Chemokine growth‐regulated oncogene 1 as a putative biomarker for gastric cancer progression

Abstract

Gastric cancer (GC) is a heterogeneous disease that is not well detected by current tumor markers. Identifying molecular markers that can predict the potential for tumor progression is important for appropriate individualized therapy. Using the Cancer Metastasis Research Center microarray database (17K cDNA microarray), we identified genes that were differentially expressed between 96 cancer and 98 normal gastric tissues using significant analysis of microarrays. From these, we selected genes that were overexpressed more than twofold in tumor tissues that encode secreted proteins. The selected genes were validated with ELISA using the sera of 96 GC patients and 48 healthy donors. Our first round of selection included 6510 genes that were differentially expressed between 96 cancer and 98 normal gastric tissues with a minimal false discovery rate of 0.005%. Out of those genes, we picked 386 that encoded secreted proteins based on the SOURCE database. Of these genes, we focused on 55 that were overexpressed more than twofold in GC compared to normal tissues. With Ingenuity Pathway Analysis, we found 34 genes related to cancer. One in particular, chemokine growth‐regulated oncogene 1, CXCL1, has been linked to cancer progression in various cancer types, but not yet to GC. Levels of CXCL1 in serum samples of GC patients were significantly higher compared with healthy donors (P < 0.05). Within GC patients, CXCL1 serum levels increased according to tumor stage and lymph node metastasis. The CXCL1 gene appears to be a candidate marker for GC progression. (Cancer Sci 2010)

Gastric cancer (GC) is one of the most fatal cancers, with approximately 880 000 new cases and 650 000 deaths per year worldwide.^{( 1} , ^{2 )} Although the development of new treatment strategies over the past few years have yielded a slight improvement in survival, GC remains one of the most common causes of cancer death.^{( 3} , ^{4 )} Radical surgery is the only potentially curative method for localized disease, with 5‐year survival rates of 70–95%, but patients who harbor advanced tumors have poor clinical outcomes with 5‐year survival rates of only 20–30%.^{( 5 )} One‐ and five‐year survival rates of stage I and II patients are 93% and 67%, respectively, compared to the much lower rates of 31% and 8% in stage III and IV patients, respectively.^{( 6 )}

Advances aimed at increasing survival rates are being made, such as in drug development. Multimodal approaches are also being improved, including neoadjuvant or adjuvant treatment with chemotherapy and radiotherapy or biomarker development for early detection or prognosis prediction. Investigation of molecular changes such as gene activation or overexpression that might occur during carcinogenesis and cancer progression can provide new insights into the disease and may identify molecular factors that could be used as novel prognostic biomarkers or therapeutic targets.

Among several approaches for investigating specific molecular changes, high‐throughput screening methods have increased the chances of identifying novel molecules. In addition, integration of gene expression and genetic changes boosts the sensitivities and specificities of molecular markers by reflecting tumor biology. Based on the high‐throughput database of the Cancer Metastasis Research Center, which includes gene expression and copy number profiles of tissues and cell lines, various approaches have been used to identify novel molecular markers and therapeutic targets in GC.^{( 7} , ⁸ , ⁹ , ¹⁰ , ^{11 )} We established a procedure for identifying potential biomarkers from the high‐throughput database and for validating their roles as diagnostic and prognostic markers in GC.

Chemokine growth‐regulated oncogene 1 (CXCL1) is a chemokine in the CXC family that promotes chemotaxis of granulocytes and endothelia. Since its identification in 1985 as an autostimulatory melanoma mitogen from the human malignant melanoma cell line,^{( 12 )} CXCL1 has been further studied in cancer because of its relationship with cell transformation, cell growth, angiogenesis, and metastasis in various cancer types.^{( 13} , ^{14 )} It is a potent mediator of tumor‐associated angiogenesis in Kaposi’s sarcoma and non‐small cell lung cancer,^{( 15} , ¹⁶ , ^{17 )} and its expression is associated with increased densities of leukocytes and blood vessels in a mouse model.^{( 18 )} The role of CXCL1 in GC has not yet been described.

In this study, we identified a novel and practical biomarker from the secretory proteins in a genome‐wide gene expression database of GC and adjacent normal tissues. We further evaluated its expression in the peripheral blood of GC patients, which can be directly translated into clinical practice.

Materials and Methods

Tissue and serum samples.  All experiments using patient tissues and sera were carried out with the approval of the Severance Hospital IRB, Yonsei University Health System (Seoul, Korea). Ninety‐six tumor and 98 adjacent normal tissues from patients who underwent surgery and were followed for at least 5 years after surgery at the Yonsei University Health System (1997–1999) were used for cDNA microarrays. Tumor tissues had at least 70% tumor content, and all tissues were stored in liquid nitrogen immediately after surgery. The total sample population of 194 tissues consisted of 85 paired and 13 normal and 11 tumor unpaired gastric tissues (Table 1). As an independent validation set, which were a different set of samples from those that were used to carry out cDNA microarrays, 96 serum samples were obtained from GC patients treated at the Yonsei University Health System, 2007–2009. All blood samples were collected at the time of diagnosis, prior to curative surgery. Patients’ clinical information was collected through medical records and American Joint Committee on Cancer (ACCC) Cancer Staging Manual (6th edition) was used for TNM classification. For normal controls, 48 sera from samples were donated by 29 male and 19 female healthy volunteers with a median age of 35 years.

Table 1.

 Clinical characteristics of gastric cancer patients who participated in this study

Patients for microarray analysis (n = 96)
Gender
Male	68
Female	28
Age (years)
Median (range)	65 (30–90)
TNM stage
I	9
II	20
III	33
IV	34
Lauren classification
Intestinal	12
Diffused	43
Mixed	33
Unkonwn	8

Patients for ELISA of CXCL1 (n = 96)
Gender
Male	53
Female	43
Age (years)
Median (range)	54 (39–75)
CXCL1
Median (range)	111.9 (37.4–578.7)
CEA
Median (range)	1.5 (0.03–111.60)
CA72‐4
Median (range)	1.4 (0.0–95.4)
CA19‐9
Median (range)	7.8 (0.1–2080.0)
Histology
Well differentiated	4
Moderately differentiated	18
Partially differentiated	28
Other	46
TNM stage
I	32
II	24
III	20
IV	20
Lauren classification
Intestinal	31
Diffused	50
Mixed	4
Unknown	11
Lymphovascular invasion
Negative	43
Positive	49
Neural invasion
Negative	42
Positive	50

CEA, carcinoembryonic antigen; CXCL1, chemokine growth‐regulated oncogene 1.

cDNA microarray experiments.  The total RNA was extracted from the homogenized tissues using TRIzol (Invitrogen, Carlsberg, CA, USA) reagent according to the manufacturer’s protocol and was further purified using the RNeasy Mini Kit (Qiagen, Valencia, CA, USA). Gene expression profiling was carried out on cDNA microarrays (GenomicTree, Daejeon, Korea) containing 17 000 human probes using a previously described method.^{( 10 )} Microarray hybridization was carried out in an indirect‐design. Each of the cDNA targets generated from tissue total RNAs (Cy5‐labeled) were competitively hybridized with cDNAs from the common reference RNA (Cy3‐labeled) following Cancer Metastasis Research Center protocols.^{( 19 )} Common reference RNA was prepared by pooling equivalent amounts of total RNAs from the AGS, MDA‐MB231, HCT 116, SK‐HEP‐1, A‐549, HL‐60, MOLT‐4, HeLa, Caki‐2, U‐87MG, SK‐MEL‐2, and Capan‐2 cell lines.^{( 20 )} Hybridized slides were scanned using a GenePix 4000B laser scanner (Axon Instruments, Union City, CA, USA).

Microarray data analysis.  After Lowess normalization, genes with more than 20% missing values in all experiments were filtered, resulting in no missing proportion of 80%, and the missing values were compensated for using the K‐NN imputation method using significance analysis of microarrays (SAM).^{( 21 )} To identify significant genes differentially expressed between normal and tumor gastric tissues, we carried out two‐class SAM using microarray data from all 196 normal and tumor samples. Selected genes were internally validated with leave‐one‐out cross‐validation based on a support vector machine algorithm. Hierarchical clustering analysis and visualization of the resulting dendrogram were done with GeneSpring 7.0 software (Silicon Genetics, Redwood City, CA, USA). Clustering was carried out by a complete linkage algorithm with an uncentered correlation. To find genes that encoded secretory proteins, we annotated the selected genes using the Stanford Online Universal Resource for Clones and Expressed sequence tags (source) (http://smd.stanford.edu/cgi‐bin/source/sourceSearch) and Ingenuity Pathways Analysis (IPA) (Ingenuity Systems, Redwood City, CA, USA).

Measurement of CXCL1 expression in serum by ELISA.  Human serum samples were quantitatively analyzed for CXCL1 with a human CXCL1/Gro‐a Quantikine ELISA kit (R&D Systems, Minneapolis, MN, USA) according to the manufacturer’s instructions. In brief, 200 μL of serum samples were added directly into wells in duplicate. After antibody conjugation and luminescence reaction, the procedure was completed with stop solution, and we measured the optical densities of each well at 450 nm and 570 nm. Final optical density was calculated by subtracting the readings at 570 nm from the readings at 450 nm. We measured standard curves in every plate with recombinant human CXCL1 in order to eliminate inter‐variation. In the analysis, we first determined the cut‐off value of CXCL1 higher than two standard deviations of the normal control’s mean level. Based on the determined cut‐off value, the positivity rate of CXCL1 was calculated. We evaluated CXCL1 levels based on different stages, number of lymph node metastases, and recurrence and compared these with other tumor markers such as carcinoembryonic antigen (CEA), CA19‐9, and CA72‐4. Differences were compared using Student’s t‐test, and the correlation coefficient was calculated by Pearson’s method. Univariate analyses were also carried out with CXCL1 and various clinical parameters using the Mann–Whitney U‐test and chi‐squared test. All statistical analysis was carried out using spss 13.0 software (SPSS, Chicago, IL, USA).

Results

Differentially expressed genes in GC.  After the preprocessing of 194 microarray results, we obtained 14 124 genes for further gene selection analysis. With these genes, we carried out two‐class SAM analysis based on 1000 permutations to identify differentially expressed genes between the 98 normal and 96 tumor gastric tissues. A significant gene set of 6510 genes was selected with a minimum false discovery rate of 0.005%. The selected gene set accurately divided the gastric tissues into normal and tumor groups (Fig. 1). To validate the classification, leave‐one‐out cross‐validation was carried out using a support vector machine algorithm. As a result, 6510 genes correctly predicted 191 of 194 samples with a 98.5% prediction rate, indicating that these genes could represent significant signatures determining the genetic features of GC (Table S1).

Figure 1.

 Differentially expressed genes between gastric normal and tumor tissues. Hierarchical clustering of 6510 selected genes from 194 gastric tissues (98 normal and 96 tumors). Red, tumor tissues; yellow, normal tissues.

Because our ultimate objective was to identify useful biomarkers that could be practically applied to patients through simple blood testing, we picked 386 genes out of the selected 6510 genes that were annotated as secretory proteins in the source database. We then focused on 55 genes that were more than twofold overexpressed in GC compared to normal tissue. This selected gene set classified the two groups with a prediction rate of 92.3% (data not shown). To determine which of the 55 genes were suitable biomarker candidates, we evaluated the genes with several criteria including functional annotations, a search of published reports for their relationship with cancer progression, and the raw microarray signal intensity that indicated the detectable level of expression. Among the 55 genes, 34 were significantly linked to cancer by biofunction in a core analysis in IPA (Tables 2 and S2). The subfunctions of these 34 genes in cancer were further analyzed (Table S3). The genes were included in the categories that were important factors in cancer progression, based on previous research by IPA. After final evaluation of the raw microarray signal intensities and the search of published reports (data not shown), we ultimately selected CXCL1, also known as the GRO1 oncogene, as a novel biomarker candidate for GC.

Table 2.

 Thirty‐four genes overexpressed more than twofold in cancer tissues and linked to biofunctions of cancer

GenBank Accession ID	Gene name	Gene symbol	Chromosomal location	Fold (T/N)
AA991590	Apolipoprotein C‐I	APOC1	19q13.2	3.64
AA873159	Apolipoprotein C‐I	APOC1	19q13.2	2.16
AA434115	Chitinase 3‐like 1	CHI3L1	1q32.1	3.66
N81029	Collagen, type XVIII, alpha 1	COL18A1	21q22.3	2.26
AA490172	Collagen, type I, alpha 2	COL1A2	7q22.1	5.25
AA150402	Collagen, type IV, alpha 1	COL4A1	13q34	2.59
AA430540	Collagen, type IV, alpha 2	COL4A2	13q34	2.16
R62603	Collagen, type VI, alpha 3	COL6A3	2q37	6.57
AA598507	Collagen, type VII, alpha 1	COL7A1	3p21.1	3.27
AI337428	Cartilage oligomeric matrix protein	COMP	19p13.1	2.68
R45054	Corticotropin releasing hormone	CRH	8q13	2.20
W46900	Chemokine (C‐X‐C motif) ligand 1	CXCL1	4q21	2.90
AA878880	Chemokine (C‐X‐C motif) ligand 10	CXCL10	4q21	2.01
AA131406	Chemokine (C‐X‐C motif) ligand 9	CXCL9	4q21	2.89
AI394706	Fibronectin 1	FN1	2q34	4.27
AA489587	Fibronectin 1	FN1	2q34	4.05
AI623173	Galanin prepropeptide	GAL	11q13.2	2.46
AA456621	Gamma‐glutamyl hydrolase	GGH	8q12.3	2.22
T53298	Insulin‐like growth factor binding protein 7	IGFBP7	4q12	2.25
AI925826	Inhibin, beta A	INHBA	7p15‐p13	16.61
AI283230	Inhibin, beta A	INHBA	7p15‐p13	2.09
R50354	Leukemia inhibitory factor	LIF	22q12.2	2.02
AA676458	Lysyl oxidase‐like 2	LOXL2	8p21.3‐p21.2	2.52
AA968896	Midkine	MDK	11p11.2	2.38
AA143331	Matrix metallopeptidase 1	MMP1	11q22.3	2.92
AA954935	Matrix metallopeptidase 11	MMP11	22q11.2|22q11.23	5.01
AA045500	Matrix metallopeptidase 11	MMP11	22q11.2|22q11.23	2.89
AA031514	Matrix metallopeptidase 7	MMP7	11q21‐q22	4.33
AA488406	Mesothelin	MSLN	16p13.3	3.13
H65030	Phospholipase A2, group VII	PLA2G7	6p21.2‐p12	2.40
AA284669	Plasminogen activator, urokinase	PLAU	10q24	3.60
R45941	Protein tyrosine phosphatase, receptor type, N	PTPRN	2q35‐q36.1	2.23
AI989728	Serpin peptidase inhibitor, clade B, member 5	SERPINB5	18q21.3	2.91
AA775616	Secreted phosphoprotein 1	SPP1	4q21‐q25	9.76
AW075162	TIMP metallopeptidase inhibitor 1	TIMP1	Xp11.3‐p11.23	4.68
AA194983	Tumor necrosis factor receptor superfamily, member 11b	TNFRSF11B	8q24	4.03
AI473336	WNT1 inducible signaling pathway protein 1	WISP1	8q24.1‐q24.3	2.25
AA922800	WNT1 inducible signaling pathway protein 1	WISP1	8q24.1‐q24.3	2.17
N78828	Wingless‐type MMTV integration site family member 2	WNT2	7q31	3.01

N, normal tissue; T, tumor tissue.

Serum levels of CXCL1 in GC patients.  To validate CXCL1 levels in serum samples, we tested 96 GC patients and 48 normal controls as described in “Materials and Methods”. We observed that mean CXCL1 levels in the sera from GC patients (146.6 pg/mL) were elevated compared with those of the normal controls (98.2 pg/mL) (P < 0.001) (Fig. 2A). We determined the cut‐off value between normal and cancer tissue as 196 pg/mL, which was higher than two standard deviations compared to the mean level of normal controls. When we carried out the univariate analysis with various clinical parameters and CXCL1, based on the cut‐off level, we did not find any significant correlation. However, although not significant, CXCL1 compared to lymphovascular invasion showed moderate correlation (Table 3). We then compared CXCL1 positive rates in GC patients with the tumor markers CEA, CA72‐4, and CA19‐9, which are currently used for GC. Based on the current standard cut‐off values for each tumor marker (CEA, 5 ng/mL; CA72‐4, 8.2 U/mL; CA19‐9, 37 U/mL), the positive rates in the 96 GC patients were as follows: 4/96 (4.2%), 11/96 (11.5%), and 6/96 (6.3%), respectively, whereas the CXCL1 positive rate was 22/96 (22.9%) (Table 4). When we added CXCL1 as a tumor marker, the frequency of patients with positivity in any tumor markers was increased from 18/96 (18.8%) to 36/99 (37.5%). These results support the possibility that CXCL1 could be a novel and additional biomarker in GC in accordance with current tumor markers.

Figure 2.

 Chemokine growth‐regulated oncogene 1 (CXCL1) levels in serum samples from 48 healthy donors and an independent set of 96 gastric cancer patients. (A) Mean CXCL1 level in serum samples from gastric cancer patients were higher (146.6 ± 107 pg/mL) than that in healthy donors (98.2 ± 49.1 pg/mL) (P < 0.001). (B) Levels of CXCL1 in serum samples according to tumor stage. The levels were significantly elevated in stage III and IV compared with that of the normal controls. *P < 0.05 by Student’s t‐test.

Table 3.

 Univariate analysis of various clinical parameters and chemokine growth‐regulated oncogene 1 (CXCL1)

	Negative CXCL1	Positive CXCL1	P‐value
Continuous variable
Age
Median (range)	54 (39–75)	56 (46–71)	0.30
CEA
Median (range)	1.5 (0.03–11.60)	1.2 (0.38–5.14)	0.62
CA72‐4
Median (range)	1.4 (0.0–95.4)	1.4 (0.8–12.5)	0.81
CA19‐9
Median (range)	9.1 (0.1–2080.0)	5.8 (0.1–78.8)	0.59
Categorical variable
Gender
Male, female	60%, 40%	45.2%, 54.8%	0.17
T stage
1, 2, 3, 4	24.3%, 29.7%, 43.2%, 66.7%	18.2%, 40.9%, 36.4%, 4.5%	0.73
N stage
0, 1, 2, 3	37.8%, 33.8%, 18.9%, 9.5%	13.6%, 40.9%, 22.7%, 22.7%	0.12
M stage
1, 2	91.9%, 8.1%	90.9%, 9.1%	0.88
Histological type
Well, moderate, partial differentiation, other	5.4%, 17.6%, 28.4%, 48.6%	0%, 22.7%, 31.8%, 45.4%	0.82
Lauren classification
Intestinal, diffused, mixed, unknown	30.9%, 55.9%, 5.9%, 7.4%	45.5%, 54.5%, 0%, 0%	0.27
Lymphovascular invasion
Negative, positive	52.1%, 47.9%	28.6%, 71.4%	0.06
Neural invasion
Negative, positive	43.5%, 56.5%	50%, 50%	0.56

Lauren classification includes diffuse, intestinal or mixed type. CEA, carcinoembryonic antigen.

Table 4.

 Positivity of known tumor markers and chemokine growth‐regulated oncogene 1 (CXCL1)

	Positive frequency (n = 96)
CEA	4 (4.2%)
CA72‐4	11 (11.5%)
CA19‐9	6 (6.3%)
CXCL1	22 (22.9%)
CEA and/or CA72‐4	14 (14.6%)
CEA and/or CA19‐9	9 (9.4%)
CA72‐4 and/or CA19‐9	15 (15.6%)
CEA and/or CA72‐4 and/or CA19‐9	18 (18.8%)
CEA and/or CA72‐4 and/or CA19‐9 and/or CXCL1	36 (37.5%)

CEA, carcinoembryonic antigen.

Gastric cancer progression marker.  To determine whether CXCL1 levels in serum samples are associated with progression of GC, we first classified the tumor samples according to TNM stages I, II, III, or IV (Fig. 2B). Levels of CXCL1 were significantly higher in stage III and IV tumors than in normal tissue (P < 0.01). Overall, CXCL1 levels in the sera of GC patients were significantly elevated in late stage GC compared to earlier stages.

We next evaluated whether CXCL1 levels in sera had an association with the number of lymph node metastases, because lymph node metastasis is one of the most significant prognostic factors reflecting gastric tumor progression. First, we calculated the Pearson’s correlation coefficient between the two parameters in the 96 GC patients and found a significant positive correlation (r = 0.29; P = 0.004) (Fig. 3A). We then confirmed levels in the sera of healthy donors and GC patients with or without lymph node metastasis. Patients with lymph node metastasis were divided into low (<15) and high (≥15) groups according to the number of metastatic nodes by the standard of the American Society of Clinical Oncology (ASCO) (Fig. 3B). There was no difference between normal and node‐negative patients. In contrast, CXCL1 levels in node‐positive patients were significantly higher than in normal controls (P < 0.05). Moreover, CXCL1 levels in the high metastasis group were also significantly greater than in node‐negative patients (P < 0.05). Next, we evaluated CXCL1 levels in patients of the same stage (low or high) with or without lymph node metastasis. In the serum samples of stage IB and stage II cases, we successfully identified the samples with lymph node metastasis numbers of zero (low, n = 10) and those with more than two (high, n = 10). Levels of CXCL1 had a tendency to be greater in the high lymph node metastasis group than in the low lymph node metastasis group (Fig. S1A,B). In stage IV tumors, CXCL1 levels were higher in the high metastasis group (>15 metastases) (Fig. S1C).

Figure 3.

 Chemokine growth‐regulated oncogene 1 (CXCL1) levels in serum samples from gastric cancer patients with or without lymph node metastasis. (A) Correlation of CXCL1 levels in serum samples and number of lymph node metastases in 96 gastric cancer patients. There was a significant positive correlation (r = 0.29; P = 0.004). (B) Levels of CXCL1 in groups of normal donors and gastric cancer patients with or without lymph node metastasis (LN mets; low, 0–15 metastases; high, >15 metastases). The CXCL1 levels in node‐positive patients were significantly elevated compared with those of the normal controls, and the high metastasis group had higher CXCL1 levels than did the node‐negative patients. *P < 0.05 by Student’s t‐test.

In addition, the rate of CXCL1 positivity was higher in patients with lymph node metastasis than in those without (P < 0.05) (Table 5). Within node‐positive patients, the frequency tended to be higher in patients with more metastases. Although the positive rates of other tumor markers also had a tendency to increase in node‐positive patients, there were no significant differences. Taken together, these data suggest that CXCL1 could increase the detection of GC when combined with other tumor markers and could itself be a marker for GC progression.

Table 5.

 Positivity of known tumor markers and chemokine growth‐regulated oncogene 1 (CXCL1) in gastric cancer patients based on lymph node metastasis status

Lymph node metastasis	CEA	CA72‐4	CA19‐9	CXCL1
Negative (n = 31)	0	2 (6.5%)	1 (3.2%)	3 (9.7%)
Positive (n = 65)	4 (6.2%)	9 (13.8%)	5 (7.7%)	19 (29.2%)
Low (<15) (n = 53)	4 (7.5%)	9 (17%)	3 (5.7%)	14 (26.4%)
High (≥15) (n = 12)	0	0	2 (16.7%)	5 (41.7%)

CEA, carcinoembryonic antigen.

Discussion

High‐throughput screening of protein, DNA, RNA, and chemical compounds, mainly through proteomics and genomics, has been used to identify novel biomarkers. There are various methods for selecting significant molecules from high‐throughput screening, depending on the purpose of the research. One method selects from a large set of molecules classified by function in order to identify significant markers. Alternately, molecules with high expression levels can be selected for final candidates. In this study, our ultimate objective was to identify novel candidate biomarkers from high‐throughput gene expression data that could be applied to blood samples of GC patients. We first selected genes that encoded secretory proteins, out of 6510 genes, by applying a fold‐change threshold. Using this method, we selected CXCL1 as a final potent biomarker candidate for GC.

Although cancer‐related functions of CXCL1 have been previously reported in several cancer types, its functions in GC have not been elucidated. We found higher levels of CXCL1 in serum samples of GC patients than in healthy donors, suggesting the possibility of a serum tumor marker for the diagnosis of GC.

When we compared CXCL1 serum levels with known tumor markers, we observed unusually low positive rates of CEA (4.2%), CA72‐4 (11.5%), and CA19‐9 (6.3%) in our cohort. Although positive rates of CEA, CA72‐4, and CA19‐9 in early GC have been reported at 9.1%, 2.3% and 11.4%, respectively, rates over 15% have been reported in advanced or overall GC.^{( 22} , ²³ , ^{24 )} The differences between published rates and our rates may be due to the random selection of samples and the small number of GC patients in our study. Even so, the addition of CXCL1 increased the sensitivity of GC detection.

We next examined whether the CXCL1 level could serve as a progression marker in different stages of GC. We observed that the levels were significantly more elevated in high‐stage GC, indicating that serum level of CXCL1 could be a progression marker as well as a diagnostic marker in GC. It is already a progression marker in renal cell carcinoma.^{( 25 )} CXCL1 may correlate to GC progression through its relationship with angiogenesis, as CXCL1 correlates with vascular endothelial growth factor (VEGF). In 2004, Bobrovnikova‐Marjon et al. ^{( 26 )} revealed that glutamine deprivation induces VEGF secretion and increases NFκB and activator protein‐1 DNA‐binding activities, and that CXCL1 is an upregulated target of NFκB. Caunt et al. ^{( 27 )} reported that CXCL1 enhances angiogenesis through angiogenic factors including VEGF, angiopoietin‐2, and MMP. Treatment with anti‐CXCL1 antibody or shRNA reduces angiogenesis and inhibits the expression of angiogenic factors. Based on these reports, we inferred that CXCL1 promotes GC progression by inducing angiogenesis, and that angiogenic factors such as VEGF mediate this process. Well‐known key molecules of signal transduction such as epidermal growth factor receptor, HER‐2, Ras, AKT, MAPK, and NFκB have been identified as regulators of CXCL1.^{( 28} , ²⁹ , ³⁰ , ³¹ , ^{32 )}

Shintani et al. ^{( 33 )} noted that CXCL1 mRNA levels tend to be higher in node‐positive oral squamous cell carcinoma tumors than in node‐negative tumors. We also observed significantly elevated CXCL1 serum levels in patients with lymph node metastasis compared with node‐negative patients, and higher CXCL1 levels correlated with greater numbers of metastatic nodes. As lymph node metastasis is one of the most significant prognostic factors in GC, CXCL1 is an excellent candidate prognostic biomarker for GC with long‐term follow‐up. However, as this was a retrospective study, we could not predict CXCL1 in patient survival because information for many patients was not available through their medical records.

In addition to cancer progression, CXCL1 is well known as an inflammation‐related protein that is elevated in inflammation‐related diseases.^{( 34} , ³⁵ , ^{36 )} Based on the close relationship between cancer and inflammation, ⁽³⁷⁾ our results must be interpreted and applied to the clinical setting with caution. Further study is needed to conclusively determine whether CXCL1 is directly related to either carcinogenesis or the progression of GC, or is only indirectly connected through inflammation.

The other 33 genes we selected that were related to cancer may also be potential candidate markers. Many of the genes were grouped into the cytokine, collagen, and MMP families (Table S3), and some are connected to each other or to other key molecules that are well known within cancer biology (Fig. S2). We plan to further investigate the genes in the connection map, particularly CXCL10, FN1, SPP1, SERPINB5, LIF, and INHBA, which are connected with CXCL1 or VEGF.

In summary, we identified CXCL1 as a candidate marker for GC progression using high‐throughput genomics and validation with serum samples. A recently published study supported this result, noting overexpression of CXCL1 in GC patients,^{( 38 )} and we further showed that CXCL1 could be an important factor in GC progression and aggressive behavior, such as lymph node metastasis. We propose that adding CXCL1 as a biomarker could increase the detection rate and predict poor prognoses of GC. We also identified other genes worth exploring for links to GC progression through validations.

Supporting information

Fig. S1. Chemokine growth‐regulated oncogene 1 (CXCL1) levels in serum samples from gastric cancer patients with low and high numbers of lymph node metastases. (A, B) In serum samples from stage IB (A) and II (B) patients, CXCL1 levels tended to be higher with more metastatic lymph nodes. High, metastasis in two or more lymph nodes; Low, no metastasis. (C) Levels of CXCL1 also tended to be elevated in the high metastasis group of stage IV patients. High, >15 lymph node metastases; Low, 0–15 lymph node metastases.

Fig. S2. Protein network of the 34 selected cancer‐related genes and other important molecules in cancer. Nineteen of the 34 genes are connected to each other or to other important well‐known molecules in cancer biology.

Table S1. Cross‐validation result of the selected 6510 genes.

Table S2. Bio‐functions of 55 genes in diseases and disorders.

Table S3. Bio‐functions of 34 genes in cancer.

Please note: Wiley‐Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

Supporting info item

Supporting info item

Supporting info item

Supporting info item

Supporting info item