Clinical characteristics of enrolled patients
From June 2017 to January 2020, 27 patients were eligible for analysis. The clinical characteristics of enrolled patients are listed in Table 1. There were 18 males and 9 females, with a median age of 65 years (37 to 77). More than half of the patients (51.9%) were gastric antrum origination. Fourteen cases had hematogenous metastasis. Four major chemotherapy regimens were identified, including platinum-based doublet regimen (Pt-2d), platinum-based triplet regimen (Pt-3d)(24), paclitaxel-based regimen (PTX) and irinotecan-based regimen (CPT-11). Fourteen patients were treated by Pt-3d and 13 patients were treated by Pt-2d as first-line regimen. Six patients and 5 patients were treated by CPT-11 and PTX as subsequent therapy, respectively. Therefore, two sequential treatment patterns, namely Pt-2d followed by PTX and Pt-3d followed by CPT-11, were identified (Table S1). The objective response rate of the first-line treatment was 51.8% (14/27). At the end of follow-up (December 31, 2020), 3 patients were still alive. The median OS of these patients was 8.5 months and the progression free survival (PFS) of first-line setting was 4.7 months.
Table 1
Clinical characteristics of 27 AGC patients
Clinical characteristics
|
|
Case (n)
|
Percentage (%)
|
Gender
|
Male
|
18
|
66.7
|
|
Female
|
9
|
33.3
|
Age
|
Median
|
65.0
|
|
|
Range
|
37–77
|
|
Tumor site
|
Fundus
|
5
|
18.5
|
|
Body
|
8
|
29.6
|
|
Antrum
|
14
|
51.9
|
Hematogenous metastasis
|
Yes
|
14
|
51.9
|
|
No
|
13
|
48.1
|
Signet cell
|
Yes
|
6
|
22.2
|
|
No
|
21
|
77.8
|
Regimens
|
Pt-3d
|
14
|
51.9
|
|
FLOT
|
11
|
/
|
|
DOS
|
3
|
/
|
|
Pt-2d
|
13
|
48.1
|
|
SOX
|
10
|
/
|
|
CAPOX
|
2
|
/
|
|
FOLFOX
|
1
|
/
|
|
PTX
|
5
|
18.5
|
|
Paclitaxel
|
2
|
/
|
|
PX
|
2
|
/
|
|
Nab-Paclitaxel
|
1
|
/
|
|
CPT-11
|
6
|
22.2
|
|
IRIR
|
3
|
/
|
|
Irinotecan
|
3
|
/
|
|
FOLFIRI
|
1
|
/
|
Treatment pattern
|
Pt-3d to CPT-11
|
6
|
22.2
|
|
Pt-2d to PTX
|
5
|
18.5
|
ORR of 1st-line
|
PR
|
14
|
51.9
|
|
SD
|
8
|
29.6
|
|
PD
|
5
|
18.5
|
Correlation between genomic alterations in ctDNA and clinical efficacy of sequential chemotherapy
During the treatment, a total of 100 blood samples were collected prospectively and plasma ctDNA were extracted. The median sampling times for patients was 3 (2 to 10). Firstly, ctDNA content fraction (CCF), copy number instability (CNI) and tumor mutation burden (TMB) of all samples were calculated. The correlations between dynamic change of CCF and patients’ survival time are illustrated in Fig. 1A. For most patients, CCF decreased from the baseline once the disease was under control, whereas it increased upon tumor progression. For patients with a relatively long-term disease control, CCF was also at a low-level. Dynamic change of CNI corresponded to the tendency of CCF (Fig. 1B).
Delta CNI defined as the difference between CNI at baseline and that in first sample after treatment was calculated to further verify the correlation between CNI and the treatment efficacy. For patients with partial response (PR) as their best overall response to the first-line regimen, the mean value of delta CNI was significantly lower than that in patients with stable disease (SD) or progression of disease (PD) (-1687.56 ± 1617.54 vs. -147.71 ± 1260.87, P = 0.011, Fig. 1C). The correlations between CNI, CCF and TMB with treatment response were further illustrated in three typical cases (Fig. 1D). All these patients responded to first-line regimens while did not respond to subsequent regimens. After disease progression of initial regimen, CNI, CCF and TMB all continued increasing during following subsequent treatment.
Fluctuating Changes Of Gene-level Cnvs During Sequential Chemotherapy
Gene-level copy number variants (CNVs) and single nucleotide variants (SNVs) were then detected utilizing a customized NSG panel including 1632 genes. For all of the samples, the top 10 recurrent genes with CNVs were NFKBIA, MCL1, HSP90AA1, MAP2K3, CREB1, PDPR, CALR, TOP1, IRS2 and PTPRT, and the top 10 recurrent genes with SNVs were TP53, LRP1B, CDH1, KMT2D, KMT2C, ARID1A, RHOA, KMT2B, APC and CTNNB1 (Fig. 2, Fig. S1). A fluctuating change of gene-level CNVs could be observed from the landscape of all of the samples. Some gene CNVs were undetectable during the period of disease control while recurred at disease progression.
Similar profiles of gene-level CNVs between baseline and endpoint of treatment
To further delineate the fluctuating change of gene-level CNVs during sequential chemotherapy, the data of samples at baseline and endpoint of all 27 patients were compared first. Genes with CNVs could be assigned into three categories according to their changes during treatment comparing with baseline data, including genes that were recurrent at endpoint, those missing at endpoint and those specific at endpoint.
For each category, the top 10 genes are presented in Fig. 3A. Overlapping genes could be observed in the three categories, where the frequency of the recurrent and missing genes was higher than that of the endpoint specific genes. The majority of detected genes at the endpoint with high frequency were shared in recurrent and missing gene sets, including HSP90AA1, CALR, NFKB1A, MCL1 and CREB1. Genes that specifically occurred at the endpoint were only observed in a few patients. These genes included NBN, IRS2, LYN, SDHB and CSMD3.
A subtle impact of subsequent regimens on gene-level CNVs driven by first-line treatment
As the profiles of genes with CNVs were similar between baseline and endpoint of treatment, this result suggested that the effect of subsequent regimens on gene-level CNVs driven by first-line regimens might be subtle. There were 11 patients who received subsequent regimens in our cohort. Therefore, gene-level CNVs in ctDNA samples collected at endpoint of first-line regimen (EP1) and endpoint of second-line regimen (EP2) were analyzed. As was expected, most genes with CNVs were recurrent between EP1 and EP2 (Fig. 3B). EP2 specific genes with CNVs comparing with EP1 and baseline data occurred in only four patients (P17, P7, P25 and P8) including CALR, HSP90AA1, MYC, FAM135B, FGFR2 and CREB1. Distribution of genes among these patients was highly heterogeneous, and no shared genes were identified among these patients.
Significant difference of gene copy number values during sequential treatment
Expect gene-level CNVs, copy number values of genes between the baseline and the endpoint were also analyzed. Unlike the CNVs, nineteen genes were identified to show significant differences. Genes with an increased copy number value after sequential treatment included CBFB, IL6ST, FBXW7, VEGFC, CYLD, FOXP1 and SMAD4. Genes with a decreased copy number value included FCRL1, UGT1A, SMARCD1, CASP8, PDCD1, CHST3, KMT5A, ATIC, ALK, RAD51D, NCOA1 and IKBKE (Fig. 4A).
Similar to the impact of subsequent regimens on gene-level CNVs, genes with significant different copy number values were not found between EP1 and EP2. For genes with CNVs detected in these samples, copy number value of genes was similar before and after the treatment of either PTX or CPT-11 as subsequent regimens (Fig. 4B).
Resistance-related genes of chemotherapy regimens defined by drug modified score (DMS)
The correlation between gene-level CNVs in ctDNA and treatment efficacy had been showed by our results. Therefore, we supposed that dynamic change of gene copy number values during treatment could be used to identified resistance-or sensitivity-related genes of chemotherapy regimens. DMS of genes was calculated based on dynamic changes of gene copy number values during treatment comparing with their baseline data. Resistance-related genes of each regimen were identified as those with increasing abnormality of copy number values (DMS increased) during treatment, while sensitivity-related genes were identified as those with decreasing abnormality of copy number values (DMS decreased). There were 69 genes for Pt-2d, 86 genes for Pt-3d, 57 genes for PTX and 41 genes for CPT-11 identified as resistance-related genes (Fig. 5A, table S2). Genes with significant differential copy number value changes during treatment were also defined as resistance- or sensitivity-related genes in the four regimens employed (Fig. 5B).
Afterwards, the oncologic signaling pathway enrichment of the identified resistance-related genes was performed (Fig. 5C). Two major pathways, RTK RAS pathway and PI3K pathway, were identified in all four regimens. Homologous recombination pathway and TGF-beta pathway were shared between Pt-3d and PTX regimes. Nrf2 pathway and mismatch repair pathway were also related with tumor resistance to Pt-3d regimen. Myc pathway was more specific to CPT-11 regimen and nucleotide excision repair pathway was specific to Pt-2d regimen.
A high inter-patient heterogeneity of involved resistance-related genes in common enriched pathways
Afterwards, we turned our focus on two common pathways, namely RTK RAS pathway and PI3K pathway. Based on the data of Pt-2d and Pt-3d regimens, the resistance-related genes in play have been labeled on the diagrams of RTK RAS pathway (Fig. 6A) and PI3K pathway (Fig. 6B) according to the types of genomic alterations detected at the baseline and their frequency in patients. CNV gain was observed in most of the oncogenes including PIK3CA, AKT, MTOR, EGFR family, FGFR and RET, where CNV loss was observed in tumor suppressor genes including PTEN, STK11, TSC1, PIK3R and PTPN11. For patients who were treated by Pt-3d (n = 14) or Pt-2d (n = 13), almost all of the genes in these two pathways could be identified as resistance-related genes, where the proportion of patients who carried those genes was relatively low. Most of such genes occurred in less than 30% of patients, which showed a high inter-patient heterogeneity.