Sage Journals: Discover world-class research

Abstract

Next-generation sequencing-based genomic profiling facilitates biomarker detection by cell-free DNA (cfDNA) liquid biopsy. However, the efficiency of mutation calling and the prognostic value of cfDNA biomarkers are disputed. We investigated 24 patients with gastric cancer in this study, using a 605-gene sequencing panel to sequence their plasma cfDNA and tumor tissue DNA. The mutation concordance between plasma cfDNA and tumor tissue DNA was 70.6% in stage IV gastric cancer and 30.2% in stage III gastric cancer, indicating insufficient mutation detection rates in stage III and early-stage cancer. When compared with total cfDNA load and blood tumor mutation burden (bTMB), the variant allele frequencies (VAF) of commonly mutated genes are highly accurate in representing disease burden. Further, VAF are a better prognostic indicator compared with serum biomarkers including carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), cancer antigen 125 (CA125), and alpha-fetoprotein (AFP). The use of cfDNA in molecular profiling of patients allows prediction of patient survival and clinical response, as well as the development of personalized therapy regimens.

Keywords

Gastroenterology cfDNA molecular characterization prognosis surveillance

Key Summary

The vast difference in cfDNA mutation detection efficiency between stages III and IV of gastric cancer shows that the mutation spectrum of stage I-III cancer revealed by cfDNA is not representative under the current sequencing settings.

The cfDNA level is a more sensitive prognostic indicator compared with TMB, and the prognostic value of TMB depends largely on the VAF of cfDNA.

The cfDNA VAF of frequently mutated genes serves as a good prognostic indicator, and is superior to serum biomarkers.

Introduction

Gastric cancer is the fifth most common cancer and the fourth leading cause of cancer-related deaths worldwide, according to the global cancer statistics,¹ with an estimated 1 million new cases and 0.8 million new deaths each year. Gastric cancer is a public health threat in China, ranking third in terms of incidence and mortality, with 47.9 new cases and 37.4 new deaths per 10 000 people.² The prevailing risk factors for gastric cancer include Helicobacter pylori (H. pylori) infection, alcohol intake, obesity, smoking, and low socioeconomic status.³ The pathology frequently involves modification of patients’ genomic landscape.

Liquid biopsy uses body fluid samples to decipher tumor genetics. It is a less invasive approach compared with needle biopsy or surgery, and reduces the risk of metastasis caused by punctures. However, it is not clear whether the cancer biomarkers identified by tissue sample sequencing can be applied directly to liquid biopsy. The mutation consistency between tissue samples and plasma samples ranged from 69% to 94% depending on the target mutation and sequencing technology.^4-7 Several studies have reported that cell-free DNA (cfDNA) levels can be used to monitor tumor recurrence in various cancer types.^8,9 The aim of the study is to investigate the efficiency and reliability of cfDNA in monitoring disease progression, and validate the prognostic value of cfDNA by comparing it with other prognostic indicators.

Methods

Patients and samples

From June 2017 to March 2021, 24 patients with histologically proven stomach cancer were included in this study at Shenzhen People’s Hospital (Shenzhen, China). The study was approved by the Ethics Committee of Shenzhen People’s Hospital (Shenzhen, China) under the approval number SYLLS201522. All patients provided written informed consent before enrollment. In this retrospective clinical study, 22 formalin-fixed, paraffin-embedded (FFPE) tumor tissue samples and 39 whole peripheral blood (10 mL) samples were obtained. Serum carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), cancer antigen 125 (CA125), and alpha-fetoprotein (AFP) levels were measured at multiple time points via immunostaining. The clinical responses of the patients were evaluated according to the RECIST 1.1 guideline.¹⁰ We collected preoperative or pretreatment whole blood samples and associated tumor tissue samples to analyze the mutation concordance between paired plasma samples and tissue samples. To evaluate the prognostic value of cfDNA in gastric cancer, we collected blood samples from 7 patients who received standard or adjuvant therapy. The samples were collected before and at the end of a regime.

DNA extraction and quantification

Aliquots of 10-20 mL whole blood samples were collected from each patient. The debris and plasma in peripheral blood lymphocytes (PBL) were separately harvested and stored. Genomic DNA (gDNA) from PBLs was extracted using the RelaxGene Blood DNA System (TianGen Biotech Co., Ltd., China). Cell-free DNA (cfDNA) was extracted from at least 2 mL plasma using the QIAamp Circulating Nucleic Acid kit (QIAGEN) following the manufacturer’s instructions. Tumor tissues measuring 0.5 cm× 0.5 cm were collected following surgical resection. Tumor DNA was extracted using the QIAamp DNA FFPE tissue kit (Qiagen) following the manufacturer’s instructions.

Library construction

FFPE tumor tissue DNA and genomic DNA from PBLs were first sheared by the enzyme NEBNext dsDNA Fragmentase. The size selection of the DNA fragments (150-250 bp) was performed using Ampure XP beads (Beckman Coulter, Inc., Brea, CA, USA). The KAPA Library Preparation kit (Kapa Biosystems, Inc., Wilmington, MA, USA) was used for library construction, according to the manufacturer’s protocol.

Targeted capture and sequencing

Targeted capture was performed using a custom set of biotinylated DNA probes containing all the exon and some intron regions of 468 cancer-related genes, and SNP loci of 137 treatment-related genes, encompassing 1.31 Mb of sequencing panel (Roche NimbleGen). The libraries were sequenced using 150 bp paired-end runs on the Illumina NovaSeq 6000 system (Illumina).

Raw data processing and alignment

Raw sequencing data were pre-processed by fastp v0.12.6 version 0.18.0, followed by sliding window trimming.¹¹ Clean reads were aligned to the hg19 genome (GRch37) using Burrows Wheeler Aligner v0.7.15 r1140 under default settings.¹² The Gencore v0.12.0 was used to remove duplicate reads.¹³ Samtools v0.1.19 was applied to generate pileup files for properly paired reads with mapping quality ⩾60.¹⁴

Mutation calling, filtering and annotation

Single nucleotide variants (SNVs) and short insertions/deletions (indels) were identified by VarScan2 v2.3.8.¹⁵ The average sequencing depth after deduplication was around 1000× (all samples ⩾ 750×) for tumor tissue DNA and 2500× (all samples ⩾1500×) for cfDNA. The variant allele frequency (VAF) threshold was set to 0.1% in cfDNA and 2% in FFPE tumor tissue DNA. The cutoff values were selected based on previous publications.^16-18 All SNVs/indels were annotated using ANNOVAR v2018-04-16.¹⁹ CNVkit v0.9.3 was used for detection of copy number variation (CNV).²⁰ The blood tumor mutational burden (bTMB) was estimated by cfDNA based on the total number of SNVs and indels with greater than 0.5% VAF per million bases. Visualizations of groups of mutations were carried out in R using the package maftools.²¹

Statistical analysis

Statistical analysis as shown in Figures 1, 5 and 6 was performed using GraphPad Prism v8.0. Normality tests, F test, and t-test were conducted as described in Figure captions. Statistical significance was evaluated as described by the figure captions. *** P < .001, ** P < .01, * P < .05, ns = not significant (P > .05). Numerical data in this study were presented as mean ±S.E.M. The mutation concordance rate was calculated as the number of shared mutations dived by the sum of mutations identified in plasma samples and tumor tissue samples.

Figure 1.

Mutation landscape of DNA extracted from tumor tissue and peripheral blood samples: (A) OncoPrint showing all mutations and copy number variations identified in tumor tissue samples of all patients. Mutation type and copy number gain or loss were color coded, (B) OncoPrint showing cfDNA mutations in patients with stage IV cancer, (C) as in (B), but involving patients with stage III cancer, (D) percentage of total mutations detected in plasma samples, tumor tissue samples, or both are shown in the boxplot, with red denoting stage IV cancer and blue representing stage III cancer. Statistical comparison was conducted by Mann-Whitney test, *denotes P < .05, **indicates P < .01, data expressed as mean ± S.E.M., and (E) as in (D), but showing the percentage of mutations that were solely identified in tumor tissue samples or plasma samples.

The statistical power and sample size were calculated with the software G*Power 3.²² Based on our study design, we set a 2-tailed t-test with 80% power and the type I error probability (α) of .05. The calculated total sample size was 26 with the actual power of 0.82. To adjust the sample size, we revisited published studies with a comparable study design. Our sample sizes are similar to those reported in previous publications.²³

Results

The mutation concordance between cfDNA and tumor tissue DNA was low in stage III cancer

Twenty-four patients who were diagnosed with gastric adenocarcinoma were enrolled in this study at Shenzhen People’s Hospital (Shenzhen, China). Paired plasma samples and fresh tissue samples were collected at first diagnosis. Follow-up plasma samples were collected at the beginning and the end of each course of treatment. The mean age of the patients was 54.33 ± 2.95 (28-75), and 50% (12/24) of them were male (Table 1). One (4.2%, n = 24) stage I patient, 9 (37.5%, n = 24) stage III patients, and 14 (58.3%, n = 24) stage IV patients were enrolled. One stage III patient and one stage IV patient harbored high microsatellite instability (MSI), while the others were stable microsatellites. The mean overall survival of the patients was 15.60 ± 2.10 (1.37-41.574) months.

Table 1.

Demographic and clinical characteristics of the patients enrolled in this study.

Characteristics	Patients
Age, years, median (range)	54.5 (28-75)
Sex, n (%)
Male	12 (50%)
Female	12 (50%)
Histopathology, n (%)
Adenocarcinoma	24 (100%)
Stage, n (%)
1	1 (4.2%)
3	9 (37.5%)
4	14 (58.3%)
MSI, n (%)
High	2 (8.3%)
Stable	22 (91.7%)
Treatment, n (%)
Surgical resection	15 (62.5%)
Chemotherapy	20 (83.3%)
Targeted therapy	14 (58.3%)
Immunotherapy	14 (58.3%)
Radiotherapy	1 (4.2%)
OS, months, median (range)	15.6 (1.37-41.57)
Baseline sample, n (%)
Blood	24 (100%)
Tissue	22 (91.7%)
Blood + Tissue	22 (91.7%)
Number of blood samples at monitoring points, n (%)
1	16 (66.7%)
2	3 (12.5%)
3	3 (12.5%)
4	2 (8.3%)

To compare the mutation concordance between cfDNA and tumor tissue DNA (tDNA), we collected preoperative or pretreated whole blood samples and corresponding tumor tissue samples. The frequent mutations of cfDNA and tDNA are displayed separately in Figure 1. The frequently mutated genes detected in tDNA were TP53 (64%), ARID1A (23%), and SYNE1 (23%, Figure 1A, Supplemental Table 1). In terms of CNV, most of the genes exhibited copy number amplification, including BRCA1/2, CCND1/3, CDKN1/2A, FGF3/4, and FGFR2/3 (Supplemental Figure 1). Copy number loss was only observed in 2 patients (FGFR3 and CDKN2A), while no structural variants were identified.

The mutation spectrum of cfDNA captured from stage IV cancer patients showed high concordance with that of paired tDNA (Figure 1B). In contrast, the cfDNA in stage I & III cancer exhibited a different mutation spectrum (Figure 1C). Unique mutations were found in both tissue and plasma samples. The tDNA of patients with stage IV and III cancer carried 86.8% ± 7.8%, and 95.2% ± 4.8% of total mutations, respectively. However, only 32.6% ± 12.1% of total mutations were identified in the cfDNA of patients with stage III cancer, which is significantly lower than the mutation detection rate of cfDNA in patients with stage IV cancer (83.8% ± 8.1%, n = 21, P < .001, Figure 1D). The mutation concordance between matched cfDNA and tDNA was evidently higher in stage IV cancer than in stage III cancer (70.6% ± 10.3% vs 30.2% ± 11.7%, n = 21, P < .05), suggesting the need for caution using cfDNA to guide clinical procedures for the management of stage III and earlier-stage cancer.

In stage III cancer, 65.5% ± 13.2% of mutations were solely detected in tDNA, while only 4.3% ± 4.3% of mutations were exclusively found in cfDNA. This tendency was reversed in stage IV cancer, with only 16.1% ± 8.0% of mutations identified in tDNA, whereas 13.3% ± 7.8% of mutations identified solely in cfDNA (Figure 1E). The low mutation concordance in patients with stage III cancer may be attributed to the low VAF of cfDNA. The mean VAF cfDNA of stage III cancer samples was 0.6 ± 0.2 (ranging from 0.1 to 41.4), significantly lower than in stage IV cancer samples (4.5 ± 0.9, ranging from 0.15 to 21.16, P < .0001, n = 21). Accordingly, the cfDNA more accurately reflects the molecular profile of late-stage cancers that release a sufficient amount of ctDNA into the circulation system.

Disease burden can be assessed by cfDNA level, tumor mutation burden, and maximum variant allele frequency

To evaluate the prognostic value of cfDNA in gastric cancer, we analyzed the mutation spectrum and VAFs of cfDNA in patients providing plasma samples at the beginning and the end of each regimen. The cfDNA level, tumor mutation burden (TMB), and maximum VAFs were compared to identify the parameter that accurately reflected the disease burden. Among the 7 patients who were monitored for 6-23 months, 4 showed consistent fluctuation in cfDNA level and TMB (Figure 2). P13 showed a decrease in cfDNA level and TMB (Figure 2A), while P5 and P18 experienced a relapse of disease after the initial decrease (Figure 2B and C). P11, however, exhibited an increase in disease burden along with the treatment (Figure 2D). However, the cfDNA level and TMB were inconsistent in 3 patients (Figure 3A-C), with 2 cases reporting the TMB below the detection threshold (Figure 3A and C) and 1 case showing inverted trends of cfDNA level and TMB (Figure 3B).

Figure 2.

Comparing the prognostic value of different metrics used to monitor disease burden: (A) the cfDNA level (ng/mL) and tumor mutation burden (mutation/Mb) of patient 13 at different time points are shown in the upper panel. The variant allele frequencies (%) of mutations identified in plasma samples at the same time points are displayed in the OncoPlot. (B) As in (A), but for patient 5. (C) As in (A), but for patient 18. (D) As in (A), but for patient 11.

Figure 3.

The cfDNA and TMB showed inconsistent results in disease monitoring. (A) The cfDNA level (ng/mL) and tumor mutation burden (mutation/Mb) of patient 10 at different time points are shown in the upper panel. The variant allele frequencies (%) of mutations identified in plasma samples at the same time points are displayed in the OncoPlot. (B) As in (A), but for patient 6. (C) Maximum variant allele frequency of patients in the consistent or inconsistent group. (D) The mean variant allele frequency of the consistent and inconsistent group. (E) The cfDNA concentration of patients with stage III and stage IV cancers (ng/mL). Shapiro-Wilk test was conducted to examine the normality of the datasets. Statistical comparisons were conducted via unpaired t-test.

The 2 groups of patients were not different in terms of staging, lymph node metastasis, overall survival (OS), molecular subtypes, and treatment regimens. One factor that may contribute to the inconsistency between cfDNA load and TMB is the maximum VAF. The maximum VAF of the cfDNA-TMB consistent group was 6.8 ± 4.8, far exceeding the maximum VAF of the cfDNA-TMB inconsistent group (1.4 ± 1.3, Figure 3D). To determine whether the difference in mutation detection rates was caused by varied tumor shedding, we compared the cfDNA concentration between stage III and IV gastric cancers. The mean cfDNA concentration of stage III was 26.74 ± 6.03 ng/mL, which was not significantly different from the cfDNA concentration in stage IV (20.28 ± 3.47 ng/mL, n = 22, P = .36, Figure 3E), demonstrating that the low technical sensitivity was not caused by differential tumor shedding. However, the changes in VAF of frequently mutated genes appear to be a more sensitive parameter for monitoring compared with the changes in total cfDNA (Figures 2 and 3). Therefore, we further assessed the prognostic value of VAF of frequently mutated genes.

The cfDNA VAF of frequently mutated genes represents a good prognostic indicator

The serum biomarkers and cfDNA loads of 6 patients was monitored at 2 to 4 time points, and the changes in disease burden of the patients were traced. The results showed that the fluctuations in VAF of frequently mutated genes more accurately reflect the progression of disease compared with the serum biomarkers. The VAFs of 3 frequently mutated genes (top gene mutations at each time point) were plotted against time, and compared with the changes in canonical serum biomarkers including CEA, CA19-9, CA125, and AFP.²⁴ During the 6 to 23 months of follow-up, the symptoms of 2 patients were relieved (Figure 4A and B), whereas 2 patients experienced disease progression (Figure 4C and D), and 2 other patients reported possible relapse of disease after the initial alleviation (Figure 4E and F). The increase or decrease in VAF of the 3 frequently mutated genes reflected disease progression or alleviation with high fidelity. In 5/6 (83.3%) cases, the change of VAF accurately depicted the fluctuations in disease burden. In P18 (Figure 4E), the drop in VAF was consistent with the partial response to neoadjuvant chemotherapy and the alleviation following surgical resection. Although the disease stopped in July 2020, a novel TP53 mutation with very low VAF (1.53%) was detected in the cfDNA measured in March 2021, indicating tumor relapse, which was confirmed by follow-up clinical examination in 2022. This increase in disease burden was also reported based on serum CEA, CA19-9, and cfDNA levels, although the fluctuation was within the normal range and may be too subtle to be identified by imaging.

Figure 4.

Monitoring cancer progression using driver genes: (A) the levels of serum CEA (ng/mL), AFP (ng/mL), CA199 (U/mL), and CA125 (U/mL) at different time points in patient 13 are shown in the upper panel. The VAF of top 3 mutations at the 3 time points are shown in the lower panel, and the clinical status is color-coded. (B) As in (A), but for patient 6. (C) As in (A), but for patient 10. (D) As in (A), but for patient 11. (E) As in (A), but for patient 18. (F) As in (A), but for patient 5.

The prognostic value of the serum markers CEA, AFP, CA19-9, and CA125 was comparatively poor. The consistencies between CA125, CA199, AFP, and CEA fluctuations and changes in disease burden were 50.0% (3/6), 33.3% (2/6), 16.7% (1/6), and 16.7% (1/6), respectively. The cfDNA load precisely reflected the disease burden in 66.7% (4/6) patients, showing a moderate accuracy. Collectively, the VAF of frequently mutated genes is a better prognostic indicator compared with serum markers and cfDNA level. Moreover, monitoring the mutation status of all driver genes instead of the initially identified mutations provides insight into tumor metastasis, given that new loci of mutations were reported in the 3 patients that exhibited disease progression (Figure 4C, D and F).

cfDNA-based molecular characterization associated with patient prognosis

Until our last follow-up on April 1, 2021, 54.2% (13/24) of patients survived, while 41.7% (10/24) of patients died, and 1 patient lost contact with the research group. The mean overall survival (OS) of the patients was 15.6 ± 2.1 months (ranging from 1.4 to 41.6 months. Two or more lines of treatment were administered to 83.35% (20/24) of the patients including surgical removal (62.5%, 15/24), chemotherapy (83.3%, 20/24), targeted therapy (58.3%, 14/24), immunotherapy (58.3%, 14/24), and radiotherapy (4.2%, 1/24, Figure 5A). Among the 14 patients who received targeted therapy, 12 carried mutations targeted by the drug (Supplemental Table 2).

Figure 5.

Prognosis of gastric cancer patients. (A) Patient OS and the treatments received are shown in the plot, n = 23, 1 patient lost contact with the research group. (B) Kaplan-Meier survival analysis of patients with low (bTMB = 0, n = 8), medium (bTMB 0.76-3.82, n = 8), and high (bTMB 5.34-10.69, n = 7) bTMB. (C) Kaplan-Meier survival analysis of patients with low (MVAF = 0, n = 7), medium (MVAF 0.01-0.04, n = 8), and high (MVAF 0.08-0.4, n = 8) MVAF. (D) Kaplan-Meier survival analysis of patients with or without TP53 mutation. (E) Kaplan-Meier survival analysis of patients with or without ARID1A mutation. (F) Kaplan-Meier survival analysis of patients with or without SYNE1 mutation. (G) Kaplan-Meier survival analysis of patients with or without CDKN mutation. Statistical analysis was conducted using log-rank test. Patient number at risk table was shown below the plot.

As each patient was assigned a unique treatment strategy based on their clinical features, it is hard to analyze how the treatment strategy affected patient survival. Both bTMB (log-rank test P = .62; multivariate analysis using Cox proportional hazards model yielded a hazard ratio of 0.39, 95% confidence interval 0.70-1.22, P = .84, Figure 5B) and maximum VAF (log-rank test P = .21; multivariate Cox proportional hazards regression analysis yielded a hazard ratio 0.92, P = .58, Figure 5C) did not correlate with patient survival. However, the molecular characterization of the patients showed a sophisticated impact on the OS of the patients regardless of the treatment choice.

Both The Cancer Genome Atlas (TCGA) and Asian Cancer Research Group (ACRG) revealed the proposed molecular subtypes of gastric cancer. However, the enrolled patients were mostly EBV-negative and microsatellite stable, while the epithelial-to-mesenchymal transition status, TP53 activity, and genomic stability could not be assessed based on the targeted sequencing results. Therefore, we examined the impact of mutation status of frequently mutated genes (mutations shared by at least 4 patients) on patient survival. No statistically significant difference was observed between the wild type group and the mutant group carrying TP53 (log-rank test P = .99; multivariate Cox proportional hazards model hazard ratio 0.93, 95% confidence interval 0.22-3.89, P = .93, Figure 5D), ARID1A (log-rank test P = .62; multivariate Cox proportional hazards model hazard ratio 0.75, 95% confidence interval 0.06-7.23, P = .75, Figure 5E), and SYNE1 (log-rank test P = .76; multivariate Cox proportional hazards model hazard ratio 0.91, 95% confidence interval 0.19-4.41, P = .91, Figure 5F). In contrast, patients carrying CDKN1C and CDKN2A/C mutations showed an OS of 8.3 ± 3.5 months, while the OS of the wild type group reached 16.6 ± 2.3 months (log-rank test P = .0079; multivariate Cox proportional hazards model hazard ratio 3.09, 95% confidence interval 0.59-16.33, P = .18, Figure 5G). Mutations of CDKN1C, CDKN2A, and CDKN2C have been identified. The genes encode a series of tumor suppressors and potential tumor suppressors that regulate cell cycle and cell proliferation similar to other proteins encoded by CDKN gene. Due to the limited patient size, it is impossible to reliably analyze the effect of CNV of CDKN genes on patient survival. To validate our findings and further investigate the question, we downloaded the DNA sequencing data of patients with gastric cancer from the TCGA database.

We analyzed the correlation between OS and CDKN1/2/3 gene mutations or CNV using the TCGA data comprising 295 patients with primary gastric adenocarcinoma.²⁵ Patients with or without CDKN3, CDKN2D, CDKN2A, and CDKN1 showed no difference in OS (24.4 ± 3.1 months vs 20.2 ± 0.1 months, log-rank test P = .23, multivariate Cox proportional hazards model hazard ratio 0.97, 95% confidence interval 0.33-2.83, P = .96). We further performed correlation analysis of the CNV data. The copy number of the chromosome region containing CDKN2A and CDKN2B exhibited a negative correlation with patient OS (Spearman’s correlation P = .05, r = −.118). The copy number amplification does not facilitate patient survival. Consistent with our finding, aberrations in CDKN genes tend to compromise the prognosis of patients with gastric cancer. Such molecular characterizations associated with patient prognosis can be identified by cfDNA, which contributes to the determination of treatment regimens based on the predicted patient survival.

cfDNA facilitates dynamic monitoring of tumor progression

The serial cfDNA test can be utilized for dynamic monitoring of tumor evolution. The mutation spectrum can be used to track the possible origin and progression or recurrence of tumors. It can also be used to construct the phylogenetic trees. The changes in mutation spectrum are shown in Figures 2, 3A and B. Further, the phylogenetic trees of p5 and p10 are presented in Figure 6. For p5, the pre-operative ctDNA (cf_20190321) was strongly consistent with the tumor tDNA (tumor) in mutation spectrum. Post-operatively, no mutation was detected in the cfDNA (cf_20190418). However, in the last cfDNA test, new mutations were detected. Besides, the mutation spectrum was totally different from that of tumor tissue, indicating the distinct origin and evolution of lung metastasis. Interestingly, although the p10 showed a similar status change as p5, the last ctDNA (cf_20201107) test showed that these 2 mutations also occurred in the tumor tDNA (tumor), which indicated the possible origin of the metastasis in the primary lesion.

Figure 6.

The phylogenetic trees of p5 and p10: (A) the phylogenetic trees of p5, including the tumor tissue and 3 cfDNA samples and (B) the phylogenetic trees of p10. The tumor tissue and 3 cfDNA samples are also included.

Discussion

The study reviewed the advantages and limitations of cfDNA in monitoring the progression of late-stage gastric cancer. The vast difference in the efficiency of cfDNA mutation detection between stages III and IV showed that the mutation spectrum of the stages I-III cancer revealed by cfDNA is not representative under the setting of 1500 × depth and the lower detection threshold of 0.1%. The staging of cancer has restricted the utility of cfDNA in mutation detection. Studies demonstrated that the mutation concordance between plasma samples and tissue samples was 32% when patients in all cancer stages were included,²⁶ which reached 83.4% if only patients with metastatic cancer were enrolled.²⁷ The low mutation concordance is also attributed to the technical limitations of NGS, as the mutation concordance was over 70% when assessed by PCR.^4,28 Increasing the DNA input and sequencing coverage depth can improve the performance and robustness of NGS tests. High coverage and read depth can be achieved by improving the capture efficiency and stability of enriched DNA, NGS library conversion, and amplification. However, this will also increase the cost of sequencing, thus limiting the use of NGS in clinical settings. Our study showed that the mean sequencing depth of 1500× at the lower detection threshold of 0.1% failed to ensure detection stringency and sensitivity. Consistent with our findings, a previous study showed that the mean fragment depth of 4000× may yield around 50% sensitivity at the VAF cutoff of 0.1%.¹⁸

The cfDNA or ctDNA level was widely utilized in prognostic monitoring. The ctDNA clearance was reported to predict prognosis under different treatment regimens.²⁹ Our results showed that cfDNA level was a more sensitive prognostic indicator compared with TMB, and the prognostic value of TMB depends strongly on the VAF of cfDNA. The VAF of frequently mutated genes was the best prognostic indicator compared with the cfDNA level, TMB, and serum markers. The detection threshold of 0.1% or less was based on pre-determined valid results. Our study proved that the accuracy of monitoring by VAF of frequently mutated genes reached 83.3% at the detection threshold of 0.1%. Notably, the monitoring of disease burden should be based on all genes and not merely on the initially identified mutations, due to the frequent emergence of novel mutations observed in metastatic and recurrent tumors.

There are many plausible hypotheses that could explain the heterogeneity in cancer. The clonal evolution model proposed that mutations occur over time and may confer a selective advantage that allows sub-clones of cancer cells to outcompete other clones.³⁰ Our findings showed that in some patients only selective mutation levels were increased after treatment (Figure 3). This observation indicates that the tumors comprised heterogeneous clones of cells, some clones responded to treatment while the others did not. ctDNA detects subclones that are resistant to current treatment. Different from total progression, this clonal progression manifests the selective advantage of certain mutations, which causes secondary drug resistance. More importantly, our data showed that ctDNA test could detect the clonal progression before tumor progression can be visualized by medical imaging. This clonal dynamic detected by ctDNA could partially explain secondary drug resistance, guide the design of treatment regimen based on the mutation spectrum of the progressive clone, and monitor disease progression ahead of imaging.

The cfDNA can also be used in the molecular characterization of patient subgroups. We identified a series of CDKN gene aberrations associated with poor prognosis. CDKN1/2/3 comprises a series of genes that are transcribed to tumor suppressors or regulators of cell cycle. CDKN1A encodes the cell cycle inhibitor p21, which is triggered by p53 in response to DNA damage, and leads to cell cycle inhibition in G1 phase.³¹ CDKN2A encodes p16 and p14, both of which are CDK4 inhibitors that regulate cell cycle.^32,33 CDKN2B is located adjacent to CDKN2A on chromosome 9p21, and are frequently deleted together in various cancer types. CDKN2B is also transcribed to a CDK4 inhibitor named p15, which regulates cell growth and cell cycle G1 phase.³⁴ CDKN3 also encodes a cyclin-dependent kinase inhibitor that binds primarily to CDK1 and CDK2 to regulate the cell cycle.³⁵ Mutant and aberrant expression of CDKN genes is frequently encountered in cancer and is associated with patient survival.^36-39 Copy number loss and large deletion of the CDKN2A/B also lead to poor prognosis.^33,34 Consistently, our study showed that mutations of CDKN1C/2A/2C and CNV of CDKN2A/B were associated with poor prognosis in gastric cancer. Identifying the molecular features of patients contributes to the prediction of patient survival, which in turn guides the treatment strategy based on the therapeutic time window available.

The major limitation of study is the small sample size, which may lead to unstable results. Although statistical methods were adjusted to address the issue, a larger cohort is required to validate the findings.

Conclusion

In this study, the clinical application of cfDNA was discussed regarding the accuracy of mutation calling, advantages over serum biomarkers, and its role in predicting prognosis. The mutation concordance between cfDNA in plasma samples and tumor DNA in tissue samples was markedly lower in patients with stage III gastric cancer compared with stage IV at the VAF detection threshold of 0.1%. Among the parameters that reflect disease burden, VAFs of frequently mutated genes are the most reliable indicator in predicting the prognosis of patients with better accuracy compared with serum biomarkers. One limitation of the study is the small sample size, highlighting the need to validate the results in a larger clinical study.

Supplemental Material

sj-jpg-1-bmi-10.1177_11772719221141525 – Supplemental material for Advantages and Limitations of Monitoring Circulating Tumor DNA Levels to Predict the Prognosis of Patients Diagnosed With Gastric Cancer

Supplemental material, sj-jpg-1-bmi-10.1177_11772719221141525 for Advantages and Limitations of Monitoring Circulating Tumor DNA Levels to Predict the Prognosis of Patients Diagnosed With Gastric Cancer by Wan He, Jingxin Yang, Xiao Sun, Shunda Jiang, Jinchan Jiang, Ming Liu, Tianhao Mu, Yingmei Li, Xiaoni Zhang, Jingxian Duan and Ruilian Xu in Biomarker Insights

Supplemental Material

sj-xlsx-2-bmi-10.1177_11772719221141525 – Supplemental material for Advantages and Limitations of Monitoring Circulating Tumor DNA Levels to Predict the Prognosis of Patients Diagnosed With Gastric Cancer

Supplemental material, sj-xlsx-2-bmi-10.1177_11772719221141525 for Advantages and Limitations of Monitoring Circulating Tumor DNA Levels to Predict the Prognosis of Patients Diagnosed With Gastric Cancer by Wan He, Jingxin Yang, Xiao Sun, Shunda Jiang, Jinchan Jiang, Ming Liu, Tianhao Mu, Yingmei Li, Xiaoni Zhang, Jingxian Duan and Ruilian Xu in Biomarker Insights

Supplemental Material

sj-xlsx-3-bmi-10.1177_11772719221141525 – Supplemental material for Advantages and Limitations of Monitoring Circulating Tumor DNA Levels to Predict the Prognosis of Patients Diagnosed With Gastric Cancer

Supplemental material, sj-xlsx-3-bmi-10.1177_11772719221141525 for Advantages and Limitations of Monitoring Circulating Tumor DNA Levels to Predict the Prognosis of Patients Diagnosed With Gastric Cancer by Wan He, Jingxin Yang, Xiao Sun, Shunda Jiang, Jinchan Jiang, Ming Liu, Tianhao Mu, Yingmei Li, Xiaoni Zhang, Jingxian Duan and Ruilian Xu in Biomarker Insights

Footnotes

Acknowledgements

We thank Dr. Yan Yin for proofreading and language editing. We thank Sridhar N (University of Toronto) and EditorBar for the language editing service.

Declarations

Supplemental material

Supplemental material for this article is available online.

References

Sung

Ferlay

Siegel

Laversanne

Soerjomataram

Jemal

, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71:209-249.

Cao

Chen

. Changing profiles of cancer burden worldwide and in China: a secondary analysis of the global cancer statistics 2020. Chin Med J. 2021;134:783-791.

Hundahl

Phillips

Menck

. The National Cancer Data Base Report on poor survival of U.S. gastric carcinoma patients treated with gastrectomy: Fifth Edition American Joint Committee on Cancer staging, proximal disease, and the "different disease" hypothesis. Cancer. 2000;88:921-932.

Zhu

Wang

Zhou

Dong

Liu

Zhang

. RAS mutational status detection in tissue, plasma, and stool samples for colorectal cancer. Biomed Res Int. 2020;2020:5419634.

Fang

Zhang

. A comparison of consistency of detecting BRAF gene mutations in peripheral blood and tumor tissue of nonsmall-cell lung cancer patients. J Cancer Res Ther. 2014;10:C150-C154.

Zhang

Dong

Ren

Zhang

. Consistency of genotyping data from simultaneously collected plasma circulating tumor DNA and tumor-DNA in lung cancer patients. J Thorac Dis. 2020;12:7290-7297.

Chae

Davis

Jain

, et al. Concordance of genomic alterations by next-generation sequencing in tumor tissue versus circulating tumor DNA in breast cancer. Mol Cancer Ther. 2017;16:1412-1420.

Luo

Zhao

Wei

, et al. Circulating tumor DNA methylation profiles enable early diagnosis, prognosis prediction, and screening for colorectal cancer. Sci Transl Med. 2020;12: eaax7533.

Saluja

Karapetis

Pedersen

Young

Symonds

. The use of circulating tumor DNA for prognosis of gastrointestinal cancers. Front Oncol. 2018;8:275.

10.

Eisenhauer

Therasse

Bogaerts

, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45:228-247.

11.

Chen

Zhou

Chen

. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics. 2018;34:i884-i890.

12.

Durbin

. Fast and accurate long-read alignment with Burrows-Wheeler transform. Bioinformatics. 2010;26:589-595.

13.

Chen

Zhou

Chen

, et al. Gencore: an efficient tool to generate consensus reads for error suppressing and duplicate removing of NGS data. BMC Bioinformatics. 2019;20:606.

14.

Handsaker

Wysoker

, et al. The sequence alignment/map format and SAMtools. Bioinformatics. 2009;25:2078-2079.

15.

Koboldt

Zhang

Larson

, et al. VarScan 2: somatic mutation and copy number alteration discovery in cancer by exome sequencing. Genome Res. 2012;22:568-576.

16.

Manca

Corallo

Busico

, et al. The added value of baseline circulating tumor DNA profiling in patients with molecularly hyperselected, left-sided metastatic colorectal cancer. Clin Cancer Res. 2021;27:2505-2514.

17.

Jennings

Arcila

Corless

, et al. Guidelines for validation of next-generation sequencing-based oncology panels: a joint consensus recommendation of the Association for Molecular Pathology and College of American Pathologists. J Mol Diagn. 2017;19:341-365.

18.

Deveson

Gong

Lai

LoCoco

Richmond

Schageman

, et al. Evaluating the analytical validity of circulating tumor DNA sequencing assays for precision oncology. Nat Biotechnol. 2021;39:1115-1128.

19.

Wang

Hakonarson

. ANNOVAR: functional annotation of genetic variants from high-throughput sequencing data. Nucleic Acids Res. 2010;38:e164.

20.

Talevich

Shain

Botton

Bastian

. CNVkit: genome-wide copy number detection and visualization from targeted DNA sequencing. PLoS Comput Biol. 2016;12:e1004873.

21.

Mayakonda

Lin

Assenov

Plass

Koeffler

. Maftools: efficient and comprehensive analysis of somatic variants in cancer. Genome Res. 2018;28:1747-1756.

22.

Faul

Erdfelder

Lang

Buchner

. G*Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav Res Methods. 2007;39:175-191.

23.

Kim

Song

, et al. Monitoring circulating tumor DNA by analyzing personalized cancer-specific rearrangements to detect recurrence in gastric cancer. Exp Mol Med. 2019;51:1-10.

24.

Feng

Tian

, et al. Diagnostic and prognostic value of CEA, CA19-9, AFP and CA125 for early gastric cancer. BMC Cancer. 2017;17:737.

25.

Cancer Genome Atlas Research Network. Comprehensive molecular characterization of gastric adenocarcinoma. Nature. 2014;513:202-209.

26.

Cao

Liu

Chen

, et al. Circulating tumor DNA is capable of monitoring the therapeutic response and resistance in advanced colorectal cancer patients undergoing combined target and chemotherapy. Front Oncol. 2020;10:466.

27.

Vandekerkhove

Lavoie

Annala

, et al. Plasma ctDNA is a tumor tissue surrogate and enables clinical-genomic stratification of metastatic bladder cancer. Nat Commun. 2021;12:184.

28.

Schmiegel

Scott

Dooley

, et al. Blood-based detection of RAS mutations to guide anti-EGFR therapy in colorectal cancer patients: concordance of results from circulating tumor DNA and tissue-based RAS testing. Mol Oncol. 2017;11:208-219.

29.

Song

Xie

, et al. Circulating tumor DNA clearance predicts prognosis across treatment regimen in a large real-world longitudinally monitored advanced non-small cell lung cancer cohort. Transl Lung Cancer Res. 2020;9:269-279.

30.

Shackleton

Quintana

Fearon

Morrison

. Heterogeneity in cancer: cancer stem cells versus clonal evolution. Cell. 2009;138:822-829.

31.

Choudhury

Djojosubroto

, et al. Cdkn1a deletion improves stem cell function and lifespan of mice with dysfunctional telomeres without accelerating cancer formation. Nat Genet. 2007;39:99-105.

32.

Bartels

van Luttikhuizen

Christgen

, et al. CDKN2A loss and PIK3CA mutation in myoepithelial-like metaplastic breast cancer. J Pathol. 2018;245:373-383.

33.

Alhejaily

Day

Feilotter

Baetz

Lebrun

. Inactivation of the CDKN2A tumor-suppressor gene by deletion or methylation is common at diagnosis in follicular lymphoma and associated with poor clinical outcome. Clin Cancer Res. 2014;20:1676-1686.

34.

Yang

Zhang

Han

, et al. Cyclin-dependent kinase inhibitor 2b controls fibrosis and functional changes in ischemia-induced heart failure via the BMI1-p15-Rb signalling pathway. Can J Cardiol. 2021;37:655-664.

35.

Fan

Chen

Huang

, et al. Overexpression of major CDKN3 transcripts is associated with poor survival in lung adenocarcinoma. Br J Cancer. 2015;113:1735-1743.

36.

Shah

Boyd

Houlston

Kaiser

. Constitutional mutation in CDKN2A is associated with long term survivorship in multiple myeloma: a case report. BMC Cancer. 2017;17:718.

37.

Dai

Fang

Cai

, et al. CDKN3 expression predicates poor prognosis and regulates adriamycin sensitivity in hepatocellular carcinoma in vitro. J Int Med Res. 2020;48:300060520936879.

38.

Chen

Shen

. Overexpression of CDKN2B is involved in poor gastric cancer prognosis. J Cell Biochem. 2019;120:19825-19831.

39.

Riese

Dahse

Fiedler

, et al. Tumor suppressor gene p16 (CDKN2A) mutation status and promoter inactivation in head and neck cancer. Int J Mol Med. 1999;4:61-65.