Bone Marrow Macrophages Contribute to Diabetic Stem Cell Mobilopathy by Producing Oncostatin M Free

Diabetes leads to multiorgan pathology, which ultimately reduces life expectancy (1). A series of consistent studies in animals and humans indicate that diabetes affects the structure and function of the bone marrow (BM) (2). Extensive remodeling of the BM microvasculature has been demonstrated in mice (3) and patients (4) with diabetes. Along with autonomic neuropathy (5), such profound alterations in the stem cell (SC) niche cause BM dysfunction, evidenced by an impaired SC mobilization in response to ischemia (6,7) and granulocyte-colony stimulation factor (G-CSF) (8,9). This novel type of chronic diabetes complication, termed “stem cell mobilopathy” (10), has implications for the care of patients with diabetes with hematological disorders. In addition, because the BM is a reservoir of vascular regenerative cells, BM alterations may pave the way to multiorgan damage (2).

On a molecular level, diabetes prevents the chemokine (C-X-C motif) ligand 12 (CXCL12) switch (11,12) (i.e., the suppression of intramarrow levels of the chemokine CXCL12 that normally allows stem cell mobilization) (13). Although a maladaptive response of the CXCL12-cleaving enzyme dipeptidyl peptidase-4 has been hypothesized (14), the exact mechanism perturbing a coordinated CXCL12 regulation in diabetes is unclear. We have previously shown that bypassing BM neuronal control, through sympathetic nervous system–independent stimuli, restores SC mobilization in diabetes (5). Indeed, diabetic mice can be effectively mobilized by the clinical-grade chemokine C-X-C receptor type 4 (CXCR4) antagonist AMD3100/Plerixafor, which desensitizes SCs to CXCL12, thereby letting them leave the BM and reach the systemic circulation (15).

Although the mobilizing activity of G-CSF is partly mediated by the sympathetic nervous system (16), G-CSF signals primarily via a receptor expressed on CD68⁺ macrophages (17), and macrophage suppression is essential to induce SC mobilization (18). Intramarrow macrophages expressing CD169 (Siglec-1) have been shown to secrete a hitherto unknown soluble protein that increases the expression and release of CXCL12 by mesenchymal stem/stromal cells (MSCs), providing a strong retention signal for SCs in the marrow (19). Identification of such a macrophage-derived factor is a primary challenge in this field, because it will eventually turn into a therapeutic target in “poor mobilizer” conditions, such as diabetes. On the basis of observations that hyperglycemia promotes myelopoiesis (20) and that diabetes alters macrophage populations (21) and is associated with a defective CXCL12 switch (11,12), in the study reported here we examined the role of BM macrophages (BMMΦ) in diabetic SC mobilopathy. We found an excess of proinflammatory macrophages in the diabetic BM and that macrophage depletion restores mobilization. We also describe the discovery that oncostatin M (OSM) is the long-sought soluble factor released by macrophages that sustains CXCL12 expression by MSCs. Neutralization of OSM is therefore a candidate therapy to restore SC mobilization and vascular repair in diabetes.

All protocols involving patients were approved by the local ethics committee and done in accordance with the principles of the Declaration of Helsinki as revised in 2008. All subjects provided written informed consent. Patients with type 1 diabetes (T1D) were recruited from the outpatient diabetes clinic of the University Hospital of Padova, whereas subjects without T1D were selected among individuals presenting for a cardiometabolic screening. Details on inclusion and exclusion criteria and clinical characterization of patients are provided in the Supplementary Data. Enrollment in the BM stimulation protocol and treatment with G-CSF in the trial NCT01102699 are described elsewhere (9). Coupled peripheral blood (PB) and BM samples were collected from patients undergoing hip replacement surgery.

All procedures were approved by the local ethics committee and from the Italian Ministry of Health. Experiments were conducted according to the Principles of Laboratory Animal Care (National Institutes of Health publication 85-23.4, revised 1985). All animals were on a C57BL/6 background. Diabetes was induced with one injection of streptozotocin. Additional details are provided in the Supplementary Data.

The following mobilization assays were used: 1) 4-day course of subcutaneous G-CSF; 2) clodronate liposome injection, followed by a course of G-CSF; and 3) antibody-mediated OSM neutralization, followed by a course of G-CSF. Details are provided in the Supplementary Data.

Human circulating monocyte-macrophages were identified and quantified as previously described (21). M1 were defined as CD68⁺CCR2⁺ cells and M2 were defined as CX3CR1⁺CD163⁺/CD206⁺ cells. Baseline and post–G-CSF levels of circulating CD34⁺ SCs were quantified as previously described (9). We used the protocol described by Chow et al. (19) to identify murine BMMΦ phenotypes. Macrophages were identified in the Gr-1⁻/CD115⁻ gate as cells expressing the macrophage marker F4/80 with low side scatter. In parallel experiments, macrophages were also identified as cells that coexpressed F4/80 and MHC-II in the CD45⁺Gr-1⁻ gate. Mouse progenitor cell levels were quantified in PB before and after mobilization: cells were stained with APC-lineage cocktail, phycoerythrin anti–Sca-1, and fluorescein isothiocyanate anti-cKit to quantify Lin⁻/c-Kit⁺/Sca-1⁺ (LKS) cells or with Alexa 647 anti-CD34 and Alexa 488 anti–Flk-1 to quantify endothelial progenitor cells. Additional details are provided in the Supplementary Data.

Human monocyte-derived macrophages were obtained as previously described (21): resting M0 cells were polarized into M1 or M2 macrophages by 48-h incubation with lipopolysaccharide (LPS) and interferon-γ (IFN-γ) or interleukin (IL)-4 and IL-13, respectively. After 48 h, the medium was removed, macrophages were kept in serum-free RPMI for a further 72 h, and then conditioned media (CM) were harvested. Human MSCs were obtained from the BM of patients undergoing orthopedic surgery at the University Hospital of Padova. BM aspirate pellets were plated on TC Petri dishes with mesenchymal medium. Murine MSCs were obtained by from C56Bl6/J mice: femurs and tibia were flushed with ice-cold PBS, and cells were cultured in minimum essential medium-α (MEM-α). Murine macrophages were obtained from BM cells cultured in RPMI 1640 supplemented with macrophage-CSF for 7 days and then polarized for 48 h with LPS and IFN-γ (M1) or IL-4 and IL-13 (M2).

Total RNA was extracted using TRIzol reagent following the manufacturer’s protocol. RNA was reverse transcribed using the First Strand cDNA Synthesis Kit, and duplicates sample cDNA were amplified on the 7900HT Fast Real-Time PCR System. Expression data were normalized to the mean of housekeeping gene ubiquitin C. Additional data can be found in the Supplementary Data.

Human and mouse MSCs at 90% confluence were incubated with cytokines or CM. CM were incubated with anti-OSM antibodies, mouse anti-human, and rabbit anti-mouse (MAB295 and AF-495-NA, respectively; R&D Systems). All experiments were conducted for 48 h, and then cells were lysed with TRIzol for gene expression analysis. Additional details are provided in the Supplementary Data.

In silico analyses were performed on gene expression data of human and mouse macrophages and MSCs retrieved from Gene Expression Omnibus (GEO) database series. Murine and human expression data were divided in two different groups [M(−) or M0 versus M(IFNγ+LPS) or M1 and M(IL4) or M2 versus M(IFNγ+LPS) or M1] and then analyzed using the GEO2R tool. We subsequently filtered data according to these stringent criteria: being upregulated in M(IFNγ+LPS) versus M(–) and versus M(IL4) macrophages at least fivefold (log-fold change >2.32) and with an adjusted P value of <0.001 in both groups. The two groups were then crossed to get a list of genes commonly upregulated in M(IFNγ+LPS) macrophages. To select genes encoding for secreted proteins, data were further filtered by comparing the human and murine gene lists obtained as above with a species-specific secreted protein list from the Metazoa Secretome and Subcellular Proteome Knowledgebase. We then manually checked for proteins with a receptor expressed in MSCs according to relevant GEO series. Technical details on this method can be found in the Supplementary Data.

Data are expressed as mean ± SE or as percentage. Normality was checked using the Kolmogorov-Smirnov test, and nonnormal data were log-transformed before analysis. Comparison between two or more groups was performed using the Student t test and ANOVA for normal variables or the Mann-Whitney U test and Kruskal-Wallis test for nonnormal variables. Linear correlations were checked using the Pearson r coefficient. Statistical analysis was accepted at P < 0.05.

We previously showed that prediabetes and type 2 diabetes are associated with imbalances in circulating monocyte-macrophage phenotypes, reflecting disequilibrium in BM populations (21,22). We herein report that patients with T1D have a significant increase in circulating CD68⁺CCR2⁺ M1-like (proinflammatory) macrophages compared with matched control subjects (clinical characteristics in Table 1). This was attributable to a significant increase in CD68 and, to a lesser extent, CCR2 expression on monocytes of patients with T1D (Fig. 1A and B). Furthermore, in patients with T1D undergoing BM stimulation with G-CSF in the NCT01102699 study (9), there was a strong negative correlation between the degree of CD34⁺ SC mobilization and the change in CX3CR1⁺CD163⁺/CD206⁺ M2-like monocyte-macrophages, which was not observed in control subjects without diabetes (Fig. 1C). These observations prompted us to explore the role of BMMΦ in the diabetic SC mobilopathy.

We first examined the percentages of macrophages in the BM of mice with streptozotocin-induced T1D and in nondiabetic mice. The Gr-1⁻CD115⁻F4/80⁺SSC^low and CD45⁺Gr-1⁻MHC-II⁺F4/80⁺ macrophage phenotypes were both increased more than twofold by T1D (Fig. 2A and B). This increase in BMMΦ appears to be independent from sympathetic innervation and from Sirt1 downregulation, which have been previously shown to mediate diabetic BM dysfunction (5): indeed, chemical sympathectomy with 6-hydroxydopamine or hematopoietic Sirt1 knockout did not affect the percentages of BMMΦ (Supplementary Fig. 1). Gr-1⁻CD115⁻F4/80⁺SSC^low cells were further characterized by FACS for the expression of classical M1 (CD86) and M2 (scavenger receptors CD301 and CD206) macrophage markers. Although CD301 was slightly more expressed on BMMΦ from diabetic mice, there was no obvious prevalence of M1 versus M2 markers (Fig. 2C). Gene expression analysis also did not allow us to unequivocally define Gr-1⁻CD115⁻F4/80⁺SSC^low cells as M1 or M2 (Fig. 2D). BM Gr-1⁻CD115⁻F4/80⁺SSC^low macrophages of diabetic mice showed significantly higher surface and gene expression of CD169 (Fig. 2E), which preferentially labels BMMΦ compared with other cell populations (Supplementary Fig. 2) and identifies macrophages provided with SC-retaining activity in mice and humans (19,23).

In nondiabetic mice, the ability of G-CSF to suppress BMMΦ content is supposed to mediate its mobilizing activity by lifting the inhibitory signal provided by CXCL12 of mesenchymal origin (17,18). We found that although G-CSF reduced BMMΦ by ∼80% in nondiabetic mice, suppression of BMMΦ content in the diabetic mice was <40% and that the percentage of post–G-CSF BMMΦ was more than fivefold higher in diabetic compared with nondiabetic mice (Fig. 3A). We therefore hypothesized that SC mobilization failure in response to G-CSF in diabetic mice is at least partly attributable to excess BMMΦ content. To clarify this point, we depleted BMMΦ using clodronate liposomes, which kill phagocytic cells by delivering toxic intracellular concentrations of clodronate (24). This approach effectively and equally depleted BMMΦ and abated BM CXCL12 gene expression in nondiabetic and diabetic mice (Fig. 3B and C). PB macrophages and Gr-1^high monocytes were also depleted, but neutrophils were unaffected. Among other niche genes, clodronate liposome treatment also reduced Vcam1 expression (Supplementary Fig. 3A and B). As a result, spontaneous and G-CSF induced mobilization of LKSs, and CD34⁺Flk-1⁺ SCs were restored toward normal levels in diabetic mice (Fig. 3D and E). These data confirm that excess BMMΦ contribute to SC mobilopathy in diabetes. However, an unrestricted targeting of BMMΦ with clodronate liposomes is unlikely to be a suitable therapeutic approach to restore BM function because it can have negative off-target effects. We therefore sought to identify the hitherto unknown soluble factor released by macrophages that prevents SC mobilization.

To better understand the meaning of CD169 overexpression in diabetic macrophages, we performed an in silico analysis of the expression of CD169 probes in the public GEO data set GDS2429, reporting gene expression profiles during typical M1 (IFN-γ+LPS) and M2 (IL-4) macrophage polarization from human monocytes (Fig. 4A). CD169 was confirmed as a macrophage marker because its expression markedly increased during monocytes-macrophage differentiation and was much more expressed in M1 than in M0 and M2 macrophages (Fig. 4B). To validate this finding, we obtained human M0, M1, and M2 macrophages using the same in vitro polarization protocol. Polarization efficiency was verified by upregulation of M1 genes Il1b, Tnfa, and iNos and downregulation of Mrc1/CD206 in cells treated with IFN-γ+LPS compared with cells treated with IL-4 (Supplementary Fig. 4). CD169 was more expressed in M1 than in M0 and M2 by FACS and quantitative PCR (Fig. 4C and D). These data indicate that CD169 is a marker of the proinflammatory M1 macrophage phenotype.

The macrophage CM is known to increase the expression and release of the retention chemokine CXCL12 by BM MSCs, thereby preventing SC mobilization (19). We therefore obtained CM from M0, M1, and M2 human macrophages, and cultured human BM-derived MSCs, most of which expressed Nestin (Fig. 4E), a niche-supporting cell marker (25). After incubating Nestin⁺ MSCs with macrophage CM for 48 h, we found that CM from M macrophages, but not from M0 and M2 macrophages, induced CXCL12 expression in MSCs (Fig. 4E). Among other niche genes, we found that M1 CM increased Angpt1 and Kitl expression (Supplementary Fig. 5). This finding, which was consistently reproducible using different batches of CM and different MSC donors, suggests that only proinflammatory macrophages (M1) promote retention versus mobilization of BMSCs through the secretion of a soluble mediator. In further support, by using mouse BM-derived macrophages polarized into M1 and M2 as well as mouse BM MSCs, we confirmed that M1 macrophages express higher CD169 levels and that M1 CM, but not M0 and M2 CM, induces CXCL12 expression (Supplementary Fig. 6A–C).

The effect of M1 CM on CXCL12 expression by MSCs was completely abolished by proteinase K but not by a protease inhibitor (Fig. 4F and G), suggesting that the M1 CM, but not the M0 and M2 CM, contains a protein that signals in MSCs and stimulates CXCL12 expression. To discover this soluble factor, we performed an in silico analysis of the public gene expression profiles of human and mouse polarized macrophages and MSCs (Supplementary Fig. 7A). Genes that were significantly upregulated in M1 compared with M0 and M2 were screened for those encoding secreted factors that had a receptor expressed on BM-derived MSCs. A list of candidate human and mouse genes/proteins was thus retrieved and scored by a literature search, looking at soluble factors produced by inflammatory macrophages that might induce CXCL12 expression by MSCs. Final candidates were validated in vitro by a dose-response stimulation of MSCs (Supplementary Fig. 7B). After unsuccessful testing of several candidates (CXCL10, CXCL11, tumor necrosis factor-α, platelet-derived growth factor-A, IL-15, and endothelin-1; Supplementary Fig. 8), we found that OSM was able to exponentially increase CXCL12 expression by MSCs (Fig. 5A). The other gp130 ligands, namely IL-6 and leukemia inhibitory factor, did not exert the same CXCL12-inducing effect (Supplementary Fig. 9). As measured by ELISA, OSM protein concentrations were markedly higher in the CM of mouse and human M1 than in M0 and M2 (Fig. 5B and Supplementary Fig. 6D), and OSM gene expression was several times upregulated in M1 verus M0 and M2 macrophages (Fig. 5C). Gene expression levels of OSM in the BM were significantly reduced after treatment with clodronate liposomes in diabetic and nondiabetic mice (Fig. 5D).

Thus, we focused on OSM as the most likely candidate SC retention factor produced by BMMΦ. Incubation of human MSCs with CM from human M1 in the presence of a neutralizing anti-human OSM monoclonal antibody completely abolished the effect on CXCL12 expression (Fig. 5E). This confirmed that OSM is the soluble factor contained in M1 CM that stimulates CXCL12 expression by MSCs. CXCL12 induction by OSM was confirmed in other cell types, such as endothelial cells (human umbilical vein endothelial cells and human aortic endothelial cells) and fibroblasts (Supplementary Fig. 10A). Although microvascular permeability affects SC mobilization (26), we did not find any effect of OSM on permeability of a human umbilical vein endothelial cells monolayer (Supplementary Fig. 10B), suggesting that CXCL12 regulation is the most likely candidate mechanism by which OSM regulates mobilization. To explore the molecular mechanisms, we analyzed classical pathways activated by the OSM receptor. Induction of CXCL12 expression by OSM was abolished when MSCs were cotreated with inhibitors of mitogen-activated protein kinase kinase (MEK) (U0126), p38 (SB202190), and signal transducer and activator of transcription 3 (STAT3) (Stattic) (Supplementary Fig. 11A). As shown by FACS, STAT3 phosphorylation by OSM was reduced by cotreatment with the p38 inhibitor (Supplementary Fig. 11B), suggesting a MEK-p38-STAT3 pathway. However, simple p38 or STAT3 activation was insufficient to induce CXCL12 in MSCs (Supplementary Fig. 11C), suggesting that recruitment of other cofactors by OSM signaling is required.

For an in vivo validation of the role of OSM as a retention factor, we treated T1D mice and nondiabetic mice with a neutralizing anti-mouse OSM antibody the day before G-CSF stimulation was started. This protocol abated circulating OSM concentrations and restored a significant CXCL12 gradient switch by G-CSF in diabetic mice by raising its PB-to-BM concentration ratio (Supplementary Fig. 12). Although the diabetic mice failed to mobilize LKS and CD34⁺Flk-1⁺ cells in response to G-CSF alone, OSM neutralization was able to restore G-CSF–induced LKS and CD34⁺Flk-1⁺ cell mobilization in diabetic mice to the level seen in nondiabetic mice and beyond (Fig. 5F–I). In view of clinical translation, we examined OSM concentrations in relation to PB and BM CD34⁺ cell distribution in six individuals with diabetes and six matched individuals without diabetes (clinical characteristics in Supplementary Table 2). The BM plasma OSM concentration and the BM-to-PB CD34⁺ cell ratio were higher in the patients with diabetes, and a close direct correlation between these parameters was found, supporting the notion that OSM regulates BM-to-PB SC mobilization (Fig. 5J–L).

In diabetes, G-CSF and ischemia-induced mobilization are both defective. We therefore sought to verify whether OSM inhibition affects the response to ischemia in diabetic mice undergoing hind limb ischemia with or without treatment with the neutralizing anti-OSM antibody. By FACS analysis, we found that OSM neutralization increased the number of circulating LKS cells after ischemia and restored the number of LKS cells homed to the ischemic muscles toward levels seen in nondiabetic mice (Fig. 6A and B). In addition, OSM neutralization restored hind limb perfusion 14 days after ischemia, as shown by laser Doppler imaging (Fig. 6C).

In this study, we show that excess proinflammatory macrophages in the diabetic BM provide a retention signal for SCs by secreting OSM and inducing CXCL12 in MSCs. Macrophage depletion and OSM neutralization were indeed able to restore SC mobilization toward normal levels in diabetic mice, which translated into improved vascular recovery after ischemia.

Diabetes induces a profound remodeling in the BMSC niche in mice and humans (3–5,27), which impairs SC mobilization in response to ischemia and growth factors (6–9). Multiple molecular pathways may mediate this dysfunction, but the exact mechanisms are unknown, thus limiting the possibility to pursue targeted therapies. Based on the notion that diabetes affects the monocyte-macrophage compartment (21,22,28) and that macrophages prevent SC mobilization (19), we assessed whether BMMΦ contribute to diabetic SC mobilopathy. Using standardized FACS protocols (19), we found two- to threefold increased levels of macrophages in the diabetic BM that were not adequately suppressed by G-CSF. This is in line with the previous finding that hyperglycemia skews SC differentiation to myeloid phenotypes through advanced glycosylation end product/receptor for advanced glycosylation end products interactions (20). Chow et al. (19) clearly demonstrated that macrophages equipped with SC retention activity are labeled by the adhesion molecule CD169 because selective depletion of CD169⁺ cells allows SC mobilization. Macrophages in the diabetic BM displayed increased expression of CD169, and we therefore hypothesized that such an excess in BMMΦ blocks SC mobilization in diabetes. To test this hypothesis, we performed macrophage depletion using clodronate liposomes, which selectively kill specialized phagocytic cells in the reticuloendothelial system, including the BM. Effective suppression of intramarrow macrophages was followed by the release of SCs in diabetic mice and restoration of response to G-CSF toward normal levels.

These data suggest that macrophage targeting can therapeutically restore BM function in diabetes. We therefore focused on the hitherto unrecognized signal(s) whereby BMMΦ affect function of the SC niche (29). By an in silico approach, we discovered that OSM is the long-sought soluble factor released by macrophages that induces the retention signal CXCL12 in MSCs, thereby preventing SC mobilization. The process that led us to this discovery was primed by the finding that the adhesion molecule CD169, which labels macrophages provided with retention activity, is markedly upregulated in proinflammatory, so-called classically activated M1 or M(LPS+IFNγ) macrophages, compared with resting M0 and M2 or M(IL-4) macrophages. The segregation of CD169 expression and CM activity enabled us to mine the widely available gene expression profiles of human and mouse M0, M1, and M2 cells in search of candidate secreted factors that signal through receptors expressed on MSCs. Potential candidates, scored by literature searches, were validated by a simple and rapid in vitro assay of CXCL12 induction. Only OSM, but not the other gp130 ligands LIF and IL-6, fulfilled all such requisites, and OSM blockade completely abolished the effects of M1 CM on CXCL12 expression in MSCs. A preliminary study of signaling pathways identified the MEK-p38-STAT3 axis as likely mediating the effect of OSM on CXCL12 expression. Upon an accurate literature review, it appeared that OSM was already known to be induced in classically activated M1 macrophages (30), able to stimulate CXCL12 production by MSCs (31), and possibly involved in SC retention within the BM (32). In addition, gp130 ligands have been shown to be upregulated in the BM of long-standing murine diabetes, whereas genetic deletion of gp130 reverses some pathologic hematopoietic features associated with diabetes (33). Similarly to what we show in the BM, macrophage-derived OSM seems to play a role also in adipose tissue inflammation (34,35), where SCs, the microvasculature, macrophages, and adipocytes form a structure resembling the BM niche.

In further support of the role of OSM in macrophage-mediated regulation of SC trafficking, we found that BM OSM expression was significantly suppressed after macrophage depletion. In view of clinical translation, we also found that OSM concentrations were higher in the BM plasma of patients with diabetes compared with patients without diabetes and were correlated with a surrogate index of steady-state mobilization, such as the ratio of CD34⁺ SCs between the BM and PB. As a final proof of concept, we show that in vivo OSM neutralization restored the SC mobilization response to G-CSF in diabetic mice. Ischemia-induced mobilization and homing of LKS cells was also improved after treating diabetic mice with the OSM neutralizing antibody, which was associated with restoration of perfusion. This indicates that any eventual peripheral effect of OSM blocking did not prevent SC from homing to ischemic tissues.

These findings have relevant implications for our knowledge of how macrophages regulate the niche and add an important piece to the complicated puzzle of the diabetic BM pathology. The pathological pathway generated by excess BMMΦ seems independent from other typical features of the diabetic BM, such as neuropathy, oxidative stress, and Sirt1 downregulation (5). However, in view of the severe BM remodeling taking place in diabetes, it is not surprising that both macrophage depletion and OSM neutralization, although effective in restoring response to G-CSF, often elicited a blunted mobilization in diabetic compared with nondiabetic mice.

On the background of the well-known hyperglycemia-driven myelopoiesis (20), our data indicate that the generation of proinflammatory macrophages represents a key event in BM dysfunction. Identification of OSM as the mediator of macrophage-retaining activity has therapeutic implications to revert diabetic SC mobilopathy and, potentially, in other “poor mobilizer” conditions. Because BM-derived cells play a major role in diabetes complications (36), restoration of BMSC mobilization with a targeted molecular approach may restore endogenous vascular regenerative capacity and improve the outcome of patients with diabetes.

Funding. This study was supported by grant GR-2010-2301676 of the Italian Ministry of Health to G.P.F. and by a European Foundation for the Study of Diabetes (EFSD)/Novartis program grant to G.P.F. M.A. is supported by the Italian Society of Diabetology.

Duality of Interest. M.A., S.C., and G.P.F. are the inventors of a patent pending, held by the University of Padova, on the use of OSM inhibition for the induction of stem cell mobilization in diabetes. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.A., N.P., S.C., R.C., L.M., and F.F. researched data. C.B. and A.C. researched data and contributed to discussion. A.A. contributed to discussion and reviewed and edited the manuscript. G.P.F. researched data and wrote the manuscript. G.P.F. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.