Genomic characterization of peritoneal lavage cytology-positive gastric cancer

Abstract

Objective

Positive peritoneal lavege cytology (CY1) gastric cancer is featured by dismal prognosis, with high risks of peritoneal metastasis. However, there is a lack of evidence on pathogenic mechanism and signature of CY1 and there is a continuous debate on CY1 therapy. Therefore, exploring the mechanism of CY1 is crucial for treatment strategies and targets for CY1 gastric cancer.

Methods

In order to figure out specific driver genes and marker genes of CY1 gastric cancer, and ultimately offer clues for potential marker and risk assessment of CY1, 17 cytology-positive gastric cancer patients and 31 matched cytology-negative gastric cancer patients were enrolled in this study. The enrollment criteria were based on the results of diagnostic laparoscopy staging and cytology inspection of exfoliated cells. Whole exome sequencing was then performed on tumor samples to evaluate genomic characterization of cytology-positive gastric cancer.

Results

Least absolute shrinkage and selection operator (LASSO) algorithm identified 43 cytology-positive marker genes, while MutSigCV identified 42 cytology-positive specific driver genes. CD3G and CDKL2 were both driver and marker genes of CY1. Regarding mutational signatures, driver gene mutation and tumor subclone architecture, no significant differences were observed between CY1 and negative peritoneal lavege cytology (CY0).

Conclusions

There might not be distinct differences between CY1 and CY0, and CY1 might represent the progression of CY0 gastric cancer rather than constituting an independent subtype. This genomic analysis will thus provide key molecular insights into CY1, which may have a direct effect on treatment recommendations for CY1 and CY0 patients, and provides opportunities for genome-guided clinical trials and drug development.

Introduction

Gastric cancer ranks as the fifth most common cancer and the third most common cause of cancer-related deaths globally (1-4). Peritoneum is a common site of metastasis for gastric cancer, occurring in up to 30% patients at diagnosis, even in countries like Japan with robust screening programs (5,6). The prognosis of patients with gastric cancer peritoneal metastasis (PM) remains dismal, with a median survival of less than 1 year (7).

Given the dismal prognosis associated with PM, an accurate diagnosis remains crucial for prognostic evaluations and therapeutic decisions. Specifically, peritoneal lavage cytology can aid in detecting disseminated cancer cells in the peritoneal cavity before formation of gross peritoneal carcinomatosis (8,9). This allows for the definition of PM approaches, particularly in cases with positive peritoneal lavege cytology (CY1) (10,11). In fact, American Joint Committee on Cancer/International Union Against Cancer Classification 8th edition currently employs washing cytology for detection of malignant cells and a more accurate tumor staging, and cytology-positive tumors even without a visible metastatic tumor (P0/CY1) are considered as advanced cancers with distant metastasis (12).

However, there is a lack of evidence on pathogenic mechanism and signature of CY1 and there is a continuous debate on CY1 therapy. For example, in America, National Comprehensive Cancer Network (NCCN) gastric cancer guidelines define CY1 as metastatic (M1) disease and recommend a palliative treatment (13), while studies from Japan mainly recommended that surgery combined with adjuvant chemotherapy may have a survival benefit in patients with CY1 gastric cancer (14-16). Therefore, exploring the mechanism of CY1 is crucial for treatment strategies and targets for CY1 gastric cancer.

As the understanding of the genomic status of gastric cancer deepens through next generation sequencing (NGS) (17-19), genomic signature of several special subtypes of gastric cancer is becoming clear. Nowadays, NGS is a powerful technology for elucidating the pathogenesis of human cancer and identifying potential therapeutic targets (20). Based on the assistance of NGS, The Cancer Genome Atlas (TCGA) and the Asian cancer research group classified gastric cancer into 4 molecular subtypes: chromosomal instability (CIN), genome stability (GS), microsatellite instability (MSI), and Epstein-Barr virus (EBV) positivity as DNA-based alterations (18,21). Apart from classification, NGS is also widely applied to guide the diagnosis and treatment of some special types of gastric cancer. For example, MSI status is determined through use of NGS, by identifying short repeating DNA segments termed microsatellites as well as alterations in mismatch repair (MMR) genes (22-24); NGS is also useful for identifying mutation and its load of HER2 mutation, and offers clues for mechanism of trastuzumab resistance (18,25).

Therefore, this study, by evaluating and differentiating negative peritoneal lavege cytology (CY0) and CY gastric cancer using NGS, aimed to figure out specific driver genes and marker genes of CY1 gastric cancer, and ultimately offer clues for potential markers and risk assessment of CY1. The genomic analysis will provide key molecular insights into CY1, potentially influencing treatment recommendations for both CY1 and CY0 patients, and creating opportunities for genome-guided clinical trials and drug development.

Materials and methods

Diagnostic laparoscopy staging

Laparoscopy was performed under general anesthesia. The patient was placed in a supine position. A 10-mm disposable trocar (observing hole) was inserted into the sub-umbilicus, and a 30° telescope was used. Another 10-mm trocar and a 5-mm trocar (operating hole) were inserted into the right and left upper quadrants, respectively. Prior to any manipulation, 250 mL of warm normal saline was infused into the subphrenic space, subhepatic space, omentum, bilateral paracolic sulci and the pouch of Douglas. Care was taken to avoid direct contact of the irrigation with the primary tumor. At least 200 mL of fluid was aspirated from the subphrenic space, subhepatic space and pouch of Douglas. The fluid was immediately sent for stimulated Raman and hematoxylin and eosin (H&E) cytological examination with 100 mL respectively. Subsequently, a systematic inspection of the abdominal cavity was performed clockwise from the right quadrant. Any suspicious lesion was biopsied and sent for pathologic examination. Patients with negative diagnostic laparoscopy results were labeled as PM−, and those with positive diagnostic laparoscopy results were labeled as PM+.

Cytology inspection of exfoliated cells

Exfoliated cells were stained with H&E, and two professional pathologists independently examined the H&E slides. According to the pathologic report system, patients with gastric cancer identified or suspected as PM were enrolled in this study and admitted to Peking University Cancer Hospital from December 2020 to July 2021. Patients were archived based on the following criteria: 1) pathologically diagnosed with gastric adenocarcinoma; 2) with PM; 3) with classic morphology of tumor cells; 4) tumor cells forming cluster excluded. In addition, various kinds of ascitic fluid specimens, including ascitic fluids, peritoneal washes and peritoneal lavage fluids were detected by H&E staining. Thirty-one patients with negative-cytology results were labeled as CY0, and 17 positive-cytology results were labeled as CY1.

Patient issue samples

According to the result of diagnostic laparoscopy staging and cytology inspection of exfoliated cells, 17 patients with P0CY1 who underwent gastrectomy at Peking University Cancer Hospital were included. Cases were enrolled in this study according to the criteria as follows: at least 20 years old; pathologically confirmed gastric adenocarcinoma; complete clinicopathological and follow-up data; without autoimmune disease or other cancer types; radical gastrectomy (R0 resection); and without previous chemotherapy/radiotherapy. In order to control potential confounding factors, 31 CY0 cases were enrolled randomly from conventional gastric cancer patients, with matched T, N stages with CY1 groups. The pathologic diagnoses and characteristics were independently determined by at least two experienced pathologists. A written informed consent was obtained prior to participation for each participant. All procedure was approved by the Ethics Committee of Peking University Cancer Hospital (No. 2022KT63).

Whole exome sequencing

Genomic DNA from fresh tumor tissues were extracted using the DNeasy Blood & Tissue kit (Qiagen, Germany). FFPE samples were de-paraffinized with xylene followed by genomic DNA extraction using QIAamp DNA FFPE Tissue Kit (Qiagen, Germany). We used KAPA Hyper Prep Kit (KAPA Biosystems, MA, United States) to prepare a whole genome library and the Illumina Rapid Capture Extended Exome Kit (Illumina Inc., USA) to perform exome capture. Enriched libraries were sequenced using the Illumina HiSeq4000 NGS platform (Illumina, USA) as paired 150-bp reads.

Bioinformatics analysis of cancer-related features

The mutational profiles of all samples were reconstructed according to the 30 previously identified mutational signatures in COSMIC (26) after preprocessing the mutation data by mutSignatures (27) using hg19 human reference genome. The somatic mutation types were annotated by Ensembl Variant Effect Predictor (VEP) (28) with RefSeq Cache. Known driver genes of gastric cancer were obtained from COSMIC cancer gene census (CGC), and the mutational frequency and categories on these genes were summarized accordingly. The subclone analysis is consist of two steps, i.e. after detecting somatic copy number variants (CNVs) of all samples by cnv_facets (29), the R package sciClone (30) was used to infer subclonal architecture of each sample with the default parameters.

Marker gene selection and functional enrichment analysis

We employed two approaches to select genes that could distinguish CY1 group from CY0 group. First, CY1-specific driver genes were predicted by MutsigCV_v1.35 (31) which was implemented on GenePattern server. Second, we established a least absolute shrinkage and selection operator (LASSO) regression model to select CY1 marker genes by using R package glmnet (32). To prevent overfitting, multiple attempts were made to fit the model using cross-validation. Lambda.min refers to the lambda value within the regularization path that corresponds to the minimum cross-validation error. In regularization methods such as ridge regression and LASSO regression, the lambda parameter is employed to control the model’s complexity, while lambda.min is utilized to select the model that yields optimal performance. In the LASSO model, the mutation statuses (yes/no) of 17,977 genes were used as the predictor features and the group attribute (CY1 and CY0) were used as the categorical label to be predicted. Finally, the gene ontology (GO) enrichment analysis and the Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of the selected genes were performed by using KOBAS 3.0 (33) with Entrez Gene ID. The functional enrichment results were considered as statistically significant if corrected P<0.05.

Results

Basic characteristics are demonstrated in Table 1, with detailed information attached in Supplementary Table S1.

Table 1. Basic characteristics of gastric cancer patients.

Variables	n (%)
	CY0 (N=31)	CY1 (N=17)
BMI, body mass index; CY0, negative peritoneal lavege cytology; CY1, positive peritoneal lavege cytology.
Age (year) [Median (range)]	64 (60−71)	63 (59−67)
Sex
Male	23 (74.2)	15 (88.2)
Female	8 (25.8)	2 (11.8)
BMI (kg/m2) [Median (range)]	24.0 (20.8−26.5)	22.8 (21.8−24.0)
History of abdominal surgery	7 (22.6)	3 (17.6)
Tumor location
Upper third gastric 　cancer	15 (48.4)	6 (35.3)
Median third gastric 　cancer	6 (19.4)	4 (23.5)
Lower third gastric 　cancer	10 (32.3)	7 (41.2)
Tumor’s long diameter (cm) [Median (range)]	4.5 (3.1−7.0)	3.5 (3.0−5.5)
Lauren classification
Intestinal	17 (54.8)	9 (52.9)
Diffuse	4 (12.9)	5 (29.4)
Mixed	9 (29.0)	3 (17.6)
Others	1 (3.2)	0 (0)
Vascular cancer embolus	19 (61.3)	12 (70.6)
Perineural invasion	10 (32.3)	15 (88.2)

Table S1. Detailed characteristics of gastric cancer patients.

Cases	cTNM	pTNM	Tumor location	Tumor size		Lauren classification	Vessel carcinoma embolus	Perineural invasion	Age (year)	Sex	BMI (kg/m2)	History of abdominal operation	Preoperative treatment
				Long diameter (cm)	Short diameter (cm)
BMI, body mass index; M, male; F, female.
CY1
1	T4bN+bM1	T4bN3bM1	Gastric antrum	12.5	−	Intestinal	1	1	69	M	24	0	0
2	T4bN2M1	T4aN2M1	Gastric antrum	2.6	2.5	Intestinal	1	1	70	M	21	0	0
3	T4aN2M1	T3N0M1	Esophago-gastric junction	2.5	−	Mixed	1	1	66	M	23	1	1
4	T4bN2M1	T4aN3aM1	Gastric body	5.5	3.0	Intestinal	1	1	57	M	22	0	0
5	T4N3M1	T4bN3bM1	Gastric antrum	6.0	3.0	Diffuse	1	1	71	F	19	0	0
6	T4aN1M1	T3N0M1	Esophago-gastric junction	7.0	5.0	Intestinal	0	1	62	M	30	1	1
7	T4bN1M1	T3N2M1	Gastric antrum	5.5	4.0	Intestinal	0	1	78	M	21	1	0
8	T4aN2M1	T4aN2M1	Gastric antrum	4.5	3.8	Intestinal	1	1	63	M	22	0	0
9	T4bN1M1	T4aN1M1	Esophago-gastric junction	3.0	2.5	Intestinal	1	1	67	M	20	0	0
10	T4aN2M1	T4bN3aM1	Esophago-gastric junction	3.0	3.0	Mixed	1	0	55	M	22	0	1
11	T3N2M1	T4aN2M1	Esophago-gastric junction	3.5	2.5	Intestinal	0	1	66	M	23	0	1
12	T3N1M1	T3N3aM1	Esophago-gastric junction	3.8	3.0	Diffuse	1	1	64	M	24	0	1
13	T4aN3M1	T3N3bM1	Gastric body	2.2	−	Diffuse	1	1	63	M	24	0	1
14	T4aN3M1	T4bN3M1	Gastric body	7.0	−	Diffuse	1	1	59	M	26	0	1
15	T2N1M1	T4aN3aM1	Gastric antrum	2.5	2.0	Mixed	0	1	57	M	27	0	0
16	T3N1M0	T2N0M1	Gastric antrum	3.0	1.5	Diffuse	0	0	61	F	22	0	0
17	T3N1M0	T3N1M1	Gastric body	3.0	1.5	Intestinal	1	1	58	M	26	0	1
CY0
18	T3N2M0	T3N2M0	Esophago-gastric junction	4.5	2.5	Mixed	1	0	64	M	20	0	0
19	T4aN1M0	T4aN2M0	Gastric body	4.5	3.5	Intestinal	0	0	72	M	28	0	0
20	T3N2M0	T2N0M0	Esophago-gastric junction	2.0	−	Intestinal	0	0	55	M	23	0	0
21	T4aN2M0	T3N1M0	Gastric antrum	9.0	−	Diffuse	1	0	66	F	25	0	0
22	T2N1M0	T2N1M0	Gastric cardia	4.0	3.2	Intestinal	1	0	58	M	28	0	0
23	T4aN2M0	T3N0M0	Gastric antrum	7.0	4.0	Diffuse	0	1	42	M	21	0	0
24	T3N1M0	T3N2M0	Gastric antrum	3.0	2.0	Mixed	1	1	66	M	20	0	0
25	T2N0M0	T4aN2M0	Esophago-gastric junction	3.0	2.0	Intestinal	0	0	61	M	29	1	0
26	T1bN0M0	T3N3aM0	Esophago-gastric junction	2.5	−	Intestinal	1	0	58	F	32	0	0
27	T4aN0M0	T4bN0M0	Esophago-gastric junction	8.0	6.0	Intestinal	0	0	66	M	26	0	0
28	T3N2M0	T3N3aM0	Esophago-gastric junction	5.0	3.0	Mixed	1	0	66	F	25	0	0
29	T3N0M0	T3N0M0	Gastric antrum	7.0	3.0	Mixed	1	0	78	F	22	1	0
30	T3N0M0	T3N2M0	Gastric cardia	3.5	3.0	Mixed	1	0	71	F	24	0	0
31	T4N2M0	T1bN2M0	Gastric antrum	2.5	2.0	Intestinal	0	0	45	F	23	1	0
32	T4aN2Mx	T4aN3aM0	Gastric cardia	4.5	3.5	Diffuse	0	1	49	M	24	0	1
33	T4aN2M0	T3N3aM0	Gastric body	3.0	−	Diffuse	0	0	71	M	23	0	0
34	T4aN0M0	T3N2M0	Esophago-gastric junction	5.5	4.0	Mixed	1	1	66	M	21	1	0
35	T4N3M0	T3N3bM0	Gastric antrum	6.0	5.0	Intestinal	1	1	66	M	18	1	0
36	T4aN2M0	T3N3bM0	Gastric antrum	4.5	4.0	Mixed	1	1	63	M	18	0	0
37	T2N0M0	T1aN0M0	Gastric body	1.5	−	Intestinal	0	0	63	M	28	0	0
38	T3N1M0	T3N3aM0	Esophago-gastric junction	4.0	2.3	Intestinal	0	1	72	M	26	0	0
39	T4aN3M0	T3N3M0	Esophago-gastric junction	7.5	7.0	Mixed	1	1	76	M	20	1	0
40	T3N2M0	T3N2M0	Gastric antrum	6.5	6.5	Intestinal	1	0	81	F	25	0	0
41	T4aN0M0	T3N0M0	Gastric body	4.5	3.0	Intestinal	0	1	57	F	30	0	0
42	T3N0M0	T1bN2M0	Gastric antrum	2.5	2.0	Mixed	1	0	62	M	22	0	0
43	T4bN3M0	T3N2M0	Gastric antrum	7.5	4.5	Intestinal	1	0	63	M	18	0	0
44	T3N1M0	T2N3aM0	Esophago-gastric junction	3.5	2.5	Intestinal	1	0	62	M	18	1	0
45	T3N2M0	T3N1M0	Esophago-gastric junction	8.0	8.0	Intestinal	1	1	45	M	27	0	0
46	T4aN1M0	T3N3aM0	Gastric body	9.0	6.5	Intestinal	1	0	77	M	23	0	0
47	T4aN3M0	T3N2M0	Gastric body	14.0	8.0	−	1	0	62	M	26	0	0
48	T4aN2M0	T3N1M0	Esophago-gastric junction	7.0	6.0	Intestinal	0	0	71	M	29	0	0

Overview of panel

The overview of somatic mutations is illustrated in Figure 1A. Missense mutations constituted the largest fraction of somatic mutations, followed by silent mutations. As for the comparison between CY1 and CY0 group, although a few samples from the CY0 group show high mutation numbers, there were no significant differences of mutational burden between the two groups (Wilcoxon test, P=0.41). We also specifically investigated the mutation numbers for major mutation types (missense and silent mutations) but observed no significant difference (Wilcoxon test, P=0.81 and 0.66, respectively) between CY1 and CY0 group.

Figure 1.

Mutational characteristics of CY0 and CY1. (A) Overview of panel demonstrated no systemic difference of mutational burden (Wilcoxon test, P=0.41) and mutation numbers for major mutation types (Wilcoxon test, P=0.81, 0.66 for missense and silent mutations, respectively) between CY1 and CY0 group; (B) Mutational signature profiles of CY1 group were similar to that of CY0 group; (C) BIRC6, ZFHX3, ATR and FAT3 were among the most frequently mutated driver genes, but there was no consistent difference of cancer genes between CY1 group and CY0 group; (D) There was no significant difference in max VAF of subclone clusters between CY1 group and CY0 group. CY0, negative peritoneal lavege cytology; CY1, positive peritoneal lavege cytology; VAF, variant allele frequencies.

There was no significant difference in mutational signatures, driver gene mutation and tumor subclone architecture between CY1 group and CY0 group

To better depict the mutational patterns between two groups, the mutational profiles of both CY1 and CY0 samples were reconstructed according to the 30 pre-defined mutational signatures in COSMIC. As shown in Figure 1B, COSMIC signature 3, which is associated with failure of DNA double-strand break-repair by homologous recombination, dominates the reconstructed mutational signature profiles. But the mutational signature profiles of CY1 group was similar to that of CY0 group.

We further investigated whether there was a significant difference in the mutation tendency for key genes in gastric cancer between two groups. Twenty-four driver genes related to gastric cancer were obtained from COSMIC CGC. BIRC6, ZFHX3, ATR and FAT3 were among the most frequently mutated driver genes, but there was no consistent difference of cancer genes between CY1 group and CY0 group (Figure 1C).

Finally, we compared the tumor heterogeneity by inferring subclone architecture between two groups by using SciClone. To avoid the influence of minor subclone, only the top three clusters of each sample were retained for next analysis. There was no significant difference in max variant allele frequencies (VAF) of the subclone clusters between CY1 group and CY0 group (Figure 1D).

In all, the above results suggest that the CY1 group and the CY0 group are similar in conventional tumor features.

CY1-specific driver genes predicted by MutsigCV

We then attempted to identify informative gene mutation features beyond the aforementioned conventional tumor features. Firstly, MutSigCV_v1.35 was used to predict cancer driver genes of CY1 group and CY0 group separately. The genes with corrected P<0.05 were considered as the candidate cancer driver genes. There were 60 cancer driver genes in CY1 group (Supplementary Table S2) and 267 cancer driver genes in CY0 group (Supplementary Table S3). The top-scored driver genes for the two groups are shown in Figure 2A. Among the 60 cancer driver genes in CY1 group, 42 genes were only observed in CY1 group, which were considered as CY1-specific driver genes. Functional enrichment analysis indicated these CY1-specific driver genes were associated with function terms like Th17 cell differentiation (Figure 2B) and pathways like protein binding (Figure 2C).

Table S2. Sixty cancer driver genes and corresponding corrected P values predicted by MutsigCV in CY1 group.

Gene	Corrected P	Gene	Corrected P
CY1, positive peritoneal lavege cytology.
TTN	<0.001	LRRIQ3	0.027
FSIP2	0.003	MAMLD1	0.027
GRB14	0.005	FANCE	0.027
CD3G	0.005	TTK	0.027
AHI1	0.007	UPF2	0.027
TP53	0.009	OR7C2	0.027
COL19A1	0.009	FAM155A	0.027
CHD4	0.011	NEGR1	0.031
SYCP2	0.012	ZNF763	0.031
CDKL2	0.012	QKI	0.031
COQ9	0.013	ZNF787	0.031
NSMCE4A	0.013	NUMBL	0.035
SEC31A	0.013	CEACAM6	0.035
BBS5	0.014	CDK12	0.035
IFNGR1	0.014	ATF1	0.035
YBX2	0.014	ZEB1	0.038
PTPN12	0.015	XDH	0.038
CDX1	0.016	TGFBI	0.038
ZC3H18	0.016	ABL1	0.039
MLLT1	0.016	SMAD4	0.040
IRAK4	0.016	LEO1	0.040
TAS2R46	0.016	RNF145	0.040
TBC1D5	0.016	RRAGD	0.040
C10orf120	0.017	MED12	0.044
FLT4	0.017	CHD5	0.044
SKIV2L2	0.019	UBE2D1	0.046
DDX10	0.021	SLIT1	0.049
OSBPL1A	0.021	GALC	0.050
LGR6	0.024	PDS5B	0.050
SLIT2	0.025	TRERF1	0.050

Table S3. Two hundred and sixty-seven cancer driver genes and corresponding corrected P values predicted by MutsigCV in CY0.

Gene	Corrected P	Gene	Corrected P	Gene	Corrected P
CY0, negative peritoneal lavege cytology.
CXCL5	<0.001	ZMAT2	0.013	FHDC1	0.032
KPNA4	<0.001	GMNN	0.013	PSME4	0.032
RASEF	<0.001	SEMA3A	0.013	MORN3	0.032
AKT3	<0.001	C4BPA	0.013	PRKCI	0.032
OFD1	<0.001	SGIP1	0.014	NPHS1	0.032
LNP1	<0.001	CUL1	0.014	OR56A5	0.032
TM9SF3	<0.001	KIF15	0.015	ZNF609	0.032
TP53	<0.001	DBR1	0.015	PPP1R9B	0.032
TLE1	<0.001	TMBIM6	0.015	CNGB3	0.032
RPP30	<0.001	EEA1	0.015	GREM2	0.032
CCDC158	<0.001	PNPT1	0.016	RASGEF1B	0.032
FSIP2	<0.001	MTAP	0.016	SEC24B	0.033
IMMP2L	<0.001	CALB1	0.016	MTOR	0.033
FLG2	<0.001	MALT1	0.016	RUNX2	0.033
SGCB	<0.001	LHFPL3	0.017	TBC1D4	0.033
PCYT1A	<0.001	CHD4	0.018	DYNC1I2	0.034
FNBP1	<0.001	ABCA5	0.018	SORL1	0.034
COL5A2	<0.001	TCTEX1D1	0.018	CLASP2	0.034
CREBBP	<0.001	SERPING1	0.018	DOCK3	0.034
SLIT2	<0.001	PTPN2	0.018	TECRL	0.034
TTN	<0.001	PDK4	0.018	USP24	0.035
CHD5	0.001	AUTS2	0.018	KDR	0.035
TBX3	0.001	NUF2	0.018	RSPRY1	0.036
RPL22	0.001	DNAH11	0.019	GOT1L1	0.036
CCND1	0.001	CAMK2D	0.019	ASB1	0.037
SAMHD1	0.001	ADAL	0.019	AP1G1	0.037
FBN1	0.001	SERPINB10	0.019	ADAMTS3	0.038
TGFBI	0.001	NAV2	0.020	ATP8A1	0.038
TRAPPC6B	0.001	ATG4C	0.020	NTNG1	0.038
PRR11	0.002	GALC	0.020	CYP4F22	0.038
EED	0.002	DYNC1LI1	0.020	FANCE	0.038
MMP17	0.002	NPAT	0.020	INA	0.038
FGFBP1	0.002	TTC26	0.020	ZC3H18	0.038
EXOC1	0.002	IGSF3	0.020	ABCB4	0.038
EIF2B3	0.002	RERE	0.020	PAPSS1	0.038
ERN1	0.002	KLRB1	0.021	MTR	0.038
SMYD5	0.002	C9orf72	0.021	AIM2	0.038
TTC37	0.003	OR2T4	0.021	ZC3H11A	0.038
MAP2K4	0.003	ITSN1	0.022	SMYD3	0.039
CLOCK	0.003	COL4A4	0.022	ICA1	0.040
CACNA1D	0.003	GABRQ	0.022	SDHB	0.041
NEGR1	0.003	MRPL35	0.023	CCDC7	0.041
ODAM	0.004	SPINK5	0.024	IRF6	0.041
GNA11	0.004	TRERF1	0.024	LPAR3	0.041
UPF2	0.004	ADAM2	0.024	STAT5A	0.041
RNF112	0.005	ILDR1	0.024	KIF13B	0.041
MAPKAP1	0.005	MCFD2	0.024	ROCK2	0.041
C10orf120	0.005	LRP2	0.024	MAP3K3	0.041
DOK6	0.005	LMAN1	0.024	PTPN14	0.041
FGFR1	0.005	DOCK8	0.024	FCN1	0.041
NARS	0.005	NCOR2	0.025	SYK	0.041
KRTAP4-1	0.006	MED12	0.025	TBCCD1	0.041
ZNF354C	0.006	MPL	0.025	TBC1D23	0.042
UACA	0.006	NUP153	0.026	ABCC2	0.042
GNAQ	0.007	PRCC	0.026	PIK3CG	0.042
NBN	0.007	ATP11B	0.027	TMEM14A	0.042
VPS36	0.007	NUP205	0.027	ARHGAP25	0.042
TIGD7	0.007	CENPA	0.027	EAF2	0.042
FAM117B	0.007	NTRK1	0.028	ADCY8	0.042
GRB14	0.007	ATP5F1	0.028	TIGD2	0.042
PLD1	0.007	GPR25	0.028	CCDC28A	0.042
STX2	0.008	OR2L13	0.028	SLC36A1	0.042
DNAH7	0.008	SETD4	0.028	ETV6	0.042
SERINC1	0.009	TRAF7	0.029	NKX6-1	0.042
TRIM33	0.009	RNF20	0.029	LAPTM4A	0.042
UBE2V2	0.009	FAM60A	0.029	SCN3A	0.042
SLC25A28	0.009	TBC1D5	0.029	ABCE1	0.042
NPY	0.010	EPHB3	0.029	GJC2	0.042
AFM	0.011	C6	0.029	ADAMTS20	0.042
DDRGK1	0.011	STAB2	0.029	TRUB2	0.042
WASF3	0.011	ATR	0.029	CDK17	0.042
NUP93	0.011	SLCO1B1	0.029	ERMP1	0.042
FLT1	0.011	ERBB4	0.029	FBXL17	0.042
C15orf39	0.011	DDX27	0.029	IRX4	0.044
C7orf26	0.011	PHIP	0.029	RRH	0.044
ZMYM4	0.011	ASCC3	0.029	KRT4	0.044
ANO3	0.011	PLCG2	0.029	BIRC3	0.044
PPARGC1B	0.011	BEND4	0.029	RHEB	0.044
PKHD1L1	0.012	TMEM40	0.030	GAB1	0.045
KDM4B	0.012	AFF3	0.030	PAXIP1	0.046
SLC20A1	0.012	HM13	0.030	REXO2	0.046
SYCP2	0.012	MMP20	0.030	COL5A3	0.046
ADSS	0.012	TRMT6	0.030	CSMD3	0.048
DHX36	0.012	BRIP1	0.031	RSL1D1	0.048
CPNE4	0.013	DMRTB1	0.031	ATAD2	0.048
TFCP2	0.013	ANLN	0.032	KLHL12	0.049
ASH2L	0.013	CACNA2D1	0.032	IGFN1	0.049
MMEL1	0.013	UFC1	0.032	ABCD2	0.049
RPS6KA2	0.013	HSPA9	0.032	NCOA6	0.049

Figure 2.

Driver genes of CY0 and CY1. (A) Top-scored driver genes of CY1 and CY0 by MutSigCV_v1.35; (B) CY1-specific driver genes were associated with function terms such as Th17 cell differentiation, chagas disease and cell cycle in KEGG enrichment analysis; (C) CY1-specific driver genes were associated with function terms such as protein binding, nucleus and nucleoplasm in GO enrichment analysis. CY0, negative peritoneal lavege cytology; CY1, positive peritoneal lavege cytology; KEGG, Kyoto Encyclopedia of Genes and Genomes; GO, gene ontology.

LASSO regression identified marker genes that distinguish CY1 group from CY0 group

We also employed the LASSO regression model as an alternative approach to identify important genes that could distinguish CY1 group from CY0 group. Based on the input mutation matrix covering all 48 samples and 17,977 genes, the LASSO algorithm identified 43 marker genes. Positive weight indicates higher mutation level in CY1 group, while negative weight indicates higher mutation level in CY0 group (Figure 3A). GO enrichment analysis was also performed on these marker genes (Figure 3B). Interestingly, five terms (ATP binding, cytoplasm, cytosol, protein binding and ventricular septum morphogenesis) were shared between CY1-specific cancer driver genes and LASSO-based CY1 marker genes, suggesting the important roles of these functional gene groups in CY1. It is also noteworthy that, among the 43 marker genes, CD3G and CDKL2 were also predicted as the CY1-specific cancer driver genes by MutsigCV.

Figure 3.

Important genes screened by LASSO regression and corresponding GO enrichment analysis. (A) Positive weight indicates higher mutation level in CY1 group, while negative weight indicates higher mutation level in CY0 group; (B) Five terms (ATP binding, cytoplasm, cytosol, protein binding and ventricular septum morphogenesis) in GO enrichment analysis were shared between CY1-specific cancer driver genes and LASSO-based CY1 marker genes, suggesting the important roles of these functional gene groups in CY1. LASSO, least absolute shrinkage and selection operator; GO, gene ontology; CY1, positive peritoneal lavege cytology; CY0, negative peritoneal lavege cytology.

Discussion

Summary

Cytology-positive gastric cancer is characterized by dismal prognosis, with high risks of PM. However, evidence is lacking on the pathogenic mechanism and signature of CY1 and there is a continuous debate on CY1 therapy. In order to figure out specific driver genes and marker genes of CY1 gastric cancer, and ultimately offer clues for potential marker and risk assessment of CY1, we explored the genomic signature and heterogeneity of CY1 gastric cancer by using whole-exome-sequencing. Seventeen CY1 patients and 31 matched CY0 patients were enrolled in this study. The LASSO algorithm identified 43 marker genes of CY1, while MutSigCV identified 42 cytology-positive specific driver genes. Interestingly, five terms (ATP binding, cytoplasm, cytosol, protein binding and ventricular septum morphogenesis) were shared between CY1-specific cancer driver genes and LASSO-based CY1 marker genes, suggesting the important roles of these functional gene groups in CY1. It is also noteworthy that, among the 43 marker genes, CD3G and CDKL2 were also predicted as the CY1-specific cancer driver genes by MutsigCV. Regarding mutational signatures, driver gene mutation and tumor subclone architecture, there was no significant difference observed between CY1 and CY0 group. From this perspective, CY1 and CY0 gastric cancer might not have distinct differences, and CY1 might represent the progression of CY0, not an independent subtype. This genomic analysis will thus provide key molecular insights into CY1, which may have a direct effect on treatment recommendations for CY1 and CY0 patients, and provides opportunities for genome-guided clinical trials and drug development.

Comparisons with other studies

Nowadays, NGS has facilitated the application of clinical genomics in the diagnosis and treatment of cancer. Pre-NGS sequencing studies on cancer somatic mutations examined only a small set of well-known cancer genes, such as TP53, EFGR and KRAS; and these focused studies had little power to discover novel cancer genes (34). In comparison, NGS is able to generate unbiased, comprehensive rather than biased, limited catalogs of various aberrations in the cancer genomes. Besides, NGS is featured by high sequence, enabling the detection of biological signals in heterogeneous cancer samples (35), and is relatively cost-effective. Therefore, NGS is common in clinical application, offering clues for early diagnosis, elucidate pathogenesis of gastric cancer and identify potential therapeutic targets (36).

Most of the driver genes and marker genes of CY1 in this study are reported to be linked with gastric cancer. For example, TP53 was considered as a driver gene of CY1 in this current study, encoding the tumor suppressor and transcription factor p53. In fact, previous genomic-based studies confirmed the high frequency of TP53 mutations in gastric cancer samples (40%−50%) (18,21). TTN mutation was associated with better response to immune checkpoint blockage in solid tumors (37), while TTK was involved in chromosome alignment at the centromere, respectively, and highly expressed in gastric cancer (38). CHD4 promotes acquired chemoresistance and tumor progression by activating the MEK/ERK axis (39).

Interestingly, this study found that CD3G and CDKL2 were specific in both driver genes and marker genes of CY1. It is well-known that integrity of the T-cell receptor/CD3 (TCR/CD3) complex is crucial for T-cell maturation. In humans, defects in CD3D, CD3E and CD3Z genes cause severe immune deficiency and present early in life with increased susceptibility to infections (40,41). In contrast, CD3G mutations lead to milder phenotypes, mainly characterized by autoimmunity. Research reported that Treg cells of patients with CD3G defects had reduced diversity, increased clonality, and reduced suppressive function (42), similar to the mechanism of tumor happening. As for cyclin-dependent kinase-like 2 (CDKL2), it is a new member of the cyclin-dependent kinase family, located on chromosome 4 (43). According to the publicly available Oncomine database, the CDKL2 level in non-tumor tissues is higher than that in tumor tissues in all reported cancer types (including brain tumor, colorectal cancer, kidney cancer, lung cancer, and breast cancer). Concerning gastric cancer, previous study (44) reported a significant downregulation of the CDKL2 protein in human gastric cancer cells and tissues, and the decreased CDKL2 level was positively correlated with Lauren classification, pathologic staging, histologic type and grade, and short patient survival. Furthermore, CDKL2 downregulation is an unfavorable prognosticator for gastric cancer, and forced CDKL2 expression in human gastric cancer cell lines hindered cell proliferation and impaired invasiveness (44). However, there is no evidence demonstrated CY1’s CDKL2 stage (44).

It should be noted there was no significant difference observed in mutational signatures, driver gene mutation and tumor subclone architecture between the CY1 group and the CY0 group. In fact, identifying cancer driver genes is crucial for understanding their role in driving malignancy and cell transformation (45). Tumor subclone architecture is a direct consequence of the unobservable evolutionary dynamics of tumor growth (46), and mutational signature could reveal environmental and endogenous sources of mutagenesis in each tumor (47). From this perspective, the similarity between CY0 and CY1 in mutational signatures, driver gene mutation and tumor subclone architecture reveals that there is little genomic difference between CY0 and CY1. Furthermore, CY1 is only the progression of CY0, not an independent subtype of gastric cancer.

Moreover, although CY1 is the golden standard for identifying intraperitoneal free cancer cells (IFCC). Cytology has, however, been criticized for its low sensitivity and the interpretive challenge of differentiating well-differentiated carcinoma cells from benign mesothelial cells (48,49). Furthermore, even with more sensitive detection techniques, the tests often fail to identify IFCC, a shortcoming that has significant management and survival implications. According to a systematic review (50), the sensitivities of conventional cytology, immunoassay, immunohistochemisty (IHC), and reverse transcriptase-polymerase chain reaction (RT-PCR) in predicting peritoneal recurrence vary considerably (11.1%−80%, 23%−100%, 22.1%−75%, and 31%−100%, respectively). In fact, patients with CY0 might also experience unexpected peritoneal recurrence after surgery (51), which might undermines the clinical proof it has come to signify. Therefore, from this perspective, differences between CY0 and CY1 might not be distinct, which leads to easy confusion of CY0 and CY1.

However, it should be noted that our study focused on P0CY1, which means “Peritoneal cytology positive for carcinoma cells but no PM”, not P1CY1. In fact, evidence reported that survival is better in patients with positive cytology without other unresectable factors than in those with macroscopic peritoneal dissemination (52,53). Previously, research reported heterogeneity between paired primary and PM samples of P1 gastric cancer at both genomic and proteomic levels, as extensive genomic heterogeneity in somatic mutations, CNVs and subclone structure were observed between primary gastric cancer and paired PM (54,55). In conclusion, it is plausible to say that P0CY1 is similar to P0CY0 and represents a relatively minor progression of P0CY0, while P1 signifies a relatively substantial progression.

Limitations

Our study has some limitations. Firstly, it has been reported that routine sequencing might not adequately reflect the complexity and heterogeneity at the individual patient level (56), and multi-omics or single-cell transcriptome sequencing might potentially provide a more comprehensive understanding of CY1 and CY0. However, it should be noted that with the increasing availability of whole-exome and whole-genome NGS data, the feasibility and efficacy of NGS are largely improving, enabling the assessment of rather small differences among individual patients with cancer, solving the problems to a large extent (57).

Besides, enrichment of CY1 cancer cells for profiling presents a significant challenge as there is currently no effective method to selectively isolate and concentrate these cells. This lack of enrichment options can make it difficult to obtain enough high-quality material for accurate and reliable profiling analysis. However, it should be noted that little difference was observed in NGS. Therefore, it might be considered that P0CY1 is indeed a severe form of P0CY0, rather than a special type, similar with the perspective of peritoneal “soil” for a cancerous “seed” (58).

Conclusions

Cytology-positive and negative gastric cancer might not exhibit distinct differences, and cytology-positive cases might represent the progression of cytology-negative gastric cancer, rather than constituting an independent subtype. This genomic analysis will thus provide key molecular insights into CY1, which may have a direct effect on treatment recommendations for CY1 and CY0 patients, and provides opportunities for genome-guided clinical trials and drug development.