Transcriptomic Landscape of Quiescent and Proliferating Human Corneal Stromal Fibroblasts

Abstract

This study analyzed the transcriptional changes in primary human corneal stromal fibroblasts (hCSFs) grown under quiescent (serum-free) and proliferating (serum-supplemented) culture conditions to identify genes, pathways, and protein–protein interaction networks influencing corneal repair and regeneration. Primary hCSFs were isolated from donor human corneas and maintained in serum-free or serum-laden conditions. RNA was extracted from confluent cultures using Qiagen kit and subjected to RNA sequencing (RNAseq) analysis. Differential gene expression (DGE) and pathway enrichment analyses were conducted using DESeq2 and Gene Set Enrichment Analysis (GSEA), respectively. Protein–protein interaction (PPI) networks were created exploiting the STRING database and analyzed with Cytoscape and the cytoHubba plugin. RNA-seq revealed 5,181 genes that were significantly differentially expressed/changed among the 18,812 annotated genes (p value <0.05). A cutoff value of a log2-fold change of ±1.5 or greater was used to identify 674 significantly upregulated and 771 downregulated genes between quiescent and proliferating hCSFs. Pathway enrichment analysis revealed significant changes in genes linked to cell cycle regulation, inflammatory, and oxidative stress response pathways, such as E2F Targets, G2M Checkpoint, and MYC Targets, TNFA signaling via NF-kB, and oxidative phosphorylation. Protein-protein interaction network analysis highlighted critical hub genes. The FGF22, CD34, ASPN, DPT, LUM, FGF10, PDGFRB, ECM2, DCN, VEGFD, OMD, OGN, ANGPT1, CDH5, and PRELP were upregulated, whereas genes linked to cell cycle regulation and mitotic progression, such as BUB1, TTK, KIF23, KIF11, BUB1B, DLGAP5, NUSAP1, CCNA2, CCNB1, BIRC5, CDK1, KIF20A, AURKB, KIF2C, and CDCA8, were downregulated. The RNA sequences and gene count files have been submitted to the Gene Expression Omnibus (accession # GSE260476). Our study provides a comprehensive information on the transcriptional and molecular changes in hCSFs under quiescent and proliferative conditions and highlights key pathways and hub genes.

1. Introduction

The cornea, an essential element of the ocular system, is instrumental in ensuring visual acuity because of its refractive capabilities and transparent nature. The transparency of the cornea is due to uniform and regular arrangement of collagen fibers, avascularity, low cellular and nuclear density in the stromal layer (primarily contains keratocytes), and the relative dehydration maintained by the corneal endothelium, each of which contributes to its overall function and structural integrity (Kumar et al., 2023a). Central among these factors are corneal stromal fibroblasts/keratocytes, which play a central role in sustaining the cornea’s physiological and reparative functions. These cells are primarily responsible for the synthesis, deposition, and modification of the extracellular matrix (ECM) and maintaining the distinct characteristics and functions of the cornea (Mohan et al., 2024). The regulatory pathways that influence corneal stromal fibroblast activity and gene expression profiles are of significant interest, as they provide potential insights into the development of innovative strategies and treatment modalities for various corneal pathologies and traumas (Wilson, 2022, Kumar et al., 2024). A comprehensive understanding of the genes and pathways contributing to corneal repair and regeneration mechanisms is imperative for sustaining corneal health and advancing functional recovery.

Environmental conditions for cellular division have been created via the use of fetal bovine serum (FBS) in fibroblast culture medium for culturing corneal stromal cells (Pilgrim et al., 2022). Corneal fibroblasts in serum-free media are quiescent (Q-hCSF), and those in serum-laden media are proliferative (P-hCSF). This is due to the nutrient-rich composition of critical biomolecules, including growth factors, hormones, and nutrients, in serum, which are vital for the sustenance and robustness of cultured cells (Palm and Thompson, 2017). However, the introduction of such nutrients into the culture environment can significantly impact the transcriptional landscape of cells (Liu et al., 2017). This impact is not uniform but varies significantly across different cell types and is also influenced by the serum concentration. Specifically, within the realm of corneal stromal fibroblasts, the addition of FBS has the potential to modulate cellular activities that are crucial for the healing and regeneration of the cornea (Kamil and Mohan, 2021). However, the detailed influence of serum on the modulation of gene expression patterns in corneal stromal fibroblasts, particularly in ways that could benefit corneal repair mechanisms, has not been thoroughly elucidated. This gap in knowledge limits the understanding of the behavior and specific transcriptional responses of hCSFs under different conditions and their effects on corneal wound healing.

Recent advances in RNA sequencing technologies have enabled detailed examination of the transcriptomic responses of cells to various treatments (Kumar et al., 2023b). Comprehensive information about specific pathways and genetic regulatory networks is vital for understanding the factors influencing cell behavior under various conditions. Furthermore, this technology holds promise for defining the gene expression landscape and identifying molecular targets that could be leveraged to develop newer therapeutic interventions for corneal repair and regeneration. Using RNA-Seq technology, this study elucidated the gene expression dynamics within Q-hCSFs versus P-hCSFs and performed differential transcriptomics to identify key genes and pathways relevant to corneal wound healing.

2. Material and Methods

2.1. Primary Culture of Human Corneal Stromal Cells

Primary hCSF cultures from donor human corneas were generated in accordance with the ethical standards set by the Declaration of Helsinki for human tissue research, the Association for Research in Vision and Ophthalmology, and the University of Missouri’s Institutional Review Board (IRB). The corneas were sourced from four donors, consisting of three females and one male, aged 49–76 years. The samples included both left and right corneas, with two from the right eye and two from the left. The use of cadaver human corneas does not constitute human subject research according to the Department of Health and Human Services regulatory definitions and is exempt from the IRB requirements. Healthy corneal tissues purchased from Saving Sight, Kansas City, Missouri, USA, were used to establish primary Q-hCSF and P-hCSF cultures following our reported method (Sinha et al, 2023). The corneas were washed with serum-free minimum essential medium (MEM), and the endothelial and epithelial layers were carefully removed using a scalpel surgical blade (#64). Afterward, the tissues were divided into 8–10 smaller pieces, placed into sterile culture dishes, and incubated at 37°C in a 5% CO2 atmosphere for 3–5 weeks; the culture medium consisted of MEM supplemented with essential amino acids, sodium pyruvate, and vitamins (Sinha et al, 2023).

2.2. Quiescent and Proliferating Human Corneal Stromal Fibroblasts

Quiescent hCSFs (passage 2) cultured to 50% confluence were subjected to serum starvation in serum-free MEM for 72 hours. Another set of passage 2 cultures (proliferating hCSFs) was cultured with MEM supplemented with 10% fetal bovine serum (FBS; CAT no. 16140–071; Thermo Fisher, Grand Island, NY, USA) for 72 hours. Both the Q-hCSFs and P-hCSFs were grown in quadruplets. The medium, with or without serum, was renewed every 24 hours until the end of the experiment. The cells cultured in serum-free media presented a Q-hCSF phenotype, in contrast to the cultures cultured in serum-enriched media, which presented P-hCSF characteristics. These changes were monitored via a phase-contrast microscope equipped with a camera.

2.3. RNA Extraction and Sequencing

RNA was extracted from cultures of well-characterized Q-hCSF and P-hCSF derived from donor human corneas, with each experiment replicated four times. The cells were washed with 1X PBS and scraped in lysis buffer, and RNA was isolated via the RNeasy Mini Kit (QIAGEN, Germantown, MD, USA) according to the manufacturer’s protocol. The desired amount of RNA was reverse transcribed into cDNA via a commercial kit (Promega Biotechnology Co., Madison, WI) following the vendor’s instructions. The integrity and concentration of the RNA samples were evaluated with a NanoDrop ND-1000 (Thermo Fisher Scientific, Wilmington, DE, USA). Furthermore, RNA samples were sequenced at Arraystar, Inc. Rockville, MD, USA. The RNA sequencing libraries were prepared, qualified, and sequenced on an Illumina NovaSeq 6000.

2.4. Differential gene expression

The quadruplet samples from both cell types, Q-hCSF and P-hCSF, were subjected to differential gene expression (DGE) analysis to identify genes exhibiting changes in expression patterns. We employed an established DGE analysis protocol from our previously published study using Python and R packages and bioinformatics tools (Kumar et al., 2023b).

2.4.1. Quality Check and Trimming

FastQC (version 0.12.1) was used to evaluate the quality of the unprocessed sequencing data (Andrews, 2010). This tool provides a range of analytics to quickly determine sequence quality based on summary graphs and tables that include basic statistics, per-base sequence quality, sequence content, per-base N content, GC content, sequence length distribution, etc. Cutadapt (version 4.6) was utilized to filter out adapter sequences, primers, poly-A tails, and other unwanted sequences from high-throughput sequencing reads (Martin, 2011). We assessed the quality of all eight RNA sequences from Q-hCSF (4 samples) and P-hCSF (4 samples), ensuring high-quality reads.

2.4.2. Read Alignment

HISAT2 (version 2.2.1) was employed for aligning trimmed RNA sequences from both the Q-hCSF and P-hCSF cell types to the human reference genome GRCh37 (Kim et al., 2017). HISAT2 utilizes a novel genome indexing scheme that employs a graph-based approach to capture a wide representation of genetic variant reads. Unlike other graph aligners (vg and bpa aligners) that use memory-intensive k-mer-based indices, HISAT2 implements the graph FM (GFM) index (Garrison et al., 2018; Rakocevic et al., 2019). HISAT2 starts by constructing a linear graph of the reference genome and subsequently introduces insertions, deletions, and mutations as alternate pathways across the graph, thus covering a wide range of genetic variants. This makes HISAT2 an efficient and practical tool for aligning raw sequencing reads to a graph that represents the entire human genome along with a large number of variants.

2.4.3. Transcript Assembly

StringTie (version 2.2.1) was used for assembling mapped reads into transcripts via a network flow algorithm and optional de novo assembly (Pertea et al., 2015; Kovaka et al., 2019). StringTie creates more comprehensive and accurate gene reconstructions and precise estimations of expression levels than other prominent transcript assembly tools, such as Cufflinks (Trapnell et al., 2010), IsoLasso (Li et al., 2011), Scripture (Guttman et al., 2010), and Traph (Tomescu et al., 2013). Unlike Cufflinks, which finds a minimal set of transcripts via a parsimony-based algorithm and then estimates their expression levels separately, StringTie assembles transcripts and estimates their expression levels concurrently. Genome-guided transcript assemblers cluster the reads and generate graph models representing all potential isoforms for each gene using reads mapped to the reference genome. StringTie repeatedly selects the heaviest route from a splice graph, builds a flow network, calculates the maximum flow to estimate abundance, and modifies the splice graph by eliminating reads assigned by the flow method. This process is repeated until all the reads are allocated. StringTie Merge (version 2.2.3) was used to construct a nonrepetitive set of transcripts identified in all previously assembled RNA-Seq datasets. It was run with all assembled transcript files (in GTF/GFF format) obtained for each sample, as well as a reference annotation file (gencode.v32.annotation.gtf), resulting in a global and unified set of transcripts from various RNA-Seq samples in a single file.

2.4.4. Differentially Expressed Features

The gene count files obtained using StringTie for each sequence profile of the Q-hCSF and P-hCSF cell types were used as inputs for DESeq2 (version 1.34.0) to identify differentially expressed features (Love et al., 2014). Additionally, we generated normalized gene counts for further analysis using gene set enrichment analysis (GSEA) (version 4.3.2) (Subramanian et al., 2005). Annotate DESeq2/DEXSeq was employed to add annotation information from the GTF file to the differentially expressed genes (DEGs). This utility enhances the DESeq2/edgeR/limma/DEXSeq output table by including gene symbols, biotypes, and gene locations. The inclusion of data can be adjusted, and this information should be included in the input GTF/GFF file as attributes of the selected feature.

2.5. Pathway Enrichment Analysis

For a focused analysis of key pathways and core genes, GSEA was employed to discern statistically significant differences in gene sets between Q-hCSF and P-hCSF. Pathways enriched in P-hCSFs relative to those enriched in Q-hCSFs were cataloged based on their normalized enrichment scores, with a stringent significance level of p < 0.01 earmarked for deeper exploration. This approach facilitated the identification of pivotal pathways and core genes implicated in the differentiation process, which were then subjected to further examination.

2.6. Protein-Protein Interaction (PPI) Network Analysis

The exploration of molecular interactions and PPI networks was conducted using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) (version 11.0) database (Szklarczyk et al., 2019). This resource provides insights into both direct and indirect protein associations and their functional implications. The STRING database aided in constructing PPI networks from the most enriched pathways identified by GSEA, highlighting critical interactions and functional correlations.

Hub genes within these networks were identified through network topology analysis performed in Cytoscape (version 3.8.2) utilizing the cytoHubba plugin (version 0.1) (Bader and Hogue, 2003; Shannon et al., 2003; Chin et al., 2014). This analysis focused on pinpointing significant nodes or probable targets within complex networks, with a particular emphasis on genes pivotal to corneal wound healing processes. The adopted methodology is depicted in Fig. 1.

Fig. 1.

Experimental design and workflow used to conduct transcriptomic and RNA-seq analyses of quiescent and proliferating human corneal stromal fibroblasts treated with or without 10% fetal bovine serum.

3. Results

3.1. RNA yield and Quality

The RNA yield and quality of the proliferating and quiescent hCSFs (four replicates each) are shown in Fig. S1. The RNA yield (normalized by cell count) is shown on the y-axis, and the blue and green bars represent the RNA yield for each sample. Higher bars indicate a greater yield of RNA. The A260/280 ratio for each sample (measure of RNA purity) was between 1.8 and 2.0, indicating high purity. The x-axis lists the sample IDs, where blue and green bars represent proliferating (A1--A4) and quiescent (C1--C4) hCSFs, respectively.

3.2. RNA sequence quality

The RNA sequences from all used P-hCSFs (P-hCSF-A1, P-hCSF-A2, P-hCSF-A3, P-hCSF-A4) and Q-hCSF (Q-hCSF-C1, Q-hCSF-C2, Q-hCSF-C3, Q-hCSF-C4) samples exhibited excellent consistency. The base call accuracy reached 99.9% based on the Phred quality score (Q). Additionally, over 92% of the bases in all proliferating and quiescent hCSF samples had a Q>30 threshold (Table 1). This high level of data quality was essential for ensuring the accuracy and reliability of subsequent gene expression profiling and robust bioinformatics analyses.

Table 1.

Quality metrics of RNA sequencing for proliferating and quiescent human corneal stromal fibroblasts.

Sample Name	Reads Count	Bases No.	Bases No. (Q≥30)	Q30 (%)
P-hCSF-A1	42668054	6400208100	5994203813	93.66
P-hCSF-A2	39871536	5980730400	5588214488	93.44
P-hCSF-A3	36289104	5443365600	5092795147	93.56
P-hCSF-A4	40134640	6020196000	5637235644	93.64
Q-hCSF-C1	33265008	4989751200	4636837418	92.93
Q-hCSF-C2	36729648	5509447200	5091129693	92.41
Q-hCSF-C3	30778970	4616845500	4298313441	93.10
Q-hCSF-C4	31019010	4652851500	4309196217	92.61

3.3. Differential Gene Expression Analysis

HISAT2 was used to determine the RNA sequence alignment with the human reference genome (GRCh37). Table 2 shows the sequence read alignment and mapping analyses for the quadruplet sequences of proliferating- and quiescent-hCSFs. The ‘Raw Pairs’ column gave an initial count of the read pairs, and the ‘Trimmed’ column showed the read pairs after the removal of 5’ and 3’ adaptors and read shorter than 20 base pairs. Specifics of the RNA content within the trimmed sequences revealed percentage of the mitochondrial RNAs (mtRNAs) and ribosomal RNAs (rRNAs) for each sample. The proportion of reads aligning to the reference genome in trimmed pairs is presented in the ‘Mapped’ column, which included mRNA, pre-mRNA, poly-A tailed lncRNA, and pri-miRNA.

Table 2.

Mapping metrics for proliferating (P-hCSF-A1 to P-hCSF-A4) and quiescent (Q-hCSF-C1 to Q-hCSF-C4) human corneal stromal fibroblasts.

Sample Name	Raw Pairs	Trimmed	mtRNAs (%)	rRNAs (%)	Mapped (%)	Unmapped (%)
P-hCSF-A1	21334027	21333984	1.50%	0.32%	90.72%	9.28%
P-hCSF-A2	19935768	19935127	1.37%	0.34%	90.72%	9.28%
P-hCSF-A3	18144552	18143966	1.99%	0.45%	91.61%	8.39%
P-hCSF-A4	20067320	20067318	1.66%	0.36%	91.17%	8.83%
Q-hCSF-C1	16632504	16632498	3.21%	0.71%	89.71%	10.29%
Q-hCSF-C2	18364824	18364819	1.20%	0.37%	85.83%	14.17%
Q-hCSF-C3	15389485	15237684	2.82%	0.62%	92.08%	7.92%
Q-hCSF-C4	15509505	15205190	1.94%	0.51%	90.51%	9.49%

Raw pairs= Number of raw read pairs obtained from sequencing; Trimmed= Number of read pairs remaining after trimming for quality and adapter sequences; mtRNAs (%)= Percentage of mitochondrial RNA reads; rRNAs (%)= Percentage of ribosomal RNA reads; Mapped (%)= Percentage of reads that successfully mapped to the reference genome; Unmapped (%)= Percentage of reads that did not map to the reference genome.

To determine gene count annotations, StringTie was used to assemble and quantify the transcripts of 8 BAM files of the P-hCSFs (n=4) and Q-hCSFs (n=4). The gene count files (18,812 annotations) were subjected to DESeq2 to obtain DEGs in the Q-hCSFs versus P-hCSFs. Fig. 2 shows a funnel chart, which effectively narrows the gene pool from total annotated genes to those that are differentially expressed, highlighting the number of significant, upregulated and downregulated genes. Total annotated (18,812), significant (5181), upregulated (674), and downregulated (717) genes were found (Table S1). A p value <0.05 was used to screen the significantly expressed genes. A total of 1391 genes were identified based on cutoff values of log2(FC) ≥ 1.5 and log2(FC) ≤ −1.5 for upregulated (674) and downregulated (717) genes, respectively.

Fig. 2.

Funnel chart illustrating the results of the differential gene expression analysis. The funnel chart depicts all annotated genes (18,812), significantly changed genes (P-adj < 0.05) (5,181), and significantly altered genes with a log2(FC) cutoff of ±1.5 (1,391 genes; 674 upregulated genes with a log2(FC) ≥ 1.5 and 717 downregulated genes with a log2(FC) ≤ −1.5).

Fig. 3 shows a volcano plot of the DEGs between the Q-hCSF and P-hCSF samples. Significantly (p ≤ 0.05) upregulated and downregulated genes are shown in red, significant genes with log2 (FC) < 1.5 are shown in blue, nonsignificant genes with log2 (FC) > 1.5 are shown in green, and nonsignificant genes with log2 (FC)< 1.5 are shown as gray dots. This demonstrated the differences in gene expression between the proliferating and quiescent hCSF cell types. Genes such as LAMA2 and GOLGA8B (upregulated) and genes such as ODC1 and ANGPTL4 CTSK (downregulated) are shown in the regions of the volcano plot. In addition, the long noncoding RNA AC011511.4 was shown to be significantly downregulated.

Fig. 3.

Volcano plot of differential gene expression in quiescent and proliferating human corneal stromal fibroblasts. The x-axis represents the log2-fold change in gene expression, and the y-axis represents the −log10 of the p value. Each point represents a gene, color-coded based on significance and magnitude of change: red dots for significant genes with log2 (FC) > 1.5, blue dots for significant genes with log2 (FC) < 1.5, green dots for nonsignificant genes with log2 (FC) > 1.5, and gray dots for nonsignificant genes with log2 (FC)< 1.5. The dashed lines indicate thresholds for statistical significance [p value < 0.05] and log2 (FC) = 1.5]. Notable differentially expressed genes are labeled.

To visualize the differential gene expression of the differentially expressed genes in Q-hCSFs and P-hCSFs, a heatmap was generated and presented in Fig. 4. In the heatmap, rows represent genes, while columns represent the cell types (Q-hCSFs and P-hCSFs). A gradient blue–yellow–red color scale (blue for the highest expression and red for the lowest expression) was used to demonstrate the differential expression of the analyzed genes. The upregulated genes in Q-hCSF are shown in blue compared with those in P-hCSF, whereas the downregulated genes are shown in the reverse pattern. Clustering in a heatmap specifies a group of genes with similar levels of expression as well as their coregulation and/or functional relationships. The most enriched pathways are color-coded: E2F targets (light blue), the G2M checkpoint (dark blue), MYC targets V1 (red), MYC targets V2 (orange), TNFA signaling via NFKB (yellow), oxidative phosphorylation (green), the UV response (purple), the inflammatory response (brown), the reactive oxygen response (dark red), IL6/JAK/STAT3 signaling (pink), DNA repair (light green), the unfolded protein response (cyan), the estrogen response late (gray), glycolysis (black), the EMT pathway (violet), spermatogenesis (light gray), TGFβ signaling (turquoise), MTORC1 signaling (olive), IL2/STAT5 signaling (beige), and Mitotic spindle (magenta). Additionally, columnwise clustering of the Q-hCSF and P-hCSF sequences revealed high intrasquence similarity and intersequence differences as well as consistency in differential gene expression. The top five up- and downregulated genes are given in Table 3.

Fig. 4.

Heatmap of gene expression and enriched pathways in quiescent and proliferating human corneal stromal fibroblasts. This heatmap displays the differential expression of genes between proliferating (P-hCSF-A1 to P-hCSF-A4) and quiescent (Q-hCSF-C1 to Q-hCSF-C4) human corneal stromal fibroblasts. The rows represent individual genes, while the columns represent different samples. Gene expression levels are color-coded, with blue indicating high expression and red indicating low expression, as shown by the color scale on the right. The left side of the heatmap displays the most enriched pathways identified through gene set enrichment analysis. Hierarchical clustering dendrograms above and to the left highlight distinct expression patterns and biological differences between Q-hCSF and P-hCSF.

Table 3.

Top five upregulated and downregulated differentially expressed genes.

DGE	Gene ID	Name	log2(FC)	P-adj
Upregulated	LAMA2	Laminin Subunit α 2	26.78	0.00E+00
	ISY1-RAB43	ISY1 Splicing Factor Homolog - RAB43, Member RAS Oncogene Family	10.91	1.01E-03
	PLCB4	Phospholipase C β 4	10.78	4.89E-05
	RIN2	Ras and Rab Interactor 2	9.85	1.82E-04
	GOLGA8B	Golgin A8 Family Member B	8.89	0.00E+00
Downregulated	AC011511.4	Noncoding RNA gene/unique sequence in the genome.	−23.68	1.00E-10
	RPS10-NUDT3	Ribosomal Protein S10 and Nudix Hydrolase 3	−12.08	1.46E-04
	SHISAL1/KIAA1644	Shisa like 1	−11.25	2.80E-04
	ANGPTL4	Angiopoietin Like 4	−10.49	0.00E+00
	LPIN2	Lipin 2	−9.27	4.54E-04

3.4. Pathway Enrichment Analysis

Pathway enrichment analysis was performed using gene set enrichment analysis to identify the top pathways that were significantly enriched in the Q-hCSF and P-hCSF samples. The pathways were ranked based on their normalized enrichment score (NES), highlighting those with the highest degree of gene overrepresentation at the top or bottom of the ranked gene list (Fig. 5A). Among the top enriched pathways were E2F targets, the G2M checkpoint, MYC targets V1 and V2, TNFA signaling via NFKB, and oxidative phosphorylation. To further elucidate the expression profiles of genes within these significant pathways, heatmaps were generated (Fig. 5B–G), illustrating the differential expression of pathway-specific genes between the Q-hCSF and P-hCSF samples. These heatmaps provide a visual representation of the distinct gene expression patterns and highlight the biological differences between the two sample groups. The five most significantly enriched pathways with a normalized enrichment score (NES) ≥ 2 and a false discovery rate (FDR) q value ≤0.05 were selected for further analysis. The identified pathways (green color) were E2F targets (NES= 2.703; FDR q value= 0.000), the G2M checkpoint (NES= 2.492; FDR q value= 0.000), MYC targets V2 (NES= 2.348; FDR q value= 0.000), MYC targets V1 (NES= 2.280; FDR q value= 0.000), TNFA signaling via NFKB (NES= 2.170; FDR q value= 0.000), and Oxidative phosphorylation (NES= 2.132; FDR q-value= 0.000). Other pathways depicted in blue color (with normalized NES <2) and listed in decreasing order of normalized NES were UV response up, Inflammatory response, Reactive oxygen response, IL3JAKSTAT3 signaling, DNA repair, Unfolded protein response, Estrogen response late, Glycolysis, EMT pathway, Spermatogenesis, TGFβ signaling, MTORC1 signaling, ILSTAT5 signaling, and Mitotic spindle. The enrichment plots for these pathways (Fig. 6) demonstrate the positive correlation and overrepresentation of pathway-specific genes in Q-hCSF, suggesting their potential involvement in the distinct biological processes and regulatory mechanisms that differentiate Q-hCSF from P-hCSF.

Fig. 5.

Pathway enrichment analysis and heatmaps of enriched genes in the Q-hCSF and P-hCSF samples. (A) The bar chart shows the results of the pathway enrichment analysis performed via gene set enrichment analysis (GSEA), which ranks pathways based on their normalized enrichment score (NES), which indicates the degree to which genes in each pathway are overrepresented at the top or bottom of the ranked list of genes. The highest ranked pathways included E2F targets, the G2M checkpoint, MYC targets V1 and V2, TNFA signaling via NFKB, and Oxidative phosphorylation. (B-G) Heatmaps depict the expression profiles of genes specifically enriched in significant pathways: (B) E2F targets, (C) G2M checkpoint, (D) MYC targets V2, (E) MYC targets V1, (F) TNFA signaling via NFKB, and (G) Oxidative phosphorylation. The columns represent individual samples of Q-hCSF and P-hCSF, and the rows represent genes, with color scales indicating expression levels from high (blue) to low (red).

Fig. 6.

Gene set enrichment analysis of key pathways in the Q-hCSF and P-hCSF samples. The enrichment plots illustrate the results of GSEA for six hallmark pathways that are significantly enriched in the comparison between the Q-hCSF and P-hCSF samples. Each panel displays the enrichment score curve, where the peak of the green line represents the enrichment score for the gene set. The plots also show the location of genes in the pathway within the ranked list of genes, with red indicating positive correlation (enrichment in Q-hCSF) and blue indicating negative correlation (enrichment in P-hCSF). The normalized enrichment score (NES) and the false discovery rate (FDR) q value are annotated in each plot.

To further elucidate the functional roles of differentially expressed genes (DEGs) in Q-hCSF compared to P-hCSF, we conducted Gene Ontology (GO) analysis to identify enriched biological processes and comodulated gene networks. GO analysis revealed significant enrichment of processes related to mitotic regulation, nuclear division, and chromosome segregation (Fig. 7A). Hierarchical clustering of these biological processes highlighted distinct categories, such as spindle checkpoint signaling and mitotic nuclear division, underscoring their critical roles in cell cycle regulation (Fig.7B). Network analysis identified key comodulated genes, including SMARCA2, which play pivotal roles in mitotic regulation and exhibit strong connectivity within the gene network (Fig. 7C). Additionally, the interaction network plot demonstrated robust interactions among the DEGs, emphasizing the coordinated regulation of these biological processes (Fig. 7D). These findings provide insights into the biological mechanisms underlying the differential expression observed in Q-hCSF.

Fig. 7.

Gene ontology analysis of DEGs and identification of biological processes and comodulated genes in Q-hCSF compared to P-hCSF. (A) Dot plot showing the enriched biological processes identified through Gene Ontology analysis of DEGs in the Q-hCSF group compared with the P-hCSF group. The x-axis represents the gene ratio, with the dot size indicating the count of genes involved in each process and the color representing the adjusted P value. (B) Cluster dendrogram displaying the hierarchical clustering of the enriched biological processes, with branches colored according to specific categories such as spindle checkpoint signaling and mitotic nuclear division. (C) Network diagram illustrating the comodulated genes and their associated biological processes, highlighting key genes. The node size represents the degree of connectivity, and the color indicates the log2-fold change. (D) Circle plot showing the interaction network of DEGs, with node size indicating the number of genes and edge thickness representing the strength of interactions among biological processes.

3.5. Protein-Protein Interaction (PPI) Network Analysis

The PPIs network among upregulated genes, when Q-hCSFs were compared with P-hCSFs is shown in Fig. 8. The network was generated using Cytoscape and analyzed with the cytoHubba plugin. Each node represents a protein encoded by an upregulated gene. The color intensity of the nodes indicates their influence within the network, with darker red representing the most influential nodes as determined by the maximal clique centrality (MCC) method. The edges represent the interactions between the proteins. The high number of interactions suggests a complex and interconnected network that indicates significant functional relationships between the upregulated proteins. The top fifteen most influential nodes, highlighted in darker red, include key proteins such as Fibroblast growth factor 22 (FGF22), CD34, Asporin (ASPN), Dermatopontin (DPT), Lumican (LUM), Fibroblast growth factor 10 (FGF10), Platelet-derived growth factor receptorβ (PDGFRB), Extracellular matrix protein 2 (ECM2), Decorin (DCN), Vascular endothelial growth factor D (VEGFD), Osteomodulin (OMD), Osteoglycin (OGN), Angiopoietin 1 (ANGPT1), Cadherin 5/VE-Cadherin (CDH5), and Prolargin (PRELP). These proteins play crucial roles in various cellular processes, including extracellular matrix organization and structural integrity (ASPN, DPT, LUM, DCN, ECM2, OMD, OGN, CD34, and PRELP), angiogenesis and vascular development (VEGFD, ANGPT1, and CDH5), and growth factors and cell signaling (FGF22, FGF10, and PDGFRB).

Fig. 8.

Protein–protein interaction network of upregulated genes in quiescent vs. proliferating primary human corneal stromal fibroblasts.

Table 4 shows the top 15 upregulated proteins in Q-hCSFs based on node rank. The log2 (fold change) values of the corresponding genes derived from DGE analysis are also shown. Each protein’s role in the cornea is described, highlighting its contributions to tissue development, repair, extracellular matrix organization, cell proliferation, migration, survival, and angiogenesis.

Table 4.

Roles and significance of the top fifteen upregulated proteins in quiescent human corneal stromal fibroblasts derived from protein–protein interaction network analysis.

Protein	PPI-network node rank	log2(FC) of corresponding gene	Role in cornea
Fibroblast Growth Factor 22 (FGF22)	1	4.110	Involved in tissue development and repair in nonocular tissues (Beyer et al, 2003). However, its role in corneal is not known.
CD34 Molecule	2	5.185	CD34 marks the cells’ potential for differentiation and repair, indicating their capacity to respond to damage. The quiescent-hCSF near the wound undergo a change in their phenotype to become more fibroblastic due to corneal trauma. This change is accompanied by a rapid disappearance of CD34 expression, indicating that the cells are becoming activated (Espana et al., 2004). However, the role of CD34 expression in quiescent-hCSF has not been fully understood, but it is believed that CD34 may be involved in regulating adhesion, differentiation, and quiescence.
Asporin (ASPN)	3	14.223	ASPN leucine-rich repeat protein family closely related to decorin and biglycan (Lorenzo et al., 2001). In nonocular tissues, ASPN inhibits TGF-β/Smad signaling upstream of TGF-β type I receptor activation in vivo by colocalizing with TGF-β1 on the cell surface and blocking its interaction with the TGF-β type II receptor (Nakajima et al., 2007). However, its role in cornea is not widely studied.
Dermatopontin (DPT)	4	4.815	DPT is involved in cell adhesion, migration, and extracellular matrix organization (Copper et al., 2006).
Lumican (LUM)	5	7.955	Lumican, a leucine-rich proteoglycan, is implicated in regulating assembly of collagen fibrils and the highly organized extracellular matrix essential for corneal transparency (Chakravarti et al., 2006).
Fibroblast growth factor 10 (FGF10)	6	2.692	Involved in corneal homeostasis, epithelium proliferation, and endothelial wound healing (Wang et al., 2021; Finburgh et al., 2023).
Platelet-derived growth factor receptor β (PDGFRB)	7	8.456	Regulates cell proliferation, migration, and survival (Mohan et al., 2024).
Extracellular Matrix Protein 2 (ECM2)	8	5.800	ECM2 contributes to the structure and function of the extracellular matrix (Yu et al., 2021)
Decorin (DCN)	9	6.445	DCN helps maintain the extracellular matrix organization and prevents corneal fibrosis (Mohan et al., 2011, Balne et al., 2021; Gupta et al., 2022).
Vascular Endothelial Growth Factor D (VEGFD)	10	7.330	VEGFD promotes angiogenesis and lymphangiogenesis (Di Zazzo et al., 2020).
Osteomodulin (OMD)	11	5.049	OMD is primarily produced by keratocytes. It helps regulate the structure and function of the stromal extracellular matrix (Tashima et al., 2015).
Osteoglycin (OGN)	12	8.717	OGN, also known as mimecan is another member of the small leucine-rich proteoglycan family. OGN regulates collagen fibril assembly and mineralization to maintain corneal structure and function (Funderburgh et al., 1997; Dunlevy et al., 2000).
Angiopoietin 1 (ANGPT1)	13	5.336	ANGPT1 is involved in blood vessel maturation and stability (Morisada et al., 2005).
Cadherin 5 (CDH5)	14	4.775	CDH5 is also known as VE-cadherin and CD144. It is crucial for maintaining endothelial cell junctions, vascular integrity, and apoptosis (Liao et al., 2000; Liu et al., 2015).
Prolargin (PRELP)	15	5.808	PRELP, similar to lumican, decorin, and biglycan, is a small leucine-rich repeat proteoglycan. It has been identified as a major proteoglycan in both the sclera and the cornea, and it binds to Type I collagen, serving as an anchor between the extracellular matrix and basement membranes (Dyrlund et al., 2012; Poulsen et al., 2018).

Fig. 9 shows the PPI network among the downregulated genes when the Q-hCSFs were compared with the P-hCSFs. The subnetworks for the top six enriched pathways discussed above are given in supplementary Fig. S2. Each node represents a protein encoded by a downregulated gene. The color intensity of the nodes indicates their influence within the network, with darker red representing the most influential nodes as determined by the MCC method. The top fifteen most influential nodes, highlighted in darker red, include key proteins such as Budding uninhibited by benzimidazoles 1 homolog (BUB1), Threonine tyrosine kinase (TTK), Kinesin family member 23 (KIF23), Kinesin family member 11 (KIF11), BUB1 mitotic checkpoint serine/threonine kinase B (BUB1B), Discs large homolog associated protein 5 (DLGAP5), Nucleolar and spindle associated protein 1 (NUSAP1), Cyclin A2 (CCNA2), Cyclin B1 (CCNB1), Baculoviral IAP repeat containing 5/Survivin (BIRC5), Cyclin-Dependent Kinase 1 (CDK1), Kinesin family member 20A (KIF20A), Aurora kinase B (AURKB), Kinesin family member 2C (KIF2C), and Cell division cycle associated 8 (CDCA8). Downregulation of these proteins may influence various cellular processes, including cell proliferation (CCNA2, CCNB1, and CDK1), cellular mitotic division (BUB1, BUB1B, TTK, and AURKB), cytokinesis (KIF23, KIF20A, and KIF2C), microtubule and spindle function (KIF11, DLGAP5, and NUSAP1), and apoptosis (BIRC5).

Fig. 9.

Protein–protein interaction network of downregulated genes in primary quiescent vs proliferating-hCSF.

Table 5 presents the top fifteen downregulated proteins in quiescent-hCSFs based on node rank. The log2 (FC) values of the corresponding genes resulting from the DGE analysis are also listed. Each protein’s role in the cornea is detailed, emphasizing their functions in the spindle assembly checkpoint, cytokinesis, cell cycle control, apoptosis inhibition, and their overall impact on corneal repair and regeneration.

Table 5.

Functions and importance of the top fifteen downregulated proteins in quiescent human corneal stromal fibroblasts identified through protein-protein interaction network analysis.

Protein name	PPI-network node rank	Log2 (FC) of corresponding gene	Role in cornea
Budding uninhibited by benzimidazoles 1 homolog (BUB1)	1	−4.770	BUB1 is a serine/threonine kinase that plays a critical role in the spindle assembly checkpoint (SAC), which ensures proper chromosome alignment and segregation during mitosis. Downregulation of BUB1 can lead to chromosomal instability and defective mitosis, resulting in compromised cell division and potentially affecting corneal repair and regeneration processes (Kim and Gartner, 2021).
Threonine tyrosine kinase (TTK)	2	−3.928	TTK is a dual-specificity protein kinase that plays a crucial role in SAC, ensuring proper chromosome alignment and segregation during mitosis. TTK is involved in monitoring and correcting kinetochore-microtubule attachments, thus preventing anaphase onset until all chromosomes are correctly attached to the spindle apparatus (Qi et al., 2021). Downregulation of TTK can lead to defective SAC function.
Kinesin family member 23 (KIF23)	3	−3.803	KIF23 is a motor protein that plays a critical role in cytokinesis. It transports organelles within cells and facilitates the movement of chromosomes during cell division (Li et al., 2021). Downregulation of KIF23 can lead to cytokinesis failure, resulting in binucleated or multinucleated cells, impaired cell proliferation, and defective wound healing.
Kinesin family member 11 (KIF11)	4	−4.359	KIF11 is also known as Eg5. It is a motor protein that plays a crucial role in the formation and maintenance of the bipolar spindle during mitosis. Its role in cancer and Retinopathy established (Birtel et al., 2017; Imai et al., 2017; Luo et al., 2022; Wang et al., 2022). However, its role in cornea is not widely studied.
BUB1 mitotic checkpoint serine/threonine kinase B (BUB1B)	5	−4.647	BUB1B, also known as BUBR1, helps maintain the integrity and proper cell division (Bolanos-Garcia and Blundell, 2011). Its role in cornea is not widely studied. Downregulation of BUB1B can lead to mitotic errors and compromised cell function, may affect corneal repair and homeostasis.
Discs large homolog associated protein 5 (DLGAP5)	6	−5.253	DLGAP5, also known as HURP (Hepatoma Up-Regulated Protein), is a microtubule-associated protein involved in the regulation of the cell cycle (Chen et al., 2024). Its downregulation may lead to cell proliferation impairment.
Nucleolar and spindle associated protein 1 (NUSAP1)	7	−4.496	NUSAP1 is a microtubule-associated protein that plays a crucial role in mitotic spindle assembly and stability. It is involved in spindle microtubule organization, chromosome segregation, and cytokinesis (Li et al., 2021).
Cyclin A2 (CCNA2)	8	−2.657	CCNA2 is a regulatory protein involved in cell cycle control. It binds to CDK1 and CDK2, playing a crucial role in the transition from G1 to S phase and from G2 to M phase. CCNA2 ensures proper DNA replication and cell division and considered as cell proliferative marker (Wu et al., 2023).
Cyclin B1 (CCNB1)	9	−6.264	CCNB1 is a regulatory protein that controls the G2/M transition in the cell cycle. It forms a complex with CDK1, known as the maturation-promoting factor (MPF), which is essential for initiating mitosis. It is also referred as cell proliferation marker (Li et al, 2021).
Baculoviral IAP Repeat Containing 5/Survivin (BIRC5)	10	−5.763	BIRC5 is a member of the inhibitor of apoptosis protein family. It inhibits caspase activation, thereby preventing apoptosis, and is also involved in regulating cell division by participating in the chromosomal passenger complex. In the corneal stroma, BIRC5 (Survivin) plays a dual role in promoting cell survival and regulating mitosis in corneal stromal fibroblasts (Konstantopoulou et al., 2018). By inhibiting apoptosis, it helps maintain the population of stromal fibroblasts, which are essential for corneal repair and regeneration. Downregulation of BIRC5 can lead to increased apoptosis and disrupted cell division, impairing corneal repair processes.
Cyclin-Dependent Kinase 1 (CDK1)	11	−4.245	CDK1 is a key regulatory kinase that drives the cell cycle progression from G2 phase to M phase, facilitating mitotic entry and progression (Kalous et al., 2020). Downregulation of CDK1 can result in cell cycle arrest, impaired cell division.
Kinesin family member 20A (KIF20A)	12	−5.305	It is primarily involved in cytokinesis, particularly in the formation and functioning of the mitotic spindle and the midbody during cell division. KIF20A ensures proper segregation of chromosomes and the completion of cytokinesis (Harvey et al., 2004). Downregulation of KIF20A can lead to cytokinesis failure, resulting in binucleated cells and disrupted cell proliferation.
Aurora kinase B (AURKB)	13	−4.935	AURKB is a serine/threonine kinase that is part of the chromosomal passenger complex. It is renowned for its essential role in mitotic progression, encompassing spindle assembly, chromosome alignment, and cytokinesis. It does these functions through precisely timed and spatially organized phosphorylation of interacting proteins at the centromere region (Wang et al., 2023). Downregulation of AURKB can lead to mitotic errors, such as aneuploidy and incomplete cytokinesis, resulting in impaired cell division and function.
Kinesin family member 2C (KIF2C)	14	−7.368	KIF2C is a microtubule depolymerase that plays a critical role in mitosis by regulating microtubule dynamics (Conrad et al., 2010). Downregulation of KIF2C can lead to mitotic defects, such as aneuploidy and chromosome missegregation, resulting in impaired cell proliferation and function.
Cell division cycle associated 8 (CDCA8)	15	−7.451	CDCA8 is a key component of the chromosomal passenger complex CPC, which also includes Aurora B kinase, INCENP, and Survivin (Xiang et al., 2022). Downregulation of CDCA8 can lead to mitotic defects, such as improper chromosome alignment and failed cytokinesis

4. Discussion

This study provides an in-depth analysis of the transcriptional responses of hCSFs under different culture conditions, i.e., quiescent (serum-free) and proliferating (serum-laden). By leveraging RNA sequencing data, pathway enrichment analysis, and PPI network analysis, we elucidated and identified key pathways influenced in Q-hCSFs compared with P-hCSFs.

The differential gene expression analysis identified 5,181 significantly regulated genes (P-adjusted < 0.05) between Q-hCSFs and P-hCSFs. Among these genes, 1,391 genes exhibited at a cutoff log2 (FC) of ±1.5 (674 upregulated and 717 downregulated genes) (Fig. 2). These results are summarized in Fig. 3, which shows a volcano plot depicting the distribution of upregulated and downregulated genes. The high number of differentially expressed genes highlights the significant impact of serum on the genetic landscape of hCSF (Fig. 4).

The gene set enrichment analysis was used to identify pathways significantly enriched in the quiescent-hCSFs compared to their proliferating counterparts. The key pathways included those related to cell cycle regulation, such as E2F Targets (NES = 2.7), G2M Checkpoint (NES = 2.49), and MYC Targets V1 and V2 (NES = 2.28 and 2.35, respectively) (Fig. 5A). These pathways are critical for cell proliferation and division, suggesting that serum supplementation activates molecular mechanisms essential for corneal cell proliferation. Additionally, pathways involved in inflammatory and oxidative stress responses, including TNFA signaling via NF-kB (NES = 2.17), oxidative phosphorylation (NES = 2.13), and IL6 JAK STAT3 signaling (NES = 1.8), were also enriched (Fig. 6).

The most significantly enriched E2F targets pathway is critical in the regulation of the cell cycle, especially the transition from the G1 phase to the S phase, where DNA replication occurs. In the context of the cornea, the E2F targets pathway contributes to several key processes, such as cell proliferation, wound healing, maintenance of corneal homeostasis, regulation of corneal fibroblasts, and response to DNA damage. E2F transcription factors are vital for the proliferation of corneal epithelial cells and stromal fibroblasts. The E2F targets pathway is activated in response to corneal injury, promoting the proliferation and migration of corneal cells to close wounds (Oshi et al., 2020). This process is crucial for maintaining the transparency and refractive properties of the cornea. E2F genes help maintain a balance between cell proliferation and cell death (apoptosis) in the cornea (Fouad et al., 2021). The cornea is exposed to various environmental stresses, including UV radiation, which can cause DNA damage, and this balance seems essential for preserving the structural and functional integrity of the corneal tissue. The pathways identified could serve as starting points for developing targeted therapies aimed at modulating corneal stromal cell activity, particularly in conditions where abnormal cell proliferation or quiescence contributes to disease. However, additional research is necessary to fully understand these mechanisms and explore their therapeutic potential in a clinical setting.

GO analysis of the DEGs between the Q-hCSF and P-hCSF groups revealed significant enrichment in key biological processes, such as mitotic regulation, nuclear division, and chromosome segregation (Fig. 7A–B). These findings suggest that the observed gene expression changes are intricately linked to critical aspects of cell cycle control, potentially contributing to the distinct biological behavior of Q-hCSF. Furthermore, the identification of comodulated genes highlights the complex regulatory networks that underlie these processes, providing a deeper understanding of the molecular mechanisms driving the phenotypic differences between Q-hCSF and P-hCSF (Fig. 7C–D).

PPI network analyses highlight the critical proteins and interactions that may play essential roles in maintaining the quiescent state or facilitating the transition between quiescent and proliferative states in hCSF, and understanding these interactions provides valuable insights into corneal cell behavior. Figs. 8 and 9 present the widespread PPI network of upregulated and downregulated genes, respectively, when the Q-hCSFs were compared to P-hCSFs. The PPI network for proteins of the upregulated genes (Fig. 8) revealed significant functional relationships among the proteins, with the top fifteen most influential nodes including FGF22, CD34, ASPN, DPT, LUM, FGF10, PDGFRB, ECM2, DCN, VEGFD, OMD, OGN, ANGPT1, CDH5, and PRELP (Table 4). These proteins are crucial for extracellular matrix organization, angiogenesis, and growth factor signaling, suggesting their key roles in maintaining corneal structure and function, facilitating tissue repair, and supporting the cellular signaling pathways critical for corneal health (Joyce, 2012). Conversely, the PPI network for corresponding proteins of downregulated genes (Fig. 9) highlights the top fifteen most influential nodes, including BUB1, TTK, KIF23, KIF11, BUB1B, DLGAP5, NUSAP1, CCNA2, CCNB1, BIRC5, CDK1, KIF20A, AURKB, KIF2C, and CDCA8 (Table 5). These proteins are essential for cell cycle regulation, mitosis, cytokinesis, microtubule function, and apoptosis (Liu et al., 2015; Liao et al., 2000; Li et al, 2021; Qi et al., 2021). The downregulation of these genes suggests significant impacts on corneal cell proliferation, division accuracy, and survival, leading to potential impairments in corneal repair and regeneration processes. The analysis of transcriptional profiles and PPI networks provides a comprehensive understanding of the regulatory mechanisms in hCSFs under different culture conditions. The identification of crucial pathways and hub genes offers potential targets for therapeutic intervention aimed at enhancing corneal repair and regeneration. Moreover, this study highlights the importance of considering the cellular microenvironment when designing therapeutic strategies. The distinct expression patterns observed under serum-free conditions highlight the need for tailored approaches that mimic the in vivo environment. Future research should focus on validating these targets in vivo and exploring their roles in corneal wound healing and fibrosis. Additionally, the integration of other omics data, such as proteomics and metabolomics data, could provide a more holistic view of the molecular mechanisms governing hCSF behavior.

Although this study offers comprehensive information regarding the transcriptomic differences between quiescent and proliferating hCSFs but has few limitations. For example, use of single serum concentration (10%) in proliferating culture which may not represent transcriptomic landscape of other serum concentrations. Another perceived limitation of the study is a difference in confluency between quiescent and proliferating hCSFs at the time of RNA extraction. To account this limitation, same quantities of RNA were used in RNASeq analysis. Further, this study used a standard in vitro adherent 2D cell culture model, which is routinely used in corneal wound healing studies. Our future studies will investigate if there are significant differences in transcriptomic profile between 2D and 3D tissue/organ in vitro models.

In conclusion, this study provides a basic framework for understanding molecular basis of hCSF behavior under quiescent and proliferative culture conditions. The new transcriptomic insights will lead a better understanding of corneal wound healing in vitro.

Supplementary Material

1

Fig. S1. The graph shows the RNA yield (normalized to the cell count) for proliferating (hCSF-A1 to hCSF-A4) (blue) and quiescent (hCSK-C1 to hCSK-C4) (green) human corneal stromal fibroblasts.

2

Fig. S2. Subnetworks for the top six enriched pathways are depicted, where each node represents a protein encoded by a downregulated gene, with the color intensity reflecting the node’s influence within the network, determined by the Maximum Clique Centrality (MCC) method. Darker red nodes indicate higher influence. The pathways include E2F Targets, G2M Checkpoint, MYC V1 and MYC V2, TNFA Signaling via NFKB, and Oxidative Phosphorylation.

Highlights.

• The study Identified 5,181 significantly regulated genes between quiescent and proliferating hCSFs, showing the serum’s impact on gene expression.

• The key enriched pathways in the quiescent-hCSFs are cell cycle regulation (E2F Targets, G2M Checkpoint, MYC Targets) and stress response (TNFA signaling, oxidative phosphorylation).

• The study emphasizes the roles of cell cycle and stress response pathways in quiescent-hCSFs.