Characterization of the prognostic values and response to immunotherapy/chemotherapy of Krüppel‐like factors in prostate cancer

Abstract At present, the overall genetic and epigenetic effects of Krüppel‐like factors (KLFs) on prostate cancer (PCa) remain unclear. Therefore, we systematically investigated the molecular differences in KLFs of transcription expression, promoter methylation and genetic alteration. Univariate and multivariate Cox proportional hazard regression was used to analyse the effect on RFS and establish the prognostic signature in the TCGA cohort, MSKCC and GSE116918 cohorts employed to validate the signature. Biological pathway enrichment and the potential response to immunotherapy and chemotherapy were inferred. The transcription levels of most KLFs are associated with the clinical outcome of PCa. Gleason score (P = .009), pathology T stage (P = .006), KLF3 (P = .034), KLF5 (P = .002) and KLF7 (P = .035) were independent prognostic factors. A prognostic signature was established in the TCGA cohort (P < .001) and validated in the MSKCC (P < .001) and GSE116918 cohorts (P = .006). Demethylation of KLF5 by 5‐azacytidine led to increased protein levels, whereas knockdown of KLF5 promoted cell proliferation. Patients in KLF‐F were more likely to respond to immunotherapy (P < .001) and bicalutamide (P < .001). In summary, we found that the KLFs and clinical feature‐based signatures may improve prognosis prediction in PCa and further promote patient stratification and disease management.


| INTRODUC TI ON
Prostate cancer (PCa) is a common malignant carcinoma among males worldwide, and accounts for the second greatest prevalence and a fifth of cancer-specific deaths. More than 300 000 deaths are caused by PCa annually, accounting for approximately 6.6% of cancer-specific mortality in males. 1-3 Additionally, PCa is the second most frequent cancer (13%) in the oldest-old males, of whom older than 85 years old, and is the primary cause of mortality in the United States (20%). 4 In China, the incidence of PCa has risen sharply from 10% to 20% over the last two decades due to the widespread use of prostate biopsies for diagnosis. [5][6][7] The overall survival rate for patients with PCa is not poor compared with that of other malignancies, with more than 80% survival during the first five years after diagnosis. [8][9][10] However, the recurrence rate of PCa is high, and most patients will enter the advanced castration-resistant PCa (CRPC) stage, which increases the risk of PCa-specific death. [11][12][13][14] Krüppel-like factors (KLFs) are zinc finger proteins that bind to the DNA transcriptional region and act as transcriptional activators or repressors. Numerous biological processes are affected by KLFs, such as cell proliferation and differentiation as well as the development of mammalian tissues and organs by maintaining the homeostasis of both tissues and systemics. 15,16 There are 18 KLF family members, and each consists of three common conserved Cys2/His2 zinc fingers. KLF family members bind to similar promoter regions in the C-terminal domains of genes, such as CACCC-, GC or GT-box. 17 The features of KLFs are both exclusive and overlapping. For example, KLF2, KLF4 and KLF6 are involved in the activation of macrophages, and KLF4 induces the M2 phenotype macrophage through IL-4, while KLF2 can inhibit NF-κB-dependent proinflammatory activation and promote M2 polarization. 18,19 In contrast, KLF6 promotes the polarization of M1-type macrophages through the inhibition of the NF-κB pathway. 20,21 Moreover, KLF2, KLF4 and KLF5 have been linked to the pluripotency of stem cells. 22 In past decades, KLFs have been found to play pivotal roles in tumorigenesis. Kim et al found that KLF12 promotes tumour growth by directly activating EGFR and serves as a prognostic marker in colorectal cancer. 23 Tsompana et al 24

| Study population
We obtained the molecular data of PCa patients from The Cancer Genome Atlas Project (TCGA). The transcriptional expression data were acquired from the public data hub UCSC Xena (https://xenab rowser.net/), and consisted of 499 PCa samples and 52 normal samples. The transcriptomic profiles of the KLFs were extracted from the whole gene transcription data under the archive of the PRAD project. The MSKCC and GSE116918 cohorts were also enrolled as validation cohorts to evaluate the prognostic value of KLFs plus clinical feature signatures. The clinicopathological information for the enrolled cohorts is summarized in Table S1.

| DNA methylation of KLFs
The β value ranges from 0 (unmethylated) to 1 (fully methylated), which indicates that the overall methylation level of the promoter region of KLFs was retrieved from MethHC (http://methhc.mbc.nctu. edu.tw/php/index.php). 25 The correlation between the promoter methylation β value and KLFs expression level was evaluated via linear regression using GraphPad Prism 8.

| Genetic alterations in KLFs
The genetic alterations of KLFs in patients with PCa were illustrated via the cBioPortal platform (http://www.cbiop ortal.org/), 26,27 which recorded the missense and truncating mutations as well as amplification and deep deletion. The RFS between patients with or without alterations was also conducted to evaluate the prognostic value of genetic alterations in KLFs. The expression of KLFs with gene amplification was depicted by GraphPad Prism 8.

| Cell culture and reagents
The PC3, 22RV1, and HEK293T cell lines were purchased from the American Type Culture Collection (ATCC, Manassas, VA). HEK293T cells were cultured in DMEM medium, while PC3 and 22RV1 cells were cultured in RPMI 1640 medium. To prepare the media, 10% foetal bovine serum and 1% penicillin and streptomycin solution were added before use. The cell incubators were maintained at 37°C and 5% CO 2 . Bicalutamide (Sigma, #90357-06-5, ≥98%) was used at 10 μM for the treatment of 22RV1 cells.

| Knockdown KLF5 plasmid design, lentivirus packaging and cell transfection
The primers used to generate the knockdown sequence of KLF5

| Colony formation assay
pLKO and shKLF5 PCa cells (PC3 and 22RV1) were seeded in six-well plates containing 800 cells per well and allowed to grow for an additional 12 days. Then, the culture solution was discarded, and the cells were rinsed twice with cold PBS. They were then fixed using 4% paraformaldehyde for 20 min and subsequently stained with 0.5% crystal violet staining solution for 20 min. The colonies were photographed and counted under a microscope.

| Western blot
Cells were washed twice with cold PBS and lysed in RIPA lysis buffer, and proteins (40-50 μg) were separated on 6%-10% SDS/PAGE gels then transferred onto PVDF membranes (Millipore). After PVDF membranes were blocked, they were sequentially incubated with primary antibodies, HRP-conjugated secondary antibodies, and visualized using an ECL system (Thermo Fisher Scientific). The primary antibodies used in the Western blot study included KLF5 (ABclonal, #A12403), E-cadherin (ABclonal, #A11492), vimentin (ABclonal, #A2666) and GAPDH (Santa Cruz, #sc-166574). The following day, anti-rabbit, antimouse or anti-goat IgG secondary antibody was used for 1 hour at a concentration of 1:5000 at 16°C and rinsed for 5 minutes with TBST three times.

| Identification of risk-associated differentially expressed genes (DEGs) and genome enrichment
The R package 'edgeR' was utilized to perform differential expression analysis with the standard comparison mode. DEGs were identified as genes that passed the threshold of P < .05 and absolute log2 foldchange >0.5. The expression levels of the top DEGs for each patient were displayed with a heatmap, and GEPIA (http://gepia.cance r-pku. cn) was used to investigate their association with patient RFS. 28 The DEGs were enrolled to generate the gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGGs) enrichment analysis using Metascape (http://metas cape.org) 29 with a threshold of false discovery rate <0.05. GSEA was used to identify pathways enriched in high-and low-risk patient groups.

| Immune infiltration, immunotherapy and chemotherapy response prediction
The infiltration of 22 subtypes of tumour-infiltrating immune cells (TIICs) was retrieved from our previous study, which was calculated by CIBERSORT, an algorithm that quantifies the proportion of TIICs with 547 signature genes. 30 We also evaluated the chemotherapy response of each patient using the public pharmacogenomics database Genomics of Drug Sensitivity in Cancer (GDSC; https://www.cance rrxge ne.org).
Chemotherapy drugs cisplatin, docetaxel and bicalutamide that are normally used to treat patients with PCa were selected for evaluation. Based on the GDSC data, the half-maximal inhibitory concentration (IC 50 ) was estimated and represented the response of the drug. Therefore, the R package 'pRRophetic' was used with 10-fold cross-validation and other parameters by default. 36

| Statistics
To obtain the normalized expression data of KLFs, we converted the raw count data to the number of fragments per kilobase of non-over- by the mean value of the signature. K-M plots present the different RFS in the two groups, and the receiver operating characteristic (ROC) curve was plotted to illustrate the predictive performance of the signature. For all statistical analyses, P < .05 was considered statistically significant.

| KLF expression and related RFS in patients with PCa
We obtained and calculated the mRNA expression of 17 KLFs from the TCGA-PRAD database, which contains 52 normal prostate tissues and 499 PCa tissues, while the mRNA level of KLF18 was not detectable and, thus, excluded from this study. As shown in Figure 2A, or KLF14 (Table S2). Furthermore, we evaluated the prognostic value of each KLF and found that the high expression of KLF7 (P = .035, Figure S1D), KLF11 (P = .024, Figure S1E) and KLF17 (P = .020, Figure   S1H) indicated a poor prognosis with RFS, while the lower expression of KLF3 (P = .015, Figure S1A), KLF4 (P = .007, Figure S1B), KLF5 (P = .002, Figure S1C), KLF14 (P = .023, Figure 2F) and KLF15 (P = .033, Figure S1G) was associated with a favourable RFS. Prognostic analyses of other KLFs were also conducted using the K-M curve, as shown in Figure S2.

| KLF promoter methylation and genetic alteration
mRNA expression is affected by promoter methylation in cells; thus, we evaluated the promoter methylation levels of KLFs. We found that increased promoter methylation in the tumour tissues of KLF3, KLF6, KLF7, KLF8, KLF10, KLF11, KLF12 and KLF13 resulted in decreased mRNA expression as compared with normal tissues (P < .05; Table S3). We also evaluated the association between DNA methylation and mRNA expression of KLFs in PCa tumour tissues and found that the mRNA expression of most KLFs was negatively associated with its DNA methylation at the promoter region (P < .05), except for KLF3, KLF9 and KLF17 ( Figure 2B).
Genetic alteration is another critical factor that affects the mRNA expression of genomic genes. Appropriately, 15% deep deletion occurred among TCGA-PRAD patients to all KLFs, while amplification accounted for approximately 10%, and the genetic mutation fraction was less than 5% ( Figure 2C). What is more, we identified a profound deep deletion rate in KLF5 (11%) and KLF12 (11%), while KLF10 had the highest amplification rate among all KLFs (7%; Figure 2D). Furthermore, this study revealed that the deletion of KLF5 (heterozygous deletion vs. homozygous deletion: P = .0103; Diploid vs. homozygous deletion: P = .0085; Diploid vs. heterozygous deletion: P = .0001; Figure S3A) and KLF12 (heterozygous deletion vs. homozygous deletion: P = .0377; Figure S3B) affected their mRNA levels. As the mRNA levels of KLFs are associated with a poor prognosis, genetic alteration and DNA methylation might be involved.

| Establishment and validation of KLF-related prognostic signature
To further explore whether and how KLFs impact the process and prognosis of PCa, we calculated the risk score for each patient based on the following formula: risk score = 0.542*Gleason score + 0.559*pathology T stage −0.040*KLF5 + 0.077*KLF13 ( Figure 3A, Table S4; see Section 2). In this formula, the Gleason score was divided into five groups with different scores. The pathological T stage included four groups: T1, T2, T3 and T4. With the risk score, patients were divided into KLF-F (low-risk, n = 208) and KLF-P (high-risk, n = 207) groups.
The recurrence rate was high at 26.09% in the KLF-P group, while only 8.17% of patients met the recurrence rate in the KLF-F group ( Figure 3B). We also evaluated the expression of KLF5 and KLF13 and identified the decreased expression of KLF5 in the KLF-P group, along with an increased level of KLF13 ( Figure 3C). Subsequently, the K-M plot was employed to test the discrimination value of the risk score, as shown in Figure 3D.  Figure 3H). Results from the GSE116918 cohort also indicated a good application of the prognostic signature with a dramatic RFS difference between KLF-F and KLF-P groups (P = .0055, Figure 3F) as well as a high AUC value (1-year AUC = 0.832; 3-year AUC = 0.574; 5-year AUC = 0.635, Figure 3I).

| KLF5 expression affected by 5-azacytidine and knockdown of KLF5 promoted cell proliferation of PCa
We first detected the baseline KLF5 protein levels in several PCa cell lines and found that KLF5 is highly expressed in PC3 and 22RV1 cells and is lower in C4-2 and C4-2B cells ( Figure 4A). Therefore, we selected PC3 and 22RV1 cell lines for the next validation. Based on the analysis between KLF expression and DNA methylation, we found that KLF5 is negatively associated with the methylation of its upstream promoter region ( Figure 2B). Therefore, we detected the protein level of KLF5 after treatment with 5-azacytidine, a commonly used inhibitor of DNA methylation. 37 As shown in Figure 4B, after treatment with 5-azacytidine, the protein levels of KLF5 increased in PC3 and 22RV1 cell lines. Subsequently, we used the shKLF5 lentivirus to knockdown KLF5 and found that cell proliferation significantly increased after knockdown of KLF5 in PC3 and 22RV1 cells ( Figure 4C,D), and the same tendency was also observed in the colony formation assay ( Figure 4E,F).

| Identification of DEGs between KFL-F and KLF-P groups and pathway enrichment
The DEGs in the KFL-F and KLF-P groups were obtained using 'edgeR' R packages with log 2 fold-change >0.5 or <−0.5 and P value <0.05.
The red dot represents the highly expressed gene in KLF-P, while the blue dot represents the highly expressed gene in KLF-F ( Figure 5A). The top 10 highly expressed genes in the KFL-F and KLF-P groups are shown in Figure 5B. The highly expressed ARHGDIG in the KLF-P group leads to a poor prognosis of PCa (P = .00028, Figure 5C), and the expression of ARHGDIG in tumours is negatively associated with the KLF5 levels (R = −0.34, P < .001, Figure 5F), confirming the prognostic value of KLF5 in PCa. Meanwhile, the high levels of the KLF-F group highly expressed LCN2 and CD38 linked to a better RFS (P < .05, Figure 5D,E), and both were positively correlated with increased levels of KLF5 (LCN2: R = 0.56, P < .001; CD38: R = 0.21, P < .001, Figure 5G,H).
To obtain an in-depth understanding of the association between and the prognosis of PCa, we performed functional enrichment  Figure 5I). The epithelial-mesenchymal transition pathway is the pivotal pathway in the cell adherens junction; therefore, we evaluated the alteration of the EMT pathway and found that after the knockdown of KLF5 (simulating KLF-P status), the protein levels of E-cadherin decreased, while vimentin increased considerably.
These WB results showed that the EMT pathway was activated after the knockdown of KLF5 ( Figure 5J). We also used Metascape to generate the overall function of the different genes in the KLF-F and KLF-P groups in GO biological processes, reactome gene sets, KEGG pathways and canonical pathways. Figure S4A displays the pathway enrichment of highly expressed genes in KLF-P.
Enrichment is primarily related to nuclear division (red and green dots) and the cell cycle of mitotic cells (blue dot). For KLF-F-related enrichment, the NABA matrisome-associated pathway (red dot) was mostly enriched, which affects the extracellular matrix. We found that the chemotaxis, tissue morphogenesis and second-messenger-mediated signalling pathways were also annotated in the KLF-F-associated gene group ( Figure S4B).

| Different immune infiltration between KFL-F and KLF-P groups
The infiltration of TIICs in tumours plays a key role in the tumour environment and affects prognosis. In the present study, we found that KLKs distinguished patients with PCa with poor a prognosis (KLF-P) or favourable prognosis (KLF-F); thus, we further investigated the different TIIC infiltrations in tumours with different prognoses. We found that the distributions of plasma cells (P = .020) and resting mast cells (P = .024) were higher in KLF-F, while M2 macrophage (P < .001) infiltration was higher in the KLF-P than the KFL-F group ( Figure 6A, Table S5). Then, we evaluated the association between KLF5 expression and the above three immune cell infiltrations and found that plasma cells and M2 macrophages were negatively associated with KLF5 expression ( Figure 6B). In our previous study, we revealed that the high infiltration of M2 macrophages is linked with the poor prognosis of patients with PCa, and this study revealed that KLF5 is the key gene against the progression of PCa in databases and in vitro. Therefore, we analysed the combined effect of KLF5 and M2 macrophages and found that patients with low KLF5 and high M2 macrophage infiltration had the worst prognosis (HR = 4.67, P < .001, Figure 6C).

| Immunotherapy and chemotherapy are more practical for KLF-F patients
We further assessed the potential response to immunotherapy in each patient using the TIDE algorithm (Table S6), and observed that patients in the KLF-F group (51.92%, 108/208) were more likely to respond to immunotherapy than those in the KLF-P group (36.23%, 75/207; P = .0015; Figure 6D). Subsequently, we analysed the

| D ISCUSS I ON
PCa is a considerable worldwide burden on public health. The symptoms of urination discomfort, bone metastasis pain and castration treatment failure, and the high rates of recurrence make it an urgent public health event. 38  There are some advantages that should not be neglect in the current study. We established and validated a novel KLF-associated prognostic signature to help predict outcomes among almost 776 patients with PCa. Therefore, clinicians could use the signature to predict underlying recurrence and carry out effective treatment. In addition, the molecular function of KLF5 was confirmed in PCa cell lines, which could affect the proliferation and treatment sensitivity of bicalutamide through the EMT pathway. Meanwhile, some limitations should also be addressed and modified in future studies. First, more clinical samples should be used to confirm the effectiveness of the KLF-associated prognostic signature. Second, although we found that the KLF-F patients could benefit more from immunotherapy, there is no difference between KLF-F and KLF-P patients in CAR-T and PD-1/PD-L1 therapy; thus, the potential novel immunotherapy should be investigated in future. Third, the potential mechanism by which KLF5 promotes cell proliferation and affects bicalutamide sensitivity should be studied.
In summary, KLF family members are essential prognostic factors for PCa. The KLFs and clinical feature-based signatures identified the unfavourable prognosis precisely, while Bicalutamide is an effective medicine to treat KLF-F patients.