Post-translational regulation of metabolism in fumarate hydratase deficient cancer cells

Deregulated signal transduction pathways and energy metabolism are hallmarks of cancer and both play a fundamental role in the process of tumorigenesis. While it is increasingly recognised that signalling and metabolism are highly interconnected, the underpinning mechanisms of their co-regulation are still largely unknown. Here we designed and acquired proteomics, phosphoproteomics, and metabolomics experiments in fumarate hydratase (FH) deficient cells and developed a computational modelling approach to identify putative regulatory phosphorylation-sites of metabolic enzymes. We identified previously reported functionally relevant phosphosites and potentially novel regulatory residues in enzymes of the central carbon metabolism. In particular, we show that pyruvate dehydrogenase (PDHA1) enzymatic activity is inhibited by increased phosphorylation in FH-deficient cells. Our work provides a novel approach to investigate how post-translational modifications of enzymes regulate metabolism and could have important implications for understanding the metabolic transformation of FH-deficient cancers.


Introduction
Cancer is thought to arise from an abnormal accumulation of somatic mutations in the genome that drive complex and profound alterations of the cellular phenotype ( 1 ) . Among these changes, dysregulated energy metabolism is gaining importance as a hallmark of cancer ( 2 ) .
Although some recent work elucidated the genetic underpinning of these metabolic changes ( 3 , 4 ) , whether cancer metabolism is tuned via post-translational changes is still largely unknown.
In yeast, several studies have shown that signalling has a broad importance in regulating the activity of metabolic enzymes involved in central carbon metabolism and other peripheral pathways ( 5 -7 ) . By contrast, regulatory phosphorylation of metabolic enzymes in human cells remains largely uncharacterised.
A particularly well-studied metabolic alteration in cancer is driven by mutations of the metabolic enzyme fumarate hydratase (FH). These mutations cause Hereditary Leiomyomatosis and Renal Cell Cancer (HLRCC) tumours, a cancer syndrome characterised by benign tumours of the skin and uterus, and a very severe and aggressive form of renal cancer ( 8 ) . FH catalyses the conversion of fumarate to malate, a reaction that takes part in the tricarboxylic acid cycle (TCA cycle). FH mutations lead to the impairment of the catalytic activity of the enzyme and thereby to the accumulation of its substrate, fumarate, and to profound metabolic changes that we and others have extensively characterised ( 8 -10 ) . Yet, whether these metabolic changes are interconnected with upstream signalling processes has not been investigated.
Here, we performed an integrative analysis to investigate at a genome-scale level the regulatory interactions between metabolism and signalling using cell lines derived from an HLRCC tumor, UOK262, and the FH reconstituted counterpart, UOK262pFH, which we previously generated ( 11 ) . In particular, we characterised signalling and metabolic changes in HLRCC cell lines by designing and acquiring phosphoproteomics, proteomics and metabolomics measurements ( Figure 1). These data-sets allowed us to study the molecular adaptations of signalling and metabolism driven by the loss-of-function of FH in HLRCC using a computational framework that integrates phosphoproteomics with in silico estimated metabolic flux rates. Pairing the metabolomics modelling with phosphoproteomics measurements allowed us to identify putative-regulatory phosphorylation-sites in metabolic enzymes involved, mostly, in the central carbon metabolism. Notably, we experimentally validated that phosphorylation of pyruvate dehydrogenase E1 component subunit alpha (PDHA1) regulated metabolism in FH-deficient cells. In summary, we present a novel computational and experimental approach to systematically identify putative regulatory phosphorylation-sites in metabolic enzymes. This approach could reveal novel regulatory networks in the metabolic transformation of cancer, with important implications for cancer therapy.

Characterisation of the (phospho)-proteome of human FH-deficient cells
We started our investigation by characterising the proteome and the phosphoproteome of human FH-deficient UOK262 cells and their FH-complemented counterpart, which we previously generated ( 10 , 11 ) . Proteomics experiment covered a total of 1,468 unique proteins (Supplementary Table 1) and, in agreement with FH loss, FH was underexpressed in UOK262 cell lines ( Figure 1B). Consistent with previous results, vimentin (VIM) is a mesenchymal marker and was identified as a top-expressed protein ( 10 ) . Reproducibility of the measurements was assessed with unsupervised hierarchical clustering where replicates showed higher correlation coefficients than all the pairwise comparisons (Supplementary Figure 1). Proteomics showed agreement with the RNA-seq transcriptomics measurements available for the same cell lines (spearman's rho (r) of 0.43, p-value = 1.7e-63) ( 10 ) (Supplementary Figure 2A). Some proteins displayed a disagreement between the protein abundance and the transcript expression, reflecting different types of regulatory mechanisms occurring at post-transcriptional and post-translational levels, consistent with previous reports ( 11 ) .
To study post-translational modifications by phosphorylation we characterised the phosphoproteome of these cell lines. In total, we measured 1,360 unique single phosphorylated phosphosites, mapping to 812 unique proteins ( Figure 1C) (Supplementary Table 1). Similarly, to the proteomics measurements, VIM also showed a significant increase in phosphorylation in the UOK262 cell lines, although these changes are associated with the increase in protein abundance. Metabolic enzymes displayed significant changes between UOK262 and UOK262pFH, in particular PDHA1 and GAPDH ( Figure 1C). Specifically, 56% (23/41) of the phosphosites in metabolic enzymes display significant changes (FDR < 5%), and these map to 20 unique enzymes. This supports the idea that metabolic enzymes are regulated by phosphorylation in FH-deficient UOK262 cell lines.

Genome-scale metabolic modelling
To investigate how phosphorylation of metabolic enzymes could regulate metabolism, we computed the intracellular metabolic fluxes of FH-deficient cells using genome-scale reconstruction of human metabolism ( 12 -14 ) . We constrained the human genome-scale model  Table 1) and the FH loss status (Figure 2A).
We used liquid chromatography coupled to mass spectrometry (LCMS)-based to measure CORE rates of 25 metabolites involved in central carbon metabolism, out of which 13 showed significant changes ( Figure 1D) (Supplementary Table 1). An unsupervised hierarchical clustering showed that UOK262 and UOK262pFH clustered separately (Supplementary Figure   1). UOK262 cells displayed increased lactate secretion ( Figure 1D), and while not significant at an FDR 5% (FDR < 10%) they also showed increase of glucose consumption (Supplementary Table 1), in line with aerobic glycolysis in FH-deficient cells ( 11 ) .
A recent version of the general human genome-scale metabolic model ( 14 ) was used to generate specific models for UOK262 and UOK262pFH cell lines separately. FH loss in UOK262 cells was represented by limiting the flux rate of its catalysed reactions to zero, while in the UOK262pFH cells they remained unaltered. The composition of the cell medium was also used to restrict the metabolites available for consumption (see Methods). The metabolic models were then constrained using the CORE rates, hence generating two context-specific metabolic models ( Figure 2A). Parsimonious FBA (pFBA) ( 15 ) was then used to simulate the two metabolic models using ATP production as the objective function (Supplementary Table 2).
UOK262 cells showed decreased biomass yield, suggesting that the impairment of the mitochondrial function by FH deletion and the increased levels of glucose intake does not lead to augmented growth rate ( Figure 2B). The models predicted increased levels of energy production of UOK262 cells ( Figure 2C). Of note, the total amount of ATP production in UOK262 was greater than what would be expected by glucose uptake alone, suggesting that other carbon sources are utilised by these cells. This is supported by the measured increased uptake of glutamine.
Glycolytic reactions displayed increased fluxes and increased levels of lactate secretion in UOK 262 cells ( Figure 2D), in line with previous observations ( 11 ) . Interestingly, the models also captured the impaired mitochondrial activity of UOK262 cells. Pyruvate dehydrogenase (PDHm) reaction shows decreased intake of pyruvate into the mitochondria and FH inactivation leads to decreased metabolic activity of several reactions in the TCA cycle, e.g. MDHm, SUCD1m.
While, these models do not account for intracellular accumulation and depletion of metabolites concentration these results are in line with the accumulation of fumarate, succinate, and succinyl-coa as aKGDm displays consistent activity. Of note, increase of fumarate is a clear biochemical feature of FH-deficient cells ( 11 ) . In summary, these models recapitulate several known metabolic phenotypes of these cell lines and also offer the possibility to explore at a genome-scale level the metabolic adaptations of UOK262 to the reactivation of FH.

Post-translational and post-transcriptional regulation of metabolism
Having evaluated the predictive capacity of the metabolic models, we then used the in silico metabolic fluxes together with the proteomics and phosphoproteomics data-sets to explore potential regulatory mechanisms of metabolism. An exploratory enrichment analysis ( 16 ) Table 2). In particular, processes involving cellular filament and cytoskeleton were identified to be significantly up-regulated in UOK262.
This result is consistent with the increased motility of UOK262 cells, which has also been associated with epithelial to mesenchymal transition (EMT) ( 10 ) . Several GO terms related with mitochondrial processes, such as respiratory chain complexes, were down-regulated, consistently with the metabolic models predictions and previously observed decreased mitochondrial activity ( 10 ) . We then assessed if protein abundance changes in metabolic enzymes could be related with metabolic flux changes predicted by the model (Supplementary Figure 2B). Correlation analysis showed no significant relationship (Spearman's r = 0.12, p-value = 5.21e-01), suggesting that enzyme abundance is insufficient to determine metabolic fluxes, which is consistent with the limited success of previous approaches to interpret metabolism using transcriptomics and proteomics data ( 18 ) . Nevertheless, several metabolic pathways displayed a consistent profile at protein and flux level. For example, the TCA cycle has decreased protein abundance and metabolic flux in UOK262. Intriguingly, glutamate metabolism shows an increase in the abundance of metabolic enzymes and decrease in metabolic flux of the whole pathway in UOK262 (Supplementary Figure 2B), and this was not in agreement with the measured increase in glutamate secretion and glutamine intake ( Figure 1D).
We validated this unexpected finding performing a separate 13 C-glutamine labeling experiment and found that indeed, whilst these cells do not accumulate glutamate ( 19 ) , they release glutamate in a time-dependent fashion, and that glutamate is predominantly generated by glutamine (See Supplementary Figure 2C). These observations suggest a broad regulatory role of post-transcriptional changes in the central carbon metabolism of UOK262 cells.
Next, we set to find potential regulatory phosphorylation sites of metabolic enzymes. We first assessed how related are the phosphorylation measurements to the respective abundance of the protein. As expected, phosphosites fold-changes were tightly correlated with the protein abundance (spearman's r = 0.56, p-value = 9.0e-44) ( Figure 3B). To focus only on phosphorylation changes that are independent of the protein abundance we only considered residues that change significantly in phosphorylation but not in abundance. This process allowed us to obtain a list of 18 phosphorylation-sites in metabolic enzymes that show significant changes in phosphorylation in FH-deficient cells ( Figure 3C). Consequently, we enquired which of these phosphosites are more likely to have a regulatory role in the metabolic enzymes using the in silico metabolic fluxes measurements as a proxy for enzymatic activity.
Using the genomic annotation in the metabolic model we mapped the selected phosphosites in the metabolic enzymes to the reactions they catalyse, covering 20 phosphosite-reaction interactions (Supplementary Table 4) ( Figure 3D). With our analysis we recapitulated 3 previously reported regulatory residues in PhosphositePlus ( 20 )  Similarly to S232, S293 phosphorylation is decreased in UOK262pFH compared to UOK262 cells ( Figure 4A). Moreover, as PDHA1 was not detected in the proteomics data-set, we measured its abundance also with WB, and confirmed that there are no significant changes in its abundance between UOK262 and UOK262pFH ( Figure 4B). These results support the hypothesis that PDHA1 activity is regulated by post-translational modifications. We then validated the predicted inactivation of PDHm by measuring the conversion of glucose-derived pyruvate to citrate, a two-step reaction that involves PDH-mediated conversion of pyruvate to acetyl-CoA and the condensation of acetyl-CoA with oxaloacetate to generate citrate. To this aim, cells were incubated with 13 C 6 -glucose and the isotopologue distribution of pyruvate and citrate was analysed by LC-MS (schematic in Figure 4C). Whilst glucose uptake is increased in FH-deficient cells ( 24 ) (Supplementary Figure 2D), the incorporation of glucose-derived molecules into citrate is significantly reduced in FH-deficient cells, consistent with the inhibition of PDH activity and the prediction of the model based on phosphoproteomics data ( Figure 3C).

Discussion
Metabolism deregulation is a hallmark of cancer ( 2 , 25 ) , and it has become increasingly clear that it is a complex process involving post-transcriptional regulation, and understanding it is a challenging task that requires an integrative perspective. Here we acquired and analysed data characterizing the proteome, phosphoproteome, and metabolome of FH-deficient cell lines and found profound alterations that are impinged by restoring FH activity (Figure 1). We found While further experimental evidence is required to confirm these findings, this hypothesis can potentially provide insights into the inactivation of GAPDH to re-route glycolysis flux into pentose phosphate pathway in response to potential oxidative stress ( 28 ) a phenomenon that has been reported in FH deficient cells ( 29 ) .
In summary, this work provides for the first time a genome-scale study of the regulatory implications of post-translational modifications in the metabolism of FH-deficient cancer cell lines. Specifically, we exemplify the utility of our approach to identify in a systematic manner potential regulatory phosphorylation residues in metabolic enzymes, which can help to shed light into the complex regulation of cancer metabolism. This approach is also generally applicable to the study of other types of post-translational modifications that fall on metabolic enzymes, for example acetylation and succination, a modification caused by increased fumarate in FH-deficient cells ( 30 ) . Pairing recent studies that have characterised the phosphoproteome of several hundreds of tumour samples ( 31 , 32 ) with metabolomics measurements will potentiate the discovery of novel therapies that exploit ubiquitous features of cancer metabolism.

Proteomics and phosphoproteomics mass-spectrometry experiment
Proteomics experiments were performed using mass spectrometry as reported ( 33 , 34 ) . Urea lysis buffer was used to lyse the cells (8 M urea, 10 mM Na3VO4, 100 mM β-glycerol phosphate and 25 mM Na2H2P2O7 and supplemented with phosphatase inhibitors (Sigma)). Proteins Raw intensities were log2 transformed followed by a scaling in each sample to account for potential pipetting differences. Only proteotypic peptides and peptides measured consistently across half of the replicates were considered. Differential protein abundance and phosphorylation analysis was performed using T-test between the UOK262 and UOK262pFH samples. P-values were controlled for false discovery rate using Benjamini-Hochberg FDR.

Consumption and release quantification of metabolites
UOK262 and UOK262pFH (1.5x10 5 ) were plated onto 6-well plates and allow to grow for 16h.

Metabolic extracts after glucose and glutamine labelling
UOK262 and UOK262pFH (1.5x10 5 ) were plated onto 6-well plates and grown overnight. The day after, medium was replaced with fresh one containing 13 C 6 glucose or 13 C 5 glutamine (Cambridge Isotope Laboratories). The following day, cells were counted using Countess (Thermo Fisher Scientific) and extracted as described before ( 10 ) at the indicated time points.

Metabolic modelling using Recon 2.2
For this analysis the human metabolic reconstruction Recon 2.2 was used ( 14 ) .  metabolomics experiments quantifying exchange rates (mmol/gDW/h). All the metabolite rates shown are significantly different (FDR < 5%) between UOK262 and UOK262pFH cells.