Determining PTEN functional status by network component deduced transcription factor activities

Abstract

PTEN-controlled PI3K-AKT-mTOR pathway represents one of the most deregulated signaling pathways in human cancers. With many small molecule inhibitors that target PI3K-AKT-mTOR pathway being exploited clinically, sensitive and reliable ways of stratifying patients according to their PTEN functional status and determining treatment outcomes are urgently needed. Heterogeneous loss of PTEN is commonly associated with human cancers and yet PTEN can also be regulated on epigenetic, transcriptional or post-translational levels, which makes the use of simple protein or gene expression-based analyses in determining PTEN status less accurate. In this study, we used network component analysis to identify 20 transcription factors (TFs) whose activities deduced from their target gene expressions were immediately altered upon the re-expression of PTEN in a PTEN-inducible system. Interestingly, PTEN controls the activities (TFA) rather than the expression levels of majority of these TFs and these PTEN-controlled TFAs are substantially altered in prostate cancer mouse models. Importantly, the activities of these TFs can be used to predict PTEN status in human prostate, breast and brain tumor samples with enhanced reliability when compared to straightforward IHC-based or expression-based analysis. Furthermore, our analysis indicates that unique sets of PTEN-controlled TFAs significantly contribute to specific tumor types. Together, our findings reveal that TFAs may be used as "signatures" for predicting PTEN functional status and elucidate the transcriptional architectures underlying human cancers caused by PTEN loss.

Figure 1. PTEN re-expression regulates transcription factor expression and activity in PTEN-inducible Pten^{Δloxp/Δloxp} MEFs.

(A) Schematic illustration of rationale and approach used in this study. To identify PTEN-controlled TFs, their activities (TFAs) in PTEN inducible system were first derived from expression of their target genes by NCA. The perturbed TFs were then examined further in mouse models and human cancers. (B) Immunoblot showing PTEN expression levels at 0, 1, and 2 days after treatment with the indicated concentration of Doxycycline in Pten^{Δloxp/Δloxp} MEFs. Isogenic WT cells (Pten^L/L) were used as a positive control. (C) Heatmaps showing the changes of expression and activity (TFA) of transcription factors in fold and log₁₀ transformed p-value of the z-test, respectively, caused by PTEN re-expression for 1 day (1/0) or 2 days (2/0).

Figure 2. PTEN re-expression downregulates activities of c-MYC and LEF1 in PTEN-inducible PC3 cells.

(A) Immunoblot showing PTEN expression levels under Doxycycline induction. (B) Bar graphs showing fold changes of c-MYC and LEF1 mRNA expression by qPCR analysis. (C and D) Bar graphs showing the target gene expression of c-MYC and LEF1 by qPCR analysis, respectively. *p<0.05 and **p<0.01.

Figure 3. PTEN-regulated TFAs are significantly increased in murine prostate cancer models in vivo.

(A) Heatmap showing changes of PTEN-regulated TFAs in PTEN inducible MEFs (PTEN null compared to PTEN re-expression or PTEN WT) and murine prostate cancer models (compared to WT control mice; Rapa: Rapamycin treatment). TFAs regulated by PTEN/AKT/mTOR pathway are marked in bold. TFAs exhibit discordant regulation between c-Myc and the PTEN/AKT/mTOR pathway are marked by *. The purple and green asterisks indicate Myc-activating and suppressing TFs respectively. (B) Triangle diagram summarizing the TFAs regulated by PTEN, AKT/mTOR and/or c-MYC.

Figure 4. PTEN-controlled TFAs predict PTEN status in human cancers.

Unsupervised clustering analysis, based on PTEN-controlled TFAs, was used to classify human tumor samples. (A) In prostate cancer, group 1 is largely composed by samples with PTEN copy number changes (CN, red) and lymph node metastases (LN met, pink); Group 2 are primary cancer samples (light blue) with normal PTEN karyotype (blue) that are separated from most of normal prostate tissues (white). TFAs that are significantly altered between group 1 and group 3 are mark by **, p<0.001. The heatmap was plotted based on relative changes to the respective average TFAs of normal samples. (B) In breast cancer, group 1 is mostly comprised of samples with PTEN-negative status (red) identified by immunohistochemistry (IHC). The majority of the samples in group 3 have positive PTEN status (blue), while group 2 includes both positive and negative PTEN samples. (C) In brain tumors, most samples in group 1 are associated with PTEN negative status (red). The PTEN negative subgroup is also correlated with higher tumor grade (green for grade 3 and purple for 4, respectively).

Figure 5. Enhanced robustness of TFA-based signatures in predicting PTEN status in human cancer.

(A) The Kaplan-Meier survival curves of patients with brain tumors stratified according to PTEN-controlled TFA and IHC analyses. (B) Log₁₀ transformed p-values of the χ² test evaluating the association of PTEN status with the hierarchical clustering-determined groups of human tumors. Clustering results are based on PTEN-controlled TFAs (red; Figure 3), prostate cancer-related TFAs (blue), TFA-based (gold) and gene expression-based (green) signatures derived from PTEN IHC data in breast tumors. When three major clusters are observed in prostate and breast cancers, the χ² tests are performed to associate different PTEN status in group 1 and groups 2 plus 3.

Figure 6. Subsets of PTEN-controlled TFAs preferentially function in specific types of tumors.

(A) t-test p-values comparing TFAs of the subgroups based on PTEN status and clustering results as shown in Figure 5 in three tumor types. In the t-tests performed on of prostate and breast cancers, PTEN positive samples in group 3 were used as the PTEN positive functional status, the PTEN negative samples in group 1 as the PTEN negative functional status. Similarly, in brain tumor PTEN IHC positive samples in group 2 and PTEN IHC negative samples in group 1 were selected for representing PTEN positive and negative functional status respectively. The red line highlighted the 6 TFAs significantly (p<0.05) altered between tumor subgroups in three tumor types. (B) Venn diagram summarizing the overlap of the TFAs that contribute to the discrimination of tumor subgroups with different PTEN status in different tumor types. (C–D) Heatmap of the absolute Pearson correlation coefficients between NCA-inferred TF activity profiles across the tumor samples from prostate (C), breast (D) and brain (E) cancers, indicating groups of co-active transcription factors may function together.