Blood samples were collected from patients who were diagnosed with GISTs and seen at the National Cancer Centre Singapore between April 1998 and September 2021. A total of 64 plasma samples and 7 whole blood samples from 46 patients were included in the final analysis. Relevant demographical and clinical information were collected and utilized for the analysis. For all GISTs diagnosed at our center, Sanger sequencing was routinely performed to detect KIT and PDGFRA gene mutations. Selected cases may undergo panel testing via next-generation sequencing at the discretion of the managing physician. All data were obtained at the time of diagnosis or subsequent follow-up. Written informed consent for use of biospecimens and clinical data was obtained in accordance with the Declaration of Helsinki. The research study was carried out with approval from the SingHealth Centralised Institutional Review Board (CIRB 2018/3182). The datasets created and analysed during this study are available from the corresponding authors upon reasonable request.

Whole blood samples were collected in 10-ml EDTA-coated tubes (BD, Cat. 368589) and processed within 2 h of collection using an in-house protocol. The samples were centrifuged for 10 min at 300 × g at room temperature to separate plasma from red blood cells. Plasma layer was harvested in 1-ml aliquots in 1.5-ml Microcentrifuge Tubes (Axygen, MCT-150-C) and spun again in a microcentrifuge at 9,720 × g at 4°C. Plasma was collected and stored at −80°C until use. QIAamp Circulating Nucleic Acid Kit (Catalog #55114, Qiagen) was used to isolate ctDNA from plasma by following the manufacturer’s instruction. Purified ctDNA was stored at −20°C.

The following is the protocol for extracting ctDNA from frozen whole blood. DNA were isolated by using DNeasy Blood & Tissue Kit (Qiagen) with the following modification to the protocol. Briefly, proteinase K was added to whole blood at the following volume ratio (proteinase K:whole blood, 1:10). Subsequently, Buffer AL (with no ethanol added) was added into thawed blood, containing proteinase K, at a volume ratio of 6:1 before incubating at 56°C for at least 10 min. Absolute ethanol was added at a volume that was equal to that of Buffer AL used. The number of DNA purification columns used was dependent on the volume of whole blood. The ratio was 1 column per 200 µl of thawed whole blood. Subsequent steps followed the manufacturer’s instruction.

Once eluted, DNA was obtained from whole blood, Mag-Bind^® Total Pure NGS (Omega Bio-tek) paramagnetic beads were used to separate the genomic DNA from ctDNA; 0.6× bead volume (calculation was based on the volume of the DNA eluent) was added to the DNA eluent, and the beads were used to remove the genomic DNA. The beads containing the genomic DNA was isolated from the supernatant by a magnetic column. Subsequently, the leftover supernatant was moved to a clean microfuge tube where 2× bead volume (calculation was based on the volume of the original DNA eluent) was added. This volume of beads would isolate out the ctDNA from the supernatant. The beads were isolated by magnet, washed, and ctDNA was eluted according to the manufacturer’s instructions. The profile of the eluted ctDNA fraction was visualized on Agilent TapeStation by using Agilent genomic DNA ScreenTape (Agilent Technologies, CA, USA) to ensure that there was no contaminating genomic DNA. If genomic DNA was still present in the eluted ctDNA fraction, the DNA clean-up would be repeated. Purified ctDNA, from plasma or whole blood, were quantified by using Qubit dsDNA assay kit (Thermo Fisher Scientific). Purified ctDNA profile was also visualized on Agilent TapeStation by using Agilent High Sensitivity D100 ScreenTape.

NGS libraries were made from, depending on individual sample ctDNA yield, 5 ng to 37 ng (median: 10.1 ng) of ctDNA input. The NGS libraries were constructed by following the Archer^® LiquidPlex™ Protocol for Illumina^® (Invitae). We are using a customized Archer^® LiquidPlex™ targeted panel, dubbed LiquidPlex NCCS GIST 18265 v1.0, that consists of probes that target specific exons of 29 genes that are known to associate with cancer and encompass known somatic mutations of GISTs ( Supplementary Table 1 ). The libraries were paired-end sequenced (2 × 150 bp) for 5–20 million raw reads with at least 5% PhiX by using the standard Illumina NGS protocol on the NovaSeq platform (Novogene).

The raw sequencing data were analyzed with Archer Analysis pipeline (version 6.2.7; https://archerdx.com/technology-platform/analysis) as previously reported (8). We used the default settings for detecting variants that were statistically significant [in our case, the variant must have an allele frequency (AF) outlier p-value < 0.05]. For ctDNA samples where no significant variants were detected by the default setting, we would manually check all the mapped NGS reads for the presence of KIT, PDGFRA, or other variants that were previously detected by Sanger sequencing of the corresponding tumor. If such variant was detected, it would only be reported as a significant variant if AF outlier p-value < 0.05.

Comparisons of tumor size with detectable ctDNA were performed using Mann–Whitney U test. All statistical evaluations were made assuming a two-sided test with a significance level of 0.05 unless otherwise stated. All tests were performed using MedCalc statistical Software for Windows version 19.0.4 (MedCalc Software, Ostend, Belgium).

A total of 46 patients were included in the study ( Supplementary Figure 1 ). The median age at study enrolment was 63.7 years (range, 30.3 to 89.1 years). Twenty-six (56.5%) were male and 20 (43.5%) were female. Primary tumor locations were stomach (n = 24, 52.2%), small bowel (n = 19, 41.3%), and rectum (n = 3, 6.5%). The majority were known to harbor KIT mutations (n = 41, 89.1%), while 3 patients (NCCS-GIST-06, NCCS-GIST-43, and NCCS-GIST-44) were PDGFRA exon 18 D842V mutants and KIT wild type. Two cases (NCCS-GIST-45 and NCCS-GIST-46) were wild type for both KIT and PDGFRA, one of which harbored KRAS exon2 G12V mutation. In terms of disease stage, 14 (30.4%) were localized GISTs that had undergone complete surgical resection. The rest (n = 32) were metastatic GISTs at various points of their treatment trajectories. Importantly, 10 patients, including 7 on TKI treatment, had evidence of disease progression at study inclusion and were the focus of the study. Clinical and demographic characteristics of all patients are summarized in Table 1 .

Blood samples of the 46 patients were drawn at study inclusion. Targeted exon panel sequencing was performed to identify mutations in plasma ctDNA. Among the 10 patients with metastatic GIST with evidence of disease progression, mutations in ctDNA were detected in 7 cases (70%). Known somatic mutations in KIT (n = 5) or PDGFRA (n = 1) in ctDNA were identified only among 6 of the 10 patients ( Figure 1 ). These KIT mutants included short sequence tandem duplication, indels, and single-nucleotide variants. The median mutant AF in ctDNA was 11.0% (range 0.38%–45.0%). In patients with metastatic progressive KIT-mutant GIST, tumor burden (as measured by the average diameter of the 3 largest lesions) was higher with detectable KIT ctDNA mutation than in those without (median, 5.97 cm vs. 2.40 cm, p = 0.0195) ( Figure 2 ). The 4 cases with undetectable primary KIT mutations include GISTs with exon 11 c.1708_1728del (NCCS-GIST-07), exon 11 c.1669_1674del (NCCS-GIST-08), exon 11 c.1654_1659del and exon 13 c.1961T>C (NCCS-GIST-09), and exon 9 c.1504_1509dup (NCCS-GIST-10). None of the known tumor mutations were detected in ctDNA for localized cases (n = 14) or metastatic cases without evidence of disease progression (n = 22). In patient NCCS-GIST-02, only the known KIT exon 11 c.1727T>C mutation, but not the exon 17 c.2466T>A mutation was detected in ctDNA. On the other hand, in patient NCCS-GIST-03, ctDNA identified KIT exon 17 c.2467T>G on top of the known exon 11 c.1653_1670del mutation—this patient had prior imatinib-resistant GIST (no molecular evaluation done apart from time of diagnosis) and was progressing despite the 4th-line treatment with pazopanib. For patient NCCS-GIST-07 with small bowel GIST progressing on 1st-line treatment with imatinib, ctDNA detected TP53 c.879_880del and SETD2 c.2238del mutations, but not the known KIT exon 11 c.1708_1728del mutation ( Table 2 ).

Mutations in SETD2, previously reported to confer worse prognosis in GIST (9), were detected in ctDNA in 3 patients with metastatic GIST. In patients with localized GIST following surgical resection (n = 14), including 5 on adjuvant imatinib, no known tumor mutations could be detected from ctDNA. Likewise, none could be detected from ctDNA from patients with metastatic GIST without evidence of disease progression (n = 22), including 15 patients on TKI treatment. In this group of patients, mutations in IDH2 (n = 3), KRAS (n =3), KIT (n = 3), and SETD2 (n = 1) were identified at low AF (median, 0.86%; range, 0.68 to 2.9%), though the corresponding mutation in the primary tumor was not known.

Serial plasma samples were available for 6 patients with metastatic GIST who underwent TKI therapy. For NCCS-GIST-17 ( Figure 3A ), the KIT exon 11 c.1673_1687del variant, as detected in the tumor, was not detected in the ctDNA from the first plasma sampling when no disease progression was evident (stable disease, SD) (week 0). In the ctDNA from the second plasma sampling at the time of disease progression in the primary stomach tumor (week 26), 2 KIT variants were detected, namely, exon 11 c.1673_1687del (AF = 3.1%) and exon 17 c.2485G>C (AF = 3.4%), the latter of which was a new variant not previously detected in the tumor. Additionally, a PIK3CA c.1633G>A variant was also detected (AF = 2.6%). Interestingly, in the ctDNA from the third plasma sampling (week 42, with progressive liver metastases), the PIK3CA c.1633G>A variant AF increased by nearly 32-fold (AF = 82.5%) while those KIT variants were reduced by at least 14.2-fold (exon 11 c.1673_1687del, AF = 0.15%; exon 17 c.2485G>C, AF = 0.24%). Generally, the detection of the KIT and PIK3CA variants in the ctDNA positively correlated with disease progression as evaluated by computed tomography (CT) scans.

For patients NCCS-GIST-04, NCCS-GIST-02, and NCCS-GIST-09, multiple plasma samplings at different time points (4–7 samples) have allowed for the monitoring of dynamic changes to the different KIT variants detected in the ctDNA in response to different types of TKI used. In NCCS-GIST-04 ( Figure 3B ) and NCCS-GIST-02 ( Figure 3C ), KIT variants detected in ctDNA progressively decreased and were not detected in ctDNA when patients were responsive to TKI treatments. However, we observed that after a period of positive drug response, known KIT variants re-emerged in ctDNA with additional new variants that were not previously detected in tumor. The emergence of additional new KIT variant correlated with resistance to corresponding TKI treatment and disease progression.

Patient NCCS-GIST-04 switched from imatinib to ripretinib (at week 0) at the time of disease progression based on imaging assessment ( Figure 3B ). Imaging at week 7 confirmed response to ripretinib therapy—correspondingly, the AF for the KIT exon 11 c.1669_1710del variant fell from 3.68% at week 0 to 0% at week 7. Subsequently from week 7 to week 19, no KIT variant was detected in the ctDNA, and this coincided with a state of continued disease stability for the patient. At week 25, disease progression was detected on CT scan, and this coincided with the re-emergence of the KIT variant exon 11 c.1669_1710del (AF = 1.1%) and the detection of a new KIT variant exon 17 c.2467T>G (AF = 0.59%). At week 43, the AF values for both KIT variants were still detectable, albeit reduced (exon 11 c.1669_1710del, AF = 0.16%; exon 17 c.2467T>G, AF = 0.18%). This coincided with the patient switching to sunitinib. Despite the reduction in AF values for the 2 KIT variants, the CT scan showed that the disease had progressed with increased tumor size.

On the other hand, NCCS-GIST-02 switched from avapritinib to imatinib and subsequently ripretinib as a result of lack of response in terms of tumor reduction in the first 8 weeks of monitoring; the KIT variants from exon 11 c.1727T>C and exon 17 c.2466T>A became undetectable from week 10 to week 16 ( Figure 3C ), correlating with positive response to ripretinib treatment. Subsequently, ctDNA collected from week 28 and week 33 contain not only the previously detected KIT variants from exon 11 c.1727T>C and exon 17 c.2466T>A, but also a new exon 13 c.1961T>C variant. This indicated the emergence of a novel TKI-resistant variant that correlated with disease progression. Conversely, for NCCS-GIST-09 ( Figure 3E ), despite disease progression, no KIT variant, as previously detected in tumor, was detected in ctDNA collected at the different time points. There was, however, a KRAS c.194G>A variant that was detected at week 41 albeit at low AF (0.84%), but it was not detected at the last sampling at week 48.

There were 2 patients with 2 serial samplings, namely, NCCS-GIST-03 and NCCS-GIST-08. For NCCS-GIST-03, 2 KIT variants were detected ( Figure 3D ). The exon 11 c.1653_1670del variant was previously detected in tumor, while the exon 17 c.2467T>G variant was only found in plasma ctDNA. Although the variant AF for both variants decreased from week 0 to week 8, it did not correlate with decrease in tumor size. For NCCS-GIST-08 ( Figure 3F ), the KIT variant previously reported in tumor was not detected in ctDNA from both samplings (week 0 and week 20). A KRAS c.194G>A variant was detected, albeit at low AF (0.8%), in the week 20 sampling.

In an exploratory analysis, we attempted to detect known variants (previously found in the tumor) from ctDNA extracted from whole blood rather than from plasma. ctDNA was extracted from whole blood from 7 patients with metastatic GIST ( Supplementary Table 2 ). NGS libraries were subsequently constructed from these extracted ctDNA. We compared the result from whole blood with those corresponding plasma-derived ctDNA as well as tumor. KIT variant was detected in 1 whole blood sample from NCCS-GIST-02. The KIT variant, exon 11 c.1727T>C, was the same as that reported for the corresponding plasma-derived ctDNA and tumor. The AF for this KIT variant in whole blood-derived ctDNA was 1.61%. This AF value was approximately 1/12 of the AF value found in plasma ctDNA (20.3%).