Cernunnos/Xlf Deficiency Results in Suboptimal V(D)J Recombination and Impaired Lymphoid Development in Mice

Xlf/Cernunnos is unique among the core factors of the non-homologous end joining (NHEJ) DNA double strand breaks (DSBs) repair pathway, in the sense that it is not essential for V(D)J recombination in vivo and in vitro. Unlike other NHEJ deficient mice showing a SCID phenotype, Xlf −/− mice present a unique immune phenotype with a moderate B- and T-cell lymphopenia, a decreased cellularity in the thymus, and a characteristic TCRα repertoire bias associated with the P53-dependent apoptosis of CD4+CD8+ DP thymocytes. Here, we thoroughly analyzed Xlf −/− mice immune phenotype and showed that it is specifically related to the DP stage but independent of the MHC-driven antigen presentation and T-cell activation during positive selection. Instead, we show that V(D)J recombination is subefficient in Xlf −/− mice in vivo, exemplified by the presence of unrepaired DSBs in the thymus. This results in a moderate developmental delay of both B- and T-lymphocytes at key V(D)J recombination dependent stages. Furthermore, subefficient V(D)J recombination waves are accumulating during TCRα rearrangement, causing the typical TCRα repertoire bias with loss of distal Vα and Jα rearrangements.


INTRODUCTION
All living organisms are exposed to DNA double strand breaks (DSBs), described as the most toxic type of DNA damage. DSBs result from either external genotoxic stresses or endogenous physiological processes (1), such as V(D)J recombination and Class Switch Recombination during T-and B-lymphocytes development and maturation (2), meiotic recombination, or during RNA Polymerase II-driven transcription of early response genes following cell activation or heat shock (3,4). Exogenous and physiological DSBs are repaired by either homologous recombination (HR) or the non-homologous end-joining (NHEJ) pathway, the later proceeding via the direct ligation of DNA ends (5). Briefly, the NHEJ pathway is composed of core factors, Ku70-Ku80 and the DNA-dependent protein kinase-catalytic subunit (DNA-PKcs) for DSBs recognition and stabilization, Artemis endo/exonuclease and Terminal-deoxynucleotydyl-Transferase (TdT), DNA polymerases µ and λ for DNA ends processing if necessary, and the Xlf-XRCC4-DNA Ligase IV complex for the final ligation step. In this last step, Xlf, and XRCC4 homodimers form a long polymeric filament tethering DNA broken ends, thus creating a "DNA repair synapse" (6)(7)(8). XRCC4 also stabilizes and activates DNA Ligase IV, which ensures the final repair of aligned DNA ends (9).
In T-and B-cell precursors in the thymus and bone marrow, respectively, variable domains of antigen receptors are somatically rearranged from V, D, and J gene loci by V(D)J recombination. Two steps of thymocyte maturation are closely associated with V(D)J recombination: first, the "TCRβ selection" at the CD4-CD8-Double Negative (DN) stage upon successful rearrangement of the Tcrb locus (10) and second, the "positive selection" at the CD4+CD8+ Double Positive (DP) stage following productive rearrangement of the Tcra locus (11). The lymphoid-specific complex RAG1/RAG2 initiates V(D)J recombination by introducing DSBs within Recombination Signal Sequences (RSS) flanking V, D, and J coding segments. RAG1/2 complexes remain bound to DSB ends as the post cleavage complex (PCC) to stabilize broken DNA ends prior to their repair by NHEJ (12). NHEJ function is critical for T-and B-lymphocyte development. Indeed it is the sole DNA repair pathway to cope with RAG1/2 generated DSBs, which occur during G0/G1 of the cell cycle. Loss of function of core NHEJ factors results in severe combined immunodeficiency (SCID) conditions in both humans and mice, owing to aborted V(D)J recombination (13). XRCC4 or DNA Ligase IV gene inactivation in mice results in SCID phenotype and embryonic lethality secondary to massive apoptosis of post-mitotic neurons (14,15). Furthermore, defects in NHEJ result in genetic instability, and unrepaired DSBs produced during V(D)J recombination may lead to T and Pro-B cell lymphomas with Tcra/d or Igh genes translocation, respectively (16). Pro-B cell lymphomas are hallmarks of NHEJ defects in Trp53 −/− mice, owing to inefficient DNA repair during V(D)J recombination. Thus, the P53 pathway plays a key role in the thymus by preventing genomic instability in case of aborted V(D)J recombination and DSBs left unrepaired (12,17). Though the P53 pathway prevents genomic instability through apoptosis of thymocytes with aborted V(D)J recombination, it is noteworthy that it is not involved in the apoptotic "death by neglect" of thymocytes with inadequate TCRαβ during positive selection (18,19).
Xlf (also known as Cernunnos) is a unique NHEJ core factor in the sense that, although it is essential for the repair of genotoxic DSBs from various origins as shown by the extreme ionizing radiation sensitivity of Xlf deficient cells (20,21), it is largely dispensable for V(D)J recombination. Indeed V(D)J recombination yields on endogenous Igκ loci and exogenous substrates are normal in vitro in v-Abl transformed Xlf −/− Pro-B cells (22,23). Xlf −/− mice show a very modest immunodeficiency with decreased lymphocyte counts in blood and a loss of cellularity in the thymus (22,24). Furthermore, Xlf −/− Trp53 −/− double Knock-Out (DKO) mice do not develop Pro-B cell lymphomas commonly seen in other NHEJ-Trp53 DKO conditions (22), further attesting for the absence of a major V(D)J recombination defect in these mice, and the thymic cellularity is even partly rescued in the Trp53 −/− background (24). We previously proposed that Xlf participates in a twotier safeguard mechanism during V(D)J recombination to avoid genetic instability (23,24). Many DNA repair factors, PAXX, ATM, H2A.X, and MRI, are functionally redundant with Xlf during V(D)J recombination, as revealed by the complete defect in V(D)J recombination in vitro, and in vivo in doubly deficient situations (25)(26)(27). This functional redundancy is mediated, at least in part, through the C-terminus region of RAG2 (23). One key feature of Xlf deficient patients and mice is a remarkable TCRα repertoire bias with the loss of distal VαJα rearrangements (24), a characteristic first revealed in the context of a reduced thymocyte lifespan such as in RORγT and TCF-1 KOs mice (28,29). Xlf −/− immune phenotype in mice is indeed associated with a decreased thymocyte lifespan owing to chronic activation of the P53 pathway and apoptosis ex vivo (24). More recently, we identified a similar TCRα repertoire skewing in various human conditions of DNA repair deficiencies or hypomorphic RAG1/2 mutations (30).
At least two non-exclusive hypotheses can be raised to account for the immune phenotype of Xlf −/− mice in the context of an apparent normal V(D)J recombination. On one hand, one may propose that subefficient V(D)J recombination is associated with the persistence of DSBs and the chronic activation of P53. We used dedicated and sensitive markers to reveal possible persistence of unrepaired DSBs in thymocytes during V(D)J recombination. On the other hand, one alternative hypothesis could be that Xlf deficiency leads to a premature thymocyte death during positive selection at the DP stage, thus interfering with the ongoing multiple TCRα rearrangements at this stage. As positive selection strictly relies on antigen presentation by MHC class I and class II molecules, we introduced the Xlf −/− onto MHC deficient background to analyze this second proposition.
Altogether, these results suggest that the process of positive selection at the DP stage is not the cause of the P53 chronic activation and resulting cell death and decreased thymocyte cellularity caused by the Xlf loss of function.

Xlf −/− TCRα Repertoire Bias Is Independent of Antigen Presentation
Having established that the MHC class I class II deficiency did not have an appreciable impact on cell death and thymocyte lifespan of Xlf deficient thymocytes, we wished to analyze the consequences on the generation of the TCRα repertoire. The V(D)J recombination of the TCRα locus is unique compared to the other TCR and Ig loci, in the sense that multiple waves of TCRα rearrangement occur until the expressed TCRαβ can recognize an antigen presented by MHC molecules for positive selection (32). Thymocyte viability directly affects the number of successive VαJα rearrangements. Decreased DP thymocyte lifespan (such as in the RORγT and TCF-1 deficient mice) leads to the loss of distal VαJα rearrangements and deficit of iNKT cells, which require the specific distal Vα14Jα18 rearrangement, in patients and mice (28,33,34). On the other hand, extending thymocyte lifespan [through Bcl-x(L) transgene for example] leads to the opposite effect; namely an overrepresentation of distal rearrangements (28,29). The TCRα repertoire is altered in Xlf deficient mice and patients possibly as a consequence of DP thymocyte lifespan decrease (24,35). We analyzed the impact of MHC deficiency on the TCRα repertoire through 5 ′ RACE RT-PCR followed by deep sequencing of mTRAV-mTRAJ junctions from whole thymocytes (Figure 2). The skewed TCRα repertoire, with strongly decreased usage of distal VαJα rearrangements and the concomitant increased usage of proximal VαJα rearrangements (Figures 2A,B), which is characteristic of Xlf deficiency, was not rescued by the MHC deficient background. Indeed, principal component analysis (PCA) and unsupervised hierarchical clustering (HC) analysis using PROMIDISα biomarker (30), which takes into account several parameters of TCR Vα and Jα usage, grouped Xlf −/− and Xlf/MHC TKO thymocytes in the same cluster, distinct from that of WT and MHC DKO thymocytes ( Figure 2C).
From this first set of analyses, we conclude that putative unrepaired DSBs that could occur in the course of T-cell activation during positive selection of DP thymocytes does not account for the decreased thymocyte viability and the skewed TCRα repertoire in Xlf −/− mice, since none of these phenotypical traits were rescued in the absence of positive selection subsequent to MHC class I and class II genes inactivation. Nevertheless, Xlf −/− thymocyte loss of viability is connected to the DP stage, when V(D)J recombination at the Tcra loci is taking place. Interestingly, biased TCRα repertoire with a similar loss of distal VαJα rearrangements has recently been described in various human conditions characterized by hypomorphic mutations in known factors of the V(D)J recombination machinery (i.e., RAG1, Artemis, DNA Ligase IV, Cernunnos/Xlf, PRKDC genes) (30). These hypomorphic mutations, which result in subefficient TCRα rearrangement waves, ultimately translate into a skewed TCRα repertoire when the repertoire of other Tcr loci appears grossly unaffected. This raises the possibility that the biased TCRα repertoire in Xlf −/− mice may not be the consequence of the decreased thymocyte viability per se (such as in RORγT deficient mice) but secondary to a subefficient V(D)J recombination activity (such as in RAG1 hypomorphic conditions).

DNA Repair Defect During TCRα Rearrangements in Xlf −/− DP Thymocytes
To test the possibility that subefficient V(D)J recombination resulting in unrepaired RAG1/2 induced DSBs may be responsible for the phenotype of Xlf −/− thymocytes we first analyzed γH2A.X and phospho S2056 DNA-PKcs DNA repair foci by immunofluorescence in total thymocytes, as surrogate markers of unrepaired DSBs. Quantification of γH2A.X (Figures 3A,B) and phospho S2056 DNAPKcs foci (Figures 3C,D) showed a significant increase in cells with 1 or 2 DNA repair foci in Xlf −/− vs. WT thymocytes, with an increase of 16.76 to 26.31% (p < 0.0001) of cells with 1 focus and 4.18 to 11.55% (p < 0.0001) of cells with 2 foci for γH2A.X and 23.05 to 27.70% (p = 0.006) of cells with 1 focus and 3.73 to 9.72% (p < 0.0001) of cells with 2 foci for phospho S2056 DNAPKcs foci in Xlf −/− vs. WT thymocytes. The frequency of thymocytes with more than 3 γH2A.X or phospho S2056 DNAPKcs foci, which may represent dying cells, also increased from 2.4 to 8.4% (p < 0.0001) and 2.3 to 6.5% (p < 0.0001) for γH2A.X and phospho S2056 DNA-PKcs respectively in WT vs. Xlf −/− thymocytes, highlighting the overall thymocyte decreased viability. To analyze whether the increased γH2A.X foci indeed corresponded to unrepaired DSBs occurring during V(D)J recombination at Tcra loci at the DP stage in Xlf −/− condition, we sorted CD4+CD8+ DP thymocytes from 6 to 9 weeks mice, and performed Tcra-γH2A.X association analysis by immuno-DNA fluorescence in situ hybridization (FISH). We used an antibody against γH2A.X as a read-out for random and RAG mediated DSBs (36) in combination with two DNA probes that hybridize to the 5 ′ and 3 ′ ends of the full length Tcra/d locus, respectively (37) (Figure 3E). The full volume of thymocyte nuclei were analyzed by confocal microscopy. The association of a γH2A.X focus with the Tcra locus specifically in DP thymocyte reflects ongoing TCRα rearrangements ( Figure 3F). It has been previously described that sorted DP thymocytes from WT mice show around 40% of cells with monoallelic Tcra-γH2A.X association, thus performing V(D)J recombination on one TCRα allele; and 10% of cells with biallelic Tcra-γH2A.X association, rearranging the two alleles at the same time (38,39). Defects in the V(D)J recombination machinery lead to various phenotypes. Atm −/− and RAG2 c/c thymocytes show a "recombination" defect, with a normal frequency of monoallelic Tcra-γH2A.X association but a severely increased biallelic association (39) and association of both Tcra-γH2A.X and Igh-γH2A.X in individual DP thymocytes (38). This misregulated V(D)J recombination with multiple simultaneous RAG cleavages is associated with early T-lymphoma of DP origin in Atm −/− (19,40) and RAG2 c/c mice crossed on Trp53 −/− (12). On the other hand, 53BP1 −/− mice show a "DNA repair" defect, with severely increased monoallelic Tcra-γH2A.X association and normal biallelic association, meaning that while the regulation of safe monoallelic RAGmediated cleavage is efficient, repair of the RAG-mediated DSB is delayed (38).
In this work, WT thymocytes showed 40.2% of monoallelic Tcra-γH2A.X association, and 9.2% of biallelic association ( Figure 3G) as previously described (38,39). In contrast, DP thymocytes from 2 independent Xlf −/− mice showed a modest but statistically significant increase in monoallelic Tcra-γH2A.X association (47.6 and 47.0% vs. 40.2%, p = 0.004 and p = 0.006), thus recapitulating a V(D)J "DNA repair" defect to some extent. Biallelic association was only very slightly and not significantly increased, with 11.3 and 11.4% in Xlf −/− cells compared to 9.2% in WT thymocytes. These results suggest that a weakened DNA repair function during TCRα rearrangements in Xlf −/− DP thymocytes could be responsible for the chronic P53 response in these cells, resulting in their reduced viability, thus explaining the immune phenotype.  (27).
The commitment of Xlf −/− DN thymocytes to the DN3 stage was not affected, with normal proportions of DN1 and DN2 subsets ( Figure 4A). In contrast, Xlf −/− thymocytes demonstrated a statistically significant increase in the DN3 subset (64.1% in Xlf −/− vs. 52.5% in WT, p = 0.009) (Figure 4A), attesting for a development delay at this stage. As a consequence, fewer cells proliferated through the DN4 stage (21.9% in Xlf −/− vs. 32.9% in WT, p < 0.0001) ( Figure 4A). More precisely, Xlf −/− thymocytes accumulated at the DN3A stage (64.4% in Xlf −/− vs. 46.5% in WT, p = 0.008) ( Figure 4B). This development delay is an evidence of an altered V(D)J recombination in the DN3A stage, which may be overcome in Xlf −/− mice through the high proliferation during the subsequent β-selection and DN4 stage. Indeed, the proliferation of thymocytes through DN4A-B-C stages (data not shown) and the TCRβ repertoire are not affected in Xlf −/− thymocytes (24). The quantification of the absolute numbers of thymocytes revealed a decrease in the various DN subsets, which suggests an overall decrease in thymocyte fitness, independent of V(D)J recombination, in Xlf mice.

Delayed Intracellular IgM Expression in
Xlf −/− Pro-B Cells in Fetal Livers B-cell lymphopenia is an important feature of Xlf −/− condition both in men and mice (20,22,25). We wished to analyze to which extent a suboptimal V(D)J recombination could also participate in this aspect of the Xlf −/− phenotype. To evaluate the   (Figure 4D). Interestingly, this was associated with a mild, yet statistically significant, diminution of B220 low sIgM+ immature B-cells (8.7% in Xlf −/− vs. 19.3% in WT, p = 0.001) (Figure 4E), suggesting an improper generation of newly immature Bcells. Furthermore, we observed a concomitant moderate, yet significant, increase in B220 low CD43+ Pro-B cells (26.3% in Xlf −/− vs. 13.2% in WT, p = 0.002) among total B220+ cells (Figure 4F)

DISCUSSION
The alteration of the TCRα repertoire observed in Xlf deficient mice (and humans), with reduced utilization of distal TRAV and TRAJ elements suggested two non-exclusive hypotheses: (1) a decrease of thymocyte lifespan at the DP stage related to the positive selection mechanism or (2) a subefficient V(D)J recombination activity. By introducing the Xlf loss of function on a MHC class I and class II DKO we showed that the molecular events associated with positive selection of T-lymphocytes in the thymus have no appreciable impact on the survival of Xlf −/− thymocytes, their P53 chronic activation, and, ultimately, their TCRα repertoire.
We therefore favor the hypothesis of a suboptimal V(D)J recombination process. According to this hypothesis, both Band T-lymphocyte development are affected in Xlf −/− mice to some extent, with a moderate developmental delay at stages that involve single-or two-steps rearrangements, such as D-to-J and V-to-DJ rearrangements of Tcra and Igh in thymus, fetal liver, and bone marrow. Interestingly, the in-depth analysis of B cell maturation in the bone marrow of an Xlf deficient patient revealed that although all the maturation steps from Pro-B to mature B cells were represented arguing against an indispensable role of Xlf for V(D)J in humans, the relative high proportion of CD22+/CD19-Pro-B cells as compared to healthy controls was indicative of a partial block of B cell differentiation compatible with a suboptimal V(D)J recombination efficiency (42).
TCRα rearrangements in DP thymocytes proceed through multiple waves of V(D)J recombination ordered from proximal to distal VαJα rearrangements until the appropriate TCRαβ expressing thymocytes undergo positive selection. In Xlf −/− DP thymocytes, we observed an increase in the number of cells harboring a DNA repair focus (γH2A.X) on one TCRα allele, suggesting a moderate "DNA repair defect" during V(D)J recombination of Tcra loci. These subefficient rearrangements would accumulate at each one of these waves, ultimately leading to the described Xlf −/− mouse phenotype, i.e., decreased DP thymocyte lifespan through P53 pathway chronic activation accompanied by a biased TCRα repertoire with the loss of distal VαJα rearrangements.
Interestingly, although skewed TCRα repertoire was primarily associated with thymocyte decreased viability, such as in RORC deficient mice and patients (29,33,34), a similar bias with loss of distal VαJα rearrangements was more broadly described in various human conditions characterized by hypomorphic mutations in known factors of the V(D)J recombination machinery (i.e., RAG1, Artemis, DNA Ligase IV, Xlf/Cernunnos, PRKDC genes) with the newly developed tool PROMIDISα (30). Xlf −/− mice analyses further demonstrated that subefficient V(D)J recombination waves could accumulate and lead to immunodeficiency with impoverished TCRα repertoire. Furthermore, the Xlf −/− phenotype only comes out in patients and mice in vivo, and has been hidden in in vitro V(D)J assays in v-Abl Pro-B cells in many settings (23,25).
Although V(D)J recombination is subefficient in mice in vivo, DSBs are ultimately repaired since Xlf −/− Trp53 −/− DKO do not develop T or pro-B cell lymphomas (22) and the immune phenotype is even rescued on a Trp53 −/− background (24). This phenotype is quite different from that of 53BP1 −/− mice for example, which exhibit a more severe "DNA repair" defect during V(D)J recombination waves at Tcrα locus (38). Indeed, 53BP1 −/− mice show severe thymic lymphomas at 2 to 4 months of age with TCRαδ translocations (43). Thus, although V(D)J recombination is subefficient in Xlf −/− DP thymocytes, the DNA broken ends are not left unrepaired. This could be explained by the several DNA repair factors that are compensatory for Xlf defect in V(D)J recombination, such as PAXX, RAG2 Cterminus, MRI, ATM, and H2A.X. The double safe-lock provided by the RAG1/2 post cleavage complex (PCC) and many DDR factors (i.e., PAXX, ATM, H2A.X and MRI) (23,(25)(26)(27) on the one hand and the Xlf-XRCC4 filament (7,8) on the other hand may ensure that DNA broken ends are not left unrepaired.
In Xlf deficient patients and mice, B cell lymphopenia is a severe feature and is associated with Class Switch Recombination defects (25,44). Furthermore, B-cells decline in bone marrow and in spleen worsen with aging, and is associated with the deterioration of hematopoietic stem cells differentiation potential, which particularly affects all hematopoietic lineages in Xlf −/− aged mice (45). The proliferative failure of Xlf −/− hematopoietic stem cells is related to accumulation of unrepaired DSBs, as exemplified in a model of human induced pluripotent stem cell (46). Here we showed that subefficient V(D)J recombination leads to delayed generation of new mature Bcells. Subefficient Igh rearrangements could have additive effects with moderate Class Switch Recombination and hematopoietic stem cell defects, which ultimately lead to the premature loss of mature B-cells in adult Xlf −/− mice and Xlf/Cernunnos deficient patients.
Lastly, one cannot exclude an additional role of Xlf in V(D)J recombination, beyond its function during the DNA repair final step. In Xlf deficient patients, coming with TCRα repertoire bias, the Ig and TCRδ repertoires are strongly impoverished because of a specific defect in N-nucleotide addition by the TdT polymerase in productive and unproductive Ig and TCR rearrangements (35). This leads to the synthesis of overall one to three amino acids shorter Ig or TCR CDR3 regions. This lower junction diversity likely induces a poorer antigen recognition potential of Xlf −/− lymphocytes (35). Whether the Xlf −/− subefficient V(D)J recombination in vivo is linked to a defect in the random nucleotide incorporation by the TdT after Artemis exonuclease activity, leading to a possible delayed synthesis of ligatable DNA ends is an interesting possibility. However this hypothesis would not explain the proliferation delay we observed at the DN3A subset in the thymus, while IJspeert et al. do not show any N-nucleotide addition defect in TCRβ rearrangements.

Flow Cytometry Analysis of Cell Populations
Cell phenotyping from 6 to 9 week old mice was performed on thymus and bone marrow after short hemolysis according to standard protocols by seven-color fluorescence analysis. The following antibodies were used: CD4, CD8, CD25, CD28, CD44, CD69, B220, CD19, and IgM (all from Sony Biotechnology, using respectively PECy7, FITC, PerCPCy5.5, PE, BV510, APC, PE, PECy7, APC fluorophores). Intracellular IgM expression in E18.5 fetal liver cells was performed as previously described (27). The following antibodies were used: CD19, B220, CD43, IgM for extracellular staining followed by cell fixation and permeabilization (Invitrogen) and intracellular IgM staining (all from Sony Biotechnology, using respectively PECy7, BV605, PE, FITC, and APC). Cells were recorded by fluorescenceactivated cell sorting (FACS) LSRFortessa X-20 immediately after incubation with Sytox Blue (Life Technologies) to exclude dead cells (except for fetal liver), and analyses were performed with FlowJo 10 software.

Thymocyte Survival Assay
Ex vivo thymocyte survival assay was performed as previously described (24) with minor modifications. Single-cell suspensions were obtained from the thymus and cultured at 3.

Quantitative Real-Time RT-PCR Analysis
TaqMan PCR was performed on triplicates of 8 ng of reversetranscribed RNA from freshly dissected total thymus, as previously described (24) with minor modifications, using predesigned primer and probe sets from Applied Biosystems [Mm01303209_m1 for mouse Cdkn1a or P21 exons 1 and 2; Mm00432050_m1 for mouse Bax exons 4 and 5; Mm00519268_m1 for mouse Bbc3/PUMA exons 3 and 4; Mm 99999915_g1 for glyceraldehyde-3-phosphate dehydrogenase (GAPDH)]. mRNA expression levels were calculated with ViiA 7 Real-Time PCR System v1.1 (Applied Biosystems). GAPDH was used for normalization of expression, and RNA from WT littermates was used as the calibrator. The relative amounts of mRNA in samples were determined using the 2- CT method, where C T is the difference between C T (C T target-C T GAPDH) sample and C T (C T target-C T GAPDH) calibrator. Final results were expressed as n-fold differences in target gene expression for tested samples compared with the mean expression value for the controls.

Analysis of Thymic TCRα Repertoire
Comprehensive TCRα repertoire analyses were performed by 5 ′ Rapid amplification of complementary DNA (cDNA) ends (5 ′ RACE) PCR/NGS (switching mechanism at the 5 ′ end of the RNA transcript, SMARTα) from total thymus RNA as previously described (24). Five hundred base-pair PCR products were gel purified and processed for single-molecule Illumina sequencing. Sequencing data were analyzed with LymAnalyzer (47)

DATA AVAILABILITY
The datasets generated for this study are available on request to the corresponding author.

AUTHOR CONTRIBUTIONS
J-PdV conceived the project. BR, VA, and JC planned and performed the experiments. BR and J-PdV co-wrote the manuscript. JC revised the manuscript. All the authors agreed to the publication of this manuscript.

FUNDING
This work was supported by institutional grants from INSERM and the Institut National du Cancer (INCa, PLBIO16-280) and by grants from La Ligue Nationale contre le Cancer (Equipe Labellisée LA LIGUE 2017). JC was supported by an ATIP-Avenir program (Inserm/CNRS, France) and the ARC foundation (France). BR was supported by the ARC foundation (France).