Purpose: Despite the wide use of antiangiogenic drugs in the clinical setting, predictive biomarkers of response to these drugs are still unknown.

Experimental Design: 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 inhibitors to the VEGF:VEGFR2 pathway. We performed extensive functional experiments, including ectopic expression of VEGFR2 mutants in different cell lines, kinase and drug sensitivity assays, and cell- and patient-derived xenografts.

Results: WES-cfDNA yielded a 77% concordance rate with tumor exome sequencing and enabled the identification of the KDR/VEGFR2 L840F clonal, somatic mutation as the cause of therapy refractoriness in our patient. In addition, we found that 1% to 3% of samples from cancer sequencing projects harbor KDR somatic mutations located in protein residues frequently mutated in other cancer-relevant kinases, such as EGFR, ABL1, and ALK. Our in vitro and in vivo functional assays confirmed that L840F causes strong resistance to antiangiogenic drugs, whereas the KDR hot-spot mutant R1032Q confers sensitivity to strong VEGFR2 inhibitors. Moreover, we showed that the D717V, G800D, G800R, L840F, G843D, S925F, R1022Q, R1032Q, and S1100F VEGFR2 mutants promote tumor growth in mice.

Conclusions: Our study supports WES-cfDNA as a powerful platform for portraying the somatic mutation landscape of cancer and discovery of new resistance mechanisms to cancer therapies. Importantly, we discovered that VEGFR2 is somatically mutated across tumor types and that VEGFR2 mutants can be oncogenic and control sensitivity/resistance to antiangiogenic drugs. Clin Cancer Res; 24(15); 3550–9. ©2018 AACR.

This article is featured in Highlights of This Issue, p. 3475

Translational Relevance

Our work illustrates the high capacity of whole-exome sequencing of plasma cfDNA (WES-cfDNA) in portraying the somatic mutation landscape in cancer and its potential as a global tumor-free genomic platform to explore the genetic causes of primary resistance to cancer therapies. The implementation of WES-cfDNA would enable discovery studies on cancer patients for whom only blood/plasma samples are available, as well as serve as a complementary assay to the standard WES-tumor by increasing the overall capacity for identifying somatic mutations and molecular targets. As a proof of concept, WES-cfDNA identified the L840F VEGFR2 clonal, somatic mutation in a highly refractory metastatic colorectal cancer patient. Our recognition of KDR/VEGFR2 somatic mutations, occurring in 1% to 3% of human cancers, as oncogenic and capable of modulating the efficacy of antiangiogenic cancer therapies in vitro and in vivo will prompt future investigation to define the potential impact of these mutants in the clinical setting.

Colorectal cancer is currently the third most frequent cancer diagnosed worldwide and is predicted to reach 2.2 million new cases per year by 2030 (1). Although bevacizumab treatment (anti-VEGF) increases the overall survival of patients with metastatic colorectal cancer (mCRC), approximately 50% and 80% of patients in first and second lines of treatment respectively are refractory and do not benefit from this therapy strategy (2, 3). The identification of predictive biomarkers of the response to bevacizumab, as well as to new agents targeting VEGFR2, such as regorafenib, is still a fundamental unmet medical necessity (4).

Genetic variants of the VEGF:VEGFR1/2 pathway could influence the outcome of antiangiogenic treatment. Although some studies have suggested the potential association of tumor response with VEGF/VEGFR germline polymorphisms (5), these results could not be confirmed in subsequent assessments (6). Interestingly, recent in vitro studies demonstrated that VEGFR2 plays a prominent role not only in endothelial cells, as is usually assumed, but also in cancer cells (7). However, it is largely unknown whether clinically relevant KDR/VEGFR2 mutations occur in tumor cells and potentially regulate drug efficiency.

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). The DNA amount was quantified with a Qubit Fluorometer (Thermo Fisher) and reported in nanogram. 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 were sheared to 300-bp fragments on a Covaris instrument (Covaris) according to standard procedures. The 2100 Bioanalyzer (Agilent) was used to access the quality and size of the preprocessed and postprocessed samples and libraries.

Routine genetic analysis

The FDA-approved Cobas mutation Kit (Roche) 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, and 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 (8).

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.), 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) 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 noninherited subclonal 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) (9). 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 nonhereditary 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), allele frequencies in population studies, such as 1000G and EXAC, and additional information are shown in Supplementary 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 Thermo Fisher online Design Tool (oligonucleotides and probes are shown in Supplementary Table S2).

Genetic/protein database and protein structure analyses

Previously reported germline and somatic variants in KDR were retrieved from general population (EXACT and ESP) and cancer (COSMIC, GENIE, and PCAWGS) 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 MutationAligner; kinase residues interacting with kinase inhibitors were mapped using the LigPlot 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 model

Liver metastasis biopsy was performed after tumor progression to capecitabine–bevacizumab rechallenge (Fig. 1). A fraction of the biopsy was used to generate the Avatar model as previously described by our group (10). 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, lenvatinib, 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 Supplementary Table S3.

Figure 1.

Discovery of the KDR/VEGFR2 L840F somatic mutation in a refractory mCRC patient. A, Thoracic computed tomography scans of the refractory mCRC patient. Imaging examinations show no response to any treatment and disease progression to FOLFIRI-cetuximab, FOLFOX-bevacizumab, afatinib-cetuximab, capecitabine-bevacizumab, oncolytic adenovirus and regorafenib. The time points of sample collection are pointed with colored arrows: red for diagnostic biopsy, blue for DNA sequencing, and green for Avatar generation and tumor DNA sequencing. Metastatic foci are indicated by black arrows. B, Noninvasive strategy for the discovery of primary therapy-resistance gene(s). Genetic and genomic analyses were performed according to the DNA sample availability. First, tumor biopsy was used (and exhausted) on routine genetic analysis that revealed no mutations in the main resistance-associated genes KRAS/NRAS/BRAF/PIK3CA. Second, whole-exome sequencing was performed using basal plasma cfDNA (to obtain both germline and cancer genetic variants/mutations) and matched leukocyte DNA (to filter out germline variants), and identified the KDR/VEGFR2 L840F somatic mutation. Third, the mutation was confirmed by Sanger and TaqMan assays in pre- and postplasma cfDNA samples and in a second biopsy. A fragment of the biopsy was used to generated the patient-derived xenograft (PDX) Avatar model and for tumor DNA exome sequencing.

Figure 1.

Discovery of the KDR/VEGFR2 L840F somatic mutation in a refractory mCRC patient. A, Thoracic computed tomography scans of the refractory mCRC patient. Imaging examinations show no response to any treatment and disease progression to FOLFIRI-cetuximab, FOLFOX-bevacizumab, afatinib-cetuximab, capecitabine-bevacizumab, oncolytic adenovirus and regorafenib. The time points of sample collection are pointed with colored arrows: red for diagnostic biopsy, blue for DNA sequencing, and green for Avatar generation and tumor DNA sequencing. Metastatic foci are indicated by black arrows. B, Noninvasive strategy for the discovery of primary therapy-resistance gene(s). Genetic and genomic analyses were performed according to the DNA sample availability. First, tumor biopsy was used (and exhausted) on routine genetic analysis that revealed no mutations in the main resistance-associated genes KRAS/NRAS/BRAF/PIK3CA. Second, whole-exome sequencing was performed using basal plasma cfDNA (to obtain both germline and cancer genetic variants/mutations) and matched leukocyte DNA (to filter out germline variants), and identified the KDR/VEGFR2 L840F somatic mutation. Third, the mutation was confirmed by Sanger and TaqMan assays in pre- and postplasma cfDNA samples and in a second biopsy. A fragment of the biopsy was used to generated the patient-derived xenograft (PDX) Avatar model and for tumor DNA exome sequencing.

Close modal

Cell lines

The human colorectal cancer 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 colorectal cancer 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 the ATCC and cultured at 37°C in 5% CO2, in RPMI Medium 1640 + GlutaMAX (Gibco) and DMEM + 2 mmol/L Glutamine (Gibco), respectively, supplemented with 10% FBS (Thermo Fisher Scientific). 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. 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 (EV) or VEGFR2 (WT or mutant) was performed with polyethylenimine (PEI). 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 minutes at room temperature and added to the cells. Following a 3-hour incubation at 37°C, the medium was changed, and the cells were allowed to grow to 100% confluence. Cells were reseeded at a series of dilutions (1:1,000–1:5,000) in antibiotic selection medium (1 mg/mL G418) to allow for single colonies to grow, whereas nontransfected 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 three 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% BSA for 4 hours at 37°C and were subsequently stimulated with 1.5 nmol/L (60 ng/mL) VEGF165 for 10 minutes at 37°C. Following stimulation, the cells were scraped in lysis buffer (50 mmol/L Tris, pH = 8.0, 120 mmol/L NaCl, and 1% NP-40) supplemented with protease inhibitors (Roche; cat. Nr 04693159001) and phosphatase inhibitors (1 mmol/L sodium orthovanadate and 20 μmol/L phernylarsine oxide) and incubated for 30 minutes on ice. Cell lysates were collected as the supernatant of a centrifugation at 30,000 × g for 15 minutes and subjected to Western blot analysis. The following antibodies were used to probe receptor activation: total KDR (Cell Signaling Technology, cat. Nr 2479) and phospho KDR at Y1175 (Cell Signaling Technology, cat. Nr 2478). The secondary antibodies used were alkaline phosphatase (AP) conjugated (Southern Biotech). All antibodies were diluted at a 1:1,000 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 (GE Healthcare), 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 minutes with a rabbit polyclonal antibody to p-ERK (1:300; Cell Signaling Technology; cat. Nr 9101) or a rabbit monoclonal antibody (D9E) to p-AKT (1:300; Cell Signaling Technology; cat. Nr 4060), followed by incubation with Alexa Fluor 555–conjugated donkey anti-rabbit IgG (1:400; Life Technologies cat. Nr A27039) at 37°C for 20 minutes. 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 10,000 cells/well and were incubated for 24 hours 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 mmol/L) 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 hours 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) and the primers described in Supplementary Table S4. 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 PBS in 4 nude mice per genotype. Tumors were measured weekly, and the animals were sacrificed within 2 months or when tumors reached the established humane endpoint. Mice injected with EV or the K868M kinase-dead mutant were kept alive and monitored weekly for 4 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), using an ÄKTA system (GE Healthcare). Fractions containing kinase domains were identified by SDS-PAGE and concentrated by ultrafiltration up to 0.2 mg/mL. Protein mutations were confirmed by in-gel enzymatic digestion followed by LC-MS/MS analysis.

Biochemical assays

The kinase activity of recombinant WT, L840F, and R1032Q VEGFR2 kinase domains, as well as that of a commercially available recombinant VEGFR2 cytoplasmic domain (residues 789–1356; PV3660; Thermo Fisher) that was used as a positive control were analyzed using the LANCE Ultra time-resolved fluorescence resonance energy transfer assay from Perkin Elmer according to the manufacturer's 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 mmol/L HEPES pH 7.4, 20 mmol/L NaCl, 1 mmol/L EGTA, 0.02% Tween 20, 10 mmol/L MgCl2, 0.1 mg/mL BGG, 2 mmol/L DTT), containing 15 μmol/L ATP and 200 nmol/L Ultralight-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 minutes at room temperature. Reactions proceeded within the linear reaction time were then terminated by the addition of 20 mmol/L EDTA and 4 nmol/L Eu-W1024–labeled PY20 antibody. After an incubation of at least 60 minutes, 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 μmol/L, 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 rehydrated through a series of decreasing ethanol concentration in water. Consecutive sections were stained with hematoxylin and eosin 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 Technology; cat. Nr 9101) for 28 minutes. 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), whereas 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).

Clinical

Case report.

A 56-year-old man was diagnosed in our center with mCRC with liver and lung metastases (cT4N2M1). Highly sensitive Cobas and BEAMing assays were performed, and no hot-spots mutations were identified in the KRAS/NRAS/PIK3CA/BRAF genes in the patient's cfDNA, nor in tumor biopsy samples. He received FOLFIRI-cetuximab as frontline treatment, but there was evidence of tumor progression after the first tumor evaluation. Consequently, he received five additional lines of treatment: FOLFOX-bevacizumab; afatinib with cetuximab; oncolytic adenovirus monotherapy (the last two lines of treatment in the context of phase I clinical trials); rechallenge with capecitabine-bevacizumab; and finally, regorafenib. However, there was no disease stability or response to any treatment, and persistent growth of his liver tumor burden was observed. The patient died within a short time period (14 months) after the initial diagnosis due to complications of his progressive disease (Fig. 1A).

Genomic analyses

Discovery of the clonal, KDR/VEGFR2 L840F somatic mutation by WES-cfDNA.

WES-cfDNA confirmed the WT status of the KRAS/NRAS/PIK3CA/BRAF genes and uncovered two known colorectal cancer driver mutations, APC c.3964G>T E1322X (COSM18702) and TP53 c.659A>C Y220S (COSM43850; Supplementary Figs. S1 and S2). Interestingly, we identified the novel KDR c.2518C>T mutation leading to the VEGFR2 L840F mutation at the protein level (Fig. 1B). The mutation was confirmed 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 (Fig. 1B; Supplementary Fig. S2). Importantly, the mutated allele frequencies (MAF) of the KDR/VEGFR2 mutation in the tumor were similar to those of trunk colorectal cancer mutations, such as those in APC and TP53 (approximately 30%, 50%, and 50%, respectively). The MAFs of KDR/VEGFR2, APC, and TP53 mutations in plasma were 11%, 8%, and 18%, respectively.

High concordance between WES-cfDNA and WES-tumor.

The concordance between WES-cfDNA and WES performed in tDNA (WES-tumor) obtained after a second liver metastasis was 73% (54/74 somatic mutation, Fig. 2A). Importantly, WES-cfDNA was able to detect a variety of mutation types, such as frameshift (including insertions and deletions), missense, noncoding (splicing), and nonsense mutations. In addition, WES-cfDNA discovered 14 high-confidence somatic mutations not identified by WES-tumor. Overall, in the absence of a tumor specimen, WES-cfDNA could identify 68 of the 88 (77.3%) total somatic mutations identified by both techniques. The complete list of the identified somatic mutations, and all the sequencing parameters and genomic annotation are depicted in Supplementary Table S1. High concordance between WES-cfDNA and WES-tumor was also observed in copy-number variation analysis (Fig. 2B).

Figure 2.

Concordance between the genomic landscape identified by whole-exome sequencing of plasma cfDNA and tumor; DNA and recurrence of KDR/VEGFR2 oncogenic mutations in human cancers. A, The histograms represent all the genes with somatic mutations identified by plasma cfDNA whole-exome sequencing (in red) and by tumor whole-exome sequencing (in blue). The list and genomic annotation of all mutations are shown in SupplementaryTable S1. The mutated allele frequencies (MAFs) of each mutation in relation to all reads are depicted in Y. B, Copy number alteration landscape portrayed by the whole-exome sequencing of tumor DNA (upper part) and pre-treatment plasma cfDNA (lower part). Gains are depicted in green, losses in red, and normal (balanced) in grey. C, Frequency of KDR somatic mutations in large cancer genomic sequencing projects including more than 70,000 cancer samples. Common germline polymorphisms were excluded from these analyses and only cancer-exclusive mutations were considered. D, Representative results from the search of genomics and protein databases, showing several mutations, structurally analogous to those of KDR, identified in other cancer-relevant kinases.

Figure 2.

Concordance between the genomic landscape identified by whole-exome sequencing of plasma cfDNA and tumor; DNA and recurrence of KDR/VEGFR2 oncogenic mutations in human cancers. A, The histograms represent all the genes with somatic mutations identified by plasma cfDNA whole-exome sequencing (in red) and by tumor whole-exome sequencing (in blue). The list and genomic annotation of all mutations are shown in SupplementaryTable S1. The mutated allele frequencies (MAFs) of each mutation in relation to all reads are depicted in Y. B, Copy number alteration landscape portrayed by the whole-exome sequencing of tumor DNA (upper part) and pre-treatment plasma cfDNA (lower part). Gains are depicted in green, losses in red, and normal (balanced) in grey. C, Frequency of KDR somatic mutations in large cancer genomic sequencing projects including more than 70,000 cancer samples. Common germline polymorphisms were excluded from these analyses and only cancer-exclusive mutations were considered. D, Representative results from the search of genomics and protein databases, showing several mutations, structurally analogous to those of KDR, identified in other cancer-relevant kinases.

Close modal

KDR/VEGFR2 somatic mutations are recurrent in cancer.

KDR bona-fide cancer mutations (polymorphisms excluded) were found in all publicly available cancer databases as follows: COSMIC v80 (691/33,320; 1.6%); GENIE (1,012/38,207; 2.6%); PCAWG (4/35; 11.5%); and MSK-IMPACT (266/10,336; 2.6%; Fig. 2C). Importantly, 15 of the 35 (43%) colorectal cancer samples in the PCAWGS study had KDR gene amplification. KDR somatic mutations found in the cancer genomics databases are shown in Supplementary Table S5.

KDR/VEGFR2 somatic mutations are analogous to known cancer-related kinase mutations.

KDR/VEGFR2 cancer mutations occur in hot-spot residues analogous to those of other kinases related to human cancers. For example, VEGFR2 L1049W is analogous to EGFR L858, and VEGFR2 D1052N/G/H is analogous to FLT3 D835, KIT D816, EGFR L861, and PDGFRA D842. Importantly, we found that R1032Q (COSM192176) is the most frequent VEGFR2 mutation in the cancer databases and possibly a mutational hot-spot of VEGFR2 because it is analogous to a number of known cancer mutations, such as 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, and ROR1 R619C/H/S (Fig. 2D).

Functional studies of the VEGFR2 cancer mutants

VEGFR2 L840F causes cancer therapy resistance.

In agreement with the phenotype we observed in our 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 (Fig. 3A). 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 (Fig. 3B; Supplementary Fig. S3). 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. 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 (Supplementary Fig. S4A–S4B). Consistent with these results, Y1175 phosphorylation of L840F VEGFR2 was significantly reduced compared with 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 (Supplementary Fig. S4C–S4D). Our kinase activity assays showed that whereas WT VEGFR2 was sensitive to axitinib, cabozantinib, dovitinib, and lenvatinib, L840F VEGFR2 was resistant to all these drugs, albeit at distinct levels (Fig. 3C).

Figure 3.

The VEGFR2 L840F mutant leads to broad and strong cancer therapy resistance and R1032Q mutant to increased sensitivity to VEGFR2 inhibitors. A, Growth inhibition of the patient-derived xenograft (PDX) Avatar model carrying the KDR/VEGFR2 L840F (red) after 3 weeks of treatment with anti-VEGF drugs (B20/murine and bevacizumab/human), VEGFR2 kinase inhibitors (axitinib, cabozantinib, cabozantinib:MEK inhibitor combo, pazopanib, regorafenib, sorafenib), or inhibitors of other kinases, such as afatinib (EGFR), crizotinib (MET), and MEK inhibitor (MEKi). A second CRC PDX model, carrying the KDR/VEGFR2 WT (blue), was used as a positive control, and was treated with some of the above drugs. To analyze the inhibition of tumor growth promoted by each tested drug, tumor volumes of the untreated mice were set as 100% growth and used as reference for the measurement of the treated animals (tumor volume of treated mice divided by tumor volume of untreated mice). B, Localization of structurally analogous L residue mutations in VEGFR2 (L840), EGFR (L718), and ABL1 (L248). A close-up view of the entrance of the ATP-binding pocket domain is shown in green for VEGFR2 L840F, in light blue for ABL1 L248V, and in orange for EGFR L718Q/R. Patients with these mutations are all refractory to treatment with tyrosine kinase inhibitors, which directly bind to these L residues (see also Supplementary Fig. S3). C, Inhibition of kinase activity of WT (black), L840F (red), and R1032Q (blue) VEGFR2 kinase domains by VEGFR2 inhibitors axitinib, cabozantinib, dovitinib, and lenvatinib. WT VEGFR2 kinase was sensitive to the four inhibitors, especially to axitinib and cabozantinib. The concentration of the mutants was increased 1,000 times to achieve a measurable kinase activity, which was consequently measured in the presence of TKIs.

Figure 3.

The VEGFR2 L840F mutant leads to broad and strong cancer therapy resistance and R1032Q mutant to increased sensitivity to VEGFR2 inhibitors. A, Growth inhibition of the patient-derived xenograft (PDX) Avatar model carrying the KDR/VEGFR2 L840F (red) after 3 weeks of treatment with anti-VEGF drugs (B20/murine and bevacizumab/human), VEGFR2 kinase inhibitors (axitinib, cabozantinib, cabozantinib:MEK inhibitor combo, pazopanib, regorafenib, sorafenib), or inhibitors of other kinases, such as afatinib (EGFR), crizotinib (MET), and MEK inhibitor (MEKi). A second CRC PDX model, carrying the KDR/VEGFR2 WT (blue), was used as a positive control, and was treated with some of the above drugs. To analyze the inhibition of tumor growth promoted by each tested drug, tumor volumes of the untreated mice were set as 100% growth and used as reference for the measurement of the treated animals (tumor volume of treated mice divided by tumor volume of untreated mice). B, Localization of structurally analogous L residue mutations in VEGFR2 (L840), EGFR (L718), and ABL1 (L248). A close-up view of the entrance of the ATP-binding pocket domain is shown in green for VEGFR2 L840F, in light blue for ABL1 L248V, and in orange for EGFR L718Q/R. Patients with these mutations are all refractory to treatment with tyrosine kinase inhibitors, which directly bind to these L residues (see also Supplementary Fig. S3). C, Inhibition of kinase activity of WT (black), L840F (red), and R1032Q (blue) VEGFR2 kinase domains by VEGFR2 inhibitors axitinib, cabozantinib, dovitinib, and lenvatinib. WT VEGFR2 kinase was sensitive to the four inhibitors, especially to axitinib and cabozantinib. The concentration of the mutants was increased 1,000 times to achieve a measurable kinase activity, which was consequently measured in the presence of TKIs.

Close modal

VEGFR2 R1032Q confers sensitivity to strong VEGFR2 inhibitors.

Kinase assay showed that similar to the L840F, the R1032Q mutation greatly reduced VEGFR2 kinase activity (Supplementary Fig. S4B), although the molecular mechanism underlying this phenotype would be distinct. As R1032Q affects the universal kinase catalytic motif DxxxxN (aa1028-DxxxRN-1033aa; ref. 11), and not the ATP-binding site as the L840F, we investigated whether R1032Q VEGFR2 would be inhibited by TKIs. In vitro kinase assays showed increased sensitivity of R1032Q VEGFR2 to TKIs (Fig. 3C; Supplementary Fig. S5A). 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 with 36.4 for WT VEGFR2] and to cabozantinib (GI50 = 2.5 for the R1032Q compared with 7.9 for WT VEGFR2; Supplementary Fig. S5B). Moreover, cabozantinib treatment of the MDST8 colorectal cancer 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. Importantly, we found that such downstream inhibition was specific to cabozantinib, a very strong VEGFR2 (0.035 nmol/L) and c-MET (1.3 nmol/L) inhibitor, and occurred in cells treated in the absence or presence of VEGF (Supplementary Fig. S5A).

VEGFR2 cancer–related mutants are oncogenic.

We further asked if in addition to modulating the response to TKIs, VEGFR2 cancer mutants could promote tumor growth. Indeed, we showed that when Colo-320 cells stably expressing D717V, G800D, G800R, L840F, G843D, S925F, R1022Q, R1032Q, and S1100F VEGFR2, when injected in mice, even in small numbers and without Matrigel, could generate tumors that reached the established humane endpoint (Fig. 4). Importantly, Colo-320–expressing 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 4.

Oncogenic potential of VEGFR2 cancer mutants in xenograft assays. Colo320 CRC cell lines were used for xenograft studies, as they resemble the mutation profiling of the studied patient's tumor, with wild-type KRAS/NRAS/BRAF/PIK3CA status, and TP53 and APC mutations. Colo320 cells stably expressing different VEGFR2 mutants were injected subcutaneously in four immune deficient mice (each blue bar represents one mouse). Results of xenograft growth after 2 months following injections or when tumors reached the established humane endpoint are shown.

Figure 4.

Oncogenic potential of VEGFR2 cancer mutants in xenograft assays. Colo320 CRC cell lines were used for xenograft studies, as they resemble the mutation profiling of the studied patient's tumor, with wild-type KRAS/NRAS/BRAF/PIK3CA status, and TP53 and APC mutations. Colo320 cells stably expressing different VEGFR2 mutants were injected subcutaneously in four immune deficient mice (each blue bar represents one mouse). Results of xenograft growth after 2 months following injections or when tumors reached the established humane endpoint are shown.

Close modal

The high capacity of WES-cfDNA for portraying the somatic mutation and copy-number variation landscapes of tumors (77.3% of concordance rate in our case) shows an immense potential for research in translational oncology (12–14). We anticipate that the implementation of WES-cfDNA will enable blood-based global genomic profiling of cancers and greatly potentiate the discovery of new biomarkers. Moreover, it has the potential to be used in clonal evolution and tumor mutation burden analyses.

Applying WES-cfDNA, we identified the VEGFR2 L840F clonal, somatic mutation in a highly refractory mCRC patient. Our validation experiments with cell lines, animal models, and biochemical assays demonstrated that VEGFR2 mutants can modulate the response to antiangiogenic agents according to their localization and functional consequence. For example, the L840F ATP-binding pocket domain mutation causes very strong and broad resistance to anti-VEGF and VEGFR2 inhibitors, whereas the R1032Q kinase domain mutation is apparently sensitive to strong VEGFR2 inhibitors, such as cabozantinib and lenvatinib. These findings demonstrate for the first time that the same well-known genetic mechanisms leading to resistance/increase sensitivity to inhibitors of EGFR, ABL1, and PDGFRA receptors do also occur in VEGFR2 and can have implications for the outcome of antiangiogenic treatments (15).

In agreement with our preclinical experimental findings, Knepper and colleagues very recently reported a prolonged complete response to pazopanib in a metastatic basal cellular carcinoma patient carrying the KDR/VEGFR2 R1032Q somatic mutation (16). Interestingly, a second mCRC patient case with the KDR/VEGFR2 R961W somatic mutation was also recently reported. After progression to 5-fluorouracil-bevacizumab, the patient responded to low-dose regorafenib with remarkable regression of the hepatic metastases, abdominal and retroperitoneal lymph nodes, and rectosigmoid colon hypermetabolic lesions (17).

Another key point of our study relates to the emerging role of loss-of-function kinase mutations to tumorigenesis and cancer therapy modulation. We investigated extensively the activity of L840F and R1032Q VEGFR2, and the results from different and independent experiments clearly show that both mutants are kinase-dead. These findings are in line with previous work from Dr. Owen Samson's and Dr. John Brognard's groups showing that recurrent MLK4 and PKC mutations in colorectal cancer are both loss of function and oncogenic (18, 19). Importantly, recent studies demonstrated that BRAF-impaired cancer mutants could 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; ref. 20). The VEGFR2 L840F and R1032Q kinase-impairing mutations would fit well in such a “rewiring mutant kinase oncogenic model (Supplementary Fig. S6).” However, tumor growth of the AvatarVEGFR2:L840F model was not significantly reduced upon treatment with inhibitors of EGFR (afatinib), MET (crizotinib), and MAPK (MEKi; Fig. 3A) and the possible rewiring partners of VEGFR2 L840F remain unknown.

In summary, the current study highlights the capability of exomic sequencing of cfDNA from plasma of cancer patients as a powerful platform for portraying the somatic mutation landscape of cancer and discovering new mechanisms of resistance to cancer therapies. 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 characterized KDR/VEGFR2 somatic mutations as potential genetic biomarkers of response to antiangiogenic cancer therapies, and these findings may serve as reference for further studies on the topic.

No potential conflicts of interest were disclosed.

Conception and design: R.A. Toledo, E. Garralda, S. Perea, C. Blanco-Aparicio, A. Cubillo, J.L. Martínez-Torrecuadrada, M. Hidalgo

Development of methodology: R.A. Toledo, J. Monsech, Á. Otero, N. Baños, Y. Durán, F. Sarno, S. Perea, A. De Martino, A. Cubillo, O. Domínguez, M. Hidalgo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.A. Toledo, E. Garralda, M. Mitsi, J. Monsech, E. Vega, Á. Otero, M.I. Albarran, N. Baños, F. Sarno, T. Sanchez-Perez, S. Perea, R. Álvarez, A. Cubillo, O. Domínguez, J.L. Martínez-Torrecuadrada, M. Hidalgo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.A. Toledo, E. Garralda, M. Mitsi, T. Pons, M. Camacho-Artacho, S. Perea, A. De Martino, D. Lietha, C. Blanco-Aparicio, M. Hidalgo

Writing, review, and/or revision of the manuscript: R.A. Toledo, E. Garralda, M. Mitsi, T. Pons, S. Perea, R. Álvarez, A. De Martino, D. Lietha, C. Blanco-Aparicio, A. Cubillo, J.L. Martínez-Torrecuadrada, M. Hidalgo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.A. Toledo, E. Garralda, V. Bonilla, S. Perea

Study supervision: R.A. Toledo, E. Garralda, A. Cubillo, M. Hidalgo

R.A. Toledo holds a Miguel Servet-I research contract by Institute of Health “Carlos III” of the Ministry of Economy (CP17/00199) and Competitiveness and is supported by a Fundacíon Olga Torres emerging researcher grant. The authors are grateful to Dr. Pedro P. López-Casas and Manuel Muñoz (CNIO Gastrointestinal Cancer Unit) for their valuable technical and administrative assistance and to Kurt Ballmer-Hofer (Paul Scherrer Institute, Switzerland) for his involvement at the beginning of the functional in vitro experiments. We would like to thank especially the patient and his family for their participation in the study.

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.

1.
Arnold
M
,
Soerjomataram
I
,
Ferlay
J
,
Forman
D
. 
Global incidence of oesophageal cancer by histological subtype in 2012
.
Gut
2015
;
64
:
381
7
.
2.
Hurwitz
H
,
Fehrenbacher
L
,
Novotny
W
,
Cartwright
T
,
Hainsworth
J
,
Heim
W
, et al
Bevacizumab plus irinotecan, fluorouracil, and leucovorin for metastatic colorectal cancer
.
N Engl J Med
2004
;
350
:
2335
42
.
3.
Wagner
AD
,
Arnold
D
,
Grothey
AA
,
Haerting
J
,
Unverzagt
S.
Anti-angiogenic therapies for metastatic colorectal cancer
.
Cochrane Database Syst Rev
2009
;
1
:
1
75
.
4.
Matos
I
,
Elez
E
,
Capdevila
J
,
Tabernero
J
. 
Emerging tyrosine kinase inhibitors for the treatment of metastatic colorectal cancer
.
Expert Opin Emerg Drugs
2016
;
21
:
267
82
.
5.
Lambrechts
D
,
Claes
B
,
Delmar
P
,
Reumers
J
,
Mazzone
M
,
Yesilyurt
BT
, et al
VEGF pathway genetic variants as biomarkers of treatment outcome with bevacizumab: an analysis of data from the AViTA and AVOREN randomised trials
.
Lancet Oncol
2012
;
13
:
724
33
.
6.
Loupakis
F
,
Cremolini
C
,
Yang
D
,
Salvatore
L
,
Zhang
W
,
Wakatsuki
T
, et al
Prospective validation of candidate SNPs of VEGF/VEGFR pathway in metastatic colorectal cancer patients treated with first-line FOLFIRI plus bevacizumab
.
PLoS One
2013
;
8
:
e66774
.
7.
Chatterjee
S
,
Heukamp
LC
,
Siobal
M
,
Schöttle
J
,
Wieczorek
C
,
Peifer
M
, et al
Tumor VEGF:VEGFR2 autocrine feed-forward loop triggers angiogenesis in lung cancer
.
J Clin Invest
2013
;
123
:
1732
40
.
8.
Toledo
RA
,
Cubillo
A
,
Vega
E
,
Garralda
E
,
Alvarez
R
,
de la Varga
LU
, et al
Clinical validation of prospective liquid biopsy monitoring in patients with wild-type RAS metastatic colorectal cancer treated with FOLFIRI-cetuximab
.
Oncotarget
2017
;
8
:
35289
300
.
9.
Toledo
RA
,
Qin
Y
,
Cheng
ZM
,
Gao
Q
,
Iwata
S
,
Silva
GM
, et al
Recurrent mutations of chromatin-remodeling genes and kinase receptors in pheochromocytomas and paragangliomas
.
Clin Cancer Res
2016
;
22
:
2301
10
.
10.
Garralda
E
,
Paz
K
,
López-Casas
PP
,
Jones
S
,
Katz
A
,
Kann
LM
, et al
Integrated next-generation sequencing and avatar mouse models for personalized cancer treatment
.
Clin Cancer Res
2014
;
20
:
2476
84
.
11.
Hanks
SK
,
Hunter
T
. 
The eukaryotic protein kinase superfamily: kinase (catalytic) domain structure and classification
.
FASEB J
1995
;
9
:
576
96
.
12.
Murtaza
M
,
Dawson
SJ
,
Tsui
DW
,
Gale
D
,
Forshew
T
,
Piskorz
AM
, et al
Non-invasive analysis of acquired resistance to cancer therapy by sequencing of plasma DNA
.
Nature
2013
;
497
:
108
12
.
13.
Butler
TM
,
Johnson-Camacho
K
,
Peto
M
,
Wang
NJ
,
Macey
TA
,
Korkola
JE
, et al
Exome sequencing of cell-free DNA from metastatic cancer patients identifies clinically actionable mutations distinct from primary disease
.
PLoS One
2015
;
10
:
e0136407
.
14.
Adalsteinsson
VA
,
Ha
G
,
Freeman
SS
,
Choudhury
AD
,
Stover
DG
,
Parsons
HA
, et al
Scalable whole-exome sequencing of cell-free DNA reveals high concordance with metastatic tumors
.
Nat Commun
2017
;
8
:
1324
.
15.
Ercan
D
,
Choi
HG
,
Yun
CH
,
Capelletti
M
,
Xie
T
,
Eck
MJ
, et al
EGFR mutations and resistance to irreversible pyrimidine-based EGFR inhibitors
.
Clin Cancer Res
2015
;
21
:
3913
23
.
16.
Knepper
TC
,
Freeman
ML
,
Gibney
GT
,
McLeod
HL
,
Russell
JS
. 
Clinical response to pazopanib in a patient with KDR-mutated metastatic basal cell carcinoma
.
JAMA Dermatol
2017
;
153
:
607
9
.
17.
Loaiza-Bonilla
A
,
Jensen
CE
,
Shroff
S
,
Furth
E
,
Bonilla-Reyes
PA
,
Deik
AF
, et al
KDR mutation as a novel predictive biomarker of exceptional response to regorafenib in metastatic colorectal cancer
.
Cureus
2016
;
8
:
e478
.
18.
Marusiak
AA
,
Stephenson
NL
,
Baik
H
,
Trotter
EW
,
Li
Y
,
Blyth
K
, et al
Recurrent MLK4 loss-of-function mutations suppress JNK signaling to promote colon tumorigenesis
.
Cancer Res
2016
;
76
:
724
35
.
19.
Antal
CE
,
Hudson
AM
,
Kang
E
,
Zanca
C
,
Wirth
C
,
Stephenson
NL
, et al
Cancer-associated protein kinase C mutations reveal kinase's role as tumor suppressor
.
Cell
2015
;
160
:
489
502
.
20.
Sen
B
,
Peng
S
,
Tang
X
,
Erickson
HS
,
Galindo
H
,
Mazumdar
T
, et al
Kinase-impaired BRAF mutations in lung cancer confer sensitivity to dasatinib
.
Sci Transl Med
2012
;
4
:
136ra70
.

Supplementary data