Loss of SDHB Promotes Dysregulated Iron Homeostasis, Oxidative Stress, and Sensitivity to Ascorbate

Succinate dehydrogenase (SDH) is a key enzyme localized in the inner mitochondrial membrane where it participates both in the tricarboxylic acid (TCA) cycle and the electron transport chain (ETC), as complex II. In the TCA cycle, SDH catalyzes the oxidation of succinate into fumarate while in the ETC, it allows the transfer of electrons from succinate to the ubiquinone pool. SDH is constituted of two anchoring subunits (SDHC, SDHD) and two catalytic subunits (SDHA, SDHB) encoded by the four SDHx nuclear genes (SDHA, SDHB, SDHC, and SDHD). SDHA is a flavoprotein to which the substrate succinate binds to be oxidized to fumarate, while the resulting electrons are channeled to ubiquinone, through the three Fe-S clusters of SDHB. Since the initial discovery of SDHD mutations in 2000 (1), all four SDHx genes have been demonstrated to act as tumor suppressor genes (2, 3, 4). Their mutations predispose to pheochromocytoma and paraganglioma (PPGL), rare neuroendocrine tumors arising from chromaffin cells of the adrenal medulla and from the sympathetic and parasympathetic nervous systems, respectively; as well as gastro-intestinal stromal tumors (GIST) and rare forms of renal cell carcinomas. It is now estimated that mutations in SDHx genes stand for almost 50% of inherited cases of PPGL. Interestingly, SDHB mutations have been shown to be the highest risk factor of malignancy in patients with PPGL: approximately 50% of PPGL patients carrying SDHB mutations will ultimately develop a metastatic form of the disease and a germline mutation in SDHB is found in up to 36% of all metastatic PPGL cases (5). In contrast, metastatic forms of the disease are found in only 5% of patients with an SDHD gene mutation. Moreover, SDHB-mutated cancers are more aggressive than non-mutated metastatic ones. The overall survival of patients with metastatic SDHB-related PPGL was initially estimated at 42 months after diagnosis of the first metastasis versus 244 months for patients with a metastatic PPGL but without an SDHB gene mutation (6). Interestingly, recent studies have revealed that increased follow-up of patients with genetically determined PPGL has significantly improved their survival, demonstrating the beneficial impact of genetic testing in these patients (7, 8).

In all cases, germline SDHx mutations are associated with loss of heterozygosity, which causes the complete loss of SDH function in the tumor (9) and a subsequent accumulation of its substrate, succinate (10). Succinate (as other TCA cycle intermediates such as fumarate and 2-hydroxyglutarate) is considered as an oncometabolite, driving aberrant activation of transcription factors and global epigenetic reprogramming. Indeed, high steady-state levels of these metabolites are known to inhibit 2-oxoglutarate (2-OG) dependent dioxygenases, such as the hypoxia-inducible transcription factor (HIF) prolyl-hydroxylases (PHD) and Ten-eleven translocation (TET) DNA demethylases (11). These enzymes belong to the large family of iron (Fe) and 2-OG-dependent dioxygenases, which activity is based on the use of iron (12), oxygen (13) and ascorbate (14) as main cofactors. Inhibition of PHDs promotes the abnormal stabilization, and thus activity of HIF1 and HIF2, while TET inhibition leads to a lack of DNA demethylation, resulting in pseudo-hypoxic and hypermethylator phenotypes, respectively. These observations were initially revealed in genomic studies performed on large PPGL tumor collections, resulting in a better understanding of the mechanisms of SDH-related tumorigenesis. In these studies, all SDHx- and FH-mutated tumors clustered together, showing a transcriptome signature characterized by activation of the hypoxic response pathway (2, 15, 16) and a DNA hypermethylator phenotype (17). Despite this apparent homogeneity, some differences have emerged, suggesting that SDHB-mutated tumors may display a particularly marked hypermethylator phenotype associated with hallmarks of neuroendocrine-to-mesenchymal transition (NMT), which were not observed in other types of SDHx-mutated cases (i.e., SDHA, SDHC, and SDHD mutated; ref. 18). To decipher these mechanisms at the cellular level, we have generated an immortalized mouse chromaffin cell (imCC) line knocked-out for the Sdhb gene (17). In these cells, we showed that Sdhb deficiency does promote a marked hypermethylator phenotype mediated by TET1 and/or TET2 blockade, which acts in synergy with HIF2 activation to promote NMT and metastatic dissemination (19, 20). Sdhb^−/− cells also undergo significant metabolic reprogramming becoming critically dependent on aspartate synthesis as a major source of carbon for anabolic purposes, which supports their viability and proliferation, maintained mainly through reductive carboxylation (21).

Despite these important advances in the understanding of the oncogenic consequences of SDHB loss of function, the exact cause of the metastatic phenotype of SDHB-related tumors remains mostly unexplained. SDHA and SDHC mutations are rare in PPGL patients, probably because of low penetrance of the disease in mutation carriers. In contrast, SDHB and SDHD are the most frequently described mutations in PPGL, with an incidence of 8%–10% and 5%–7% among all PPGL cases, respectively (22). Genotype–phenotype correlation studies performed in patients with SDHB or SDHD mutations have undoubtedly demonstrated their different clinical behavior in terms of invasiveness. To address this crucial question in the field, we have generated Sdhd knocked-out cell lines in the imCC model. In this new and unique model, we have investigated the different pathways previously identified in SDH-related tumors to identify the specificities that may explain their different clinical behavior.

Detailed methods are provided in the Supplementary Information section.

Complete knock-out of Sdhd gene was achieved in immortalized mouse Chromaffin Cells (imCC) by transfecting Wild-Type imCC, previously generated by our lab (17), with in silico designed targeted gRNA, using the PrecisionX Cas9 SmartNuclease RNA System Kit (System Biosciences). Sdhd-rescue clones were established by transfecting Sdhd^−/− imCC with an SDHD-expressing vector (ORIGENE, NM003002). Screening of these clones was performed by SDH activity test, qRT-PCR, and direct sequencing of the Sdhd gene by Sanger method. Cells used for experiments were maintained in DMEM high-glucose Glutamax (Gibco) unless when grown for metabolic or respirometry experiments (see Supplementary Information). Wound scratch assays, proliferation and adhesion tests were performed to assess cellular phenotype. When stated, cells were treated with indicated doses of iron Chloride (Sigma), MitoTEMPO (Sigma), l-Ascorbic acid (Sigma), talazoparib (MedChemTronica), or N-acetyl-cysteine (Sigma).

RNA sequencing was performed for 7 samples: 3 Sdhb^−/−, 2 Sdhd^−/−, and 2 WT imCC. Enriched samples were used to generate sequencing libraries with the Illumina “TruSeq Stranded mRNA Sample Prep” Kit and associated protocol provided by the manufacturer. Libraries were sequenced by IntegraGen on an Illumina HiSeq 2000 as paired-end 75-bp reads. Reduced Representation Bisulfite Sequencing (RRBS) was also performed by Integragen SA, as described previously (17). Details of the process leading to the evaluation of differential expression and methylation rates are provided in the Supplementary Information. The sequencing data reported in this article have been deposited to the EGA (European Genome-phenome Archive) database (European Genome-phenome Archive) database (accession number EGAS00001005279).

All cell types underwent metabolite extraction using Methanol:H2O:Chloroform at ratio 1:1:1, following ¹³C-glucose and ¹³C-glutamine labeling. The polar fractions were used for gas chromatography – mass spectrometry (GC-MS) analysis allowing acquisition of the isotope data. High-resolution respirometry was performed by loading Intact or permeabilized cells in an Oxygraph-O2 k (Oroboros instruments) chamber, using Mir05 respiratory media and digitonin for the latter. A pseudo dual-wavelength Varian CARY50 spectrophotometer was used to assess respiratory chain activities as described previously (23). For the assessment of oxidative stress features, cells treated as stated, were incubated with either CellRox, MitoSox, or Bodipy C11 Reagent (Thermo Fisher) and Flow cytometry data were acquired using an LSR FORTESSA analyzer (BD Biosciences).

For cellular iron and copper assessment, cells were incubated with the relevant metal-specific probe prior to being analyzed by flow cytometry on a BD Accuri C6 (BD Biosciences). Laser ablation–inductively coupled plasma-mass spectrometry (LA-ICP-MS) was performed on dewaxed and rehydrated formalin-fixed paraffin-embedded tumor sections. Details of LA-ICP-MS experiments are provided in the Supplementary Information. LA-ICP-MS images were acquired in a fixed dosage mode, with a vertical and horizontal spatial resolution of 20 μm, a detailed list of sample settings is available in Supplementary Table S1. Average ⁵⁶Fe and ⁶³Cu intensities were calculated from pixels containing a positive elemental tissue marker [(31P)] ensuring background regions did not influence these calculations.

For qRT-PCR, total mRNA was extracted from cell pellets using RNeasy Plus Mini Kit (Qiagen). Reverse transcription was fulfilled using iScript cDNA Synthesis Kit (Bio-Rad) and qRT-PCR was performed on CFX96 Real-Time System C1000 Touch Thermal Cycler (Bio-Rad) using SuperScript SybrGreen (Bio-Rad) to assess the expression of Sdhb, Sdhd, Snai1, Fn1, Epas1, MMP9, and 18S. Western blot analysis was performed to assess SDHB, SDHA, HIF2α, and TUBULIN expression. RIPA lysis buffer was used for extraction of total protein and concentrations were determined with Bradford colorimetric method. Results were read with a LAS-4000 Mini (Fuji). Actin staining and γH2AX immunofluorescence were performed on cells plated on glass slides, while γH2AX IHC was performed on 6-μm sections from paraffin-embedded tumors. Details of primers, primary and secondary antibodies, immunostaining, and experimental conditions are provided in the Supplementary Information.

Data are represented as mean (of at least three independent experiments) ± SEM. Data were analyzed by one-way or two-way ANOVA. Statistical tests were carried out using the GraphPad Prism software. All analyses considered a value of P ≤ 0.05 to be statistically significant. For enrichment analysis, we used an in-house adaptation of the GSEA method to identify gene sets from the MSigDB database overrepresented among up/downregulated and hyper/hypomethylated genes.

To investigate the mechanisms underlying the metastatic characteristics of SDHB-mutated PPGL, but not of SDHD-deficient tumors, we knocked-out the Sdhd gene using the CRISPR/Cas9 procedure in wild-type (WT) imCC (17). We obtained two independent Sdhd^−/− clones (ClA and ClB) each containing a different homozygous point mutation on exon 3 of Sdhd, c.119delC and c.120_121delCA, respectively, both leading to premature stop codons (Fig. 1A). These mutations were predicted to lead to the translation of truncated proteins of 79 and 68 amino acids, respectively, instead of the 159 amino acids of the entire WT SDHD protein. To attest for the absence of off-target effects in Sdhd^−/− cells, we generated a third cell line, in which SDHD loss was rescued by stably transfecting an Sdhd-expressing vector in the Sdhd^−/− ClA (clone Sdhd-R). Sdhd-mutated cells showed a decrease in Sdhd RNA levels (Supplementary Fig. S1A) when compared with WT and Sdhd-R cells, which may be suggestive of nonsense-mediated decay in the mutant clones, and particularly in clone A.

SDHD protein expression could not be estimated due to the lack of efficient antibody in mouse. It is worth noting that Sdhd mRNA levels were also reduced in Sdhb-deficient cells. In contrast, Sdhb mRNA levels were absent in Sdhb^−/− cells, but not significantly reduced in Sdhd^−/− imCC (Supplementary Fig. S1B). It has previously been reported that disassembly of the SDH complex in case of SDHx mutations leads to the degradation of the SDHB subunit in all SDHx-mutant cells (24). However, our data show that Sdhd^−/− clones retain a small but significant level of SDHB protein (Fig. 1B). Similarly, evaluation of SDHB protein by IHC in SDH-related PPGL frequently leads to a weak diffuse signal in SDHD-deficient tumors, while it is totally negative in SDHB-mutated ones, suggesting that some SDHB remains present in this type of mutants (25). Nevertheless, they exhibited a complete loss of SDH enzymatic activity, similar to that observed in Sdhb-deficient cells (Fig. 1C), and in accordance with observations in mutated human PPGL (9, 26). Interestingly, it has previously been suggested that succinate accumulation might be less important in SDHD-deficient tumors, which may account for their different metastatic potential (27). Analysis of intracellular succinate levels revealed a substantial accumulation of succinate in Sdhd^−/− cells that reached that observed in the Sdhb^−/− imCC model (Fig. 1D). Similarly, levels of the related metabolites, fumarate, and 2-oxoglutarate, were strongly reduced in all SDH-deficient cells regardless of the subunit inactivated. Sdhd-R cells showed a profile similar to that of WT imCC (Fig. 1E and F; Supplementary Fig. S1C and D).

Next, we assessed the metabolic and respiratory phenotypes of both types of SDH-deficient cells. ¹³C-glucose tracing experiments revealed that in accordance with previous work (21), cells deficient in SDH-activity undergo a metabolic rewiring in the absence of a fully functional TCA cycle (Fig. 2A). Like Sdhb^−/− cells, the synthesis of aspartate by Sdhd^−/− imCC cannot rely on oxidative TCA cycle metabolism (Fig. 2B and C, m+2 isotopomer) and instead increase the aspartate produced through pyruvate carboxylation (PC; Fig. 2B and C, m+3 isotopomer) for anabolic purposes. Equally, their synthesis of citrate from glucose, a central metabolite for cellular anabolism, was greatly reduced (Fig. 2B). The metabolism of another key nutrient, glutamine was therefore examined. Although both cell lines retain some oxidative metabolism of glutamine to produce succinate, Sdhd-/- imCC appear to significantly increase their synthesis of aspartate and citrate using reductive carboxylation (Fig. 2C and D), which is not observed to the same degree in Sdhb^−/− imCC. We then measured oxygen consumption of intact and permeabilized imCC (Fig. 2E and F). In both cases, Sdhb^−/− imCC exhibited no significant change in basal respiration, while Sdhd^−/− cells showed a decrease, suggesting that a more significant defect in respiration exists in the Sdhd^−/− cells compared with Sdhb^−/−. Maximal respiratory capacity (FCCP-mediated uncoupled respiration) was significantly reduced in both SDH-deficient imCC models, in compliance to that previously reported in Sdhb^−/− cells (28). The relative decrease in respiratory activity observed in Sdhd^−/− cells compared with those deficient in Sdhb was consistent with the increased reliance on reductive carboxylation for the synthesis of aspartate and citrate (Fig. 2C), an effect previously observed in hypoxia (29). Given the decreased oxygen consumption observed in the Sdhd-deficient cells compared with those lacking Sdhb, we examined whether there was a compensatory increase in pyruvate reduction to lactate to regenerate cytosolic NADH and maintain glycolysis. We indeed saw that while Sdhb^−/− cells show a small increase in lactate production, this is further increased in Sdhd-deficient cells (Fig. 2G).

Altogether, these results suggest that massive succinate accumulation and rewired metabolism do not constitute relevant triggers for SDHB-related malignancy, although the extent of mitochondrial respiratory deficit may vary between the two models.

We previously reported that Sdhb^−/− imCC exhibit a mesenchymal-like phenotype with increased adhesion, migration, and invasion capacities (19). Morphologic observation following phalloidin staining qualitatively revealed that Sdhd^−/− cells appear to be generally smaller than Sdhb^−/− cells, and display a less extensive and rather peripheral actin mesh unlike Sdhb^−/− imCC (Fig. 3A), suggesting that only Sdhb^−/− cells undergo mesenchymal transformation. Accordingly, markers of NMT activation such as Snai1, Mmp9, Fn1 (Fig. 3B) were overexpressed in Sdhb^−/− imCC but not in Sdhd^−/− cells. In accordance with the loss of neuroendocrine features, NCadh was downregulated in Sdhb^−/− imCC, while it was actually overexpressed in Sdhd^−/− cells (Supplementary Fig. S2A). Sdhd^−/− cells showed moderate increase in adhesion (Fig. 3C), and no increase in migration, evaluated both in collective (Fig. 3D) and individual migration assays (Supplementary Fig. S2B). Sdhd^−/− imCC also exhibited a much slower cell growth than Sdhb^−/− cells (Supplementary Fig. S2C). Hence, despite similar SDH loss-of-function and succinate levels, Sdhd-deficient cells do not display the mesenchymal and migratory phenotype of Sdhb KO cells, as observed in human paraganglioma (18).

We have previously established that SDHB-mutated PPGL and Sdhb^−/− imCC display pseudohypoxic and hypermethylator phenotypes, caused by inhibition of PHD and TET enzymes and involved in the mesenchymal-like hallmarks associated with SDHB deficiency (17, 20). We next performed reduced representation bisulfite sequencing (RRBS) to precisely quantify 5-methyl-cystosine (5mC) modifications along the genome after enrichment for CpG-rich regions in the different cell lines. Compared with WT, both Sdhb and Sdhd KO cells displayed a DNA hypermethylator phenotype, but this phenotype was less pronounced in Sdhd^−/− cells (Fig. 3E). In total, Sdhb^−/− cells displayed 6,847 significantly hypermethylated gene–based features (TSS ± 500 bp and gene bodies) against only 2,564 for Sdhd^−/− cells (Fig. 3F; Supplementary Table S2). Hypermethylator phenotypes in Sdhb^−/− and Sdhd^−/− cells displayed a significant overlap and correlation (Fig. 3G and H), with 1,084 common hypermethylated features representing 42% and 16% of hypermethylated features in Sdhd^−/− and Sdhb^−/− imCC, respectively.

Surprisingly, Western blot analysis in our cell lines revealed stabilization and massive increase of HIF2α protein levels in Sdhb^−/− cells, which was not seen in either Sdhd^−/−, WT, or Sdhd-R imCC (Fig. 3I). We thus analyzed the transcriptome of all cell lines using RNA-sequencing. Hierarchical clustering of expression profiles across the 1,000 most variant genes (based on SD) showed that Sdhd^−/− imCC remain closer to wild-type cells and display less gene expression changes than Sdhb^−/−cells (Fig. 3J). To investigate more specifically the expression of hypoxia-inducible genes, we analyzed the expression of a previously published list of 52 HIF targets (3), among which, 47 were present in the transcriptome of imCCs (Supplementary Table S3). Twenty-four (24) of them were overexpressed in Sdhb-deficient cells compared with WT (mean value >120% of WT), while 17 were overexpressed in Sdhd-deficient cells compared with WT (13 overlapping with those high in Sdhb KO cells; Fig. 3K). Comparison of both KO cells showed that 19 of the 28 overexpressed genes showed stronger expression in Sdhb^−/− cells (Fig. 3L). These data therefore suggested that although HIF2α was hardly detectable by Western blot analysis, Sdhd^−/− cells did show a mild pseudohypoxic signature. To validate this observation, we further analyzed previously published transcriptome data generated in a collection of 188 human PPGL. In this study, 3 SDHD, 1 SDHA, and 2 SDHC-mutated PPGL were evaluated together with 17 SDHB-mutated ones (3), which all clustered together using the expression of the 52 HIF-target genes. We have reanalyzed these data, limiting them to the 41 highly expressed ones, and we evaluated their mean expression in SDHB versus SDHD-mutated tumors. We observed that most genes were overexpressed in SDHB-related tumors while the subset of genes over-represented in SDHD-mutated tumors corresponded to genes expressed by endothelial cells or associated with angiogenic processes (Supplementary Fig. S3A), therefore reflecting the very high vascular density described in these tumors (15, 30). Similar results were obtained on the TCGA cohort of PPGL (Supplementary Fig. S3B; ref. 31).

Altogether, these findings indicate that Sdhd inactivation induces less DNA methylation and gene expression changes than Sdhb deletion. We next aimed at identifying the causes of the seemingly greater inhibition of 2OG-dependent dioxygenases in Sdhb^−/− cells.

2OG-dependent dioxygenases are iron-containing enzymes. Because SDHB is the iron-sulfur subunit of complex II, we wondered whether its complete loss might lead to disturbances in iron metabolism, which would in turn modulate the activity of 2OG-dependent dioxygenases. We thus used specific fluorescent probes to detect cytosolic and mitochondrial iron in all cell types, as well as assessed iron levels in tumor samples from patients with SDHB mutations. We also evaluated mitochondrial copper in all cell types. Indeed, iron and copper are both essential metals with close metabolic interactions that have been known for a long time (32).

We first used the commercial Calcein-AM probe that reveals both iron(II) and iron(III) levels, but also other ions such as calcium. This experiment showed a decrease in global labile iron in SDH-deficient cells compared with WT imCC (Fig. 4A). To validate this observation in human PPGL, we performed laser ablation-inductively coupled plasma-mass spectrometry (LA-ICP-MS) in 6 pheochromocytomas carrying NF1 (n = 2), RET (n = 2), or SDHB (n = 2) mutations, as well as 4 paragangliomas with SDHB (n = 2) or SDHD (n = 2) gene mutations. We observed that average ⁵⁶Fe counts were significantly decreased in tumors with SDHB mutations compared with tumors with NF1, RET, or SDHD mutations (Fig. 4B–F). This decrease was greater in SDHB-mutated paragangliomas (Fig. 4D) than in SDHB-mutated pheochromocytomas (Fig. 4C). These results obtained in tumors from patients consolidate the hypothesis that loss of the SDHB subunit leads to an altered iron metabolism. Calcein-AM probe being poorly specific, we next examined the subcellular localization of Iron(II) using specific probes of both this ion, and specific cell compartments. Unexpectedly, evaluation of the labile pool of cytosolic iron(II) revealed a significant increase in Sdhb^−/− cells only (Fig. 4G). This suggests that the calcein-AM data would actually be imputable to a strong decrease in cytosolic levels of Iron(III) in SDH-deficient cells. In opposition, we observed a significant decrease in the mitochondrial pool of chelatable iron [iron(II)] in both knockout models, a decrease that was especially pronounced in Sdhb^−/− cells and rescued in Sdhd-R imCC (Fig. 4H).

We next evaluated mitochondrial copper levels in all cell types. Concerning the pool of labile mitochondrial Cu(II), no significant changes were observed in the SDH-deficient cells, while mitochondrial Cu(I) was significantly increased in Sdhb KO cells only (Supplementary Fig. S4A and B). LA-ICP-MS was also used to quantify the levels of ⁶³Cu in 10 PPGL tumors. This evaluation of global copper was not able to confirm the difference in SDHB versus RET or NF1 tumors. However, it suggested that SDHD-deficient PGL had a decrease in copper counts compared with the other genotypes (Supplementary Fig. S4C). Hence, there was a dysregulation in copper and especially in iron pools in cells and tumors depending on their genotypes. In support of these observations, we evaluated the mRNA expression of several iron and copper transporters as well as of chaperones or different actors of iron/copper homeostasis, which revealed that these pathways were indeed strongly affected in Sdhb KO cells (Fig. 4I and J). In particular, these cells showed a differential expression of transporters implicated in the entry of iron and copper in the cell and the mitochondria: high DMT1 (Slc11a2) and low SLC25A37 would sustain the increased cytosolic distribution of iron(II), while low CTR1 (Slc31a1) and high Slc25a3, the high mitochondrial Cu(I) levels. Consistently with these observations, human PPGL transcriptome data showed an increase in DMT1 mRNA levels in SDH tumors as compared with other types of mutated PPGL. SDH tumors also exhibited a slight increase in the expression of other actors involved in iron binding and transport, such as the hypoxia-inducible genes TF and TFRC. SLC31A1 was significantly decreased in both SDH and VHL tumors compared with RET or NF1-mutated ones. No differences were detected for TFR2 nor SLC25A3 expression (Supplementary Fig. S4D).

Sdhb-KO cells show increased mitochondrial copper(I) and cytosolic iron(II) levels, which could cause important oxidative stress in these cell compartments (33). An increase in reactive oxygen species (ROS) might also participate to the sustained inhibition of 2OG-dependent dioxygenases activity (34). Flow cytometry analyses of fluorescent probes revealed that Sdhb-deficient cells showed increased ROS both in the cytosol (Fig. 5A) and especially in the mitochondria (Fig. 5B). They also displayed a significant increase in lipid peroxidation levels (Fig. 5C) and in mitochondrial activity (Fig. 5D) but not in overall mitochondrial mass (Fig. 5E). Unexpectedly, none of the Sdhd-deficient clones showed any changes in ROS levels as compared with the WT cells.

Because Sdhb^−/− cells display such a significant increase in ROS, we postulated that they would probably show increased anti-oxidant responses, either through superoxide dismutase (SOD) or through activation of the NRF2 pathway. However, the global superoxide dismutase activity did not differ between the different cell types (Supplementary Fig. S5A). Moreover, RNA-sequencing analysis showed a significant downregulation of mitochondrial SOD2, thus demonstrating that Sdhb-deficient cells do not detoxify the excess in mitochondrial ROS (Fig. 5F).

We performed gene set enrichment analyses using RNA-seq data of 58 NRF2-target genes (Supplementary Table S4), which showed no activation of the pathway in Sdhb nor in Sdhd KO cells, compared with WT (Supplementary Fig. S5B). In addition, treating the cells with N-acetyl cysteine (NAC) had no effect on the levels of mitochondrial ROS in any of the cell types studied (Fig. 5G), neither affected their proliferation (Fig. 5H). To fully validate this observation in the human context, this list of genes was also tested in the transcriptome data of the human PPGL COMETE collection (Fig. 5I). Mean values of NRF2 targets levels in the groups of sporadic non-mutated tumors (n = 49), RET (n = 19), NF1 (n = 37), VHL (n = 40), SDHA, C, D- (n = 6) and SDHB-mutated tumors (n = 17) were used to perform unsupervised classification, which confirmed that the NRF2 pathway is not activated in SDHB-deficient tumors.

It has previously been reported that SDHB loss is associated with increased DNA damage (35), and a sensitivity to inhibitors of PARP (35, 36, 37). In addition to the recently demonstrated mechanism associated with oncometabolite-induced inhibition of the lysine demethylase KDM4B, increased ROS levels may participate to such phenotype. We therefore evaluated DNA damage after staining PPGL tumor tissues and imCCs with anti-γH2AX antibody (Supplementary Fig. S5C and D). Although it was not statistically significant, we observed a slight increase in γH2AX levels in SDHB-deficient tumors, which was not seen in Sdhb^−/− imCCs. Treating cells with increasing doses of the PARP inhibitor talazoparib reduced cell survival in all cell types, with an actually higher efficacy in WT and Sdhd-R cells (Supplementary Fig. S5E). In this cell model, it seemed that talazoparib efficiency was not related to SDH-deficiency, but was rather related to cell proliferation indexes of the treated cell types.

Altogether, these data suggest that in Sdhb KO cells, a combination of succinate accumulation, dysregulation of iron homeostasis and high ROS levels may account for the increased inhibition of TET and PHD hydroxylases. To demonstrate this link, we treated cells with iron(II) (50 μmol/L), which had no effect on cytosolic ROS (Supplementary Fig. S6A) but reduced mitochondrial ROS, only in Sdhb^−/− cells (Fig. 6A). This was associated with a decrease in HIF2α protein levels in Sdhb^−/− cells as well (Fig. 6B). Treating cells with increasing doses of ferrous iron revealed a specific sensitivity of Sdhb-deficient cells. Indeed, after 48 hours of treatment, 0.1 mmol/L doses of iron(II) led to a 60% decrease in Sdhb^−/− cells survival while it only moderately affected WT, Sdhd-KO, or Sdhd-R cell types (Fig. 6C).

To further demonstrate that increased mitochondrial ROS were indeed responsible for the massive HIF2α stabilization observed in Sdhb-deficient cells, we then treated them for 12h with MitoTEMPO, a mitochondrially targeted antioxidant that acts as a specific scavenger of mitochondrial superoxide. MitoTEMPO (MTT) at 2.5 μmol/L had no effect on cytosolic ROS (Fig. 6D), but significantly decreased mitochondrial ROS in Sdhb^−/− imCC (Fig. 6E), which was associated with a substantial reduction of HIF2α levels (Fig. 6F). High doses of MTT are suspected to show a reverse effect, promoting a pro-oxidant response. We therefore treated cells with 2.5 mmol/L of the compound and showed that such high dose increased cytosolic ROS in all cell types and especially in WT and Sdhd^−/− ClA cells (Fig. 6G), as well as mitochondrial ROS in WT and Sdhb^−/− cells (Fig. 6H). Strikingly, such treatment promoted a strong pseudohypoxic response in WT, Sdhd^−/−, and Sdhd-R cells, thus demonstrating that increasing ROS was sufficient to mediate HIF2α stabilization. In contrast, this led to the opposite effect in Sdhb^−/− cells (Fig. 6I) in which HIF2α protein was almost lost. It actually appeared that high dose MTT was highly toxic for Sdhb^−/− cells exclusively (Fig. 6J), with perinuclear vacuoles emerging after 12 hours and leading to the death of all Sdhb^−/− imCC after 72 hours.

In an attempt to evaluate another pro-oxidant treatment that might be used in a clinical context, we finally tested the vulnerability of imCC to high dose ascorbate. A dose response curve showed that Sdhb^−/− cells were indeed much more vulnerable to ascorbate with significant increase in cell death appearing from a 1 mmol/L dose onward, a complete lethality being reached at a 2.5 mmol/L ascorbate dose (Fig. 6K). At the higher dose, 85% of Sdhb-deficient cell loss was achieved after 24 hours, while a similar response was only apparent in Sdhd and WT cells after 72 hours (Fig. 6L). To evaluate the mechanism associated with ascorbate-induced lethality, we measured cytosolic and mitochondrial ROS in cells treated for 48 hours with 1 mmol/L ascorbate. At such a dose, there was already a significant lethality of Sdhb-KO cells and enough cell material was still available to allow the analyses (Supplementary Fig. S6B). Interestingly, we observed that cytosolic ROS were not modified by ascorbate in any of the cell models (Supplementary Fig. S6C). In contrast, ascorbate led to a massive increase in mitochondrial ROS, specifically in Sdhb^−/− cells (Fig. 6M).

These data suggest that in Sdhb-KO cells ascorbate interacts with the increased iron(II) labile pool to enhance oxidative stress, leading to rapid cell death.

It is well-established that a mutation in the SDHB gene is a high risk factor for metastatic PPGL (38) and poor prognosis (6), unlike mutations in other SDHx genes (SDHA, SDHC, or SDHD) that rarely cause metastatic forms of the disease. This question has been unresolved for 20 years. Many studies have been performed in the past years, mainly using cell lines exhibiting a decrease or complete lack in the expression of Sdhb. Thereby, previous work from our team has shown that metastatic SDHB tumors exhibit a NMT phenotype (18, 19), associated with increased DNA methylation in comparison with other SDHx-mutated tumors (20, 39). Furthermore, we have recently demonstrated that this hypermethylated profile is due to TET silencing and that the NMT phenotype observed in SDHB mutated tumors is the result of synergistic roles of TET repression and pseudohypoxia, through HIF2 activation (20). Although these data have shed light on some mechanisms undeniably involved in the tumorigenesis of SDHB-mutated tumors, they do not explain the different clinical outcomes observed in patients carrying mutations in the different SDHx genes. In this study, we were able to characterize and compare two physiologically appropriate cellular models exhibiting a complete knock-out in Sdhb or Sdhd genes, which strikingly recapitulate most of the phenotypic characteristics of SDHB and SDHD-mutated tumors.

Succinate accumulation is probably the major consequence of SDH loss of activity and increased succinate levels in tumor samples are used as a biomarker of SDH deficiency in PPGL patients (27, 40–42). For a long time, it has therefore been considered to be the main trigger of PPGL tumorigenesis. The role of succinate as an oncometabolite was initially demonstrated in 2005, with the demonstration of its inhibiting role on 2OG-dependent PHD enzymes, leading to the subsequent stabilization of HIFα subunits even under normoxia (43, 44). Later, it was demonstrated that succinate is also a major epigenetic modifier, inhibiting TET enzymes and promoting a genome-wide hypermethylated profile (17, 39, 45). It has been suggested that a stronger inhibition of 2OG-dependent dioxygenases observed in SDHB-mutated tumors versus tumors with mutations in other SDHx genes, might be explained by a fuller inactivation of SDH enzyme associated with higher succinate (27). Here, we firmly invalidate this long-held assumption, both Sdhb^−/− and Sdhd^−/− imCC models showing the same loss of SDH activity and similar levels of succinate and fumarate (Fig. 1). Indeed, many of the metabolic characteristics investigated were highly similar between the two cell models, suggesting that merely measuring overall steady-state succinate and fumarate levels are inadequate for predicting phenotype, and smaller changes in metabolic function may drive downstream. Surprisingly, HIF2α protein was hardly detectable in Sdhd-deficient cells while it was strongly activated in Sdhb^−/− imCC. This provides further evidence that although aspects of the overall metabolic phenotype were similar between Sdhb^−/− and Sdhd^−/− models, as yet unexplored metabolic factors may play a role in the inactivation of the PHD enzymes to stabilize HIF2α. This was unexpected as previous transcriptome analyses of human tumor samples have classified all SDHx-mutated tumors together with VHL-mutated ones in a “pseudohypoxic” cluster (16, 17). However, such classifications were based on global transcriptome and obviously included many other pathways. To our knowledge, only three studies have classified PPGL tumors based on the expression of hypoxia-induced genes. Hensen and colleagues compared hypoxia-inducible genes in sporadic head and neck PGL (HNPGL) to HNPGL carrying PGL2 or SDHD gene mutations (46). This study was not able to identify any differences in the hypoxic signature of these three types of tumors. Besides, Fliedner and colleagues evaluated the expression of HIF target genes in a larger collection of SDHD (n = 14) and SDHB (n = 15) related PPGL and also revealed some differences between the two types of tumors, with some genes being less overexpressed in SDHD-mutated PGL (47). On the basis of previously published transcriptome data (3), our current study shows that most HIF-target genes were overexpressed in SDHB-related tumors while the subset of targets over-represented in SDHD-mutated tumors corresponded to genes expressed by endothelial cells. Hence, although it is most probable that, as previously suggested in other studies (15, 48–50), SDHD tumors display some induction of pseudo-hypoxia in vivo, the combination of our in vitro data and human PPGL studies does suggest that this effect is at least much more limited in non-SDHB–mutated cells and can therefore not be imputable to succinate accumulation only.

Our data reveal that while SDHB expression is completely abolished in Sdhb^−/− cells, a small fraction of SDHB protein remains detectable in Sdhd^−/− clones, consistently with IHC data on tumors that frequently report a weak diffuse SDHB staining in SDHD-mutated PPGL (25, 51). SDHB is the Fe-S subunit of Complex II and contains two highly conserved L(I)YR motifs that are essential for acquisition of Fe-S clusters by recruiting the Fe-S transfer machinery. We postulate that the loss in mitochondrial Fe-S clusters present in the SDHB subunit, may be responsible for the significant imbalance in cytosolic and mitochondrial iron distribution, that we observe in Sdhb^−/− cells. Interestingly, Saxena and colleagues reported that 37% of disease-causing missense mutations in SDHB were located in either the L(I)YR Fe-S transfer motifs or in the 11 Fe-S cluster-ligating cysteines. Moreover, analyses of reported missense mutations in patients with SDH-RCC or GIST revealed that 50% of these patients had SDHB mutations in the L(I)YR motif amino acid residues (52). Altogether, these observations point to a central role of Fe-S clusters loss in SDHB-related disease, which are further reinforced by our present observations.

This labile iron overload may in turn be responsible for or exacerbate the overproduction of ROS observed in Sdhb^−/− cells, thereby indirectly activating pseudohypoxia and DNA methylation. Indeed, it has been previously demonstrated in a number of studies that in the context of excessive ROS levels, these oxidant molecules can act as signals able to trigger the stabilization of HIFα transcription factors in normoxic conditions (53–55; 34). Besides, in this study, we show that the use of MitoTEMPO, a mitochondrially targeted antioxidant at low concentrations led to a significant decrease in mitochondrial ROS, resulting in turn in reduced HIF2α levels. Conversely, high doses of MTT appear to exert a prooxidant action leading to increased cytosolic ROS in all of the cell types, resulting in a strong pseudohypoxic response. Altogether, these results demonstrate the direct link between ROS and HIF2α stabilization and suggest that the elevated ROS levels observed in Sdhb^−/− cells are probably responsible for their stronger pseudohypoxic phenotype. The association between SDH-deficiency and ROS production has been controversial, with some studies showing an increase in ROS levels (28, 53) and others not (56). Actually, these apparently contradictory data are in total accordance with our observations. Indeed, Selak and colleagues first reported the absence of oxidative stress following SDH inactivation, but this was observed in an SDHD KD cell model (56). In contrast, Guzy and colleagues proposed that the mechanism explaining PPGL tumorigenesis following SDH inhibition was an increase in ROS production, acting as messengers able to activate pseudohypoxic responses and probably resulting in oxidative damage to DNA, genomic instability, and tumorigenesis (53). In that study, an Sdhb-Knock Down (KD) model was compared with a Sdha-KD one, and ROS increase was only observed in Sdhb KD cells. At that time, it was suspected that SDHA gene mutations were not associated with PPGL, an assumption that we demonstrated to be untrue several years later (3). On the basis of these observations, their conclusions were that loss of all but the SDHA subunits of SDH would produce such increase in ROS production, explaining SDHB/C/D–mediated tumorigenesis. In view of the actual knowledge on SDH genetics, the combination of these data with our current observations suggests that ROS production is a specific feature of SDHB-mutated cells, and may be responsible, not for SDH-related tumorigenesis in general, but for SDHB-related aggressive phenotype.

Our data suggest that the increased oxidative stress observed in Sdhb-deficient cells could be the consequence of the dysregulated iron/copper homeostasis. Indeed, redox-active iron strongly interacts with hydrogen peroxide through the Fenton/Haber–Weiss reaction. This usually slow reaction is catalyzed by labile iron, yielding mainly the highly reactive and toxic hydroxyl radical (•OH) from hydrogen peroxide (H₂O₂) and superoxide ions (•O₂⁻): iron(III) + •O₂⁻ → iron(II) + O_2; iron(II) + H₂O₂ → iron(III) + OH⁻ + •OH (57). Oxidative damage arising from excessive levels of ROS are observed in a variety of pathologies such as neurodegenerative diseases and cancer, and have been linked to malignant transformation, together with an imbalance in iron homeostasis (58). Similar assumptions can be made for copper (59). Our results are also consistent with a recent study demonstrating the role of histone H3-H4 as a copper reductase enzyme that binds Cu(II) and catalyzes its reduction to Cu(I). They showed that active H3-H4 is required for proper utilization of copper for mitochondrial respiration and SOD function. Thereby, the excess of mitochondrial Cu(I) observed here predominantly in Sdhb^−/− imCC suggests a greater inactivation of H3-H4 and thereby may explain the inhibition of Sod2 expression and the mitochondrial respiratory deficit observed following Sdhb-deficiency (60). In addition, the lack of ROS changes observed in Sdhd KO cells might be associated with a residual activity of Complex II in Sdhb KO cells. Indeed, while changes in SDH activity coupled to complexes I and III have been shown in a number of different systems to generate ROS, SDH itself has also been suggested to generate ROS when electron flow is compromised (61). It was recently suggested that cells lacking SDHB maintain some form of SDH subunit-containing complex (62), which maintains some level of activity. Although we have no data to validate this hypothesis, one could hypothesize that SDHB-deficient cells can generate ROS due to the retention of an “SDH-like” complex, while SDHD-deficient cells avoid this through complete loss of SDH.

In this regard, our findings are consistent, although different, with those recently proposed by Liu and colleagues (63). Using an Sdhb KD cell model, these authors also describe ROS accumulation associated with an increase in the intracellular labile iron pool. Surprisingly, these authors observed an increase in iron pool using the calcein probe while in our model, and in human PPGL, this experiment actually showed a decrease in Sdhb KO cells. A possible explanation stands in the fact that the model used in this study is, as the human PPGL studied, a true genetic KO with no SDHB expression, while the model used by Liu and colleagues is a KD in which 30% of SDHB protein expression remains. Hence, these different models point to similar pathways, but with major differences in the adaptive responses of the cells. Indeed, Liu and colleagues propose that the iron overload would be due to increased expression of TF and TFR2, while, as previously mentioned such overexpression was not observed in our model. We did observe a slight increase in TF mRNA levels in SDH tumors as compared with other types of mutated PPGL, but not in TFR2. However, in both cases, levels of expression were actually extremely low in all PPGLs and therefore probably not biologically relevant (Supplementary Fig. S4). Furthermore, these authors showed in the same model, that antioxidant responses were increased following SDHB KD, notably through the upregulation of the nuclear factor erythroid 2-related factor 2 (NRF2). We did not observe any activation of neither the NRF2 pathway nor superoxide dismutase (SOD) activity in any of our different cell types and these data were further validated by our analysis of the transcriptome of human PPGL. Besides, N-acetyl cysteine (NAC) treatment, an inhibitor of the NRF2 pathway, had no effect on the production of mitochondrial ROS by Sdhb^−/− cells. Hence, it is quite interesting to notice that although similar pathways seem to be affected in the Sdhb KO and KD cells (i.e., iron homeostasis, oxidative stress), it seems that the mechanisms and the adaptive responses involved are not similar.

Collectively, these data demonstrate that cells carrying a complete abolition of Sdhb gene are not able to neutralize the oxidative damage caused by the mitochondrial ROS excess, through the upregulation of any intrinsic anti-oxidant pathway. Because ROS generation seems to be a critical hallmark of SDHB loss, testing an antioxidant treatment that could be used for clinical purposes on our Sdhb^−/− model appeared relevant. Ascorbate offers many assets, as it presents few side effects in healthy tissues, even at pharmacologic concentrations and is easy to access. However, although at physiologic concentration (micromolar), ascorbate acts as an antioxidant, able to decrease ROS levels (64), it has been shown to have pro-oxidant actions when reaching high concentrations (65). In addition to its strong oxidative role, ascorbate also affects iron homeostasis, as it reduces ferric iron [iron(III)] in a cycle of reactions, leading to the fully oxidized form of ascorbate (DHA), ferrous iron [iron(II)], and eventually resulting in increased intracellular ROS levels (66, 67). Therefore, in a context of redox imbalance associated with an elevated pool of labile iron, ascorbate treatment could lead to unbearably high ROS levels for the cells and thus aggravate an existing oxidative stress burden. Use of ascorbic acid at pharmacologic concentrations was already suggested as a promising therapeutic strategy on various types of tumor cells exhibiting increased ROS levels, increased labile iron and metabolism impairments (58). All the more that, for the past few years many studies have demonstrated the efficacy of ascorbate as an anticancer treatment, killing cancer cells in vitro (68, 69) and slowing tumor growth in vivo (70). In compliance with these reports, our data reveal that high-dose ascorbate treatment led to a higher accumulation of ROS solely in the mitochondria of Sdhb^−/− imCC, resulting promptly in cell death, thus validating the increased sensitivity of Sdhb^−/− cells for ascorbate, mediated by ROS overload. Interestingly, Liu and colleagues also recently reported that ascorbate treatment increased apoptosis in SDHB KD cells in vitro and delayed the growth of tumors as well as of hepatic lesions in xenografted mice models in vivo (70). Numerous clinical trials are currently evaluating the efficacy of intravenous high-dose pharmacologic ascorbate injection for treating different cancer types (67, 71). In that context, ascorbate intravenous injections at pharmacologic doses appear as a worthy and highly promising therapeutic strategy to treat SDHB-deficient malignancies, which often exhibit a poor answer to current treatments strategies (radiotherapy, chemotherapy, targeted therapy) and could also be tested in combination with other therapies.

Overall, the comprehensive comparison of the first relevant Sdhb and Sdhd knock-out cellular models reveals the common and distinctive pathways used by these cells to adapt to these deficiencies, and highlight specific hallmarks associated with Sdhb mutations. Lifting the veil on a long-lasting question, we show that Sdhb-deficient cells exhibit stronger PHD and TET inhibition than their Sdhd-deficient counterparts, explaining their aggressive phenotype and mesenchymal morphology. We demonstrate that Sdhb^−/− imCC manifest a significantly exacerbated imbalance in copper and iron homeostasis, resulting in an increased labile iron pool, associated with significant ROS accumulation. High-dose ascorbate treatment of Sdhb^−/− imCC offers promising results, as it highly aggravated the oxidative stress burden endured by Sdhb-deficient cells, thus rising as a potential original therapeutic approach to treat malignant PPGL.

K. Kluckova reports grants from The Paradifference Foundation during the conduct of the study. A. Thakker reports grants and other support from Paradifference Foundation during the conduct of the study. L. Vettore reports grants from Cancer Research UK during the conduct of the study. D.A. Tennant reports grants from The Paradifference Foundation and grants from Cancer Research UK during the conduct of the study. No disclosures were reported by the other authors.

J. Goncalves: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing–original draft. S. Moog: Data curation, formal analysis, supervision, investigation, methodology. A. Morin: Resources, data curation, formal analysis, supervision, investigation, methodology. G. Gentric: Resources, data curation, formal analysis. S. Müller: Resources, data curation, formal analysis, supervision. A.P. Morrell: Data curation, formal analysis. K. Kluckova: Data curation, formal analysis. T.J. Stewart: Data curation, formal analysis. C.L. Andoniadou: Data curation, formal analysis, supervision. C. Lussey-Lepoutre: Data curation. P. Bénit: Resources, data curation, formal analysis. A. Thakker: Data curation, formal analysis, validation, methodology. L. Vettore: Data curation, formal analysis. J. Roberts: Resources, data curation, formal analysis, validation. R. Rodriguez: Resources, formal analysis, supervision, funding acquisition, validation. F. Mechta-Grigoriou: Conceptualization, resources, data curation, formal analysis, validation, methodology. A. Gimenez-Roqueplo: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, methodology. E. Letouzé: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration. D.A. Tennant: Conceptualization, resources, data curation, formal analysis, funding acquisition, validation, methodology. J. Favier: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration.

The authors thank all members of the Genetics Department, Biological Resources Center and Tumor Bank Platform, Hôpital Européen Georges Pompidou (BB-0033–00063), and the Metabolic Tracer Analysis Core (MTAC) at the University of Birmingham for access to their technology platform, as well as the London Metallomics Facility funded by the Wellcome Trust (grant reference 202902/Z/16/Z). We thank Dr Stijn J.M. Van Malderen for assistance with shift correction. This work was supported by The Plan Cancer, Epigénétique et Cancer (EPIG201303 METABEPIC), The Paradifference Foundation, la Ligue Contre le Cancer (Equipe Labellisée) and the Cancer Research for Personalized Medicine - CARPEM project (Site de Recherche Intégré sur le Cancer - SIRIC). The London Metallomics Facility is funded by the Wellcome trust (grant no 202902/Z/16/Z). S. Moog is the recipient of a fellowship from la Fondation pour la Recherche Médicale. J. Goncalves is the recipient of a fellowship from la Ligue Nationale contre le Cancer. The R.R. research group is funded by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement no 647973), the Fondation Charles Defforey-Institut de France, and Ligue Contre le Cancer (Equipe Labellisée).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.