Abstract

Objective

   This study aimed to analyze immune-related genes and immune cell
   components in the peripheral blood of patients with acute myocardial
   infarction (AMI).

Methods

   Six datasets were obtained from the GEO repository comprising 88
   healthy samples and 215 AMI samples. We performed the weighted gene
   co-expression analysis (WGCNA) and five machine learning (ML) methods
   to identify immune-related genes and construct diagnostic models.
   CIBERSORT algorithm was adopted for the assessment of the degree of
   immune infiltration. Finally, RT-PCR, immunofluorescence double and
   immunohistochemistry were conducted to analyze the expression level of
   the identification of featured immune-related genes and localization
   relationship in heart tissue of AMI mouse model.

Results

   A total of 496 immune-related DEGs were obtained between AMI and normal
   samples. WGCNA finally determined the co-expression modules that showed
   the most significantly positively associated with AMI (r=0.41;
   P<0.001). Among the five ML models, XGBoost had the highest AUC (0.849)
   and accuracy (0.812) to discriminate patients with AMI from normal in
   the validation sets. Furthermore, we found that the proportion of
   chemokine receptor (CCR), macrophages, neutrophils, and Treg cells in
   the AMI groups was significantly higher than that in the normal groups.
   In vitro RT-PCR verification revealed that SOCS3, MMP9, and AQP9
   expression increased significantly in the AMI mouse model. Among the 22
   immune cells, AQP9, MMP9, and SOCS3 displayed the strongest positive
   correlation with neutrophils. In MI-mice, MPO stained strongly along
   the lateral cardiomyocytes, whereas it was weaker in sham mice.
   Combined immunofluorescence was observed in same parts of the cytoplasm
   of cardiomyocytes in myocardial infarction area, indicating
   co-localization of MPO with MMP9 and SOCS3 in these areas,
   respectively.

Conclusion

   Immune-related genes and immune cells are intimately related to AMI.
   Constructing different ML models based on these biomarkers could be a
   valuable approach to diagnosing AMI in clinical practice.

   Keywords: acute myocardial infarction, immune-related genes, immune
   infiltration, machine learning, diagnostic model

Introduction

   Acute myocardial infarction (AMI) is a leading cause of cardiovascular
   disease mortality around the world, causing a tremendous burden to
   global public health. In America, someone experiences a myocardial
   infarction approximately every 40 seconds.[46]^1 Two thirds of all AMIs
   present no ST-elevation.[47]^2 As for the cause of AMI, plaque rupture
   remains the most common etiology, but unique syndromes such as plaque
   erosion, coronary microvascular dysfunction, spontaneous coronary
   artery dissection, and coronary spasm related to drug use are more
   prevalent in this age group.[48]^3 Cardiovascular disease remains the
   main cause of morbidity and mortality among young people worldwide,
   even as significant progress has been made in improving diagnosis and
   therapies.[49]^4 At present, we face the challenges consisting of
   atypical and delayed presentation, nonadherence to treatment, and
   syndromes unique to young men. Therefore, we need to look for a
   potential method that is sufficiently stable and reliable to help
   diagnose AMI in the early stages regardless of atypical symptoms.[50]^5

   Numerous studies have been performed to better understand the
   mechanisms involved in the pathogenesis of AMI. From these, immune
   responses have been reported to play an important role in AMI, and
   require the coordinated efforts of multiple immune-related genes and
   various immune cells.[51]^6–8 Various immune cells participate in
   immunomodulation after AMI, collectively working to remove necrotic
   tissue and rebuild the injured myocardium[52]^9 The inflammatory
   response plays a central role in the development of myocardial
   necrosis, ventricular remodeling, dysfunction, and coronary heart
   failure after AMI.[53]^7^,[54]^10 For example, MMP9 expression in AMI
   patients has been found to be significantly higher than in controls,
   and there is a time-dependent relationship between plasma levels of
   levels of NT-proBNP and MMP-9.[55]^11 Enhanced activation of
   cardioprotective signaling pathways by inhibiting myocardial suppressor
   of cytokine signaling 3 (SOCS3) expression prevented left ventricular
   remodeling after AMI.[56]^12 In addition, studies have shown that
   immune cell infiltration plays an increasingly important role in the
   development and progression of AMI.[57]^6^,[58]^7^,[59]^13 Various
   types of cell death have been identified in AMI, including necrosis,
   apoptosis, autophagy-related cell death, and pyroptosis.[60]^7^,[61]^14
   Numerous preclinical studies have shown that macrophages, mast cells,
   and eosinophils, which influence cardiac function after myocardial
   infarction (MI), are involved in the immune regulation of the infarcted
   heart.[62]^7^,[63]^13 Therefore, therapies aimed at regulating the
   immune response after AMI are a good choice for the treatment of AMI,
   and are promising in reducing complications and mortality.[64]^15

   The roles played by immune-related genes and immune cells, as well as
   the associations between them in patients with AMI, have not been fully
   investigated. Therefore, we analyzed immune-related genes and immune
   cell components in peripheral blood of patients with AMI using
   bioinformatic strategies. In addition, machine learning is a very
   effective approach when dealing with high-dimensional datasets,
   including genomic, epigenomic, proteomic or metabolomic data
   sets.[65]^16 Therefore, we compared five different machine learning
   algorithms to obtain better diagnostic efficacy and identify immune
   related genes in patients with AMI, which is expected to provide new
   potential targets for diagnosis and prognosis, and thus improve the
   outcome of patients with AMI.

Materials and Methods

Systematic Search of Microarray Expression Profiling Datasets

   AMI-related keywords were searched in the Medical Subject Headings
   (MeSH) library ([66]https://www.ncbi.nlm.nih.gov/mesh). We then
   conducted a systematic search in the GEO repository
   ([67]https://www.ncbi.nlm.nih.gov/gds/) using the following search
   sentence: (((myocardial infarction) AND ((blood) OR PB)) AND Homo
   sapiens [Organism]) AND Expression profiling by array [Filter]. A
   report was included in the analysis if the following criteria were
   satisfied: (1) used a case-control design; (2) patients did not have
   any diseases other than AMI; and (3) patients were medication free. We
   finally obtained a total of six datasets ([68]Figure 1).

Figure 1.

   [69]Figure 1
   [70]Open in a new tab

   Data processing workflow.

Initial Data Processing and Removal of Batch Effect

   We used edgeR (version 4.1.4) to process all data from gene expression
   profiles. Through the package “limma”,[71]^16 we merged all samples and
   deleted those with coronary heart disease, stable angina, and repeated
   measurements ([72]Figure 1). Batch effects were removed by implementing
   the “sva” package in the R language.[73]^17 Numerically larger
   expression values from six data sets were log2 transformed prior to
   cross-platform normalization. For systematic evaluations, ComBat
   outperforms other tools. Therefore, we chose ComBat to remove the batch
   effect between the two platforms.[74]^17 The ComBat method was used to
   normalize the expression values from different batches or platforms.
   Principal component analysis (PCA) was then conducted to confirm
   whether the batch effect was removed.

Differentially Expressed Genes

   The “limma” package was applied to select differentially expressed
   genes (DEGs) between AMI and control samples. We set the significance
   criteria as |logFC|>0.5 and P<0.05 and the DEGs were visualized by
   volcano plot.[75]^18 Then we chose the top 50 up-regulated genes and 50
   down-regulated genes to plot the heatmap using the “pheatmap” package.

Functional Analyses of DEGs

   To further reveal the biofunction of the selected DEGs, Gene Ontology
   (GO) enrichment analysis including biological process (BP), cellular
   component (CC) and molecular function (MF) analysis was performed using
   the “clusterProfiler” package. The criteria were set as P<0.01, the
   outcome was visualized in a barplot. Moreover, GObubble and GOChord
   were applied with the “GOplot” package to illustrate the functional
   analysis data.[76]^19 Likewise, Kyoto Encyclopedia of Genes and Genomes
   (KEGG) pathway analysis was performed by applying the “clusterProfiler”
   package to illustrate the enrichment of pathways in DEGs; the results
   were displayed in bar graph and circle graph.

Differentially Expressed Immune-Related Genes

   We obtained immune-related genes identified in previous studies from
   two databases and merged them ([77]https://www.immport.org/home;
   [78]https://www.innatedb.ca/). We then evaluated the expression of
   these genes in our 303 samples. The “limma” package was then applied to
   select immune-related DEGs between control and AMI samples and we set
   the significance criteria as P <0.05. The most significantly 20
   upregulated and 20 downregulated genes between the normal and AMI
   groups were inserted in the heat map graphically.

Weighted Gene Co‐Expression Network Analysis

   Weighted gene coexpression analysis (WGCNA) was performed to identify
   potential functional modules that could characterize the biological
   function of the samples.[79]^20 The minimum module size was set to 50
   genes. The soft threshold for the scale-free network was determined
   based on the maximal R^2 (power=6). The similarity of the topological
   overlap matrix was used to evaluate the distance between each pair of
   genes. Furthermore, hierarchical clustering analysis with the average
   method and the dynamic method was used to build the cluster tree and
   classify the genes into modules, respectively. Finally, we identified
   three functional modules. Then Spearman correlation coefficients and
   the corresponding p-values between two groups and functional modules
   were calculated by the cor function with using the Spearman method in
   the WGCNA package. The module most closely related to AMI was selected
   for further analysis. Next, we constructed a heatmap to illustrate the
   expression of three modules between normal and AMI samples.

Protein–Protein Interaction Network and Co-Expression Network

   As mentioned above, we selected the genes from the module most closely
   related to AMI. We mapped a protein–protein Interaction (PPI) network
   including the top 20 core genes using the online mapping tool “STRING”
   ([80]https://string-db.org/cgi/input.pl) and the minimum interaction
   distance was 0.9. Next, we performed a coexpression network analysis on
   these core genes using the “igraph” R package to explore the
   correlation intensity between these genes, and the results were plotted
   using a cut-off correlation coefficient of 0.4.[81]^22

Functional Analyses of Three Functional Modules

   As mentioned previously, GO enrichment analysis including BP, CC, and
   MF analysis was performed using the “clusterProfiler” package to reveal
   the biofunction of three modules. Meanwhile, KEGG pathway analysis was
   performed applying the “clusterProfiler” package to illustrate the
   enrichment of pathways in three modules, the results were displayed in
   bubble charts.

Establishment of the Machine Language Classifier

   We extracted the 235 immune-related gene expression profiles from the
   turquoise module and performed univariate logistic regression. The data
   set was randomly divided into 2:1 (202:101) as the training and
   validation sets. The following ML algorithms, LASSO, SVM, random
   forest, XGBoost, and neural network, were used to build diagnostic
   models for gene expression data and the R package “glmnet, e1071,
   randomForest, caret, and neuralnet” were used.[82]^21–25 ML models with
   default parameters, as defined in the Scikit Learn library
   ([83]http://scikit-learn.org/stable/), and from ML models after
   hyperparameter optimization were applied. Furthermore, the AUC,
   accuracy, precision, recall, specificity, and F1-score were calculated
   from Confusion Matrices to evaluate the classification capability of
   each model.[84]^26 Finally, the two models with the highest average AUC
   in validation sets were chosen without affecting the accuracy of
   classification efficiency. We selected the random forest and the
   XGBoost models and mapped the most important top 20 genes in both
   models, respectively, for further validation.

Immune Cell Infiltration Analysis

   To calculate the scores for infiltrating immune cells and to evaluate
   the activity of immune-related pathways in the control and AMI samples,
   the CIBERSORT algorithm was used to estimate the relative proportions
   of 22 types of infiltrating immune cells. This algorithm was run with
   the default LM22 signature matrix downloaded from the CIBERSORT portal
   and 1000 permutations using the “CIBERSORT” R package. Only samples
   with CIBERSORT p-value smaller than 0.05 were included. The correlation
   between immune genes and immune cells was determined by Spearman
   correlation.

Mouse Myocardial Infarction Model

   All animal experiments were approved by the Institutional Review and
   Ethics Board of the First Affiliated Hospital of Nanjing Medical
   University, Jiangsu Province Hospital, and were performed in accordance
   with the international ARRIVE (Animal Research: Reporting of In Vivo
   Experiments) guidelines for animal experiments.[85]^27 Mice were
   anesthetized with 2% isoflurane. The heart was retrieved from a small
   hole in the fourth intercostal space. The descending branch of the left
   coronary artery was located, sutured, and ligated approximately 3-mm
   from its origin with 6‐0 silk thread. After ligation, the heart was
   returned to the chest and the muscles and skin were sutured closed.
   Except for ligation, the sham operation group was the same as the
   control group. Each group contained six mice. After 3 days, the mice
   were administered a high-dose of anesthesia with 2% isoflurane and
   specimens were collected after myocardial perfusion with normal saline.
   The tissue was stored in liquid nitrogen.

Quantitative RT-PCR Assessed the Expression Level of Potential Immune-Related
Genes

   Total RNA was isolated from mice heart tissues using Trizol and cDNA
   was synthesized using reverse transcription kits (Takara, Shiga, Japan)
   according to the manufacturer’s instructions. Quantitative reverse
   transcription PCR was performed using SYBR Green Mix on an ABI7900HI
   (Thermo Fisher Scientific). The expression of genes was normalized to
   that of reference genes, including 18S. The primer sequences are listed
   in [86]Supplementary Table S2.

Double Immunofluorescence

   Immunohistochemical Staining performed after being dewaxed of mice
   tissue sections. Sections permeabilized with 0.1% Triton X-100, and
   blocked with 10% goat serum at room temperature for 1 h. Samples were
   incubated with anti-rabbit SOCS3 antibody (for;1:500, ab16030, Abcam,
   USA), anti-rabbit MMP9 (1:500, ab76003, Abcam, Cambridge, MA, USA),
   anti-rabbit-AQP9 antibody (1:600, ab84828, Abcam, Cambridge, MA, USA),
   and anti-rabbit-MPO antibody (1:600, [87]EPR20257, Abcam, Cambridge,
   MA, USA) at 4°C overnight. On the 2nd day, slides were incubated with
   goat anti-rabbit/rabbit FITC secondary antibody for 1h at room
   temperature. The nuclei were dyed by 4′,6-diamidino-2-phenylindole
   (DAPI) for 5min (4′,6-diamidino-2-phenylindole, Vector, ZsBio, Beijing,
   China). Images were obtained using Zeiss fluorescence microscope
   (Germany) and Nikon microscope (Nikon, Tokyo, Japan). At least five
   fields for each coverslip were captured. The number of cells, antigen
   distribution, and cellular fluorescence intensity that represented
   antigenicity were measured using the captured images. The image
   analysis was performed with ImageJ (NIH, Bethesda, MD, USA).

Results

Data Set Collection, Study Population, and Preprocessing of Data

   In this study, six microarray datasets from the GEO database were
   included for analysis: [88]GSE29111, [89]GSE48060, [90]GSE61145,
   [91]GSE62646, [92]GSE66360, and [93]GSE123342, with a sample size of
   52, 52, 57, 98, 99, and 192, respectively ([94]Supplementary Table S1).
   We merged all the samples and deleted the ones with coronary heart
   disease, stable angina, and repeated measurements. Finally, 88 healthy
   controls and 215 patients with AMI were included ([95]Figure 1). In
   addition, we removed the batch effect from different platforms and
   different batches between the datasets ([96]Figure 2).

Figure 2.

   [97]Figure 2
   [98]Open in a new tab

   Principal component analysis (PCA) of gene expression datasets. (A)
   Gene expression profiles without the removal of the batch effect. (B)
   Gene expression profiles with removal of batch effect.

Identification of DEGs Between Normal and AMI Samples and Functional Analyses

   Gene expression levels were compared in the 88 normal and 215 AMI
   samples and we identified 3279 DEGs. After removing genes with low
   expression levels, the heatmap of the top 50 up-regulated and 50
   down-regulated genes selected according to the screening criteria
   (|logFC|>0.5, P<0.05) are shown in [99]Supplementary Figure S2. GO
   enrichment analysis was performed for the 157 DEGs selected to identify
   significantly enriched GO terms. GObar revealed the significantly
   enriched GO terms (P < 0.01). The results of the most significantly
   enriched GO terms were as follows: BP: neutrophil activation involved
   in immune response and neutrophil activation; CC: secretory granule
   lumen, cytoplasmic vesicle lumen, and vesicle lumen; MF: carbohydrate
   binding. GOChord was carried out to define the relationship between
   DEGs and their associated terms. We could see that DEGs were enriched
   into eight GO terms, and neutrophil activation was involved in the
   immune response, neutrophil activation, neutrophil degranulation, and
   neutrophil-mediated immunity contained the most DEGs
   ([100]Supplementary Figure S2A and [101]S2B). Therefore, the findings
   indicated that neutrophils were closely related with DEGs and may play
   an important role in the progression of AMI. Furthermore, through KEGG
   pathway analysis, we discovered that osteoclast differentiation was the
   most significantly enriched pathway among the DEGs ([102]Supplementary
   Figure S2C and [103]S2D).

Identification of Differentially Expressed Immune-Related Genes

   After downloading the immune-related genes from the website, we
   compared their expression between 88 normal and 215 AMI samples and
   identified 496 DEGs (all P<0.05). Among these, 167 genes were
   down-regulated, while 329 other genes were enriched in the AMI group.
   The 20 most significant upregulated (such as TLR2, IL1R2, AQP9) and
   downregulated (such as PTGDR, PRF1, ADRB2) genes between the normal and
   AMI groups were graphically illustrated in a heatmap ([104]Figure 3).

Figure 3.

   [105]Figure 3
   [106]Open in a new tab

   Heatmap showing the top 20 up-regulated and 20 down-regulated
   immune-related DEGs in 88 normal and 215 AMI samples.

Identification of Gene Co-Expression Modules for Differentially Expressed
Immune-Related Genes

   To uncover the differences in immune-related genes between the normal
   and AMI groups, WGCNA was conducted to identify potential functional
   modules. The correspondence between immune-related DEGs and groups is
   shown in [107]Figure 4A and [108]B. We defined three modules (blue,
   turquoise, and grey) and each module was assigned with a unique color.
   Of the three modules, MEturquoise showed the most significant
   correlations with AMI (r=0.41; P<0.001). Furthermore, MEturquoise and
   MEgrey modules were positively correlated with AMI samples, while
   MEblue was negatively correlated with the AMI group ([109]Figure 4C).
   In addition, from the heatmap, we detected that the expression of the
   turquoise module genes was significantly different between the normal
   and AMI groups and tended to be up-regulated in the AMI samples
   ([110]Figure 4D).

Figure 4.

   [111]Figure 4
   [112]Open in a new tab

   Weighted gene co‐expression analysis (WGCNA) based on the
   immune-related DEGs. (A) Analysis of the scale-free index and mean
   connectivity for various soft-threshold powers. (B) Gene dendrogram
   obtained by average linkage hierarchical clustering. The row underneath
   the dendrogram shows the module assignment determined by the Dynamic
   Tree Cut. (C) Positive and negative correlation coefficients of the
   WGCNA modules between the normal and AMI groups. (D) The heatmap showed
   the expression of three modules between normal and AMI samples.

Construction of the PPI Network and the Co-Expression Network

   To illustrate the association between genes in turquoise module, we
   constructed a corresponding PPI network ([113]Figure 5A). Each node in
   the graph represented a gene and the connecting line connected the
   interacting genes in pairs. The colors of the connecting lines showed
   different degrees of PPI. STAT3, SYK, RELA, MAPK14, LYN, GRB2, MAPK1,
   AKT1, FOS, RAC1, JUN, MYD88, IL1B, HCK, TLR4, PRKCD, FCER1G, FGR,
   NFKBIA, and TLR2 were identified as hub genes ([114]Figure 5B). Next,
   we drew the co-expression network diagram of the top 20 core genes in
   red to indicate the positive correlation and in blue to indicated a
   negative correlation ([115]Figure 5C).

Figure 5.

   [116]Figure 5
   [117]Open in a new tab

   Maps illustrating the association between MEturquoise genes. (A) The
   PPI network of MEturquoise genes generated by the STRING platform. (B)
   Bar chart showing the top 20 hub genes. (C) The coexpression network
   showing correlation intensity of the top 20 hub genes and the cutoff of
   correlation coefficient is 0.4.

Functional Analyses of Three Functional Modules

   GO enrichment analysis for the three modules was performed to identify
   the significantly enriched GO terms (P<0.05): MEblue: BP: T cell
   activation; CC: external side of plasma membrane; MF: cytokine receptor
   activity. MEgrey BP: epithelial cell proliferation; CC: membrane raft;
   MF: receptor ligand activity. MEturquoise: BP: neutrophil activation;
   CC: secretory granule lumen, cytoplasmic vesicle lumen, and vesicle
   lumen; MF: immune receptor activity. ([118]Figure 6A–C). In addition,
   according to P adjusted < 0.05, significantly enriched KEGG pathways
   were screened: T cell receptor signaling pathway, Th17 cell
   differentiation, cytokine-cytokine receptor interaction, lipid and
   atherosclerosis, and NF-kappa B signaling pathway ([119]Figure 6D).

Figure 6.

   [120]Figure 6
   [121]Open in a new tab

   Functional analyses of three modules. Enrichment of cell functions
   evaluated by GO functional enrichment analysis of (A) biological
   process (BP), (B) cellular component (CC), and (C) molecular function
   (MF), (D) enrichment of cell function pathways evaluated by KEGG
   pathway enrichment analysis.

Establishment of the Diagnostic Model Based on Five ML Methods

   Univariate logistic regression of the 235 DEG of the turquoise module
   was carried out and we finally obtained 180 potential genes (P<0.01).
   Five ML classification methods were used for the diagnostic model. All
   models produced an average AUC > 0.85 in the training sets ([122]Figure
   7A). Among the five ML models, XGBoost had the highest AUC (0.849) and
   accuracy (0.812) to discriminate patients with AMI from normal in the
   validation sets. In addition, the random forest classifier also
   performed well, and the AUC and accuracy were 0.830 and 0.812
   respectively ([123]Figure 7B). Next, we mapped the most significant top
   20 DEGs using random forest (TLR2, FPR1, IRAK3, AQP9, NFIL3, C5AR1,
   TLR4, PTAFR, LILRB3, SOCS3, PLAUR, MMP9, BMX, CXCL16, JUN, S100A12,
   CD14, CLEC4D, SIGLEC9, PELI1) and XGBoost (AQP9, IL1R2, TLR2, SIGLEC9,
   HSPA6, MMP9, PGLYRP1, SOCS3, IRAK3, LILRB3, CD14, LYZ, LY96, VAV3,
   CSF2RB, HMGB2, NR4A3, JUN, ADRM1, PELI2) for further validation
   ([124]Figure 7C and [125]D). Nine overlapping genes were defined (AQP9,
   TLR2, SIGLEC9, MMP9, SOCS3, IRAK3, LILRB3, CD14 and JUN) in both two
   models ([126]Figure 7E).

Figure 7.

   [127]Figure 7
   [128]Open in a new tab

   Identification of featured immune-related genes in patients with acute
   myocardial infarction using machine learning (ML) methods. (A)
   Comparison of the AUC of the models with different ML classification
   algorithms in the training set. (B) Comparison of diagnostic
   performance of machine learning models with different feature selection
   methods in the validation set. (C) The most important top 20 genes
   selected by random forest. (D) The most important top 20 genes selected
   by XGBoost. (E) Venn diagram showed the overlapping genes screened by
   two machine learning.

   Abbreviations: LASSO, least absolute shrinkage and selection operator;
   SVM, support vector machine; XGBoost, extreme gradient boosting; AUC,
   area under the ROC curve.

Identification of Immune-Related Gene Expression in the Acute Myocardium
Infarction Mice Heart

   The expression of the nine overlapping genes (AQP9, TLR2, SIGLEC9,
   MMP9, SOCS3, IRAK3, LILRB3, CD14, and JUN) selected by two ML models
   were validated in the mouse myocardium. Between the sham and AMI
   groups, we found that the AQP9, SOCS3, and MMP9 genes had a
   significantly increased expression in AMI mice (P<0.001). However,
   other six genes had not shown any significant difference (all P > 0.05)
   ([129]Figure 8).

Figure 8.

   [130]Figure 8
   [131]Open in a new tab

   The expression levels of nine overlapping genes selected by two machine
   learning in mouse myocardial tissue analyzed by quantitative RT-PCR. *p
   < 0.05, **p < 0.01 and ***p < 0.001.

Immune Cell Infiltration Analysis

   Applying the CIBERSORT algorithm, we investigated the proportions of
   immune infiltration among patients with AMI and control samples in 22
   subpopulations of immune cells ([132]Supplementary Figure S3). Weak to
   moderate correlations were found between different subpopulations of
   immune cells ([133]Supplementary Figure S4). In addition, the
   proportion of neutrophils, monocytes, mast cells activated and T cells
   CD4 memory activated in the AMI groups was significantly higher than
   that in normal groups, while the proportion of B cells memory, T cells
   CD8, T cells CD4 naive, T cells CD4 memory resting, T cells gamma
   delta, and mast cells resting was lower than in normal groups
   ([134]Figure 9A).

Figure 9.

   [135]Figure 9
   [136]Open in a new tab

   Immune cell infiltration analysis. (A) Violin plot showing the
   significant changes of the immune infiltration level in the patients
   with AMI compared to the control group. (B) Lollipop plots illustrated
   the correlation between AQP9 expressions and the immune cells.

   We further explored the correlation between immune-related genes (AQP9,
   MMP9 and SOCS3), which are highly expressed in a mouse model of acute
   myocardial infarction, and immune cells. The results showed that among
   the 22 immune cells, AQP9 displayed the strongest positive correlation
   with neutrophils (r = 0.79, p < 0.001) and the strongest negative
   correlation with T cells CD8 (r = –0.46, p < 0.001) ([137]Figure 9B).
   MMP9 was also positively correlated with neutrophils (r = 0.63, p <
   0.001) and negatively correlated with T cells CD8 (r = −0.39, p <
   0.001) ([138]Supplementary Figure S5A). In addition, SOCS3 also showed
   positively correlated with neutrophils (r = 0.44, p < 0.001) and
   negatively correlated with T cells CD8 (r = −0.31, p < 0.001)
   ([139]Supplementary Figure S5B).

Expression of MPO and MMP9, SOCS3, and AQP9 in Acutely MI-Mice

   Healing from any injury requires granulocyte to remove debris and dead
   cells, and promote healing and repair. SOCS3, MMP9 and AQP9 proteins
   were significantly increased and positively correlated with neutrophil
   infiltration. To reveal the relationship between the distribution of
   MPO, SOCS3, MMP9, and AQP9 antigens in MI-mice and sham groups, we used
   double immunofluorescence. In MI-induced mice, MPO stained strongly
   along the lateral cardiomyocytes, whereas it was weaker in sham mice.
   Combined immunofluorescence was observed in same parts of the cytoplasm
   of cardiomyocytes in myocardial infarction area, indicating
   co-localization of MPO and MMP9, and MPO and SOCS3 in these areas,
   respectively ([140]Figure 10A and [141]B). In addition, our results
   revealed that SOCS3 protein was localized to the extra-neutrophils and
   peri-cardiomyocyte, while MMP9 protein was localized to the
   intracellular cardiomyocyte and extra-neutrophils. In the case of sham
   control, no significant labeling of MPO, MMP9 and SOCS3 was found.
   However, the results for AQP9 failed to show fluorescent
   co-localization with MPO, and immunohistochemical staining also
   revealed high AQP9 expression consistent with area of myocardial
   infarction, mainly focused on the cytoplasmic stroma ([142]Figure 10C).

Figure 10.

   [143]Figure 10
   [144]Open in a new tab

   Expression of MPO and MMP9, SOCS3, and AQP9 in acutely MI-mice
   myocardium by immunohistochemistry and immunofluorescence double
   staining were performed using histological sections of acute
   MI-associated injury, as well as the relationship between expression of
   individual MMP9, SOCS3, AQP9 and MPO invasion. (A) Immunofluorescence
   double staining with antibodies against MPO (red) and SOCS3 (green);
   MPO is concentrated in clusters over infarction area, and SOCS3 protein
   deposition over myocardial cell bodies. Hoechst 33258 dye stains the
   nuclei blue. Bars=20 μm. (B) Immunofluorescence double staining with
   antibodies against MPO (red) and MMP9 (green); MPO is concentrated in
   clusters over infarction area, and SOCS3 protein deposition over
   myocardial cell bodies. Hoechst 33258 dye stains the nuclei blue.
   Bars=20 μm. (C) Immunohistochemistry of AQP9 MI-mice cardiac tissues
   stained with anti-AQP9 antibodies showed positive staining within
   myocardial cytoplasm compare with sham-operation mice. Bars=20 μm. Data
   are expressed mean ± SD, n=4 sections per group. 400 magnifications. *P
   < 0.05 vs sham-operated mice.

Discussion

   AMI is the most serious manifestation of coronary artery disease,
   leading to millions of deaths annually in both developed and developing
   countries.[145]^28^,[146]^29 Despite recent achievements in reperfusion
   strategies and pharmacological treatment approaches, AMI remains a
   major problem for global health, affecting more than 7 million people
   worldwide every year.[147]^30 A large number of immune cells
   participate in immunomodulation after AMI to remove necrotic tissue and
   repair damaged myocardium. The innate immune response comprising the
   inflammatory, proliferative, and maturation phases is a necessary
   regulator during the process.[148]^31–33 Improvements in early
   interventions, such as reperfusion therapy and pharmacological
   treatment after AMI, have dramatically increased survival in recent
   decades. However, the process of myocardial reperfusion will
   paradoxically induce cardiomyocyte death and myocardial injury in
   itself, and is defined “myocardial reperfusion injury,” which accounts
   for approximately 50% of the final MI size.[149]^34 Therefore, new
   therapies that can limit the size of the myocardial infarct, prevent
   adverse ventricular remodeling, and reduce eventual heart failure due
   to structural heart disease after AMI are urgently required.

   AMI triggers a strong inflammatory response in the affected cardiac
   tissue. Most recently, new therapeutic tools based on immune-related
   intervention therapy may modulate unbalanced inflammation in damaged
   cardiac tissue, contributing to the resolution of this pathological
   condition. Currently, the results of a Phase II clinical trial
   involving 182 patients with non-ST-segment elevation myocardial
   infarction (NSTEMI) with anakinra, which is an IL-1 blocker (100 mg
   daily for 14 days) showed a reduction in the acute inflammatory
   response.[150]^35 In addition, another IL-1 blocker, named canakinumab,
   has been reported to significantly reduce the risk of recurrent acute
   coronary syndromes and the need for revascularization.[151]^36 As for
   other immune responses, in mouse models of permanent coronary ischemia
   and ischemia/reperfusion, it has been shown that short-term blockade of
   S100A9 reduces the local and systemic inflammatory response and
   significantly improves cardiac function.[152]^37 However, all of these
   therapies are still in the experimental stage and none of them has been
   applied clinically. Therefore, we should explore the specific
   mechanisms underlying human AMI and try to find both a valuable and a
   practical method to improve the outcome of patients undergoing AMI.

   In our study, we searched for the gene expression profiles of AMI
   samples from the GEO database and obtained 88 normal samples and 215
   AMI samples. The results of the bioinformatics analysis implied that
   the DEGs between normal and AMI samples were mainly enriched in
   neutrophil activation and regulation of inflammation signaling
   pathways. Although differential immune-related genes were stratified
   into three modules (MEblue, MEturquoise and MEgrey) showing the most
   significant correlations with AMI samples; the expression of the genes
   of the turquoise module were significantly upregulated in the AMI
   samples. There was a close association between the genes from the
   turquoise module and AMI. The biofunction of three modules were
   significantly different. Differential gene groups and their functional
   enrichment should be better defined in future studies.

   Apoptosis plays a major role in the pathophysiological processes
   following MI.[153]^38 Our study found that the STAT3 was the most
   important upregulated gene in the MEturquoise module, and the RT-PCR
   analysis indicated SOCS3 expression was markedly increased in MI mice.
   The STAT3/SOCS3 axis influences the T-cell repertoire, with SOCS3
   providing protection against autoimmune diseases by blocking Th17
   development.[154]^39 Previous animal studies have shown that myocardial
   STAT3 is a potent antiapoptotic, pro-survival signaling molecule during
   ischemia-reperfusion.[155]^40 Furthermore, cardiac-specific deletion of
   SOCS3 induced sustained activation of STAT3 and reduced myocardial
   apoptosis that prevented ischemia and reperfusion injury in mice.
   Furthermore, SOCS3 is the major negative modulator of Th17 function
   through STAT3 activation,[156]^41 and also constrains Th17 cell
   differentiation in T cells.[157]^42

   Our study showed that Th17 acts as the main activation pathway and
   IL-17 is the main effector of Th17 cells. IL-17 is an early promoter of
   T-cell-induced inflammation, which amplifies the inflammatory response
   by promoting the release of pro-inflammatory cytokines. After binding
   to the receptor, Il-17 plays its biological role through the MAP kinase
   pathway and the Nuclear factor kB (NFkB) pathway.[158]^43 Th17 cells
   can secrete and produce IL-17a, IL-17f, IL-6, and tumor necrosis factor
   alpha (TNF-a), which collectively mobilize, recruit, and activate
   neutrophils.[159]^44 It is well known that neutrophils are recruited
   into the infarcted area during the inflammatory phase, being attracted
   by cell debris and inflammatory signals.[160]^45 IL-17 produced by Th17
   cells can effectively mediate the excitatory process of neutrophil
   mobilization, thus effectively mediating the inflammatory response of
   tissues.

   We performed the CIBERSORT algorithm to explore the infiltration of
   immune cells in the normal and AMI group. The proportion of
   neutrophils, monocytes, mast cells activated and T cells CD4 memory
   activated in the AMI groups was significantly higher than that in
   normal groups, while the proportion of B cells memory, T cells CD8, T
   cells CD4 naive, T cells CD4 memory resting, T cells gamma delta, and
   mast cells resting was lower than in normal groups.

   Immunomodulation can be a two-edged sword. Considering repair of
   damaged tissue depends on timely suppression of inflammation, which is
   accompanied by mesenchymal activation to preserve tissue integrity. The
   relationship between myocardial function and structural integrity is
   complex, and the insufficient inhibition of inflammation in the damaged
   heart can lead to disastrous results, such as loss of cardiac muscle
   cells, inhibition of systolic function, ventricle enlargement, loss of
   integrity of the ventricular wall, and heart rupture of the
   heart.[161]^46 Considering the samples in our study were within 7 days
   of MI and not all were obtained within 24–72 hours, we speculated that
   the peak systemic inflammatory reaction had overcome the compensatory
   mechanism. However, MMP9, an indicator of fibrosis was markedly
   increased in the MI mouse model. It is worth affirming that
   inflammation stimulates fibrosis and accelerates healing at the site of
   MI. From another standpoint, many immune cells and immune activity
   participate in the initial pro-inflammatory response to AMI and in the
   anti-inflammatory reparative phase following AMI. In AMI patients, both
   of gene expressions related to complement, natural killer cells, CD3^+,
   CD8^+ T cells, B cells and the quantity of these immune cells have been
   reported to decrease.[162]^47 Another study indicated that CD8^+ T
   cells might contribute to acute coronary events through their
   pro-inflammatory cytotoxic effects.[163]^9^,[164]^48 In addition,
   neutrophils provide the first line of defense against invading
   microorganisms and tissue injury, and they also play an important role
   in the resolution of inflammation and repair.[165]^49 In our study, we
   found that CD8^+ T cells were enriched in normal samples while
   neutrophils were enriched in AMI samples. Therefore, we can develop new
   protocols targeting these immune cells to improve the outcome of
   patients with AMI.

   Several subsets of early recruited and activated neutrophils in cardiac
   necrosis are closely associated with post-infarction myocardial
   injury.[166]^50^,[167]^51 However, controversial results may be due to
   the fact that this treatment also eliminates macrophages and some
   systemic inflammatory molecules,[168]^52^,[169]^53 raising major
   concerns about the potential induction of immunosuppression and
   delaying clinical trials in humans. It is important to clarify the
   expression of neutrophil-associated cardiomyocyte proteins. Our study
   explored the correlation between AQP9, MMP9 and SOCS3 and immune cells
   in the acute MI-mice model, which were positively correlated with
   neutrophils. We elucidated for the first time that immune regulation by
   neutrophils may be able to target AQP9, MMP9 and SOCS3 for myocardial
   repair, providing a theoretical basis for immune-targeted therapy for
   MI.

   Recently, researchers have investigated new immunotherapy treatment
   schemes for AMI. Immune-related genes may be a good choice. For
   example, defects in the TLR2 or TLR4 gene in mice can prevent cardiac
   remodeling, effectively reducing maladaptive ventricular remodeling
   after AMI.[170]^54 The participation of FPR1 in the inflammatory
   response can protect against myocardial ischemia-reperfusion
   injuries.[171]^55 In rat cardiomyocytes, it was found that activation
   of FPR1 could contribute to the cardioprotective actions of Ac-ANX-A1
   (2-26).[172]^56 Furthermore, the study found that AQP9 could be
   inhibited by miR-212 overexpression by activating the PI3K/Akt
   signaling pathway, which results in the promotion of vascular
   regeneration, decreased cardiomyocyte apoptosis, and improving
   ventricular remodeling in rat models of AMI.[173]^57 In addition, the
   IL1R2 gene, which controls cell metabolism, as well as the immune
   response induced by many cytokines, plays a key role in AMI. The study
   also reported that IL1R2 contributes to atherosclerosis, which is
   closely associated with AMI.[174]^58^,[175]^59 In general, the DEGs
   TLR2, FPR1, AQP9, and IL1R2 also emerged in our study.

   We used five ML methods to identify featured immune-related genes in
   patients with AMI. Among these five classifiers, random forest and
   XGBoost performed well. The TLR2 and FPR1, AQP9, and IL1R2 were the
   most significant top 20 genes in the random forest and the XGBoost
   diagnostic model, respectively. Moreover, AQP9 showed significant
   differences in the in vivo animal model in our study. These genes can
   provide a novel therapeutic target for AMI and contribute to
   personalized medicine, which will benefit more patients. Based on the
   immune-related genes closely associated with the progression of AMI, we
   may formulate a new diagnosis workup, estimate prognosis, and provide
   targeted treatment. Definitely, significant study is need before
   targeted therapy can be introduced for clinical application.

   To date, targeted therapy for AMI is currently being evaluated in
   clinical trials. In this study, we comprehensively analyzed
   immune-related genes and immune cell components in the peripheral blood
   of patients with AMI using bioinformatics strategies. We used five ML
   methods to identify featured immune-related genes in AMI patients,
   among which high expression of AQP9, MMP9 and SOCS3 genes were
   validated in MI mouse model. Therefore, constructing ML models based on
   these biomarkers could be a valuable approach for diagnosing AMI in
   clinical practice. However, there were still some limitations in our
   study. First, our sample size was relatively small and we need more
   samples for further research. Second, we did not compare the
   characteristics between different groups of patients. Third, the
   specific usage of the diagnostic models was not elaborated.
   Furthermore, more clinical characteristics such as abdominal obesity or
   smoking should be involved to better reflect the true characteristics
   of AMI. Overall, our model still needs more validation in in vitro and
   in vivo experiments to confirm its clinical application.

Acknowledgments