JMJD5 mRNA expression in various types of tumor tissues was examined by the TIMER [16, 17, 18] database (http://timer.cistrome.org/). The differential mRNA expression between tumor and normal tissues for JMJD5 has been detected using the “Gene-DE” module in TIMER across The Cancer Genome Atlas (TCGA) tumor resources. JMJD5 mRNA expression was displayed using box plots, showing the median, spread and outliers by RNA-Seq normalized by transcript per million (TPM) across normal and cancerous tissues.

All these cancer patients’ samples were obtained from Huaihe Hospital of Henan University. The present research has been approved by the Ethics Committee of Huaihe Hospital of Henan University, under the condition of written consent by each patient. All cases were diagnosed histologically by following the World Health Organization classification. All tissues were fixed in 4% buffered formaldehyde and then paraffin embedded to construct TMAs. Eight separate TMAs were generated, containing 14 different types of cancers (Table 1). The detailed IHC protocol has been previously published [19]. The following antibodies were used: rabbit anti-human JMJD5 polyclonal antibody (1:250, Abcam #28883, USA) and HRP-Polymer anti-Rabbit IHC Kit (Maixin, Fuzhou, China). Stained sections were scanned using a ScanScope T2 automated slide scanner (Aperio Technologies, Vista, CA, USA). The IHC staining was independently evaluated by two authors without knowledge of the clinicopathological information. The quantification of IHC was transformed to parameters which give the mean optical density measured using Image-Pro Plus 2.0 (Media Cybernetics, USA), the software determined the final date through optical density cumulative value divided by the target distribution area.

The effect of JMJD5 on relapse-free survival (RFS) in the above significantly expressed cancers were investigated by using Kaplan–Meier plotter [20] (www.kmplot.com). Survival analyses were performed to generate Kaplan-Meier plots. Taking 95% confidence intervals (CIs) as hazard ratios (HRs), we obtained log-rank p values.

In order to ascertain the correlations between JMJD5 expression and six tumor-infiltrating immune cells (B cells, CD4${}^{+}$ T cells, CD8${}^{+}$ T cells, macrophages, neutrophils and myeloid dendritic cells) in BRCA, LIHC, LUAD, LUSC and STAD, we chose to use the TIMER database. The TIMER database is a web resource which is quite smart in terms of systematic evaluations of the clinical outcomes of different immune cells in various types of cancers. The “Immune” module in TIMER was used to find the relation between genomic changes and immune infiltrates in TCGA. At the first step, we selected the “Gene” module and entered in JMJD5 for gene expression, and secondly, enter immune infiltrating cell (B cells, CD4${}^{+}$ T cells, CD8${}^{+}$ T cells, macrophages, neutrophils and myeloid dendritic cells) separately, then a heatmap with numbers was generated to show the Spearman’s Rho through various cancer types adjusted by tumor purity. Once the interested cell on the heatmap was selected, a scatter plot was created to show the relationship between the JMJD5 expression and infiltrate estimation value. Furthermore, we selected the tumor “Purity” to adjust our analysis as most immune cell types have negative correlation with tumor purity. Partial Spearman’s association analysis was used to determine the correlation coefficient.

Patients with BRCA, LIHC, LUAD, LUSC and STAD were divided into four groups as follows: (1) low JMJD5 expression + low tumor-infiltrating immune cells; (2) low JMJD5 expression + high tumor-infiltrating immune cells; (3) high JMJD5 expression + low tumor-infiltrating immune cells; and (4) high JMJD5 expression + high tumor-infiltrating immune cells. A Cox proportional hazards model was used to draw Kaplan–Meier plots for JMJD5 expression and immune infiltrates to visualize the survival differences. The expression of JMJD5 and six immune infiltrates was divided into low and high levels by 50%. p values of the log-rank test for comparing survival curves of four groups (2 vs. 1 and 4 vs. 3) are shown in each plot.

The data were analyzed with GraphPad Prism 5 and presented as the means $\pm$SD. Statistical significance was calculated with a t-test. The correlation of JMJD5 expression and immune infiltrates was calculated through partial Spearman’s correlation analysis. p 0.05 was treated as statistically significant.

We investigated the JMJD5 mRNA expression in various cancers using TIMER. Our results presented that the JMJD5 mRNA expression was significantly higher in PRAD (**) and STAD (***) but lower in BRCA (*), CHOL (***), colon adenocarcinoma (COAD) (***), kidney renal clear cell carcinoma (KIRC) (*), LIHC (***), LUAD ($\cdot$), LUSC (***) and uterine corpus endometrial carcinoma (UCEC) (***) by comparing each tissue with its normal one (Fig. 1A).

To confirm JMJD5 protein expression and estimate its clinical significance in cancers, we investigated the JMJD5 protein expression in TMAs using IHC. It is now clear the JMJD5 protein expression was significantly higher in bladder urothelial carcinoma (BLCA) (*), cervical squamous cell carcinoma (CESC) (*), ovarian cancer (OV) (*), pancreatic invasive ductal carcinoma (PAAD) (*), PRAD (**) and STAD (***) compared to the respective normal tissues (Fig. 1B) but was significantly lower in BRCA (*), CHOL (*), KIRC (**), LIHC (*), LUAD (**) and LUSC (*) compared to the respective normal tissues (Fig. 1C). The JMJD5 protein level was higher in COAD than in the respective normal tissue, but there was no statistically significant difference (Fig. 1B). The JMJD5 protein level was lower in UCEC compared to the respective normal tissue, but there was no statistically significant difference (Fig. 1C). IHC staining showed that JMJD5 was localized in different parts of tumor cells, including nuclei, cytoplasm or both, as follows: only in nuclei in BRCA and LUAD (Fig. 1C); only in cytoplasm in COAD, OV, PAAD, PRAD, STAD, CHOL, KIRC, LIHC, LUSC and UCEC (Fig. 1B,C); and in both nuclei and cytoplasm in BLCA and CESC (Fig. 1B). Additionally, we found that the protein expression and localization of JMJD5 varied according to the different pathological types of tumors. For example, the protein level of JMJD5 was lower in LUC (LUAD, LUSC and large cell carcinoma (LULCC)) compared to the normal tissues; however, there was an obvious difference in LUAD (**) and LUSC (*) (Fig. 1C) but no obvious difference in LULCC (data not shown). JMJD5 protein expression was observed only in nuclei in LUAD (Fig. 1C), only in the cytoplasm in LUSC (Fig. 1C) and in both nuclei and cytoplasm in LULCC (data not shown).

Fig. 1.

Human mRNA and protein expression of JMJD5 in various tumor tissues compared to normal tissues. (A) Box plots showing the distributions (median, spread and outliers) of the JMJD5 mRNA levels (log2 TPM) by RNA-seq data as displayed in gray columns when normal data were available. The number of samples is shown at the bottom. p value significance is indicated as follows: 0 $\le$ *** 0.001 $\le$ ** 0.01 $\le$ * 0.05 $\le$$\cdot$ 0.1. (B) The protein expression of JMJD5 was significantly higher in BLCA, CESC, OV, PAAD, PRAD and STAD than in the respective normal tissues. (C) JMJD5 protein expression was significantly lower in GHOL, KIRC, LIHC, LUAD and LUSC compared to the respective normal tissues. The quantification of IHC results was transformed to the mean optical density measured by Image-Pro Plus 2.0. Bars show the means $\pm$ SD. Statistical differences were noted as *p 0.05, **p 0.001 and ***p 0.0001. Original magnification: 20$\times$. Scale bar: 100 $\mu$m.

We next determined whether the JMJD5 expression has any effect on the prognosis of cancer patients. Among the cancers (BLCA, BRCA, CESC, OV, PAAD, PRAD, STAD, CHOL, KIRC, LIHC, LUAD and LUSC) with significantly different expression of JMJD5, we identified JMJD5 as a prognostic marker in BRCA (n = 3955, RFS: HR = 0.75, 95% CI from 0.67 to 0.83, log-rank p = 1.4 $\times$ 10${}^{-7}$) (Fig. 2A), LIHC (n = 364, RFS: HR = 0.6, 95% CI from 0.42 to 0.85, log-rank p = 0.0033) (Fig. 2B), LUC (n = 1927, RFS: HR = 0.81, 95% CI from 0.71 to 0.92, log-rank p = 0.001) (Fig. 2C) and STAD (n = 881, RFS: HR = 1.25, 95% CI from 1.05 to 1.49, log-rank p = 0.011) (Fig. 2D). However, the expression of JMJD5 had no correlation with the survival rate in other types of cancer (BLCA, CESC, CHOL, KIRC OV, PAAD and PRAD) (Supplementary Fig. 1), except for the above four types of cancer (BRCA, LIHC, LUC and STAD). These results indicated that BRCA, LIHC and LUC patients with significantly higher expression of JMJD5 had an improved survival rate. In contrast, STAD patients with low JMJD5 expression were associated with a better survival rate. Thus, the expression of JMJD5 impacts RFS and may serve as a prognostic marker in BRCA, LIHC, LUC and STAD.

Fig. 2.

Kaplan–Meier survival curves comparing the low and high expression of JMJD5 in BRCA, LIHC, LUC and STAD patients. (A) RFS of BRCA. (B) RFS of LIHC. (C) RFS of LUC. (D) RFS of STAD. The red curve indicates patients with high JMJD5 expression, while the black curve indicates patients with low JMJD5 expression. p 0.05 was supposed to have statistical difference.

We focused on and analyzed the correlation between JMJD5 expression and the quantity of six infiltrating-immune cells (B cells, CD4${}^{+}$ T cells, CD8${}^{+}$ T cells, neutrophils, macrophages and myeloid dendritic cells) in BRCA, LIHC, LUAD, LUSC and STAD. Our results demonstrated that in BRCA, JMJD5 expression significantly negatively correlated with tumor purity and neutrophil infiltration and positively correlated with the infiltration of B cells, CD8${}^{+}$ T cells and macrophages but has no relationship with CD4${}^{+}$ T cells and myeloid dendritic cells (Fig. 3A). The JMJD5 expression in LIHC had no relationship with tumor purity or infiltration of CD4${}^{+}$ T cells, macrophages, neutrophils and myeloid dendritic cells, but it was significantly negatively correlated with infiltrated B cells and significantly positively correlated with CD8${}^{+}$ T cells (Fig. 3B). In LUAD, the expression of JMJD5 significantly negatively correlated with tumor purity, and positively correlated with the infiltration of B cells, CD4${}^{+}$ T cells and myeloid dendritic cells but had no relationship with CD8${}^{+}$ T cells, macrophages or neutrophils (Fig. 3C). The JMJD5 expression in LUSC significantly negatively correlated with tumor purity and positively correlated with the infiltration of B cells, CD4${}^{+}$ T cells, CD8${}^{+}$ T cells, neutrophils and myeloid dendritic cells, but had no relationship with macrophages (Fig. 3D). The JMJD5 expression in STAD had no relation with tumor purity but was significantly positively correlated with the infiltration of B cells, CD4${}^{+}$ T cells, CD8${}^{+}$ T cells, macrophages, neutrophils and myeloid dendritic cells (Fig. 3E). In conclusion, these results suggested that the JMJD5 expression strongly correlates with immune cell infiltration in BRCA, LIHC, LUAD, LUSC and STAD.

Fig. 3.

Correlation of JMJD5 expression with six tumor-infiltrating immune cells in BRCA, LIHC, LUAD, LUSC and STAD. (A) In BRCA, JMJD5 expression significantly negatively correlated with tumor purity and neutrophil infiltration and positively correlated with the infiltration of B cells, CD8${}^{+}$ T cells and macrophages but has no relationship with CD4${}^{+}$ T cells and myeloid dendritic cells. (B) In LIHC, the JMJD5 expression had no relationship with tumor purity or infiltration of CD4${}^{+}$ T cells, macrophages, neutrophils and myeloid dendritic cells, but it had a significant negative correlation with infiltrated B cells and significant positive correlation with CD8${}^{+}$ T cells. (C) In LUAD, the JMJD5 expression had a significant negative correlation with tumor purity and a positive correlation with the infiltration of B cells, CD4${}^{+}$ T cells and myeloid dendritic cells as well as no relation with CD8${}^{+}$ T cells, macrophages or neutrophils. (D) In LUSC, the JMJD5 expression significantly negatively correlated with tumor purity and positively correlated with the infiltration of B cells, CD4${}^{+}$ T cells, CD8${}^{+}$ T cells, neutrophils and myeloid dendritic cells as well as no relation with macrophages (E) In STAD, the expression of JMJD5 had no relationship with tumor purity but a significantly positively correlated with the tumor-infiltrating B cells, CD4${}^{+}$ T cells, CD8${}^{+}$ T cells, macrophages, neutrophils and myeloid dendritic cells. p 0.05 was considered as significant.

The Kaplan–Meier plots showed that the survival rate of low JMJD5 expression patients with higher B cell tumor infiltration were much better than those with lower B cell tumor infiltration in BRCA (n = 1100, OS: HR = 0.658, p = 0.0487) (Fig. 4A; 2 vs. 1) and LUAD (n = 515, OS: HR = 0.67, p = 0.039) (Fig. 4B; 2 vs. 1). The survival rate of high JMJD5 patients with lower macrophage tumor infiltration were much better than those with higher macrophage infiltration in LIHC (n = 371, OS: HR = 1.72, p = 0.0158) (Fig. 4C; 4 vs. 3). The survival rate of low JMJD5 expression patients with lower macrophage infiltration is much better than those with higher macrophage infiltration in STAD (n = 415, OS: HR = 2.26, p = 0.00035) (Fig. 4D; 2 vs. 1). In summary, the combination of JMJD5 expression with either B cell tumor infiltration or macrophage tumor infiltration may serve as a new tumor prognostic marker.

Fig. 4.

Kaplan–Meier plots for expression level and immune cell infiltrates to visualize the survival differences in BRCA, LUAD, LIHC and STAD. (A and B) The survival rate of low JMJD5 expression patients with higher B cell tumor infiltration were much better than those with lower B cell tumor infiltration in BRCA (A) and LUAD (B). (C) LIHC patients with high expression of JMJD5 had an improved survival rate with lower macrophage tumor infiltration compared to those with higher macrophage tumor infiltration. (D) The survival rate of low JMJD5 expression patients with lower macrophage infiltration is much better than those with higher macrophage infiltration in STAD.