Metabolic control of PPAR activity by aldehyde dehydrogenase regulates invasive cell behavior and predicts survival in hepatocellular and renal clear cell carcinoma

Background Changes in cellular metabolism are now recognized as potential drivers of cancer development, rather than as secondary consequences of disease. Here, we explore the mechanism by which metabolic changes dependent on aldehyde dehydrogenase impact cancer development. Methods ALDH7A1 was identified as a potential cancer gene using a Drosophila in vivo metastasis model. The role of the human ortholog was examined using RNA interference in cell-based assays of cell migration and invasion. 1H-NMR metabolite profiling was used to identify metabolic changes in ALDH7A1-depleted cells. Publically available cancer gene expression data was interrogated to identify a gene-expression signature associated with depletion of ALDH7A1. Computational pathway and gene set enrichment analysis was used to identify signaling pathways and cellular processes that were correlated with reduced ALDH7A1 expression in cancer. A variety of statistical tests used to evaluate these analyses are described in detail in the methods section. Immunohistochemistry was used to assess ALDH7A1 expression in tissue samples from cancer patients. Results Depletion of ALDH7A1 increased cellular migration and invasiveness in vitro. Depletion of ALDH7A1 led to reduced levels of metabolites identified as ligands for Peroxisome proliferator-activated receptor (PPARα). Analysis of publically available cancer gene expression data revealed that ALDH7A1 mRNA levels were reduced in many human cancers, and that this correlated with poor survival in kidney and liver cancer patients. Using pathway and gene set enrichment analysis, we establish a correlation between low ALDH7A1 levels, reduced PPAR signaling and reduced patient survival. Metabolic profiling showed that endogenous PPARα ligands were reduced in ALDH7A1-depleted cells. ALDH7A1-depletion led to reduced PPAR transcriptional activity. Treatment with a PPARα agonist restored normal cellular behavior. Low ALDH7A1 protein levels correlated with poor clinical outcome in hepatocellular and renal clear cell carcinoma patients. Conclusions We provide evidence that low ALDH7A1 expression is a useful prognostic marker of poor clinical outcome for hepatocellular and renal clear cell carcinomas and hypothesize that patients with low ALDH7A1 might benefit from therapeutic approaches addressing PPARα activity. Electronic supplementary material The online version of this article (10.1186/s12885-018-5061-7) contains supplementary material, which is available to authorized users.

Experimental design: To generate adult-specific, and spatially restricted transgene expression, UAS-transgenes were placed under apterous-Gal4 control. Gal4 activity was inhibited during larval and pupal development using the temperature sensitive form of Gal80 ts , by rearing animals at the permissive temperature (18°C). Newly emerged adult flies were shifted to 29°C, inactivating the Gal80 inhibitor and allowing the apterous-Gal4 transgene to direct expression of UAS-EGFR, together with a UAS-GFP marker to label the tissue. A UAS-RNAi transgene targeting CG9629 was used to test tumor formation in the context of EGFR overexpression (left half). GFPexpressing tumors were found in 76% of animals expressing EGFR and the CG9629 RNAi transgene. None were found in flies expressing EGFR alone or the CG9629 RNAi transgene alone.
This assay system has been used to identify context-dependent tumor suppressors based on transgene expression during larval stages (Herranz et al., 2012Genes Dev 26, 1602-1611. Modification for use to screen for tumor formation in the adult will be described elsewhere (Kugler et al., in preparation)  (B) Heatmap of the correlation between ALDH7A1 mRNA expression and EGFR RNA and EGFR phosphorylation in all cancer types. Red: positive correlation coefficients, blue: negative correlation. Correlations with significant p-values are indicated. ALDH7A1 expression positively correlates with EGFR phosphorylation status in GBM, LIHC and kidney (KIRP, but not KIRC), and with EGFR mRNA levels in GBM and LIHC. There is weak negative correlation in colon (COAD) and thyroid (THCA) cancer.
(C) Cox proportional hazard regression analysis of the association between ALDH7A1 mRNA and EGFR levels for liver and kidney cancer. The box and horizontal lines represent the estimated Hazard Ratio (HR) and corresponding confidence interval. In the liver cancer dataset association between ALDH7A1 expression and poor survival outcome is dependent on EGFR status: survival was significantly worse for patients with low ALDH7A1 in a high EGFR expression/phosphorylation group while it was not significant in low EGFR group. This was not the case for the kidney cancer patients: ALDH7A1 expression was significantly associated with poor clinical outcome in both low and high EGFR groups according to EGFR RNA expression and low EGFR group according to EGFR phosphorylation levels.
Supplemental Figure S3A: Gene set and pathway analysis comparing low vs high ALDH7A1 tumors (A) For each method and corresponding annotation set, significantly affected pathways and biological processes where selected (p<0.05). Pathways and biological processes that were changed in both LIHC and KIRC patients with low ALDH7A1 are shown. ↑ -activated; ↓-inactivated pathways and biological processes; ↕-direction is not provided. In all three cell lines, lactose levels decreased and glucose levels increased. In Huh7 cells we see a reduction in glycerophosphocholine (GPC), phoshocholine and choline levels. We were not able to detected phosphocholine and glycerophosphocholine (GPC) in caki2 cells as it was below the detection level. The effects on amino acids were cell line dependent.
Huh7 and Caki2 1 H NMR spectra were processed and analyzed as described for BJ cells in the methods section. Briefly spectra were normalized against total intensity of a spectral region (above 1.5). "CluPA" algorithm was used to align peaks. "Rolling ball" algorithm (span -50) was applied to correct shifting baseline. Baseline correction, data binning (bin=4), normalization and peak alignment was done using R package "ChemoSpec".
Supplemental Figure S5: assessment of correlation between PPAR activity and ALDH7A! on other cancers (A) The "low activity", "intermediate" and "normal like" PPAR signature groups were identified as described for Figure 5. (B) Survival outcome was compared as described in Figure 5. The HNSC, LUSC, KIRP, BRCA patient groups with low PPAR activity did not exhibit significantly lower overall survival probability compared to "normallike" PPAR group for these cancers. BLCA showed worse survival for both the low and intermediate PPAR groups compared to the normal-like group.
(C) ALDH7A1 expression was assessed in the three PPAR activity groups, as described in figure 5. There was no correlation between low PPAR activity and low ALDH7A1 levels.
Supplemental Figure S6: effects of PPAR agonists (A) Immunoblots showing ALDH7A1 protein in BJ-4F3 cells treated with the PPAR agonists. The PPARα agonist ciprofibrate (Cipr) was used at 400 and 600 μM; the PPARβ agonist GW501516 (GW) was used at 60 and 80 μM; the PPARγ agonist rosiglitazone (Rosi) was used at 100 μM and 150 μM. Sh-1 and sh-2 show the effect of shRNA mediated depletion of ALDH7A1. Control 1 (C-1) indicates cells transduced with the empty vector. Control 2 (C-2) expressed a non-targeting shRNA Anti-actin was used to control for loading.
(B-C) Quantification of wound healing assays after 24h migration. Cells were treated with PPARβ and PPARγ agonist or DMSO as a control. The migrated distance was measured (µm), and averages from three independently transduced cell lines were calculated (± SEM) (D) Quantification of cell invasion through Matrigel over 24h. BJ-4F3 cells were treated with PPARβ agonist or DMSO as a control. The bar plots show the percent of cells that crossed the matrigel barrier (average of 3 independent experiments ± SEM). The two-tailed Mann Whitney test was used to calculate p-values.
(E-F) RT-PCR of PPAR transcriptional targets. Light grey -control cells transduced with the empty vector and non-targeting shRNA, accordingly. Black -ALDH7A1 depleted cells transduced with two independent shRNAs (sh-1 and sh-2). Data represent average ± standard error of the mean (SEM) from 3 independent experiments normalized to β-actin, kif1 and tbp (in the case of Huh7 cells) and kif1 (Caki2 cells). The twotailed Mann Whitney test with adjustment for False Discovery Rate was used to calculate p-values.
(G) RT-qPCR of PPAR transcriptional targets in cells treated with Ciprofibrate. Light greycontrol cells transduced with non-targeting shRNA and ALDH7A1 depleted cells transduced with shRNAs (sh-1). Cells were seeded and allowed to attach overnight and then treated with Ciprofibrate or DMSO. Cells were collected for RNA extraction and RT-qPCR. β-actin was used as normalization control. Friedman rank sum test with pairwise post-hoc test for multiple comparisons with holms adjustment was used to calculate p-values between groups with and without Ciprofibrate treatment.
Data represents average ± SEM from 2 independent experiments.

S7
: Assays on cancer cell lines Notes: 1) Scratch assays were performed as described in Figure  1. 2) "PPAR signature" was examined by qPCR for selected PPAR targets, as in Figure  5. Yes indicates changes in PPAR target expression. 3) "Rescue" indicates suppression of the cellular phenotype by treatment with the PPAR alpha agonist, as in Figure  6. 4) Hep3B cells detach very easily, so the scratch assay cannot be done. All other kidney cell lines we tested do not grow as a monolayer that lends itself to this kind of assay.