We utilized Tumor Immune Estimation Resource version 2 (TIMER2) to analyze MMP1 expression levels between tumors and paired normal tissues. In tumor types with no or limited normal tissues, we investigated MMP1 expression by using ‘Gene Expression Profile’ section of Gene Expression Profiling Interactive Analysis (version 2) (GEPIA2) to combine data from normal tissues in the Genotype-Tissue Expression (GTEx) database, under parameters of log₂FC (fold change) cutoff = 1 and P-value cutoff = 0.01. We also used ‘Pathological Stage Plot’ section to obtain MMP1 expression in TCGA cancers at each pathological stage.

UALCAN tool was utilized to perform protein expression analysis. Protein expression data were from Clinical Proteomic Tumor Analysis Consortium (CPTAC) dataset. Besides, we downloaded and analyzed immunohistochemical images of MMP1 protein expression in human pancreatic normal tissues and tumor tissues from the Human Protein Atlas (HPA).

We obtained overall survival (OS) and disease-free survival (DFS) significance maps and Kaplan–Meier (K-M) survival plots of MMP1 by utilized ‘Survival Analysis’ section of GEPIA2. We used cut-off value of 50% to split the high and low expression cohorts. Log-rank test was used to perform hypothesis testing.

We obtained RNA-seq data in transcripts per million reads (TPM) format from TCGA and GTEx uniformly processed by the Toil process (17) in UCSC XENA. Tumor data from TCGA and corresponding normal tissue data from GTEx were extracted. Expression comparisons between samples were performed after log2 transformation of RNA-seq data in TPM format. Finally, we utilized R package pROC (version 1.17.0.1) to analyze data and R package ggplot2 (version: 3.3.3) to visualize the results.

The cBioPortal tool was employed to analyze MMP1 genetic alterations. Alteration frequency and type and mutated site for MMP1 from TCGA Pan-Cancer Atlas Studies were obtained and analyzed. We input ‘MMP1’ to ‘Quick Search’ module, and genetic alterations and mutated site data can be obtained from ‘Cancer Types Summary’ and ‘Mutations’ modules.

According to whether MMP1 was mutated or not, we divided the limited tumor sample data available from the TCGA into MMP1 altered and unaltered groups and compared their prognoses, including OS, disease-specific survival (DSS) and progress-free survival (PFS).

MCPCOUNTER, TIDE and EPIC algorithms were employed to investigate the correlation between MMP1 expression and infiltration of CAFs in the ‘Immune Association’ section of TIMER2.

We obtained the tumor dataset from the UCSC database. Immediately, the MMP1 expression data in each sample were extracted. Then, we chose samples from which the samples were derived from Primary Blood Derived Cancer-Peripheral Blood, Primary Tumor Correlation test and Metastatic samples of TCGA-SKCM. Besides, tumor gene expression profiles were extracted. Subsequently, we mapped them to GeneSymbol. Finally, R package ESTIMATE (version 1.0.13) was used to calculate stromal, immune, and estimate scores, and the corr. test function of R package psych (version 2.1.6) was utilized to calculate Pearson’s correlation between MMP1 expression and immune infiltration scores.

We obtained tumor data from UCSC, and MMP1 expression data were extracted. Then we chose samples from which the samples were derived from Primary Blood Derived Cancer-Peripheral Blood and Primary Tumor. We also obtained the Simple Nucleotide Variation dataset of level4 for TCGA samples from GDC treated by MuTect2 software. TMB score were calculated by using maftools R package (version 2.8.05). Besides, we obtained MSI score from a previous study (18). Finally, we explored Spearman’s correlations between TMB and MSI and MMP1 expression.

GeneMANIA website was utilized to find functionally similar genes using a large body of genomics and proteomics data, and we used it to search for genes that are functionally similar to MMP1. Also, a human MMP1-binding protein co-expression network can be obtained by STRING tool. Next, we set the main parameters: max number of interactors to show (‘no more than 20 interactors’), minimum required interaction score [‘highest confidence (0.700)’], active interaction sources (select all), meaning of network edges(‘evidence’), and Network type (‘full STRING network’). All gene symbols were used as input gene symbols for Gene Ontology pathway enrichment analysis with R package clusterProfiler (version: 3.14.3). In the end, we utilized R package ggplot2 (version: 3.3.3) to visualize the results.

Human PAAD cell lines (AsPC-1, PANC-1 and SW1990) were purchased from Shanghai Institute of Biochemistry and Cell Biology. Cells were passaged and cryopreserved by the Central Laboratory of Hepatobiliary Pancreas, Nanjing Drum Tower Hospital. Human normal pancreatic cell line (HPDE) and PAAD cell lines (Capan-2 and BxPC-3) were derived from our laboratory. Cells were cultured in DMEM (Dulbecco’s modified Eagle’s medium) or RPMI 1640 containing 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (WISENT, China). All cells were cultured at 37°C with 5% CO2. Moreover, all siRNAs (si-NC, si-MMP1) that purchased from RiboBio (Guangzhou, China) were transfected into BxPC-3 and Capan-2 cells by using Lipofectamine 3000 (Invitrogen, USA). At 2 days after transfection, cells were collected for following experiments. The sequences of siRNAs were displayed in Supplementary Table 1 .

RIPA buffer (Beyotime, China) was used to lyse cells to extract total protein. The treated protein samples were then subjected to SDS-PAGE. We transferred the separated protein onto PVDF membranes after electrophoresis. Membranes were blocked with blocking solution (5% skim milk) and then primary antibodies (MMP1, Abcam; GAPDH, Abcam) were used to incubate membranes overnight. After incubation with secondary antibodies, proteins were visualized with high-sensitivity electrochemiluminescence agent (Beyotime, China) followed by autoradiography.

Culture-Inserts 2 Well for self-insertion (Ibidi, Germany) were placed into wells of 6-well plates. Cells were digested with trypsin, and resuspended with DMEM medium containing 1% FBS and 1% penicillin–streptomycin. 3×10⁴ cells were seeded in each well of the insert. After one night, the insert was removed. Cell migration was observed by photography in an Olympus CKX41 inverted microscope at 0 and 48 h. The scratch area at 0 and 48 h was measured using Image J software.

Transwell inserts (Corning, USA) were coated with Matrigel (BD biosciences, USA). 1 × 10⁵ cells were seeded in the upper chamber with serum-free DMEM, and serum containing DMEM was added to the lower chamber. Cells were incubated in 37°C incubator for 24h. Then, cells attached to the chamber were fixed in 4% paraformaldehyde. After 20 minutes, the chamber was moved into 0.1% crystal violet. Finally, cells in the upper chamber and Matrigel were gently removed with cotton swabs, and the total number of invaded cells was counted under an Olympus IX73 inverted microscope.

PAAD cells transfected with si-MMP1 and si-NC in logarithmic growth phase were seed in 96-well plates at 1000 cells/well density, and 5 multiple wells were designed for each group. After incubation for 24h 48h and 72h, a volume of 10 μL CCK8 solution (Vazyme, China) was added to each well and incubated in an incubator for 1 h. The wavelength at OD450 was measured in a multiplate reader.

According to manufacturer’s instructions, samples were stained with Annexin V-FITC and propidium iodide (Vazyme, China). Then, an Aria flow cytometer (BD Biosciences, USA) was utilized to display the apoptotic cell population. Finally, FlowJo software was used to analyze data.

Data were analyzed using GraphPad Prism, version 6.00 and SPSS, version 20.0. We employed One-way ANOVA to compare multiple groups, and Student’s t-test to compare two groups. Data were reported as means ± SD. P ≤ 0.05 was considered statistically significant.

TIMER2 was utilized to investigate MMP1 expression levels between tumor tissues and control tissues from the TCGA database. MMP1 expression levels were significantly higher in 19 tumor types than in corresponding controls. Notably, the mRNA levels of MMP1 were only lower in KICH than in control tissues. In addition, MMP1 expression levels were not different in PRAD compared to normal tissues ( Figure 1A ). As for tumor types without normal tissue data in TCGA, we analyzed MMP1 expression by combining TCGA and GTEx data. We found higher expression of MMP1 in tumor tissues from DLBC, SKCM and UCS ( Figure 1B ).

Next, we obtained MMP1 protein levels by utilizing the CPTAC dataset from the National Cancer Institute. The data revealed that MMP1 protein levels in COAD, HNSC, LUAD and PAAD were significantly higher than in control tissues ( Figure 1C ). Moreover, GEPIA2 were utilized to investigate the correlation between MMP1 expression and pathological stages. MMP1 expression was significantly related to the tumor stage in 9 cancer types ( Figure 1D ). Taken together, these data showed that MMP1 expression was elevated in most tumor types from TCGA.

To understand the correlation between MMP1 expression and prognosis, we investigated the correlation between MMP1 high and low expression groups and prognosis using the TCGA dataset. High MMP1 expression was related to poor OS in 9 tumor types. However, in OV, high MMP1 expression may be related to better OS ( Figures 2A, B ). In addition, high MMP1 expression was related to poor DFS in 6 tumor types ( Figures 2C, D ).

We also used ROC curves to evaluate the diagnostic accuracy of MMP1 in human tumors. The data suggested that MMP1 had a high accuracy (AUC > 0.9) in predicting the diagnosis in 15 cancer types ( Figure 3 ).

As we know, genetic alterations promote tumor development. To investigate the genetic alterations of MMP1, we utilized the cBioPortal website to analyze data from TCGA. We found that MMP1 gene alterations were most frequent (> 8%) in cervical squamous cell carcinoma among all tumors, and the predominant type of alterations was amplification ( Figure 4A ). As shown in Figure 4B , the most frequently mutated region (P412S/H/L) in MMP1 protein was mutated four times in total.

Moreover, we further investigated the relationship between MMP1 gene alterations and prognosis, and we presented meaningful results in Figure 4C . STAD patients with MMP1 gene alterations had poor DFS, DSS, OS, and PFS. In patients with BRCA, people with MMP1 gene alterations had a poor OS. However, with MMP1 gene alterations, OV patients had a better prognosis in DSS.

Malignant solid tumor tissue includes many types of cells. In addition to tumor cells, stromal cells and infiltrating immune cells are also considered to play an important role in tumor progression. It is unclear whether the MMP1 expression influences the recruitment of immune cells. We used TIDE, EPIC and MCPCOUNTER algorithms to analyze the correlation between immune cells infiltration and MMP1 expression in pan-cancer based on the TIMER2. We found that CAFs was positively related to MMP1 expression in 31 tumor types, and CAFs was significantly related to MMP1 expression in 18 tumor types ( Figure 5 ).

We also analyzed immune cells and stromal cells to determine the stromal, immune, and estimate scores. Eventually, we found a significant correlation between MMP1 expression and stromal score in 24 tumor types, and all of them were positively correlated ( Supplementary Figure 1 ). Moreover, MMP1 expression was positively and significantly related to immune score in 16 tumor types. However, THYM and TGCT were negatively related ( Supplementary Figure 2 ). In the relationship between MMP1 expression and estimate score, MMP1 expression was significantly associated with estimate score in 22 tumor types. A negative correlation was found in THYM, and other tumor types were positively related to MMP1 expression ( Supplementary Figure 3 ). Above all, MMP1 expression had a correlation with immune infiltration.

We calculated the spearman correlation between MMP1 expression and TMB and MSI, and observed a positive correlation between MMP1 expression and TMB in ACC (r=0.30, p=2.0e-2), BRCA (r=0.29, p=1.35e-20), KICH (r=0.43, p=3.9e-4), SARC (r=0.18, p=6.8e-3), STAD (r=0.14, p=4.3e-3), THYM (r=0.29, p=1.5e-3) and UCS (r=0.33, p=1.3e-2). In contrast, MMP1 expression was significantly and negatively related to TMB in BLCA (r=-0.13, p=8.4e-3), GBM (r=-0.18, p=3.0e-2), HNSC (r=-0.14, p=1.3e-3) and KIRP (r=-0.14, p=2.8e-2) ( Figure 6A ). In COAD (r=0.21, p=3.6e-7), TGCT (r=0.22, p=7.4e-3) and UCEC (r=0.18, p=1.3e-2), MMP1 expression had a positive correlation with MSI, whereas in GBM (r=-0.12, p=3.5e-2), HNSC (r=-0.11, p=1.7e-2), LUAD (r=-0.12, p=2.0e-4), LUSC (r=-0.11, p=1.8e-2) and PAAD (r=-0.16, p=3.9e-2), it was adversely connected with MSI ( Figure 6B ).

We extracted the top 20 genes most closely related to MMP1 via GeneMANIA ( Figure 7A ). Subsequently, GO enrichment analysis showed that these 20 genes were mainly correlated with extracellular structure organization/extracellular matrix organization/extracellular matrix disassembly/collagen metabolic process in the enriched biological processes (BP), and the enriched cellular components (CC) were mainly associated with collagen−containing extracellular matrix, platelet alpha granule platelet alpha granule lumen and tertiary granule lumen. Besides, molecular functions (MF) were related to endopeptidase activity, metallopeptidase activity, serine−type endopeptidase activity and metalloendopeptidase activity ( Figure 7B ).

Moreover, we utilized the STRING tool to obtain 20 MMP1-interacting proteins and create a PPI network, which was showed in Figure 7C . These proteins were also enriched in extracellular structure organization/extracellular matrix organization/extracellular matrix disassembly, and the major enriched KEGG pathways were related to IL−17 signaling pathway, EGFR tyrosine kinase inhibitor resistance, and inflammatory bowel disease ( Figure 7D ).

To validate the function of MMP1, we chose human pancreatic cancer for further investigation. Compared with HPDE, WB results suggested that MMP1 protein had a high expression in pancreatic cancer cells ( Figure 8A ). We also obtained the immunohistochemical pictures of MMP1 expression levels in pancreatic cancer from the HPA database ( Figure 8B ). Immunohistochemical results showed that MMP1 was overexpressed in pancreatic cancer. Then, we designed an siRNA against MMP1 and transfected it into Bxpc-3 and Capan-2 cells to examine the effect of MMP1 in PAAD cells. The results of WB demonstrated that siRNA had efficiency in inhibiting MMP1 expression ( Figure 8C ). Next, we found that silence of MMP1 could reduce the PAAD cells migration and invasion ability ( Figures 8D, E ). Subsequently, we examined the changes in cell viability after transfection in the normal cell group and the siMMP1-transfected group using the CCK8 kit. The results showed that the proliferation ability of PAAD cells was significantly reduced after 48 hours of MMP1 knockdown ( Figure 8F ). Meanwhile, we used flow cytometry to demonstrate that the knockdown of MMP1 also promoted PAAD cells apoptosis ( Figure 8G ). Therefore, these results indicated that MMP1 promoted the metastasis and inhibited the apoptosis of PAAD cells.