Fumarate induces vesicular release of mtDNA to drive innate immunity

Mutations in fumarate hydratase (FH) cause hereditary leiomyomatosis and renal cell carcinoma1. Loss of FH in the kidney elicits several oncogenic signalling cascades through the accumulation of the oncometabolite fumarate2. However, although the long-term consequences of FH loss have been described, the acute response has not so far been investigated. Here we generated an inducible mouse model to study the chronology of FH loss in the kidney. We show that loss of FH leads to early alterations of mitochondrial morphology and the release of mitochondrial DNA (mtDNA) into the cytosol, where it triggers the activation of the cyclic GMP–AMP synthase (cGAS)–stimulator of interferon genes (STING)–TANK-binding kinase 1 (TBK1) pathway and stimulates an inflammatory response that is also partially dependent on retinoic-acid-inducible gene I (RIG-I). Mechanistically, we show that this phenotype is mediated by fumarate and occurs selectively through mitochondrial-derived vesicles in a manner that depends on sorting nexin 9 (SNX9). These results reveal that increased levels of intracellular fumarate induce a remodelling of the mitochondrial network and the generation of mitochondrial-derived vesicles, which allows the release of mtDNAin the cytosol and subsequent activation of the innate immune response.

(1) The link between a metabolite and mtDNA is new finding. I am not aware of metabolites linked to mtDNA. Previous results have largely been confined to TFAM +/-or chemotherapeutic agents. This opens the possibility of metabolites under physiological or pathological conditions activating mtDNA dependent cGAS-STING inflammatory response. A key aspect of the paper is that elevation of fumarate but not succinate triggers release of mtDNA.
(2) Previous studies have surmised a variety of mechanisms to suggest how mtDNA is released. None of them to this reviewer are satisfactory. All of them rely on pathways linked to cell death including BAX/BAK, PTP or VDAC. This has raised the possibility that mtDNA is released in dying cells. How mtDNA is released from matrix without releasing cytochrome c has always puzzled me. However, this study links MDVs as a possible mechanism of mtDNA release. I like the Snx9-dependent release of mtDNA experiments and this could be broad mechanism by which mtDNA is released under a variety of stimuli beyond fumarate elevation.
I have a few key experiments that would bolster this excellent paper.
(1) Could the authors reconstitute the FH deficient cells with cytosolic FH? Does the fumarate action dependent on cytosolic elevation of fumarate?
(2) Previously the authors have shown that fumarate likely through succination causes defects in the mitochondrial respiratory chain in mouse cells. Human cancer cells with FH deficiency also demonstrate mitochondrial complex I defects. Could the authors reconstitute their cells with NDI1 that complements the loss of mitochondrial complex I? Does elevation of fumarate lead to complex I defects that are necessary for mtDNA release? How do the authors think the elevation of fumarate triggers MDVs?
(3) I would like to see RNAseq with siRNA against Snx9 in FH deficient cells. How much the inflammatory response is dependent on Snx9.
(4) cGAS-STING pathway occurs when cGAS-binds to STING dimers residing in the ER membrane leading to STING activation and trafficking to ER-GOLGI compartment where TBK1 is recruited. Do they observe Snx9 positive MDVs fusing or proximal to ER-GOLGI compartment? (5) Beyond fumarate and succinate, the other key members of this metabolite family are alpha-ketoglutarate (aKG) and L-2hydroxyglutarate (L-2HG). aKG is required for 75 plus dioxygenases including enzymes that participate in DNA, RNA, histone demethylases as well as prolyl hydroxylases for HIF and collagen regulation. Fumarate, succinate and L-2HG inhibit these enzymes. Thus, I am curious whether exogenous addition of cell permeable (octyl) L-2HG or aKG would trigger cGAS-STING inflammatory response. I would test whether this occurs before doing any experiments on mtDNA release with these two metabolites.
Referee #3 (Remarks to the Author): The manuscript submitted by Zecchini et al focuses on the influence of Fh1 tumour suppressor and the chronology of its loss in the kidney, by developing a novel mouse model. By using this mouse model FH (in which FH loss is induced) and cell lines taken from kidneys of these mouse, the authors suggest that loss of Fh1 leads to fumarate accumulation, which results in mitochondrial network remodeling, mtDNA release into the cytosol, and activation of the innate immune response. Additionally, this work investigates the role of MDVs in fumarate accumulation, and suggests the fumarate phenotype is mediated through MDVs in a Snx9 dependent manner.
The manuscript is well explained and well written, and the experiments are mostly well planned and focused, which overall reflects the hard work of the research team. Nevertheless, I have some concerns regarding some of the methodological and analytical aspects of the research, and these should be addressed by the authors. Also, additional experiments should be done to further validate the suggested mechanism.
If the authors can address the concerns of this reviewer, I believe the manuscript potentially merits publication in Nature.

General comments:
The major concern is the physiological relevance of this work which is unclear. What would be the outcome of inhibition of the fumarate Snx9-Irf3 axis? Will inhibition of this axis effect tissue physiological outcome? Moreover, Fh1 KO causes changes in respiratory capability of the cells (Warburg effect) so it is not surprising that the expression of many genes change. In other words, it is not clear whether the loss of Fh1 directly causes the release of DNA which according to the authors, is suggested to affect the immune response. We know that loss of Fh1, causes in the long run, accumulation of fumarate, and effects histone di-methylases. The authors do not characterise the vesicles they claim are formed, whether these are specifically loaded with mtDNA, and released.
Reviewer: Fig 1-(c)-The change in fumarate (the most relevant metabolite) levels, does not appear to be significant, and with such a change, one would not expect the outcome suggested by the authors. While the other metabolites are affected, this is still not convincing. We agree with the claim that this mouse is an excellent model, but results must be refined.
Reviewer: There is no reference in the text to the differences in the results between day 5 and day 10; on day 5 in most pro-inflammatory cytokines there is no significant different between WT and KO mouse. What is the possible explanation?
Reviewer: The Western blot in in EtOH, 3 days post induction shows a decrease in Fh1 levels even though it is a control and no influence should be observed.
Reviewer: Fig 2-(e)-The authors need to demonstrate by FISH the presence of other mitochondrial genes beside TOM20 within the cytosol. It also important to characterize the physiological state of the cells and mitochondria, with respect to respiration and membrane potential. Referee #4 (Remarks to the Author): In the manuscript entitled "Fumarate induces mtDNA release via mitochondrial-derived vesicles and drives innate immunity", Zecchini and authors show that loss of the tumor suppressor fumarate hydratase leads to increased mtDNA release and subsequent activation of interferon stimulated genes (ISGs). They create an inducible model of FH1 deletion to investigate the temporal dynamics of fumarate induction and the downstream effects on mitochondrial dynamics and innate immunity. They use epithelial cell lines derived from the mouse kidney for the majority of their studies and show that loss of fumarate increases the number of swollen and elongated mitochondria, the release of mitochondrial DNA via mitochondrial-derived vesicles, and the induction of the cGas-Sting-Tbk1 pathway. Mitochondrial-derived vesicles are highly understudied, and these findings indicate a new context and physiological role for these structures. While the authors have designed an inducible model to dissect the acute changes and immediate effects of FH1 loss, the timeline of events does not fully support the conclusions presented. The authors suggest that fumarate accumulation causes the effects on mitochondria, mtDNA release, and p-Stat1, but all these effects are seen between 3-6 days after induced FH1 loss, while fumarate is not increased until much laterbetween days 10-15. Later in their studies they add exogenous monomethylfumarate (MMF) which also affects mitochondria dynamics, mtDNA and pStat1 and ISGs. Does the dose of MMF correspond to the amount of fumarate at days 3-6, when there appears to be very little? Can the authors explain how fumarate is mediating these effects prior to its accumulation at days 3-6? Can low doses of MMF have similar effects to higher doses to recapitulate the effects of a small increase in fumarate? Because of this discrepancy, the data in the MMF treated cells is less convincing. The authors should show more of the mechanism in the inducible FH1 KO lines including the EtBr experiment to eliminate mtDNA as well as the effects on mitochondrial derived vesicles. Finally, how does this mechanism contribute to the tumor suppressive function of FH1 in HLRCC? Additional experiments to address this point would strengthen the relevance of these findings.
Referee #5 (Remarks to the Author): Zecchini et al. report that deletion of fumarate hydratase (FH) in mouse kidneys and murine kidney epithelial cell lines leads to an accumulation of fumarate, subsequent disruptions to mitochondrial morphology, and the release of mtDNA into the cytosol. They propose that cytosolic mtDNA engages the cGAS-STING signaling axis leading to elevated expression of type I interferon responses in FHdeficient cells in vitro and both interferon and pro-inflammatory responses in FH-deficient kidneys in vivo. Furthermore, the authors show that addition of exogenous monomethylfumarate (MMF) is sufficient to phenocopy the alterations to mitochondrial morphology, mtDNA release, and innate immune activation in vitro. Finally, they report that mtDNA release is not dependent on Bax-Bak or other canonical players, but instead occurs via the Snx9-dependent generation of mitochondrial derived vesicles (MDVs). Depletion of Snx9 is sufficient to mitigate mtDNA release into the cytosol and inhibit activation of the innate immune signaling. Overall, this is an interesting study with robust datasets to support that FH deficiency triggers innate immune responses through mitochondrial stress. Although proposed by others, this paper also reveals that MDVs may be a mechanism of mtDNA release. Despite these positives, there are significant mechanistic gaps that must be further explored to solidify the proposed pathways. Specific comments follow below: Major Comments: 1. Fig 1: The FH conditional deletion model here provides a powerful tool to examine the effects of FH loss on immune signatures in a tissue specific manner. However, the authors do not provide any data from other organs. Do the levels of fumarate increase in other organs, and if so, is this correlated with increasing immune signatures? Additionally, the authors state that there are no gross morphological changes in kidneys 10 days after tamoxifen-mediated Cre deletion of Fh1. A careful analysis of the H&E image of Figure 1 (d) seems to show many more nuclei between kidney cells. Moreover, even though FH expression is markedly down 5 days after tamoxifen exposure, the immune signature doesn't come up strongly until day 10. Both of these lines of evidence are indicative of elevated kidney-infiltrating immune cells, as opposed to a kidney epithelial cell-intrinsic immune response. Tissue RNAseq data will not distinguish kidney epithelial cell-intrinsic signatures of inflammation versus those generated by infiltrating immune cells. In agreement with the latter, the pathway analysis and gene expression signature seem to represent a generally more proinflammatory response as opposed to type I interferon. For example, IL-6, CD14, C3, TLR2, etc. are more consistent with a signature of macrophage/monocyte/neutrophil infiltration, as opposed to a cell-intrinsic type I interferon and interferon stimulated gene response. Given that the conditional knockout approach relies on the ubiquitously expressed ROSA-ERCre that will target FH in many body cells and tissues, it is important that the authors do more work to define the in vivo mechanisms.
2. The evidence in support of cGAS-STING-IFN-I pathway activation is generally not sufficient. Although crosses of FH cKO mice onto cGAS or STING KO mice to show ablation of the immune signature in vivo would be the most convincing, cGAS/STING inhibitors are readily available for mouse in vivo pre-clinical studies. These would be great to include as a way to link the in vivo phenotypes more cohesively with the cellular mechanisms presented in the later figures. In addition, the authors do not sufficiently explain how the kidney epithelial cell lines were immortalized and derived. This is important because many immortalization/transformation procedures render the cGAS/STING pathway non-functional at several levels (see paper from Stetson et al (Science 2015) on SV40 lgT and other oncogenes downregulating STING signaling). Finally, the authors never show that genetic ablation of cGAS/STING by siRNA, Crispr, etc is sufficient to ablate the interferon and ISG signature observed in vitro. The RU.521 experiments of Figure 2 (m) and 3 (n) do not convincingly show that cGAS inhibition abolishes the induction of immune genes after FH deletion of MMF treatment. It would be much more convincing to target these pathways using siRNA or other means (the authors use cGAS, STING, Rig-I siRNAs in Extended figure 6 but never examine immune gene expression after knockdown in FH deficient or MMF treated cells). Given that mtRNA has also been noted as a ligand driving immune activation downstream of mitochondrial stress, it is important to rule out the mtRNA sensors Rig-I/Mda5 and perhaps other nucleic acid sensing TLRs. This could be easily accomplished using the cell lines and siRNAs reported in this paper. 3. Comprehensive work from Dr. Frezza's lab and others has revealed fumarate as an oncometabolite in HLRCC via its actions on Keap1 and the NRF2 pathway, among others. Several recent papers have revealed that NRF2 pathway activation directly represses STING and counter regulates type I interferon responses (PMID: 30158636, 31487581, 31487581, 31487581). Fumarate accumulation after FH loss upregulates the NRF2 target Hmox1 in the cells used here (Extended Data Figure 2 (c)), so this suggests that NRF2 is activated. It is thus unclear how fumarate-dependent elevations in NRF2 would support robust STING activation. It is therefore important that the authors examine these connections and more thoroughly and explore innate immune pathways distinct from cGAS-STING that might trigger the fumarate-dependent expression of innate immune genes in vitro (and in vivo). 4. The Snx9-MDV pathway as a route for mtDNA release in FH deficient cells is interesting. It is unclear, however, exactly how membrane bound mtDNA would be sensed by cGAS in the cytosol in a cell-autonomous manner. Do the authors see cGAS around the PDH/DNA+ MDVs? Is it possible these vesicles might be released and fuse with neighboring cells in vitro (or immune cells in vivo), leading to a paracrine activation of cGAS? It may be interesting to consider supernatant transfer from FH deficient cells onto WT and cGAS knockdown cells to further explore/clarify the mechanisms of mt-nucleic acid sensing. Additionally, the authors show that knockdown of Snx9 leads to a reduction in cGAS/STING pathway activation. Does this inhibition also lead to a reduction in expression of the ISGs and pro-inflammatory cytokines and chemokines reported in Figure 1 (f,g)? Similarly, does Snx9 knockdown have any effect on mitochondrial morphological changes? 5. The authors claim in the discussion that this paper represents the first report of a disease relevant mutation causing release of the mtDNA into the cytosol. This is a significant overstatement. Recently, human mutations in ATAD3A causing bona fide mitochondrial disease have been shown to liberate mtDNA into the cytosol and engage cGAS. Moreover, loss of TDP-43, CLPP, YME1L, OPA1 and others, all of which are linked to human disease, induce mtDNA release and activation of innate immunity. Perhaps the authors meant to restrict their discussion of impact to kidney diseases, but even so, loss of Tfam in kidney tubule cells has been shown to induce mtDNA leakage and cGAS-STING inflammation in vivo and in vitro. There is no evidence provided here to link the mtDNA release and immune phenotypes shown to HLRCC, so I believe it would be prudent to tone down these statements of novelty and properly reference the literature. It will not decrease the impact of the author's findings to reference these recent papers. Technical Comments: 1. A significant portion of the analyses herein rely on manual scoring of immunofluorescence images. While manual scoring itself is not itself problematic, the individual(s) conducting the scoring should be blinded to the experimental groups wherever possible in order to minimize the impact of unintentional bias in the results. This is particularly important for subjective measures, such as mitochondrial morphology, where there is no clearly defined delineation between different morphological classifications. 2. Figures 3 (f-h, j, n), 4 (j) and Extended Data Figures 3 (j-l), 4 (i-l), 5 (a, h-k) -For these figures, each individual replicate for the control group has been set to one. This has the effect of artificially reducing the variance of the control group to zero, resulting in an inflated level of significance for groups that are being directly compared to the controls. Control values should be normalized to the mean of the control group, rather than set to one on an individual basis, preserving the variance and allowing an accurate significance level to be calculated. , 5 (f,g) -Each of these pairs of figures seem to present the same data (e.g., same means, same p values), just with two slightly different figure formats. What is the rationale for this duplication of data?

Reviewer #1 [mito biology/cancer metabolism]
This is an interesting paper that links endogenous fumarate elevation to release of mtDNA resulting in the activation of cGAS-STING dependent inflammatory response. I like the findings in the paper as it uncovers two important observations.
(1) The link between a metabolite and mtDNA is new finding. I am not aware of metabolites linked to mtDNA. Previous results have largely been confined to TFAM +/-or chemotherapeutic agents. This opens the possibility of metabolites under physiological or pathological conditions activating mtDNA dependent cGAS-STING inflammatory response. A key aspect of the paper is that elevation of fumarate but not succinate triggers release of mtDNA.
(2) Previous studies have surmised a variety of mechanisms to suggest how mtDNA is released. None of them to this reviewer are satisfactory. All of them rely on pathways linked to cell death including BAX/BAK, PTP or VDAC. This has raised the possibility that mtDNA is released in dying cells. How mtDNA is released from matrix without releasing cytochrome c has always puzzled me. However, this study links MDVs as a possible mechanism of mtDNA release. I like the Snx9-dependent release of mtDNA experiments and this could be broad mechanism by which mtDNA is released under a variety of stimuli beyond fumarate elevation. I have a few key experiments that would bolster this excellent paper.
The authors thank the reviewer for their positive feedback, detailed discussion and pertinent suggestions. We believe we have addressed all the comments in the best possible way and hope the reviewer will be satisfied with our revised manuscript. To help the review of this document, we have incorporated panels from the figures as they appear now in the revised version of the manuscript. Please, note that the order of the sub-panels may differ from the figure in the manuscript for presentation purposes.
(1) Could the authors reconstitute the FH deficient cells with cytosolic FH? Does the fumarate action dependent on cytosolic elevation of fumarate?
The general issue of subcellular compartmentalization of metabolites, and, specifically, fumarate in the context of this paper, is very pertinent. As the reviewer suggested, we expressed a cytosolic form of Fh1 in our Fh1-deficient cell line to generate Fh1 -/-CL1+cytoFh1 (R#1 Figure 1a, b). We observed a partial rescue of mitochondrial basal respiration (R#1 Figure 1d), a reduction in the levels of fumarate as well as a partial rescue in the levels of S-(2-succinyl)cysteine (2SC) (R#1 Figure 1e, f). In line with the original data we presented in the manuscript, we also observed a decrease in the number of cytosolic DNA foci in the Fh1 -/-CL1+cytoFh1 cells (R#1 Figure 1g) together with a reduction in the activation of Tbk1 and Irf3 (R#1 Figure  1h), and a partial rescue in the transcriptional activation of downstream Interferon-stimulated genes (ISGs) (R#1 Figure 1i).
The overall reduction in fumarate observed in Fh1 -/-CL1+cytoFh1 cells likely reflects a reduction of both mitochondrial and cytoplasmic pools because of mitochondrial fumarate moving into the cytosol. Importantly, however, immunofluorescence staining with the anti-2SC antibody confirmed a remaining positive signal in the mitochondria in the cytosolic FH rescue Fh1 -/-CL1+cytoFh1 cells (R#1 Figure 1a), strongly suggesting an accumulation of mitochondrial fumarate in these cells. Therefore, we conclude that the cytosolic DNA foci phenotype observed in Fh1 -/-CL1+cytoFh1 can be attributed to the remaining levels of mitochondrial fumarate. Of note, the re-expression of the full-length FH, which leads to its mitochondrial localization, fully rescued all the different phenotypes analyzed, including respiration, fumarate and 2SC levels, cytosolic DNA foci number, as well as the inflammation phenotype compared to the partial rescue observed in Fh1 -/-CL1+cytoFh1 cells, highlighting the importance of the mitochondrial levels of fumarate in the different observed phenotypes. We have amended the manuscript and included this new important data (revised manuscript Extended Data Fig. 7).

Author Rebuttals to Initial Comments:
the loss of mitochondrial complex I? Does elevation of fumarate lead to complex I defects that are necessary for mtDNA release?
We thank the reviewer for requesting this critical control. In order to answer this comment, and as suggested by the reviewer, we constitutively expressed NDI1 in our Fh1-deficient cells to generate the Fh1 -/-CL1+EGFP:NDI1 line (R#1 Figure 2a, b). Despite a partial rescue of respiration, which was still significantly lower than control (R#1 Figure 2c) due to persistent defects in complex II and TCA cycle (as shown before in Tyrakis et al., Cell Reports 2018), NDI1 expression did not significantly alter the release of mitochondrial nucleic acids into the cytosol (R#1 Figure 2d), nor the activation of the downstream cytosolic nucleic acid sensing pathway as indicated by unaffected Tbk1 and Irf3 phosphorylation (R#1 Figure 2e) or ISGs transcriptional activation (R#1 Figure 2f). Altogether, these data suggest that the phenotype we observe in Fh1-deficient cells is not triggered by an isolated defect in complex I. We have amended the manuscript accordingly to incorporate the data relating to NDI1 and complex I (revised manuscript Extended Data Fig. 5). How do the authors think the elevation of fumarate triggers MDVs?
Based on the new results, we hypothesize that a fumarate-dependent succination mechanism within the mitochondria could be involved in the formation of MDVs. Indeed, we now show that MMF treatment, which recapitulates the effects of FH loss and mtDNA release in the cytosol, induces a progressive increase of succinated proteins, accumulating particularly inside mitochondria (revised version Extended data Fig. 6e and R#1 Figure 3). Interestingly, the intensity of succination, here monitored by an anti-2SC antibody, correlated with the size of swollen mitochondria. Absolute metabolite quantification by LC-MS confirmed a rapid accumulation of 2SC in MMF-treated cells comparable to that observed in the inducible line iFh1 -/-CL29 (R#1 Figure 3). Moreover, new experiments performed in mtDNA-depleted Rho 0 (ρ0) cells showed that MMF treatment in these cells did not trigger the formation of TOM20 -PDH + MDVs (revised manuscript Extended Data Fig. 12e, f and R#1 Figure 4). This unexpected but crucial finding suggests that mtDNA is required for the formation of these particular MDVs. Interestingly, it has been reported that nucleoid-related proteins like TFAM and TWINKLE, can be succinated in human FH-deficient tumours and this has been associated with a loss of mtDNA copy number (Crooks et al., Science Signalling 2021). While it is not clear at this point which protein is the main culprit, as cysteine residues of many proteins are succinated upon FH loss, this result lends support to the hypothesis that fumarate-driven succination may trigger MDVs formation. This hypothesis is further corroborated by the finding that only fumarate, but not other metabolites such as succinate, αKG, and 2HG, can elicit the formation of MDVs and mtDNA-release-dependent innate immunity (revised manuscript Extended Data Fig. 8). We have added this discussion in the revised version of the manuscript. (3) I would like to see RNAseq with siRNA against Snx9 in FH deficient cells. How much the inflammatory response is dependent on Snx9.
In this study, we show that fumarate triggers the transcriptional activation of a specific subset of ISGs (see RNASeq dataset and Fig. 1f). ISGs expression is modulated by complex and intricate regulatory feedback loops (both positive and negative), often regulated at the post-translational level, acting at different times for different targets that make it difficult to assess this response as a whole at a single time point. In our manuscript, we observe different expression dynamics for each of our transcriptional targets that is likely to reflect such a complex transcriptional regulation. The phosphorylation of the downstream elements of the cascade i.e. Tbk1 and Irf3 might arguably constitute a more stable and reliable read-out of the pathway activation. Consequently, unless a time course is used, we believe that a genome-wide approach would yield limited information about the specific inflammatory pathway(s) that might be affected.
However, to better documents the inflammatory response observed upon loss of Snx9, a wider panel of targets where Snx9 expression is knocked-down following treatment with MMF is now presented in Extended Data Fig. 12g and R#1 Figure 5.

Reviewer #3 [fumarate/mito]
The manuscript submitted by Zecchini et al focuses on the influence of Fh1 tumour suppressor and the chronology of its loss in the kidney, by developing a novel mouse model. By using this mouse model FH (in which FH loss is induced) and cell lines taken from kidneys of these mouse, the authors suggest that loss of Fh1 leads to fumarate accumulation, which results in mitochondrial network remodeling, mtDNA release into the cytosol, and activation of the innate immune response. Additionally, this work investigates the role of MDVs in fumarate accumulation, and suggests the fumarate phenotype is mediated through MDVs in a Snx9 dependent manner.
The manuscript is well explained and well written, and the experiments are mostly well planned and focused, which overall reflects the hard work of the research team. Nevertheless, I have some concerns regarding some of the methodological and analytical aspects of the research, and these should be addressed by the authors. Also, additional experiments should be done to further validate the suggested mechanism. If the authors can address the concerns of this reviewer, I believe the manuscript potentially merits publication in Nature.
The authors thank the reviewer for the careful reading of the manuscript, detailed discussion and pertinent suggestions. We have now addressed all comments raised by the reviewer, which we trust has significantly increased the quality of our manuscript. To help the review of this document, we have incorporated panels from the figures as they appear in the revised version of the manuscript. Please, note that the order of the sub-panels may differ from the figure in the manuscript for presentation purposes.

General comments:
The major concern is the physiological relevance of this work which is unclear. What would be the outcome of inhibition of the fumarate Snx9-Irf3 axis? Will inhibition of this axis effect tissue physiological outcome? Moreover, Fh1 KO causes changes in respiratory capability of the cells (Warburg effect) so it is not surprising that the expression of many genes change. In other words, it is not clear whether the loss of Fh1 directly causes the release of DNA which according to the authors, is suggested to affect the immune response.
We apologise for the ambiguity in our manuscript and we have attempted to answer the reviewer's comment as best as possible below. The text has also been modified accordingly to better highlight the direct contribution of mtDNA release to the immune response, and new set of experiments/analyses have been performed to establish a link between FH-deficient renal cancer and the immune response.
(1) What would be the outcome of inhibition of the fumarate Snx9-Irf3 axis? This is an interesting point which we also pondered on. Our study has uncovered a novel function of fumarate (via the loss of Fh1 activity) in the activation of Pattern Recognition Receptor (PRR) pathways both in cellulo and in vivo, that links Snx9-dependent mitochondria-derived vesicles (MDVs) formation to the downstream activation of the Sting/Tbk1/Irf3 signalling cascade that ultimately results in the stimulation of an inflammatory response via sustained ISGs transcriptional activation. As mutations in Fumarate Hydratase (FH) are associated with Hereditary Leiomyomatosis and Renal Cell Cancer (HLRCC), which results in the development of an aggressive form of kidney cancer, we hypothesise that a chronic low-grade inflammation resulting from fumarate-dependent activation of the Snx9-Irf3 pathway could contribute to tumorigenesis in these patients. In support of this hypothesis, we provide additional data in the revised version of the manuscript that show a robust immune signature associated with tumour tissue from HLRCC patients (R#3 Figure 1a, b), corroborating our in vivo and in vitro data. We also showed using deconvolution from bulk transcriptomics data (see Material and Methods for the experimental details) that the contribution of immune cells to this signature is higher in FH-deficient HLRCC tumours compared to other renal tumour types (R#3 Figure 1c); indicating an immune environment in FH-deficient tumours. In addition, we showed the presence of IL6 and IL10 in the supernatant of HLRCC tumour tissue (R#3 Figure 1d), corroborating the activation of the innate immune response. Together, this new set of data supports the critical role of this pathway not only in vivo in our mouse model, but also in patients.
We agree with the Reviewer that it would be interesting to elucidate the physiological outcome of the inhibition of this pathway, but this will require the generation of new mouse models (e.g. Fh1-deficient mouse model crossed with cGAS-KO or Sting-KO mouse model), which is not feasible in the time allowed for the revision of this manuscript, and we think beyond the scope of this initial study. It is worth noting here that currently, none of the mouse models of Fh1 loss leads to overt carcinoma, which would also complicate the design of this experiment. We are currently working to overcome this issue, and we hope to address this relevant question in a near future. The data presented below has now been incorporated in the revised manuscript (Fig. 5a-d). (2) Will inhibition of this axis effect tissue physiological outcome?
Our study shows that the Snx9-Irf3 signalling pathway is stimulated in Fh1-deficient cells and in kidney from our new generated inducible mouse model. In this revised version of our manuscript, we showed that the inhibition of the Snx9-Irf3 axis, either via the silencing of Snx9 or the in vivo inhibition of cGAS (revised Fig. 3n and ED Fig. 9g) leads to a significant reduction in the ISGs expression, suggesting that the inhibition of this pathway could lead to an anti-inflammatory effect that could contribute to tumour suppression. Similarly to our reply to question (1) of this reviewer, elucidating the in vivo outcome of inhibition of this pathway is beyond the scope of this initial study and will be investigated in a follow-up study.
(3) Moreover, Fh1 KO causes changes in respiratory capability of the cells (Warburg effect) so it is not surprising that the expression of many genes change. In other words, it is not clear whether the loss of Fh1 directly causes the release of DNA which according to the authors, is suggested to affect the immune response.
This is an appropriate comment, which we would like to elaborate on here. Several crucial findings support the specific role of Fh1 loss and increased levels of fumarate in causing the release of mtDNA to trigger the innate immune response. First, to rule out that the observed mtDNA release is a general consequence of TCA cycle inhibition and mitochondrial dysfunction (and the "Warburg effect" mentioned by the referee), we duplicated most of our experiments in cells lines harbouring a deletion of succinate dehydrogenase b (Sdhb), another component of the TCA cycle and complex II of the electron respiratory chain, also linked to hereditary cancers. As reported by Cardaci and co-workers, the loss of Sdhb activity results in a Warburg-like phenotype and a near-complete loss of oxygen consumption (Cardaci et al., Nat Cell Biol, 2015). However, unlike the loss of Fh1 activity, Sdhb deletion did not induce the cytosolic release of mtDNA, Tbk1/Irf3 activation or transcriptional activation of ISGs (revised manuscript Extended Data Fig. 4). The lack of an inflammatory response in Sdhb-deficient cells was corroborated by the absence of immune infiltrates in SDH-deficient renal tumours in our new analyses; in contrast with the significant inflammatory response in FH-deficient tumours (revised manuscript Extended Data Fig. 5c). Second, we also showed that the expression of the NADH oxidase NDI1, which rescues complex I defects observed in Fh1-deficient cells (Tyrakis et al., Cell Reports, 2017 and response to referee 1, point (2)), did not prevent mtDNA release, activation of TBK1/Irf3 nor the transcriptional upregulation of ISGs, corroborating the prerequisite and specificity of Fh1 loss and fumarate accumulation -rather than a bioenergetic defect -for mtDNA release driving the immune response. In further confirmation of this hypothesis, our study reveals that the observed effect is mediated specifically by fumarate but not succinate, which accumulates upon loss of Sdhb. Indeed, in contrast to fumarate treatment, succinate treatment showed no effect on the cytosolic release of mtDNA, Tbk1/Irf3 phosphorylation nor ISGs transcriptional activation (revised manuscript Extended Data Fig. 8a-h). In the revised version of the manuscript, we have also extended the treatment to other known "oncometabolites", namely αKG and 2HG (see response to referee 1, point (5)). These metabolites are also generated due to alterations to the TCA cycle. However, treatment with these metabolites had no effect on our specific read-outs (revised manuscript Extended Data Fig. 8i-l). Finally, we also show that re-expression of full-length and also cytosolic-only Fh1 in Fh1-deficient cells rescues the phenotype (revised manuscript Extended Data Fig. 7). Of note, the cytosolic rescue of Fh1, which still accumulates mitochondrial fumarate even though to a lower level than the Fh1 -/-, reduces only partially the inflammatory phenotype, thus ascribing to mitochondrial fumarate the trigger of the cascade we describe. Altogether, these results indicate that the phenotype we observe in Fh1-deficient cells is not triggered indirectly by changes in the respiratory capability of the cell but are due to the specific loss of Fh1 activity and fumarate accumulation. We thank the reviewer for this useful comment and trust that the new data described here and that we have now incorporated in the revised manuscript has alleviated the reviewer's concerns.
(4) We know that loss of Fh1, causes in the long run, accumulation of fumarate, and effects histone dimethylases. The authors do not characterise the vesicles they claim are formed, whether these are specifically loaded with mtDNA, and released.
We agree with the reviewer that the specific loading of vesicles with mtDNA is an important point for this work. Mitochondrial-derived vesicles (MDVs) are characterised by their size but, more importantly, by cargo specificity. Indeed, MDVs only harbour a subset of mitochondrial proteins, which allows their identification and characterization compared to mitochondria. In the manuscript, we have shown that upon fumarate accumulation, cells exhibited the presence of MDVs in the cytosol compared to control cells. These MDVs are characterized by the presence of the mitochondrial matrix marker, Pyruvate Dehydrogenase (PDH), but negative for the outer mitochondrial membrane marker, TOM20, two well-known markers of different types of MDVs (Sugiura et al., EMBO J 2014, König et al., Nat Cell Biol. 2021, Soubannier et al., Curr Biol. 2012 To the question of whether the vesicles are actually loaded with mtDNA, our original microscopy data showed co-localisation of DNA foci with PDH, but not TOM20, in the vicinity of MMF-treated mitochondria, indicating that these DNA foci contained in MDVs are likely of mtDNA origin (revised manuscript Fig. 4c and image below, red arrows).
increase. There is a time lag before the effects of Fh1 deletion can be observed because, whilst the locus will be affected, some protein is still present within the cells. This explains why the effect in vivo is stronger at day 10 compared to day 5. In addition, secondary effects, such as activation of downstream signalling cascades and regulatory feedback loops are triggered later on. The molecular events we report here are a snapshot of the transcriptional landscape at days 5 and 10. A more refined picture would require many more time points. We hope that these more detailed explanations address the question about the difference between day 5 and day 10. (4) Authors: Figure 2-(e) Representative confocal images of mitochondrial morphology and DNA foci in iFh1fl/flCL29 cells treated with vehicle (EtOH) or 4-OHT (iFh-/-CL29) for 24 hours at day 6, 10 and 15 post-induction. Mitochondria and DNA were labelled using anti-TOM20 and anti-DNA antibodies, respectively. White arrows indicate cytosolic DNA foci. Scale bar: 10 μm.
Reviewer: Fig 2-(e)-The authors need to demonstrate by FISH the presence of other mitochondrial genes beside TOM20 within the cytosol. It also important to characterize the physiological state of the cells and mitochondria, with respect to respiration and membrane potential.
Based on this comment and the other comment of the same reviewer, we realized that there is some ambiguity regarding this specific point in the manuscript and will attempt to clarify the issue below.
Following the observation of nucleic acid fragments in the cytosol of Fh1-deficient cells and given the striking morphology of Fh1-deficient mitochondria, we hypothesized that mitochondrial material may be leaking into the cytosol. We used digital PCR (ddPCR) to demonstrate and quantify the presence of mtDNA fragments of mitochondrial origin in the cytosol using primers that cover three different regions of the mitochondrial genome -namely the D-loop, Co3 and ND1 loci. Although these three loci do not cover the entirety of the mitochondrial genome, they are almost equidistant within the circular mtDNA molecule (and thus provide a good coverage; R#3 Figure 7) and the experiments we performed confirmed the accumulation of these mtDNA fragments into the cytosol of Fh1-deficient cells and under other conditions of fumarate accumulation.
In the manuscript entitled "Fumarate induces mtDNA release via mitochondrial-derived vesicles and drives innate immunity", Zecchini and authors show that loss of the tumor suppressor fumarate hydratase leads to increased mtDNA release and subsequent activation of interferon stimulated genes (ISGs). They create an inducible model of FH1 deletion to investigate the temporal dynamics of fumarate induction and the downstream effects on mitochondrial dynamics and innate immunity. They use epithelial cell lines derived from the mouse kidney for the majority of their studies and show that loss of fumarate increases the number of swollen and elongated mitochondria, the release of mitochondrial DNA via mitochondrial-derived vesicles, and the induction of the cGas-Sting-Tbk1 pathway. Mitochondrialderived vesicles are highly understudied, and these findings indicate a new context and physiological role for these structures. While the authors have designed an inducible model to dissect the acute changes and immediate effects of FH1 loss, the timeline of events does not fully support the conclusions presented. The authors suggest that fumarate accumulation causes the effects on mitochondria, mtDNA release, and p-Stat1, but all these effects are seen between 3-6 days, after induced FH1 loss, while fumarate is not increased until much later -between days 10-15. Later in their studies they add exogenous monomethylfumarate (MMF) which also affects mitochondria dynamics, mtDNA and pStat1 and ISGs. Does the dose of MMF correspond to the amount of fumarate at days 3-6, when there appears to be very little? Can the authors explain how fumarate is mediating these effects prior to its accumulation at days 3-6? Does the dose of MMF correspond to the amount of fumarate at days 3-6, when there appears to be very little?
We thank Reviewer 4 for their comments, and we hope that they will be satisfied by our new set of experiments addressing their concerns. We have incorporated in the response the figures as they appear now in the revised version of the manuscript. To help the review of this document, we have incorporated panels from the figures as they appear now in the revised version of the manuscript. Please, note that the order of the sub-panels may differ from the figure in the manuscript for presentation purposes.
Based on the reviewer's first comment, which is similar to the comment of Reviewer 2 specific point 1, we realise there might be some ambiguity regarding the effects of fumarate accumulation and the timing of events. This is indeed an important issue and we apologise for not making it clearer.
Upon loss of Fh1 activity, fumarate cannot enter the enzymatic reaction that converts it into succinate but, instead, reacts chemically with the thiol side chain within cysteine in proteins via a Michael addition reaction (Blatnik et al., Diabetes 2008). This chemical modification that adds fumarate to cysteine residues is called succination. This non-enzymatic post-translational modification of cysteine also generates S-2(succinyl)cysteine (2SC), which is considered a metabolic marker of loss of FH activity. Through the process of succination, fumarate alters protein function (Reviewed in Schmidt et al., Semin Cell Dev Biol 2020). Another consequence of fumarate accumulation in the cytosol is the reversal of argininosuccinate lyase, a urea cycle enzyme, that results in the generation of argininosuccinate from arginine and fumarate (Zheng et al., Cancer Metabolism 2013). Succination and reversal of argininosuccinate lyase (ASL) are adaptive mechanisms that allow the cell to limit the accumulation of fumarate to survive the mutation. Indeed, the inhibition of either pathway is lethal to cells (Reviewed in Schmidt et al., Semin Cell Dev Biol, 2020). These reactions can be seen as "buffering tanks" that mop up the excess of fumarate. Consistently, with our new inducible model, both in cellulo and in vivo, we show that 2SC and argininosuccinate levels increase well before free fumarate is detected, and only when these "buffering tanks" are at near-saturation, we observe a more significant accumulation of fumarate in the tissue. Consequently, a modest increase in fumarate can still be accompanied by a strong downstream effect due to protein succination. A stronger accumulation of fumarate is seen later on (>>>day 10) in kidney tissue from our mouse model, and the higher levels observed in the different chronic models (Reviewed in Schmidt et al., Semin Cell Dev Biol, 2020) reached upon several months of culture. We have amended the manuscript and provided a cartoon (R#4 Figure 1) below that has also been integrated into the manuscript (revised manuscript Extended Data Fig.1c) to illustrate the concept and better explain the sequence of events leading to 2SC, arginosuccinate and fumarate levels accumulation. with either vehicle or 4-OHT (right-hand side plots) or in cells treated with either vehicle or 400 μM MMF for the indicated period of time (left-hand side plots). Figure 3. MMF-mediated mitochondrial succination. 2SC immunofluorescence: representative confocal images of cFh1 fl/fl treated with 400 µM MMF or vehicle (DMSO) for the indicated period. Mitochondria were labelled using anti-TOM20, and succinated substrates were labelled with an antibody recognising 2SC modifications. Scale bar: 10 µm.

R#4
Can low doses of MMF have similar effects to higher doses to recapitulate the effects of a small increase in fumarate? Because of this discrepancy, the data in the MMF treated cells is less convincing To address this point, we have broadened range of MMF concentrations used to treat the cells, particularly lower concentrations (R#4 Figure 4) as suggested by the reviewer. Our new results show a clear dosedependent effect on the transcriptional activation of target ISGs with the strongest effect at the higher dose (400 µM). For most of the targets, the lower concentrations (25 µM) show no or little effect. In line with the succination process detailed above, the lower dose is not sufficient to trigger a consistent response compared to higher doses. This data has now been incorporated into a new figure (revised manuscript Extended Data Fig. 8l) that combines the effect of lower doses of MMF on the ISG response and that of other metabolites (αKG and 2HG; see response to reviewer 1 point 5). that alterations of mtDNA or its nucleoid structure caused by fumarate is at least in part responsible for the formation of MDVs. R#4 Figure 6. mtDNA is required for fumarate-driven MDVs formation. Left panel. representative confocal images of iFh1 fl/flCL29 ρ0 treated with 400 µM MMF for 1-8 days. Mitochondria were labelled using anti-TOM20 and anti-PDH antibodies, and DNA using an anti-DNA antibody. Scale bar: 10 μm. Right panel: Quantification of TOM20 -PDH + vesicle number from (e). n=3 independent experiments Finally, how does this mechanism contribute to the tumor suppressive function of FH1 in HLRCC? Additional experiments to address this point would strengthen the relevance of these findings This is a very pertinent comment and we thank the reviewer for suggesting the additional experiments. Indeed, mutations in Fumarate Hydratase (FH) are associated with Hereditary Leiomyomatosis and Renal Cell Cancer (HLRCC), which can ultimately result in the development of an aggressive form of kidney cancer. In our study, we hypothesize that the chronic low-grade inflammation resulting from the early activation of the innate immune response may play a role in the development of tumours in FH-deficient kidney tissue.
We want to emphasise there that there are no in vivo models so far that recapitulate the type of tumours observed in HLRCC patients and the mouse model we used only develops premalignant cysts only in the late stage of the mouse life. Therefore, while working on generating a more suitable mouse model, we resorted to human data. First, GSEA analysis of a published expression profiling dataset (Ashrafian et al., Cancer Research, 2010;GSE20896) of HLRCC vs Normal tissue shows a strong enrichment in the innate immune response (R#4 Figure 7a). Second, despite the paucity of HLRCC tumour material, we teamed with the group of Dr Maxine Tran to provide additional data showing a transcriptional activation of ISGs in tumour tissue from HLRCC patients vs normal tissue that is in line with the phenotype we report in the manuscript in across our models (R#4 Figure 7b and revised manuscript Fig. 5). Third, using a deconvolution method (https://github.com/Danko-Lab/TED) to determine the cellular composition of kidney tumour tissue across a panel of FH-deficient RCCs and other renal cancer subtypes including, ccRCC, papillary-type RCC (pRCC), SDH deficient tumours (SDH_renal), chromophobe RCCs (chRCC) and metabolically different chRCC (MD-chRCC) compared to normal tissue (Normal), we confirmed an elevated contribution in lymphocytes, reflecting an inflammatory microenvironment, in FH-deficient tumours but not in SDH-deficient tissues vs normal tissue (R#4 Figure 7c) in line with our hypothesis. Fourth, high levels of Interleukin-6 (IL6) have been shown to be present in the tumour microenvironment; thereby reflecting the strong association between inflammation and cancer (Kumari et al., Tumour Biol., 2016;Hirano et al., Int Immunol, 2021). Its overexpression has been reported in almost all types of tumours and it has been shown to play a role in the pathogenesis of chronic inflammatory diseases, autoimmune diseases and cancer. In line with this, we also observed a slight but significant elevation in IL6 and IL10 levels in the FH-deficient HLRCC tumours by ELISA (R#4 Figure 7d). This data is now incorporated in the revised version of the manuscript in the revised Fig. 5. Of note, as reported in a recent publication from the Linehan lab (Crooks et al., Science Signalling, 2021), human HLRCC kidney tissue displays a mitochondrial phenotype that is identical to that we report in our study in Fh1-deficient animals. We now refer to this phenotype in the revised version of the manuscript. Altogether, this new dataset indicates that some of the features we observe in our in vivo and in vitro models of Fh1-deficiency are recapitulated in human tumour tissue associated with loss of Fumarate Hydratase (FH), the human ortholog of mouse Fh1. . We applied a deconvolution method (https://github.com/Danko-Lab/TED) to determine cellular composition of mouse kidney tissue at day 5 and day 10 post-induction. This approach indicated no significant differences in lymphocytes or myeloid lineage contribution between the control and Fh1-deficient mouse kidney tissues (pairwise comparison using Wilcoxon rank sum exact test and Benjamini-Hochberg p-value adjustment shows no significant differences between the control and Fh1-deficient mouse kidney tissues). (f). Immunohistocytological staining of Fh1 +/+ vs Fh1 -/mouse kidney tissue at day 10 post-induction. Left: 2SC staining showing an accumulation of 2SC in Fh1 -/mouse kidney tissue (bottom) vs Fh1 +/+ control animals (top) in line with our metabolomics data confirming the Fh1-deficient status of the tissue. Right: no difference in staining for CD14, a marker for myeloid cells, was observed between Fh1 +/+ control and Fh1 -/mouse kidney tissue.

2.
The evidence in support of cGAS-STING-IFN-I pathway activation is generally not sufficient. Although crosses of FH cKO mice onto cGAS or STING KO mice to show ablation of the immune signature in vivo would be the most convincing, cGAS/STING inhibitors are readily available for mouse in vivo pre-clinical studies. These would be great to include as a way to link the in vivo phenotypes more cohesively with the cellular mechanisms presented in the later figures.
We thank the reviewer for this suggestion. Following their advice, we designed an in vivo study using a Sting inhibitor (R#5 Figure 3a). Mice that were treated with the combination of Tamoxifen and Sting inhibitor showed a dose-dependent reduction in ISGs expression. Noticeably Sting inhibitor H-151 had a strong effect, reducing the levels of the cytokines Cxcl10, Ccl20 and Ccl2 but a milder effect on other markers like Ifitm10, Ifi202b and Areg. Although the levels of Ifnb1 were decreased, the use of Sting inhibitor didn't fully abrogate its production. This suggested that other pathways could also contribute to the type I response, a point that we are now discussing in the manuscript and in other answers to this reviewer. This data has now been incorporated in the revised manuscript in Fig. 3n and Extended Data Fig. 9g.
Finally, the authors never show that genetic ablation of cGAS/STING by siRNA, Crispr, etc is sufficient to ablate the interferon and ISG signature observed invitro. The RU.521 experiments of Figure 2 (m) and 3 (n) do not convincingly show that cGAS inhibition abolishes the induction of immune genes after FH deletion of MMF treatment. It would be much more convincing to target these pathways using siRNA or other means (the authors use cGAS, STING, Rig-I siRNAs in Extended figure 6 but never examine immune gene expression after knockdown in FH deficient or MMF treated cells). Given that mtRNA has also been noted as a ligand driving immune activation downstream of mitochondrial stress, it is important to rule out the mtRNA sensors Rig-I/Mda5 and perhaps other nucleic acid sensing TLRs. This could be easily accomplished using the cell lines and siRNAs reported in this paper.
We thank the reviewer for mentioning the possibility that other pathways involved in cytosolic nucleic acid recognition upstream TBK1/IRF3 signalling pathway could be involved in our mechanism. Indeed, recent publications have acknowledged that several PRR proteins can cooperate in the regulation of ISG responses (Zevini et al., Trends Immun, 2017). Indeed, as the reviewer underlined here, both our cellular models and in the new in vivo experiments highlight the inability to completely rescue the ISG response observed with the cGas or Sting inhibitors, RU.521 or H-151, respectively. Therefore, we decided to investigate more broadly other PRR proteins potentially involved in the signalling of the ISG response. We conducted a siRNA targeted screen in the inducible model of Fh1 deletion using phosphorylation of Irf3 as a read out, since it showed the strongest activation in our cellular models and is a common marker of most PRR driven type I inflammation. (R#5 Figure 4). This experiment revealed that in addition to silencing cGas and sting, siRNA directed against Rig-1 was also able to mitigate Irf3 Phosphorylation; while silencing of Mda5, Mavs, Ifi202b, or Tlr receptors had no effect. Although this result does not completely rule out a role for Toll-like receptors in the immune response we observe, considering the strong effect of the combination of Rig-I and cGas silencing (see also R#5Figure 5), we chose not to investigate their potential role at this moment and decided instead to investigate further a potential role of RIG-I. We further confirmed these results in MMF-treated cells where we show that silencing cGas and Rig-I individually had a mild effect on Tbk1/Irf3 activation but strikingly, combined treatment of siRNAs against cGas and Rig-I or cGas, Rig-I together with Sting obliterated Tbk1/Irf3 activation (R#5 Figure 5 and revised manuscript Fig. 3o, p). Based on these results, we concluded that 1) when cGas is silenced, Rig-I may be able to compensate for the absence of the protein and vice versa; and 2) that Rig-I and cGas are the major regulators involved in the ISGs response we observe. RIG-I recognizes short viral doublestranded RNA, However, strikingly, the silencing of Mda5 and Mavs, the other key components involved

5.
The authors claim in the discussion that this paper represents the first report of a disease relevant mutation causing release of the mtDNA into the cytosol. This is a significant overstatement. Recently, human mutations in ATAD3A causing bona fide mitochondrial disease have been shown to liberate mtDNA into the cytosol and engage cGAS. Moreover, loss of TDP-43, CLPP, YME1L, OPA1 and others, all of which are linked to human disease, induce mtDNA release and activation of innate immunity. Perhaps the authors meant to restrict their discussion of impact to kidney diseases, but even so, loss of Tfam in kidney tubule cells has been shown to induce mtDNA leakage and cGAS-STING inflammation in vivo and in vitro.
We thank the reviewer for pointing this out and apologise for the oversight. The point we sought to make was that it is the first time that an alternative mechanism for mtDNA release other than involving mitochondria loss of integrity, and therefore compatible with cell survival and a persistent immune signature, is proposed. We have amended the text and references accordingly.
There is no evidence provided here to link the mtDNA release and immune phenotypes shown to HLRCC, so I believe it would be prudent to tone down these statements of novelty and properly reference the literature.
It will not decrease the impact of the author's findings to reference these recent papers.
We understand the reviewer's comment and will modify the text accordingly. Nevertheless, we now provide additional data obtained from human patients showing a similar immune response signature and cytokines production profile in HLRCC tumours to that seen in our different mouse models (R#5 Figure  11 and revised manuscript Fig. 5a-d). We hope that our findings will catalyse an in-depth characterisation of the role of fumarate-dependent inflammation in the aetiology of HLRCC, and its progression. Technical Comments: 1. A significant portion of the analyses herein rely on manual scoring of immunofluorescence images. While manual scoring itself is not itself problematic, the individual(s) conducting the scoring should be blinded to the experimental groups wherever possible in order to minimize the impact of unintentional bias in the results. This is particularly important for subjective measures, such as mitochondrial morphology, where there is no clearly defined delineation between different morphological classifications.
We appreciate the reviewer's concerns, and we wish to explain in detail here why we choose to report the mitochondrial morphology phenotype this way: first, due to the unusual mitochondrial morphology observed in these cells, we wanted to provide a comprehensive assessment. In addition, with our classification, we wanted to describe the heterogeneity of the mitochondrial network in different cells, which wouldn't otherwise be reflected by assessing average size of mitochondria. Nevertheless, we now provide new unbiased quantification of mitochondria area and number (R#5 Figure  12) using Fiji software calculated from randomized region of interest analyses (see Material and Methods section). These new results show that the loss of Fh1 induces a decrease of mitochondrial count mirrored by an increase of mitochondrial area., corroborating the changes in the mitochondrial network in cFh1 CL1 and cFh1 CL19 (revised Extended Data fig. 3f), iFh1 -/-CL29 and iFh1 -/-CL33 (revised Fig. 2i, j and  Extended Fig. 2j,k) and in MMF treated cells (revised Fig3c,d). Figures 3 (j-l), 4 (i-l), 5 (a, h-k) For these figures, each individual replicate for the control group has been set to one. This has the effect of artificially reducing the variance of the control group to zero, resulting in an inflated level of significance for groups that are being directly compared to the controls. Control values should be normalized to the mean of the control group, rather than set to one on an individual basis, preserving the variance and allowing an accurate significance level to be calculated.

Figures 3 (f-h, j, n), 4 (j) and Extended Data
We understand the concern of the reviewer and we want to reassure here that we didn't manipulate the data to change the significance of our results. The digital droplet PCR experiments assess cytosolic mtDNA copy after a brief cell extraction with digitonin and we define mtDNA release by the fold change observed in different samples to this baseline. Thus, we would argue that Fold Change to control is the biological value we want to report here. We report below the statistics as they would be calculated plotting each control value (R#5 Figure 13). Also, we have documented the absolute copy number data in the raw data section. understanding that this complex is not a major driver of IFN and NF-kB signaling in physiological conditions. However, it might be nice to cite these reports and add a few lines to the discussion. Given this unique RIG-I-dependent but MAVS-independent result, I request that the authors include the the p-IRF3 data from response Fig. 4 to the paper, and also add MAVS siRNA knockdown experimental data to Fig. 3 or ED Fig. 9. I think it is important to clarify that this is a unique response downstream of FH inhibition and fumarate accumulation, which reveals unexpected complexity in the innate immune response influenced by mitochondrial metabolites and mtDNA release. After incorporating these data, references, and discussion points, I will support publication of the paper in Nature.

Response to the referees Zecchini, Paupe et al
Referee: plain black text Authors: plain blue text Referee #4 (Remarks to the Author): The authors have provided a substantial amount of new data and I can appreciate the amount of work that has gone into this revision. It is still unclear to me how these effects are mediated by fumarate if fumarate is not yet accumulated when the key phenotypes are observed (i.e. phosphor-IRF3, cytosolic DNA fragments, and induction of MDVs). The MMF experiments do not clarify this point as the addition of MMF causes an increase in fumarate. The new schematic presented (Extended data Figure 1C animation ) suggests that the early (and most important) phenotypes are not mediated by fumarate itself but instead by the adaptive buffering response to increased fumarate, perhaps the increased 2SC. It seems the more appropriate experiments would include manipulation of 2SC instead of MMF. The MMF experiments would only make sense with a dose that causes an increase in 2SC but no change in fumarate as that would best mimic the induced loss of Fh1. The results are very striking, but the mechanisms is still unclear as to what is mediating which phenotypes. Clearly MMF has an effect, but there is also clearly an effect before fumarate is accumulated. The data shown do not differentiate these two mechanisms, nor does the data show they are one in the same.
We thank the reviewer for assessing our revised manuscript critically and appreciating our efforts to address their comments. The referee points out that "the new schematic presented (Extended data Figure 1C animation ) suggests that the early (and most important) phenotypes are not mediated by fumarate itself but instead by the adaptive buffering response to increased fumarate, perhaps the increased 2SC". Based on this comment, we need to clarify some important aspects of our findings.
First, we would like to elaborate on an important point regarding succination, which may not have been sufficiently clear in our earlier response. In FH-deficient cells, fumarate cannot be converted into malate and we surmise that the loss of Fh1 leads to its increased availability. On one hand, this initial fumarate accumulation leads to the formation of the metabolite 2SC, i.e. the adduct between fumarate and free cysteine, which can be detected by mass spectrometry. On the other hand, fumarate can lead to the succination of cysteine residues of proteins that can be visualised using the dedicated anti-2SC antibody. Based on our current model, in FH-deficient conditions, fumarate would only start to accumulate, as measured by mass spectrometry, once a significant pool of available cysteine residues has undergone succination, and other buffering systems have reached saturation. This model is corroborated by the observation that in vivo, at day 5 upon Fh1 loss, even though fumarate levels are not yet significantly up by that time 2SC is increased. Thus, fumaratemediated effects of FH loss on protein can appear before fumarate accumulates in the cell. Yet, increased succination arises from the dysregulated homeostasis of fumarate levels, so we postulate that the effects are de facto mediated by fumarate. We have amended the text to clarify this point and have been careful in distinguishing between free 2SC and protein succination.
In reply to the point as to whether more appropriate experiments would include manipulation of "2SC instead of MMF", we want to emphasise that succination, and therefore the production of 2SC, are inescapably linked and can only be achieved in vivo or in vitro by the addition of (exogenous or endogenous) fumarate (or MMF) to cysteine thiol residues. It is, therefore, technically not possible to manipulate protein succination without first increasing fumarate. Yet, we agree with the reviewer that our study does not provide definitive evidence to dissect the effects of fumarate accumulation from fumarate-driven succination, and which fumarate by-product is responsible for the observed phenotype. We have now amended the text and discussion to mitigate any ambiguity about our conclusions.