MMP3-based prognostic nomogram and risk stratification for vascular mimicry in oral squamous cell carcinoma progression

Abstract

Oral squamous cell carcinoma (OSCC) severely affects facial aesthetics and the quality of life of patients. Although clinicians continue to improve diagnostic and treatment methods, they still cannot effectively improve the survival rate. Previous studies have found that Matrix Metalloproteinases (MMP2 \MMP9) can promote the formation of vascular mimicry (VM) in tumor tissues, but the relationship between MMP3 and VM in tumors tissues has not been clarified. In this study, we investigated the relationship between MMP3 and VM in OSCC tissues. Lasso regression analysis identified VM formation as an independent risk factor for patient prognosis. Subsequent survival analysis revealed a significant difference in prognosis between the VM-positive group and the VM-negative group (P = 0.00054). The constructed nomogram demonstrated that higher MMP3 expression levels were associated with increased risk scores and decreased 1- and 3-year expected survival rates. ROC curve analysis indicated excellent predictive performance of the model, with 1-year AUC = 0.906 and 2-year AUC = 0.970. Decision curve analysis (DCA) further confirmed the clinical utility of the model. Collectively, these findings suggest that MMP3 is a negative prognostic biomarker in OSCC. Research into MMP3 as a therapeutic target could represent a new and valuable treatment option.

Background

OSCC is a group of cancers primarily originating from the oral mucosa and jawbones [1]. It exhibits a high morbidity and mortality in Southeast Asian countries, creating an urgent need to identify novel therapeutic targets [2–4]. Research has revealed that during progression to advanced stages, cancer stem cells mutate to form “Vascular mimicry” [5, 6]: under the induction of a series of genes related to carcinogenesis, hypoxia, angiogenesis, and autophagy, cancer stem cells differentiate into vascular-like cells, forming channel-like structures. These structures can provide nutrients and oxygen to promote the metastasis of cancer cells. However, at present, there are no markers that can effectively mark VM, Therefore, this experiment verified the relationship between MMP3 and VM.

Previous studies have demonstrated that MMPs are extensively distributed in various human tissues [7], and participate in multiple pathological processes such as tumor angiogenesis [8, 9]. Currently, the investigation of the mechanisms underlying individual MMPs is predominantly employed to identify therapeutic targets for MMP inhibitors [10]. Prior research has revealed that MMP2 and MMP9 exert a promotional effect on the formation of VM [11], however, the association between MMP3 and VM remains poorly characterized. In our previous research, we analyzed the differentially expressed genes in OSCC using the Cancer Genome Atlas (TCGA) database. Through KEGG pathway enrichment analysis of these differentially expressed genes, it was found that within the MMP family, MMP3 was significantly upregulated in OSCC, with its expression levels gradually increasing as OSCC progressed [12]. Therefore, in this experiment, we attempted to utilize MMP3 as a marker for VM formation, analyzing the impact of VM on the prognosis of OSCC.

In recent years, prognostic nomograms have been increasingly used as predictive methods for most types of cancer, mainly because they can generate individual survival rates for OSCC patients by integrating different clinical variables. Compared with the traditional AJCC tumor staging system, they have higher clinical value and better ability to identify high-risk patients in early stages [13–15]. To predict the prognosis of OSCC patients, this study will systematically analyze the relationship between MMP3, tumor VM formation as well as patient prognosis, establish a prognostic risk prediction model for OSCC, providing more references for the diagnosis and therapeutic strategies.

Methods

Research subjects

A total of 60 OSCC tissue samples and 15 normal gingival tissue specimens were collected from patients who visited the Affiliated Hospital of Southwest Medical University between January 1, 2022, and December 31, 2023. All tissue specimens were confirmed by at least two tissue pathologists. HE and immunohistochemical staining were performed to assess and quantify MMP3 expression and VM formation.

Experimental methods

Surgically excised tissues from patients with OSCC were collected, first subjected to HE staining, and diagnosed as OSCC by two pathology experts. Subsequently, double immunohistochemical staining with PAS+CD31 was performed on consecutive sections. Select five fields of view at 100 μm magnification and count the formation quantity of VM, classifying them according to a scoring system from 0 to 4. At the same time, MMP3 staining was carried out to quantify its expression level. The average optical density value (AOD) of MMP3 was calculated using Fiji (ImageJ). A higher AOD value indicates a higher expression level of MMP3. The specific steps are as follows: (1) Image preprocessing: Open the Fiji (ImageJ) software, import the MMP3 immunohistochemical image, and convert the image to an 8-bit grayscale image (Image - Type − 8-bit); (2) Color deconvolution: To accurately extract the positive staining signal, the blue staining of the cell nucleus needs to be subtracted. Use the “Color Deconvolution” function, select the DAB (brown) channel; (3) Calibration to Optical Density (OD): Convert the grayscale values to OD values. Find “Calibrate” in the “Analyze” and select “Uncalibrated OD” for calibration; (4) Set measurement parameters: Under the “Analyze” menu, select “Set Measurements”, check “Area”, “Integrated Density” and “Limit to threshold”; (5) Adjust Threshold and Perform Measurements: Use the Threshold tool to adjust the slider and outline all positively stained regions (typically displayed in red). Finally, click “Analyze” – “Measure”. The software will automatically calculate values such as Area and IntDen, and the AOD value is the result of dividing IntDen by Area.

Finally, clinical data of the patients (sex, age, smoking history, alcohol consumption history, degree of tissue differentiation, presence or absence of lymph node metastasis, performance or not of cervical lymph node dissection, administration or not of radiotherapy, as well as clinical stage) were collected, and both a prognostic nomogram and a risk prediction model were constructed. The clinical characteristics of the subjects are presented in the following table (Table 1).

Table 1.

Statistics of VM scores and MMP3 concentrations of 75 subjects

Characteristic	Group 1 (n = 30)	Group 2 (n = 30)	Group 3 (n = 30)	P-value
Age (years)	60.1 ± 10.9	60.7 ± 14.5	40.5 ± 11.2	< 0.001
Gender, n (%)				0.865
Male	18 (60.0%)	18 (60.0%)	16 (53.3%)
Female	12 (40.0%)	12 (40.0%)	14 (46.7%)
Stage, n (%)				< 0.001
Stage I	8 (26.7%)	15 (50.0%)	0 (0.0%)
Stage II	11 (36.7%)	6 (20.0%)	0 (0.0%)
Stage III	3 (10.0%)	3 (10.0%)	0 (0.0%)
Stage IV	8 (26.7%)	2 (6.7%)	0 (0.0%)
Missing/other	0 (0.0%)	4 (13.3%)	30 (100%)*
Grade, n (%)				< 0.001
G1	9 (30.0%)	13 (43.3%)	0 (0.0%)
G2	14 (46.7%)	11 (36.7%)	0 (0.0%)
G3	7 (23.3%)	2 (6.7%)	0 (0.0%)
Missing	0 (0.0%)	4 (13.3%)	30 (100%)*
Smoker, n (%)				< 0.001
No	16 (53.3%)	19 (63.3%)	23 (76.7%)
Yes	14 (46.7%)	11 (36.7%)	7 (23.3%)
Alcohol history, n (%)				0.082
No	13 (43.3%)	19 (63.3%)	22 (73.3%)
Yes	17 (56.7%)	11 (36.7%)	8 (26.7%)
Radiation therapy, n (%)				< 0.001
No	13 (43.3%)	23 (76.7%)	30 (100.0%)
Yes	17 (56.7%)	7 (23.3%)	0 (0.0%)
Lymph node dissection, n (%)				< 0.001
No	17 (56.7%)	25 (83.3%)	30 (100.0%)
Yes	13 (43.3%)	5 (16.7%)	0 (0.0%)
MMP3 expression	0.31 ± 0.05	0.14 ± 0.06	0.05 ± 0.02	< 0.001
Score	3.9 ± 0.9	0.0 ± 0.0	0.0 ± 0.0	< 0.001
Vital status (event), n (%)				< 0.001
Alive (0)	18 (60.0%)	30 (100.0%)	30 (100.0%)
Dead (1)	12 (40.0%)	0 (0.0%)	0 (0.0%)

Based on the results of the double-staining immunohistochemistry PAS(+) and CD31(-), the formation of VMs was scored as follows: Since this experiment was conducted based on the two-dimensional plane of tissue sections for statistics, 0 points are assigned to normal blood vessels, 1 point to blood vessels invaded by tumors with an invasion area less than 1/4, 2 points for invasion greater than 1/4 but less than 1/2, 3 points for invasion greater than 1/2, and 4 points to blood vessels completely formed by tumor cells (Fig. 1). According to the above scoring criteria, the microscopic image field was 100 μm, and five fields were randomly selected, and the total VM score of each field was calculated; and the average score of the five fields was calculated as the sample VM score. Finally, the positive area ratio of MMP3 in each field is measured using ImageJ, the average ratio of the five fields of view is obtained, and statistical analysis is performed.

Fig. 1.

A1–A3 represent normal blood vessels, scored as 0; B1–B3 show tumor cell invasion involving less than 1/4, scored as 1; C1–C3 show tumor cell invasion involving less than 1/2, scored as 2; D1–D3 show tumor cell invasion involving more than 1/2, scored as 3; E1–E3 represent cavities completely composed of tumor cells, scored as 4. (A1, B1, C1, D1, E1 are HE staining; A2, B2, C2, D2, E2 are PAS+CD31 staining; A3, B3, C3, D3, E3 are MMP3 staining)

Acceptance criteria

The inclusion and exclusion criteria for the subjects are as shown in Table 2.

Table 2.

Statistics of VM scores and MMP3 concentrations of 60 subjects

Criteria	Inclusion criteria	Exclusion criteria
Survival time	Complete survival data available	Incomplete survival data
Underlying diseases	No significant underlying systemic diseases	Presence of significant uncontrolled comorbidities
Surgical treatment	Underwent the specified surgical procedure.	Did not undergo the specified surgical procedure
Postoperative follow-up	Complete postoperative follow-up records available	Incomplete postoperative follow-up records
Informed patient consent	Provided written informed consent	Declined or unable to provide informed consent
Pathological diagnosis	Pathologically confirmed OSCC	Pathological diagnosis other than OSCC

Statistical analysis

First, the RNA-seq data and clinical data of TCGA-HNSC were downloaded from the TCGA database. Data without corresponding clinical information were discarded, and only data belonging to the oral cancer site were retained. Then, the survival package of R version 4.2.1 (http://www.r-project.org/) was used to perform a proportional hazards hypothesis test and Cox regression analysis. Following this, the rms package was employed to construct a nomogram-based prediction model and visualize the results. Clinically, 60 patients with OSCC were collected, along with gingival tissues from 15 healthy individuals as a control group. Clinical factors (including sex, age, smoking, alcohol consumption, surgical history, lymphatic infiltration, tumor stage, pathological classification, lymph node dissection, history of radiotherapy) were statistically analyzed for each group. Overall survival (OS) and the expression level of MMP3 in tissues were also assessed. At the same time, 60 patients were evaluated using the VM score. Using R version 4.1.0 (http://www.r-project.org/), Lasso regression analysis was performed to screen out the independent risk factors for survival in OSCC patients, and a nomogram was developed to predict 1-year and 2-year survival based on the VM score and the level of MMP3 expression. Finally, to further validate the model’s effectiveness, 1-year and 2-year Kaplan–Meier (KM) survival curves, ROC calibration curves, and decision curve analysis (DCA) curves were constructed.

Result

Prognostic nomogram and overall survival of OSCC patients in TCGA

We established a risk prediction model for 528 OSCC patients selected from the TCGA database. The included clinical features include gender, age, smoking and alcohol history, tumor stage, pathological grade, and lymph node invasion, etc. The clinical characteristics of the included patients are as follows (Table 3). Based on these clinical data, a prognostic nomogram (Fig. 2) is established. On the left side of the figure are the names of clinical variables, and on the right side are the corresponding value ranges. The longer the horizontal line, the more significant the impact of this factor on clinical events. Factors such as increasing age, smoking and alcohol history, and lymph node invasion can reduce patients’ postoperative survival rate. Additionally, changes in MMP3 levels also have a significant impact on OS.

Table 3.

Clinical characteristics of 528 subjects in TCGA

Characteristic	Overall (N = 528)
Age (years)	60.6 ± 12.0
Gender, n (%)
Male	386 (73.1%)
Female	142 (26.9%)
Race, n (%)
White	452 (85.6%)
Black or African American	48 (9.1%)
Asian	13 (2.5%)
Other/not reported	15 (2.8%)
Clinical stage, n (%)
Stage I	27 (5.1%)
Stage II	74 (14.0%)
Stage III	81 (15.3%)
Stage IVA	258 (48.9%)
Stage IVB	28 (5.3%)
Stage IVC	5 (0.9%)
Not reported/missing	55 (10.4%)
Vital status, n (%)
Alive	299 (56.6%)
Dead	229 (43.4%)
Alcohol history, n (%)
Yes	344 (65.2%)
No	167 (31.6%)
Not reported	17 (3.2%)
Anatomic site, n (%)
Oral cavity	316 (59.8%)
Larynx	117 (22.2%)
Oropharynx	80 (15.2%)
Hypopharynx	10 (1.9%)
Other	5 (0.9%)

Fig. 2.

Prognostic nomogram for 528 OSCC patients in TCGA

The relationship between the level of MMP3 expression, VM score, and the occurrence and development of tumors

Initially, a statistical analysis was conducted on the 75 samples(Table 4). Based on the presence or absence of VM, the samples were divided into two groups: a VM-forming group and a non-VM-forming group. In the VM-forming group, VM was evaluated, and the MMP3 content was quantified in both groups. Clinical variables (weight loss, gender, age, smoking, alcohol consumption, presence or absence of lymphatic infiltration, lymphadenectomy, radiotherapy, etc.) were included in the LASSO regression analysis. Using the LASSO method for parameter selection, a regression analysis was performed with 10-fold cross-validation; and its trajectory is shown in Fig. 3. This allowed for the selection of variables, and at lambda.min = 0.01131492, a model with excellent characteristics and the minimum number of variables was obtained (Fig. 4). The selected variables included VM score, stage, alcohol history, grade, lymphnode neck dissection— these are independent prognostic factors in OSCC patients.

Table 4.

Statistical results of VM score and MMP3 concentration in 75 subjects (Thirty subjects with VM formation, 30 subjects without VM formation, and 15 healthy individuals)

Examinee	VM score	MMP3
1	2	0.236
2	5.2	0.3
3	5.4	0.393
4	5	0.232
5	4.4	0.33
6	4.2	0.267
7	3.6	0.283
8	4.6	0.304
9	3.2	0.271
10	2	0.342
11	4.2	0.274
12	3	0.376
13	3	0.268
14	3.6	0.353
15	3	0.253
16	3.9	0.203
17	3.4	0.266
18	5.2	0.359
19	4.6	0.35
20	4.2	0.343
21	4.4	0.308
22	4	0.334
23	3.8	0.331
24	4.8	0.292
25	3.5	0.291
26	2.6	0.296
27	5	0.355
28	4.1	0.373
29	3.1	0.342
30	2.7	0.255
1	0	0.047
2	0	0.279
3	0	0.091
4	0	0.093
5	0	0.122
6	0	0.128
7	0	0.096
8	0	0.107
9	0	0.127
10	0	0.122
11	0	0.193
12	0	0.083
13	0	0.158
14	0	0.13
15	0	0.073
16	0	0.149
17	0	0.069
18	0	0.099
19	0	0.083
20	0	0.115
21	0	0.178
22	0	0.153
23	0	0.213
24	0	0.21
25	0	0.219
26	0	0.255
27	0	0.21
28	0	0.173
29	0	0.22
30	0	0.184
1	0	0.041
2	0	0.097
3	0	0.038
4	0	0.031
5	0	0.02
6	0	0.069
7	0	0.022
8	0	0.068
9	0	0.074
10	0	0.046
11	0	0.025
12	0	0.061
13	0	0.044
14	0	0.031
15	0	0.033

Fig. 3.

Path diagram of variable selection based on Lasso regression

Fig. 4.

Illustrates the process of selecting the optimal value of the parameter λ in the Lasso regression model through the cross-validation method

The relationship between MMP3 and tumor occurrence and development

Based on the aforementioned known prognostic risk factors (gender, age, weight loss, smoking, alcohol consumption, tumor stage, pathology grade, VM status, cervical lymph node infiltration, neck lymph node dissection, and radiation therapy), and patients were grouped according to VM status to construct a Kaplan–Meier survival analysis graph based on MMP3 expression levels (Fig. 5). The figure shows that in patients with OSCC who exhibit VM formation, the survival rate decreases over time compared to the group without VM formation.

Fig. 5.

p < 0.05, which indicates statistical significance. When VM formation occurs, the survival prognosis is poorer

Construction of a survival prediction nomogram for 60 patients with OSCC

The VM score and MMP3 content were included in the prognosis nomogram (Fig. 6), along with clinically relevant factors affecting the survival of patients: the influencing factors are shown on the left, horizontal lines indicate the scores impacting survival prognosis, and the total score is the sum of all indicators. The results demonstrated that MMP3 content and VM score possess a certain prognostic ability in predicting the survival of OSCC patients. As the content of MMP3 in the tissue increased, the VM score rose, the OS of the patients decreased, and the prognosis was poor. MMP3 content and VM score may serve as independent markers for clinical diagnosis and prognosis of OSCC.

Fig. 6.

Prognostic nomogram for 60 OSCC patients in the validation group

Prognostic model based on MMP3 and VM scores: efficacy and clinical applicability

Based on the risk prediction models of MMP3 and VM scores above-mentioned, we employed the receiver operating characteristic (ROC) curve to assess their effectiveness and the area under the curve (AUC) to test the efficacy of the model. As shown in Fig. 7, the AUC value at 1 year was 0.906, and at 2 years — 0.970; both values exceeded 0.5, indicating a high diagnostic accuracy of this model. Furthermore, to further evaluate the practicality, optimize clinical decision-making to assist physicians in making more precise diagnostic and therapeutic choices in clinical practice, we established DCA curves for the 1st and 2nd years to verify the model’s clinical predictive capability. As shown in Fig. 8, the decision curves indicate that if patients’ threshold probability is 45% at 1 year and 70% at 2 years, applying this risk prediction model in making decisions about whether to pursue treatment or forgo treatment for patients with OSCC yields greater benefit.

Fig. 7.

1-year and 2-year receiver operating characteristic curves (ROC) of the subjects

Fig. 8.

Benefit decision curve analysis (DCA) based on the risk prediction model for one year and two years

Discussion

In 1962, MMPs were discovered in the tail of tadpoles [16]. Subsequent studies revealed that the MMP family comprises 26 proteins that play a crucial role in many biological processes and serve as valuable therapeutic targets [17]. Previous researches found that, in many tumors, elevated MMP expression leads to collagen degradation both inside and outside the cell, the permeability and integrity of the extracellular matrix are disrupted, tumor cells invade and metastasize, and drugs are prevented from penetrating to the tumor site, resulting in treatment difficulties [18–20]. Thus, as proteolytic enzymes, MMPs can degrade collagen and other proteins in the ECM, penetrate the basal membrane and surrounding fibrillar collagen matrix, destroying connective tissue. In this process, the proteolytic activity of MMP3 determines the extent of local tumor invasion and metastasis [18, 21–23]. To further investigate the mechanism of MMP3 in tumorigenesis and progression, in this study, we collected tumor tissues from 60 OSCC patients and performed MMP3 immunohistochemical staining to assess MMP3 expression levels. The results demonstrated elevated expression of MMP3 in OSCC. Consequently, the VM scoring system was further incorporated into our analysis to further investigate the relationship between MMP3 and the generation of VM within the tumor. Previous studies have shown a significant correlation between MMP3 expression and the pathological stage of patients with head and neck squamous cell carcinoma [24–27]. However, the underlying mechanism remains unclear. In this study, we conducted VM scoring on 60 OSCC patients. The results demonstrated that as the formation of VM increased and the expression level of MMP3 rose, the prognosis of OSCC patients deteriorated. MMP3 and VM can promote tumor progression and metastasis. Nevertheless, the specific interaction mechanism between VM and MMP3 has not yet been further elucidated.

In tumor tissues, the process of angiogenesis is structurally and temporally different from normal vascular formation. Therefore, inhibiting angiogenesis in tumor tissues represents a valuable direction for formulating treatment strategies [28]. VM is formed by tumor stem cells differentiating into vascular endothelial cells and create pseudo-vascular channels that transport nutrients to tumor tissue, and it is associated with poor prognosis in cancer patients. However, the precise mechanism of this phenomenon remains unclear [29–31]. Therefore, in this experiment, we performed PAS+CD31 double staining on OSCC patient tissues to detect VM formation. Through Lasso, Nomogram and KM survival curves, we found that the survival prognosis of the group with elevated MMP3 expression and VM formation were worse. Based on these observations, we speculated that there might be a correlation between MMP3 and VM formation. However, since this study only analyzed their synergistic role in OSCC progression, the precise relationship between MMP3 and VM remains undefined. Further experiments are warranted to explore their interaction in greater depth.

Due to the high recurrence rate and low survival following surgery for OSCC, staging systems such as American Joint Committee on Cancer (AJCC) are widely used for diagnosing and managing patients with OSCC. However, most of these assessments are based on pathological features and do not take into account complex clinical factors or the impact of key molecules on survival prognosis in OSCC. In this study, building on previous research regarding the pro-angiogenic effect of MMP3 in OSCC, we incorporated the VM assessment standard and developed a risk prediction model based on VM and MMP3. Compared with traditional staging system, the model demonstrated better predictive efficiency; however, due to the limitation of follow-up time, this study could only predict 1-year and 3-year survival rates. In subsequent experiments, the follow-up period could be extended to obtain more predictive results.

Conclusion

In summary, we developed and validated a risk prediction model based on the MMP3 and VM scoring system, which demonstrates high diagnostic and prognostic value for OSCC patients. The formation of MMP3 and VM proved to be closely interconnected and significantly correlated with clinical outcomes in OSCC. Our findings provide clinicians with new, effective approaches to the treatment and management of patients with OSCC. Furthermore, these data may serve as a foundation for new applications in other types of tumors.

Author contributions

XQL and MHN conceived the project. FQ, DQ, RD, XZ and YT collected the data and conducted the experiment. FQ and LH analyzed the data. FQ and LH wrote the manuscript. XQL and MHN edited and revised the manuscript. All authors reviewed and approved for publication.

Funding

This project was financially supported by the Project of Oral Health Promotion and Oral Medicine Development in Western Regions for Clinical Research Fund of the Chinese Stomatological Medical Association (CSA-W2025-07) and Southwest Medical University Technology Program (2024KQZX03).

Data availability

The datasets generated and analysed during the current study are available in the Zenodo repository, Zenodo. [https://doi.org/10.5281/zenodo.18051434](https:/doi.org/10.5281/zenodo.18051434) .

Declarations

Ethics approval and consent to participate

The protocol was approved by the Biomedical Research Ethics Committee of Southwest Medical University and by the local National Health Service research ethics committee in accordance with the ethical standards for medica research involving human subjects, as laid out in the1964 Declaration of Helsinki and its later amendments. Prior to participation, all subjects were fully informed of the study’s purpose, procedures, potential risks and benefits, and their right to withdraw at any time without penalty. Written informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.