Development and Validation of a New Prediction Criteria for Neoadjuvant Chemotherapy Response in Locally Advanced Gastric Cancer Based on Ferroptosis‐Related Biomarkers

ABSTRACT

Background

The efficacy of neoadjuvant chemotherapy (NAC) combined with surgery and postoperative adjuvant chemotherapy in improving the prognosis of locally advanced gastric cancer (LAGC) has been confirmed. Nevertheless, there is still no effective tool to predict the response to NAC in gastric cancer patients.

Methods

This study included a total of 184 patients (118 in the training cohort and 66 in the testing cohort) treated with oxaliplatin + fluorouracil‐based NAC from January 2010 to July 2022. Immunohistochemistry staining of biomarkers was carried out on the endoscopy biopsy specimens before receiving treatment. Logistic regression was performed to explore the predictive biomarkers of response to NAC, and internal verification was carried out by using the bootstrap method.

Result

Ferroptosis increased significantly in NAC‐sensitive LAGC. Among the molecular markers related to ferroptosis, Fe²⁺, ACSL4, DHODH, GCH1, and FSP1 showed significant differences between the sensitive and nonsensitive groups. Logistic regression analysis revealed Fe²⁺, ACSL4, DHODH, and GCH1 as predictors of the efficacy of NAC. A nomogram prediction model was established based on identified biomarkers. The area under the ROC curve (AUC) of the nomogram model was 0.89, which indicated that the prediction results of the model were highly consistent with the actual situation (Hosmer–Lemeshow test p = 0.684). The AUC of the nomogram model in the external validation set (N = 66) was 0.82, indicating that the model has certain predictive ability.

Conclusion

This study revealed the critical value of biomarkers related to ferroptosis serving as novel predictive biomarkers for patients with LAGC receiving NAC.

Abbreviations

4HNE

4‐hydroxynonenal

ACSL4

acyl‐CoA synthetase long‐chain family member 4

AIFM2/FSP1

ferroptosis suppressor protein 1

DHODH

dihydroorotate dehydrogenase

GPX4

glutathione peroxidase 4

GSH

glutathione

LAGC

locally advanced gastric cancer

LPCAT3

lysophosphatidylcholine acyltransferase 3

MDA

malondialdehyde

NAC

neoadjuvant chemotherapy

OS

overall survival

PFS

progression‐free survival

pCR

pathologic complete response

PUFA‐PLs

polyunsaturated‐fatty‐acid‐containing phospholipids

TRG

tumor regression grade

xCT/SLC7A11

solute carrier family 7 member 11

1. Introduction

Gastric cancer is the fifth largest malignant tumor in the world, and the third leading cause of cancer death [1]. The proportion of early gastric cancer (EGC) is low in China, and most patients are diagnosed with advanced or late disease. Patients with advanced gastric cancer are treated with surgery combined with drugs. Despite recent advances in drug therapy for gastric cancer, the overall response rate to chemotherapy is still less than 50% [2, 3, 4].

Neoadjuvant chemotherapy (NAC) has become one of the most important strategies for patients with locally advanced gastric cancer (LAGC), because it not only reduces tumor size and thereby improves the rate of R0 resection during operation but also facilitates assessment of individual responses perioperatively. Based on the results of the MAGIC trial and the FNCLCC/FFCD clinical trial, the National Comprehensive Cancer Network (NCCN) recommends neoadjuvant therapy (including NAC and neoadjuvant chemoradiotherapy) for gastric cancer patients with stage above cT2 and no distant metastasis, and the main chemotherapy regimen is FLOT [5, 6, 7]. The European Society for Medical Oncology (ESMO) recommends perioperative chemotherapy and radical surgery for patients with gastric cancer higher than stage IB [8].

A great number of clinical trials have revealed that patients who achieve pathological complete response (pCR) after NAC have better prognosis than those who do not [9, 10]. However, patients are still trapped in a low proportion of pCR currently. To improve pCR rate is a vital goal of NAC. Therefore, identifying patients with superior response to NAC in early time naturally becomes the main focus in the research on gastric cancer at this stage.

In recent years, ferroptosis has gained a lot of interest as a new form of programmed cell death, characterized by iron‐dependent accumulation of lipid reactive oxygen species (ROS) to lethal levels [11, 12]. As a programmed cell death mode, ferroptosis may potentially reflect patients' sensitivity to chemotherapy. The occurrence conditions of ferroptosis mainly include the synthesis and peroxidation of polyunsaturated‐fatty‐acid‐containing phospholipids (PUFA‐PLs), iron metabolism, and mitochondrial metabolism. Iron‐dependent lipid peroxidation is the key step in ferroptosis; regulation of lipid metabolism is a key determinant of ferroptosis susceptibility. ACSL4 (acyl‐CoA synthetase long‐chain family member 4) and LPCAT3 (lysophosphatidylcholine acyltransferase 3) promote the development of PUFA (polyunsaturated fatty acids) combined with phospholipids to form PUFA‐PLs, which are susceptible to the oxidation of free radicals mediated by arachidonic acid lipoxygenases (ALOXs) [12]. Doll et al. found that the expression of ACSL4 appears to be a predictive marker for ferroptosis sensitivity in different cellular contexts [13].

Unbalanced regulation of iron metabolism in cells can promote or inhibit ferroptosis. Ferrous iron (Fe²⁺) exists in the form of an unstable intracellular iron pool, and its increase and decrease can correspondingly promote or inhibit the occurrence of ferroptosis [14]. Ferroptosis is regulated by the accumulation of intracellular ROS and metabolic pathways related to glutathione depletion, including iron metabolism, lipid peroxidation, and glutathione metabolism. The balance between the ferroptosis induction pathway and defense systems also affects the degree of ferroptosis [11]. The classical cellular defense mechanism of ferroptosis is through the glutathione peroxidase 4 (GPX4)–solute carrier family 7 member 11 (SLC7A11/xCT) signaling axis. Sha R. et al. revealed the critical value of ACSL4 and GPX4 serving as novel predictive and prognostic biomarkers for patients with breast cancer receiving NAC [15]. Ouyang et al. revealed that ferroptosis negative regulation (FNR) signatures were correlated with the chemoresistance of gastric cancer [16]. Therefore, we hypothesized that these biomarkers' expression status might help predict neoadjuvant chemosensitivity for patients with LAGC, which was elucidated in this retrospective study of our prospective clinical trials.

2. Methods

2.1. Study Cohorts

In this study, we included adult patients with histologically confirmed LAGC without distant metastasis (M0). SOX/XELOX was administered every 3 weeks for three cycles. Oxaliplatin was administered intravenously at a dose of 130 mg/m² on Day 1 of the 21‐day cycle. S‐1 was administered orally at 80 mg/m² (80–120 mg/day total dose depending on the patient's body surface area as follows: < 1.25 m², 80 mg; 1.25–1.5 m², 100 mg; and > 1.5 m², 120 mg). Capecitabine was taken in two daily doses of 1250 mg/m² each (total daily dose of 2500 mg/m²), approximately 12 h apart. S‐1/Capecitabine was administered daily for 14 days, followed by a 7‐day rest period. D2 surgery was conducted sequentially after NAC. Between January 2010 and January 2019, 118 patients with qualified biopsy specimens before NAC and treated with surgery were available for the analysis. The research was ethically approved by the First Affiliated Hospital, Sun Yat‐sen University (approval ID [2023]082).

Inclusion criteria include (1) age ≥ 18 years old; (2) preoperative pathological diagnosis of gastric adenocarcinoma; (3) no preoperative radiotherapy, immunotherapy, targeted therapy, or other antitumor therapies were received; (4) complete medical records.

Exclusion criteria include (1) recurrent gastric cancer and remnant gastric cancer; (2) underwent emergency surgery due to tumor bleeding, perforation, or obstruction during NAC; (3) with unresectable distant metastatic lesions; (4) complicated with other gastrointestinal malignancies; (5) with anemia or hematological diseases before operation.

2.2. Clinical Characteristics and Histologic Assessment

The clinicopathological and baseline characteristics include age, gender, BMI, hemoglobin, AFP, CEA, CA199, CA125, tumor site, histologic grade, Borrmann classification, clinical T stage, clinical nodal status, and TNM stage. Histologic grade was defined as well (G1), moderately (G2), poorly differentiated (G3), or undifferentiated (G4).

Pathologic response to NAC was evaluated using post‐chemotherapy resection specimens. Patients were divided into the sensitive group and the nonsensitive group according to the pathological evaluation of the tumor regression grade (tumor regression grade, TRG), proposed by the 8th AJCC TNM classification or the NCCN guidelines. Sensitivity was defined as TRG 0‐2, and nonsensitivity was defined as TRG 3. The tumor regression scores are as follows: Grade 0, complete response (no viable cancer cells, including lymph nodes); Grade 1 (single cells or rare small groups of cancer cells); Grade 2 (residual cancer cells with evident tumor regression but more than single cells or rare small groups of cancer cells); Grade 3 (extensive residual cancer with no evident tumor regression).

2.3. IHC and Lillie's Ferrous Iron Stain Kit

IHC was performed on formalin‐fixed paraffin‐embedded tissue sections, with the following antibodies: anti‐ACSL4 antibody (1:400; Cat#ab155282, Abcam, United Kingdom), anti‐GPX4 antibody (1:100; Cat#ab125066, Abcam, United Kingdom), anti‐GCH1 antibody (1:1500; Cat#ab236387, Abcam, United Kingdom), anti‐AIFM/FSP1 antibody (1:1000; Cat No.20886‐1‐AP, Proteintech, United States), anti‐DHODH antibody (1:100; A6899, Abclonal, China), and anti‐SLC7A11/xCT antibody (1:600; Cat No.26864‐1‐AP, Proteintech, United States). Slides were incubated with primary antibodies overnight at 4°C, followed by wash and incubation with goat anti‐rabbit secondary antibodies (DAKO??) at room temperature for 30 min. Negative controls were treated identically without the primary antibodies. Biopsy tissue sections were photographed using a digital slice scanner (KF‐PRO‐020, KFBIO, China) and analyzed with three representative regions of tumor cells randomly captured at ×200 magnification using the K‐viewer software (version: 1.6.0.28). The IHC staining intensity was analyzed by investigators and Image‐Pro Plus v6.0 (Media Cybernetics Inc., USA), respectively. Protein expression was presented by mean optical density, which was calculated as the integrated optical density (IOD) of positively stained area divided by tumor area in each image. The staining intensity was graded on a scale of 0–3 (0, none; 1, weak; 2, moderate; 3, strong). The percentage of positive tumor cells was scored as follows: 0, 0%; 1, 1%–25%; 2, 26%–50%; 3, 51%–75%; 4, 76%–100%. A final semiquantitative score (0–12) was derived by the intensity score multiplied by the percentage score.

We investigated the ferrous iron accumulation by the Lillie's Ferrous Iron Stain Kit (G3320, Solarbio Life Sciences, Beijing, China). The positive control was lung tissue. The slices were washed with distilled water and dipped in a Lillie staining solution for 25–30 min. Subsequently, the slices were immersed in nucleus staining solution for 5–10 min and washed with distilled water for 5 s. Finally, the dehydrated slices were used for measuring iron content after being washed with distilled water again. For each patient, three fields were selected to obtain the average number of Lillie‐positive cells per square millimeter.

2.4. Training and Validation Set

Patients with tumors showing Grades 0 to 2 were divided into the sensitive group, and patients with tumors showing Grade 3 were divided into the nonsensitive group.

In the training set, we retrospectively included a total of 118 LAGC patients receiving platinum and fluorouracil‐based NAC in the First Affiliated Hospital, Sun Yat‐sen University, from January 2010 to December 2019. We included 66 patients from the First Affiliated Hospital, Sun Yat‐sen University (from January 2020 to July 2022); the Seventh Affiliated Hospital, Sun Yat‐sen University (from January 2020 to January 2022); the First Affiliated Hospital of Guangzhou Medical University (from January 2019 to July 2022); and the First Affiliated Hospital of Shantou University Medical College (from January 2017 to April 2022) as a multicenter external validation set.

2.5. Statistical Analyses

Statistical analysis was performed using the GraphPad Prism 9 software (Graph Pad Software, San Diego, CA) and SPSS 25.0 software (SPSS Inc., Chicago, IL, USA). The clinicopathological and baseline characteristics were analyzed as categorical variables by the chi‐squared test or Fisher's exact test and as continuous variables by the Spearman's rank correlation test. Univariate and multivariate logistic regression analyses were performed to derive odds ratios (OR) and 95% confidence intervals (95% CI) when analyzing the predictive value of ferroptosis‐related biomarkers for chemosensitivity in the whole group and subgroups and the correlation between these biomarkers and clinicopathological variables.

3. Results

3.1. Baseline Characteristics and Treatment Response

A total of 118 patients with LAGC were identified (Figure S1). Of those, the response was evaluated according to the TRG pathological response system: 48 out of 118 patients (41%) had a good tumor regression (TRG 0–2), while 70 out of 118 (59%) had a poor tumor regression (TRG 3) in the training cohort. The mean age at the time of diagnosis was 55 years, and 67% of patients were male. The average BMI was within the normal reference range (18.5–23.9).

Classification variables were used to describe the tumor stages and pathological features. In the training set, the tumor sites were mostly gastric cardia (38.1%) and antrum (30.5%), the main differentiation type was poorly differentiated adenocarcinoma (66.2%), and Borrmann type III was the most common type (66.9%). The clinical TNM stage before treatment was mainly stage III (77.1%), and there were no statistical differences in the clinical information and pathological features between the training set and testing set before treatment (Table S1).

3.2. Association of Tumor Response Grade and Sensitivity of Ferroptosis

The expression of MDA and 4HNE was detected in postoperative patients' tumor tissues to quantify the expression level of ferroptosis. The results indicated that MDA and 4HNE expression in the sensitive group was higher than that in the nonsensitive group, indicating that the sensitivity to ferroptosis was correlated with the efficacy of NAC. Tumors with higher sensitivity to ferroptosis had a more obvious pathological regression response to NAC (p < 0.05; Figure 1).

FIGURE 1.

Correlation analysis between the ferroptosis sensitivity and TRG. (A) Expression of MDA and 4HNE in pre‐NAC gastric cancer specimens. (B) MDA density was higher in the sensitive group than in the nonsensitive group. (C) 4HNE density was higher in in the sensitive group than in the nonsensitive group. Scale bar = 100 μm, 25 μm. MI = IOD/area.

The study identified patients with tumors showing Grades 0 to 2 as the sensitive group, and patients with tumors showing Grade 3 as the nonsensitive group. We analyzed the differences in the expression of molecular markers related to the ferroptosis positive pathway between the efficacy‐sensitive group and the insensitive group, including the expression level of ACSL4 and ferrous irons (Fe²⁺). The results indicated that the density of average iron‐stained positive cells in the efficacy‐sensitive group was significantly higher than that in the insensitive group (p < 0.001). The expression of ACSL4 in tumor tissues was artificially scored under a microscope, and the expression levels of ferroptosis positive regulators were higher in the sensitive group (p < 0.001; Figures 1 and 2).

FIGURE 2.

Expression of Fe²⁺ in endoscopic biopsy specimens before NAC. (A) The green arrows show the ferrous deposition stained in specimens. (B) In the training cohort, CD is higher in the sensitive group than nonsensitive group. Scale bar = 100 μm, 50 μm. CD: cell density.

The lethal accumulation of lipid peroxides involves an antagonism between ferroptosis execution and ferroptosis defense systems in cells. The GPX4‐xCT/SLC7A11 system is considered to be the major ferroptosis defense system in cells. The expression levels of GPX4 and xCT in gastric tumor tissues before treatment were determined by IHC and scored. The results showed that the expression level of GPX4 and xCT in the NAC‐sensitive group was slightly lower than that in the nonsensitive group, but the difference between the two groups was not statistically significant.

The DHODH‐CoQH2 system, the GCH1‐BH4 system, and the FSP1‐CoQH2 system are GPX4‐independent ferroptosis defense systems. They are equally likely to play a role in ferroptosis regulation. Therefore, we examined the expression levels of DHODH, GCH1, and FSP1 in endoscopic specimens by IHC. The staining scores of DHODH, GCH1, and FSP1 in the nonsensitive group were significantly higher than those in the sensitive group (p < 0.01) (Figure 3).

FIGURE 3.

Expression of ferroptosis‐related factors in endoscopic biopsy specimens before NAC. (A) Representative immunohistochemistry pictures of high or low expression of ferroptosis‐related factors. (B) In the training cohort, the expression of ACSL4 was higher in sensitive group than nonsensitive group (p < 0.001). GPX4 and SLC7A11 had not significant difference. The expression of FSP1, DHODH, and GCH1 were similar with ACSL4, higher in sensitive group (p < 0.01).

3.3. Establish the Predictive Nomogram Model

The cutoff values of the above molecular markers were determined by performing ROC curve analysis, and the results suggested that the best cutoff value of the density of Fe²⁺ positive staining cells was 11.58/mm². According to the above results, the continuous variables were transformed into binary variables, and univariate and multivariate logistic regression analyses were performed.

After screening by univariate logistic analysis, ACSL4, GCH1, FSP1, DHODH, and Fe²⁺ were included in the multivariate logistic regression equation. The results suggested that the expression levels of ACSL4, GCH1, DHODH, and Fe²⁺ in tumor tissues before treatment could be used as predictive factors for the efficacy of NAC (p < 0.05; Table 1). Then, we established the nomogram model, which can predict the risk of nonsensitivity to NAC in gastric cancer patients, and carried out internal verification by using the bootstrap method in the training set. The area under the ROC curve (AUC) of the nomogram model was 0.89, suggesting that the model has good discrimination. Fitting in good condition after the model calibration prompts the model predicted results to be consistent with the actual situation (Hosmer—Lemeshow test p = 0.684; Figure 4).

TABLE 1.

Univariate and multivariate logistic regression in the training cohort.

Variable	Univariate logistic				Multivariate logistic
	OR	95%CI		p	OR	95%CI		p
GPX4	1.132	0.944	1.358	0.181
ACSL4	0.768	0.669	0.882	< 0.001	0.612	0.483	0.776	< 0.001
GCH1	1.217	1.072	1.382	0.002	1.241	1.006	1.531	0.044
FSP1	1.301	1.094	1.547	0.003	1.150	0.891	1.483	0.283
DHODH	1.211	1.061	1.382	0.005	1.366	1.092	1.709	0.006
SLC7A11	1.097	0.960	1.254	0.174
Fe2+†	0.853	0.790	0.922	< 0.001	0.832	0.760	0.911	< 0.001

^†The cutoff value of the density of Fe²⁺ positive staining cells was 11.58/mm².

FIGURE 4.

Establish the predictive nomogram model. (A) Nomogram model. (B) shows ROC curve, and (C) shows calibration curve of the prediction model for the risk of nonsensitive to NAC in the training cohort (AUC = 0.89, and corrected c‐index is 0.88). (D) Predictive value of the model in testing cohort. AUC = 0.82.

3.4. Predictive Value of the Model in Testing Set

We include 66 patients with LAGC in the testing set (Figure S2). The prediction model was validated in a multicenter external testing set. In the validation set, the characteristics of patients, including age, sex, and BMI, were similar to the distribution of patients in the training set (Table 2). Almost 42% of patients had a good tumor regression (TRG 0–2), while 58% of patients had a poor tumor regression (TRG 3) in the testing cohort. The external validation of the prediction model in the multicenter validation set showed that the AUC was 0.82, indicating that the model had good prediction performance (Figure 4).

TABLE 2.

Baseline characteristics of patients in the training and validation cohorts.

Characteristics	Training cohort (n = 118)				Validation cohort (n = 66)
	Total (n = 118)	Sensitive group (n = 48)	Nonsensitive group (n = 70)	p	Total (n = 66)	Sensitive group (n = 28)	Nonsensitive group (n = 38)	p
Age				0.466				0.594
Mean ± SD	55.1 ± 12.7	53.9 ± 13.5	56.0 ± 12.2		57.9 ± 12.5	58.5 ± 12.4	57.5 ± 12.6
Gender (n [%])				0.652				0.484
Male	79 (66.99)	31 (64.6)	48 (68.6)		44 (66.7)	20 (71.4)	24 (63.2)
Female	39 (33.1)	17 (35.4)	22 (31.4)		22 (33.3)	8 (28.6)	14 (36.8)
BMI (kg/m 2 )				0.901				0.814
Mean ± SD	21.1 ± 2.9	21.1 ± 2.5	21.1 ± 3.1		21.2 ± 2.8	21.9 ± 2.4	21.2 ± 3.2
Hb (g/L)				0.569				0.582
Mean ± SD	119.5 ± 18.2	120.6 ± 18.4	118.7 ± 18.2		117.1 ± 22.0	118.4 ± 19.2	116.1 ± 24.0
AFP				0.353				0.721
Increased (%)	9 (7.6)	5 (10.4)	4 (5.7)		6 (9.1)	3 (10.7)	3 (7.9)
CEA				0.077				0.709
Increased (%)	35 (29.7)	10 (20.8)	25 (35.7)		13 (19.7)	5(17.9)	8 (21.1)
CA125				0.872				0.937
Increased (%)	14 (11.9)	6 (12.5)	8 (11.4)		5 (7.6)	1 (7.1)	3 (7.9)
CA19‐9				0.576				0.041
Increased (%)	12 (10.2)	4 (8.3)	8 (11.4)		15 (22.7)	3 (10.7)	12 (31.6)
cTNM stage ‡				0.429				0.210
I	1 (0.8)	0	1 (1.4)		0	0	0
II	10 (8.5)	4 (8.3)	6 (8.6)		7 (10.6)	5 (17.9)	2 (5.3)
III	91 (77.1)	36 (75.0)	55 (78.6)		48 (72.7)	19 (67.9)	29 (76.3)
IV	16 (13.6)	8 (16.7)	8 (11.4)		11 (16.7)	4 (14.3)	7 (18.4)
cT stage				0.697				0.498
T1–2	5 (4.2)	2 (4.2)	3 (4.3)		4 (6.1)	2 (7.1)	2 (5.3)
T3	27 (22.9)	12 (25.0)	15 (21.4)		16 (24.2)	5 (17.9)	11 (28.9)
T4	86 (72.9)	34 (70.8)	52 (74.3)		46 (69.7)	21 (75.0)	25 (65.8)
cN stage				0.919				0.196
N0‐1	48 (40.7)	19 (39.6)	29 (41.4)		33 (50.0)	16 (57.1)	17 (44.7)
N2‐3	70 (59.3)	29 (60.4)	41 (58.6)		33 (50.0)	12 (42.9)	21 (55.3)
cM stage				0.268				0.392
M0	101 (85.6)	39 (81.3)	62 (88.6)		56 (84.8)	25 (89.3)	31 (81.6)
M1	17 (14.4)	9 (18.8)	8 (11.4)		10 (15.2)	3 (10.7)	7 (18.4)
ypTNM stage ‡				< 0.001				< 0.001
pCR	5 (4.2)	5 (10.4)	0		6 (9.1)	6 (21.4)	0
I	13 (11.0)	9 (18.8)	4 (5.7)		9 (13.6)	7 (25.0)	2 (5.3)
II	29 (24.6)	17 (35.4)	12 (17.1)		17 (25.8)	6 (21.4)	11 (28.9)
III	54 (45.8)	13 (27.1)	41 (58.6)		28 (42.4)	8 (28.6)	20 (52.6)
IV	17 (14.4)	4 (8.3)	13 (18.6)		6 (9.1)	1 (3.6)	5 (13.2)
ypT				< 0.001				0.002
T0	5 (4.2)	5 (10.4)	0		6 (9.1)	6 (21.4)	0
T1‐2	24 (20.4)	17 (35.4)	7 (10.0)		19 (28.8)	10 (35.7)	9 (23.7)
T3	29 (24.6)	11 (22.9)	18 (25.7)		12 (18.2)	3 (10.7)	9 (23.7)
T4	60 (50.8)	15 (31.3)	45 (64.3)		29 (43.9)	9 (32.1)	20 (52.6)
ypN				0.004				< 0.001
N0‐1	55 (46.6)	30 (62.5)	25 (35.7)		40 (60.6)	24 (85.7)	16 (42.1)
N2‐3	63 (53.4)	18 (37.5)	45 (64.3)		26 (39.4)	4 (14.3)	22 (57.9)
ypM				0.058				0.114
M0	99 (83.9)	44 (91.7)	55 (78.6)		59 (89.4)	27 (96.4)	32 (84.2)
M1	19 (16.1)	4 (8.3)	15 (21.4)		7 (10.6)	1 (3.6)	6 (15.8)
Local				0.984				0.403
U	45 (38.1)	20 (41.7)	25 (35.7)		29 (43.9)	14 (50.0)	15 (39.5)
M	34 (28.8)	10 (20.8)	24 (34.3)		9 (13.6)	3 (10.7)	6 (15.8)
L	36 (30.5)	17 (35.4)	19 (27.1)		26 (39.4)	11 (39.3)	15 (39.5)
UML	3 (2.5)	1 (2.1)	2 (2.9)		2 (3.0)	0	2 (5.3)
Grade				0.489				0.057
Gx	3 (2.5)	3 (6.2)	0		7 (10.6)	6 (21.4)	1 (2.6)
G1	1 (0.8)	1 (2.1)	0		1 (1.5)	0	1 (2.6)
G2	36 (30.5)	15 (31.3)	21 (30.0)		23 (34.9)	8 (28.6)	15 (39.5)
G3	78 (66.2)	29 (60.4)	49 (70.0)		35 (53.0)	14 (50.0)	21 (55.3)
Borrmann				0.320				0.211
I–II	18 (15.3)	10 (20.8)	8 (11.6)		17 (25.8)	10 (35.7)	7 (18.4)
III	79 (66.9)	30 (62.5)	49 (70.0)		32 (48.5)	12 (42.9)	20 (52.6)
IV	19 (16.1)	7 (14.6)	12 (17.1)		9 (13.6)	3 (10.7)	6 (15.8)
Unknown	2 (1.7)	1 (2.1)	1 (1.4)		8 (12.1)	3 (10.7)	5 (13.2)

Abbreviations: CA125, cancer antigen 125; CA19‐9, carbohydrate antigen 199; CEA, carcinoembryonic antigen; pCR, pathological complete response.

^‡ According to the 8th edition of the American Joint Committee on Cancer (AJCC) guidelines.

4. Discussion

As far as we know, this study for the first time reported the predictive value of ferroptosis‐related factors for tumor regression in LAGC treated with NAC. Our study first revealed that the efficacy of NAC (pathological regression grade) is correlated with the sensitivity of ferroptosis in advanced gastric cancer. Patients with high ferroptosis sensitivity have a higher degree of tumor regression after NAC.

This study demonstrated that the efficacy of NAC is related to the occurrence of ferroptosis in gastric cancer. One of the important characteristics of ferroptosis is lipid peroxidation. We assayed intracellular malondialdehyde (MDA) and 4‐hydroxynonenal (4HNE), which are often tested as cell lipid peroxidation level indicators [17]. We found that MDA and 4HNE expression levels were significantly lower in the nonsensitive group than in the sensitive group. Moreover, the levels of MDA and 4HNE are associated with the grade of tumor regression after NAC, implying that ferroptosis is strongly correlated with the NAC response in gastric cancer.

Therefore, we speculated that patients with higher positive ferroptosis‐related indicator expression might respond better to NAC due to their hypersensitivity to ferroptosis. Given the antagonism between ferroptosis execution and ferroptosis defense systems in cells, this study also examined the negative ferroptosis‐related signature expression levels in the tumor. Firstly, we found that the expression of Fe²⁺ was higher in the sensitive group, and there was a significant difference between the different response groups (p < 0.001). Both univariate and multivariate logistic regression analyses suggested that Fe²⁺ could be used as an independent predictor of NAC response. ACSL4 also have similar results, but the determination of GPX4 with previous research is different. Several studies and reviews expounded that GPX4 is the essential regulator of ferroptotic cancer cell death. Ouyang's study [16] had shown that ferroptosis negative regulation (FNR) features are closely related to the progression and chemoresistance of gastric cancer. FNR‐related genes (GPX4, SLC7A11, and FTH1) are upregulated in 5‐FU‐resistant cells, and GPX4 expression is higher in tumors than in corresponding adjacent normal gastric tissues. It has also been proposed that high levels of GPX4 or FTH1 in gastric cancer are significantly associated with poor survival. Our study showed that the expression level of GPX4 in the nonsensitive group was higher than that in the sensitive group, but the difference between the two groups was not significant (p = 0.057). We further measured the expression of SLC7A11/xCT, and the difference between the two groups was not different, either. Considering the different processes, method designs, and sample selections of the two studies, the possible reason for this inconsistency is that the content of gastric cancer in this study was determined by endoscopic biopsy tissue without any treatment, which has not been induced by chemotherapy drugs and may have different expression at the biochemical level. We still need to further explore in follow‐up studies.

Additionally, we discovered that the GPX4‐independent ferroptosis defense systems have a more significantly different expression in gastric cancer tissues without treatment. Lei et al. [11] concluded that the ferroptosis defense system can be divided into the GPX4‐dependent and GPX4‐independent ferroptosis defense systems, and cancer cells with low expression of components of GPX4‐independent systems (such as ferroptosis suppressor protein 1 (FSP1), dihydroorotate dehydrogenase (DHODH), or GTP cyclohydrolase 1 (GCH1)) depend on GPX4 for survival and therefore are vulnerable to GPX4 inhibition. Conversely, cancer cells with low expression of GPX4 are sensitive to the inactivation of components of GPX4‐independent systems. For further confirmation, we identified the levels of DHODH, GCH1, and FSP1 in biopsy tissues. We found that the staining scores of DHODH, GCH1, and FSP1 in the nonsensitive group were significantly higher than those in the sensitive group. The predictive values of GCH1 and DHODH were determined by univariate and multivariate logistic regression analysis and included in the prediction model for validation. These results indicate that they may affect the development of drug resistance and can be further explored as therapeutic targets for reversing drug resistance.

Our study showed that ACSL4, Fe²⁺, DHODH, and GCH1 could be independently prognostic for gastric cancer treated with NAC, thereby establishing a nomogram prediction model to predict the efficacy of NAC in gastric cancer. After correction in the training set and internal validation, the nomogram prediction model had good discrimination, and the positive predictive value and negative predictive value of predicting the risk of insensitivity were 0.91 and 0.77, respectively. In order to screen out gastric cancer patients who are not sensitive to NAC, early surgical treatment or other treatment methods can be performed to reduce disease progression or even loss of operation time due to NAC treatment, which seriously affects the prognosis.

For possible clinical usage of this model, patients first undergo endoscopic biopsy for pathological examination. Once the pathological examination confirms gastric adenocarcinoma, the expression levels of relevant indicators can be detected through IHC, and then, the score can be determined based on this model. Based on this prediction model, we think the higher the score, the higher the risk of insensitivity to NAC. For patients who are not sensitive and can be treated with R0 resection, surgical treatment and postoperative chemotherapy are recommended. For sensitive patients, neoadjuvant therapy is recommended before surgery.

There are still some limitations in this study. First, the sample size was relatively small; we need a larger sample size for validation. Second, the patients in this study mainly received SOX regimens, so the predictive ability of response to other regimens' treatment is unclear.

5. Conclusions

In conclusion, our study for the first time reported that the expression of ferroptosis‐related factors could serve as novel predictive biomarkers for NAC for patients with gastric cancer. It might help screen candidate responders and determine chemotherapy strategies. Basic research is required to elucidate the mechanism of ferroptosis affecting chemosensitivity.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Comparison of clinicopathological characteristics between the training and validation cohorts.

Figure S1: Inclusion and exclusion flow chart.

Figure S2:. Flowchart of validation group.

Shen, M. , Cai, Q. , Wang, R. , Zhang, F. , Chen, H. , Wang, X. , Xie, R. , Zhang, J. , Liu, B. , He, Y. , Hou, X. , and Yang, D. (2025) Development and Validation of a New Prediction Criteria for Neoadjuvant Chemotherapy Response in Locally Advanced Gastric Cancer Based on Ferroptosis‐Related Biomarkers. Journal of Gastroenterology and Hepatology, 40: 2913–2924. 10.1111/jgh.70124.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.