Abstract

Background

   Although surgical methods are the most effective treatments for colon
   adenocarcinoma (COAD), the cure rates remain low, and recurrence rates
   remain high. Furthermore, platelet adhesion-related genes are gaining
   attention as potential regulators of tumorigenesis. Therefore,
   identifying the mechanisms responsible for the regulation of these
   genes in patients with COAD has become important. The present study
   aims to investigate the underlying mechanisms of platelet
   adhesion-related genes in COAD patients.

Methods

   The present study was an experimental study. Initially, the effects of
   platelet number and related genomic alteration on survival were
   explored using real-world data and the cBioPortal database,
   respectively. Then, the differentially expressed platelet
   adhesion-related genes of COAD were analyzed using the TCGA database,
   and patients were further classified by employing the non-negative
   matrix factorization (NMF) analysis method. Afterward, some of the
   clinical and expression characteristics were analyzed between clusters.
   Finally, least absolute shrinkage and selection operator regression
   analysis was used to establish the prognostic nomogram. All data
   analyses were performed using the R package.

Results

   High platelet counts are associated with worse survival in real-world
   patients, and alternations to platelet adhesion-related genes have
   resulted in poorer prognoses, based on online data. Based on platelet
   adhesion-related genes, patients with COAD were classified into two
   clusters by NMF-based clustering analysis. Cluster2 had a better
   overall survival, when compared to Cluster1. The gene copy number and
   enrichment analysis results revealed that two pathways were
   differentially enriched. In addition, the differentially expressed
   genes between these two clusters were enriched for POU6F1 in the
   transcription factor signaling pathway, and for MATN3 in the ceRNA
   network. Finally, a prognostic nomogram, which included the ALOX12 and
   ACTG1 genes, was established based on the platelet adhesion-related
   genes, with a concordance (C) index of 0.879 (0.848–0.910).

Conclusion

   The mRNA expression-based NMF was used to reveal the potential role of
   platelet adhesion-related genes in COAD. The series of experiments
   revealed the feasibility of targeting platelet adhesion-associated gene
   therapy.

   Keywords: Platelet adhesiveness, gene, non-negative matrix
   factorization, transcription factor POU, colon adenocarcinoma

Background

   Platelets are involved in tumor angiogenesis and cancer progression,
   and high platelet counts in various cancers are associated with shorter
   disease survival.^[41]1–[42]3 Furthermore, platelets impact the disease
   burden and treatment outcomes of cancer patients and are involved in
   cancer progression, including cancer metastasis. Moreover, platelets
   protect cancer cells from chemotherapy-induced apoptosis and maintain
   the integrity of the vascular system that supplies the
   tumor.^[43]4,[44]5 The increase in platelet count has been identified
   as a predictor of cancer in patients with occult malignancies,^ [45]6
   and this has consistently been associated with the worsening of
   progression-free survival and/or overall survival (OS) of patients with
   ovarian cancer,^[46]7,[47]8 breast cancer,^ [48]9 colorectal cancer,^
   [49]10 lung cancer,^ [50]11 and gastric cancer.^[51]12,[52]13

   Neovascularization is essential for tumor growth, in order to ensure a
   sufficient supply of nutrients and oxygen. Cancer cells usually secrete
   certain growth factors, such as vascular endothelial growth factor, in
   order to induce vascularization, and platelets exhibit similar
   abilities.^ [53]14 In addition, platelets release some factors that
   help maintain the integrity of the microvasculature, which in turn
   reduces the ability of immune cells to infiltrate tumor tissues,
   ultimately promoting tumor growth.^ [54]15

   The ability to evade attacks by natural killer (NK) cells and fluid
   shear stress is crucial for cancer cells to successfully metastasize.
   Cancer cells can accomplish this by adhering to platelets through a
   process known as platelet cloaking.^[55]16,[56]17 Mice experiments have
   revealed that reducing the platelet number or function loss of
   platelets can significantly reduce the ability of cancer cells to
   metastasize.^ [57]18 Meanwhile, the lack of P-selectin (CD62P), which
   is a cell adhesion molecule in activated platelets, may inhibit
   intestinal tumor growth by preventing platelet adhesion to tumor
   cells.^[58]19 These data show that it is possible to develop novel
   cancer therapeutic strategies by studying platelet adhesion.

   The present study aims to identify the molecular clusters of platelet
   adhesion-associated genes in colon adenocarcinoma (COAD) and
   investigate the potential mechanisms of action by exploring the
   association between these identified platelet-associated genes, and
   tumor growth patterns, clinical efficacies, immune microenvironments,
   tumor cell stemness, transcription factor regulatory networks,
   competing endogenous RNA (ceRNA) regulatory networks, and drug
   sensitivities. The analysis was conducted through multiple dimensions,
   in order to gain a deeper understanding of the mechanisms of COAD
   development and identify novel therapeutic targets. Furthermore, a
   prognostic nomogram was developed to guide future clinical treatments.

Methods

   The flowchart for the present study is presented in [59]Figure 1.

Figure 1.

   [60]Figure 1.
   [61]Open in a new tab

   Flowchart for the selection process used in the study.

Participants

   In order to determine the relationship between platelets and OS, the
   clinical data of COAD patients, who were treated in the First
   Affiliated Hospital of Guangxi Medical University between January 2017
   and December 2022, were collected. In order to detect the mRNA levels,
   tissue samples were obtained from patients treated in the Department of
   Medical Oncology of the First Affiliated Hospital of Guangxi Medical
   University in April 2023. In order to detect the protein levels, tissue
   samples were obtained from patients, who were treated in the Department
   of Medical Oncology of the First Affiliated Hospital of Guangxi Medical
   University between April 2022 and April 2023. The present study was
   approved by the Ethics Committee of the First Affiliated Hospital of
   Guangxi Medical University (No. 2023-E186-02), and the privacy of the
   patients was well-protected. All patients agreed to participate in the
   study and provided a signed informed consent.

Online data analysis

   Online data were used to determine the relationship between the
   alterations in platelet adhesion-related genes and OS. Initially,
   platelet adhesion-related genes (Gene Ontology [GO]:0070527, downloaded
   on 18 July 2022) were obtained from the MsigDB database
   ([62]http://www.gsea-msigdb.org/gsea/msigdb/index.jsp). Then, the
   cBioPortal database ([63]http://www.cbioportal.org/) was used to
   explore the alterations in platelet adhesion-related genes and their
   effect on survival.

Downloading of The Cancer Genome Atlas (TCGA) data and bioinformatic analysis

   The COAD gene expression matrices for 41 normal and 480 tumor tissues,
   with the corresponding clinical characteristic data, were downloaded
   from the TCGA database ([64]https://cancergenome.nih.gov).

   In order to determine the role of platelet adhesion-related genes in
   COAD, a differential analysis was conducted to compare these genes
   between normal and tumor groups. The differentially expressed genes
   (DEGs) were defined as false discovery rate (FDR) < 0.05 and log[2]
   fold change (FC) > |1|. Then, the identified genes were subjected to
   the non-negative matrix factorization (NMF) clustering analysis, in
   order to determine the corresponding characteristics among the
   different clusters. Afterward, the OS was compared among clusters using
   survival curves.

   In order to determine the specific differences among clusters, the copy
   number variants (CNVs) were initially explored, since these have been
   identified as an important causal factor for various human diseases,
   and play an increasingly important role in human oncological
   diseases.^[65]20,[66]21 Then, the patient-related CNV data was
   downloaded from the TCGA website, and the R software was used to
   process and visualize the data. Afterward, Kyoto Encyclopedia of Genes
   and Genomes (KEGG) enrichment analysis was performed for the identified
   genes using the Database for Annotation, Visualization, and Integrated
   Discovery (DAVID; [67]https://david.ncifcrf.gov/) tool. Next, a
   single-sample gene set enrichment analysis was performed to determine
   the immune infiltration differences among clusters. Then, cancer
   stemness analysis was conducted to compare the cancer stemness among
   clusters.

   The inter-cluster drug sensitivity analysis was conducted using the R
   package “oncoPredict,” based on the Genomics of Drug Sensitivity in
   Cancer (GDSC) database ([68]https://www.cancerrxgene.org/).^ [69]22

Hub genes screening and bioinformatic analysis

   The hub platelet adhesion-related genes were identified using the gene
   expression matrix, and these hub genes were defined as FDR < 0.05 and
   log[2] FC > |1.0|.

   GO and KEGG functional enrichment analyses were performed for the hub
   genes using the DAVID tool. Then, the obtained hub genes were subjected
   to transcription factor enrichment analysis and visualization with R
   and Cytoscape 3.9.1.

   Finally, a predictable ceRNA network was established to identify the
   ceRNAs involved in the inter-cluster variation. The hub gene targets
   were initially analyzed using the ENCORI database, and the remaining
   genes were filtered according to the survival data (p < 0.05).
   Cytoscape 3.9.1 was used to visualize the data.

Real-time quantitative polymerase chain reaction

   Total RNA was extracted from tissues using the VeZol reagent (Vazyme),
   and the mRNA was reverse transcribed into cDNA using the RevertAid
   First Strand cDNA Synthesis Kit (Thermo). Next, real-time quantitative
   polymerase chain reaction (RT-qPCR) was performed using SYBR™ Green PCR
   reagents (Thermo). The primer sequences used for the present study are
   presented in [70]Supplemental Table S2. β-actin served as the
   endogenous reference gene for normalization. The relative expression
   was calculated using the 2^−ΔΔCt method, and each experiment was
   performed in triplicate.

Immunohistochemistry

   The paraffin-embedded fixed tissue was cut into sections. After
   deparaffinizing the slides for 20 min at 80°C, the slides were immersed
   in a xylene bath for 5 min, three times. Then, these were rehydrated in
   100% ethanol for 30 s, twice, fixed in 95% ethanol for 30 s, and fixed
   in 75% ethanol for 30 s. Next, antigen retrieval was performed using a
   repair solution for 3 min in a pressure cooker set at high pressure,
   and the slides were washed with phosphate-buffered saline (PBS). Then,
   these slides were immersed in 3% hydrogen peroxide for 10 min, and
   incubated overnight at 4°C with the appropriate primary antibody
   (POU6F1, PA5-61403, Invitrogen; MATN3, ab238893, Abcam; ACTG1,
   ab123034, Abcam; ALOX12, ab211506, Abcam) diluted in PBS. After
   overnight incubation at 4°C, the slides were rinsed with PBS, and
   incubated with the appropriate secondary antibody for 30 min at 37°C.

   In order to conduct a semi-quantitative analysis of
   immunohistochemistry, the following criteria were used. Based on
   staining intensity, the cells were scored as follows: 0 negatively
   stained, 1 weakly stained (pale yellow), 2 moderately stained (brownish
   yellow), and 3 strongly stained (sepia). Based on the percentage of
   positive cells, the positive ratio scores are as follows: 1, 25% or
   less; 2, 26%–50%; 3, 51%–75%; 4, 76% or more. We calculated the final
   score by multiplying staining intensity scores and positive ratio
   scores.

Nomogram construction and verification

   Least absolute shrinkage and selection operator (LASSO) Cox regression
   analysis was initially performed for the platelet adhesion genes, in
   order to identify the genes related to the prognosis. Then, the
   screened genes were subjected to univariate analysis, and all
   significant variables obtained from the univariate analysis were
   entered into the multivariate analysis ([71]Supplemental Figure S1).
   All included variables met the proportional risk assumption, based on
   the log survival plot. The prognostic factors were identified according
   to the forward stepwise Cox proportional risk regression model, and a
   nomogram was constructed. Then, the nomogram was validated through
   discrimination and calibration. The model discriminative power was
   assessed using Harrell's consistency (C) index, which handles censored
   data, and estimates the probability of events. A higher C index
   indicates better discriminative power. The differences between the
   predicted and actual survival rates were compared using calibration
   plots.

Statistical methods

   For the survival analysis of real-world data, all analyses were
   conducted using the limma, survival, or survminer packages in R
   (version 4.2.1). The patients were divided into high and low groups,
   according to the median platelet count, which was set as the cutoff
   value in the survival analysis. The statistical analysis was conducted
   using the R software (version 4.2.1), and data visualization, nomogram
   construction, and nomogram validation were performed. The p-values were
   determined using Student's t-test or analysis of variance. A p-value of
   <0.05 was considered statistically significant.

Results

Platelet count and its correlation with OS in COAD patients

   A total of 945 patients from real world met the inclusion criteria, and
   these were used for the analysis. Among them, 46 were at Stage I, 285
   at Stage II, 272 at Stage III, and 342 at Stage IV. There was a
   significant difference in platelet count across the tumor pathologic
   stages. The median value was 233.4  ×  10^9/L in Stage I,
   269.8  ×  10^9/L in Stage II, 307.8  ×  10^9/L in Stage III, and
   374  ×  10^9/L in Stage IV. The results presented in [72]Figure 2 show
   that the platelet count was proportional to the pathologic stage at
   presentation ([73]Figure 2(a)). Furthermore, the OS was inversely
   correlated with the platelet count ([74]Figure 2(b), p < 0.001).

Figure 2.

   [75]Figure 2.
   [76]Open in a new tab

   The association between platelet counts and the clinicopathological
   stage or survival. (a) The distribution of platelet counts in patients
   in different pathological stages. (b) The Kaplan–Meier survival curves
   between groups with high and low platelet counts (ns, p > 0.05,
   *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

Alterations in platelet adhesion-related genes inversely correlate with OS

   A total of 70 genes involved in platelet adhesion were identified from
   the MsigDB database. The expression of these 70 genes was compared
   between the COAD and control samples using the cBioPortal database. A
   total of 1471 samples from four studies were included for the analysis:
   619 studies from the Dana-Farber Cancer Institute (DFCI; Cell Reports,
   2016), 74 studies from Genentech (Nature 2012), 640 studies from The
   Cancer Genome Atlas (TCGA; Firehose Legacy), and 138 studies from
   Memorial Sloan Kettering (MSK; Genome Biol 2014). Then, the genetic
   alterations in platelet adhesion-related genes were explored. Mutations
   were identified in all four studies, while amplifications, profound
   deletions, and multiple alterations were only identified in the TCGA
   data. These mutations were the most prevalent genetic alterations in
   studies from DFCI, Genentech, and MSK, while amplification was the most
   prevalent in studies from TCGA ([77]Figure 3(a)). The survival analysis
   revealed that the alterations in platelet adhesion-related genes
   inversely correlate with OS ([78]Figure 3(b), p < 0.001).

Figure 3.

   [79]Figure 3.
   [80]Open in a new tab

   Genetic alteration mining and survival analysis. (a) The alterations of
   the 70 platelet adhesion-related genes were explored in four different
   colon adenocarcinoma (COAD) study cohorts. (b) The Kaplan–Meier
   survival curves between groups, with and without alterations
   (p < 0.001).

Differentially expressed platelet adhesion-related genes and characteristics
of patients with different DEGs

   Using the data downloaded from the TCGA database, 23 DEGs were
   identified: ADAMTS18, BLK, CEACAM1, COMP, CSRP1, CTSG, FGA, FGB, FGG,
   FLNA, HBB, ILK, ITGB3, MMRN1, MYL9, P2RY12, PDPN, PIK3CG, SERPINE2,
   SLC7A11, STXBP1, TYRO3, and WNT3A. These DEGs are presented in the
   heatmap in [81]Figure 4.

Figure 4.

   [82]Figure 4.
   [83]Open in a new tab

   Heatmap for the 70 platelet adhesion-related genes (the DEGs are marked
   in dark blue color).

   The further NMF clustering analysis of these 23 DEGs revealed two
   unique clusters of 433 COAD patients from TCGA, with 54 patients in
   Cluster1 and 379 patients in Cluster2 ([84]Figure 5(a)). The survival
   curves were drawn by matching the survival data for each patient. The
   results revealed that the OS of patients was better in Cluster2, when
   compared to Cluster1 ([85]Figure 5(b)).

Figure 5.

   [86]Figure 5.
   [87]Open in a new tab

   Identification and survival comparison of Cluster1 and Cluster2. (a)
   Two clusters were identified through non-negative matrix factorization
   (NMF). (b) The Kaplan–Meier curve for Cluster1 and Cluster2.

Difference in CNVs between clusters

   The analysis of CNVs in the two clusters revealed that 16 platelet
   adhesion-related genes had greater copy number gain than loss (GNAS,
   MYL9, PPIA, IL6, ACTG1, PRKCA, WNT3A, ACTB, C1QTNF1, COMP, EMILIN2,
   GATA1, ITGB3, SYK, TLN1, and VCL), and 30 genes had greater copy number
   loss than gain (LYN, PTPN6, CEACAM1, FGA, FGB, FGG, GP6, METAP1, MMRN1,
   SLC7A11, ACTN1, ALOX12, BLK, BLOC1S4, CELA2A, CSRP1, DMTN, HBB, IL6ST,
   ILK, MYH9, PDPN, PIK3CB, PRKCD, PRKCQ, PRKG1, SH2B3, STXBP3, TSPAN32,
   and TYRO3) in Cluster1 ([88]Figure 6(a)). In Cluster2, 29 genes had
   greater copy number gain than loss (ACTG1, GNAS, MYL9, C1QTNF1, CD9,
   PRKCA, PTPN6, ACTB, F11R, MPL, VCL, PEAR1, CLIC1, CSRP1, IL6, EMILIN2,
   PRKCQ, FLNA, MYL12A, PPIA, TSPAN32, MYH9, P2RY12, PIK3CB, RAP2B,
   PDGFRA, TLN1, ITGB3, and PLEK), and 36 genes had greater copy number
   loss than gain (LYN, GP6, PRRG1, CELA2A, IL6ST, PDPN, PIK3CG, FN1,
   SERPINE2, WNT3A, ABAT, BLOC1S4, JAK2, STXBP3, CEACAM1, HBB, STXBP1,
   ALOX12, BLK, HSPB1, ILK, MMRN1, SH2B3, SLC7A11, TYRO3, UBASH3B, COMP,
   DMTN, FGA, FGB, FGG, METAP1, PRKCD, ADAMTS18, FERMT3, and FIBP)
   ([89]Figure 6(b)). Next, a further KEGG enrichment analysis was
   conducted for these genes with changed copy numbers. The results
   revealed that genes with increased CNVs in Cluster1 were mainly
   enriched in the Rap1 signaling pathway, leukocyte transendothelial
   migration, and the PI3K-Akt signaling pathway ([90]Figure 6(c)).
   Furthermore, genes with decreased CNVs in Cluster1 were mainly enriched
   in the B cell receptor signaling pathway, Fc gamma R-mediated
   phagocytosis, and the T-cell receptor signaling pathway ([91]Figure
   6(d)). The increase in CNVs in Cluster2 mainly concentrated in
   leukocyte transendothelial migration, the Janus kinase-signal
   transducers and activators of transcription signaling pathway, and the
   PI3K-Akt signaling pathway ([92]Figure 6(e)). The decrease in CNVs in
   Cluster2 was primarily concentrated in the chemokine signaling pathway
   ([93]Figure 6(f)).

Figure 6.

   [94]Figure 6.
   [95]Open in a new tab

   Copy number variation (CNV) and enrichment analysis in clusters. (a)
   CNV plots for Cluster1. (b) CNV plots for Cluster2. (c and d)
   Enrichment analysis for the copy number gain of genes (c) and loss of
   genes (d) in Cluster1. (e and f) Enrichment analysis for the copy
   number gain of genes (e) and loss of genes (f) in Cluster2.

Differences in immune infiltration and cancer stemness between clusters

   Next, the immune infiltration and activity in tumors were quantified
   between the two clusters. The results revealed that five immune cells
   were significantly different between the two groups ([96]Figure 7(a)):
   incompletely mature NK cells (CD56^bright), fully mature NK cells
   (CD56^dim), gamma delta T cells, T helper (T[h])17 cells, and T[h]2
   cells. Furthermore, tumor stem cells were the most important cause of
   treatment failure in malignant tumors.

Figure 7.

   [97]Figure 7.
   [98]Open in a new tab

   Profile of signatures of the 433 patients with COAD from TCGA and
   validation. (a) Difference in immune cell infiltration between the two
   clusters. (b) Difference in cancer stemness index between the two
   clusters. (c) Volcano plots for the hub genes. (d) GO and KEGG
   enrichment analysis for the hub genes. (e) Network of transcription
   factors. (f) Network for ceRNAs. (g) RT-qPCR results for POU6F1 or
   MATN3 expression at the mRNA level. (h) immunohistochemistry results
   were evaluated semiquantitatively for POU6F1 or MATN3 expression at the
   protein level.

   RT-qPCR: real-time quantitative polymerase chain reaction; ceRNA:
   competing endogenous RNA; GO: gene ontology; KEGG: Kyoto Encyclopedia
   of Genes and Genomes.

   In order to understand the differences in cancer stemness between the
   two clusters, the cancer stemness index was explored. This index was
   used to assess the similarity of tumor cells to stem cells, since
   having an active biological process similar to stem cells would result
   in a higher degree of tumor dedifferentiation. The analysis results
   revealed that cancer stemness was higher in Cluster2, when compared to
   Cluster1. However, the survival rates were better in Cluster2, when
   compared to Cluster1 ([99]Figure 7(b)).

Differences in drug sensitivity between clusters

   In order to determine the sensitivity of each cluster to various
   oncology drugs, and provide a reference for clinical treatment, the
   drug sensitivity difference between the two clusters was further
   analyzed. A total of 13 drugs with significant differences were
   identified ([100]Figure 8 and [101]Table 1): alisertib, crizotinib,
   dabrafenib, Eg5_9814, elephantin, epirubicin, ERK_2440, ERK_6604,
   PAK_5339, PD0325901, SCH772984, trametinib, and ulixertinib. According
   to the analysis results, Cluster2 responds better to treatment with the
   above-mentioned drugs, when compared to Cluster1.

Figure 8.

   [102]Figure 8.
   [103]Open in a new tab

   Significant differences identified in the drug sensitivity analysis:
   (a) alisertib, (b) crizotinib, (c) dabrafenib, (d) Eg5_9814, (e)
   trametinib, (f) elephantin, (g) epirubicin, (h) ERK_2440, (i) ERK_6604,
   (j) ulixertinib, (k) PAK_5339, (l) PD0325901, and (m) SCH772984 (refer
   to [104]Table 1 for the drug mechanisms of action).

Table 1.

   Drugs with significant differences between the two clusters after the
   drug sensitivity analysis.
   Drug Antitumor mechanism P
   Alisertib Alisertib has antitumor activity and induces apoptosis and
   autophagy by targeting the AKT/mTOR/AMPK/p38 pathway in leukemic
   cells.^[105]23 <0.001
   Crizotinib In vitro, Crizotinib was shown to be a potent inhibitor of
   tumor growth, through inhibition of tyrosine phosphorylation of NPM-ALK
   and tyrosine phosphorylation of c-Met.^ [106]24 <0.001
   Dabrafenib Dabrafenib is an ATP-competitive Raf inhibitor.^ [107]25
   <0.001
   Elephantin Elephantin is a sesquiterpenoid tumor inhibitor.^ [108]26
   <0.001
   Epirubicin Epirubicin is a Foxp3 inhibitor that inhibits the inhibition
   of DNA and RNA synthesis.^ [109]27 <0.001
   PD0325901 (Mirdametinib) PD0325901 is a selective and
   non-ATP-competitive MEK inhibitor that inhibits p-ERK1/2 expression and
   induces apoptosis.^ [110]28 <0.001
   SCH772984 [111]CH772984 is a highly selective and ATP-competitive ERK
   inhibitor with antitumor activity.^ [112]29 <0.001
   Trametinib Trametinib is a potent MEK inhibitor that activates tumor
   cell autophagy and induces apoptosis.^ [113]30 <0.001
   Ulixertinib Ulixertinib is a highly selective and ATP-competitive
   ERK1/2 inhibitor that inhibits the phosphorylation of ERK2 and the
   downstream protein RSK.^ [114]31 <0.001
   Eg5_9814 GDSC database <0.001
   ERK_2440 GDSC database <0.001
   ERK_6604 GDSC database <0.001
   PAK_5339 GDSC database <0.001
   [115]Open in a new tab

   AKT/mTOR/AMPK: protein kinase B/mammalian target of
   rapamycin/AMP-protein activated protein kinase; NPM-ALK:
   nucleophosmin-anaplastic lymphoma kinase; ATP: adenosine triphosphate;
   MEK: mitogen-activated extracellular signal-regulated kinase; ERK:
   extracellular signal-regulated kinase; RSK: ribosomal S6 kinase.

Identification of hub genes and enrichment analysis

   A total of 16 hub genes were obtained, based the criteria of FDR of
   <0.05 and log[2] FC >|1.0|: MATN3, SPACA3, MICU3, ZFP37, RIPPLY3,
   LY6G6F-LY6G6D, COMMD7, CHN2, DCN, BCHE, PRAMEF12, ZNF461, EMCN, CD36,
   GPRC5D, and CCDC152. As shown in [116]Figure 7(c), 15 of these genes
   were downregulated, and one gene was upregulated. The GO enrichment
   analysis of these hub genes revealed that the main molecular function
   involved the extracellular matrix structural constituent, amyloid-beta
   binding, Toll-like receptor binding, and low-density lipoprotein
   particle receptor activity ([117]Figure 7(d)). The KEGG enrichment
   analysis revealed that these genes are involved in the adipocytokine
   signaling pathway, cholesterol metabolism, fat digestion, and
   absorption ([118]Figure 7(d)). The transcription factor analysis
   ([119]Figure 7(e)) revealed the enrichment of POU6F1 (POU class 6
   homeobox 1). The established ceRNA network revealed that one gene
   (MATN3), three miRNAs (hsa-miR-577, hsa-miR-20a-5p, and
   hsa-miR-186-5p), and 23 lncRNAs (DUBR, MIR99AHG, [120]AC127070.2,
   LINC02381, [121]AC015712.2, [122]AC015871.3, [123]AL031123.2, PINK1-AS,
   LINC02035, [124]AL157392.3, ARRDC1-AS1, THAP7-AS1, SGMS1-AS1,
   [125]AC008393.1, BOLA3-AS1, NIFK-AS1, OIP5-AS1, ARHGAP5-AS1, LINC01128,
   SCAMP1-AS1, ARMCX5-GPRASP2, and ERVK13-1) were involved in the
   inter-cluster variation.

Inter-cluster transcription factor regulatory network

   Transcription factors control chromatin accessibility and transcription
   by recognizing specific DNA sequences and forming complexes to direct
   the gene expression. The results would lead to DNA-binding
   transcriptional repression activity, RNA polymerase II specificity, and
   sequence-specific double-stranded DNA-binding activity. In addition,
   these are involved in the negative regulation of transcription by RNA
   polymerase II.

Nomogram construction and verification

   LASSO logistic regression analysis was performed to select the
   important predictive features in platelet adhesion-related genes and
   prevent overfitting ([126]Figure 9(a)). A total of seven genes were
   selected using the LASSO logistic regression algorithm (the specific
   LASSO coefficients of genes are presented in [127]Supplemental Table
   S2). Then, the biomarkers that improved the prognosis of COAD were
   assessed by univariate and multifactorial analyses, according to
   p < 0.05, and the prediction model was established ([128]Figure 9(b)).
   Two platelet adhesion-related genes, ALOX12 and ACTG1, were included.
   The clinical features (primary therapy outcome and pathologic stage)
   were also included in the prognostic model of Harrell's C index, and
   these were used to evaluate the discriminatory power of the model
   (0.879 [0.848–0.910]). The nomogram presented in [129]Figure 9(e) was
   based on the results of this model and was created to predict the
   one-year and three-year OS rates. The calibration plots presented in
   [130]Figure 9(f) reveal the excellent calibration of the nomogram, and
   the receiver operating characteristic (ROC) curves for the nomogram for
   OS (area under the curve [AUC] for one-year  =  0.749, AUC for three
   years  =  0.712). Furthermore, compared to the other prognostic models
   in the four separate studies in [131]Figure 9(g),^[132]32–[133]35 this
   novel nomogram, which contained the platelet adhesion gene data, had
   good predictive power.

Figure 9.

   [134]Figure 9.
   [135]Open in a new tab

   Construction and verification of the nomogram. (a) Plots for the LASSO
   regression coefficients over the different values of the penalty
   parameter. (b) The LASSO regression analysis result. (c) RT-qPCR
   results for ALOX12 or ACTG1 expression at the mRNA level. (d)
   immunohistochemistry results were evaluated semiquantitatively for
   ALOX12 or ACTG1 expression at the protein level. (e) Nomogram for OS.
   (f) Calibration plots for the nomogram, and ROC curves for the
   nomogram. (g) Comparison of prognostic models in the different studies.

   CR: complete response; PR: partial response; SD: stable disease; PD:
   progressive disease; LASSO: least absolute shrinkage and selection
   operator; RT-qPCR: real-time quantitative polymerase chain reaction;
   OS: overall survival; ROC: receiver operating characteristic.

Expression validation of key genes

   The expression of two key genes (POU6F1 and MATN3) in the regulatory
   network and two genes (ALOX12 and ACTG1) were further investigated in
   the clinical samples. The mRNA expression in normal and tumor tissues
   obtained from four COAD patients from real world was detected by
   RT-qPCR, and the protein expression in normal and tumor tissues
   obtained from four COAD patients from real world was detected by
   immunohistochemistry. As shown in [136]Figure 7(g) and (h), the mRNA
   and protein expression levels of POU6F1 were significantly lower in
   tumor tissues, when compared to normal tissues, while the mRNA
   expression levels of MATN3 were not statistically different between
   these tissues. The expression of MATN3 protein was, however,
   statistically different. Furthermore, the mRNA and protein expression
   levels of ALOX12 were significantly lower in tumor tissues, when
   compared to normal tissues ([137]Figure 9(c) and (d)), while the mRNA
   and protein expression levels of ACTG1 were significantly higher in
   tumor tissues, when compared to normal tissues ([138]Figure 9(c) and
   (d)).

Discussion

   Colon cancer rapidly progresses after initiation. Therefore, more
   effective drugs or therapeutic targets are required to improve the
   survival rate. Through the multidimensional analysis of tumors, it is
   possible to uncover the weaknesses of tumors and target these for
   treatment. There is increasing evidence that points to the involvement
   of platelets in the development of tumors.^[139]2,[140]36 The present
   study initially revealed that high platelet count is associated with
   poor OS. Using the data obtained from four large studies, it was found
   that the mutations, amplifications, profound deletions, and multiple
   alterations in these genes lead to poor survival.

   Based on the DEGs associated with platelet adhesion, the COAD tumor
   expression matrix was further divided into two groups using NMF
   clustering analysis. The survival analysis of these data revealed that
   the patient survival rates were better in Cluster2, when compared to
   Cluster1. Normally, alterations to copy numbers are detrimental, and
   these may be contributing factors to various human diseases.^ [141]37
   In the present study, 16 genes were identified to have more copy number
   increases than deletions in Cluster1, and 30 genes were identified to
   have more copy number deletions than increases. In Cluster2, 29 genes
   had more copy number increases than deletions, and 36 genes had more
   copy number deletions than increases. Next, an enrichment analysis was
   performed for these genes to gain a deeper understanding of the
   differences between the two clusters. It was found that the PI3K-Akt
   signaling pathway was present in both clusters in the copy number
   amplification groups, suggesting that platelet adhesion-related genes
   may play a role in the development of COAD through the PI3K-Akt
   signaling pathway. These results also revealed that platelet
   adhesion-related genes may be involved in the remodeling of the tumor
   immune microenvironment.

   It is important to investigate the differences in immune cell
   infiltration between clusters, since tumor tissues contain stromal
   cells, fibroblasts, and immune cells, which all contribute to the tumor
   microenvironment.^[142]38,[143]39 A previous study revealed that the
   degree of immune cell infiltration in tumor tissues plays an important
   role in tumor development and treatment.^ [144]40 The results of the
   present study revealed that five immune cells were significantly
   different between the two groups: CD56^bright, CD56^dim, γδT cells,
   T[h]17 cells, and T[h]2 cells. These data were consistent with the
   enriched pathways, according to the CNV genes identified above.
   Combined with the survival analysis results for the two clusters, the
   increase in mature NK cells and T[h] cells at the tumor site may be a
   breakthrough point to improve the immune microenvironment of COAD.

   A small subset of cancer stem cells (CSCs) can self-renew, indefinitely
   proliferate and differentiate in multiple directions. These are the
   primordial cells responsible for tumorigenesis and recurrence.^ [145]41
   Previous studies have revealed that CSCs are the most important cause
   of malignancy treatment failure. This is due to the tumorigenesis, and
   the phenotypic differences in the differentiated progeny of tumor
   cells. Furthermore, it was found that cancer stemness was higher in
   Cluster2, when compared to Cluster1. Moreover, the survival rates were
   better in Cluster2, when compared to Cluster1. This illustrates the
   complexity of the development process of COAD. In order to provide
   guidance in clinical medication, the two clusters were assessed for
   drug sensitivity. A total of 13 drugs were identified. These drugs have
   been or will be used in clinics for the treatment of tumors.

   Next, the differences in gene expression between the two clusters were
   explored by performing an enrichment analysis. Interestingly, the GO
   and KEGG analysis results revealed that the identified hub genes were
   involved in lipid and cholesterol metabolism in COAD. A growing number
   of studies have demonstrated the importance of these two pathways in
   tumor development.^[146]42–[147]45 Meanwhile, the transcription factor
   pathway enrichment analysis of hub genes was performed using the DAVID
   tool to construct a transcription factor regulatory network.
   Transcription factors are important hubs that connect the signaling and
   protein expression in vivo. The construction of transcription factor
   regulatory networks among clusters can help in understanding the
   potential regulatory mechanism of platelet adhesion-related genes in
   COAD. The analysis revealed that 10 hub genes were enriched in POU6F1,
   which is a member of the POU factor family of transcription factors.^
   [148]46 Interestingly, in the present study, POU6F1 was identified to
   have a low expression in tumor tissues, when compared to normal
   tissues. Another study reported that downregulation of POU6F1 in lung
   adenocarcinoma (LUAD) is associated with poor prognosis.^ [149]47 The
   enrichment results revealed that platelet adhesion-related genes may be
   involved in the regulation of gene expression, providing novel insights
   into the treatment of COAD. However, further experimental
   investigations are required.

   In order to elucidate the potential regulatory mechanisms of ceRNA in
   COAD, another ceRNA regulatory network was constructed to identify new
   therapeutic targets at the lncRNA–miRNA–mRNA level. A survival analysis
   was conducted for all hub genes, and three relevant miRNAs and 23
   lncRNAs were obtained in MATN3 as a single target gene. The MATN3 gene
   encodes a member of the von Willebrand factor A domain, and this has
   been considered to be involved in the formation of filamentous networks
   in extracellular matrices.^ [150]48 Furthermore, the MATN3 gene has
   received increasing attention in recent years, especially in gastric
   cancer research, due to its potential to serve as both a diagnostic
   biomarker and a prognostic biomarker.^[151]49–[152]51 Moreover, MATN3
   acts as an oncogenic factor in gastric cancer, which is in agreement
   with the findings of the present study. Although the MATN3 expression
   was not statistically different in the COAD and normal tissues,
   treatments that target MATN3 can be a promising therapeutic strategy
   for other cancers. Unfortunately, the mechanism, in which the MATN3
   gene influences cancer development and progression, has received
   minimal attention to date.

   Meanwhile, the miRNAs and lncRNAs identified in the present study have
   received increasing attention in tumor research. Hsa-mir-577 has been
   mentioned in several studies,^[153]52–[154]55 but this appears to have
   a dual nature, depending on the cancer type. Atypical endometrial
   hyperplasia may be associated with the aberrant expression of
   hsa-miR-577, which targets IGF1 and IGF1R, while it was found that
   hsa-miR-577 significantly downregulates the translation efficiency of
   CYP3A4 mRNA in the liver.^[155]55,[156]56 A study reported that the
   downregulation of hsa-mir-20a-5p by atorvastatin can increase the
   low-density lipoprotein receptor (LDLR) mRNA and protein expression in
   HepG2 cells, while the overexpression of has-mir-20a-5p can reduce the
   LDLR mRNA and protein expression.^ [157]57 In another study on COAD,
   hsa-mir-186-5p was described as a tumor suppressive factor that
   inhibits the SMAD6 and SMAD7 expression, thereby regulating the tumor
   growth factor-β signaling pathway.^ [158]58

   lncRNAs can serve as diagnostic biomarkers for a variety of cancers.
   Four lncRNAs (ARRDC1-AS1, OIP5-AS1, BOLA3-AS1, and NIFK-AS1) were
   mentioned in previous COAD studies.^[159]59–[160]62 Collectively, these
   factors may lead to the further exploration of COAD progression and can
   be investigated as biomarkers and therapeutic targets for COAD.

   DUBR, MIR99AHG, PINK1-AS, SGMS1-AS1, LINC01128, and SCAMP1-AS1 were
   investigated in LUAD.^[161]63–[162]68 Furthermore, it was reported that
   ARHGAP5-AS1 is upregulated in chemoresistant gastric cancer cells and
   that the knockdown of the lncRNA reverses the resistance.^ [163]69
   [164]AL031123.2 can be used as a prognostic factor in cervical cancer.^
   [165]70 There is a significant correlation between the [166]AL157392.3
   expression and glycolysis across a wide range of cancers.^ [167]71
   THAP7-AS1 is significantly upregulated in gastric cancer tissues, when
   compared to non-tumorous gastric tissues.^ [168]72 However, the
   specific mechanistic role of these aforementioned lncRNAs in COAD
   remains unclear. Thus, more in-depth experimental studies are required.

   Finally, in the LASSO logistic regression analysis, two prognostic
   factors (ALOX12 and ACTG1) were identified. The ALOX12 gene encodes a
   member of the lipoxygenase family of proteins. This encoded enzyme acts
   on different polyunsaturated fatty acid substrates to generate
   bioactive lipid mediators, including eicosanoids and lipoxins.^ [169]73
   Interestingly, ALOX12 is a pro-oncogenic factor, which may promote
   tumor development through lipid metabolic pathways, and this has been
   confirmed in several studies.^[170]74,[171]75 In addition, ACTG1 is
   associated with cell migration and adhesion.^ [172]76 The clinical
   utility of the nomogram was demonstrated using ROC curves and
   calibration plots. The calibration plots and ROC curves of the nomogram
   were well-calibrated.

   The present study had several limitations. First, further validation of
   the efficacy and standardization of the two clusters in clinical
   diagnosis is necessary, but remains challenging. Second, the present
   study lacked experimental studies on platelet adhesion in patients.
   Finally, the multidimensional analysis results for platelet-associated
   adhesion genes had significant practical implications. Therefore, there
   is an urgent need to further explore existing results, in order to
   identify novel treatment options for COAD, which could not be completed
   in the present study.

Conclusion

   Based on the presented data, it was revealed that the effects of
   platelet adhesion-related genes in COAD are complex. On the one hand,
   platelet adhesion-related genes can affect the OS of COAD patients by
   regulating lipid and cholesterol metabolism, which may be caused by the
   differences in CNVs. On the other hand, several immune cells have
   varying degrees of infiltration, according to the different platelet
   adhesion-related gene expression levels, which may cause the tumors of
   different patients to respond differently to the same drug. Finally,
   the constructed nomogram can be used as a reference in clinical
   practice.

Supplemental Material

   sj-docx-1-dhj-10.1177_20552076231203902 - Supplemental material for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm
   [173]Click here for additional data file.^ (1.4MB, docx)

   Supplemental material, sj-docx-1-dhj-10.1177_20552076231203902 for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm by
   Xiao-jv Chi, Yi-bei Song, Deng-he Liu, Li-qiang Wei, Xin An, Zi-zhen
   Feng, Xiao-hua Lan, Dong Lan and Chao Huang in DIGITAL HEALTH
   sj-doc-2-dhj-10.1177_20552076231203902 - Supplemental material for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm
   [174]Click here for additional data file.^ (402KB, doc)

   Supplemental material, sj-doc-2-dhj-10.1177_20552076231203902 for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm by
   Xiao-jv Chi, Yi-bei Song, Deng-he Liu, Li-qiang Wei, Xin An, Zi-zhen
   Feng, Xiao-hua Lan, Dong Lan and Chao Huang in DIGITAL HEALTH
   sj-xlsx-3-dhj-10.1177_20552076231203902 - Supplemental material for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm
   [175]Click here for additional data file.^ (8.9KB, xlsx)

   Supplemental material, sj-xlsx-3-dhj-10.1177_20552076231203902 for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm by
   Xiao-jv Chi, Yi-bei Song, Deng-he Liu, Li-qiang Wei, Xin An, Zi-zhen
   Feng, Xiao-hua Lan, Dong Lan and Chao Huang in DIGITAL HEALTH
   sj-xlsx-4-dhj-10.1177_20552076231203902 - Supplemental material for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm
   [176]Click here for additional data file.^ (10.3KB, xlsx)

   Supplemental material, sj-xlsx-4-dhj-10.1177_20552076231203902 for
   Significance of platelet adhesion-related genes in colon cancer based
   on non-negative matrix factorization-based clustering algorithm by
   Xiao-jv Chi, Yi-bei Song, Deng-he Liu, Li-qiang Wei, Xin An, Zi-zhen
   Feng, Xiao-hua Lan, Dong Lan and Chao Huang in DIGITAL HEALTH

Acknowledgements