agilent SureSelect V5, ref. 5190-6208, protocol G7530-90000 version B1). xGen Blocking Oligos (IDT, Iowa, USA) were used as suggested by Rubicon Genomics. Captured targets were subsequently enriched by 11 cycles of PCR with KAPA HiFi HotStart (Kapa Biosystems), with a Tann of 60° and the following primers, which target generic ends of Illumina adapters: AATGATACGGCGACCACCGAGAT and CAAGCAGAAGACGGCATACGAGAT. For sequencing, magnetic bead-purified libraries with similar concentrations of cfDNAs and tDNA, and half the concentration of gDNA were pooled in order to increase coverage and favor the detection of non-inherited sub-clonal mutations. Sequencing was carried out in the Illumina HiSeq4000 platform. All sequencing data are going to be deposited in the European Nucleotide Archive (ENA) under the accession number ENA#202177, at the time of publication.
Somatic mutation call
Bioinformatics analyses were performed using the NEXTGEN software (Softgenetics, USA), as previously described2. The detailed parameters used for the alignment and mutation call are provided as supplementary material. Briefly, FastaQ files were aligned using the BWA pipeline and the variants were processed by sequential stringent filters to exclude low-confidence variants. Only variants that passed the following filters were classified as high-quality and considered in the study: overall and allele scores ≥ 12; coverage ≥ 20; number of mutated reads ≥ 20; percentage of mutated reads ≥ 3% of cfDNA / tDNA and ≥ 35% of gDNA; F:R read balance ≥ 0.1; and F:R read percentage ≥ 0.45. The list of non-hereditary mutations detected by WES-cfDNA and WES-tumor was generated after disregarding germline variants (obtained by WES-gDNA). A detailed genomic annotation of the somatic mutations we identified, prediction of mutation pathogenicity based on predictor algorithms (SIFT, Polyphen2, LRT, Mutation Taster, Mutation Assessor, and other software packages included in the dbNSFP3), allele frequencies in population studies, such as 1000G and EXAC, and additional information are shown in Table S1.
TaqMan SNP genotyping assay
A custom TaqMan® genotyping assay for the detection of the KDR c.2518C (L840L) and KDR c.2518C>T (L840F) alleles was designed using the Thermofisher online Design Tool (oligonucleotides and probes are shown in Table S3).
Genetic/protein database and protein structure analyses
Previously reported germline and somatic variants in KDR were retrieved from general population (EXACT4 and ESP5) and cancer (COSMIC6, GENIE7, PCAWGS8) sequencing public projects. The VEGFR2, EGFR, and ABL1 protein structures were obtained from the RCSB data bank; structurally analogous mutations in other cancer-relevant kinases were identified using MutationAligner9; kinase residues interacting with kinase inhibitors were mapped using the LigPlot10 software. Computational modeling of inhibitor binding to WT and L840F VEGFR2 was performed as previously described.
Generation and treatment of the Avatar patient-derived xenograft (PDX) model
Liver metastasis biopsy was performed after tumor progression to capecitabine-bevacizumab rechallenge (Figures 1 and 1S). A fraction of the biopsy was used to generate the Avatar model as previously described by our group11,12. Expanded cohorts (five to six animals per arm) were treated with: anti-VEGF drugs (B20/murine and bevacizumab/human), VEGFR2 kinase inhibitors (axitinib, cabozantinib, cabozantinib:MEK inhibitor combo, lenvantinib, pazopanib, regorafenib, and sorafenib), and inhibitors of other kinases, such as afatinib (EGFR), crizotinib (MET), and MEK inhibitor (MAPK). Information on the treatment regimens is shown in Table S4.
Cell lines
The human CRC cell lines used in the current study were selected based on their genotype in order to be as informative as possible for each experiment. Thus, we chose the Colo-320 cell line to interrogate the phenotypic changes caused by the overexpression of VEGFR2 mutants because it has the same genetic background as the patient´s tumor (mutated TP53/APC and WT KRAS/BRAF). The MDST8 CRC cell line was used for drug sensitivity studies because it naturally harbors the KDR/VEGFR2 R1032Q mutation, which we found to be a hot-spot VEGFR2 mutation in human cancers.
Colo320 and MDST8 colorectal cell lines were obtained from ATCC and cultured at 37°C in 5% CO2, in Roswelll Park Memorial Institute (RPMI) Medium 1640 + GlutaMAX (Gibco, USA) and Dulbeccos’ modified Eagle’s medium (DMEM) + 2 mM Glutamine (Gibco, USA), respectively, supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher Scientific, USA). Porcine aortic endothelial (PAE) cell lines, kindly provided by Dr. Kurt Ballmer-Hofer, were grown in DMEM supplemented with 10% FBS.
Generation of stable colorectal and endothelial cell lines
Colo-320 and PAE cell lines were used to generate cell lines stably expressing the VEGFR2 mutants using a previously described protocol12. Briefly, cells were seeded in 10-cm plates in the appropriate medium and were grown to 70% confluence. Transfection with constructs carrying either the empty vector or VEGFR2 (WT or mutant) was performed with polyethylenimine (PEI) as previously described13. Briefly, 30 μg of WT or mutant VEGFR2 plasmid (in the pBE vector containing the neomycine resistance gene, which confers resistance to the selection antibiotic G418) was mixed with 60 μl PEI (1 mg/ml in H2O) in 2 ml serum-free DMEM, incubated for 10 min at room temperature and added to the cells. Following a 3-h incubation at 37°C, the medium was changed, and the cells were allowed to grow to 100% confluence. Cells were re-seeded at a series of dilutions (1:1000-1:5000) in antibiotic selection medium (1 mg/ml G418) to allow for single colonies to grow, while non-transfected cells were dying. Individual colonies were consecutively transferred to 24-well and 6-well plates and screened by western blotting for VEGFR2 expression. To reduce polyclonality, colonies with the highest expression levels were subjected to 3 additional rounds of subcloning.
VEGF stimulation and western blotting
Transiently transfected HEK293 cells or stable PAE cell lines expressing WT or L840F-KDR were starved in DMEM supplemented with 1% bovine serum albumin (BSA) for 4h at 37°C and were subsequently stimulated with 1.5 nM (60 ng/ml) VEGF165 for 10 min at 37°C. Following stimulation, the cells were scraped in lysis buffer (50 mM Tris pH = 8.0, 120 mM NaCl, 1% NP-40) supplemented with protease inhibitors (Roche, cat. Nr 04693159001) and phosphatase inhibitors (1 mM sodium orthovanadate and 20 uM phernylarsine oxide) and incubated for 30 min on ice. Cell lysates were collected as the supernanant of a centrifugation at 30,000 × g for 15 min and subjected to western blot analysis. The following antibodies were used to probe receptor activation: total KDR (Cell Signaling, cat. Nr 2479), phospho KDR at Y1175 (Cell Signaling, cat. Nr 2478). The secondary antibodies used were alkaline phosphatase (AP) conjugated (Southern Biotech). All antibodies were diluted at a 1:1000 ratio in 5% BSA in Tris-buffered saline, containing 0.05% Tween20 (TBST) buffer. The chemiluminescence signal was developed with the Novex AP Chemiluminescence substrate (Invitrogen, cat. Nr 100002906), recorded with an Amersham Imager 600 (Amersham), and quantified by ImageJ (NIH). Activation of KDR was assessed by the ratio of phospho-to-total signal.
Tissue immunofluorescence
Immunofluorescence staining was performed to detect p-ERK and p-AKT. Formalin-fixed and paraffin-embedded tumors from Avatar models were cut into 3-mm-thick sections, deparaffinized, and preincubated with FBS to prevent nonspecific binding. The sections were incubated at room temperature for 30 min with a rabbit polyclonal antibody to p-ERK (1:300; Cell Signaling #9101) or a rabbit monoclonal antibody (D9E) to p-AKT (1:300, Cell Signaling #4060), followed by incubation with Alexa Fluor 555–conjugated donkey anti-rabbit IgG (1:400; Life Technologies#A27039) at 37°C for 20 min. Nuclei were counterstained with DAPI (Molecular Probes) at 1:1,000 dilution, and the slides were mounted with Mowiol 4-88 (Calbiochem). Images were acquired with a confocal TCS-SP5 (AOBS-UV) (Leica Microsystems) confocal microscope, equipped with a 20xHCX PL APO 0.7 N.A. objective.
Proliferation assays
Proliferation assays were performed using the CellTiter-Glo® Luminescent Cell Viability Assay (Promega). Briefly, cell lines were seeded in 96-well microtiter plates at a density of 10000 cells/well and were incubated for 24 h before adding the various drugs. A “mother plate” containing drugs at a concentration 200× higher than the final concentration to be used in the cell culture was prepared by serial dilutions of stock solutions of the drugs (10 mM) in DMSO. The appropriate volume from each drug (usually 2 μL) was added automatically (Beckman FX 96 tip) from this plate to the cell culture plate to reach the final concentration for each drug. Each concentration was assayed twice. The final concentration of DMSO in the tissue culture media did not exceed 1%. The cells were exposed to the drugs for 72 h and then analyzed using the CellTiter-Glo® Luminescent Cell Viability Assay (Promega). Cell proliferation values were plotted against drug concentrations and fitted to a sigmoid dose-response curve using the Activity base software from IDBS in order to calculate growth inhibition (GI50) values versus DMSO.
Cloning and mutagenesis
The L840F and eight additional KDR mutations of interest identified on cancer databanks, as well as the K868M kinase-dead mutation, were generated by site-directed mutagenesis of WT KDR/VEGFR2 cloned on the pBE vector, using the QuikChange Kit (Agilent, USA) and the primers described in Table S5. Mutations were confirmed by Sanger sequencing of the entire open reading frame.
Transfection and xenograft models
Colo-320 cell line-derived xenografts were generated from subcutaneous injections of 4 × 105 cells resuspended in phosphate-buffered saline (PBS) in four nude mice per genotype. Tumors were measured weekly and the animals were sacrificed within two months or when tumors reached the established humane endpoint. Mice injected with empty vector or the K868M kinase-dead mutant were kept alive and monitored weekly for four months.
Production of recombinant kinase domains of WT, L840F, and R1032Q VEGFR2
The kinase domains (residues 806-1171) of WT, L840F, and R1032Q VEGFR2 without the kinase insert domain (aa 940-989) were cloned, tagged with 6×His at their C-terminus, and expressed in the baculovirus-infected insect cell system. Proteins were purified by affinity chromatography on HisTrap columns, followed by size-exclusion chromatography on a HiLoad 16/600 Superdex 200 prep grade column (GE Healthcare, USA), using an ÄKTA system (GE Healthcare, USA). Fractions containing kinase domains were identified by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and concentrated by ultrafiltration up to 0.2 mg/ml. Protein mutations were confirmed by in-gel enzymatic digestion followed by liquid chromatography–mass spectrometry (LC-MS)/MS analysis.
Biochemical Assays
The kinase activity of recombinant WT, L840F, and R1032Q VEGFR-2 (KDR) kinase domains, as well as that of a commercially available recombinant KDR cytoplasmic domain (residues 789-1356) (PV3660; ThermoFisher) that was used as a positive control were analyzed using the LANCE®Ultra time-resolved fluorescence resonance energy transfer (TR-FRET) assay from Perkin Elmer according to the manufactures’ instructions.. Briefly, the enzymes were titrated starting from an initial concentration of 5 µg/ml and proceeding with 1:4 serial dilutions, and were added to the reaction buffer (15 mM HEPES pH 7.4, 20 mM NaCl, 1 mM EGTA, 0.02% Tween 20, 10 mM MgCl2, 0.1 mg/ml BGG, 2 mM DTT), containing 15 μM ATP and 200 nM UltralightTM-labeled Poly GT substrate in a total volume of 20 μl. The reaction was allowed to proceed in an Optiplate 384 from PerkinElmer for 60 min at room temperature. Reactions proceeded within the linear reaction time were then terminated by the addition of 20 mM EDTA and 4 nM Eu-W1024-labeled PY20 antibody. After an incubation of at least 60 min, the samples were excited with a Light Unit laser at 337 nm, and the emission of the LANCE Eu/APC (615/665 nm) was measured with an Envision reader (PerkinElmer). To test the effect of known VEGFR2 inhibitors on kinase activity, 0.3 ng WT VEGFR2 and 300 ng mutant VEGFR2 were used. The starting concentration of the inhibitors tested was 10 µM, followed by 1:5 serial dilutions. In order to calculate IC50 values of inhibition versus DMSO, the data were plotted against the inhibitor concentration and fitted to a sigmoid dose-response curve using the Activity base software from IDBS.
Immunohistochemistry
Avatar tumor samples were fixed in 10% neutral buffered formalin (4% formaldehyde in solution) and paraffin-embedded. Subsequently, 3-μm-thick sections were cut from the samples, mounted in superfrost®plus slides, and dried overnight. Before staining, the sections were deparaffinized in xylene and re-hydrated through a series of decreasing ethanol concentration in water. Consecutive sections were stained with hematoxylin and eosin (H&E) and by immunohistochemistry, using an automated immunostaining platform (Ventana Discovery XT, Roche or Autostainer Plus Link 48). Antigen retrieval was first performed with high or low pH buffer (CC1m, Roche), endogenous peroxidase was blocked (3% hydrogen peroxide), and the slides were incubated with an anti p-ERK rabbit polyclonal primary antibody (1:300; Cell Signaling #9101) for 28 min. Subsequently, the slides were incubated with the corresponding visualization system (OmniRabbit, Ventana, Roche) with signal amplification conjugated with horseradish peroxidase. The signal was developed using 3,30-diaminobenzidine tetrahydrochloride (DAB) as a chromogen (Chromomap DAB, Ventana, Roche or DAB solution, Dako), while the nuclei were counterstained with Carazzi’s hematoxylin. Finally, the slides were dehydrated, cleared, and mounted with a permanent mounting medium for microscopic evaluation. The entire slide was scanned with a slide scanner (Axio Z1, Zeiss), and images were captured with the ZEN software (Zeiss) after evaluation by a trained veterinary pathologist. Image analysis and quantification were performed using the AxioVision software package (Zeiss).
Kaplan-Meier analysis of mCRC patients
Kaplan-Meier survival data of 1,303 patients from four different cohorts of CRC patients with global gene expression and survival datasets available (GSE24551, GSE14333, GSE17538, GSE39582) were queried using the R2 microarray analysis and visualization platform (http://hgserver1.amc.nl/cgi-bin/r2/main.cgi). Concerning KDR gene expression in the combined datasets, 304 tumors had high levels of expression and 999 had low levels of expression.
Acknowledgments
We are grateful to Dr. Pedro P. López-Casas and Manuel Muñoz (CNIO Gastrointestinal Cancer Unit) for their valuable technical and administrative assistance; to Dr. Javier Muñoz and Eduardo Zarzuela (CNIO Proteomics Core Unit) for assistance with protein mutation analysis; to Dr. Diego Megías and Manuel Pérez (CNIO Confocal Microscopy Core Unit) for their assistance with the tissue immunofluorescence preparation; to Prof. Patricia L. Dahia (Department of Medicine of the University of Texas Health Science Center at San Antonio) and for Dr. Rodrigo Dienstmann (Vall d’Hebron Institute of Oncology, Barcelona, Spain) for her critical reading and suggestions on the manuscript; to Kurt Ballmer-Hofer (Paul Scherrer Institute, Switzerland) for his involvement at the beginning of the functional in vitro experiments. RA Toledo was a recipient of a research fellowship from the National Council for Scientific and Technological Development (CNPq). The authors would like to thank especially the patient and his family for their participation in the study.
Footnotes
Conflict of interest: The authors have no conflict of interest to declare.