Samples were obtained in agreement to the French and German ethical laws. MMC were purified from the BM of 131 patients with newly-diagnosed MM (median age, 59 years) after written informed consent was given. The study has been approved by the ethic boards of Heidelberg University and Montpellier University hospitals. According to Durie-Salmon classification, 14 patients were of stage IA, 24 of stage IIA, 76 of stage IIIA, 14 of stage IIIB, one had a plasma cell leukaemia and 2 were of undetermined stage. Human normal bone marrow samples were obtained from filtration residues of the bone marrow harvested from healthy donors for stem cell allograft after agreement of the Center of Biological Resources of the Montpellier University Hospital. Buffy coats of peripheral blood cells were purchased from the French Blood Center according to a written approved convention between French Blood Center and Montpellier University Hospital. Normal BM plasma cell (BMPC) and primary MMC were purified using using autoMACS with anti-CD138 MACS microbeads (Miltenyi-Biotec, Paris, France) as previously described 9 . For the isolation of peripheral blood memory B cells (MB), monocytes, NK and T cells were first removed using anti-CD14, anti-CD16 and anti-CD3 magnetic beads (Dynal), and MB cells were then positively selected using anti-CD27 MACS microbeads (Miltenyi Biotec). Polyclonal plasmablasts (PPC) were generated from purified peripheral blood CD19⁺ B cells in vitro, as described 10 11 . Whole BM samples (WBM) were obtained from 75 of the 131 patients with MM whose MMC were purified. According to the Durie-Salmon classification, 12 patients were in stage IA, 15 in stage IIA, 45 in stage IIIA, three in stage IIIB. In addition, BM T lymphocytes (CD3+), monocytes (CD14+) and polymorphonuclear neutrophils (PMN, CD15+) were simultaneously purified from the BM of five of those 75 patients, as described previously 9 . Stromal cells (BMSC, n = 5) and osteoclasts (n = 7), which cannot be harvested in BM aspirates, were generated in vitro from the BM of MM patients, as described 9 .

Microarray experiments were performed at the Institute of Research in Biotherapy http://irb.chu-montpellier.fr/en/laboratories_microarray.html in the Montpellier University Hospital. RNA was extracted with the RNeasy Kit (Qiagen) or the SV-total RNA extraction kit (Promega) and Trizol (Invitrogen), in accordance with the manufacturer's instructions. Biotinylated complementary RNA (cRNA) was amplified with a double in-vitro transcription, according to the Affymetrix small sample labelling protocol vII (Affymetrix). The biotinylated cRNA was fragmented and hybridized to the HG-U133 Plus 2.0 GeneChip oligonucleotide arrays according to manufacturer's instructions (Affymetrix). Fluorescence intensities were quantified and analyzed using the GCOS software (Affymetrix). Arrays were scaled to an average intensity of 100. A threshold of 1 was assigned to values under 1. Gene expression data were analysed with our bioinformatics platforms: RAGE, http://rage.montp.inserm.fr 12 and "Amazonia!", http://amazonia.montp.inserm.fr 13 .

All microarray data presented in this paper have been deposited in the ArrayExpress public database, under accession numbers E-MEXP-2360 for BMPC 11 and E-TABM-937 for B cells, PPC, MMC and environment population samples. We also used Affymetrix data of 22 normal BMPC samples and a cohort of 345 purified MMC from previously-untreated patients from the Arkansas Research Group (Little Rock) 14 15 . These data are publicly available in the NIH Gene Expression Omnibus (GEO) http://www.ncbi.nlm.nih.gov/geo/ under accession number GSE2658 for MMC samples and GSE5900 for 22 normal BMPC.

For supervised analyses, differentially expressed genes were identified by a student t-test, and p-values were adjusted for multiple comparisons using the Benjamini and Hochberg correction. The threshold for significance was set to a p value ≤ 0.05. Among those genes, those with a fold change ≥ 2 were retained. Among the genes found to be statistically overexpressed with a fold change ≥ 2 in a given population, those with 100% absent call in this population were considered not to be biologically relevant and were removed. The call ("present" or "absent") is determined by Affymetrix GCOS-software and indicates whether a gene is reliably expressed or not. Regarding analysis on the different environment populations, whole bone marrow samples (WBM) were not included for supervised analysis. They include both MMC and environment cells and the expression of a given gene could be the reflection of its expression in MMC or in the environment or both.

Based on the literature data, we have listed 32 growth factors that have already been described as growth factors for MMC (Table 1A). As VEGFs, Wnts and FGFs are members of families with strong homologies, we extended the study to all members of those families, even if only some of them have already been shown to play a role in MM (Table 1B). Gene expression of 24/32 known MGF, 3/4 VEGFs, 13/22 FGFs and 11/19 Wnts could be interrogated with Affymetrix U133 Plus 2.0 arrays (Tables 1A and 1B). Other genes could not be studied due to a lack of probe set (one gene: Wnt3A) or to a "non-informative" probe set, i.e. with a low signal and an absent call in ≥ 95% of the samples (20 genes, see additional file 1). Thus 51 growth factors could be interrogated by 147 probe sets. When several probe sets were available for a given gene, the probe set with the highest percentage of "present call" among all samples or the one with the highest variance in case of 100% presence was selected. Probe sets are listed in Tables 1A and 1B.

In order to identify which MGF are expressed by each cell component of the BM, we compared the expression of the 51 MGF genes among all BM subpopulations. 3/51 MGF genes - FGF7, NRG3, Wnt10A - were statistically significantly overexpressed in purified myeloma cells (MMC, n = 131) compared to each other BM population: T cells (CD3, n = 5), monocytes (CD14, n = 5), CD15+ polymorphonuclear neutrophils (PMN, n = 5), bone marrow stromal cells (BMSC, n = 5), and osteoclasts (n = 7) (p ≤ .05, fold-change ≥ 2). They were defined as "myeloma MGF" (Figure 1 and additional file 2, table S-I). 22/51 genes were overexpressed in cells from the environment compared MMC (p ≤ .05, fold-change ≥ 2). 14 out of those 22 MGF genes were overexpressed in cells of the putative tumor niche, i.e BMSC (11 genes) and osteoclasts (3 genes), compared to MMC and other BM subpopulations (CD3, CD14 and PMN). They were defined as "myeloma niche MGF" (Figure 1 and additional file 2, table S-II). 8/22 genes were overexpressed in at least one BM subpopulation (CD3, CD14 and/or PMN) compared to MMC, BMSC and osteoclasts (p ≤ .05, fold-change ≥ 2). They were defined as "environment MGF" (Figure 1 and additional file 2, table S-III). 17/51 genes were not classified by the supervised analysis. For the majority of them (12 genes), the reason is that they were highly expressed both in MMC and in at least one environment population (Figure 1 and additional file 2, table S-IV). The five remaining genes (NRG2, Wnt4, Wnt11, Wnt16 and FGF18) were considered as a "myeloma MGF" although they were not statistically significantly overexpressed in MMC compared to environment cell populations. Indeed, they displayed a present call in some of the MMC samples whereas they were absent in all other BM populations (see additional file 2, table S-I). Finally, 9/51 MGF were not/weakly expressed, neither in MMC nor in the environment. They had less that 5% of presence in MMC samples and BM subpopulations and they were not expressed in BMSC and osteoclasts (data not shown). Those genes include 5 members of the FGF family (FGF3,4,6,11,14), 3 members of the Wnt family (Wnt2A,7A,8A) and NRG4. This lack of expression could not be attributed to a lack of detection by a non-functional Affymetrix probe set, as expression of these genes could be detected using Affymetrix microarrays in various normal tissues (data not shown, see additional file 1). Altogether, those data indicate that the majority of MGF are overexpressed by the environment compared to MMC, and suggest that BMSC are the main source of MGF.

The 3 "myeloma MGF" as well as FGF18, NRG2, Wnt4, Wnt11 and Wnt16 (100% absent call in all environment populations) are represented in Figure 2. For clarity of the presentation, only 44/131 MMC samples as well as 44 whole BM samples (including MMC and cells of the environment; WBM), both obtained from the same patients, are depicted here. "Myeloma MGF" include 2 members of the EGF family (NRG2 and NRG3), 2 members of the FGF family (FGF7 and FGF18) and 4 members of the Wnt family (Wnt4, Wnt10A, Wnt11 and Wnt16). Except for FGF7, those 8 MGF displayed 100% absent call in all other BM cell populations (table S-I). Hence, the expression detected in the WBM samples is due to the MMC (Figure 2, right panel) and those factors are exclusively autocrine for the MMC. Three of these 8 genes - FGF7, Wnt4 and Wnt10A - were also expressed in normal plasma cells (Figure 2).

The 14 "myeloma niche" MGF are represented in Figure 3. 11/14 genes were overexpressed in BMSC compared to all other BM populations (p ≤ .05, fold-change ≥ 2): BNDF, 3 members of the FGF family (FGF1, FGF2, FGF5), GDF15, IL-6, JAG1, LIF, VEGFC and 2 members of the Wnt family (Wnt3 and Wnt5B) (Figure 3A). 3/14 genes were overexpressed in osteoclasts compared to all other BM populations (p ≤ .05, fold-change ≥ 2): IGF1, IL10 and April (Figure 3B). IGF-1 and IL-6 have been extensively described as paracrine myeloma growth factors 5 6 . Accordingly, they belong to the "myeloma niche" category. Interestingly, IL-6 and IGF-1 were also expressed by the MMC themselves, in 60% and 100% of the patients, respectively (% of present call, table S-II), but at a lower level than in BMSC for IL-6 (median IL-6 expression in MMC = 55, versus 1867 in BMSC, table S-II) and in osteoclasts for IGF-1 (median IGF-1 expression in MMC = 424, versus 1104 in osteoclasts, table S-II).

The 8 "environment" MGF are represented in figure 4A and listed in additional file 2, Table S-III. CD14-positive monocytes overexpressed 6/8 "environment" MGF (p ≤ .05, fold-change ≥ 2): 2 TNF-superfamilly members (TNFSF13B/BAFF and TNF-α), OSM, IL-1β, IL15 and HB-EGF. CD15-positive PMN also overexpressed BAFF and OSM, as well as FGF13, compared to MMC, BMSC and osteoclasts. Thus, monocytes and PMN may be an important source of growth factors for the tumor, especially because PMN represent around 50% of the cells in the BM 16 . CD3+ T cells overexpressed FGF9 (Table S-III).

12/51 genes were not classified by the supervised analysis (Table S-IV). For seven of them, the reason is that they were highly expressed both in MMC and in at least one environment cell population: AREG, Wnt5A, CCL3, CLCF1, VEGF, FGF12 and HGF (Figure 4B and table S-IV). AREG was expressed in MMC and WBM samples, as well as in CD3, CD14 and PMN subpopulations, but it was weakly or not expressed by BMSC and osteoclasts. In contrast, Wnt5A was not expressed in WBM samples and in the subpopulations, but it was expressed in osteoclasts and BMSC. It should be noted that HGF expression in MMC (median = 218) was not significantly higher that that in BMSC (median = 194, p = 0.4), CD14 (median = 99, p = 0.1) and PMN (median = 102, p = 0.1). However, HGF expression in MMC among 131 patients was very heterogeneous and it could be very high in some patients' MMC (range 1 to 5378). Other 5 genes (PTN, NRG1, JAG2, VEGFB and Wnt6) were also expressed in MMC and environment cells, but without any change in signal between present or absent call (data not shown). This may suggest that those probe sets poorly work. It may also be that the expression level is at the limit of sensitivity and thus can hardly be detected by Affymetrix probe sets. This is likely the case for NRG1 whose gene expression measured by real-time PCR has been already documented in MM 17 .

Because the aim of this study was to describe the expression of growth factors that sustain myeloma cell growth through paracrine or autocrine stimulation, we focused on the analysis of receptor expression in MMC only. However, we cannot exclude that MGF receptor expression by cells from the environment (data not shown) could indirectly influence myeloma cell growth as well. Among 48 MGF receptors that are listed in Table 2, 36 could be interrogated with Affymetrix microarrays. Half of the MGF receptor genes (18/36) were expressed in more than 50% of the MMC samples (present call, Table 2). Among them, 4 MGF receptor genes were expressed in 100% of the patients (IL-6ST, IL-10 RB, BCMA or Notch 2), 1 MGF receptor gene was expressed in 99% of them (IL-6R) and 5 MGF receptor genes were expressed in more than 90% of the patients (IL-10RA, TACI and Frz receptors 4,7 and 9). This indicates that the vast majority of the patients, if not all, have receptors for IL-6, IL-10, Baff/April and Wnt. 14/36 MGF receptor genes were expressed in 10-50% of the patients and 4/36 of them were expressed in less than 10% of the patients (table 2). CLCF1R was absent is 100% of the MMC samples and IL1R in 99% of them (in agreement with previous data 18 ), suggesting that their respective ligands produced by the environment (see figure 4) have no biological relevance to MMC.

In order to investigate the expression of MGF and MGF receptors throughout normal PC differentiation and in the generation of malignant plasma cells, we compared the expression of the 51 MGF and 36 MGF receptor genes in memory B cells (MB, n = 6), in in vitro-generated plasmablasts (PPC, n = 7) in purified bone marrow plasma cells (BMPC, n = 7), and MMC (n = 131). Data are shown in additional file 3, figures S-I and S-II and additional file 2, tables S-V and S-VI. The expression of 8 MGF and 10 MGF receptor genes was statistically significantly different between B cells and "PPC+BMPC" (p ≤ .05, fold-change ≥ 2). 4 MGF genes (OSM, Wnt6, Wnt16 and FGF9) were up-regulated in B cells, whereas others 4 (FGF7, CCL3, Wnt5B and IGF1) were up-regulated in "PPC+BMPC". 1 MGF receptor gene (IL10RA) was up-regulated in B cells, whereas other 9 genes were up-regulated in "PPC+BMPC" (p ≤ .05, fold-change ≥ 2): 3 FRZ-family members (FRZ1,3,6), the 2 members of the IL6R complex (IL-6R and IL-6ST), IL15RA, Met, TNFRSF17 and IL21R.

Interestingly, the expression of 12 MGF and 12 receptors was significantly different between PPC and BMPC, 11/12 MGF and 11/12 MGF receptor genes being up-regulated in BMPC compared to PPC (p ≤ .05, fold-change ≥ 2; figures S-I and S-II). Those 11 MGF genes include 2 Wnt-family members (Wnt5A, Wnt10A), 2 VEGF-family members (VEGFB and C), 1 FGF-family member (FGF2), 1 EGF-family members (AREG), 2 TNF-family members (Baff, April), PTN, CCL3 and JAG1. MGF receptor genes comprise 5 members of the FRZ family (FRZ1,2,6,7,8), IL6ST, NTRK2, Notch4, TNFRSF1A, CCR1 and LIFR. IL21R was upregulated in PPC. These data emphasize an increased expression of genes encoding for MGF and their receptors throughout plasma cell differentiation, in particular during late plasma cell stage, i.e in BMPC. It is noteworthy that FRZ receptor expression is a hallmark of PC differentiation.

We then compared the expression of the 51 MGF and 36 receptor genes in MMC and in BMPC. Only 3 MGF and 1 MGF receptor gene (FZD2) were significantly differentially expressed between MMC and BMPC (p ≤ .05, fold-change ≥ 2). FZD2 was overexpressed in BMPC, whereas all three MGF genes were overexpressed in MMC: AREG, NRG3 and Wnt5A. AREG and WNT5A were also upregulated in BMPC compared to PPC. In contrast, NRG3 was not expressed in any normal population (100% absent call) but it was expressed in 73% of the patients'samples with a very strong signal in some of them (see Figure 2). Those 3 MGF genes were also found to be overexpressed in MMC versus BMPC (p ≤ .05, fold-change ≥ 2) in an independent publicly available data set of 22 BMPC and 345 MMC patients' samples from the Arkansas Research Group (Little Rock, additional file 2, table SVII-A). Of note, 10/51 MGF and 5/36 receptor genes were "absent" in all BMPC samples (data not shown) whereas they were expressed in more than 10% (present call) of the MMC samples (see additional file 2). Hence, they may also be considered as being aberrantly expressed in MMC compared to normal BMPC although they are not statistically significantly overexpressed in MMC. Those MGF genes include HGF, 2 EGF-family members (HB-EGF and NRG2), 3 Wnt-family members (Wnt5B, Wnt11, Wnt16), 2 FGF-family members (FGF9, FGF18), BDNF and TNF-α (data not shown). MGF receptor genes include FGFR3, IL11R, IGF1R, FRZ4 and LRP6. Additional normal samples would be required to conclude about their relative expression in normal versus tumor cells. Among them, one MGF (HGF) and one MGF receptor (IL-11R) genes were statistically significantly overexpressed in MMC versus BMPC in the publicly available data set of 22 BMPC and 345 MMC patients' samples (p ≤ .05, fold-change ≥ 2; additional file 2, table S-VII).