CAR T-cells targeting FGFR4 and CD276 simultaneously show potent antitumor effect against childhood rhabdomyosarcoma

Chimeric antigen receptor (CAR) T-cells targeting Fibroblast Growth Factor Receptor 4 (FGFR4), a highly expressed surface tyrosine receptor in rhabdomyosarcoma (RMS), are already in the clinical phase of development, but tumour heterogeneity and suboptimal activation might hamper their potency. Here we report an optimization strategy of the co-stimulatory and targeting properties of a FGFR4 CAR. We replace the CD8 hinge and transmembrane domain and the 4-1BB co-stimulatory domain with those of CD28. The resulting CARs display enhanced anti-tumor activity in several RMS xenograft models except for an aggressive tumour cell line, RMS559. By searching for a direct target of the RMS core-regulatory transcription factor MYOD1, we identify another surface protein, CD276, as a potential target. Bicistronic CARs (BiCisCAR) targeting both FGFR4 and CD276, containing two distinct co-stimulatory domains, have superior prolonged persistent and invigorated anti-tumor activities compared to the optimized FGFR4-specific CAR and the other BiCisCAR with the same 4-1BB co-stimulatory domain. Our study thus lays down the proof-of-principle for a CAR T-cell therapy targeting both FGFR4 and CD276 in RMS.

In the submitted manuscript, the authors describe the development of a bispecific CART approach for rhabdomyosarcoma (RMS) that consists of targeting FGR4 and CD276, two established targets for RMS.The authors compare several bispecific CART population and identify that co-expression of a FGR4.28z and a CD276.41BBzCAR in T cells endows CARTs with superior antitumor activity in several models.The authors also perform extensive correlative studies, demonstrating that the expression CARs with CD28 and 41BB costimulatory domains has additive and/or synergistic effects.While new novel mechanisms are explored, believe that the conducted study is very impactful and present a significant step forward in developing bispecific CART products for solid tumors such as RMS.The data is well presented and the manuscript is well written.I mainly have some minor concerns, which are summarized below.
Minor 1) Figure Panels 6G-J show that bispecific CARTs have superior anti-tumor activity than monospecific CARTs against single antigen positive tumors in vivo; the bioluminescence data (6I,J) is most likely explained by 'bleed through' of the bioluminescence signal of the contralateral, growing tumor.However, even when one reviews the leg volume measurements (6G,H), FGR4.28zCARTs have decreased antitumor activity in comparison to bispecific CARTs against CD276KO tumors (6G), which does not occur with FGR4KO cells and CD276-CARTs (6H); also, Mock Ts seem to have significant antitumor activity against FGR4KO cells.All these issues are not described and/or discussed, which I believe creates confusion on multiple levels.
2) As the authors pointed out, they cannot evaluate toxicity in their in vivo models.However, recommend comparing GMCSF and IL6 production of mono-and bi-specific CART population post antigen-specific activation; two cytokines that have been implicated in CART-associated cytokine release syndrome.6) The authors provide a very nice phenotypic analysis of the different FGR4-CART populations (S2C-F).However, such an analysis is not provided for the bispecific CART population (S5A)believe that it would be very reassuring if a similar analysis is provided for these constructs.In particular, since one of the bispecific CART population is identified 'as the winner' in subsequent experiments.
7) Lines 204-207: the authors states that they generated bicistronic CARs; this is technically incorrect: the CAR is not bicistronic -they generated a bicistronic lentiviral vector encoding 2 CARs separated by a self-cleaving peptide -please correct.8) Line 257: do not believe that these tumors are 'large' based on the provided Figure panelrecommend removing.9) While the authors cite Hirabayashi et al (ref 28), they do not discuss that these authors published a different bispecific CART design (same TM), only one zeta domain -believe that this deserved to be discussed.
Reviewer #2 (Remarks to the Author): Tian et al. developed a new bicistronic CAR T cell therapy to treat rhabdomyosarcoma by targeting FGFR4 and CD276.The authors mainly demonstrate that: 1. CD276 can be a novel antigen of rhabdomyosarcoma for CAR-T therapy 2. Bicistronic CAR T cell with CD28 and 4-1BB co-stimulatory signal synergizes CAR-T activation signal, leading to enhancement of overall CAR-T anti-tumor response.
Overall, this paper is well-structured, and the logic is easy to follow.However, some issues need to be addressed.Comments 1.In Fig 1 .the authors claim that substituting HTM from CD8 to CD28 can enhance tumor killing by demonstrating FGFR4-28HTM-BBz showed superior anti-tumor activity in vivo (Fig. 1E).However, it is quite interesting that no significant difference in tumor killing in vitro and cytokine produce between FGFR4-28HTM-BBz.Please provide potential reasons that could explain the difference in vivo and in vitro data.
2. In Fig 1N, there was a 100-fold higher number of FGFR4-28HTM-BBz detected in mice blood compared to FGFR4-28HTM-28z.Typically, better CAR-T expansion shows better tumor control, but authors found that FGFR4-28HTM-28z with lower number of CAR-T in mice blood has better tumor control activity compared to FGFR4-28HTM-BBz.Although the authors claim that CD28-based CAR-T showed faster killing but was easily exhausted, I think it is still quite challenging to understand that less tumor control activity found in CAR-T, which can expand 100 times more.It would be appreciated if the author could explain this insistency between tumor killing and CAR-T expansion.
3. Authors identified that bicis CAR with FGFR4-28HTM-BBz and CD276-8HTM-BBz showed higher exhausted phenotype than CAR with one CD28 and one BB in co-stimolatory domain.Please explain why two BB co-stimulatory signals can cause more exhaustion on the T cells.

Like Fig 1,
there is an inconsistency between tumor control and CAR-T expansion in mice blood (i.e., bicis CAR with FGFR4-28HTM-BBz and CD276-8HTM-BBz has almost no tumor killing in vivo in 4E, but second higher expansion capability found in mice blood (Fig 4F )).Please provide an explanation.
Reviewer #3 (Remarks to the Author): The manuscript by Tian and colleagues reports on the effect of bicistronic CARs targeted against FGFR4 and CD276 in rhabdomyosarcoma.Optimized CARs containing two different co-stimulatory domains increased anti-tumour activity significantly.The bicistronic CAR approach has the potential to reduce the risk of tumour escape, but also allows for additive and even synergistic activation as two co-stimulatory domains could result in increased down-stream signalling.
The manuscript is well-written and experiments nicely presented.The animal experiment using individual CRISPR knockouts of the targets is a particularly elegant approach to show specificity of the targeting and increased activity of the dual CARs.
FGFR4 levels are claimed by the authors to be low in all normal tissues and are clearly high in RMS tumours.However, FGFR4 has several important functions, including regulating bile acids, cholesterol and lipid metabolism.It is therefore a concern for side effects in children treated with a FGFR4-targeted CAR.The targeting of both FGFR4 and CD276 could also increase unwanted side effects.
Comments: 1.As the animal studies are performed in NSG mouse, it would be interesting to know if the CARs also recognize the mouse forms of FGFR4 and CD276?If they do, non-tumour targeting activities of the CARs could be analyzed in the animal models.2. The authors suggest that the limited efficacy of the FGFR4 CAR in RMS559 may be due to heterogeneous FGFR4 expression (line 159).From the flow cytometry measurements in figure 2C, this does not seem to be the case.In any case, the hypothesis should be investigated more closely.Possibly, immunofluorescence microscopy could be used to investigate if FGFR4 is heterogeneously expressed on the cell-surface of RMS559.As FGFR4 is mutated in this cell line (FGFRV550L), the localization or stability of the receptor may be different.3. It is surprising that the expression levels of FGFR4 and CD276 are so different in RMS559 compared to its xenograft counterpart RMS559 CDX (Fig 2D).This warrants an explanation.4. It is claimed that dual stimulation leads to increased and more persistent downstream signalling compared to stimulation by FGFR4 and CD276 alone (Fig 7E).This is difficult to see in figure 7D, especially in the case of p-PLCgamma, pERK1/2 and pAKT.Actually, the dual stimulation signals seem very similar to FGFR4 stimulation alone.Western blotting is not very quantitative, and in this experiment, only one value for each condition is shown, which is not very convincing.Even if one representative western blot is shown, the quantification should be performed and analysed for all three independent experiments together.Then error bars and significant levels can be calculated.
In the submitted manuscript, the authors describe the development of a bispecific CART approach for rhabdomyosarcoma (RMS) that consists of targeting FGR4 and CD276, two established targets for RMS.The authors compare several bispecific CART population and identify that co-expression of a FGR4.28z and a CD276.41BBzCAR in T cells endows CARTs with superior antitumor activity in several models.The authors also perform extensive correlative studies, demonstrating that the expression CARs with CD28 and 41BB costimulatory domains has additive and/or synergistic effects.While new novel mechanisms are explored, believe that the conducted study is very impactful and present a significant step forward in developing bispecific CART products for solid tumors such as RMS.The data is well presented and the manuscript is well written.I mainly have some minor concerns, which are summarized below.
Thank you for your careful review.We agree that although we attempted to gate carefully on the specific FGFR4KO and CD276KO tumor, the bioluminescence data (Fig. 6, I and J) reflect some 'bleed through' of signal from the contralateral tumor.Nevertheless, the results demonstrate that individual CAR T cells are effective against tumors with high expression of the target antigen, and the bispecific CAR T-cells are effective in eliminating both KO tumors.We agree with the reviewer's observation concerning the decreased efficacy of FGFR4.28HTM.28zCAR T-cells compared to BiCisCAR against CD276KO tumor.We believe that this is due to its short persistence and proneness to exhaustion, shown in current Fig. 1, L and M. On the contrary, BiCisCAR T-cells using two different co-stimulatory domains (CSDs) demonstrated better expansion and prolonged persistence in different RMS IM models (Fig. 4, H and J; Fig. S7G).We have added a description of this observation to clarify the results (lines 392-396).
Regarding the smaller tumor volume in the mock T-cells treated group compared with the CAR T-cells group, we believe this resulted from a significantly higher tumor burden in this cohort.We observed that these control mice had cachexia and were much frailer than those in CAR T-cell-treated groups.We believe the cachexia resulting from higher tumor burden was associated with slower growth of both CD276 KO (Fig. 6G) and FGFR4 KO (Fig. 6H) treated with mock T-cells.Moreover, this effect is further exacerbated by the fact that KO of FGFR4 leads to reduced growth of RMS tumors, as we have previously demonstrated [PMID: 19809159] and is readily demonstrated in comparing the tumor growth of the CD276 KO and FGFR4 KO cells compared with the parental cell line shown below in Rebuttal Figure 1.
Rebuttal Figure 1: Leg volume measured after tumors were intramuscularly implanted into NSG mice showed slower growth of RH30 tumors with FGFR4 KO.
2) As the authors pointed out, they cannot evaluate toxicity in their in vivo models.However, recommend comparing GMCSF and IL6 production of mono-and bi-specific CART population post antigen-specific activation; two cytokines that have been implicated in CART-associated cytokine release syndrome.
We thank the reviewer for their suggestion.In this study, we measured cytokine release of IFN-γ, IL-2, and TNF-α, but not specifically of GMCSF or IL-6.However, we expect they are all produced by CAR T-cells upon antigen-specific activation.Additionally, we did not observe any adverse effects, such as weight loss, in mice treated with BiCisCAR T cells from two independent donors (Rebuttal Figure 2).We do recognize this potential side-effect associated with CAR T therapy, which will be closely monitored in future clinical trials.

Rebuttal Figure 2. Mice bearing RMS559 IM xenografts didn't lose weight after Mock T or BiCisCAR T-cells treatment. Each line represents the relative weight of a mouse before (0 day) and after CAR T-cell treatment from two T-cell donors.
3) Fig 1H,M: Plotting Kaplan-Meier curves in which one group has a 'n=3' raises significant concerns regarding scientific rigor and reproducibility; either increase group size or remove.
We appreciate the carefulness of the reviewer and agree to remove these two panels from the original Fig. 1.We also repeated these two experiments using CAR T-cells derived from another donor with an increased group size (6 mice per group) at the reviewer and the editor's request.The new experiment data are consistent with those in Fig. 1 and have been added in the new Fig.S3.Because we simultaneously measured 4 exhaustion markers by flow cytometry, we used R to plot current Fig. 1M showing the expression of >2 markers on each same cell.As requested, representative flow plots for two markers for the FGFR4-targeting CAR T-cells are shown below (Rebuttal Figure 3).Additionally, the current flow software does not generate figures with >2 markers, and these are difficult to visualize.

Rebuttal Figure 3. Representative flow cytometry plots show the expression levels of CD39, PD-1, LAG-3, and Tim-3 on FGFR4 targeting CAR T-cells obtained from mouse blood after 32 days postinfusion into RH30 I.M. xenografts model.
6) The authors provide a very nice phenotypic analysis of the different FGR4-CART populations (S2C-F).However, such an analysis is not provided for the bispecific CART population (S5A) -believe that it would be very reassuring if a similar analysis is provided for these constructs.In particular, since one of the bispecific CART population is identified 'as the winner' in subsequent experiments.
Thanks for the reviewer's comments and suggestions.We have now added a similar figure of phenotype analysis for three BiCisCAR T-cells in Fig. S6.7) Lines 204-207: the authors states that they generated bicistronic CARs; this is technically incorrect: the CAR is not bicistronic -they generated a bicistronic lentiviral vector encoding 2 CARs separated by a self-cleaving peptide -please correct.
We agreed and modified the text accordingly."we generated 3 bicistronic lentiviral constructs encoding 2 CARs to target both We agree with this comment and have removed the word "large" here.
activation [PMID: 32193224].These observations indicate that CD28 HTM imparts the CAR with greater potency.Additionally, by comparing the CAR T-cell counts circulating in the blood, we found FGFR4.28HTM.BBz CAR T-cells have higher persistence than those with CD8 HTM on day 31 after CAR T-cell infusion (Fig. S1F).We, therefore, confirmed that the CD28 HTM domain enhances the activity and persistence of the CAR T cells in vivo.We have added this in the conclusion to our manuscript in lines 124-125.
We believe there may have been some misunderstanding by the reviewer, for which we apologize.The current Fig. 1L (original Fig. 1N) refers to the RH30 model, in which two modified FGFR4-CARs showed similar high anti-tumor activity (Fig. 1, E to G).For the orthotopic RMS559 models, FGFR4-28HTM-BBz showed no difference in its persistence compared to FGFR4-28HTM-28z, except on day 32 (Former Fig S4G,new Fig S5G).We agree with the reviewer that a better CAR-T expansion is generally associated with better tumor control.So FGFR4.28HTM.BBz CAR indeed controls the growth of orthotopic xenografts in the RH30 model, the same as FGFR4.28HTM.28zCAR (Fig. 1, E to G).Even though this CAR showed significantly weaker killing activity and lower cytokine releases than that of FGFR4.28HTM.28zCAR (Fig. 1, B and C), the prolonged persistence and less exhaustion make this CAR functional for a long time, resulting in greater control of RH30 xenograft growth.
However, in the more aggressive, fast growing tumor, RMS559 model, rapid and greater anti-tumor activity endowed by the CD28 co-stimulatory domain becomes much more important in controlling and killing the tumor within the first 14 days (Fig. 1, I to K) before FGFR4.28HTM.28zCAR T-cells diminished (Fig. S5G).Although FGFR4.28HTM.BBz CAR showed modestly higher persistence on day 32 post-infusion (Fig. S5G), it couldn't control the growth of these aggressive RMS559 xenografts due to its weaker anti-tumor activity.Therefore, FGFR4-28HTM-28z is better at controlling rapidly growing tumors compared to FGFR4-28HTM-BBz.We have conveyed this idea in lines 126-129 of our manuscript.
3. Authors identified that bicis CAR with FGFR4-28HTM-BBz and CD276-8HTM-BBz showed higher exhausted phenotype than CAR with one CD28 and one BB in co-stimolatory domain.Please explain why two BB co-stimulatory signals can cause more exhaustion on the T cells.
We thank the reviewer for raising this important question.We compared TCR activation signaling between two BiCisCARs following co-responding antigen stimulation in Figure S12D.BiCisCAR using the two 4-1BB CSDs induced lower phosphorylation levels of CAR-CD3z, PLCγ1, AKT, and ERK, compared with BiCisCAR using CD28 and one BB in co-stimulatory domains after dual-stimulation, suggesting a weaker activation of BiCisCAR T-cells using two 4-1BB CSDs.Prolonged exposure to high levels of antigen is a major cause of T cell exhaustion (PMID: 26205583 and PMID: 21555851).Therefore, we believe prolonged exposure to antigen stimulation at the tumor site is likely the cause of more exhaustion for the BiCisCAR T-cells with two 4-1BB co-stimulatory domains.
Tumor control by the CAR T-cells is not only dependent on counts in blood at day 21 but also on rapid expansion and cytolytic activity of CAR T cells at an earlier time point to control rapidly growing tumors.
We attempted to detect CAR T cells in blood circulation before 11 days without much success due to few CAR T cells detected by this flow cytometry method.Thus, although there was a comparable CAR T cell count at day 21 for both BiCisCARs, the tumor cytotoxicity for FGFR4.28HTM.28z-CD276.8HTM.BBz CAR T cells were superior to that of FGFR4.28HTM.BBz-CD276.8HTM.BBz CAR T cells (Fig. 3F).We have a sentence in the discussion to reflect this in lines 506-509.Thanks for raising this interesting question.We didn't switch the co-stimulatory signals in BiCisCARs, and it would be an interesting experiment.This important question is the subject of a future study.

Reviewer #3 (Remarks to the Author):
The manuscript by Tian and colleagues reports on the effect of bicistronic CARs targeted against FGFR4 and CD276 in rhabdomyosarcoma.Optimized CARs containing two different co-stimulatory domains increased anti-tumour activity significantly.The bicistronic CAR approach has the potential to reduce the risk of tumour escape, but also allows for additive and even synergistic activation as two co-stimulatory domains could result in increased down-stream signalling.
The manuscript is well-written and experiments nicely presented.The animal experiment using individual CRISPR knockouts of the targets is a particularly elegant approach to show specificity of the targeting and increased activity of the dual CARs.
FGFR4 levels are claimed by the authors to be low in all normal tissues and are clearly high in RMS tumours.However, FGFR4 has several important functions, including regulating bile acids, cholesterol and lipid metabolism.It is therefore a concern for side effects in children treated with a FGFR4targeted CAR.The targeting of both FGFR4 and CD276 could also increase unwanted side effects.
3) Fig 1H,M: Plotting Kaplan-Meier curves in which one group has a 'n=3' raises significant concerns regarding scientific rigor and reproducibility; either increase group size or remove.4) Fig 7D: please provide the data for the other 2 donors as supplementary data.5) Fig 1O: please provide exemplary flow plots in the appendix.
4) Fig 7D: please provide the data for the other 2 donors as supplementary data.We have added the T-cell activation signaling data for the BiCisCAR T-cells from the other 2 donors to Fig. S13.5) Fig 1O: please provide exemplary flow plots in the appendix.
FGFR4 and CD276, separated by a self-cleaving linker allowing co-257: do not believe that these tumors are 'large' based on the provided Figure panelrecommend removing.