In this retrospective study, we included 141 samples from patients who received curative treatment at Shanghai General Hospital. All patients met the following inclusion criteria: 1) available baseline data and histologic material, 2) HCC diagnosed by a physician based on unambiguous histologic characteristics (Sun et al., 2011). Age, gender, hepatitis B virus surface antigens (HBsAg), liver cirrhosis, α-fetoprotein (AFP), Child-Pugh grade, tumor size, tumor number, tumor differentiation grade, tumor distribution, and pathological Tumor Node Metastasis stage (TNM stage) were all systematically recorded. All patients provided informed consent, and the academic study was conducted in accordance with ethical standards.

The primary antibody was anti-KLF4 (1:500, Santa Cruz Biotechnology). The IHC procedures were performed carried out precisely as described previously (Sun et al., 2016; Sun et al., 2017). KLF4 staining was classified into three categories based on the intensity of the staining and the percentage of positive cells: weak/negative, moderate, and strong (Wei et al., 2005). In addition, in order to differentiate between low and high KLF4 expression, negative/weak expression was deemed low, whereas moderate/strong expression was deemed high.

The TCGA database (https://portal.gdc.cancer.gov/) was used to obtain genomic data from 10,363 tumors and 730 adjacent normal tissues, as well as corresponding clinical information from 33 tumor types (Cancer Genome Atlas Research Network et al., 2013). We utilized level 3 RNA-seq data from LIHC for HCC, which included normal (n = 50) and tumor (n = 374) samples. All fragments per kilobase per million (FPKM) data were converted to transcripts per million (TPM) format and log2 (TPM+1) conversion for further research. KLF4 expression levels were assessed for each cancer type using log2 (TPM+1) of that gene, and mRNA expression data were characterized by median ± upper and lower quartiles. The ggplot2 package was used to create the visualization.

The prognostic ability of KLF4 was also determined by survival analyses between the low- and high-expression groups, which were separated by median expression thresholding. Subsequently, we performed various subgroup analyses based on clinical characteristics to investigate the overall survival (OS) and disease-free survival (RFS) of KLF4.

The clinical characteristics and demographics of the HCC cohort were presented using descriptive statistics, with comparative statistical methods described in the table legend. Survival analyses were performed using the survminer (version 0.4.9) and survival package (version 3.2-10). To test the independent prognostic value of KLF4, we used Cox regression via the survival package’s Cox regression function (coxph). Significant variables associated with survival on univariate analysis were included in the multivariate models for OS and RFS. Nomograms for prognosis prediction were also created using the RMS package (version 6.2-0) and survival package (version 3.2-10), and all independent predictors from Cox models were integrated. Finally, calibration curves and DCA were plotted to evaluate the calibration ability and the clinical significance.

The differentially expressed genes (DEGs) between two groups of patients (low-/high-KLF4 expression groups) were identified using the DESeq2 package (version 1.26.0) based on the median counts of KLF4 expression. Genes with an absolute value of log2 of fold change >1 and adjusted p-value < 0.05 were considered significantly differentially expressed, and their potential functions were investigated by gene ontology (GO) and KEGG enrichment using the clusterProfiler package (version 3.14.3) (Yu et al., 2012).

Immune cell infiltration levels in the HCC cohort were estimated using ssGSEA with the GSVA package (Hänzelmann et al., 2013). In our study, the deconvolution algorithm allowed us to distinguish 24 distinct subtypes of immune cells. The relationship between the level of expressed KLF4 and the infiltration levels of each distinct immune cell population was then characterized using Spearman’s correlation analysis. In addition, the levels of infiltration of various immune cells associated with antitumor immunity were compared. Spearman’s correlation was also used to examine the association between the KLF4 gene, the classical immune cell gene marker, and immune checkpoints. Concurrently, the interaction between KLF4 and immune scores was investigated in the CAMOIP database (https://www.camoip.net/) (Lin et al., 2022).

The validation cohort TMA consisted of 268 cases of HCC with matched non-tumor adjacent tissue obtained from Shanghai General Hospital. Primary antibodies used included anti-CD8 (Invitrogen, 14-0081-82), anti-CD68 (Abcam, ab955), and anti-KLF4 (Abcam, ab215036). CD8 and CD68 were used to identify CD8⁺ T cells and macrophages, respectively. TMA was processed for IHC analysis of immune cell infiltration and KLF4 expression. For the score of stained markers, we used a histologic score (H-score) based on the intensity and percentage score (Paschalis et al., 2019). H-score for each scanned section was determined automatically by Servicebio (Ding et al., 2019). In addition, patients were grouped based on the median H-score of KLF4 expression. To validate the relationship between infiltrating immune cells and KLF4, differences and correlations between low/high KLF4 expression CD8⁺ T cells and macrophages were examined.

All statistical analyses and visualizations were carried out using R version 3.6.3 (R Foundation) and the R package ggplot2. The Mann-Whitney U-test or Wilcoxon signed-rank test was applied to compare KLF4 expression between tumor and non-tumor tissues. The Wilcoxon signed-rank test, the Chi-square test, and the Fisher exact test were used to assess the association between KLF4 expression and clinicopathologic features. Log-rank test was utilized to compare survival curves. Cox regression was employed in both univariate and multivariate analyses. Three methods were used to estimate the property of a nomogram: DCA, calibration curve, and Concordance index (C-index). Infiltration levels of macrophages and CD8⁺ T cells were estimated using the Wilcoxon signed-rank test. All correlation analyses incorporated either Pearson’s or Spearman’s correlation tests. The difference was considered statistically significant when the p-value was less than 0.05 (*p < 0.05, **p < 0.01, ***p < 0.001).

Compared to normal tissues, the pan-cancer data analysis found that KLF4 expression was low in tumor tissues. 11 tumors exhibited low KLF4 expression: bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), colon adenocarcinoma (COAD), head and neck squamous cell carcinoma (HNSC), kidney chromophobe (KICH), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), stomach adenocarcinoma (STAD), thyroid carcinoma (THCA), and uterine corpus endometrial carcinoma (UCEC) (Figure 1A). Notably, we found that KLF4 was significantly underexpressed in HCC (Figure 1C). For additional validation, a larger sample size study of unpaired samples was done. Several tumors, including BLCA, BRCA, COAD, HNSC, KICH, LIHC, LUAD, LUSC, pheochromocytoma and paraganglioma (PCPG), rectum adenocarcinoma (READ), STAD, THCA, and UCEC, exhibited a lower expression level than normal tissue, although cholangiocarcinoma (CHOL) had a greater expression level (Figure 1B). Notably, KLF4 expression appears to be considerably decreased in HCC (Figure 1D).

141 HCC patients were classified according to four levels of KLF4 expression, as illustrated by representative IHC images in Figure 1E. KLF4 expression was used to categorize samples into negative/weak, moderate, and strong groups. Tumor tissues exhibited 46.1%, 47.52%, and 6.38%, whereas normal tissues exhibited 31.21% in the moderate group and 68.79% in the strong group, indicating that KLF4 expression was lower in tumors than in normal tissues (Figure 1F). These findings suggested that KLF4 could be a useful biomarker for detecting HCC.

Based on their IHC scores, 141 patients with HCC were divided into two groups. We categorized weak/negative as the group with low expression, and the remainder (moderate/strong) as the group with high expression. According to Table 1, low KLF4 expression was related to a high AFP level, a large tumor, and an advanced TNM stage. Other clinical characteristics in our HCC sample did not correlate significantly with KLF4 expression. Reduced KLF4 expression was significantly related to rapid HCC progression to advanced stages.

Figure 2A demonstrated that patients with a higher KLF4 expression had a better OS (HR = 0.24). This difference was also observed in the subgroup of patients with HCC (Figures 2B–I). In addition, high KLF4 expression was associated with an improved RFS prognosis in HCC (HR = 0.27) (Figure 2J). Analysis of subgroups revealed that highly expressed KLF4 had a longer RFS (Figures 2K–R). According to the results, KLF4 may have far-reaching effects on the prognosis prediction of HCC.

The results of univariate and multivariate analyses of OS and RFS were used to evaluate the clinical significance of KLF4 expression in HCC patients (Table 2). Age, liver cirrhosis, tumor number, tumor size, AFP level, tumor differentiation grade, TNM stage, and KLF4 expression were all significantly associated with OS and RFS in the univariate analysis. On the contrary hand, gender, HBsAg, tumor distribution, and Child-Pugh grade were not associated with OS or RFS. In addition, in multivariate analysis, only three of 8 prognostic factors remained independent, including KLF4 expression, tumor differentiation, and TNM stage, remained independent prognostic factors. In HCC patients, high KLF4 expression was consistently found to be an independent prognostic predictor of OS (HR = 0.274, 95% CI = 0.152-0.495) and RFS (HR = 0.325, 95% CI = 0.187–0.563).

We developed two nomograms that incorporate KLF4 expression, tumor differentiation, and TNM stage to establish more accurate prognostic models for OS and RFS in HCC patients. In OS and RFS nomograms, a worse prognosis was associated with a higher total sum point of the assigned number of points for each factor (Figures 3A, B). For instance, HCC with low KLF4 expression, tumor stage III, and tumor differentiation grade I/II would have a total of 191 points (91 for KLF4 expression, 0 for tumor differentiation grade I/II, and 100 for tumor stage III), corresponding to a predicted 1-year OS of 42.0%. To report the clinical net benefit of our models, we also developed a conventional prognostic staging system (TNM status). C-indices of OS and RFS in TNM status were 0.710 (0.685-0.735) and 0.706, respectively, demonstrating the prognostic capability of our model (0.682-0.730). The new models based on KLF4 expression had a greater net benefit for predicting OS (Figures 3C, D) and RFS (Figures 3E, F) than the TNM models for specific threshold probabilities (0.20-0.87) exhibiting a wider range in DCA at 1 and 3 years. The 5-year cut-off probabilities ranged between 0.25 and 0.60 in OS (Figure 3G) and between 0.25 and 0.75 in RFS (Figure 3I). Notably, RFS prediction models exacerbated these differences. The accuracy of our nomograms was then evaluated using calibration plots. Calibration plots for the 1-, 3-, and 5-year OS (Figure 3H) and RFS (Figure 3J) run very close to the diagonal, indicating that predicted and actual observed outcomes are consistent. Overall, the models we developed could aid clinicians in promoting optimal clinical outcomes for patients with HCC.

According to Figure 4A, 1,536 genes had a significant upregulated expression in high-KLF4 expression groups (red dots), while 357 genes had a significant downregulated expression (blue dots). Enrichment analysis of all DEGs subsequently revealed that KLF4 was involved in numerous pathways (Figure 4B). The relationship between KLF4 and immune response-activating cell surface receptor signaling pathway, regulation of lymphocyte immunity, phagocytosis, T cell activation, external side of the plasma membrane, antigen binding, and so on, was revealed by GO and KEGG annotation (Figures 4C–F). These findings suggested that the KLF4 expression network influenced the TIME in HCC.

In our study, we found a correlation between KLF4 expression and immune population levels. HCC exhibited significant variations in macrophages, active DC, T cells, T helper cells, and other immune cells (Figure 5A). Overall, KLF4 expression increased antitumor immune cell infiltration, suggesting KLF4 may play a significant role in antitumor immune response (Figure 5B). KLF4 was also found to be significantly associated with CD8+T cells, Th1 cells, DC, B cells, NK cells, and macrophages gene markers (Table 3). When measured using immune-related scores, all outcomes remained constant (Figure 5C). Patients with elevated KLF4 expression had elevated scores for IFN-y response, macrophage regulation, Th1 cells, and lymphocyte infiltration. In the meantime, it was discovered that KLF4 is strongly associated with the genes involved in immune checkpoints, particularly the genes responsible for T-cell exhaustion (Figures 5D, E). It is widely believed that HCC patients with elevated KLF4 levels are more susceptible to immune checkpoint blockades (ICBs) and that targeting KLF4 could affect exhausted T cell reprogramming to improve immune responses to cancer cells.

As previously stated, the TIME appeared to have some distinct characteristics in HCC patients with high KLF4 expression. The remarkable correlation between immune cell infiltration levels and KLF4 expression was confirmed by our clinical cohorts. The high KLF4 expression group in HCC tissues contained significantly more CD8⁺ T cells and macrophages than the low KLF4 expression group. Additionally, CD8⁺ T cells and macrophage infiltration were positively correlated with KLF4 expression levels (Figures 6A, B). Notably, only macrophages infiltrated adjacent normal tissues similarly. KLF4 expression was associated with increased macrophage infiltration, but not CD8⁺ T cell infiltration (Figures 6C, D). Again, the significant relationship between KLF4 expression and immune infiltration in HCC, specifically macrophages, was evident, suggesting that KLF4 may play crucial roles in the TIME of HCC.