Monitoring the Outcomes of Systemic Chemotherapy Including Immune Checkpoint Inhibitor for HER2-Positive Metastatic Gastric Cancer by Liquid Biopsy

Abstract

Purpose

For precision medicine, exploration and monitoring of molecular biomarkers are essential. However, in advanced gastric cancer (GC) with visceral lesions, an invasive procedure cannot be performed repeatedly for the follow-up of molecular biomarkers.

Materials and Methods

To verify the clinical implication of serial liquid biopsies targeting circulating tumor DNA (ctDNA) on treatment response, we conducted targeted deep sequencing for serially collected ctDNA of 15 HER2-positive metastatic GC patients treated with anti-PD-1 inhibitor in combination with standard systemic treatment.

Results

In the baseline ctDNAs, 14 patients (93%) harbored more than one genetic alteration. A number of mutations in well-known cancer-related genes, such as KRAS and PIK3CA, were identified. Copy number alterations were identified in eight GCs (53.3%), and amplification of the ERBB2 gene (6/15, 40.0%) was the most recurrent. When we calculated the mean variant allele frequency (VAF) of mutations in each ctDNA as the molecular tumor burden index (mTBI), the mTBI trend was largely consistent with the VAF profiles in both responder and non-responder groups. Notably, in the longitudinal analysis of ctDNA, mTBI provided 2–42 weeks (mean 13.4 weeks) lead time in the detection of disease progression compared to conventional follow-up with CT imaging.

Conclusion

Our data indicate that the serial genetic alteration profiling of ctDNA is feasible to predict treatment response in HER2-positive GC patients in a minimally invasive manner. Practically, ctDNA profiles are useful not only for the molecular diagnosis of GC but also for the selection of GC patients with poor prognosis for systemic treatment (ClinicalTrials.gov identifier: NCT02901301).

Graphical Abstract

INTRODUCTION

Gastric cancer (GC) occurs at a high incidence in Asia and is the third leading cause of cancer-related deaths worldwide.¹,² Genetic alterations are one of the crucial causative events in GC development. To comprehensively elucidate genetic alterations in GC, several research groups have recently analyzed GC genomes using whole genome or exome sequencing.³,⁴,⁵,⁶,⁷ They found recurrent mutations or copy number alterations (CNAs) of genes, such as TP53, PIK3CA, ARID1A, and ERBB2. Molecular profiling of GC genomes has revealed druggable targets, and based on this, various targeted agents and immune checkpoint inhibitors (ICIs) provide survival benefits for cancer patients. Until now, the molecular subtype feasible for targeted therapy has been the HER2-positive subgroup, which has unique biology and clinical behavior compared to HER2-negative GC patients. Recently, Epstein-Barr virus (EBV)-related or microsatellite instability-high (MSI-H)/mismatch repair-deficient (dMMR) subgroup showed the potential benefit with ICIs, such as anti PD-1/PD-L1 inhibitors, in advanced GC.

For precision medicine, exploration and monitoring of molecular biomarkers are essential. Most of tissue retrieval of GC is done through an esophagogastroduodenoscopic (EGD) biopsy. However, in advanced GC with visceral lesions or previously gastrectomized patients, an invasive procedure cannot be performed repeatedly for the follow-up of molecular characteristics. More importantly, tissue biopsy has a severe limitation in view of the pronounced genetic and phenotypic heterogeneity of tumor tissues.⁸ Considering that most of the molecular targeted therapy or immunotherapy-based phase III clinical trials have failed, possibly due to the genetically and histologically heterogeneous characteristics of GC,³ a less invasive technique capable of both capturing tumor heterogeneity and overcoming the difficulty in tissue sampling during the course of therapy is needed.

Circulating tumor DNA (ctDNA) is a small fragment of DNA released from cells undergoing apoptosis or necrosis in tumor tissues.⁹ Recent studies have shown that genetic alterations in tumor tissues, including driver events, can be detected by liquid biopsy of ctDNA.¹⁰,¹¹,¹² Moreover, the clinical use of analytical tests to assess genetic alterations in ctDNA from cancer patients is increasing, especially in clinical trials.¹³

The first-line anti-human epidermal growth factor receptor 2 (HER2) antibody trastuzumab in combination with cytotoxic chemotherapy is the standard treatment for HER2-positive GC patients, as recommended by the Trastuzumab for Gastric Cancer study.¹⁴ However, as most patients develop intrinsic or acquired drug resistance within a year, recent prospective phase II studies have shown the efficacy of combining anti-PD-1 inhibitors with first-line trastuzumab and chemotherapy for HER2-positive GC.¹⁵,¹⁶

To verify the clinical implications of serial liquid biopsies targeting ctDNA in patients with advanced GC on systemic chemotherapy, we conducted an exploratory study by serial collection of ctDNA from treatment-naïve, recurrent, or metastatic HER2-positive advanced GC patients treated with anti-PD-1 inhibitor (pembrolizumab) in combination with standard chemotherapy.

MATERIALS AND METHODS

Patients and samples

In this study, ctDNA was serially collected from a subgroup of patients included in the prospective open-label phase Ib/II study of first-line pembrolizumab in combination with trastuzumab, capecitabine, and cisplatin in HER2-positive GC (ClinicalTrials.gov identifier: NCT02901301).¹⁶ The patients received 200 mg pembrolizumab and 6 mg/kg (after 8 mg/kg load) trastuzumab [Herzuma^® (CT-P6), Celltrion Inc.] intravenously on day 1, 1000 mg/m² capecitabine orally twice daily on days 1–14, and 80 mg/m² cisplatin intravenously on day 1 every 3 weeks. The trial was conducted in accordance with the Declaration of Helsinki and the Guidelines for Good Clinical Practice. The trial protocol was approved by the Institutional Review Board of Severance Hospital (Seoul, Korea; IRB no. 4-2016-0190), and all patients provided written informed consent before enrollment. Clinical responses of all patients were evaluated according to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1.¹⁷ For tumor burden analysis, the long diameters (or short diameters of the lymph nodes) of all measurable lesions were used.

Baseline tumor tissues were obtained before treatment initiation. HER2-positive was defined as either immunohistochemistry (IHC) 3+ or IHC 2+ in combination with in situ hybridization (ISH) +, as assessed by a local laboratory in primary or metastatic tumor tissues. To assess the HER2 status, anti-HER2/neu antibody (Clone 4B5; Ventana Medical Systems, Oro Valley, AZ, USA) was used for IHC, and the HER2 expression scoring system was applied according to the guidelines.¹⁸ In addition, in case of HER2 2+ by IHC, HER2 DNA amplification was evaluated by the silver-enhanced ISH (SISH) method. SISH was performed using INFORM^® HER2 DNA and chromosome 17 (CEP17) probes (Ventana Medical Systems) using a Ventana Benchmark XT automated staining system according to the manufacturer’s instructions. HER2 DNA amplification was defined as a HER2/CEP17 ratio ≥2.0.

MSI/dMMR status was determined by IHC for MLH1, MSH2, MSH6, and PMS2 in formalin-fixed, paraffin-embedded tissue sections. PD-L1 status was determined using the Dako PD-L1 IHC 22C3 pharmDx kit (Agilent Technologies, Santa Clara, CA, USA) according to the manufacturer’s instructions. PD-L1 expression was determined using the combined positive score, which is the number of PD-L1 stained cells (tumor cells, lymphocytes, and macrophages) divided by the total number of viable tumor cells, multiplied by 100. EBV status was determined using EBV-encoded small RNA (EBER) ISH.¹⁹

Blood samples were collected at three different time points for each patient: before treatment initiation (baseline), at cycle 1 day 14, and at cycle 2 day 1 (or at cycle 3 day 1). One or two additional blood samples were also collected during the follow-up according to the clinical courses of individual patients, including at the time of progressive disease.

Extraction of ctDNA

A 10-mL aliquot of whole blood collected in an EDTA tube was centrifuged at 1600×g for 10 min within 2 hours. The separated plasma was transferred to a dry tube and followed by a second centrifugation at 4000×g for 10 min. DNA extraction was performed using the QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany) using 1 mL of plasma.

Targeted next-generation sequencing (NGS) and identification of somatic variants

We performed targeted NGS by previously described methods.⁷ In brief, custom NGS panels, OncoChase-AS and OncoChase-HS targeting 95 and 157 cancer-related genes, respectively (ConnectaGen, Seoul, Korea), were used to generate sequencing libraries (Supplementary Table 1, only online). For OncoChase-AS, sequencing libraries were generated using the AmpliSeq Library Kit 2.0 (Thermo Fisher Scientific, Waltham, MA, USA) and sequenced using the Ion S5 system (Thermo Fisher Scientific) according to the manufacturer’s instructions. Sequencing reads were aligned to UCSC hg19, and genomic variants were called using Torrent Suite v5.6. For OncoChase-HS, sequencing libraries were generated using the SureSelect XTHS target enrichment system (Agilent Technologies) and then sequenced using the HiSeq2000 platform (Illumina, San Diego, CA, USA) according to the manufacturer’s instructions. Sequencing reads were aligned to UCSC hg19, and genomic variants were called using the Agilent SureCall software v4.1.1.5. The ANNOVAR package²⁰ was used to select somatic variants located in the exonic sequences and to predict their functional consequences. Stringent post-filtering processes were conducted to ensure reliable and robust mutation calling. First, known polymorphic sites for East Asians (>0.1% of minor allele frequency) in public databases (dbSNP137, ESP6500, and the 1000 genomes project) were filtered out as germline polymorphisms. Subsequently, the variant with a total read depth <50 or variant support read depth <3 was filtered. The remaining variants were considered candidate somatic mutations. In addition, somatic variants detected in ctDNA prior to chemotherapy were re-examined for the presence of sequencing reads supporting the corresponding variants. The number of sequencing reads for serial ctDNAs in each case was extracted using a bam-readcount (https://github.com/genome/bam-readcount). If the number of sequencing reads supporting the corresponding variants was three or more, the variant was manually rescued. In each ctDNA sample, the molecular tumor burden index (mTBI) was analyzed using the mean variant allele frequency (VAF) of the detected mutations.²¹ ΔmTBI was calculated based on the mTBI of the ctDNA sample at baseline.

DNA copy-number analysis

DNA CNAs were estimated using the targeted NGS data. The multiscale reference module and rank segmentation statistical algorithm in NEXUS software v10.0 (Biodiscovery, El Segundo, CA, USA) were used to define the CNAs of each sample. Targeted NGS data of normal circulating DNA from 20 healthy donors were used as the pooled normal reference. Segments were classified as copy number gains and losses when the log2 ratio was greater than 0.25 and less than -0.25, respectively. All identified CNA events were manually curated based on the sequencing depth ratio.

RESULTS

Patient characteristics

Fifteen HER-2 positive GC patients were included in this exploratory study between February 2017 and August 2020. The median follow-up duration was 19.3 months. The median age of patients was 61 years (range, 34–79 years), and the majority were male (73.3%). As expected, none of the patients had EBV-positive or MMR-deficient GC (Table 1). Blood samples were collected from all patients at three different time points: before treatment initiation (baseline), at cycle 1 day 14, and at cycle 2 day 1 (or at cycle 3 day 1) to evaluate the earlier change. During follow-up, one or two additional blood samples were also collected according to the clinical courses of the individual patient for longitudinal analyses, which included blood samples at a time point near the disease progression for patients GC08 to GC15. The time points of each blood sampling for ctDNA extraction are depicted in Fig. 1.

Table 1. Clinicopathologic Characteristics of 15 GC Patients at Baseline.

Patient ID	Sex/age	Pathology	Metastatic organ	HER2§	EBV*	MSI type†	PD-L1‡	Best response	PFS
GC01	M/34	APD	LN	3+	Negative	pMMR	0%	CR	42.8¶
GC02	M/60	AMD	Peritoneum, LN	2+ (2.9)	Negative	pMMR	0%	PR	30.8¶
GC03	F/54	AMD	Liver, LN	3+	Negative	pMMR	0%	PR	21.9
GC04	M/61	AMD	Peritoneum, LN	3+	Negative	pMMR	0%	SD	35.2¶
GC05	F/63	AMD	LN	2+ (3.7)	Negative	pMMR	<1%	CR	32.4¶
GC06	F/56	SRC	Peritoneum, LN, bladder	2+ (>2.0)	N/A	N/A	15%	CR	32.2¶
GC07	M/73	AMD	Peritoneum, LN	3+	Negative	pMMR	0%	PR	12.4
GC08	M/57	APD	Peritoneum, LN	3+	Negative	pMMR	<5%	PR	7.7
GC09	M/46	APD+SRC	LN	3+	Negative	pMMR	0%	PR	5.6
GC10	M/59	APD	LN	2+ (2.05)	Negative	pMMR	<5%	PR	9.8
GC11	M/62	AMD	Liver, lung	3+	Negative	pMMR	10%	PR	6.2
GC12	M/64	AMD	Bone, lung	2+ (2.05)	Negative	pMMR	5%	PR	5.5
GC13	F/77	AWD	Liver, LN	3+	Negative	pMMR	5%	PR	11.5
GC14	M/69	AMD	Lung, peritoneum, liver, LN	3+	Negative	pMMR	0%	PR	6.2
GC15	M/79	AMD	Adrenal gland, LN	3+	Negative	pMMR	5%	PR	2.7

GC, gastric cancer; EBV, Epstein-Barr virus; MSI, microsatellite instability; M, male; F, female; APD, adenocarcinoma poorly differentiated; AMD, adenocarcinoma moderately differentiated; SRC, signet ring cell; AWD, adenocarcinoma well differentiated; LN, lymph node; PFS, progression-free survival (months); N/A, not available; pMMR, proficient mismatch repair genes.

^*EBV-associated type by EBER in situ hybridization; ^†Mismatch repair deficiency type by immunohistochemistry; ^‡Immunohistochemical stain results (combined positive score) using PD-L1 22C3 antibody; ^§HER2-positivity by immunohistochemistry with HER2:CEP17 ratio by in situ hybridization for HER 2+ patients; ^¶Finished 2-year treatment without progression.

Fig. 1. Swimmer plot of the treatment and blood sampling time points. Each lane represents a single patient’s treatment history of first-line therapy. X-axis represents treatment duration. Yellow triangles represent the time point of blood sampling for ctDNA extraction. GC, gastric cancer; ctDNA, circulating tumor DNA.

Genetic alterations in the baseline ctDNA samples

We first analyzed the genetic alteration profiles of baseline ctDNA samples from 15 GC patients using targeted deep sequencing. The mean sequencing depth was 3027× (range, 91× to 5443×) across the entire target genome (Supplementary Table 2, only online). The majority of baseline ctDNA of GCs (14/15, 93.3%) had non-silent mutations in at least one target gene, including well-known cancer-related genes such as KRAS (p.G12D), PIK3CA (p.E542K), KDM6A (p.Y1208C), ERBB3 (p.G284R), EGFR (c.1881-2A>G), and KIT (p.K642Q) (Fig. 2 and Supplementary Table 3, only online). Of the mutated genes, five were recurrently detected across multiple ctDNA samples (>2), which was consistent with previous reports³,⁵,⁶: TP53 (n=9, 60.0%), AR (n=3, 20.0%), APC (n=2, 13.3%), ARID1A (n=2, 13.3%), and RET (n=2, 13.3%) (Fig. 2). No ERBB2 mutations were observed in the 15 GCs.

Fig. 2. Genetic alterations identified in baseline ctDNA from 15 GC patients. Seventeen genes with 26 mutations and 11 copy number alterations are shown. Upper panel represents the frequency of genetic alterations for each patient. Right-hand panel represents the number of genetic alterations for each gene. Asterisks represent the reported variants in the COSMIC database. ctDNA, circulating tumor DNA; GC, gastric cancer.

When we analyzed CNA profiles, eight of the 15 GCs (53.3%) harbored at least one CNA (median of 1 CNA, range 0–2), whereas the other seven cases did not have any (Supplementary Table 4, only online). As expected, the most recurrent CNA was an amplification of the ERBB2 gene (6/15, 40.0%) (Supplementary Fig. 1, only online). Among the six patients with ERBB2 amplification, five were HER2 IHC 3+ and one patient was HER2 2+ with SISH+. CCNE1 amplification was detected in two cases, while other CNAs did not recur (MYC amplification: GC01, KRAS amplification: GC08, and FGFR2 amplification: GC09). Together, 14 of the 15 baseline ctDNAs harbored mutations and/or CNAs (8 cases harbored both mutation and CNA and 6 cases harbored only mutation), and one was silent.

Monitoring the genetic alteration profiles after treatment

We next analyzed the genetic alteration profiles of serial time point ctDNA samples aftertreatment of the 15 GC patients (52 ctDNAs, mean sequencing depth: 3181×, range 369× to 6513×) to explore the clinical implication of serial liquid biopsies on ICI treatments. For this, the 15 GC patients were divided into two groups based on their clinical response to the treatment. Seven patients (GC01–GC07) whose progression-free survival (PFS) was >12 months (six patients showed PFS >20 months) were categorized into the responder group, while eight patients (GC08–GC15) whose PFS was <12 months (six patients showed PFS <8 months) were categorized into the non-responder group. One patient in the responder group (GC04) did not show any genetic alterations and was, therefore, excluded from this analysis.

In all patients except one (GC05) from the responder group, baseline levels of VAF of detected mutations and CNAs decreased dramatically within 2 weeks after treatment and remained up to 70 weeks (Supplementary Fig. 2A, only online). In the GC05 case, VAFs decreased 2 weeks after treatment, bounced back in the 4th week, and then gradually diminished until the 12th week. In contrast, the baseline levels of VAF and CNAs in the non-responder group decreased at 2 weeks after the first treatment and then gradually increased until the diagnosis of clinical disease progression (Supplementary Fig. 2B, only online). No further genetic alterations were found in any patient post-treatment compared to their pre-treatment status (Supplementary Table 3, only online), suggesting a low possibility of acquiring further genetic alterations following systemic treatment.

We next calculated the mean VAF of detected mutations in each ctDNA as mTBI, and the mTBI trend was largely coherent with the VAF profiles of detected mutations in both responder and non-responder groups (Fig. 3). In addition, the average mTBI at first blood sampling after treatment was significantly lower than that of baseline in both responder (p=0.031) and non-responder groups (p=0.008) (Supplementary Fig. 3, only online). Of note, the average mTBI at the last blood sampling after treatment was significantly higher than that of first blood sampling in the non-responder group (p=0.016), but not in the responder groups (p>0.05) (Supplementary Fig. 3, only online). In the non-responder group, mTBI provided 2–42 weeks (mean 13.4 weeks) lead time in the detection of disease progression compared to conventional follow-up with CT imaging (Fig. 3B).

Fig. 3. Serial comparison of tumor burden and mTBI from ctDNA. mTBI at baseline were compared serially with tumor burden for each patient, except one patient GC04 who did not harbor any genetic alterations. Green line and black dotted-line represent the ΔmTBI and linear trend in ΔmTBI, respectively. The ΔmTBI was calculated based on the mTBI of the ctDNA sample at baseline. Longitudinal analyses of ctDNA were performed and analyzed among two groups according to their response pattern to the treatment: (A) responder group (GC01–GC07) and (B) non-responder group (GC08–GC15). mTBI, molecular tumor burden index; AU, arbitrary unit; ctDNA, circulating tumor DNA.

Representative clinical cases of the responder and non-responder groups

In the case of GC01 (responder group), the baseline levels of VAFs and CNAs were not detected at 2 weeks after initial treatment and remained unchanged until the 70th week, where routine follow-up with CT imaging showed no metastatic lymph node progression (Fig. 4A). Conversely, for GC13 (non-responder group), AR mutation detected at baseline was not detected in the 3rd week after treatment but rebounded the VAF from the 4th week (Fig. 4B). Although the CT image revealed partial response to the treatment at the 35th week, ctDNA sampled at the 35th week showed much higher VAF of AR mutation than that sampled at the 3rd and 4th weeks. Disease progression in this patient was finally confirmed by CT imaging at the 49th week (Fig. 4B). Interestingly, no increase in mean VAFs and CNAs for GC14 (non-responder group) was observed during longitudinal ctDNA analysis until the diagnosis of disease progression (Fig. 4C). In this case, the metastatic lesions were stable after the treatments, but only the primary tumor in the stomach progressed (cancer invasion progression up to the esophagogastric junction on EGD), implying that ctDNA mainly originated from metastatic tumors rather than primary tumor.

Fig. 4. Representative clinical cases for responders and non-responders. Representative case of responder group (A) and non-responder group (B) from ctDNA in comparison with tumor burden and CT imaging. (C) An interesting case of primary tumor progression in the stomach (inlet) without increasing genetic alterations in ctDNA is shown. ctDNA, circulating tumor DNA; GC, gastric cancer; VAF, variant allele frequency; CN, copy number; AU, arbitrary unit; RECIST, Response Evaluation Criteria in Solid Tumor.

DISCUSSION

In this study, we evaluated whether genetic alteration profiling of serial liquid biopsy is useful for monitoring treatment outcomes in real-time throughout the treatment of GC patients prospectively. We serially collected ctDNA of 15 HER2-positive GC patients from prospective open-label phase Ib/II study of the first-line pembrolizumab in combination with trastuzumab, capecitabine, and cisplatin in HER2-positive GC; the prospective trial study enrolled 43 patients and showed objective response rate of 76.7% and median PFS of 8.6 months.¹⁶

Liquid biopsy is an essential tool for the non-invasive monitoring of cancer patients and overcomes the limitations of needle biopsy, such as false negatives due to tumor heterogeneity, since ctDNA is derived from either primary or metastatic cancer cells.²² Liquid biopsy has been suggested as a useful non-invasive tool for the early detection and tracking of treatment outcomes in GC.²³,²⁴ However, more data is required to verify its applicability for the management of GC patients after chemotherapy and to improve their performance.

Through targeted NGS analysis, we identified more than one genetic alteration in ctDNAs from most of the GC patients (14/15, 93%), suggesting that liquid biopsy has enough sensitivity to monitor genetic alterations in patient blood, which enables real-time tracking of treatment outcomes. The VAF profiles of somatic mutations and CNAs identified by liquid biopsy were largely consistent with the clinical response levels to the treatment, suggesting that serial liquid biopsies can reflect the treatment responses properly; therefore, this method can be applied in the clinical setting.

In the baseline ctDNAs, a number of mutations in well-known cancer-related genes, such as KRAS, PIK3CA, KDM6A, ERBB3, EGFR, and KIT, were identified, some of which were recurrently detected: TP53 (60%), AR (20%), APC (13.3%), ARID1A (13.3%), and RET (13.3%). Our results were consistent with those of previous reports that screened mutations in GCs,³,⁴,⁵,⁶,⁷ further supporting the reliability of our data. Regarding CNA profiles, CNAs were identified in more than half of the 15 GC cases, and the most recurrent CNA was an amplification of the ERBB2 gene (6/15, 40.0%). Although the concordance rate of 40% between HER2-positive status based on IHC and SISH (3+ on IHC or 2+ on IHC and SISH+) and ERBB2 amplification in ctDNA was lower than our expectation, this result was largely consistent with previous studies. In a report on breast cancer by Sakai, et al.,²⁵ the concordance rate between HER2 positivity (IHC3+ or IHC2+ and FISH positive) and HER2 amplification in ctDNA was only 31%. In a study on GC by Niu, et al.,²⁶ the concordance rate was 64.7%. The discrepancy between IHC/SISH and ctDNA may be attributed to several factors, such as the time difference between tumor tissue sampling and blood sampling, intratumor heterogeneity, and biological factors that affect ctDNA shedding.²⁵ It has been commonly reported that the detection sensitivity of genetic alterations in ctDNA is lower than that in primary tissue,²⁷ suggesting that the lower detection sensitivity of genetic alterations in ctDNA is another cause for this discrepancy. Of note, when we compared the genetic alteration profiles of ctDNAs after treatment to those at baseline for each patient, there were no newly detected mutations or CNAs after treatment, suggesting that the possibility of acquisition of new genetic alterations after ICI treatment is very low.

Tumor mutational burden (TMB) is being increasingly recognized as a potential biomarker for the response to ICIs in diverse cancers, including GC.²⁸,²⁹,³⁰ However, targeted NGS-based TMB could be substantially distorted due to various factors, such as library construction, variant calling pipeline, and target region to be sequenced, leading to over- or underestimation of the TMB.³¹,³² In this study, we evaluated mTBI, defined as the mean VAF of detected mutations,²¹ which can help standardize the applicability and accuracy of ctDNA in monitoring tumor burden in HER2-positive GC patients. The longitudinal analysis of ctDNA from baseline to several time points after treatment revealed a mean 13.4 weeks lead time in the detection of disease progression compared to RECIST during conventional follow-up with CT. This finding was consistent with a previous report³³ in which mTBI of ctDNA was largely correlated with radiographic response. Indeed, most patients in the non-responder group showed elevated mTBI before disease progression confirmed by radiography, suggesting that post-treatment changes in ctDNA were powerful predictors of both response and progression in GC. Moreover, we found two insightful cases: the GC05 case represents the sensitivity of this ctDNA detection method, whereas the GC14 case shows that ctDNA detection might be useful only for distant hematogenous metastasis progression.

This study had several limitations. First, although our ctDNA monitoring was investigated among patients undergoing cancer immunotherapy regimen, we could not detect any immune-specific marker, which is a common limitation of DNA sequencing methods, especially with ctDNA (except for tumor mutation burden). However, we have proposed that mTBI might predict treatment response to immunotherapy. For comprehensive immune-specific monitoring, combining immune cell profiling or RNA sequencing would be a good integrative approach. Second, although all TP53 mutations detected in this study were reported as somatic mutations in the COSMIC database, the potential origin from clonal hematopoiesis cannot be ruled out.³⁴,³⁵ Further investigation, including targeted deep sequencing of matched leukocyte DNA, is needed to rule out whether hematopoietic cells are potential source of cell-free DNA and contribute somatic variants to the ctDNA pool. Third, we did not compare the tissue-based and ctDNA-based NGS results, since the main aim of this study was to verify the clinical implications of serial liquid biopsies targeting ctDNA in patients with advanced GC on systemic chemotherapy rather than judging the consistency of the genetic alteration profiles between primary tissue and ctDNA. Fourth, we did not find any newly acquired mutations or CNAs during serial sampling, which could explain the clinical outcomes. This may be due to the limited number of target genes in our NGS panel. The development of new NGS-panels covering more targets, including GC-specific mutations and CNAs, is required to overcome intratumor heterogeneity and suggest new target agents. Fifth, the patients enrolled in this study showed relatively good responses to the combination therapy compared to traditional treatments; therefore, there was a limitation in identifying the genetic alterations associated with the treatment outcome. Lastly, we collected the patient’s blood when they visit for chemotherapy (every 3 weeks), but did not evaluate the optimal frequencies and intervals of blood sampling in this study. Further prospective trials for ctDNA profiling with a larger number of GC patients will provide clinically applicable alteration markers related to treatment outcomes, and an optimal strategy for blood collection for detection of progression in advance than CT.

In conclusion, our prospective study suggests that serial genetic alteration profiling of ctDNA is a feasible way to predict treatment response in HER2-positive GC patients in a minimally invasive manner. Practically, ctDNA profiles are useful not only for the molecular diagnosis of GC but also for the selection of metastatic GC patients with poor prognosis for systemic chemotherapy.

SUPPLEMENTARY MATERIALS

Supplementary Table 1

The Target Genes for OncoChase Cancer Panels

Supplementary Table 2

The Description of Targeted Deep Sequencing Data

Supplementary Table 3

Somatic Point Mutations and Indels Identified in ctDNAs from 15 GC Genomes

Supplementary Table 4

Somatic Copy Number Alterations Identified in ctDNAs from 15 GC Genomes

Supplementary Fig. 1

Amplification of ERBB2 gene. Amplifications of ERBB2 were detected in the ctDNA samples of GC patients by targeted NGS. The x-axis represents genomic position, and the y-axis represents the relative depth ratio (tumor/normal) in log2 scale. Red circle represents the ERBB2. ctDNA, circulating tumor DNA; GC, gastric cancer; NGS, next-generation sequencing.

Supplementary Fig. 2

Serial comparison of tumor burden and genetic alterations from ctDNA. VAF of mutations and absolute copy number of CNAs detected in baseline were compared serially with tumor burden for each patient, except one patient GC04, who did not harbor any genetic alterations. Longitudinal analyses of ctDNA were performed and analyzed among two groups according to their response pattern to the treatment: (A) responder group (GC01 to GC07) and (B) non-responder group (GC08 to GC15). In the GC05 case, the second cycle of treatment was delayed for 2 weeks due to grade 3 thrombocytopenia; therefore, the third blood sampling was performed before the second cycle of treatment. ctDNA, circulating tumor DNA; GC, gastric cancer; CNAs, copy number alterations; VAF, variant allele frequency; CN, copy number; AU, arbitrary unit.

Supplementary Fig. 3

Serial comparison of mTBI from ctDNA. The average mTBI at first sampling was significantly lower than that of baseline in both responder (p=0.031) and non-responder groups (p=0.008). Of note, the average mTBI at the last blood sampling after treatment was significantly higher than that of first blood sampling in the non-responder group (p=0.016), but not in the responder groups (p>0.05). There was also no significant difference between baseline and last sampling in the non-responder group. mTBI, molecular tumor burden index; ctDNA, circulating tumor DNA.