Background
Gastrointestinal stromal tumor (GIST) is a rare cancer of mesenchymal origin, with an incidence rate of approximately 1 case/100,000/year [
1,
2]. Oncogenic activation of KIT or PDGFRA receptor tyrosine kinases (RTKs) is central to GIST biology, and are present in 85–90% of the patients [
3,
4]. Specifically, two thirds of GISTs harbor a wide array of primary mutations in
KIT juxtamembrane domain, encoded by exon 11. Similar complexity is found in other
KIT regions (exons 9, 13 and 17) [
5]. Likewise, mutually exclusive primary mutations in PDGFRA are found in homologous regions [
6]. Although most advanced GISTs respond to first-line inhibitor imatinib [
7], disease progression eventually occurs in 20–24 months after treatment initiation. Acquired resistance to imatinib is due in 70–90% of GIST patients to the expansion of subpopulations harboring different KIT secondary mutations [
8‐
10] that cluster in the ATP-binding pocket and the activation loop [
5,
8‐
10]. Resistance mechanisms after several lines of treatments are yet to be fully elucidated [
11].
Importantly, KIT/PDGFRA primary and secondary genotype is relevant for GIST clinical management because it predicts GIST clinical behavior and efficacy from tyrosine kinase inhibitors (TKIs) with KIT inhibitory activity in the first line [
12] – imatinib – and in any line of treatment after imatinib failure, including standard second- (sunitinib) and third-line treatments (regorafenib) [
13‐
17]. Therefore, detection and monitoring of GIST primary and resistance mutations in circulating tumor DNA (ctDNA) has the potential to improve molecular profiling, surveillance and treatment decision-making.
qPCR or digital PCR-based technologies have the highest analytical sensitivity for mutation detection [
18‐
20]. While PCR plasma genotyping is preferred for recurrent predictable aberrations, technologies based on next-generation sequencing (NGS) have the potential to asses more broadly the variety of primary and resistance mutations [
21‐
23]. Thus, the complexity and diversity of KIT primary and secondary mutations in imatinib-sensitive and –resistant patients favors the use of NGS over PCR for the detection of cfDNA mutations. NGS technologies employ various strategies for enriching specific target regions, and some of them are commercially available for their use in plasma [
24,
25]. By contrast, amplicon-based target enrichment, although less sensitive, has a widespread use in molecular screening programs using tumor tissue, and it is progressively emerging as an alternative approach for extensive cfDNA assessment [
26,
27]. This, in turn, would potentially facilitate the implementation of cfDNA evaluation in oncology centers with expertise in NGS.
Overall, there is an urgent need for real-time tumor biomarkers to guide therapy selection in GIST. Nevertheless, until ctDNA is proven to render the genomic information detected in solid tissue, it cannot replace the need for metastatic tissue biopsy of patients, nor guide clinical decisions [
28]. To address this, we orthogonally validated an amplicon-based NGS panel for routine molecular prescreening in a cohort of localized and advanced GIST patients with matched tissue and plasma samples, and in a second cohort with serial plasma determinations.
Methods
GIST patient cohorts
Localized and metastatic KIT- or PDGFRA-mutant GIST patients were prospectively enrolled in a tissue and plasma acquisition protocol. Consenting patients were distributed in two cohorts: cohort A (matched tissue/plasma) included localized or metastatic GIST patients with matched tissue and plasma samples obtained simultaneously. Localized patients were imatinib-naïve and tissue samples were obtained through surgical removal of the primary tumors. Tumor tissue in metastatic patients included resections of unifocal progressive disease. In all cases plasma samples were collected 7 to 14 days before tumor resection. Plasma samples from patients on TKI treatment were obtained while on drug. Cohort B (serial) included metastatic patients with serial plasma samples throughout the course of their treatment, together with tumor tissue at the time of diagnosis.
This study was approved by the Institutional Review Board from each participating center and written informed consent was obtained from all patients to donate blood samples and tumor tissue.
Blood sample collection and plasma processing
Peripheral blood was collected into EDTA tubes (Beckton Dickinson) and plasma was extracted within 4 h of blood collection through two centrifugation steps of 10 min each, the first at 1600 g and the second at 3000 g. Single-use 1.5 mL plasma aliquots were obtained and stored at − 80 °C until use. cfDNA was obtained from 3 mL of plasma using the QIAamp Circulating Nucleic Acids kit (Qiagen) and quantified with a Qubit Fluorometer (ThermoFisher Scientific).
Tumor tissue specimen collection and processing
Representative formalin fixed, paraffin-embedded (FFPE) tumor tissue blocks were retrieved from each case and reviewed by a GIST expert pathologist (S.L.). Five 10-μm tissue sections with more than 20% tumor area were obtained. DNA extraction was performed with the automated system Maxwell16 FFPE plus LEV DNA purification kit (Promega). DNA quality and concentration were measured with a NanoDrop 1000 spectrophotometer (Thermo Scientific, Waltham, MA).
In order to optimize VHIO amplicon-sequencing pipeline, seven additional FFPE primary tumor samples were retrieved from our GIST series database with known
KIT exon 11 long insertions and/or deletions (indels) (> 15 base pairs) through Sanger Sequencing [
29].
DNA from GIST cell lines
DNA from two human GIST cell lines with known long
KIT exon 11 in-frame deletions was also used to optimize the NGS pipeline for the detection of long indels (> 15 base pairs). GIST-T1 has deleted 19 aminoacids (
KIT exon 11 p.V560_Y578del), and GIST430 has 17 (
KIT exon 11 p.V560_ L576del) [
30].
Tumor and plasma mutational analysis by amplicon sequencing
An initial multiplex-PCR with a proof-reading polymerase was performed on all samples. Tumor and plasma DNA were sequenced with an in-house developed amplicon-sequencing panel of over 1330 primer pairs targeting frequent mutations in oncogenes and several tumor suppressors, totaling 60 genes (Additional file
1: Table S1) including KIT and PDGFRA, which contain reported hotspots for primary and secondary mutations in GIST [
31]. 500 ng of DNA from each tissue sample, or total cfDNA from 3 mL plasma samples were used for library preparation according to our established protocols. Duplicate chemistries were performed for each sample. Plasma analyses were carried out blinded to clinical information such as tumor genotype.
Amplicon sequencing was performed as previously described [
31‐
33]. Specifically for this study, indexed libraries were pooled and sequenced in a HiSeq 2500 instrument (2X100) at an average coverage of 1000x for tissues samples and 5000x for plasmas. Variants were called using VarScan2 (v2.3.9) with the following parameters: minimum variant allele frequency (VAF) of 3% for FFPE samples and 1% for plasma samples; total coverage ≥10 reads; variant coverage ≥7 reads, and a
p-value < 0.05. Germline mutations were manually excluded.
Recent studies revealed shortcomings of state-of-the-art variant callers that might fail to detect complex indels [
34]. For this reason, an alternative pipeline for the detection of large indels in
KIT exon 11 was also used. Filtered reads were re-mapped using bwa with relaxed values in parameters for long gaps: increasing maximum number of gap extensions (−e) and the maximum occurrences for extending a long deletion (−d), and decreasing the penalties for opening (−O) or extending a gap (−E). The resulting SAM file was used as input for the indel caller SOAPindel.
Droplet digital PCR (ddPCR)
The QX200 ddPCR System (Bio-Rad Laboratories, Hercules, CA) was used to confirm in plasma the presence of variants detected in tumor tissue and plasma by amplicon sequencing. Genomic DNA from tumor tissue from the same patient was run as positive control. Custom Taqman SNP genotyping assays for ddPCR were designed to detect KIT and PDGFRA mutations (Additional file
2: Table S2). 1.5 mL of plasma was used for ddPCR validations.
Briefly, the 20 μL final volume of TaqMan PCR reaction mixture was assembled with 1x ddPCR Supermix for Probes (no dUTP), 900 nM of each primer, 250 nM of each probe and 8 μL of cfDNA or 30 ng per reaction in FFPE (positive controls). Each assay was performed in triplicate in separate mixes and loaded in different wells for amplification. The thermal cycling program was performed according to specifications of the manufacturer. After PCR, droplets were read in the Droplet Reader and analyzed with QuantaSoft version 1.7.4. Human reference genomic DNA was included as negative control and used to determine the cutoff for allele calling in each assay. ddPCR validations were carried out blinded to tumor genotype information.
Statistical analysis
Descriptive statistics were used to characterize patients at study entry. Fisher’s exact test, Mantel-Haenzel test and Mann-Whitney test were used, depending on each variable type, to determine the association between clinicopathological and molecular features with detection of ctDNA. Concordance of VAFs between plasma ddPCR and NGS was calculated using Pearson Correlation. All statistical tests were conducted at a 2-sided significance level of 0.05. Analyses were performed using GraphPad Prism 6 (GraphPad Software, Inc., La Jolla, CA) or IBM SPSS Statistics 20.0 (Chicago, IL).
Discussion
ctDNA evaluation might impact particularly in tumor types associated with recurrent driver genetic events, such as GIST, a rare neoplasm of mesenchymal origin whose course of disease is governed by KIT or PDGFRA oncogenic activation [
5]. The diversity of primary and secondary mutations across well-known exonic regions of the
KIT gene [
3‐
6,
8‐
10,
12,
16] positions NGS as a more suitable technology than digital PCR for exhaustive evaluation of driver and resistance mutations in plasma. Nonetheless, strict criteria must be followed prior to the implementation of liquid biopsy into the clinic [
28].
We investigated the validity and utility of amplicon-based plasma NGS to detect molecular alterations in GIST patients. To this purpose, matched tumor tissue was collected at the time of plasma sampling or at diagnosis and served as the reference standard. ddPCR was used for orthogonal validation of mutations found by NGS in both tumor tissue and plasma. Our amplicon pipeline was also improved for the detection of long indels in KIT, which is a common challenge across NGS-based technologies. The overall sensitivity for detection of tumor tissue mutations in cfDNA was 28.6%, showing high concordance with ddPCR in the confirmation analyses. No KIT or PDGFRA primary mutations were detected in the 5 localized GIST. Conversely, amplicon-sequencing detected cfDNA mutations in 38% metastatic GIST patients. Few case series and reports have addressed the role of ctDNA in GIST. This evidence shows that mutant KIT and PDGFRA can be detected and quantified in plasma of GIST patients, also with a predilection for patients with high tumor burden [
35‐
37]. However, our results differ in some regards from these studies. Prior NGS-based analyses detected ctDNA in 17–70% localized, and in 100% metastatic GIST patients [
36‐
39]. Conversely, NGS detection rate in our population was lower. Even the more sensitive and specific ddPCR technology only reached an overall sensitivity of 42.9% in our series, failing to detect mutations in localized GIST. Several factors may have accounted for these disparities. First, we applied a stringent criteria for NGS variant calling AF at ≥1%, also increasing the average coverage for plasma samples to 5000x. Consequently, we validated that our panel was robust to detect cfDNA mutations at AFs ≥1%, while most of the discrepancies observed between amplicon sequencing and ddPCR were shown at low AFs. These validations involved orthogonal NGS of tumor tissue, as the reference standard, and plasma variants cross-validation with ddPCR, thus following recent ASCO Guidelines recommendations [
28]. Second, prior NGS studies in GIST did not incorporate variant calling algorithms for variants at low AF (< 5%) and mostly relied on manual inspection of raw data based on mutational findings in tumor tissue. This method is biased, since it enhances ctDNA detection. Moreover, ctDNA findings were not validated with a different technology. Therefore, this approach, although feasible, lacks clinical utility because it is not systematic to be implemented in the routine clinical care. Third, there is no established optimal lower limit of detection of ctDNA, and it varies depending on each assay and its intended use. Nonetheless, several studies have recently shown that the lower the variant AFs (< 1%), the lower the concordance between plasma and tissue genotyping and the higher the rate of discrepancies among NGS platforms [
24,
40,
41].
ctDNA was found in a low proportion of GIST patients (27.8%) and at low AF (6.2%, range 1–14%) compared to the majority of neoplasms [
21‐
24,
42]. These findings are unexpected since the bulk of disease in metastatic GIST patients is usually higher than in other cancer types, as reflected by a median tumor burden of 15.2 cm in our series. Accordingly, prior data in GIST have also reported low AF of mutations found in plasma, which is also in line with the scarce works studying ctDNA in sarcomas [
42,
43] and further supports the sensitivity reached by NGS in our series. Thus, this collective evidence indicates that ctDNA shedding appears to be low in malignant mesenchymal neoplasms. Although inter-studies comparisons are challenging, the proportion of metastatic GIST patients with ctDNA detected by NGS or ddPCR lies in the medium-to-low range compared with other epithelial neoplasms analyzed with several high sensitive techniques, including NGS [
21]. Thus, intrinsic GIST biological characteristics might condition a lower ctDNA shedding than expected.
We found amplicon sequencing of ctDNA informative in a subset of GIST, mainly in metastatic, progressive disease after imatinib failure. Therefore it has the potential to avoid tumor biopsies when tumor genotyping is required. Additionally, serial ctDNA assessment reproduces the course of the disease and provides information on subclonal dynamics. Notably, we confirmed that monitoring of known KIT or PDGFRA mutations in plasma with ddPCR is useful in a bigger subset of GIST patients, and that might predict tumor progression before radiological evaluation. Nonetheless, NGS of plasma advantages digital PCR-based technologies in the detection of the higher variety of mutations found in GIST. This is non-trivial, since KIT secondary genotype predicts response to TKIs after imatinib failure [
15‐
17,
30], and therefore, serial plasma determination of cfDNA mutations help to guide treatment decisions in GIST patients. For instance, NGS of plasma in patient 18 adds further evidence supporting that regorafenib is predominantly active against secondary mutations in the activation loop [
15,
30], and suggests that regorafenib dose is critical for the effective suppression of resistant subclones.
Unlike prior reports, an important focus of our studies was on imatinib-resistance disease, with 11 out of 18 patients in this setting. We did not identify substantial heterogeneity of KIT secondary mutations neither in plasma nor in tumor tissue, which agrees with previous PCR-based studies [
9,
10,
16,
44] and more recent plasma NGS reports [
36,
37,
39]. Likewise, we did not observe either enrichment in KIT-downstream molecules as a resistance mechanism to TKIs with KIT inhibitory activity [
11], although phenotypical changes were shown in three patients with matched NGS of tissue and plasma. This, in turn, highlights unknown KIT-independent underlying mechanisms of resistance not captured with the NGS panel. Thus, the only mechanism of resistance identified in our series in GIST patients progressing to sunitinib or regorafenib consisted on secondary mutations in the KIT activation loop. These data will need to be verified in further series.
The main limitation from our study is the cohort size: despite exhaustive inter-platform analysis and cross-validations, these low numbers cannot capture the biological complexity of this disease from paucisymptomatic localized tumors to TKI-refractory disease. This limitation, also affecting prior publications in GIST and sarcomas [
35,
36,
38,
39,
42,
43,
45‐
47], would benefit from international consortiums delving deeper the clinical utility of ctDNA in malignant mesenchymal neoplasms. Likewise, the role of other circulating markers [
48] or epigenetic biomarkers [
49] with potential role in tumor diagnosis, monitoring and response evaluation is yet to be defined in GIST.
Novel ultra-deep NGS assays for plasma sequencing have the potential to detect a wider array of mutations at lower AFs, and therefore, to provide more thorough information regarding monitoring and determination of resistance mechanisms. However, the aforementioned challenges with variants at low AF (< 1%) are yet to be technically addressed in the forthcoming years [
24,
40,
41], particularly in a disease like GIST with low ctDNA shedding and AF. Amplicon sequencing of plasma with high coverage (5000x), when correctly validated for detection of cfDNA mutations, has the advantages to detect robustly plasma mutations at AFs ≥ 1%, less expensively, and with the potential to be successfully implemented in a higher number of oncology centers with expertise in NGS, given the widespread use of amplicon-based NGS platforms in molecular prescreening programs. Although likely less sensitive than other approaches, recent studies support its use for ctDNA determination [
26,
27].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.