TNF antagonist sensitizes synovial fibroblasts to ferroptotic cell death in collagen-induced arthritis mouse models

Ferroptosis is a nonapoptotic cell death process that requires cellular iron and the accumulation of lipid peroxides. In progressive rheumatoid arthritis (RA), synovial fibroblasts proliferate abnormally in the presence of reactive oxygen species (ROS) and elevated lipid oxidation. Here we show, using a collagen-induced arthritis (CIA) mouse model, that imidazole ketone erastin (IKE), a ferroptosis inducer, decreases fibroblast numbers in the synovium. Data from single-cell RNA sequencing further identify two groups of fibroblasts that have distinct susceptibility to IKE-induced ferroptosis, with the ferroptosis-resistant fibroblasts associated with an increased TNF-related transcriptome. Mechanistically, TNF signaling promotes cystine uptake and biosynthesis of glutathione (GSH) to protect fibroblasts from ferroptosis. Lastly, low dose IKE together with etanercept, a TNF antagonist, induce ferroptosis in fibroblasts and attenuate arthritis progression in the CIA model. Our results thus imply that the combination of TNF inhibitors and ferroptosis inducers may serve as a potential candidate for RA therapy.

I do have some queries which authors should address. • Can the authors provide clinical demographics or RA and OA patients, and of the high disease activity vs low disease activity RA. • Could the authors clarify was this arthroplasty (ie end stage) or key-hole arthroscopy where biopsies were obtained. As above it would be important to include demographic and clarify medication data as the authors state that the RA patient had active disease, with mean duration of 10 years, but weren't on any steroid or second line medication which would be unusual with active disease and a disease duration of 10years? • Figure 1A should quantify sub-lining vs lining layer as it would be interesting to see if expression is more localised to fibroblasts in a specific anatomical location. • Should show RA images of High vs Low disease activity for MDA with IgG controls • Similarly did the authors stain high vs low disease activity for 4HNe and 8-oxdG • How did the author define High disease activity vs low disease activity • Figure 1F -should quantify Fab and F4/F80 • Line 130 the authors state' Interestingly, the surviving FAPα+ fibroblast cells that evaded IKEinduced ferroptosis were mainly surrounded by macrophages (Extended Data Fig. 1h), implying that macrophages protect fibroblasts against lipid oxidative stress and ferroptotic death' . Do the authors know if these macrophages are the recently described protective CX3CR1 macrophages in the RA joint.
• For Histology Authors need to include IgG controls throughout. • Were the Fib-a clusters or Fib-b clusters Thy1 + or Thy1-. • Were FMO controls used for flow cytometry • The authors used cellular communication networks to examine potential receptor-ligand pair interactions between macrophages and Fibroblasts. There is a spectrum of macrophages in the joint, did the author identify specific macrophages subtypes associated with these interactions • For the Analysis of the potential receptor-ligand pair interactions can the authors clarify if the analysis was intra-analysis ie they examined the interaction of macrophages with Fib-a and Fib-a within each RA patient, and then combined analysis. • In figure 4 the authors show that TNF-α protects RA fibroblasts from ferroptosis by enhancing the GSH biosynthetic pathway via activation of NF-κB. The manuscript would be greatly enhanced if the authors could show some functional outputs of the fibroblasts in response to the various treatments (RSL3; si GCLC, SINFKB) including fibroblasts invasiveness, migration capacity and secretion of key MMPS and pro-inflammatory mediators • In the last figure the author show that TNF-α antagonist sensitizes RA fibroblasts to ferroptosis induction in CIA model mice. Similar histological staining in Human RA tissue pre/post TNFi would demonstrate the translational aspect of the study.
Reviewer #2 (Remarks to the Author): The manuscript by Jiao Wu and colleagues examines alleviation of RA through activation of ferroptosis in synovial fibroblasts and the role of TNFa as a driver of ferroptosis resistance in a subset of the cells. This is an interesting story which, I think, convincingly suggest that synovial fibroblasts may have elevated susceptibility to ferroptosis inducers in the RA patients. It also demonstrates that a subset of fibroblasts may be protected from ferroptosis induced by an exogenous trigger through TNF-dependent signaling. At the same time, I don't think that the manuscript in the current form is sufficiently impactful for the publication in Nature Communications. First, I don't see evidence that ferroptosis is playing a significant role in RA in the absence of IKE treatment, for example there is no difference in scores in Fig. 1d-e in CIA and CIA+Lip-1 group. Therefore, impact of the findings is dependent on the feasibility of using ferroptosis inducers in the context of RA< Second, while activation of ferroptosis may be a good strategy in more acute settings, I find it questionable that such approach will be productive in a chronic and non-life threatening disease like RA, including long term death-associated inflammation, local tissue damage and damage to a variety of other tissues. To make discovery of the effect off ant-TNFs more impactful, the authors need to convince the readers first that exogenous activation of ferroptosis is a viable approach to RA. Therefore, I would recommend the manuscript in its present form for consideration in a more specialized journal.
Reviewer #3 (Remarks to the Author): The concepts introduced in the study are novel and interesting, and there is a wealth of data supporting their hypothesis. However, there are some issues that should be addressed. Overall the manuscript is extremely dense and each figure has up to 20 sections. The authors should focus their studies on the key data required to make their case and prove the mechanism and eliminate speculation and conclusions based on assumptions like much of Figure 2 and negative data related to the other cytokines. A shorter manuscript will be much more accessible and understandable.
1. A previous study reported that TNF stimulation induces NF-B activation through TNFR1/PKC /IKK / , results in p38 MAPK-, JNK-1/2-dependent NOX/ROS pathways in RA SFs (Mediators Inflamm. 2014;2014:279171). Since NOX/ROS induce lipid oxidation and ferroptosis, this could conflict with the present study. The authors should discuss difference between the studies. 2. Line 130: The authors should quantify the contention that surviving FAP+ fibroblasts are more likely to be surrounded by macrophages. 3. The data from Figure 2l through 2r are primarily computational and from biologic studies. Comments like "TNF signaling through NFkB was enriched in fibroblast clusters 1 and 2", "Macrophages also initiated TGFß signaling on fibroblasts", "TNF released from macrophages showed a higher interaction with TNFR2 in Fib a than in Fib b" and many others are inferences rather than the result of biologic validation. The text could be significantly shortened and the comments here and throughout the manuscript should distinguish between computational predictions, potential associations based on informatics and biological validation. The authors should test some of these predictions and present data supporting them. 4. The fact that TNF, IL-1 and IL-6 had different effects on ferroptosis provides a way to dissect the signaling mechanisms. The authors conclusion that NFkB is the primary mechanism related to the TNF effect seems unlikely because IL-1 also induces the same transcription factor. A more detailed signaling study defining how TNF leads to GSH production would be important. If it is through NFkB, than methods to prevent NFkB activation would be an important step to proving their hypothesis. 5. The translational approach of testing etanercept and IKE is interesting and important, although they have not shown whether the effects are due to modulating adaptive or innate immune mechanisms.
For example, what was the effect on type II collagen antibodies? The authors should test their treatment regimen in established disease (typically day 35 or later in mouse CIA) or in a model that is not dependent on adaptive immunity such as a serum transfer model to determine the mechanism of action. Prolonged treatment of mice with human etanercept leads to neutralizing antibodies and occurs within a couple weeks. This issue can be avoided with shorter term studies. 6. RSL3 concentrations used in these experiments were 0.125 μM. Others, however, require higher concentration (2 μM, 3 μM, and 5 μM) than those (Oncogene. 2017 Oct 5;36(40) 5593-5608;Cell Death Diff. 2020 Dec 17;Front Pharmacol. 2018 Nov 22;9:1371). Prior to do in vitro cell experiment with IKE, RSL3, and others, the author should perform dose responses to determine the optimal concentrations. 7. The manuscript should be reviewed by a statistician to assess the methods, especially Fig. 1d, e, and k and Fig. 3a. 8. Minor points: in line 110, the authors comment on FAP+ fibroblasts in RA synovium and cultured RA synovial fibroblasts. As noted by several authors, cultured SF are almost all FAP+/CD90+/podoplanin+ and have a phenotype that is a combination of lining and sublining fibroblasts. Also, the title should be modified because the authors do not actually test their hypothesis in patients with RA.
The figures needed to address each concern were combined and presented as Response Figures. Detail of patients clarified.
Response Figure 1 4. Statistics of MDA, 4-HNE and 8-OHdG of RA in high vs low disease activity.
Response Figure 2 5. Definition of high, moderate and low disease activity in RA.
Formula for DAS28-ESR and standards for disease activity added.
6. Quantification of FAP and F4/80 in fibroblasts. Response Figure 3 7. Explore CX3CR1+ macrophage in the RA joint. Response Figure 4 8. Provide all IgG controls of IHC assay. Response Figure 5 9. Analyze the expression of THY1 in fibroblast clusters in RNA-seq.
Response Figure 6 10. FMO control in Flow Cytometry assay. Response Figure 7 11. Analyze specific macrophage subtypes interacts with fibroblast in RNA-seq.
Response Figure 8 12, Clarify if the interaction analysis was within each RA patient and combined.
Response Figure 9 13. Detect the invasiveness, migration capacity, secretion of key MMPs and pro-inflammatory mediators of the fibroblasts in response to various treatments.
Response Figure 10 14. Human RA tissue pre/post TNFi Difficult was explained.
REVIEWER 2 1. Demonstrate if ferroptosis plays an important role in RA without IKE treatment, such as CIA and CIA+Lip-1 group.
Response Figure 11 2. Question that whether ferroptosis will be productive in a chronic and non-life-threatening disease like RA, including long term death-associated inflammation and local tissue damage and damage to a variety of other tissues.
Response Figure 12 REVIEWER 3 1. Discuss the difference between this paper and a previous study about TNF-NOX signaling.
Response Figure 13 2. Quantification the contention between survival FAP+ fibroblasts and macrophages.
Response Figure 14 3. Some data are redundant. Validate some computational inferences like"TNF released from macrophages showed a higher interaction with TNFR2 in Fib a than in Fib b".
Redundant data were deleted. Response Response Figure 17 6. In vitro cell experiment perform dose responses with IKE and RSL-3.
Response Figure 18, Response Figure 19 7. The manuscript should be reviewed by a statistician to assess the methods.
All data were reviewed. Response Figure 20 8. Modify the title and comment about the characteristics of RA synovial fibroblasts.
Title and comment were modified.

REVIEWER 1
The (females, n = 14; high disease activity, n = 19, moderate disease activity, n = 7); OA n = 21 (females, n = 12). For the single-cell RNA sequencing, the following patients were included: RA n = 5 (females, n = 5, high disease activity n = 5). For the measurement of MDA, 8-OH-dG and iron levels in joint fluid, the following RA patients were included: n = 20 (females, n = 16; high disease activity, n = 12; moderate disease activity, n = 8). For the isolation of synovial fibroblasts from joint fluid and circulating fibrocytes from peripheral blood mononuclear cells, the following active RA patients were included: n = 6 (females, n = 4; high disease activity, n = 3; moderate disease activity, n = 3).
• Could the authors clarify was this arthroplasty (ie end stage) or key-hole arthroscopy where biopsies were obtained. As above it would be important to include demographic and clarify medication data as the authors state that the RA patient had active disease, with mean duration of 10 years, but weren't on any steroid or second line medication which would be unusual with active disease and a disease duration of 10years?

Response:
We thank the reviewer for this comment. The biopsies were obtained from key-hole arthroscopy (during arthroscopic cleansing of the knee) and the synovial fluid samples were obtained during joint aspiration. Both arthroscopic cleansing of the knee and joint aspiration are used as routine treatments for therapeutic reasons, which help to remove damaged or infected tissue and to reduce the inflammation. In our original manuscript, we stated that "No patients included in the study were being treated with corticosteroids or second-line drug agents", which we intend to mean they were not under these treatments at the time of, and shortly before, the surgery (we set the standard as one month before the surgery). In clinic, a lot of rheumatoid arthritis patients stopped taking their medications within two years after they started them when it went into remission. However, there is no cure for rheumatoid arthritis.
Thus, there are a lot of patients with relapsed active disease who don't receive treatments for a not long term before the therapeutic surgery. We apologize for our lack of clarity, leading to the misunderstanding that the patients were not under treatment for the whole period since their diagnosis, which is not the case. We have made this clear in the revision.
• Figure 1A should quantify sub-lining vs lining layer as it would be interesting to see if expression is more localised to fibroblasts in a specific anatomical location.

Response:
We agree this is an important question to address. In Figure 1a of original version (revision Fig. 1d), we used MDA Detection Kit to measure MDA concentration in joint fluid, not IHC. I guess the reviewer means the staining of 4-HNE (revision Fig 1b). In the revised version, we did TSA-based immunofluorescent multiplex assay and quantify the sub-lining vs lining layer as the reviewer suggested.
Fibroblasts of lining layer was stained with VCAM-1 and fibroblasts of sub-lining layer was stained with CD248 + . We found the staining of 4-HNE remained unchanged in the fibroblasts of lining layer and sub-lining layer (revision Fig. 1c).
• Should show RA images of High vs Low disease activity for MDA with IgG controls Response: We thank the reviewer for this suggestion. In the original version Figure   1a (revision Fig. 1d), we already measured MDA concentration in joint effusion of RA patients with high or moderate disease activity with a MDA Detection Kit, which is not a IHC result. To avoid misunderstanding and distinguish the results from synovium and joint effusion, we labeled each plot in the revision.
• Similarly did the authors stain high vs low disease activity for 4HNe and 8-oxdG

Response:
We thank the reviewer for this suggestion. Since RA patients with low disease activity usually do not meet the indications for surgery or joint aspiration, there're not enough samples from patients with low disease activity. All patients involved in our study are of high or moderate disease activity. In the original version, 8-OHdG was measured by ELISA and the concentration in the joint fluid of RA patients with high and moderate disease activities was shown in Extended Data Figure   1b (revision Supplementary Fig. 1e). • Figure 1F -should quantify Fab and F4/F80 Response: We think the reviewer means Figure 1i (revision Fig. 2b). We quantified FAP and F4/80 as suggested using HALO ™ Image Analysis Software. IKE treatment strikingly decreased the population of FAPα+ fibroblasts (revision Fig. 2b). Although IKE seemed to increase the mean fluorescence intensity of F4/80+ macrophage, the Cohen's d is only 0.21, which indicates a small effect size (revision Fig. 2b).
• Line 130 the authors state 'Interestingly, the surviving FAPα+ fibroblast cells that evaded IKE-induced ferroptosis were mainly surrounded by macrophages (Extended Data Fig. 1h), implying that macrophages protect fibroblasts against lipid oxidative stress and ferroptotic death'. Do the authors know if these macrophages are the recently described protective CX3CR1 macrophages in the RA joint. The number of CX3CR1 + F4/80 + macrophages within the 50 μm radium of surviving FAPα+ fibroblasts that evaded IKE-induced ferroptosis significantly increased in joints of IKE-treated CIA mice (Response Figure 4). However, the percentage of CX3CR1 + F4/80 + cells among total F4/80 + macrophages remained unchanged.
Although these data didn't indicate specific enrichment of CX3CR1 + macrophages around survival fibroblasts, it's worthy to explore if there are specific macrophages clusters playing a key role in the protective effect against ferroptosis.
• For Histology Authors need to include IgG controls throughout.

Response:
We thank the reviewer for this suggestion. For all IHC staining and Tyramide signal amplification (TSA)-based immunofluorescent multiplex, we did IgG controls. Since all the antibodies used for normal IHC and TSA-based immunofluorescent multiplex are produced by mouse or rabbit, representative images were shown in the revision. We showed immunohistochemistry on formalin-fixed paraffin embedded human RA synovium or mouse CIA joints using 10 µg/mL of rabbit IgG1 isotype control or mouse IgG1 isotype control followed by an antimouse/rabbit avidin-biotin HRP detection system in revision Supplementary Fig. 1a.
The IgG controls for PTGS2 and GPX4 in mouse CIA joints were shown in revision Supplementary Fig. 1k. Fluorescent multiplex IHC staining of mouse CIA joints using 10 µg/mL of rabbit IgG1 isotype control or mouse IgG1 isotype control followed by TSA detection kit was shown in revision Supplementary Fig. 1d.

Response:
We checked our single cell RNA-seq data. We found the expression of Thy1 showed no difference between Fib-a cluster and Fib-b cluster (Response Figure   6). were added (revision Supplementary Fig. 2a-b).
• The authors used cellular communication networks to examine potential receptorligand pair interactions between macrophages and Fibroblasts. There is a spectrum of macrophages in the joint, did the author identify specific macrophages subtypes associated with these interactions

Response:
We agree this is an important question to address. We tried to identify specific macrophages subtypes and analysis the receptor-ligand interactions between different macrophage clusters with fibroblast clusters.
Primary cell cluster analysis was performed using Seurat's FindClusters function

Response:
We agree such data from clinical samples will be very helpful. However, practically it is not feasible for us to obtain RA tissues from patients receiving arthroscopic cleansing of the knee, both before and after anti-TNF treatment without therapeutic reasons.

REVIEWER 2
The manuscript by Jiao Wu and colleagues examines alleviation of RA through activation of ferroptosis in synovial fibroblasts and the role of TNFa as a driver of ferroptosis resistance in a subset of the cells. This is an interesting story which, I think, convincingly suggest that synovial fibroblasts may have elevated susceptibility to ferroptosis inducers in the RA patients. It also demonstrates that a subset of fibroblasts may be protected from ferroptosis induced by an exogenous trigger through TNF-dependent signaling. At the same time, I don't think that the manuscript in the current form is sufficiently impactful for the publication in Nature Communications. First, I don't see evidence that ferroptosis is playing a significant role in RA in the absence of IKE treatment, for example there is no difference in scores in Fig. 1d-e in CIA and CIA+Lip-1 group. Therefore, impact of the findings is dependent on the feasibility of using ferroptosis inducers in the context of RA.

Response: We truly appreciate the reviewer' valuable comments and suggestions.
First, as we demonstrated in Figure 1, hyperplastic rheumatoid synovium and synovial fluid of RA patients showed increased levels of iron and lipid peroxidation, indicate an increased sensitivity to ferroptosis induction but not necessarily ongoing ferroptosis in fibroblasts. In this sense, the increased ferroptosis tendency without actual ongoing ferroptosis is exactly the basis of our proposed ferroptosis inductionbased therapy. The reviewer is correct to conclude that such therapy depends on the availability/feasibility of clinical ferroptosis inducers. However, to conduct mechanistic and preclinical animal model investigation -the goal of this paper -is crucial to prove this concept and to define the underlying mechanism, which will set the stage for the development of ferroptosis-inducing clinical agents. We also agree that the role of lipid peroxidation and increased ferroptosis tendency in the entire progression of RA development needs to be explored. In our previous experiment, liproxstatin-1 was administrated to CIA mice that already developed active inflammation and failed to suppress the development of inflammation and joint damage (revision Fig. 1f and 1g). In the revision, we started liproxstatin-1 treatment from an earlier time point, which helps us to better investigate the role of lipid peroxidation in different stages of RA development (revision Supplementary Fig. 1g).
We Second, while activation of ferroptosis may be a good strategy in more acute settings, I find it questionable that such approach will be productive in a chronic and non-life threatening disease like RA, including long term death-associated inflammation, local tissue damage and damage to a variety of other tissues. To make discovery of the effect off anti-TNFs more impactful, the authors need to convince the readers first that exogenous activation of ferroptosis is a viable approach to RA. Therefore, I would recommend the manuscript in its present form for consideration in a more specialized journal.

Response:
We thank the reviewer for bringing up this crucial concern. We truly understand the reviewer's concern about the potential adverse side effects of the ferroptosis-inducing therapy. Indeed, this is always a major consideration for any potential therapy and can only be formally tested by a clinical trial, which is outside the scope of this paper. We agree that RA is a non-life threatening disease. Further, we collected the main organs, including heart, liver, spleen, lung and kidney of ferroptosis inducer (IKE)-treated mice (IKE 20mg/kg, twice a week, endpoint >57 days from CIA induction). We did H&E staining and didn't find significant pathological changes (revision Supplementary Fig. 12a). Although we didn't observe obvious pathological changes in main organs of CIA mice treated with low dose IKE, it's noteworthy that long-term administration of ferroptosis inducers may increase the risk of death-associated inflammation and damage to normal tissues.
Activated fibroblasts showed higher sensitivity to ferroptosis inducers compared to other main cell types in the hyperplastic synovium, which partially increases the specificity of ferroptosis induction. In addition, fibroblasts-directed ferroptosis strategies should be developed. Future studies will potentially identify surface proteins that are more specific to fibroblasts, which help to facilitate the ferroptosis strategies targeting fibroblasts therapeutically. Moreover, two FDA-approved rheumatoid arthritis drugs, sulfasalazine and auranofin, were also proven to effectively trigger ferroptosis (Signal Transduct Target Ther. 2020 Jul 31;5(1):138. Cancer Res 2021;81:1896-908. Oncol Rep. 201942(2):826-838). The combination of these drugs together with anti-TNFα therapy could be tested to reduce the risk of ferroptosis induction of proto-drug.

REVIEWER 3
The concepts introduced in the study are novel and interesting, and there is a wealth of data supporting their hypothesis. However, there are some issues that should be addressed. Overall the manuscript is extremely dense and each figure has up to 20 sections. The authors should focus their studies on the key data required to make their case and prove the mechanism and eliminate speculation and conclusions based on assumptions like much of Figure 2 and negative data related to the other cytokines. A shorter manuscript will be much more accessible and understandable.

Response:
We thank the reviewer for the positive and encouraging assessment. For the presentation issue the reviewer pointed out, we revised the manuscript to make it less dense and flow better. The negative data related to IL1β and some assumptions based on single cell RNA sequencing were deleted. In addition, more experiments were added to prove the key mechanisms.

Response:
We agree this is an important question to address. We found that TNF-α enhances the cystine uptake and cellular GSH biosynthetic pathway via activation of NF-κB to protect fibroblasts from oxidative stress and excessive iron to escape ferroptosis. Interestingly, it has been reported that TNF-α/TNFR1/NF-κB signaling could promote the activation of NOX, which is responsible for the generation of ROS (Mediators Inflamm. 2014;2014:279171). Since ferroptosis can be induced by accumulation of superoxide and hydrogen peroxide upon upregulation of NOX, we wondered if TNF-α/NFκB may also activate ferroptosis-sensitizing signaling. In the revision, we showed that short-term exposure of fibroblasts to TNF-α markedly increased ROS levels, which was reduced by pretreatment with the inhibitor of NOX (NOXi) (revision Fig. 6p). However, long-term exposure of fibroblasts to TNF-α failed to result in permanent increase of ROS (revision Fig. 6p). Of note, NOXi further enhanced the protective effect of TNF-α against ferroptosis triggered by RSL3 (revision Fig. 6q). Although both TNFR1 and TNFR2 signaling have been proven to initiate NF-κB function via activation of TRAF-2, the increase in NF-κB activity in response to TNFR1 activation is rapid and transient, whereas TNFR2 activation results in a much slower but persistent response (J Biol Chem2004;279:32869-32881).
To confirm the involvement of TNFR2 in the regulation of NF-κB activity in RA fibroblasts, we treated fibroblasts with anti-TNFR1 and anti-TNFR2 antibodies, respectively, under long-term stimulation of TNF-α. We found that blocking of TNFR2 abrogated NF-κB activation compared to blocking of TNFR1, measured by IκB phosphorylation (revision Fig. 4j), indicating that long-term stimulation of TNF-α tent to activate NF-κB signaling through TNFR2 other than TNFR1. These data suggest that the defensive effect of TNF-α/TNFR2/NF-κB against ferroptosis overcomes the ferroptosis-sensitizing contribution of TNF-α/TNFR1/NFκB-induced ROS production.
2. Line 130: The authors should quantify the contention that surviving FAP+ fibroblasts are more likely to be surrounded by macrophages.

Response:
We quantified the distance between FAP+ fibroblasts and F4/80+ macrophages using HALO™ Image Analysis Software. The number of F4/80+ macrophages within the 50 μm radium of surviving FAPα+ fibroblasts that evaded IKE-induced ferroptosis significantly increased in joints of IKE-treated CIA mice (revision Fig. 2f-g).
3. The data from Figure 2l through 2r are primarily computational and from biologic studies. Comments like "TNF signaling through NFkB was enriched in fibroblast clusters 1 and 2", "Macrophages also initiated TGFß signaling on fibroblasts", "TNF released from macrophages showed a higher interaction with TNFR2 in Fib a than in Fib b" and many others are inferences rather than the result of biologic validation. The text could be significantly shortened and the comments here and throughout the manuscript should distinguish between computational predictions, potential associations based on informatics and biological validation. The authors should test some of these predictions and present data supporting them.

Response:
We agree that the text could be significantly shortened and some conclusions based on single-cell RNA sequencing and assumptions should be eliminated. Thus, we deleted all comments about TGFβ, PDGF and EMT signaling based on single-cell RNA sequencing. We also agree that some key predictions should be tested to provide evidence of biologic validation and to support our points of view.
We tested the interaction between TNFα and TNFR1/TNFR2 as well as the activation of NFκB signaling, which will also help to explain the question #1. From single-cell RNA sequencing, we predicted the molecular interactions involved in the signaling of crucial cytokines in RA and found that fibroblasts transmitted most of the TNF-α signals via TNFR1 and TNFR2, and TNF-α provided by macrophages showed a higher interaction with TNFR2 in Fib a than in Fib b cluster (revision Fig. 4i). To confirm the involvement of TNFR2 in the regulation of NF-κB activity in RA fibroblasts, we treated fibroblasts with anti-TNFR1 and anti-TNFR2 antibodies, respectively, under long-term stimulation of TNF-α. We found that blocking of TNFR2 abrogated NF-κB activation compared to blocking of TNFR1, measured by IκB phosphorylation (revision Fig. 4j), indicating that long-term stimulation of TNF-α tent to activate NF-κB signaling through TNFR2 other than TNFR1. Thus, the stronger interaction between TNF-α from macrophages and TNFR2 in Fib a cluster is consistent with the enrichment of TNF-α signaling through NF-κB in these fibroblasts.
4. The fact that TNF, IL-1 and IL-6 had different effects on ferroptosis provides a way to dissect the signaling mechanisms. The authors conclusion that NFkB is the primary mechanism related to the TNF effect seems unlikely because IL-1 also induces the same transcription factor. A more detailed signaling study defining how TNF leads to GSH production would be important. If it is through NFkB, then methods to prevent NFkB activation would be an important step to proving their hypothesis.

Response:
We thank the reviewer for this suggestion. Since the reviewer suggested to "focus studies on the key data and eliminate negative data related to the other cytokines" to make the paper more accessible and understandable, we decided to delete IL1β results. In the original version, we already used PS1145, an inhibitor of IκB kinase (IKK) that specifically inhibits IKK-mediated IκB phosphorylation to prove that TNF promotes GSH production through NFκB (original version Figure 3kn, revision Fig. 6k-o). We found that PS1145 effectively inhibited NF-κB activation and abrogated the upregulation of SLC7A11, GCLC and GCLM expression (revision Fig. 6k-l). In addition, PS1145 obviously hindered TNF-α-mediated defense of ferroptosis and lipid ROS accumulation (revision Fig. 6m-n). Consistently, TNF-αinduced GSH accumulation was effectively terminated upon PS1145 treatment (revision Fig. 6o).  Supplementary Fig. 1f), although not started at day 35 or later. In the revision, we tested the effect of etanercept and IKE on type II collagen antibodies in serum of CIA mice. We found that the anti-type II collagen antibody in serum of CIA mice did not differ significantly between mice treated with etanercept in combination with IKE and mice treated with vehicle (revision Supplementary Fig.   12b), suggesting that the reduced inflammation and joint damage are not due to modulating autoantibodies production.

Response:
We thank the reviewer for this question. The dose response of fibroblasts to IKE and RSL3 was shown in original version Figure 3a (revision Fig. 5a-b), and the dose response to TNFα and other cytokines was shown in original version Extended Data Figure 6a-c (revision Supplementary Fig. 8a-c). The 0.125μM dose for RSL3 was chosen for following experiments because we found TNF-α could strongly protect fibroblasts from low-dose RSL3, while high doses of RSL3 still induced potent ferroptosis in the presence of TNF-α. Besides, we did find that fibroblasts are much more sensitive to ferroptosis induction than most of the epithelial tumor cells.
That is consistent with our previous ferroptosis paper, which found intercellular interactions between epithelial cells suppress ferroptosis by activating the intracellular NF2 (Nature, 2019;572(7769):402-406). To further prove the sensitivity of fibroblasts to ferroptosis induction, we analyzed the sensitivity of a panel of fibroblast cell lines (PG13, IMR-90, MRC5, and MH7A) to ferroptosis induction. All the tested cell lines were sensitive to 0.125 μM RSL3 induction for 14 h (Response Figure 19).
These results also suggest the potential of ferroptosis therapy in other diseases related to abnormal activation of fibroblasts.
7. The manuscript should be reviewed by a statistician to assess the methods, especially Fig. 1d, e, and k and Fig. 3a.

Response:
We thank the reviewer for bringing up this important suggestion. The manuscript has been reviewed by a statistician and the methods for all plots were indicated in the figure legends. For example, original version Fig. 1d and 1e (revision Fig. 1f and 1g) showed joint inflammation measured by arthritis score and paw thickness. One-way ANOVA followed by Multiple Comparisons was performed to compare the means of arthritis score or paw thickness at the end point (day 22 after disease onset). Original version Fig. 1k (revision Fig. 2d and 2e) showed cell death and lipid ROS production in circulating fibrocytes from PBMCs and synovial fibroblasts from inflamed joint fluid of RA patients treated with RSL3. Two-tailed ttest was performed to compare the means of circulating fibrocytes and synovial fibroblasts under the stimulation of RSL3. Original version Fig. 3a (revision Fig. 5a and 5b) showed relative viability of fibroblasts primed with TNF-α, IL-6 or TGF-β, followed by treatment with different concentrations of IKE or RSL3. IC50 values were calculated using nonlinear regression analysis (log(inhibitor) vs. response --Variable slope (four parameters)). We found that in human fibroblasts TNF-α conferred significant resistance to ferroptosis induced by both IKE and RSL3. The   We also agree it's inappropriate to use the former title with the description "therapy" since we didn't test the hypothesis in patients. Thus, we modified the title of the paper to "TNF-α Antagonist Sensitizes Synovial Fibroblasts to Ferroptotic Cell Death in Rheumatoid Arthritis".