ABSTRACT
Among hematopoietic cells, osteoclasts (OCs) and immature dendritic cells (DCs) are closely related myeloid cells with distinct functions: OCs participate skeleton maintenance while DCs sample the environment for foreign antigens. Such specificities rely on profound modifications of gene and protein expression during OC and DC differentiation. We provide global proteomic and transcriptomic analyses of primary mouse OCs and DCs, based on original stable isotope labeling with amino acids in cell culture (SILAC) and RNAseq data. We established specific signatures for OCs and DCs, including genes and proteins of unknown functions. In particular, we showed that OCs and DCs have the same α- and β-tubulin isotype repertoire but that OCs express much more of the β tubulin isotype Tubb6 (also known as TBB6). In both mouse and human OCs, we demonstrate that elevated expression of Tubb6 in OCs is necessary for correct podosomes organization and thus for the structure of the sealing zone, which sustains the bone resorption apparatus. Hence, lowering Tubb6 expression hinders OC resorption activity. Overall, we highlight here potential new regulators of OC and DC biology, and illustrate the functional importance of the tubulin isotype repertoire in the biology of differentiated cells.
INTRODUCTION
Osteoclasts (OCs) and monocyte-derived immature dendritic cells (DCs) are closely related cell types belonging to the monocyte lineage of myeloid hematopoietic cells. Although closely related and capable of trans-differentiation (Madel et al., 2019), OCs and DCs fulfill very distinct biological functions. On one hand, OCs reside in the bone tissue and are specialized for bone resorption; they ensure the maintenance of bone health in coordination with osteoblasts: the bone forming cells. OC function is also crucial for hematopoietic stem cell mobilization and they are closely linked with the immune system (Cappariello et al., 2014). On the other hand, immature dendritic cells are sentinels of the immune system that reside in peripheral tissues where they constantly sample the surrounding environment for pathogens, such as viruses or bacteria; upon non-self body encountering, DCs mature or differentiate into antigen-presenting cells to fulfill their immune-initiating function (Tiberio et al., 2018). Myeloid cell differentiation is associated with the expression of distinctive proteins related to their specific functions, such as the cathepsin K and the metalloprotease MMP9 in OCs for bone collagen fiber degradation (Cappariello et al., 2014) or the chemokine CCL17 in DCs for their migration and T cell chemo-attraction (Real et al., 2004). In particular, OCs have a very specific capacity to organize their actin cytoskeleton into a belt of podosomes, which is the backbone of the bone resorption apparatus (Georgess et al., 2014a; Touaitahuata et al., 2014), whereas DCs display clusters of podosomes that participate in antigen sampling (Baranov et al., 2014).
Unraveling the fine molecular mechanisms controlling the specific functions of OCs and DCs is key to understanding the homeostasis and pathological dysfunctions of the bone and immune system, and then to designing targeted genetic and pharmacological therapeutic strategies. In fact, improving knowledge about DC biology and pathology paved the way for the manipulation of DCs to enhance immune responses, in particular in the context of cancer immunotherapy (Sabado et al., 2017; Wculek et al., 2019). The exacerbated activity of OCs causes osteoporosis, a serious public health problem leading to significant morbidity and mortality; this abnormal increase in OC activity is associated with a number of pathologies ranging from age-related sexual hormone decay to chronic inflammation and cancer (Khosla and Hofbauer, 2017). Better understanding of OC biology allowed the development of more focused therapeutic solutions against osteoporosis, as demonstrated by the recent approval of Denosumab, an antibody against the receptor activator of NF-κB Ligand (RANKL), the key cytokine for OC differentiation (Compston et al., 2019), and the promising targets cathepsin K (Drake et al., 2017) or Dock5 (Vives et al., 2015).
Differential proteomics and transcriptomics proved powerful tools for identifying molecular pathways involved in specific cell functions. The comparison between OC and DC transcripts indeed revealed new regulators of bone resorption by OCs (Gallois et al., 2010; Georgess et al., 2014b) and combined OC transcriptomics and proteomics shed light on the changes in cellular functions associated with OC differentiation (An et al., 2014). Nevertheless, the very few studies available in myeloid cells are difficult to compare owing to different cellular systems and experimental set up. Furthermore, the only proteomics data available for OCs were obtained using OCs derived from the RAW264.7 cell line, which proved to be very different from primary OCs, in particular regarding cytoskeleton regulation (Ng et al., 2018), which is key for bone resorption (Georgess et al., 2014a; Touaitahuata et al., 2014) and a promising therapeutic target against osteoporosis (Vives et al., 2015).
To overcome the limitation of currently available data, we provide here comparative quantitative proteomics and transcriptomics data of primary OCs, DCs and bone marrow macrophages (MOs), differentiated ex vivo from the same bone marrow cells. By minimalizing the culture condition differences to the cytokine cocktails used to differentiate each myeloid cell type, we could establish transcriptomic and proteomic signatures for OCs and DCs. We then exploited the data to uncover potential new regulators of OC biology and highlighted the β-tubulin isotype Tubb6 as key for OC cytoskeleton organization and bone resorption capacity.
RESULTS
Global proteome and transcriptome of primary myeloid cells
Osteoclasts (OCs) and immature dendritic cells (DCs) are myeloid cells that perform very distinct biological functions. In order to gain new insight into the molecular mechanisms underlying their specific activities, we sought to compare the global proteome and transcriptome of primary OCs and DCs, which are post-mitotic myeloid cells, using bone marrow macrophages (MOs), which retain the ability to divide in our experimental conditions, as reference monocytic myeloid cell type. To achieve this, and to minimize irrelevant variations due to culture conditions, we set up conditions to derive the three myeloid cell types from the same mouse bone marrow cells with the only variation in culture medium composition being restricted to cytokines: GM-CSF for DCs, M-CSF for MOs and M-CSF+RANKL for OCs (Fig. 1A,B). In each experiment, we monitored DC and OC differentiation during and at the end of differentiation by controlling cell morphology and the cytoskeleton using phase-contrast and fluorescence microscopy (Fig. S1A,B).
To quantify proteins, we performed stable isotope labeling with amino acids in cell culture (SILAC) with isotope-labeled Arg and Lys: heavy-isotopic labeling (H) for OCs, intermediate-isotopic labeling (M) for DCs and no isotopic labeling (L) for MOs (Fig. 1A), incorporating isotope-labeled amino acids at the time of cytokine addition in order to label only proteins neo-synthetized during DC and OC differentiation. Protein levels were compared between OCs and MOs (H+L), DCs an MOs (M+L), and OCs and DCs (H+M). To quantify transcripts, we performed RNAseq in identical culture conditions as SILAC except for labeled amino acids (Fig. 1B). Principal component analysis showed that SILAC and RNAseq biological replicates were grouped, whereas the three SILAC clusters (H+L, M+L and H+M) and the three RNAseq clusters (OCs, DCs and MOs) were clearly separate (Fig. S2A,B), as confirmed by multi-scatter plotting (Figs S3 and S4). We further used these global SILAC and RNAseq data to identify the proteomic and transcriptomic specificities of primary OCs and DCs.
SILAC identified a total of 2820 protein groups considering one peptide for adequate protein identification, among which 2769 corresponded unambiguously to a single UniProt identifier and to a single protein (Table S1). Among these, 1942 were present in all three myeloid cell comparisons, the others in only one or two (Fig. 1A). The normalized log2 protein abundance ratios were in the same range in all comparisons, between around −7 and +7 (Fig. 1C and Table S1). Considering only genes annotated as ‘protein coding’ in the Ensembl database, the RNAseq identified 15,597 mouse Ensembl genes in OCs, 16,197 in DCs and 15,875 in MOs, representing collectively 17,199 different genes, with most of the transcripts (14,663) present in the three myeloid cell types (Fig. 1B and Table S2). Differential expression analysis of the four replicate experiments picked 4781 significant transcript differences in OCs versus MOs, 6903 in OCs versus DCs and 6766 in DCs versus MOs (padj<0.05 in Table S2). The log2 fold changes were in the same range for in all comparisons, between roughly −10 and +10 (Fig. 1D and Table S2). For gene expression profiling, we excluded the 7624 genes with similar expression levels in OCs, DCs and MOs (italic in Table S2) and the 540 genes with a mean expression level below 10 fragments per kilobase million (FPKM) in the three cell types (italic in Table S2). Hierarchical clustering analysis of the remaining 9035 genes revealed five gene clusters, in particular cluster E corresponding to elevated expression in OCs, and clusters A and C corresponding to elevated expression in DCs (Fig. 2).
Analysis of differential protein and gene expression between myeloid cells
We then identified proteins and genes characteristic of OCs and of DCs. For proteins, expression levels were considered different when the normalized log2 abundance ratio was ≥0.75 or ≤−0.75. Thereby, we established an OC protein signature of 144 proteins, present at lower levels in both DCs and MOs (Fig. 3A and Table S3). Similarly, the immature DC signature comprised 181 proteins (Fig. 3A and Table S4). Conversely, 80 proteins were less abundant in OCs (Fig. 3A and Table S5) and 147 proteins were less abundant in DCs (Fig. 3A and Table S6). A similar analysis for the transcripts analyzed by hierarchical clustering led to OC and DC transcriptional signatures of 1207 and 2419 genes, respectively (Fig. 3B and Table S2). Finally, 554 genes were expressed at lower levels in OCs (Fig. 3B and Table S2) and 1587 genes were expressed at lower levels in DCs (Fig. 3B and Table S2).
R2 correlation coefficients between proteomics and transcriptomics were low, around 0.3 (Fig. 3C), in agreement with the correlations reported in various biological and experimental systems, including in OCs derived from the RAW264.7 cell line (An et al., 2014; Ghazalpour et al., 2011). In fact, among the 144 proteins of the OC signature, only 72 genes were also in the OC transcriptional signature, two were repressed (Table S3, pink and purple, respectively) and 70 changed differentially or did not change significantly when compared with DCs and MOs. Similarly, among the 181 proteins of the immature DC signature, 138 genes were also in the transcriptional signature and five were repressed (Table S4, pink and purple respectively). Not withstanding these discrepancies, RNAseq provides a much deeper view than SILAC of genome expression during OC and DC differentiation. For example, no proteins were detected in SILAC for genes essential in OC biological functions present in the OC signature, such as Siglec15 (Hiruma et al., 2013) and Bcar1 (also known as p130Cas) (Nagai et al., 2013).
Finally, we examined the molecular function (MF) and biological process (BP) gene ontology (GO) terms associated with OC and DC signatures. All the proteins and all the genes identified by SILAC and RNAseq were, respectively, used as reference proteome and transcriptome. In the OC protein signature, all enriched GO terms pointed to energy metabolism by the mitochondria (Fig. 4A and Table S7), in agreement with the upregulation of tricarboxylic acid cycle enzymes during osteoclastogenesis and elevated energy metabolism in OCs (An et al., 2014; Lemma et al., 2016). Transcriptionally enriched GO terms in OCs also additionally included mRNA translation (Fig. 4B and Table S8). For DC signatures, all enriched GO terms related to immune functions (Fig. 4C,D and Tables S9, S10). Consistent with the OC GO term analysis, the main protein-protein interaction (PPI) networks among the OC transcriptional signature were linked to mRNA translation and mitochondrial energy production (Fig. 4E-G).
Overall, the proteomic and transcriptomic approaches in primary myeloid cells proved complementary to identify biologically relevant species characteristic of OCs and DCs. Moreover, the OC and DC signatures highlighted species of unknown function in these cell types, which could bring novel insights into the specific biological functions of these myeloid cells.
Identification of potential new regulators of OC biology
The 144 proteins of the OC signature comprised 27 known regulators of OC differentiation and activity (green highlight in Table S3), including cathepsin K (CATK) and tartrate-resistant acid phosphatase type 5 (PPA5) (Cappariello et al., 2014), the vATPase complex, a3 subunit Tcirg1 (Ochotny et al., 2011) and Dock5 (Vives et al., 2011, 2015), and other vATPase subunits, mitochondrial proteins and proteins involved mRNA translation. Interestingly, there were also 45 proteins of unknown function in OCs (light- and dark-blue highlight in Table S3), including proteins involved in the cytoskeleton and intracellular trafficking, key processes for bone resorption (Cappariello et al., 2014). Conversely, of the 80 less abundant proteins in OCs, seven were known inhibitors of OC differentiation or function (green highlight in Table S5), such as Serpinf1 (Pedf), which is involved in human type VI osteogenesis imperfecta (Homan et al., 2011) and Fkbp5, which is associated with human Paget's disease (Lu et al., 2017). So far, the other 73 proteins have not been linked to OC biology and could be new OC inhibitors (Table S5).
We then compared our data with the only SILAC global proteomic data available so far for OCs, concerning osteoclastogenesis in RAW 264.7 cells (An et al., 2014). The RAW 264.7 cell line was derived from the lymphoma of a mouse infected with Abelson murine leukemia virus; it can differentiate into OCs in the presence of RANKL. As shown by a recent SILAC study, the proteome of RAW 264.7-derived OCs is very distinct from that of primary OCs, in particular regarding cytoskeletal organization; unfortunately, as the data are not available, we could not compare it with ours (Ng et al., 2018). 2073 proteins were identified by SILAC during RAW 264.7 osteoclastogenesis (An et al., 2014), among which 1506 were common to our SILAC in primary OCs. They comprised 92 proteins of our primary OC protein signature, but only 56 were upregulated during RAW264.7 OC differentiation (Table S3, ↑), even though the log2 expression ratio cut off set at 0.5 in that study was less stringent than ours. The other proteins were downregulated or unchanged during RAW264.7 differentiation (orange highlight in Table S3; respectively, ↓and =), including known regulators of OC biology induced during primary OC differentiation, such as tensin 3 (TNS3) (Touaitahuata et al., 2016) and CD68/macrosialin (Ashley et al., 2011). Moreover, 52 proteins from our OC signature were not detected in RAW 264.7 OCs (red highlight in Table S3, n.f.). Overall, our primary OC signature comprised 89 proteins not highlighted in RAW264.7 OCs, the only SILAC proteome data available so far for OCs (An et al., 2014), among which unknown regulators of osteoclast biology could be present.
These 89 proteins comprised: 15 known actors of OC biology (green highlight in Table S3) and 36 proteins involved in biological processes activated in OCs (members of the v-ATP complex, actors of mitochondrial energy metabolism and mRNA translation) and 38 proteins of unknown function in OCs (dark-blue highlight, Table S3). We picked about half of the remaining proteins for a siRNA screening, covering a variety of functions but avoiding the proteins with a role on chromatin (Tep1, Hmgb3, Histone H1.4 and H1.5) (bold lines in the dark-blue proteins of Table S3). To avoid interference with the early differentiation process, primary OC cultures were transfected at day 2 of differentiation with a commercial SmartPool of 4 siRNAs (Table S11) and grown for 2 more days to obtain OCs. We stained actin, tubulin and DNA, acquired mosaic images of entire wells on an automated microscope with a low-magnification objective and full-well images were reconstituted. The screen was performed in triplicate with three independent mouse OC cultures. None of the 17 test siRNAs provoked cytoskeleton organization defects visible at that image resolution; moreover, no significant changes in OC size were measured (not shown). Remarkably, with the five siRNAs targeting the AKR1C3 homolog Akr1c18 (AKC1H), Dmxl1, Trip11, Vps52 and Wdr7 (Fig. S5A), we observed a high proportion of OCs exhibiting very big and unusual intracellular structures delineated with actin and tubulin (Fig. 5A,B). The induction of OC differentiation markers Src and cathespin K was not affected by the siRNAs (Fig. 5C,D). The ‘vacuole-like structures’ had a regular circular shape and spanned over 10 µm in all directions (Fig. 5E). We hypothesized that such structures may arise from the disturbance of OC intracellular traffic. In fact, it was shown that the inhibition of lipid kinase PIKFYVE, a regulator of endolysosomal traffic, induces the formation of large endolysosomes in RAW264.7 cells (Hazeki et al., 2013). We did observe very large vesicles in OCs treated with the PIKFYVE inhibitor YM201636; these were positive for the endolysosomal marker Lamp3 (CD63) and not delineated by actin (Fig. S5B). This is in contrast to the vacuole-like structures elicited by the five siRNAs in OCs, which were surrounded by actin but were not delineated by any of the early and late endosomal marker we tested, including CD63 (Fig. 5E). This suggests that these unusual actin and microtubule-delineated structures are not membrane compartments; we could not relate them to any subcellular compartment reported so far and they had no effect on the amount of cathespin K inside the OCs or secreted, or on its maturation, as assessed by western blot (data not shown). These results suggest that our OC signature comprises new regulators of OC biology, which should be examined in more details to understand their functions.
Tubb6 regulates the OC cytoskeleton and bone resorption
The SILAC and RNAseq data also revealed changes in the expression of β-tubulins between OCs, DCs and MOs. Among all α- and β-tubulin genes, these cells express the same four α and four β isotypes: namely Tubb1a, Tubb1b, Tubb1c and Tubb4a, and Tubb2a, Tubb4b, Tubb5 and Tubb6. Tubb5 (tubulin beta class I) and Tubb6 (tubulin beta 6 class V) represent more than 80% of the Tubb reads in the three cell types (Table S2). The protein levels of the four α and four β isotypes levels were not different between OCs, DCs and MOs, except for Tubb6 (Table S1). Indeed, Tubb6 showed 1.375 and 1.256 log2 protein fold increases when compared with primary DCs and MOs, respectively, whereas such increases were not observed during RAW264.7 osteoclastogenesis (Table S3).
Tubb6 has been reported to have a microtubule destabilizing effect in cycling cells, in contrast with Tubb5 (Bhattacharya and Cabral, 2009; Bhattacharya et al., 2011). As microtubule dynamic instability is important for correct organization of actin and for bone resorption (Biosse Duplan et al., 2014; Guimbal et al., 2019), we explored the biological relevance of increased Tubb6 levels in OCs. Q-PCR analyses confirmed that, among all β-tubulin isotypes, Tubb6 expression uniquely increased during primary OC differentiation (Fig. 6A). To specifically target Tubb6, we generated the two siRNAs si1077 and si1373 (Table S11), which affect neither the expression of Tubb2a, Tubb4b and Tubb5 RNAs nor the global Tubb protein level in OCs (Fig. 6B,C) and do not modify the expression of the OC differentiation markers Src and Ctsk (Fig. S6A,B). Interestingly, we observed that Tubb6 siRNAs strongly perturbed the organization of the podosome belt, provoking podosome scattering at the cell periphery (Fig. 6D,E). This was accompanied by a severe modification of microtubule morphology at the cell periphery with the appearance of highly buckled microtubules (Fig. 6F). Using specific antibodies raised against the C terminus of mouse Tubb6 (Spano and Frankfurter, 2010), we found that the distribution of Tubb6 was indistinguishable from that of pan-β-tubulin, except for a lower amount of Tubb6 in the mononucleated cells surrounding osteoclasts (Fig. 6G), consistent with increased expression of the Tubb6 during OC differentiation (Fig. 6A).
Functionally, Tubb6 knockdown resulted in severe reduction in the activity of OCs (Fig. 7A). Our former analysis of Affymetrix data of human myeloid cells also revealed increased expression of TUBB6 mRNA in OCs derived from human CD14+ peripheral blood monocytes (Gallois et al., 2010; Maurin et al., 2018). Thus, we examined the effect of TUBB6 siRNAs in human OCs differentiated from peripheral blood of two independent donors. Consistent with the results in mouse OCs, the proportion of human OCs presenting a normal podosome belt was also strongly reduced (Fig. 7B). We further analyzed the importance of Tubb6 on bona fide sealing zones, examining OCs plated on mineralized substrates. In mouse OCs plated on coverslips coated with collagen mineralized with calcium phosphate (ACC), we observed that Tubb6 siRNAs provoked the appearance of numerous and very small sealing zones per OC, instead of the usual one or two bigger sealing zones in control OCs (Fig. 7C-E and Fig. S6C,D). This indicates that Tubb6 siRNAs induce a defect in sealing zone maturation. TUBB6 siRNAs also reduced the size of the sealing zones in human OCs differentiated from the peripheral blood of two independent donors and plated on bone (Fig. 7F-H). These results suggest that higher levels of Tubb6 are important for efficient bone resorption, impacting on the organization of osteoclast podosomes and the structure of the sealing zone. Tubb6 overexpression was found to cause dystrophy in both human and mouse myotubes, correlated with expression changes of muscular heavy chain myosin genes (Randazzo et al., 2019). Osteoclasts express two myosin heavy chain genes: Myh9 (myosin IIa) and Myh10 (myosin IIb) (Table S2). Myosin IIa, but not myosin IIb, is enriched at the podosome belt and the sealing zone; the silencing of myosin IIa was found to increase sealing zone size (McMichael et al., 2009). But in OCs treated with Tubb6 siRNAs, there was no modification of myosin IIa protein levels (data not shown).
Altogether, these results establish that the β-tubulin isotype repertoire is of functional importance in myeloid cells. In particular, the specific increase in Tubb6 protein levels that occurs during OC differentiation is essential for microtubules to ensure correct patterning of OC podosomes for efficient bone resorption.
DISCUSSION
We have combined SILAC proteomics, RNAseq and homogenous experimental conditions to provide the first comparative global proteomes and transcriptomes of primary mouse OCs and DCs. From the data, we propose specific transcriptomic and proteomic signatures for each cell type, including proteins of unknown function that represent new candidate regulators of OC and DC biology. Using a siRNA approach, we explored several of these new candidates in OCs. In particular, we found that OCs, MOs and DCs express the same repertoire of four α- and four β-tubulin isotypes. Within this repertoire, only Tubb6 protein is differentially expressed, with higher levels in OCs. We further demonstrated that this increased expression of Tubb6 is essential for podosome organization and the control of sealing zone size in both mouse and human OCs, and that it is key for OC resorption function.
The hematopoietic cells of the monocytic lineage are innate immune cells characterized by a great level of plasticity: mouse bone marrow macrophages and human CD14+ monocytes can differentiate in DCs and OCs; moreover, DCs have the capacity to transdifferentiate into OCs (Madel et al., 2019). Still, each cell type has very specific biological functions requiring particular cellular processes: OCs are the professional bone-resorbing cells that participate in the maintenance of skeleton health, whereas DCs are the sentinel antigen-presenting cells that sample their environment for ‘non-self’ entities. Unraveling the molecular pathways underlying the biology of OCs and DCs is key to understanding both their physiological and pathological roles. There are only a few proteomic datasets available for OCs; furthermore, because of the difficulty in generating sufficient primary OCs for sample preparation, all were obtained using RAW264.7 cells as OC precursors (Segeletz and Hoflack, 2016). RAW264.7 is a transformed macrophage-like cell line with osteoclastogenic potential, but this occurs in a M-CSF-independent manner; it was reported recently that there are major proteome discrepancies between primary OCs and OCs derived from RAW264.7 cells, in particular regarding cytoskeleton regulatory pathways (Ng et al., 2018). Here, we have developed an experimental approach to make primary MOs, OCs and DCs amenable to quantitative SILAC proteomics. Thereby, we provide the first global quantitative proteomic dataset for primary OCs. In the present study, we focused our analyses on the OC signature. Of note, the DC signatures contain genes of unknown functions in these cells, including strongly differential genes that are poorly studied overall, such as BC035044, D630039A03Rik or Adgrg5, some of which could be relevant for DC biology. For example, Plet1 was recently shown to be involved in interstitial migration of murine small intestinal DCs (Karrich et al., 2019).
To obtain the three myeloid cell types, we started from the same mouse bone marrow and used the same culture medium, just changing the cytokines added. The aim was to minimize the non-relevant changes that would not be related to the differentiation programs of the each cell types. To remain compatible with the technical requirements of our study, in particular the amount of material required for SILAC, we used plastic as a substrate for OCs, DCs and MOs. Although not their physiological environment, such a procedure provides cells competent for their physiological functions. For example, on plastic or glass, OCs have a ruffled border and they secrete active cathepsin K (Fuller et al., 2010; Touaitahuata et al., 2014); when transferred onto bone, such OCs are able to resorb it. Yet previous reports show that seeding OCs onto bone induces higher expression of various genes as compared to plastic (Crotti et al., 2011; Purdue et al., 2014), thus we may miss some of those genes in our OC transcriptomic signature. These articles explicitly mention 25 genes whose expression levels are increased in OCs plated on bone when compared with plastic. Examining how these genes were classified in our study, we found that 12 genes were indeed in our OC signature and four genes were not expressed (Table 1). Three genes were falling into our DC signature and their fold induction on bone does not reflect genes characteristic of resorbing OCs when compared with DCs. Finally, for the remaining six genes, the fold change in induction on bone was specified only for three genes: Bikunin, Xdh and Max. According to induction reported on bone and the expression we found in OCs, MOs and DCs, they may represent transcripts characteristic of the bone-resorbing OCs absent from the OC signature we established on plastic. Xdh and Max appear relevant in the bone-resorbing OCs. Max (Myc-associated factor X) is a partner of Myc that modulates global gene expression. Myc was indeed shown recently to drive metabolic reprogramming during OC differentiation and function, switching the metabolism to an oxidative state involving the TCA cycle and oxidative phosphorylation (Bae et al., 2017). On the other hand, Xhd encodes the xanthine dehydrogenase (or xanthine oxidase) that participates in the generation of reactive oxygen species (ROS) through xanthine metabolism. The ROS generated by xanthine dehydrogenase have been shown to increase OC formation and bone resorption (Fraser et al., 1996; Garrett et al., 1990). The gene Bikunin (also known as Ambp), is difficult to interpret as it encodes a precursor glycoprotein further cleaved into two proteins: the lipocalin alpha-1-microglobulin (AMBP) and the protease inhibitor bikunin (trypstatin). Overall, this confirms that OC differentiation on plastic does recapitulate the transcriptional differentiation program of functional bone-resorbing OCs. Still, the transcriptional profile elicited by the transcription factor Max could be interesting for highlighting genes specifically linked to the oxidative metabolism in the bone-resorbing osteoclast.
We compared our primary OC signature to the only SILAC global proteomics data reported for OCs, which was obtained from RAW264.7 cells (An et al., 2014). Among the proteins of our primary OC signature also detected in RAW264.7 OCs, one-third were not changed or downregulated during RAW264.7 osteoclastogenesis, including proteins induced during primary OC differentiation and essential for bone resorption, such as Tensin 3 (TENS3) (Touaitahuata et al., 2016) and Macrosialin (CD68) (Ashley et al., 2011). This confirmed the potential relevance of the proteins, but also likely the genes, in our primary OC signatures that remain of unknown function. The siRNA screen revealed five siRNAs provoking the appearance of large vacuole-like structures in OCs. The target genes have diverse assigned functions: Dmxl1 and Wdr7 encode essential vATPase accessory proteins (Merkulova et al., 2015); Trip11 codes for GMAP-210, which is involved in vesicle tethering to Golgi (Roboti et al., 2015); the product of Vps52 is involved in endosome recycling (Schindler et al., 2015); and Akr1c18 is the mouse homolog of aldoketo-reductase AKR1C3, which is involved in progesterone conversion (Lanišnik Rižner and Penning, 2019). Such intracellular structures delineated with actin and microtubules, but without endosomal membrane marker, have not been described previously, making it difficult to hypothesize what molecular mechanism may be perturbed. We found that inhibition of the lipid kinase PIKFYVE could induce very large intracellular structures of similar size in OCs but they were not delineated by actin and were endolysosomal membrane compartments, as reported in RAW264.7 cells (Hazeki et al., 2013). More studies are needed on these potential new regulators of OC biology, which would require stable knockdown or knockout models to understand their precise functions.
The OC signature also contained tubulin β-isotype Tubb6 (tubulin beta 6 class V). In OCs, MOs and DCs, we found that the tubulin repertoire was restricted to the expression of only four α- and four β-tubulin genes; they do not express RNA for other tubulin isotypes such as Tubb1, which is needed for pro-platelet elongation in megakaryocytes, another myeloid lineage hematopoietic cells (van Dijk et al., 2018). This suggests the existence of tight transcriptional control over the tubulin proteins during hematopoiesis, but the mechanisms regulating the transcription of tubulin genes remain very poorly understood (Gasic and Mitchison, 2019). How different tubulin isotypes affect microtubule dynamics to influence various processes in differentiated cells is an open question. The recent advances in producing recombinant tubulin isoforms started to shed light on the different biochemical properties of tubulin β isotypes that can influence microtubule dynamics in vitro (Roll-Mecak, 2019). Tubb6 has the unusual property of destabilizing microtubules in cycling cells, whereas this is not the case for Tubb5 (tubulin β-cell I); this effect appears to rely specifically on the core domain of Tubb6 and not on its tail (Bhattacharya and Cabral, 2009; Bhattacharya et al., 2011). Two amino acids, Ser239 and Ser365, in the core domain are crucial for the microtubule-destabilizing effect of Tubb6, and they are substituted for Cys in Tubb5. Interestingly, Tubb3 bears the Ser239 and Ser365, as in Tubb6; on the other hand, Tubb2b and Tubb5 have Cys residues at those positions. Biochemical studies revealed that Tubb3 and Tubb2b have distinct effects in vitro: on microtubule dynamics; on their association with microtubule-associated proteins (MAPs); and on their structure (Pamula et al., 2016; Ti et al., 2018). Therefore, it is very likely that the level of Tubb6 can influence microtubule intrinsic properties as well as their dynamic behavior and association with MAPs in OCs. Indeed, we found that the reduction of Tubb6 in OCs has a strong effect on microtubule and podosome organization, sealing zone size, resulting in impaired resorption activity. Contrary to Tubb5, which is usually a major β-tubulin isotype with Tubb4b, the expression of Tubb6 is low in most tissues (Leandro-García et al., 2010), and the tight control of Tubb6 expression levels appear key to various processes in differentiated cells. In fact, TUBB6 is among the most upregulated genes in Duchenne muscular dystrophy (DMD) and persistent elevation of Tubb6 levels drives microtubule disorganization in the fibers of muscles in DMD mice (Randazzo et al., 2019). TUBB6 levels also control the execution of pyroptosis by bacteria-infected lymphoblastoid cells (Salinas et al., 2014). We show here that higher amounts of Tubb6 in OCs are key to actin organization, sealing zone size and OC resorption function. Recent reports have highlighted the importance of microtubule dynamics and their crosstalk with actin to control podosome patterning in OC and bone resorption (Biosse Duplan et al., 2014; Guimbal et al., 2019; Zalli et al., 2016). Thus, the enrichment of Tubb6 in OCs is very likely to participate in the molecular processes controlling actin and microtubule crosstalk to ensure efficient bone resorption. Whether it functions through affecting intrinsic microtubule dynamics or MAP binding, or both, remains to be elucidated.
In summary, we designed a homogeneous experimental set-up to generate protein and RNA samples of primary OCs and DCs, leading to relevant protein and transcript signatures of the two myeloid cell types. In particular, we found that high levels of the β-tubulin isotype Tubb6 are characteristic of OCs and functionally relevant. Further studies on the specific properties of Tubb6 will bring better knowledge about the crosstalk between microtubules and the actin cytoskeleton, which remains poorly understood, taking advantage of the unique functional context of bone resorption by OCs. This should also be valuable in the context of osteolytic bone diseases and pave the way for novel strategies to treat osteoporosis by targeting the OC cytoskeleton.
MATERIALS AND METHODS
Animals, human samples and ethics statement
Four-week-old C57Bl/6J mice were purchased from Harlan France and maintained in the animal facilities of the CNRS in Montpellier, France. Procedures involving mice were performed in compliance with local animal welfare laws, guidelines and policies, according to the rules of the regional ethical committee. Monocytes from healthy subjects were provided by Etablissement Français du Sang (EFS) (Toulouse, France), under contract 21/PLER/TOU/IPBS01/20130042.
Production of bone marrow macrophages, osteoclasts and immature dendritic cells for SILAC and RNAseq
Bone marrow cells were extracted from long bones of C57BL/6J mice at 6-8 weeks of age, as described previously (Guimbal et al., 2019), and cultured in α-minimal essential medium (α-MEM, Lonza) containing 10% fetal bovine serum (FBS, BioWest) (heat inactivated at 56°C for 30 min), 2 mM glutamine (Lonza) and 100 U/ml penicillin and streptomycin (Lonza). Cell density was 3×107 cells per 150 mm dish incubated at 37°C in 5.5% CO2, in a humidified incubation for 24 h. Non-adherent cells were then collected and used to differentiate the three myeloid cell types and prepare samples for stable isotope labeling by amino acids (SILAC) and RNAseq (Fig. 1A,B). For bone marrow macrophages (MOs) and osteoclasts (OCs), non-adherent cells were then plated at a density of 5×106 cells per 10-cm plate in the same medium containing 100 ng/ml macrophage colony-stimulating factor (M-CSF, Miltenyi) for 5 days. Cells were then detached using Accutase (Sigma), washed in PBS and resuspended in Roswell Park Memorial Institute (RPMI) medium: SILAC RPMI 1640 (Lonza) supplemented with 10% dialyzed FBS (Life Technologies), 2.4 mM L-Prolin (Sigma), 2 mM glutamine (Lonza) and 100 U/ml penicillin/streptomycin (Lonza). For MO RNAseq and SILAC sample preparation, cells were then plated at 0.8×105 cells per well in 24-well plates in RPMI medium supplemented with 30 ng/ml M-CSF and with the L-enantiomers amino acids Lys (0.4 mM) and Arg (0.8 mM) (Sigma) non-isotope-labeled (L amino acids). For OC RNAseq sample preparation, cells were plated at 1×105 cells per well in 24-well plates in RPMI medium supplemented with 30 ng/ml M-CSF and 50 ng/ml receptor activator of NF-κB ligand (RANKL, Miltenyi) and with L amino acids as above. For OC SILAC samples, L amino acids were substituted for the same concentrations of heavy isotope-labeled (H) amino acids 13C6-15N2 L-Lys:2HCl and 13C6-15N4 L-Arg (Cambridge Isotope Laboratories). For immature dendritic cell (DC) RNAseq sample preparation, cells were plated at 4×105 cells per well in 24-well plates in RPMI medium supplemented with 20 ng/ml granulocyte macrophage colony stimulating factor (GM-CSF, Miltenyi) and with non-labeled amino acids, as above. For DC SILAC sample preparation, L amino acids were substituted for the same concentrations of intermediate isotope-labeled (M) amino acids 4,4,5,5-D4 L-Lys:2HCl and 13C6 L-Arg (Cambridge Isotope Laboratories). All media were changed every other day for 8 to 10 days. Under these conditions, H and M amino acid incorporation in OCs and DCs, respectively, was over 95%, and the Arg to Pro conversion rate was below 0.4%.
Protein and RNA extraction
Following differentiation, OC cultures were incubated with 120 µl per well Accutase (Sigma) for a few minutes to remove mono-nucleated cells. Then all cultures were rinsed once with PBS and processed for protein or RNA extraction. For SILAC samples, cells were lysed in 8 M urea (Sigma) and 10 mM HEPES (Sigma) (pH 8.0), with 12 wells per cell type in a total of 400 µl. Cell lysates were centrifuged at 20,000 g for 2 min; protein concentration was determined using BCA reagent (ThermoFisher) and was around 25 µg/ml per sample. Lysates were stored at −80°C until SILAC proteomic analysis. For each SILAC analysis, OC, DC and MO protein samples were derived from the same mouse; the experiment was carried out in triplicate with three independent mouse bone marrow extractions.
For RNAseq, total RNA was extracted and purified using RNeasy Mini Kit and QIAshredder spin columns (Qiagen) according to manufacturer's instructions (eight wells for DCs, 12 wells for MOs and four wells for OCs) to obtain around 10 µg of RNA. RNA concentration was measured using a Nanodrop UV-visible spectrophotometer (Thermo Scientific) and RNA quality was assessed using a Bioanalyzer 2100 (Agilent Technologies). Lysates were stored at −80°C until RNAseq analysis. For each RNAseq analysis, OC, DC and MO RNA samples were derived from the same mouse, the experiment was carried out four times with four independent mouse bone marrow extractions.
RNAseq analyses
The 12 RNA samples (2 µg) were processed and analyzed in parallel by Fasteris SA (Switzerland), according to the HiSeq Service Stranded Standard Protocol’ (https://support.illumina.com/sequencing/sequencing_instruments/hiseq-3000.html). The stranded mRNA libraries were sequenced by HiSeq 4000 Illumina technology, generating single reads of 1×50 bp. Adapter sequences were removed from the obtained 1×50 bp reads and adapter trimmed reads were used for further analysis. About 30 million raw reads were obtained per sample (from 26,717,590 to 36,916,924), with around 99% of the reads mapping on reference mouse genome GRCm38. Multiple mapping percentages ranged between 23.62 and 32.43% according to sample (Fig. S1C). Sequence mapping (Mus musculus genome GRCm38, from iGenome downloaded on the 2017-07-13), normalization and estimation of transcript abundances (FKPM) were performed using the Tuxedo suite of short read mapping tools (Bowtie v2.0.5, Tophat v2.0.6, Samtools 1.2 and Cufflinks v2.1.1). Differential expression analysis was performed with DESeq2 R package from Bioconductor v2.13. For each comparison by pairs, the mean of the normalized counts obtained for the four replicates within each group of samples was calculated as well as the log2 fold change. The p, adjusted for multiple testing with the Benjamini-Hochberg procedure, was used to qualify fold changes as significant (padj<0.05).
Mass spectrometry sample preparation, LC-MS-MS analysis and protein quantification
Within each of the three replicate SILAC experiments, equal amounts of OC and MO, OC and DC, or DC and MO proteins were mixed, leading to a total of nine samples that were processed in parallel. All MS grade reagents were from Thermo Fisher Scientific, grade Optima. For each sample, a total of 10 µg of proteins were digested with 0.5 µg LysC (Wako) in 600 µl of 100 mM triethylammonium bicarbonate for 1 h at 30°C. Then samples were diluted three times in the buffer and 1 µg trypsin (Gold, Promega) was added overnight at 30°C. Peptides were then desalted using OMIX (Agilent) and analyzed online by nano-flow HPLC-nanoelectrospray ionization using a Qexactive HF mass spectrometer (Thermo Fisher Scientific) coupled to a nanoLC system (U3000-RSLC, Thermo Fisher Scientific). Desalting and preconcentration of samples were performed online on a Pepmap precolumn (0.3×10 mm; Thermo Fisher Scientific). A gradient consisting of 0-40% B in A for 120 min (A: 0.1% formic acid, 2% acetonitrile in water; B: 0.1% formic acid in 80% acetonitrile) at 300 nl/min, was used to elute peptides from the capillary reverse-phase column (0.075×500 mm, Pepmap, Thermo Fisher Scientific). Data were acquired using the Xcalibur 4.0 software. A cycle of one full-scan mass spectrum (375-1500 m/z) at a resolution of 60,000 (at 200 m/z), followed by 12 data-dependent MS/MS spectra (at a resolution of 30,000, isolation window 1.2 m/z) was repeated continuously throughout the nanoLC separation.
Raw data analysis was performed using the MaxQuant software (version 1.5.5.1) with standard settings. Databases used consist of mouse entries from UniProt (reference proteome UniProt 2016_11), 250 contaminants (MaxQuant contaminant database) and corresponding reverse entries. Each protein identified was associated to the corresponding gene in Mus musculus genome GRCm38. Relative protein quantifications were calculated on the median SILAC ratios to determine log2 fold changes. In each comparison, proteins with |log2 abundance ratio|≥0.75 were considered differentially abundant.
Other bioinformatic analyses
Principal component analyses of both proteomic and transcriptomic results were made using R software. For RNAseq, the hierarchical clustering analysis was performed in MeV tool v4.9 (Saeed et al., 2003) using the four normalized expression counts for each cell type. Before analysis, values were normalized by rows using the software. An average linkage algorithm from a distance matrix calculated with the Pearson correlation coefficient was employed to make the analysis. Comparison between RNAseq and SILAC results and linear regressions were made using Graphpad Prism 5 software. GO term enrichment analyses were performed using Panther software (pantherdb.org/). The total lists of proteins and genes identified in this study were used as references. UniProtIDs and Ensembl IDs were employed as input type IDs for proteomic and transcriptomic results, respectively. GO terms with P<0.05 were considered significantly enriched. Histograms of enriched or diminished GO terms were represented using Graphpad Prism 5 software. Protein-protein interactions (PPIs) were determined using the Search Tool for Retrieval of Interacting Genes/Proteins (STRING) software v10.5 (string-db.org/) (Szklarczyk et al., 2015). Only interactions based on experimental source with interaction score ≥0.7 were considered. PPIs networks were then represented using Cytoscape v3.6.1 (www.cytoscape.org/index.html) (Shannon et al., 2003).
Mouse OC cultures for siRNA treatment and immunofluorescence
Mouse primary OCs were differentiated from bone marrow cells of 6- to 8-week-old C57BL/6J mice as described previously (Guimbal et al., 2019). At day 2 of differentiation, siRNAs were transfected with siImporter in OptiMEM medium (Life Technologies) containing 30 ng/ml M-CSF and 50 ng/ml RANKL, as described previously (Touaitahuata et al., 2016) using 100 nM siRNA: either Dharmacon siGenome Smartpools or custom oligonucleotides from Eurogentec (Table S11). After 3 h, medium was replaced with α-MEM medium containing 30 ng/ml M-CSF and 50 ng/ml RANKL. When relevant, OCs were transferred after another 4 h onto ACC-coated coverslips prepared as described previously (Maurin et al., 2018). Medium was changed every other day for 3-4 days, until OCs differentiated. For PIKFYVE inhibition, OCs were treated for 1 h with 1 µM YM201636 (Cayman) in 0.005% DMSO.
OCs were then fixed in 3.2% paraformaldehyde and 10 µM Taxol (Sigma) in PHEM (PIPES 60 mM, HEPES 25 mM, EGTA 10 mM, MgSO4 4 mM at pH 6.9) for 20 min. For tubulin detection, cells were permeabilized for 1 min with 0.1% Triton X-100 in PBS and saturated with blocking buffer (1% BSA in PBS) before incubation for 1 h with anti-α-tubulin antibodies (T5168, Sigma; 1:2000) in PBS containing 1% BSA. For CD63 (LAMP-3), cells were permeabilized for 2 min with 0.1% saponin in PBS and saturated with blocking buffer before incubation for 1 h with antibody R5G2 (MBL Life Science; 1:200) in PBS containing 1% BSA. Alternatively, for Tubb6 localization, OCs were fixed for 10 min in methanol at −20°C and saturated with blocking buffer (1% BSA in PBS) before incubation for 1 h in PBS containing 1% BSA with anti pan-β-tubulin antibodies (E7, Developmental Studies Hybridoma Bank; 1:2000) and anti-mouse Tubb6 antibody, a generous gift from Dr Anthony Frankfurter (Virginia University, Charlottesville, VA) who generated and purified the antibody (Spano and Frankfurter, 2010). In all cases, cells were then incubated for 1 h in Alexa-conjugated secondary antibodies and/or F-actin marker Alexa-conjugated phalloidin (Life Technologies; 1:1000) or rhodamine-labeled phalloidin (Sigma; 1:10,000) in PBS containing 1% BSA and bisBenzimide Hoechst dye (SIGMA) to stain DNA when relevant.
For siRNA screening, images were acquired at low resolution directly in the 24-well plates with an automated ArrayscanVTi microscope (Thermo) equipped with a 10× EC Plan Neofluar 0.3 NA objective, as described previously (Maurin et al., 2018). For confocal imaging, samples were mounted in Citifluor mounting medium (Biovalley) and images were acquired with a SP5 confocal microscope equipped with Leica LAS-AF software with objectives 40× HCX Plan Apo CS 1.3 NA oil or a 63× HCX Plan Apo CS 1.4 NA oil (Leica), or a Zeiss Axioimager Z2 microscope equipped with MetaMorph 7.6.6 Software (Molecular Devices) with objectives 40× EC Plan Neofluar 1.3 NA oil or 63× Plan Apochromat 1.4 NA oil (Zeiss). For Airyscan confocal studies, imaging was performed with a 63× Plan Apo 1.4NA of a Zeiss LSM880 confocal microscope equipped with an Airyscan detector (32 GaAsp detector) in super resolution mode; Alexa 488 was excited at 488 nm with argon laser and Alexa 647 at 633 nm with helium/neon laser. Airyscan analysis was carried out using Zen software with default settings.
Human OC cultures for siRNA treatment and immunofluorescence
For differentiation to human OCs, monocytes purified by CD14+ purification kit (Miltenyi) were seeded on slides in 24-well plates at a density of 5×105 cells per well in RPMI supplemented with 10% FBS (Sigma), M-CSF (50 ng/ml) and RANKL (30 ng/ml); the medium was then replaced every 3 days with medium containing M-CSF (25 ng/ml) and RANKL (100 ng/ml) (Miltenyi) until OC differentiation as described previously (Raynaud-Messina et al., 2018). Targets silencing was performed at day 5 using reverse transfection protocol as previously described (Troegeler et al., 2014). Shortly, cells were transfected with 200 nM of ON-TARGETplus SMARTpool siRNA targeting TUBB6 (Table S11) or the ON-TARGETplus Non-targeting control pool (Dharmacon) using HiPerfect transfection system (Qiagen) in RPMI. Four hours post-transfection, cells were incubated for 24 h in RPMI-1640 medium, 10% FBS, 20 ng/ml of M-CSF and RANKL (100 ng/ml) for 7 additional days. Alternatively, OCs were transferred onto bone slices after 4 days and cultured for another 3 days. Cells were then fixed with 3.7% PFA and 30 mM sucrose in PBS. After permeabilization with 0.3% Triton X-100, cells were saturated with blocking buffer, washed and incubated with Alexa Fluor 555 phalloidin (33 mM, ThermoFisher Scientific) and DAPI (500 ng/ml, Sigma) in blocking buffer for 1 h. Coverslips were mounted with Fluorescence Mounting Medium (Dako). Imaging was performed with a SP5 confocal microscope as described above. For human OC sealing zones, bone slices were imaged with a Leica DM-RB fluorescence microscope or on a FV1000 confocal microscope (Olympus) as described previously (Raynaud-Messina et al., 2018).
Image quantification, OC activity assays, Q-PCR and western blotting
The podosome belt status of OCs was determined considering that the podosome belt was abnormal when, in over half of OC periphery, actin staining was fragmented and/or weak and/or thin and/or absent and/or podosomes were scattered as described previously (Maurin et al., 2018). For automated imaging of the 24-well plates treated with siRNAs, images were acquired with an automated Arrayscan VTi microscope (Thermo) equipped with a 10x EC Plan Neofluar 0.3NA objective. Each well was imaged as a spiral mosaic of ∼300 partially overlapping fields, which were combined in ImageJ 1.51w to obtain an image of the whole well. The measurement of sealing zone size was performed using ImageJ 1.51w software. Total RNA and protein extractions, real time Q-PCR and western blots were performed as reported previously (Maurin et al., 2018) using the primers described in Table S11, primary antibodies against pan β-tubulin (sc-398937 1:1000, Santa Cruz) and Gapdh (2118 1:10,000, Cell Signaling), and horseradish peroxidase-conjugated anti-mouse or anti-rabbit secondary antibodies (respectively NA931V and NA934V, 1:10,000, GE Healthcare). Mineral dissolution activity of OCs was measured as described previously (Maurin et al., 2018). Briefly, at day 3 of differentiation (24 h after siRNA transfection), OCs were detached with Accutase (Sigma) for 5-10 min at 37°C and seeded for 3 days onto inorganic crystalline calcium phosphate (CaP)-coated multiwells (Osteo Assay Surface, Corning), eight wells per siRNA. For each siRNA, four wells were then stained for tartrate resistant acid phosphatase (TRAP) activity to count OCs and four wells stained with Von Kossa to measure CaP dissolution as described previously (Brazier et al., 2009) and imaged with a Nikon SMZ1000 stereomicroscope equipped with a Nikon DXM 1200F CCD camera. Quantification of resorbed areas per wells was carried out with ImageJ 1.51w software. In each experiment, OC-specific activity was expressed as the average area resorbed in the four wells stained with von Kossa normalized by the average number of the OCs in the four wells stained with TRAP. Statistical analyses were performed with GraphPad Prism 5.0. All imaging was performed at the Montpellier Ressources Imagerie (MRI) imaging facility (www.mri.cnrs.fr) except human OC sealing zone images, which were made at the TRI-Genotoul imaging facility (trigenotoul.com).
Acknowledgements
We acknowledge the imaging facility for MRI, a member of the national infrastructure France-BioImaging supported by the French National Research Agency (ANR-10-INBS-04, ‘Investments for the future’). We also thank Dr Anthony Frankfurter (Virginia University, Charlottesville, USA) for his generous sharing of Tubb6 antibodies. We are very grateful to Frank Comunale, Christelle Dantec, Philippe Fort and Peggy Raynaud (CRBM Montpellier, France) for helpful advice and discussions during transcriptomic analysis and for reagent sharing. We thank Oana Vigy (Functional Proteomics, Platform, Montpellier, France) for SILAC data deposition and acknowledge the help of Dominique Helmlinger (CRBM, Montpellier, France) for RNAseq data deposition.
Footnotes
Author contributions
Conceptualization: S.U., A.B.; Methodology: D.G., P.M., A.M., S.U., A.B.; Validation: D.G., P.M., A.M., J.M., C.V., B.R.-M., S.U.; Formal analysis: D.G., P.M., A.M., S.U.; Investigation: D.G., P.M., A.M., J.M., C.V., B.R.-M., S.U.; Resources: C.V., B.R.-M., S.U.; Data curation: P.M., S.U.; Writing - original draft: D.G., P.M., A.M., J.M., C.V., B.R.-M., S.U., A.B.; Writing - review & editing: D.G., P.M., A.M., J.M., C.V., B.R.-M., S.U., A.B.; Visualization: D.G., P.M., A.M., J.M., S.U., A.B.; Supervision: A.B.; Project administration: A.B.; Funding acquisition: A.B. and C.V.
Funding
This study was supported by the Centre National de la Recherche Scientifique (CNRS) and by the Université de Montpellier, and by grants from the Fondation pour la Recherche Médicale (DEQ20160334933), the GEFLUC Languedoc Roussillon (A.P. 2015), the Société Française de Rhumatologie (2676 Subvention 2014), the Fondation ARC pour la Recherche sur le Cancer (PJA 20191209321 to A.B. and PDF2016-1205179 to P.M.), the Agence Nationale de la Recherche France (ANR-16-CE13-0005) to C.V., and the Human Frontier Science Program (HFSP: RGP0035/2016) to the IPBS.
Data availability
The RNAseq data have been deposited to the NCBI Gene Expression Omnibus GEO repository with the dataset identifier GSE149887. SILAC proteomics: The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (Perez-Riverol , 2019) with the dataset identifier PXD018832.
Peer review history
The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.239772.reviewer-comments.pdf.
References
Competing interests
The authors declare no competing or financial interests.