The Role of Osteopontin on Prostatic Fibrosis and Inflammation

Published on January 19th, 2022 by Axel Martinelli

For this post, I will re-analyse the transcriptomic data from a recently published article on the role of osteopontin on prostatic fibrosis and inflammation (Popovics et al, 2021). Fibrosis occurs as a consequence of injury or damage to internal organs and can interfere with the proper functioning of an organ. Osteopontin has pro-inflammatory and pro-fibrotic properties and was shown to be particularly abundant in the prostate of patients suffering from  Lower Urinary Tract Symptoms (LUTS). To evaluate the role of osteopontin in prostate inflammation and fibrosis, the authors infected wild type (WT) and osteopontin KO mice with E. coli to induce an infection in the prostate. They observed that KO mice had lower levels of inflammation and fibrosis, as well as no urinary dysfunction, compared to WT mice. RNAseq sequencing further confirmed their observation, with a noticeable reduction in the expression of inflammatory and fibrotic biomarkers (such as Mmp3) in infected KO mice compared to infected WT mice.

Materials and Methods

The read counts from the experimental RNAseq data were obtained from the GEO repository (GSE179655) and injected into the Bigomics Analytics platform. The 12 samples were organised into four groups: Uninfected WT mice (WTU), infected WT mice (WTEC), uninfected KO mice (KOU) and infected KO mice (KOEC). 

Geneset clustering was performed using the covariance based on the log CPM (counts per million) of individual component genes and coloured by the standard deviation across samples.

For the differential gene expression analysis, the DESEQ2 algorithm with parameters FDR<0.05 and logFC> |0.5| was used. For the GO term  analysis, the fgsea algorithm with parameters FDR<0.05 and logFC> |0.5| was used. L1000 Drug connectivity map analysis used an FDR<0.05.


For the clustering analysis I opted for a PCA plot to replicate one of the figures in the article. As in the published plot, WTEC samples (blue stars) partially clustered separately from the other samples (Fig. 1). Grouping samples by phenotype and infection in a heatmap consisting of the top 50 dysregulated genes indicated three main gene clusters, namely S1, S2 and S4,  that distinguished WTEC samples from the rest (Fig. 2A). Functional annotation of the gene clusters with the Hallmark collection revealed that the upregulated clusters S2 and S4 contained datasets linked to Inflammatory and stress responses. Meanwhile, cluster S1 contained downregulated genes including SPINK5, which is thought to have anti-inflammatory properties, and various secretoglobins (sgbs), which play a role in modulation of inflammation, tissue repair and tumorigenesis (Jackson et al, 2011) and thus are highly relevant in the process of fibrosis (Fig. 2B).

Figure 1. PCA plot annotated by phenotype. WTU: uninfected wild type, WTEC: infected wild type, KOU: uninfected KO, KOEC: infected KO.

Figure 2. (A) Sample cluster heatmap by phenotype of the top 50 dysregulated genes. (B) Functional annotation of the four main clusters. WTU: uninfected wild type, WTEC: infected wild type, KOU: uninfected KO, KOEC: infected KO.

 I also used a new approach for annotating clusters recently implemented in the Omics PlayGround platform that produces UMAP plots based on the logCPM or logFC at the gene or geneset level.  The geneset level analysis (spanning >50,000 gene sets) indicates a strong level of divergence in a group of gene sets, highlighted by a black box in Fig. 3A. When comparing the same area across sample groups, a distinct difference between WTEC and the other groups (including the infected KO mice, KOEC) was observed (Fig. 3B). A close-up of the region indicates the presence of many Hallmark gensets related to inflammation, as well as genesets related to fibrosis, such as epithelial–mesenchymal transition and angiogenesis (Fig. 3C).

Figure 3. Geneset clustering. (A) UMAP clustering of gene sets coloured by standard-deviation (red= high, blue=low). Proximity indicates high covariance. A black box highlights the region of high variability between phenotype groups.(B) UMAP clustering of gene sets coloured by relative log-expression of the phenotype group (red= high, blue=low). Proximity indicates high covariance. (C) Close-up of the region highlighted in A with annotation of the most variable Hallmark gene sets.

Differential Gene Expression

For the differential gene expression Analysis, I focused on two pairwise comparisons, namely WTEC vs WTU and KOEC vs WTEC, and selected two gene families that had been described as dysregulated (collagens and ccl chemokines), as well as gene cyp2b10. The figure shows collagens (Fig. 4A) and chemokines (Fig. 4B) are upregulated in WTEC vs WTU, but downregulated in KOEC vs WTEC, while the reverse is true for cyp2b10 (Fig. 4C). This reflects the main observations in the original article.

Figure 4. (A) Gene expression levels (in logFC) of various collagen genes in the WTEC vs WTU and KOEC vs WTEC comparisons. (B) Gene expression levels (in logFC) of various ccl chemokines genes in the WTEC vs WTU and KOEC vs WTEC comparisons. (C) logFC expression summary barplot for gene cyp2b10 across all pairwise comparisons; the WTEC vs WTU and KOEC vs WTEC comparisons are highlighted by a red box.

Gene Set Enrichment and Drug Connectivity Map  Analysis

Analysis of GO terms across the pairwise comparisons in the dataset indicated, as expected, the upregulation of inflammatory and pro-fibrotic and the downregulation of cholesterol biosynthesis pathways following bacterial infection in the WT but not in the KO cells (Fig. 5A).

Figure 5. Activation matrix of the top 50 dysregulated GO terms across the pairwise comparisons in the experimental dataset. Size of the dots indicates statistical power, while colour indicates upregulation (red) or downregulation (blue).

As osteopontin inhibition looks like a promising target for intervention against LUTS, I consulted the L1000 Drug connectivity map to identify existing drugs that could mimic the gene expression profile generated by the suppression of osteopontin in the KOEC vs WTEC pairwise comparison. The analysis indicated that histone deacetylase (HDAC) inhibitors, estrogen receptor agonists and antagonists and EGFR inhibitors were the most significantly correlated modes of actions to replicate the suppression of osteopontin at a gene expression level (Fig. 6A). 

A quick search in the available literature revealed that estrogens have an antifibrotic effect and that their suppression exacerbates dermal fibrosis (Avouac et al, 2020). Meanwhile, EGFR signalling promotes renal fibrosis (Chen et al, 2012) and is a therapeutic target in lung fibrosis (Vallath et al, 2014). Furthermore, upregulation of osteopontin is correlated to susceptibility to anti-EGFR therapies in breast cancer (Anborgh et al, 2018), which could indicate a link between EGFR suppression and osteopontin-mediated fibrosis. Finally,  the anti-fibrotic potential of HDAC inhibitors has been recognised (Yoon et al, 2019). In particular, Panobinostat, which the platform identifies as a strong inhibitor of osteopontin-mediated alterations (Fig. 6B), has been tested as a drug against idiopathic pulmonary fibrosis (Korfei et al, 2018).

Figure 6. L1000 drug connectivity map analysis. (A)  Modes of action of the most frequent positively correlated drug profiles. (B) Enrichment plot of panobinostat experiments against the pairwise comparison KOEC vs WTEC indicates a significant positive correlation.


Through the Omics Playground platform, I was able to quickly replicate the main findings by Popovics and colleagues, in particular the dampening of pro-fibrotic and pro-inflammatory genes in mice with a knocked-out osteopontin gene. Furthermore, I was able to identify drugs that could mimic the inhibition of osteopontin and thus have therapeutic potential. The most common modes of actions (e.g. HDAC inhibitors) have been described in the literature as potential anti-fibrosis therapies, thus confirming the reliability of these findings.

Read also: