All stromal cells [MS5, C3H10T1/2, MC3T3-E1-4, MC3T3-E4-24 bone marrow stromal lines, AFT024 and BFC012 stromal lines from fetal liver (41) or embryonic stem cell-derived fibroblasts as in (42, 43)] were cultured in complete medium consisting of Minimum Essential Media alpha (MEMα) with GlutaMax™ (Gibco, catalog no. 32561) and 1% Penicillin/Streptomycin (P/S, Gibco, catalog no. 15140) supplemented 10% fetal bovine serum (FBS, Gibco, catalog no. 10270-106) at 37°C and 5% CO2 with media changed every 2-3 days, and differentiated as described below (44). For subsequent analysis, OP9 cells were plated in 24-well plates (Falcon) with CaF2 substrates (Crystran) for Raman microspectroscopy and RT-qPCR, or in 6-well plates (Falcon) for HPLC analysis, and in flat-bottom tissue culture-treated 96-well plates (Falcon) for co-culture assays. OP9 cells were either plated subconfluently at a density of 5,000 cells per cm2 (undifferentiated OP9 cells, uOP9) or plated at confluency of 20,000 cells per cm2 in complete medium (spontaneous OP9 cells, sOP9) for either 7 or 17 days, as specified in figure legends. Both OP9 cells and OP9-EGFP cells (16) were authenticated by the ATCC on June 24, 2021, by short tandem repeat (STR) analysis, indicating an exact match of the ATCC reference OP9 cell line (CRL-2749). OP9 cells are also directly available from the ATCC (CRL-2749), but this source has not been tested for adipogenesis or hematopoietic support in our laboratory.

Fixed 24-well plates were kept in PBS at 4°C and shipped to the University of Lille. Raman acquisitions were done on a LabRAM HR800 equipped with an immersion objective (Nikon, obj x100, numerical aperture = 1, Japan) and a diode laser λ=785 nm. The laser power at the sample was 30mW. The lateral resolution was 1-2µm. Spectral acquisition was made in the 400–1800cm^-1range and spectral resolution was 4cm⁻¹. The acquisition time was set at 60s per spectrum. Raman spectra were processed using Labspec software (HORIBA, Jobin-Yvon, France). The water immersion objective focused the laser on the center of individual lipid droplets where one spectrum corresponds to one adipocyte lipid droplet. In total 2944 spectra were measured over 120 sOP9-adipocytes (on average 24 spectra per adipocyte) and 2971 spectra over 138 iOP9-adipocytes (with an average of 21 spectra per adipocyte). The number of spectra per well was between 60 to 110. The unsaturation ratio was calculated as ratio of area under the curve of bands 1654cm^-1/1441cm^-1. The optical image of the adipocyte was saved for each acquisition. The diameter of LDs was evaluated from the optical image by using the software FIJI (45).

The instrument is also equipped with a XYZ motorized stage which allows the acquisition of Raman images. Raman images were acquired using the point-by-point imaging mode (Pt-Img). The laser beam was focused perpendicular to the sample surface. Acquisition time was set to 3sec (×2) for each spectrum. The laser beam was stepped in two dimensions (x and y), and a spectrum was recorded at each position (x,y). The step was set to 1 µm between 2 positions. The Pt-Img mode generated x×y spectra.

Cells were fixed with PFA as described above for adipocytic quantification. After washing, 500µl of PBS was left in the wells, plates were sealed with parafilm and stored at 4°C until processing for HPLC analysis. Cells were detached from the plate with 1mL (2 x 0.5ml) of PBS and lipids extracted with chloroform/methanol (2:1; v/v) under agitation for 30min at room temperature. The resulting lipid extract was then subjected to saponification, followed by fatty acid derivatization into naphthacyl esters as described previously (46). Fatty acid derivatives were applied into the HPLC Alliance system (2695 Separations Module, Waters, Saint-Quentin-en-Yvelines, France) equipped of an autosampler (200ml-loop sample), a photodiode array detector (model 2998), and the Empower software for data analyses. Fatty acid derivatives were eluted on a reverse phase YMC PRO C18 column (3mm, 4.6x150mm, 120Å) by using two solvent systems: a) methanol/acetonitrile/water (64:24:12; v/v/v) and, b) methanol/dichloromethane/water (65:28:7; v/v/v) under identical conditions reported previously (46). Fatty acid derivatives were detected at 246 nm and quantified by an external standard curve realized with the naphthacyl 19:0 derivative. Note that 19:0 was also used as internal standard to evaluate FA recoveries. Lipidomics LC-HRMS standards and solvents are provided in Table S1 .

At day 7 or 17 of adipocytic differentiation, RNA was extracted with Trizol and cDNA was synthesized using Taq DNA polymerase (catalog no. 10342020, Life Technologies) according to manufacturer’s instructions. Quantitative PCR was performed in technical triplicates using the Fast SYBR Green qPCR Mastermix with 250 μM of primer concentration (Applied Biosystems) on QuantStudio 6 (Life Technologies). Primers were pre-validated for high efficiency and housekeeping gene primers were internally tested not to vary upon OP9 differentiation. The geometric mean of housekeeping genes (RPL13 and YWHAZ) was used as reference to calculate the fold expression of each gene relative to the samples of undifferentiated OP9 cells as described for the delta-delta normalization method (47, 48). Primer sequences obtained from Microsynth are listed in Supplementary Table S2 .

Total bone marrow was extracted from eight-week-old B6 ACTb-EGFP females housed in 12-hour day-night light cycles and provided ad-libitum sterile food and water as described previously, in accordance to Swiss law and ARRIVE guidelines, with the approval of cantonal authorities (Service Veterinaire de l’Etat de Vaud). Total bone marrow cells were extracted from femur, tibia and pelvis by crushing using a mortar and pestle in ice-cold PBS supplemented 1mM ethylenediaminetetraacetic acid (EDTA, catalog no. 15575020, Thermo Fisher Scientific). The samples were dissociated and filtered through a 70µM cell strainer (catalog no. 352350, Falcon), lysed for 30sec at room temperature in red blood cell lysis buffer (Biolegend, catalog no. 420301), washed with ice-cold PBS-EDTA and centrifuged at 1300rpm for 10min at 4°C. For HSPC sorting prior to co-culture with OP9 stroma, the cell pellet was stained with 50µl biotinylated ‘lineage’ antibody cocktail (BD, catalog no. 558451) in 1ml PBS-EDTA per six bones for 15min on ice. The sample was washed with PBS-EDTA and stained with 50µl magnetic beads of the same kit in PBS-EDTA for 10min on ice. The samples were then washed and filtered through a 70µm cell strainer (catalog no. 352350, Falcon) prior to lineage depletion using the AutoMACS Pro (Miltenyi Biotec, USA). After depletion the negative fraction was resuspended in an antibody mix containing antibodies against Streptavidin-TxRed (1:200), cKit-PECy7 (1:200), Sca1-APC (1:100) and PI (1:1000). Cell sorting was performed on a FACSAria Fusion (Becton Dickinson, USA) cell sorter.

OP9 cell seeding as detailed in Section 2.1, was followed by short-term in vitro adipocyte differentiation of OP9 cells, and then a gentle wash with PBS and pre-warmed Iscove’s Modified Dulbecco’s Medium (IMDM, catalog no. 12440053, Gibco) supplemented with 10% FBS and 1% Pen/Strep (P/S) was added. The washing was done using the Caliper Sciclone ALH 3000 (Caliper Life Sciences, USA) and consisted of four cycles of removing 140µl media and adding 140µl fresh IMDM media, to dilute the previous media so as to keep mature adipocyte detachment to a minimum. The wells were left with 100µl IMDM 10% FBS and 1% P/S. HSPCs extracted from B6 ACTb-EGFP mice were plated at a ratio of 1:10 initial OP9 cells (in volume of 100µl IMDM, totaling 200µl volume per well).

After seven days of co-culture, the 96-well plates were removed from the incubator and placed on ice. To count non-adherent cells, the cells in suspension were retrieved through repeated washes and stained with CD45-PacBlue at final concentration of 1:200, PI at 1:1000 and 5µl CountBright beads (catalog no. C36950, Invitrogen) directly added to the media. Without any further manipulation, the plates were analyzed using the High Throughput Sampler (HTS) module of a LSRII (Becton Dickinson, USA) flow cytometer. For accuracy, the settings were programmed to mix the wells thoroughly before sample uptake. To count adherent cells, a separate set of duplicate plates were taken from the incubator, the media manually removed, and wells washed once with 200µl PBS. 40µl of Trypsin EDTA (0.5%, catalog no. 25300-054, Gibco) was added to each well and incubated for five minutes at 37°C. 160µl of ice-cold PBS containing FBS (to neutralize the trypsin), CD45-APCCy7 (1:200), PI (1:1000) and 5µl CountBright beads were then added to the plates. The plates were stained on ice and directly analyzed by flow cytometry.

This study complied with the Declaration of Helsinki and the local ethical authorities (CER-VD). All patients signed a specific consent for the reuse of biological samples in the context of our study. For HPLC, iliac crest BM aspirates from female and male patients undergoing treatment for acute myeloid leukemia (between ages 48 and 73, n=5) or surgical debris from patients undergoing hip replacement surgery (between ages 56 and 80, n=4) were received for analysis from the Lausanne University Hospital (Centre Hospitalier Universitaire Vaudois, CHUV). Floating adipose tissue, bone marrow serum, and oil fractions were isolated and processed for lipid extraction and processed as described for OP9 lipidomics.

Values are shown as mean plus or minus the standard deviation or standard error of the mean as indicated. Student’s t-test was performed for all experiments when comparing two conditions only, or a Two-Way ANOVA when comparing multiple conditions, with P-values indicated for statistical significance.

Raman spectra were processed by using the PLS Toolbox and the MIA toolbox (v8.7 eigenvector Research, Inc., USA). The spectra were baseline corrected, normalized and mean centered. Two processes were done on Raman spectra. First, cluster analysis (CA) was performed using Ward’s method. CA clustered individual Raman spectra into 3 categories (saturated-rich lipids, unsaturated-rich and mixture of saturated and unsaturated lipids). Second, principal component analysis (PCA) was used which highlighted the difference in spectral features between the spontaneous and induced conditions. All data are reported as mean ± standard deviation. All statistical analyses were performed using GraphPad Prism (version 8.0.0, GraphPad Software). Mean values were compared by Mann and Whitney test, or two-way analysis of variance (ANOVA) followed by Bonferroni’s post hoc test. Statistical significance was accepted for P<0.05, and reported.

We chose murine BM-derived OP9 stromal cells as a model for this study because of their propensity to differentiate into adipocytes while displaying both hematopoietic support and dual adipogenic and osteogenic differentiation capacities as compared to other commonly used murine stromal lines including BM-derived MS5 and C3H10T1/2 lines or MC3T3 subclones ( Figures S2A-C and [44)]. Note retention of alkaline phosphatase (AP) activity in confluent and fully mature, lipidated OP9-derived O Red Oil positive adipocytes ( Figure S2D ), reminiscent of properties assigned to hematopoietic supportive AP+ reticular adventitial cells in the bone marrow (53) and possibly of AdipoCARs (54).

The non-clonal OP9 stromal cell line (55, 56) was originally derived from the calvaria of newborn osteopetrotic mice [(C57BL/6xC3H)F2-op/op] deficient in macrophage colony-stimulating factor (M-CSF), and were shown optimal for studies of hematopoietic cell development and differentiation sensitive to M-CSF (57, 58). OP9 cells have been described as preadipocytes with multilineage differentiation capacity which readily generate mature adipocytes, and express the transcriptional activators CCAAT/enhancer binding proteins (C/EBP) α and β, together with the master regulator of adipocyte differentiation peroxisome proliferator activated receptor-γ (PPARg), and perilipin (PLIN), a phosphoprotein associated with the surface of lipid droplets (LDs) (22, 56). Thanks to their multipotency, hematopoietic support, and adipocytic differentiation capacities, OP9 cells are thus ideal for in vitro studies on BM adipogenesis. Due to their non-clonal nature and inherent heterogeneity, they serve as a BM model for studying BMAd differentiation at the single cell level.

In vitro, the OP9 cell line spontaneously differentiates into adipocytes upon confluency in the presence of serum (spontaneous OP9 adipocytes, sOP9), a phenomenon that can be more robustly promoted by exposure to a classical adipocytic differentiation cocktail consisting of insulin, dexamethasone, and 3-isobutyl-1-methylxanthine (induced OP9 adipocytes, iOP9) (22, 55). After 17 days in culture, both sOP9 and iOP9 culture conditions contained numerous differentiated OP9-adipocytes ( Figure 1 ) with a significantly greater prevalence of Oil Red O staining of lipids in the induced condition (iOP9) (p<0.01) as compared to control baseline undifferentiated OP9 cells plated at sub-confluency (uOP9) ( Figure 1G ). Spontaneous OP9-adipocytes formed under the minimal confluency condition (sOP9) and stained with Oil Red O at intermediate levels between undifferentiated uOP9 cells and induced iOP9 cultures ( Figure 1G ). No significant amount of mature adipocytes could be visualized in the uOP9 condition ( Figures 1A, D ). Adipocytes were defined as previously described, where cells containing at least four identifiable lipid droplets were accounted as mature adipocytes (59). Morphologically, both sOP9 and iOP9 adipocytes stained with Oil Red O at day 17, but iOP9 adipocytes constituted the majority of the cultures and often contained large lipid droplets, while sOP9 adipocytes were sparser and their lipid droplets seemed smaller.

Given the intermediate phenotype of sOP9 adipocytic cultures, containing a mixture of mature adipocytes and non lipidated stromal cells likely including preadipocytes, we wished to determine whether the transcriptional profile and lipid composition of these cultures more resembled the undifferentiated uOP9 cultures or the highly lipidated iOP9 adipocytic cultures. We thus compared transcriptional adipocyte differentiation markers by RT-qPCR and the global fatty acid (FA) composition by high performance liquid chromatography (HPLC) for the three conditions.

RNA transcripts for the canonical adipocyte differentiation and maturation program (Fabp4, Adipoq, Cebpa, Pparg, Lpl, Pnpla2 -also known as Atgl-) in the sOP9 condition most resembled iOP9 adipocytic cultures, with average gene expression more than 10 times higher for the sOP9 and iOP9 conditions than for the control uOP9 condition ( Figure 4A ). These results were further validated in the OP9-EGFP cell line used in previous studies [ Figures S3A, B (15)]. Further analysis of the desaturation and elongation pathway for fatty acid synthesis showed a similar pattern ( Figure 4B ), with overall higher expression for the sOP9 and iOP9 adipocytic conditions as compared to the uOP9 undifferentiated condition. Of note RNA transcripts for Sds1 and Fads2 were however significantly between sOP9 and iOP9 cultures. We thus conclude that, at the population level, expression of genes related to the adipocyte differentiation and fatty acid desaturation programs is similar for both sOP9 and iOP9 adipocytic conditions after 7 days of culture, suggesting that the transcriptional program is triggered in sOP9 cells prior to lipid droplet accumulation.

Conversely, total FA content and profiles were similar for control uOP9 cells and the sOP9 adipocytic condition as measured by HPLC analysis. Classical adipogenic induction of iOP9 cells resulted in a 16-fold increase of total FA content as compared with sOP9 cells or uOP9 cells (2.4x105 ± 2.2x104pmol/cm2 versus 1.5x104 ± 7.0x103pmol/cm2 and 17.4x104 ± 3.7x104pmol/cm2 respectively, Figure 2C ). FA profiles revealed a preferential enrichment of MUFAs in iOP9 cells (47% MUFAs versus 38% in uOP9s and 36% in sOP9 cells), mostly to the detriment of PUFAs (only 2% in iOP9 cells versus 6% PUFAs in uOP9s and 10% in sOP9 cells, p<0.0001) ( Figures 2D, G ). In terms of individual FAs, the most striking finding was the marked high ratio of palmitoleic acid (16:1) in iOP9 cells, representing 29% of total FA (versus 9% in uOP9 or sOP9 cells, p=0.007) ( Figures 2D, E ). This result was associated to a reduction of 18 carbon FAs in iOP9 cells and a trend towards a consistent reciprocal enrichment of 18 carbon FAs in sOP9 cells: stearic acid (18:0, sOP9 11% vs. iOP9 4%, p=0.02), oleic acid (18:1, sOP9 26% vs. iOP9 16%, p=0.07) and linoleic acid (18:2 n-6, sOP9 4% vs. iOP9 1%, p=0.007). The most abundant PUFA species was arachidonic acid (20:4 n-6). In relative terms, linoleic (18:2 n-6) and arachidonic acid (20:4 n-6) were significantly enriched in the sOP9-adipocytes at 5% of total FA while making up only 1% of total FA in iOP9-adipocytes. Notably, the ratio of palmitoleic acid (16:1) versus its saturated form (stearic acid, 16:0) in iOP9 adipocytes was three times higher than in sOP9 adipocyte cultures or uOP9 control cells ( Figure 2E ). The ratio of oleic acid (18:1) versus its saturated form (stearic acid, 18:0) showed a similar but non-significant trend ( Figure 2F ). These results are very similar to those obtained on rats’ tibiae when comparing lipid profiles from constitutive adipocytes versus regulated adipocytes (1). Our data is also congruent with the results shown for rabbit BMAds when comparing FA composition of rabbit BMAs from rBMAd-rich versus cBMAd-rich sites ( (35) and summarized in Table 1 ).

Overall, these results indicate that, at the population level, sOP9 adipocytic cultures share transcriptional similarities in the adipocyte differentiation program with iOP9 cells, but are less lipidated and present lower abundance of unsaturated FAs. This constellation would be consistent with our hypothesis that sOP9 adipocytic cultures present similarities with regulated BMAds, and thus an enrichment in saturated FA, as compared to cBMAd-like iOP9 adipocytic cultures. Given the heterogeneous lipidation of sOP9 adipocytic cultures and the inevitably bulk nature of our RT-PCR and HPLC-based analysis, we set out to validate our hypothesis through a single cell analysis pipeline that could directly compare sOP9 and iOP9 adipocytes on their relative FA acid composition at the single cell level.

Raman microspectroscopy represents a powerful yet non-invasive and label-free method to assess the lipid composition of adipocytes in vitro. The unsaturation ratio is recognized as a measure of the proportion of unsaturated FA obtained by the ratio of the peak area assigned to unsaturated bonds (C=C) divided by the peak area assigned to saturated bonds (CH2) (Equation 1) (60–63).

Equation 1: Lipid unsaturation ratio measured by Raman microspectroscopy.

Our feasibility test with Raman microspectroscopy imaging showed that it was possible to acquire Raman spectra at single lipid droplet (LD) resolution in our model ( Figure 3 ). After 17 days of culture, Raman spectra, representative of the molecular composition of individual LDs, were acquired for sOP9 and iOP9 adipocytes through high resolution Raman imaging. No analysis was performed for the undifferentiated (uOP9) condition as it presented with very few, if any, LDs. Raman images of individual adipocytes were processed by PCA. The PCA shows that a mixture of spectra exists between the adipocytes but also within the LDs themselves, as illustrated by the Raman images ( Figures 3B, E ). In Figure 3C , the PC1 scores (96.68%) show a representative spectrum rich in unsaturated lipids for the iOP9 adipocyte shown in Figures 3A, B . In Figure 3F , the PC1 scores (95.69%) show a spectrum representative of a saturated rich lipids for the sOP9 adipocyte shown in Figures 3D, E . Figure S4 shows the additional variation captured by PC2 and PC3 (<5%) for the same two adipocytes shown in Figure 3 . We conclude that LDs composed of predominantly unsaturated-rich or saturated-rich lipids can be identified through Raman microspectroscopy within OP9-derived adipocytes at single LD resolution.

In order to determine if significant differences exist in the unsaturation ratio of FAs accumulated within sOP9 versus iOP9 adipocytes, we thus proceeded to the acquisition of Raman spectra from individual LD in both conditions and throughout 3 separate experimental campaigns (iOP9: n = 138 adipocytes and 2971 spectra; sOP9: n = 120 adipocytes and 2944 spectra; lateral resolution 1-2µm), and combined all data for aggregated principal component (PC) analysis in both conditions. Figure S5 shows the classification of spectra using unsupervised hierarchical cluster analysis, which predicted three main categories: saturated-rich, unsaturated-rich and mixture. The representative spectrum of each category is shown in Figure 4A . PCA was then performed on the averaged spectra per adipocyte. The PC1 loading plot contains characteristic spectra where the positive peaks correspond to saturated lipids (1061, 1129, and 1296 cm-1, represented in red for Figures 4A, B ) and the negative peaks correspond to unsaturated lipids (1080, 1267, and 1655 cm-1, represented in blue for Figures 4A, B ) (60). PC1 captures 78.20% of the variability of the overall dataset, and separates the Raman spectra based on the saturated-rich and unsaturated-rich profiles ( Figure 4B ). On the PCA score plot, each point represents an averaged spectrum per adipocyte ( Figure 4C ). Most of the spectra from iOP9-adipocytes (81%) form a cluster in the negative score along PC1, while the spectra of sOP9-adipocytes are scattered along PC1. This indicates that the molecular composition of LDs in sOP9-adipocytes is heterogeneous compared to iOP9-adipocytes, and that iOP9-adipocytes are enriched in unsaturated-rich lipids. Score plots in Figures 4C, D are equivalent, but the points in Figure 4C are colored according to PC1 scores, and the points in Figure 4D are colored according to the culture condition. Specifically, the points in dark red and dark blue in Figure 4C correspond to Raman spectra of saturated-rich and unsaturated-rich lipids, while intermediate colors point to Raman spectra of a rather balanced mixture of saturated and unsaturated lipids. PCA analysis shows that LDs in iOP9-adipocytes are mainly rich in unsaturated lipids, while LDs in sOP9-adipocytes are composed of a rather balanced mixture of unsaturated and saturated lipids ( Figure 4D ). However, when comparing the averaged raw unsaturation ratio (based on averaged spectra per adipocyte) we found no significant difference between iOP9 adipocytes and sOP9 adipocytes ( Figure 4E ). At the adipocyte level, both conditions have a similar averaged unsaturation ratio. This result is explained by PCA analysis which shows that LDs of both conditions are not solely composed of either uniformly saturated or unsaturated lipids, but rather by different mixtures thereof. We thus took advantage of the 1-2μ resolution of our setup and performed the analysis at the single LD level.

First, we hypothesized that the smaller diameter of LDs in sOP9 adipocytes may be related to their lower unsaturation ratio, owing to the higher mobilization rate of unsaturated FAs (64). We thus plotted LD diameter versus unsaturation ratio ( Figures 5A, B ). The LD diameter did not correlate with the unsaturation ratio. Specifically, correlation analysis for the LD diameter to unsaturation ratio shows that the parameters are not linearly correlated in either condition ( Figures 5A, B , R²spont=0.0013; R²ind=0.0107). Even when the spectra are separated according the 3 types of lipids (saturated-rich, unsaturated-rich and mixture) the LD diameter and the unsaturation ratio are not linearly correlated. All the R² are inferior to 0.05 ( Figure S6 ). Since, no correlation was observed, the LD diameter and the unsaturation ratio were explored separately.

The averaged LD diameter was significantly lower (p<0.0001) in the sOP9-adipocytes (5.24 ± 1.85µm) than the iOP9-adipocytes (7.87 ± 3.83µm) ( Figure 5C ). When we further separated LDs by type of lipids, we found that saturated-rich, unsaturated-rich or mixture LDs are not different in their average composition when originating from iOP9 or sOP9 adipocytes, but unsaturated LDs are more frequent in iOP9s while saturated LDs are more frequent in the sOP9 condition ( Figures 5D, E ; S5B ). Unsaturated-rich LDs are overall bigger in size than saturated-rich LDs, with a predominance in iOP9 adipocytes, particularly above the diameter of 10 µm. In iOP9 adipocytes, the number of unsaturated-rich LDs is higher compared to sOP9 adipocytes for LD diameter superior to 10 µm.

Then, the averaged unsaturation ratio per condition were obtained from individual spectra of LDs. The unsaturation ratio of iOP9 condition (0.17 ± 0.05) is significantly higher than sOP9 condition (0.14 ± 0.05, p=0.0014). Also on average, LDs of iOP9 condition have more unsaturated-rich lipids compared to the LDs in sOP9 condition. Since differences are observed between sOP9 and iOP9 around a LD diameter of 10 µm ( Figure 5E ), the unsaturation ratio of LDs was investigated as function of the threshold of 10 µm ( Figure 5G ). The larger LDs (>10µm) have the highest unsaturation ratio compared to smaller LDs. Smaller LDs (<10µm) from iOP9 adipocytes have a higher unsaturation ratio compared to sOP9-adipocyte. Then, we investigated the unsaturation ratio as a function of the 3 types of LDs, namely saturated-rich, mixture or unsaturated-rich. Although we found consistent differences between the three classes of LDs, there were no significant differences between LDs of the same class in sOP9 adipocytes versus iOP9-adipocytes ( Figure 5H ). Not surprisingly, mixture LDs showed the greatest heterogeneity in their unsaturation ratio ( Figure 5H ). Overall, we conclude from this data that iOP9 adipocytes contain significantly bigger LDs and more frequently unsaturated LDs than sOP9 adipocytes. In vivo, both higher LD diameter and higher FA unsaturation are characteristics of cBMAds, while smaller LD diameter and lower unsaturation ratio are characteristics of rBMAds. We could not find a consistent relationship between LD diameter and unsaturation ratio.

Another in vivo property of rBMAds as compared to cBMAds is their differential capacity to serve as a supportive niche to hematopoietic progenitor cells. Indeed, cBMAds are associated to the hematopoietic poor -yellow- regions of the marrow, while rBMAds are associated to the hematopoietic-rich -red- regions of the marrow. To determine whether sOP9 versus iOP9 cultures behave different in terms of hematopoietic support, we co-cultured murine hematopoietic stem and progenitor cells (HSPCs) isolated as cKit+Lin-Sca1+ (KLS) cells for seven days with the differentiated OP9 cells at a ratio of 1 HSPC to 10 OP9 cells. As in previous work (65), this high ratio of KLS to OP9 cells serves to avoid dispersion due to HSPC heterogeneity and avoids downstream stochastic, clone effects on analysis secondary to differences in the content of the most primitive and thus proliferative HSPCs across wells, which although small, would be exponentially exaggerated after co-culture. At the end of the seven-day co-culture period, flow cytometric analysis revealed that total hematopoietic expansion of 2,000 seeded HSPCs was 4.5-fold lower (p<0.0001) when iOP9 cultures were used as feeder as compared to the sOP9 condition ( Figure 6A ). Of note, total hematopoietic cells in suspension were significantly lower in the iOP9 versus sOP9 condition (p<0.05) whereas the more immature progenitor cells that are typically in close contact and adherent to the feeder cells were proportionally lower in the sOP9 condition. This suggests that iOP9 cultures may preferentially favor support of the more primitive hematopoietic progenitors, which is congruent with previous data from purified human BMAd as compared to undifferentiated stromal cells (9). The more hematopoietic progenitor supportive function of the sOP9 condition is also illustrated by the quantification of colonies formed by day 7 of co-culture ( Figure 6B ). The total number of hematopoietic colonies per initial number of HSPCs seeded, was significantly higher in the sOP9 coculture (0.13 colonies per initial KLS seeded) than in the iOP9 coculture (0.01 colonies per initial KLS seeded). All three colony categories, from the shiny consisting of more mature hematopoietic cells to the mixed and cobblestone colonies which contain the most immature hematopoietic progenitors were significantly increased (P<0.05) in the sOP9 condition. To understand to which extent secreted factors from either condition contributed to hematopoietic expansion, HSPCs were cultured for two days in the absence of feeder cells but in the presence of different concentrations of 48-hour conditioned media from either uOP9, sOP9 or iOP9 cultures. For the iOP9 condition we tested both a short (5 + 2 days) and a long (17 + 2 days) DMI adipogenesis induction protocol ( Figure 7C ). Hematopoietic expansion was greatest for HSPCs cultured in the highest concentration of undifferentiated-conditioned medium. HSPCs cultured in conditioned medium from iOP9-adipocytes did not show a significant expansion after two days in culture, independently of the length of the iOP9 DMI induction protocol. However, the conditioned medium from sOP9 cultures was substantially more supportive to HSPC maintenance than iOP9, although less than those cultured in conditioned media from undifferentiated OP9 cells. This effect increased with higher ratios of conditioned-to-basal medium. We therefore conclude that the in vitro hematopoietic progenitor support capacity of sOP9 cultures is superior to that of iOP9 cultures, aligning with the in vivo functionality of regulated versus constitutive BMAds.

In order to contextualize and validate the value of the FA unsaturation ratio used throughout this work as a surrogate marker for regulated versus constitutive BMAds, we set to test whether we could find differences in the unsaturation ratio of lipids extracted from one of the most extreme BM remodeling scenarios. Patients suffering from acute leukemias receive intensive aplastic chemotherapy, a regime that eliminates leukemic cells but also virtually all short-term hematopoietic progenitors in the BM, leading to a rapid adipocytic conversion of the marrow. BMAds formed in this context should represent a stereotypical example of rBMAd, and we thus set out to compare the unsaturation ratio of lipids obtained from iliac crest samples from these patients as compared to the lipid fraction obtained from surgical debris of adult patients undergoing elective hip replacement surgery, which is thought to represent a cBMAd-rich site. Analysis of these primary human BM lipid samples reflected a preferential enrichment of MUFAs in the post-chemotherapy iliac crest BM samples (regulated BMAd site) versus the femoral head BM samples (constitutive BMAd site) from hip replacement surgery (39.6% versus 51.2%, p=0.02) to the detriment of PUFAs (18.6% versus 17.0% p=0.80) and SFAs (41.8% and 36.6%, p=0.04) ( Figure 7D ). While rapid conversion to adipocytic marrow is induced by chemotherapy, and thus iliac crest samples are interpreted as rBMAd-rich, the differences in fatty acid profiles displayed could also be an inherent function of the anatomical site. The largest difference between FA species in the regulated versus constitutive marrow sites was seen in the 18:1 to 18:0 unsaturation ratio (4.40 versus 7.91, p=0.007), and the 16:1 versus 16:0 unsaturation ratio (0.11 versus 0.19, p=0.031) ( Figures 7B, C ). We thus conclude that human BM samples from a rBMAd-rich site in the context of BM remodeling driven by hematopoietic demand present a lower unsaturation ratio, as measured by 16:1/16:0 -and 18:1/18:0- FA content, than samples obtained from a cBMAd-rich site, which complements previously reported results for rabbit and rats marrow at steady-state (1, 35–37) as summarized in Table 1 .

Taken together, we conclude that the spontaneous versus induced OP9 in vitro adipogenesis model presented here replicates key differences in lipid droplet size, FA unsaturation ratio and hematopoietic support respectively associated in vivo to regulated versus constitutive BMAd subtypes. We additionally validate the value of the 16:1/16:0 FA unsaturation ratio as a surrogate marker to segregate rBMAd-rich versus cBMAd-rich tissues both in vivo and for our in vitro model, and the feasibility of Raman microspectroscopy to reveal equivalent differences in the unsaturation ratio at the single adipocyte level.