Analysis of differentially expressed genes in bacterial and fungal keratitis.

Purpose: This study was aimed at identifying differentially expressed genes (DEGs) in bacterial and fungal keratitis. The candidate genes can be selected and quantified to distinguish between causative agents of infectious keratitis to improve therapeutic outcomes.
Methods: The expression profile of bacterial or fungal infection, and normal corneal tissues were downloaded from the Gene Expression Omnibus. The limma package in R was used to screen DEGs in bacterial and fungal keratitis. The Co-Express tool was used to calculate correlation coefficients of co-expressed genes. The "Advanced network merge" function of Cytoscape tool was applied to obtain a fusional co-expression network based on bacterial and fungal keratitis DEGs. Finally, functional enrichment analysis by DAVID software and KEGG analysis by KOBAS of DEGs in fusion network were performed.
Results: In total, 451 DEGs in bacterial keratitis and 353 DEGs in fungal keratitis were screened, among which 148 DEGs were found only in bacterial keratitis and 50 DEGs only in fungal keratitis. Besides, 117 co-expressed gene pairs were identified among bacterial keratitis DEGs and 87 pairs among fungal keratitis DEGs. In total, nine biological pathways and seven KEGG pathways were screened by analyzing DEGs in the fusional co-expression network.
Conclusion: TLR4 is the representative DEG specific to bacterial keratitis, and SOD2 is the representative DEG specific to fungal keratitis, both of which are promising candidate genes to distinguish between bacterial and fungal keratitis.

Diseases of the cornea are a major cause of blindness worldwide.[1] The etiology of corneal blindness encompasses a wide variety of inflammatory and infectious eye diseases that ultimately cause functional blindness.12] Keratitis is a type of corneal inflammation resulting in vision loss. It typically arises due to noninfectious causes such as eye trauma but can manifest as a result of microbial infection by pathogens such as fungi, bacteria, viruses or amebae.[3] Until now, infectious keratitis remains one of the main causes of corneal blindness and poses a diagnostic dilemma due to its varied presentation and visual morbidity.[14]

Currently, bacterial keratitis and fungal keratitis are the most common corneal infectious diseases posing a risk to patient vision.[56] The major causative pathogens for bacterial keratitis are Staphylococcus aureus and Pseudomonas aeruginosa.[7] Bacterial keratitis frequently leads to severe visual impairment from corneal ulceration, perforation, and scarring.[8] Following an infection, topical antimicrobial therapy is crucial for managing symptoms.910] Risk factors of fungal keratitis include ocular trauma, topical steroid use, ocular surface disease, and contact lens use.[11] Aspergillus spp., Fusarium spp., Candida spp., are the major causative pathogens of fungal keratitis among many.[12] Fungal keratitis commonly leads to poor visual acuity,[13] and is typically managed by polyenes and azoles.[14]

Each case of infectious keratitis must be confirmed by evaluating corneal infiltrate cultures.[1516] Clinically, corneal ulcers are often treated empirically without the use of microbiological analysis due to urgent requests for treatment to achieve optimal therapeutic outcomes.[16] In order to rely on empirical treatment, the clinician must distinguish between infectious agents based on clinical history, symptoms and characteristics. This method remains highly subjective and risky as incorrect identification of the pathogen facilitates further development of the corneal infection, ultimately leading to a worsened therapeutic outcome. Hence, it is necessary to identify and develop novel approaches to quickly recognize or identify bacterial versus fungal keratitis.

For the specific treatment of infectious keratitis, it is important to reveal the functional and molecular aspects of the disease to develop a possible treatment strategy. In recent years, the analysis of differentially expressed genes (DEGs) in disease has attracted a lot of attention, and may be a promising approach to develop more efficient treatments for keratitis. In this study, we aimed to screen the DEGs in bacterial and fungal keratitis by comparing total gene expression levels in infected versus healthy corneal tissues. This strategy will allow for the identification of candidate genes that can be therapeutically targeted to treat keratitis originating from different infectious agents.

Methods

Affymetrix microarray data

The transcription profile of GSE58291 was downloaded from the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/), a public functional genomics data repository that archives and freely distributes high-throughput molecular abundance data at the National Center for Biotechnology Information. In total, 30 corneal tissue samples were acquired. Among them, three samples showed empty expression profile data, numbered as GSM1406007 (fungal infection), GSM1406009 (bacterial infection), and GSM1406015 (normal control). Hence, 27 tissue samples (12 normal corneas, 7 bacteria-infected corneas, and 8 fungi-infected corneas) were reserved for bioinformatic analysis. Detailed information of the 27 samples is listed in Table 1. The causative organisms for bacterially infected corneas included Streptococcus pneumonia (n = 6) and Pseudomonas. aeruginosa (n = 1). The causative organisms for fungal keratitis were Fusarium sp. (n = 5), Aspergillus sp. (n = 2, A. flavus and A. terreus) and Lasiodiplodia sp. (n = 1). Platform information was GPL10558 Illumina Human HT-12 V4.0 expression beadchip. Platform annotation information of the chip expression profiles was also downloaded.

Table 1

Characteristic information of 27 samples

Source name	Sample	Comment (sample_title)	Causative organism
GSM1406021 1	Cornea	Cornea_bacterial_keratitis_rep8	Streptococcus pneumoniae
GSM1406018 1	Cornea	Cornea_fungal_keratitis_rep9	Aspergillus terreus
GSM1406017 1	Cornea	Cornea_bacterial_keratitis_rep7	Pseudomonas aeruginosa
GSM1406016 1	Cornea	Cornea_fungal_keratitis_rep8	Fusarium sp.
GSM1406013 1	Cornea	Bacterial keratitis rep6	Streptococcus pneumoniae
GSM1406011 1	Cornea	Cornea_bacterial_keratitis_rep5	Streptococcus pneumoniae
GSM1406008 1	Cornea	Cornea_bacterial_keratitis_rep3	Streptococcus pneumoniae
GSM1406006 1	Cornea	Cornea_fungal_keratitis_rep6	Fusarium sp.
GSM1406003 1	Cornea	Cornea_fungal_keratitis_rep5	Fusarium sp.
GSM1406002 1	Cornea	Cornea_fungal_keratitis_rep4	Fusarium sp.
GSM1406001 1	Cornea	Cornea_fungal_keratitis_rep3	Lasiodiplodia
GSM1406000 1	Cornea	Cornea_bacterial_keratitis_rep2	Streptococcus pneumoniae
GSM1405999 1	Cornea	Cornea_fungal_keratitis_rep2	Aspergillus flavus
GSM1405996 1	Cornea	Cornea_fungal_keratitis_rep1	Fusarium sp.
GSM1405994 1	Cornea	Cornea_bacterial_keratitis_rep1	Streptococcus pneumoniae
GSM1405992 1	Cornea	Cornea_normal_rep1	Normal noninfected tissue
GSM1405993 1	Cornea	Cornea_normal_rep2	Normal noninfected tissue
GSM1405995 1	Cornea	Cornea_normal_rep3	Normal noninfected tissue
GSM1405997 1	Cornea	Cornea_normal_rep4	Normal noninfected tissue
GSM1405998 1	Cornea	Cornea_normal_rep5	Normal noninfected tissue
GSM1406004 1	Cornea	Cornea_normal_rep6	Normal noninfected tissue
GSM1406005 1	Cornea	Cornea_normal_rep7	Normal noninfected tissue
GSM1406010 1	Cornea	Cornea_normal_rep8	Normal noninfected tissue
GSM1406012 1	Cornea	Cornea_normal_rep9	Normal noninfected tissue
GSM1406014 1	Cornea	Cornea_normal_rep10	Normal noninfected tissue
GSM1406019 1	Cornea	Cornea_normal_rep12	Normal noninfected tissue
GSM1406020 1	Cornea	Cornea_normal_rep13	Normal noninfected tissue

Data preparation and differential gene expression analysis

The raw expression profile data in text format were mapped to the corresponding gene names using the GPL10558 Illumina HumanHT-12 V4.0 expression beadchip platform. The average expression value was calculated as the single expression value of this gene, when multiple probes matched to the same gene. Then, logarithm to the base 2 (log2) of expression values was calculated to acquire approximately normally distributed gene expression data, which were continuously subjected to median normalization.1718] According to sample infection types, comparisons were performed in the bacterial infection versus normal control group, as well as in the fungal infection versus normal control group. Here, samples of normal cornea tissues were classified as the normal control group. The limma[19] package in R was used to screen DEGs by analyzing the gene expression data of corneal tissues from the above three groups. The Bonferroni's method[20] in multi-test package was applied to adjust raw P values for false discovery rate (FDR).[21] FDR <0.05 and the absolute value of log2FC >1 were used as cut-off criteria.

Comparisons of gene expression profiles

Gene expression profiles are species-specific, suggesting that gene expression is significantly altered in diseased tissues.[22] According to the expression profile of screened DEGs in bacteria versus fungal and normal samples, we extracted the expression value of DEGs in each sample from the downloaded expression value files. Then, the pheatmap package in R was used to generate expression values by biclustering[2324] based on Euclidean distance.[25] The results are shown as a heatmap.

Calculations of co-expression correlation coefficient among DEGs

Although there are approximately 25,000 genes in the human genome, only a fraction of these genes are expressed simultaneously in a single cell or specific tissues during a specific developmental stage.[26] There are many methods to identify whether co-expression exists between two genes, iamong which the most common method is to use Pearson's correlation coefficient.[27] To obtain DEGs with correlations, the CoExpress tool[28] (http://www.bioinformatics.lu/CoExpress/) was used to calculate correlation coefficients among co-expressed DEGs in the bacterial versus normal group or the fungal versus normal group. Finally, gene pairs with the absolute value of correlation coefficients >0.9 were retained.

Difference between bacterial versus normal DEGs and fungus versus normal DEGs

Through comparison of the gene expression profile between bacterial and normal groups, we acquired the screened DEGs, which are referred to as DEGs1. Similarly, through comparison between fungal and normal groups, we acquired more screened DEGs, which are referred to as DEGs2. To compare the differences between DEGs1 and DEGs2, a Venn diagram was used.[29]

The fusion of co-expression network

Based on DEGs1 and DEGs2, we acquired two corresponding co-expression networks by gene co-expression network analysis. Then, the DEGs1-based co-expression network was merged with the DEGs2-based co-expression network using the Cytoscape tool "Advanced network merge" to obtain a unique network.[30]

Functional enrichment analysis of DEGs in fusion co-expression network

Currently, there are multiple tools for gene function enrichment analysis, among which DAVID has been widely used.[31] Using DAVID software, the biological pathways significantly enriched by DEGs in the fusion co-expression network were identified. A P value less than 0.05 was used as a screening threshold.

Pathway analysis of DEGs in fusion co-expression network

Continuously, the pathway annotations and enrichment analysis were completed using KOBAS[32] based on algorithm of accumulative hypergeometric distribution. A P value less than 0.05 was used as a screening threshold.

Results

Data preprocessing and DEGs’ identification

To remove system errors under sequencing, the data were preprocessed. Through data preparation described in the methods section, we obtained the normalized gene expression data. After preprocessing, the medians of expression values in all the samples were relatively linear, suggesting that the expression data were well-normalized [Fig. 1]. A total of 451 DEGs were obtained from the comparison between the bacterial and normal groups and 353 DEGs from the comparison between the fungal and normal groups.

Figure 1

The boxplot of expression profiling at prestandardization (a) and poststandardization (b). Boxes with white, light gray, and dark gray represent normal cornea, bacterial-infected cornea, and fungus-infected cornea samples, respectively. Y-axis represents gene expression value

We observed that the screened DEGs could significantly distinguish bacteria/fungus-infected from normal corneal samples [Figs. 2a and b]. These results indicated that significant sample differences existed among screened DEGs between bacteria-infected and normal groups, as well as between fungus-infected and normal groups.

Figure 2

The heap map of screened DEGs. (a) Heap map of DEGs screened between bacterial-infected and normal groups. (b) Heap map of DEGs screened between fungus-infected and normal groups. Red color represents high expression, while blue color represents low expression. Color changes from blue to red indicate the corresponding expression value change from lower to higher

Difference between DEGs between bacterial versus normal and fungus versus normal groups

To compare the difference in DEGs in bacterial versus normal and that in fungal versus normal, a Venn diagram was constructed. We observed that the number of overlapped DEGs was 303, which accounted for 67.18% (303/451) among bacterial versus normal DEGs and 85.84% (303/353) among fungal versus normal DEGs, respectively [Fig. 3]. There were 148 DEGs specific to bacterial keratitis, such as CD34 (CD34 molecule, P = 1.40E-09, low expression), HK2 (hexokinase 2, P = 6.79E-07, overexpression), and TLR4 (toll-like receptor 4, P = 2.35E-09, overexpression) and 50 specific DEGs in fungal keratitis, such as ADH7 (alcohol dehydrogenase 7, P = 3.85E-04, low expression), ASGR1 (asialoglycoprotein receptor 1, P = 5.49E-08, overexpression), and SOD2 (superoxide dismutase 2, P = 7.89E-05, overexpression) [Table 2]. These results suggest that there were a large number of DEGs identified both in bacteria-infected and fungus-infected corneas, exhibiting tremendous similarities in the above two keratopathies. In addition, the same DEG showed homodromous up- or down-regulation in bacteria and fungus-infected cornea samples, also exhibiting complete uniformity in the above two keratopathies.

Figure 3

Venn diagram of DEG sets between bacterial vs. normal and fungus vs. normal groups

Table 2

DEGs specific in bacterial and fungal keratitis

Groups	Gene symbol	P	logFC	Gene symbol	P	logFC
Bacterial keratitis	CD34	1.40E-09	−3.550401	PFKFB3	3.02E-06	2.1137928
	OLFML1	5.96E-07	−2.892602	HK2	6.79E-07	2.1140937
	HTRA1	6.45E-07	−2.624773	SKAP2	1.81E-09	2.1184403
	RPPH 1	3.17E-05	−2.483578	ERO1L	2.21E-07	2.1187003
	CTSF	5.41E-10	−2.477922	HLA-B	2.53E-10	2.1210097
	IRX2	5.82E-09	−2.47353 1	CASP5	2.97E-07	2.1239787
	JAM3	3.29E-07	−2.459587	ACSL5	1.43E-07	2.1263413
	ADRB2	8.62E-06	−2.428554	AP1S2	2.54E-08	2.128140 1
	ISLR	1.57E-07	−2.407548	CMTM6	4.39E-08	2.1737369
	MFAP4	6.08E-06	−2.397586	ARRB2	2.28E-10	2.1842634
	CXCL14	3.46E-05	−2.357675	STX11	0.000133	2.2101693
	THNSL2	1.95E-06	−2.348504	FAM49B	2.40E-07	2.2180724
	ELF3	0.000778	−2.341577	ZEB2	1.98E-08	2.224024
	CLDN5	0.00025	−2.299214	KLRB1	2.96E-06	2.2241235
	PLA2G2A	0.000888	−2.29423 1	CYB5R4	5.12E-08	2.2409857
	IGFBP2	1.07E-08	−2.289433	RIPK2	2.26E-05	2.2449883
	CCDC3	1.50E-05	−2.26505 1	ADORA3	2.83E-08	2.245635
	FAM46B	8.10E-08	−2.262456	TLR4	2.35E-09	2.2484145
	GLT8D2	4.16E-05	−2.237313	HLA-F	7.22E-08	2.2579246
	SERPINF1	7.60E-06	−2.232497	RILPL2	8.36E-06	2.259585
	SCNN1A	1.16E-05	−2.225126	CCL2	0.000469	2.2641285
	RCAN2	1.85E-07	−2.222192	UPB1	8.46E-07	2.268115
	PDGFRL	0.000144	−2.217872	LMNB1	2.24E-07	2.2795219
	SERPINA5	3.09E-08	−2.211005	ARHGAP15	1.36E-10	2.2814229
	ZNF750	0.000718	−2.207514	PIK3CG	2.45E-10	2.2903533
	HBZ	4.51E-07	−2.206249	C5AR1	2.28E-09	2.2964389
	CPXM2	1.60E-05	−2.204688	GAPT	1.00E-07	2.2981819
	MT1X	1.76E-05	−2.203755	PLIN2	6.08E-06	2.3070316
	KAZALD1	1.18E-08	−2.203342	GPR65	1.30E-06	2.317418 1
	F10	1.30E-07	−2.195377	CXCL16	2.30E-09	2.3231413
	FBLN5	1.72E-05	−2.189242	MPP 1	1.35E-11	2.3255188
	PHGDH	2.23E-06	−2.180239	ANXA3	9.88E-09	2.3397564
	EMX2	2.76E-07	−2.178918	SIGLEC10	2.60E-06	2.3605042
	LAMB2	4.64E-07	−2.171439	OLR1	5.30E-05	2.3724875
	SOX15	8.35E-05	−2.16497 1	DRAM 1	2.76E-11	2.3752319
	SVEP1	5.96E-05	−2.148164	LY96	6.33E-08	2.3793451
	SNORD97	1.30E-05	−2.147336	IL18RAP	3.49E-06	2.4051193
	AHNAK	6.83E-08	−2.115137	HIF1A	4.07E-08	2.418505
	PDGFD	0.000363	−2.112166	HNRNPA3P1	1.83E-06	2.4269378
	TCEAL2	1.22E-07	−2.103969	GNG2	5.36E-08	2.4428582
	TMEM100	1.40E-08	−2.100933	SIGLEC5	2.75E-07	2.4445636
	SETBP1	3.40E-07	−2.099829	CYP27A 1	4.13E-07	2.4615939
	BMP4	1.51E-09	−2.090556	TMEM71	9.94E-10	2.4708218
	PRODH	3.54E-10	−2.089715	PTGER4	7.66E-08	2.4890422
	GPRC5C	2.02E-06	−2.085608	LAMC2	3.97E-05	2.5039927
	EPHX2	2.72E-09	−2.082969	IL10RA	1.32E-1 1	2.5234075
	WFDC1	1.40E-07	−2.074921	STEAP 1	1.33E-08	2.5253655
	MOXD1	0.00125	−2.065241	NRP 1	3.61E-09	2.5407649
	COL8A1	4.38E-06	−2.056859	EVI2A	1.35E-11	2.5469826
	CYP26A1	0.00309	−2.042533	BID	1.14E-07	2.5633001
	MGP	1.58E-06	−2.03842	CXCL2	0.00644	2.578593
	RIPK4	0.00049	−2.01784	CMTM2	1.02E-07	2.5786635
	TOB1	5.62E-08	−2.012904	EPB41L3	5.06E-08	2.603111
	COX7A1	1.54E-06	−2.009043	VNN3	4.45E-09	2.6083142
	LCP2	5.27E-06	2.0062235	GBP1	3.42E-08	2.6092806
	ZMYND15	1.93E-10	2.0095573	SNX10	7.88E-08	2.6197063
	MS4A4A	4.93E-08	2.016654 1	S100A12	1.05E-07	2.6254425
	SLC43A3	1.59E-06	2.0215348	CTSS	1.95E-11	2.6559225
	BATF	4.95E-08	2.029239	PTPRE	5.31E-11	2.6785904
	PDE4B	5.10E-06	2.0297734	KRT6C	0.0182	2.6803395
	IRAK2	0.000354	2.0404054	SLC16A10	1.07E-08	2.7276327
	GZMA	7.15E-08	2.0425125	MXD1	2.41E-08	2.7510621
	ANTXR2	2.61E-07	2.0522199	FYB	9.74E-12	2.778599
	TFPI2	3.09E-06	2.0532942	GCA	1.01E-10	2.7942486
	PRDM8	1.35E-06	2.0542583	NPL	3.18E-11	2.79455
	CXCL1	0.000476	2.0544949	PIK3AP1	3.44E-08	2.799228
	PLSCR1	8.98E-08	2.0548479	HMOX1	2.92E-06	2.8822109
	NFE2	2.39E-07	2.0548741	LILRB2	1.62E-10	2.8849687
	TGM3	0.000106	2.0656169	EMR3	7.77E-10	3.0604573
	BASP1	7.63E-08	2.088951	HLA-DRB1	0.00407	3.2639598
	PTGS2	0.0041	2.0901709	IL6	0.000243	3.4041672
	IL4R	1.74E-06	2.101003	IL1A	0.000232	3.5103275
	RP2	2.04E-08	2.1014075	HLA-DRB5	0.00321	3.6076907
	PLEKHO2	1.18E-08	2.106695	MS4A7	1.16E-10	3.7087315
Fungal keratitis	ADH7	0.000385	−2.41104	KLF4	3.21E-09	-2.086789
	AGR2	0.000124	−2.382602	LY86	4.45E-11	2.2194223
	AKR1C2	5.87E-09	−2.122723	MATN2	9.05E-05	-2.021565
	ASGR1	5.49E-08	1.9998781	MYO1G	6.08E-10	2.4582294
	BST2	1.92E-09	2.2396599	NEK6	2.09E-10	2.0478274
	C1QTNF1	0.000141	2.0340408	NQO1	3.37E-07	-2.670235
	CAMP	1.45E-05	2.3958655	PDXK	1.41E-11	2.05441
	CCL22	1.21E-06	2.077171	PLEKHO1	2.85E-10	2.2511074
	CD74	2.59E-09	2.4476072	PMEPA 1	1.26E-06	2.1131556
	COL1A1	4.88E-05	3.0325672	PTPRO	3.88E-10	2.0927626
	COL22A1	4.45E-07	2.211018	RARRES2	5.15E-06	2.0625649
	COL5A1	1.81E-07	2.6823189	RASAL3	7.37E-12	2.1098773
	CTSZ	1.16E-08	2.0937448	S100A7	9.22E-06	3.0074117
	CXCL10	3.08E-06	2.2824941	S100A7A	0.000117	2.0777797
	CXCL13	0.000252	2.1880856	SBSN	0.000988	2.4539646
	DSG1	0.000938	−2.007756	SLAMF9	1.17E-05	2.077392 1
	FCGR1A	2.39E-12	2.1689304	SOD2	7.89E-05	3.154099
	GJB6	2.81E-05	−2.088691	SPRR2D	0.0105	2.4678331
	GPR68	3.20E-07	2.2930453	SPRR2F	0.0135	2.2326478
	GPX2	5.67E-06	−2.625696	SPRR3	0.0131	2.2941334
	GZMB	2.67E-05	2.0633337	STEAP3	5.39E-07	2.0188332
	HBA2	0.000319	2.9052848	TM4SF19	1.57E-06	2.0371738
	HBB	0.0028	2.7257732	TMEM176A	4.23E-10	2.1360348
	ISG15	1.83E-07	2.5742654	TNC	2.55E-08	3.1398672
	ITGB7	8.13E-09	2.0142005	TRPV2	1.54E-09	2.1827889

DEGs: Differentially expressed genes

The number of co-expressed gene pairs was 117 pairs among bacterial versus normal DEGs and 87 pairs among fungal versus normal DEGs, respectively. The co-expression networks were visualized using Cytoscape tool to obtain the corresponding network graphs.

The fusion of co-expression networks

After fusion of bacterial versus normal and fungal versus normal coexpression networks, a novel fusion coexpression network was generated [Fig. 4]. This fusion co-expression network included 79 DEG nodes and 190 connecting edges. Among these 79 DEG nodes, 19 were unique to bacterial versus normal DEGs, 5 were unique to fungal versus normal DEGs, and 55 were present in both comparisons.

Figure 4

The fusion coexpression network merged from bacterial vs. normal and fungus vs. normal coexpression networks. Dark gray and light gray represent bacterial vs. normal DEGs and fungus vs. normal DEGs, respectively. Triangle and inverted triangle represent up- and down-regulation DEGs, respectively. White rhombus represents DEGs identified both in bacterial vs. normal DEGs and fungus vs. normal DEGs

Function enrichment analysis of DEGs in fusion co-expression networks

Through analysis of DEGs in fusion co-expression networks using DAVID, we searched nine biological pathways in total that were significantly differentially regulated [Supplemental Table 1]. Among these nine biological pathways, the immune response was the most significant pathway. Notably, the other eight biological pathways were mainly associated with the immune system.

Supplemental Table 1

Biological pathways searched based on DEGs in fusion co-expression network

Term	Count	P
GO: 0006955~immune response	15	2.58E-06
GO: 0002504~antigen processing and presentation of peptide or polysaccharide antigen via MHC class II	5	1.58E-05
GO: 0009611~response to wounding	11	1.54E-04
GO: 0019882~antigen processing and presentation	5	5.96E-04
GO: 0006952~defense response	10	0.002058
GO: 0006954~inflammatory response	7	0.003948
GO: 0034097~response to cytokine stimulus	4	0.005919
GO: 0050777~negative regulation of immune response	3	0.005977
GO: 0055114~oxidation reduction	9	0.009274

In total, DEGs in fusion co-expression network were involved seven KEGG pathways [Supplemental Table 2], among which antigen processing and presentation (hsa04612) was the most striking. Specifically, there were five DEGs identified both between the bacterial versus normal groups and the fungal and normal groups, including HLA-DRB3, IFI30, HLA-DPA1, HLA-DMB, and HLA-DMA, involved in antigen processing and presentation pathways. Among them, HLA-DRB3, HLA-DPA1, HLA-DMB, and HLA-DMA were also involved in the immune response.

Supplemental Table 2

KEGG pathways analyzed based on DEGs in fusion co-expression network

	Pathway	P	FDR
hsa04612	Antigen processing and presentation	0.001309	0.015919
hsa05332	Graft-versus-host disease	0.001468	0.014292
hsa04940	Type I diabetes mellitus	0.001822	0.014782
hsa04672	Intestinal immune network for IgA production	0.002844	0.019737
hsa05320	Autoimmune thyroid disease	0.003189	0.019373
hsa00010	Glycolysis/Gluconeogenesis	0.005056	0.027221
hsa05416	Viral myocarditis	0.008085	0.038996

FDR: false discovery rate; P: P

Discussion

In this study, we primarily screened the DEGs in bacterial keratitis and fungal keratitis through analyzing the gene expression profiles of corneal tissues. In total, there were 451 DEGs identified from bacterial keratitis versus normal corneal tissues and 353 DEGs identified from fungal keratitis versus normal corneal tissues. The number of overlapping DEGs between bacterial keratitis and fungal keratitis was 303, which accounted for a larger proportion in corresponding keratitis. In addition, through co-expression network analysis, 117 co-expressed gene pairs were identified in bacterial keratitis DEGs and 87 pairs in fungal keratitis DEGs. After constructing the fusional co-expression network based on bacterial and fungal keratitis co-expression DEGs, nine biological pathways by function enrichment analysis and seven KEGG pathways by KEGG analysis were identified as significant.

Toll-like receptor 4 (TLR4) is a crucial pattern recognition molecule that participates in the innate immune response to lipopolysaccharide, a vital component of Gram-negative bacteria.[33] It is reported that TLR4 mRNA levels were significantly upregulated in bacterial (P. aeruginosa) infected mouse cornea tissue.[34] In accordance with this study, our data confirmed that TLR4 levels were significantly increased by approximately 5-fold in human cornea tissues with bacterial keratitis compared to those in cornea tissues from healthy donors.[34] The deficiency of TLR4 in mouse could result in increased polymorphonuclear neutrophil infiltration and proinflammatory cytokine production, as well as decreased β-defensin-2 and inducible nitric oxide synthase production in mouse with P. aeruginosa infection of the cornea.[34] Yan et al. reported that TLR4 found in corneal macrophages could regulate P. aeruginosa keratitis by signaling through myeloid differentiation factor 88 (MyD88)-dependent and -independent pathways.[35] In addition, TLR4 was also reported to regulate fungal keratitis such as fusarium keratitis[36] and A. fumigatus keratitis,[37] although from our analysis, there were no significant differences in TLR4 expression between corneal tissues infected with a fungal pathogen versus normal tissues. This suggests that TLR4 is a candidate target gene to distinguish bacterial keratitis from fungal keratitis. A promising strategy for the diagnosis of infectious keratitis may be developed based on TLR4 expression.

Among the 50 non-overlapping DEGs in fungal keratitis, SOD2 levels were found to be significantly increased by about 9-fold in human corneal tissues with fungal keratitis compared to those in normal human corneal tissues. It has been previously reported that SOD2 expression is significantly increased by about 2-fold in mouse corneas with fungal (Candida albicans) keratitis compared to that in healthy mouse corneas.[38] We speculated that the differences in SOD2 fold-change between mouse and human could be due to species-specific differences and fungal species. Moreover, SOD2 was a pivotal DEG node in the fusional co-expression network, which was derived from fungus keratitis DEGs, and co-expression is associated with MYOC, KRT6B, and CSF1R. Meanwhile, SOD2 was identified to participate in responses to wounding and oxidation reduction pathways. Although there is currently no report describing a role for SOD2 in keratitis, SOD2 still remains a potential candidate gene to distinguish between bacterial and fungal keratitis due to its specific association with fungal keratitis.

Furthermore, through functional analysis of DEGs in fusion co-expression networks, we identified nine biological pathways such as pro-inflammatory and anti-inflammatory responses, antigen processing and presentation,[39] and wounding responses.[40] The majority of the identified pathways are associated with the immune system. Through KEGG analysis of DEGs in fusion co-expression network, we identified seven KEGG pathways, of which antigen processing and presentation[39] intestinal immune network for IgA production, autoimmune thyroid disease, and viral myocarditis[41] are more associated with the immune system. This suggests that perturbations in the immune system induced by pathogen exposure in the cornea leads to the malignant advance of infectious keratitis. Based on our findings, we speculate that strategies aimed at controlling inflammation are a compensatory therapy to alleviate the pain experienced by patients with keratitis excepting to eliminate pathogens, which requires further investigation of the identified immune-related DEGs in infectious keratitis.

In this study we identified novel DEGs associated with bacterial or fungal keratitis; however, our study does have limitations. Firstly, we were limited in the clinical materials including verification of our samples. Secondly, due to the limitations of obtaining human tissue samples, the sample size remains small. Future studies will be required to corroborate our findings using larger sample sizes. Finally, the expression levels of the candidate genes may be affected by the nature of the pathogen, the stage of the disease, or the genetic background of the host. Thus, studies with large sample sizes are warranted to validate our findings in the near future.

Conclusion

In summary, our work screened 451 DEGs in corneas with bacterial keratitis and 353 DEGs in corneas with fungal keratitis, in which 148 DEGs were found to be specific to bacterial keratitis and 50 DEGs specific in fungal keratitis. TLR4 was an upregulated gene specific in bacterial keratitis and SOD2 was an upregulated gene specific to fungal keratitis. Both genes are promising candidate targets to distinguish bacterial and fungal keratitis.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.