Background
A growing body of evidence links changes in metabolism to cancer [
1,
2]. In addition to the well-known shift of cancer cells to aerobic glycolysis, mutations or changes in the expression of metabolic enzymes have been identified as potential cancer drivers. Mutations and/or altered expression of metabolic enzymes such as succinate dehydrogenase, pyruvate kinase and isocitrate dehydrogenase are linked to tumor initiation, development and drug resistance [
3‐
6]. Changes in metabolite levels can affect expression profiles, epigenetic marks and chromatin organization in cancer, with resulting changes in cellular phenotypes, metastatic potential, as well as on the tumor microenvironment [
7].
The human ALDH family comprises 19 enzymes that catalyze NAD(P)+ dependent oxidation of aldehydes to their corresponding carboxylic acids and NAD(P)H [
8]. Notably, ALDH1 is thought to be oncogenic in breast cancer. Cells with high ALDH1 activity have been linked to poor outcome in some cancers [
9,
10], albeit not in others [
11,
12]. Evidence of the roles of other ALDH isoforms in cancer remains equivocal.
In this study, we provide evidence for a role of ALDH isoform 7A1 (ALDH7A1) in human cancer, and link this to regulation of PPAR activity. PPARs (Peroxisome proliferator-activated receptors) are ligand-activated transcription factors, regulated by cellular metabolites [
13,
14]. Metabolite-regulated control of PPAR activity contributes to cellular homeostasis through feedback regulation on the expression on enzymes involved in glucose, amino acid and lipid metabolism [
15]. Metabolic profiling showed that ALDH7A1-depletion reduced the levels of metabolites that serve as activating ligands for PPARs. Analysis of cancer RNAseq data from TCGA showed that low ALDH7A1 mRNA levels correlate with a low PPAR activity signature, and with poor survival prognosis in patients with hepatocellular carcinoma and renal clear cell carcinoma. Importantly, the cellular phenotypes associated with ALDH7A1-depletion, increased migration and invasiveness, were corrected by restoring PPAR activity. We hypothesize that metabolic changes resulting from low ALDH7A1 expression may be linked to clinical outcome through their effects on PPAR activity. PPARs are pharmaceutical targets for metabolic disorders including diabetes, dyslipidemia, obesity, chronic inflammation and atherosclerosis [
16,
17]. Immunohistochemical staining of clinical samples suggests that low ALDH7A1 expression may be a useful prognostic marker of poor clinical outcome for hepatocellular and renal clear cell carcinomas. Our findings suggest a route to identifying cancer patients who might benefit from PPAR agonist therapy.
Methods
Cells
Primary BJ cells were originally obtained from ATCC (Cat# ATCC® CRL-2522™). hTert-expressing BJ cells were engineered to express p53 and p16 shRNAs (4F3). These genetic modifications enable cells to migrate and invade well in migration and invasion assays. Cells were expanded to passage 5, and frozen. All subsequent experiments were performed using this parental polyclonal 4F3 cell line. BJ cells were tested for mycoplasma every 6 months and examined for consistent phenotype and behavior on an ongoing basis. Information on the other cell lines used in this study is provided in Additional file
1: Figure S7.
Viral transduction
Lentivirus particles were produced by calcium phosphate transfection of 293 T cells and harvested after 24 h–48 h using standard procedures. One to two passages after thawing, BJ-4F3, HUH7, CAKI2 cells were transduced with control shRNAs (Sigma: SHC001 as empty vector, SHC002 as non-targeting shRNA control) or ALDH7A1-specific shRNAs (Sigma: TRCN0000028424 (sh1) and TRCN0000028447 (sh2)) for 24 h, allowed to recover for 24 h, and placed under puromycin selection (2 μg/ml) for 6 days. Experiments were performed within the next 10 passages. All experiments were performed at least 3 times with independently transduced cells. Knockdown efficiency was assessed by quantitative RT-PCR (qPCR) (forward primer: CATGGCGTGAGTGAAGGAC, reverse primer: CAGGGCAATAGGTCGTAATAACC), and/or by immunoblotting of cell extracts using rabbit anti-ALDH7A1 (Sigma: HPA023296).
Quantitative qPCR
Total RNA was isolated with the “RNeasy Plus Mini Kit” following the manufacturer’s instructions. After DNase treatment (RQ1 RNase-Free DNase; Promega) a cDNA was synthesed using a SuperScript™ III First-Strand Synthesis System (Invitrogen) using 0.5–1 μg of total RNA. qPCR was carried out on a QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems) using HOT FIREPol® EvaGreen® qPCR Mix Plus (ROX) (SOLIS BIODYNE). Total RNA from each sample was normalized to β-ACTIN, KIF and TBP for the BJ and HUH7 cell line or KIF in the case of CAKI2 cell line. Significance was determined using the Mann–Whitney U test after adjusting for False Discovery Rate. The following primers were used: CYP27A1 (forward primer: GGTGCTTTACAAGGCCAAGTA, reverse primer: TCCCGGTGCTCCTTCCATAG), FABP3 (forward primer: TGGAGTTCGATGAGACAACAGC, reverse primer: CTCTTGCCCGTCCCATTTCTG), ACSL1 (forward primer: CTTATGGGCTTCGGAGCTTTT, reverse primer: CAAGTAGTGCGGATCTTCGTG), CPT2 (forward primer: CTGGAGCCAGAAGTGTTCCAC, reverse primer: AGGCACAAAGCGTATGAGTCT), ACOX1 (forward primer: ACTCGCAGCCAGCGTTATG, reverse primer: AGGGTCAGCGATGCCAAAC), FADS2 (forward primer: AATCAGCAGGGGTTTCAAGA, reverse primer: GGCACTACGCTGGAGAAGAT), APOA1 (forward primer: TTGCTGAAGGTGGAGGTCAC, reverse primer: TGGATGTGCTCAAAGACAGC), β-ACTIN (forward primer: GATGCGTAGCATTTGCTGCATGG, reverse primer: TGAGGCTAGCATGAGGTGTGTG), TBP (forward primer: CGCCGAATATAATCCCAAGC, reverse primer: TCCTGTGCACACCATTTTCC), KIF (forward primer: TTGCCTCCTTTGGCAACATTCG, reverse primer: ACACAGCACCAATACCCATGATAC).
BJ cells were treated with PPARα agonist (Ciprofibrate) or DMSO as a control. Cells were collected for RNA extraction and qPCR as described above. 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.
Cell culture
Unless specifically mentioned, all cell lines were cultured in high glucose DMEM (Dulbecco’s Modified Eagle Medium; Sigma-Aldrich) with 10% Fetal Calf Serum (Sigma) and 1% Penicillin-Streptomycin (Sigma), 1% GlutaMAX™-I (Gibco) and 1% pyruvate. Cells were cultured at 37 °C in a humidified environment containing 5% CO2.
Phenotypic assays
Cell proliferation assays were performed by plating BJ-4F3 cells at a density of 2 × 104 cells/cm2 in triplicate wells. Cells were grown for 3, 24, 48, 72 or 96 h and then fixed with 4% formaldehyde (Sigma-Aldrich). The number of DAPI-stained nuclei was counted in representative images of each well, at each time point. Data are presented as the fold change in cell number over time (± standard error of the mean).
Wound healing assays were performed by plating transduced BJ-4F3 cells (at 4 × 104 cells/cm2), HUH7 cells (7 × 104 cells/cm2) and CAKI2 cells (5 × 104 cells/cm2). Cells were allowed to form a monolayer for 24 h. A stripe was cleared by dragging a pipet tip across the surface of the plate, and the culture medium was changed to wash away floating cells. The initial state was recorded by taking 2–4 images at defined places (4x magnification; t = 0). Cells were allowed to migrate for 24 h and images were taken of the same regions. The area devoid of cells was measured and the average migrated distance calculated.
Invasive migration assays (transwell) were performed using transduced BJ-4F3 cells, that had been serum-starved for 24 h. Matrigel invasion chambers with 8.0 μm Polyethylene Terephthalate membranes were used according to manufacturer’s protocol (Fisher Scientific, #11553570). Complete DMEM supplemented with 20% FCS was used as attractant at the bottom of the well. 5 × 104 cells were seeded on top of the Matrigel in serum-free complete DMEM. After 24 h the chamber was washed once with PBS and cells were fixed with 4% formaldehyde. Nuclei were stained with DAPI and counted to determine the number of cells in the upper invasion chamber. The inside of the chamber was then cleared and the cells that had migrated through the gel were counted. Ten pictures were taken per chamber at 10x magnification. The total number of cells in the invasion chamber was used for normalization. Cell number and migrated distance were measured with ImageJ Fiji software.
Cells were treated with Ciprofibrate, GW501516, or Rosiglitazone at the concentration indicated in the figures at t = 0 of the scratch assay, and at seeding time in the invasion assays (both chambers).
Expression, correlation and survival analysis
Statistical analysis was performed using R Software. For mRNA expression, significance was determined using the Mann–Whitney U test. To calculate comparison between multiple groups, pairwise Wilcoxon test with Bonferroni correction for multiple testing was applied. For overall survival analysis, cancer patients were divided into three equally sized groups based on ALDH7A1 mRNA expression levels (low, middle, high). Cox proportion hazard regression models were used to calculate p-values between groups.
For Additional file
1: Figure S4 patients were divided into two equally sized groups based on ALDH7A1 expression, EGFR expression and sum of scaled and centered relative protein levels EGFR_pY1068 (CST; 2234), EGFR_pY1173 (Abcam; ab32578). Hazard Ratio for low ALDH7A1 expression and associated
p-value was calculated in EGFR low and high groups separately. For correlation analysis Spearman coefficients and corresponding
p-values were calculated between ALDH7A1 RNA expression and EGFR RNA expression, and the sum of scaled and centered phosphorylated EGFR protein levels.
Pathway and gene set enrichment analysis
The R/bioconductor package limma [
19] was used to identify genes differentially expressed between the top and bottom third ALDH7A1 expression groups. Data were filtered using RSEM > 10 in at least in 33% of samples to reduce noise from low expressed transcripts. Genes with log
2-fold change +/− 0.4 with adjusted
p-value threshold < 0.05 were defined as differentially expressed between groups. All genes not eliminated by filtering were used to define the “gene universe” for pathway enrichment analysis.
The following algorithm packages were used for analysis: SPIA [
20], CEPA [
21], GRAPHITE [
22], PIANO [
23], GAGE [
24], ESEA [
25]. The following databases were employed: REACTOME (
http://reactome.org/), BIOCARTA (
http://www.biocarta.com/), please note that the biocarta server is not available anymore. NCI (
http://www.ndexbio.org/#/), KEGG (
http://www.genome.jp/kegg/) [
26,
27], MSigDB (
http://software.broadinstitute.org/gsea/index.jsp) (H: hallmark gene sets, CP:BIOCARTA: BioCarta gene sets, CP:KEGG: KEGG gene sets, CP:REACTOME: Reactome gene sets). Unless otherwise specified pathway databases included in these packages were used. For SPIA analysis pathways were downloaded directly from KEGG. For GAGE and PIANO, annotation sets were downloaded from the Molecular Signatures Database v5.2 (
http://software.broadinstitute.org/gsea/msigdb).
After SPIA analysis with the KEGG annotation database, pathways and biological processes most likely to be affected were selected after filtering results by criteria - pG < 0.05, NDE > 3. pG represents the combined
p-value from gene enrichment and probability of perturbation accumulation in the pathway and NDE represents differentially expressed genes per pathways. The same criteria were applied for Graphite “runSpia” analysis with the Reactome, Biocarta and NCI annotation databases. For Piano gene set enrichment analysis, an adjusted
p-value of < 0.05 for up and down regulated gene sets was set as filtering criterion. Minimum and maximum number of genes per set was defined as 3 and number of DE/5. Piano analysis was run using MSigDB “Hallmark gene sets” (h.all.v5) and “CP:BIOCARTA: BioCarta gene sets”, “CP:KEGG: KEGG gene sets”, “CP:REACTOME: Reactome gene sets” annotation sets. In case of CEPA pathway analysis, affected pathways were selected if 3 out of 6 (equal.weight, in.degree, out.degree, betweenness, in.reach, out.reach) statistics were
p-value < 0.05 for all annotation databases used. For GAGE analysis log
2-fold change for all genes after filtering were used and a
p-value < 0.05 was set as filtering criterion for the results. For ESEA analysis, the expression matrix of all genes after filtering was used. NOM
p-value was used as significant criterion for Gain-of-correlation and Loss-of-correlation result filtering. Affected pathways and biological processes that were not detected at least by 2 different methods with the same annotation database were filtered out. Only changes that occurred in both LIHC and KIRC patients with low ALDH7A1 expression were kept. KEGG pathway maps were rendered with “Pathview” [
28].
1H NMR spectrometry
Twelve control and twelve ALDH7A1-depleted cell samples from 3 independently transduced polyclonal cell lines were analyzed in duplicate. Samples were extracted in chloroform-methanol-water [
29]. The aqueous supernatant was lyophilized and stored at − 80 °C. Immediately before measurement, samples were rehydrated in 200 μl of 50 mM phosphate buffer (pH 7.4) in D
2O, and 180 μl was transferred to a 3 mm NMR tube. The buffer contained the chemical shift reference 3-(trimethylsilyl)-propionic acid-D4, sodium salt and NaN
3.
NMR measurements were performed at 25 °C on a Bruker Avance III HD 800 spectrometer, operating at a
1H frequency of 799.87 MHz, equipped with a 3 mm TCI cold probe.
1H NMR spectra were acquired using a single-90°-pulse experiment with a Carr-Purcell-Meiboom-Gill (CPMG) delay added, in order to attenuate broad signals from high-molecular-weight components. The total CPMG delay was 194 ms and the spin-echo delay was 4 ms. The water signal was suppressed by excitation sculpting. A total of 128 transients of 32 K data points spanning a spectral width of 20 ppm were collected, corresponding to a total experimental time of 6.5 min. The spectra were processed using iNMR (
http://www.inmr.net). An exponential line-broadening of 0.5 Hz was applied to the free-induction decay prior to Fourier transformation. Spectra were referenced to the TSP signal at − 0.017 ppm, automatically phased and baseline corrected.
Drifting baseline of NMR spectra was corrected using the “rollingBall” algorithm. Spectra from BJ-4F3 cells were normalized against total intensity by dividing each intensity value by the sum of all intensity values. This method was chosen since total concentration of metabolites should be comparable across all samples. However, in our case spectra contained large peaks with significant variation between control and ALDH7A1 down-regulated cells, which could drastically affect total intensity values. Therefore, spectra were normalized against total intensity of a spectral region (above 4) that does not contain large peaks with significant variation. The “CluPA” algorithm was used to align peaks. Baseline correction, normalization and peak alignment was done using R package “ChemoSpec” (
https://cran.r-project.org/web/packages/ChemoSpec/). Principal Component Analysis with “pareto” scaling was performed using R package “muma” [
30]. One sample was excluded from analysis due to technical problems. Two samples were defined as outliers in the PCA analysis and were therefore also excluded.
To identify metabolites that are changed in ALDH7A1 depleted cells, the intensity values for signals above the baseline threshold defined as mean + 1SD of all intensity signals were compered. Non-parametric pairwise Wilcoxon-Mann Whitney U test with Benjamini-Hochberg correction for multiple testing was used to calculated p-values.
In the case of HUH7 and CAKI2 cells, 1H NMR spectra were processed and analyzed as above with minor adjustments. Six control and six ALDH7A1-depleted cell samples from 3 independently transduced cell lines were analyzed. 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”.
Gene expression clustering
PPAR transcriptional targets were selected from KEGG database (
http://www.genome.jp/kegg-bin/show_pathway?hsa03320). Low expressed genes were filtered out. Unsupervised hierarchical clustering analysis was applied to cluster LIHC and KIRC patient normal and tumor tissues into groups based on median centered log
2 PPAR target gene expression values. Control and tumor samples were clustered separately.
Immunohistochemistry
Liver and kidney cancer arrays presenting tumors and adjacent normal tissue biopsy samples were obtained from US Biomax (Rockville, MD, USA; HLiv-HCC180Sur-02, HLiv-HCC180Sur-03 and HKid-CRC180Sur-01). Additionally, 72 archival patient samples from the pathology department, Rigshospitalet Copenhagen were examined. Ethical approval was obtained from the Danish National Committee on Biomedical Research Ethics. Immunostaining was performed using rabbit anti-ALDH7A1 (Sigma: HPA023296) [
31] and the streptavidin–biotin peroxidase complex method according to the manufacturer’s instructions (UltraVision HRP DAB system, Thermo). Sections were examined by an experienced pathologist to confirm the tissue identity and assigned a score: 0 (no staining), 1 (weak staining up to 10% of tissue), 2 (weak staining 10–25% of tissue), 3 (weak to moderate staining ≥50% of tissue), 4 (moderate to strong staining of 50–75% of tissue) and 5 (moderate to strong staining > 75% of tissue). The score for each tumor was calculated by subtracting the score of the normal tissue from that of that tumor.
Multivariate regression analysis
Patients with complete set of information on survival time, status, stage and ALDH7A1 regulation were included in the study. Hepatocellular carcinoma patients with stage I (7) and stage IV (3) disease were excluded from multivariate analysis due to small sample size. We also excluded patients with hepatic cirrhosis. All covariates were tested for the proportional hazards assumption, and the multivariate Cox proportional hazards regression models were created using R package “Survival” (
https://cran.r-project.org/package=survival). Different models were compered by Likelihood ratio test and chi-square test. A forest plot was produced from the regression model with R package “forestmodel” (
https://cran.r-project.org/web/packages/forestmodel). Likelihood ratio test were used to calculate
p-values for the Kaplan Meier plots.
Conclusions
To date, little is known about the role of ALDH7A1 in cancer. Metabolic roles of ALDH7A1 include protecting cells from oxidative stress by metabolizing aldehydes derived from lipid peroxidation [
37], and protecting cells from osmotic stress by metabolizing betaine aldehyde to betaine, which serves as a cellular osmolyte [
38]. Our metabolic profiling data now links ALDH7A1 activity to the levels of activating ligands for the PPAR transcription factors. We have provided evidence that the effects of ALDH7A1 on cellular migration and on invasive behaviors is mediated through decreased PPAR activity. These data suggest a mechanism by which the ALDH7A1 activity can influence a wide range of metabolic pathways and cellular functions, with the potential to impact disease progression.
Literature on the role of ALDH7A1 in cancer suggests a somewhat complex scenario, with different outcomes in cancers of different tissue origin. In some reports, high ALDH7A1 has been linked to more severe disease. Positive ALDH7A1 protein staining correlates with increased cancer recurrence in non-small cell lung carcinoma [
39]. High ALDH7A1 protein expression has been reported in ovarian cancer, with highest expression in invasive ovarian cancer cells comparing to healthy ovarian epithelia [
40]. On the other hand, our analysis of 17 cancer types, using TCGA RNAseq data showed that ALDH7A1 expression was lower in several cancer types and that lower expression correlated with poor clinical outcome for HCC and renal ccRCC. ALDH7A1 activity impacts a number of metabolite pathways directly, and acts indirectly via PPARs on others. These metabolic shifts appear to impact different types of cancer differently. Why are liver and kidney cancer sensitive to the effects of low ALDH7A1? ALDH enzyme family members have distinct activities and substrate specificities. ALDH7A1 expression is high in the metabolically active kidney and liver tissues, whereas lung and prostate tissue express only low or moderate ALDH7A1 levels. ALDH7A1 is also to protect cells from osmotic stress. This might be important in liver and kidney, where turnover of osmolites such as betaine and glycerophosphocholine are tightly regulated. If high ALDH7A1 expression is important for liver and kidney homeostasis, it is tempting to speculate that low expression of this enzyme might contribute to cancer development in these tissues to a greater extent than in other tissue types.
PPARs as therapeutic targets for cancer
PPARs are ligand activated transcription factors that play an important role as regulators of metabolism and cellular homeostasis. PPARs are known to regulate fatty acid synthesis, uptake and storage, mitochondrial and peroxisomal fatty acid oxidation and ketogenesis, insulin sensitivity, glucose metabolism, drug metabolism and amino acid metabolism. In addition, PPARs have anti-inflammatory and immune suppressive functions. Given these wide-ranging effects on cellular metabolism and defense mechanisms, it may not be surprising that PPARs have been implicated as oncogenes in some cancer models and as tumor suppressors in others [
41,
42].
PPAR agonists are in clinical use for metabolic disorders and have been considered as cancer therapeutics. However, a number of safety concerns have been raised due to unwanted side effects and cancer development in rodent models [
17]. PPAR activators used as dietary supplements induced liver enlargement accompanied by oxidative stress in rats and mice [
43]. The PPARγ agonist Rosiglitazone was withdrawn due to increased risk of myocardial infarction [
44]. A meta-analysis found a modest but clinically significant increase in overall risk of bladder cancer upon long term treatment of another PPARγ agonist Pioglitazone [
45]. However, two large meta-analysis studies showed no statistically significant association between Fibrate (PPARα agonist) and cancer incidence [
46,
47]. To exploit the potential of PPARs as drug targets for cancer, we will require a more nuanced understanding of the role of specific PPAR isoforms in specific cancers, as well as means to identify patient groups who might benefit from therapeutics targeting PPARs.
Our studies reveal a striking three-way connection between low ALDH7A1 abundance, low PPAR activity and poor clinical outcome. Notably, patients with low PPAR activity also have low ALDH7A1 levels, suggesting a causal link between these two. This is likely due to the effects of ALDH7A1 on PPAR ligand levels, and is reflected experimentally by reduced PPAR target expression in cells depleted of ALDH7A1. Low PPAR activity is a useful predictor of poor clinical outcome, but it is difficult to measure in a clinical setting. We have provided evidence that scoring for ALDH7A1 levels by IHC may be a useful surrogate for PPAR activity to predict clinical outcome for patients with HCC and ccRCC.
Clinical trials are evaluating PPARα activation for treating non-alcoholic fatty liver disease and primary biliary cirrhosis, in combination with existing treatments (ClinicalTrials.gov identifier: NCT00262964, NCT00575042, NCT02823353, NCT02965911, NCT02823366). A trial evaluating the effect of the PPARα agonist Fenofibrate on patients with multiple myeloma (NCT01965834) is ongoing. However, to our knowledge, trials evaluating the effects of PPARα agonists on patients with HCC or ccRCC have not been started. We propose that selecting HCC or ccRCC patients according to ALDH7A1 IHC status might be a promising avenue for future study.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (
http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.