Upregulation of FOXM1 leads to diminished drug sensitivity in myeloma

Levels of FOXM1 mRNA in myeloma cells were determined using Affymetrix U133Plus 2.0 microarrays (Santa Clara, CA) as previously described [15, 16]. Statistical analysis of microarray data relied on GCOS1.1 software (Affymetrix, Santa Clara, CA). Patients at UAMS were treated using the Total Therapy 2 regimen, the backbone of which is high-dose melphalan therapy (HDT) and autologous stem cell transplantation (ASCT). Half of the patients received thalidomide both during intensive therapy and as maintenance therapy. The therapeutic approach to relapsing disease was not uniform and depended mainly on the time to relapse, the pace of relapse (slow versus aggressive), the presence or absence of organ dysfunction, and the patient’s overall health status, physical and mental fitness and treatment preference.

Four IgA-producing HMCLs, designated CAG, XG1, H929 and ARP1, were included in this study. The identity of the cell lines was validated as previously described [12], using chromosomal translocation status and gene expression spikes as main parameters. Cells were propagated in vitro at 37 °C and 5% CO₂ using RPMI1640 cell culture medium (Gibco) supplemented with 10% heat-inactivated fetal bovine serum (Atlanta Biologicals) and antibiotics (100 units/mL penicillin and 100 μg/mL streptomycin, Sigma). In some experiments, CAG and XG1 cells over-expressing FOXM1 (FOXM1^Hi) were compared to cells containing normal amounts of FOXM1 (FOXM1^N) [12]. In other experiments, H929 and ARP1 cells in which FOXM1 expression had been knocked down using shRNA (FOXM1^Lo) were compared to parental FOXM1^N cells [12]. Chemicals including myeloma drugs were purchased from Sigma (doxorubicin [Dox], thiostreptone [TS]), Millennium Pharmaceuticals (bortezomib [Bz]), or Invitrogen (propidium iodide, RNase A).

For cell cycle analysis, cells were fixed in ice-cold ethanol (1 h, 4 °C), washed in PBS, re-suspended in propidium iodide (PI) solution (40 μg/ml, 3 h, 4 °C) supplemented with 50 μl RNase A (10 μg/ml), and evaluated by flow cytometry using a FACScan (Becton Dickinson, San Jose, CA). For determination of clonogenicity, 10⁴ myeloma cells were seeded in soft-agar plates (0.5 ml RPMI1640 supplemented with 0.33% agar and 10% FBS) and grown for 2 weeks at 37 °C and 5% CO₂ – in some cases exposed, during week 2, to myeloma drugs. Myeloma clones, defined as tight aggregates of ≥40 tumor cells, were enumerated on digital images of soft-agar plates analyzed with the help of Image J. For measurement of proliferation and viability, cells were counted using a hemocytometer and evaluated for exclusion of trypan blue (0.4% dye in PBS, pH 7.3), respectively. For determination of apoptosis, the flow-cytometric Annexin V APC assay (eBioscience, San Diego, CA) was used according to the manufacturer’s instructions. For determination of drug-efflux capacity, the flow-cytometric eFluxx-ID™ multidrug resistance assay was employed. MCF7 human breast cancer cells were included as benchmark. For determination of senescence-associated β-galactosidase (β-gal) activity, a kit from Cell Signaling Technology (Cat# 9860) was used. Briefly, cells were fixed (15 min), washed (PBS) and incubated in β-gal staining solution overnight at 37 °C.

Protein levels were determined by Western analysis using antibodies to FOXM1, Rb and pRb obtained from Santa Cruz Biotechnology (FOXM1, sc-500) or Cell Signaling Technology (Rb, 9309; pRb, 8515). Cells were lysed with assistance of the Mammalian Cell Extraction kit (K269–500) from Biovision, Milpitas, CA. 10-μg samples of protein were fractionated on 4–12% SDS-PAGE gels, followed by transfer to nitrocellulose membranes blocked with 5% non-fat dry milk in Tris-buffered saline containing 0.05% Tween-20. Incubation with primary antibodies occurred overnight at 4 °C. Proteins were visualized using HRP-conjugated secondary antibody and SuperSignal West Pico (Pierce, Rockford, IL). Membranes were subsequently stripped and re-probed for β-actin (Santa Cruz Biotechnology, sc-47778), which served as loading control.

To compare the drug response in CAG myeloma cells expressing normal and elevated FOXM1 levels, respectively, 2 × 10⁶ FOXM1^N and FOXM1^Hi cells were injected subcutaneously (SC) into the right or left flank of NSG mice (Jackson Laboratory, Bar Harbor, Maine). Viability of FOXM1^N and FOXM1^Hi cells was comparable (≥90%). Seven days later, one group of mice was treated with bortezomib (1 mg/kg) administered intraperitoneally (IP) twice weekly. Another group of mice, designated untreated control, was injected with drug vehicle, normal saline (0.9% sodium chloride). In all cases, tumor growth was measured using a pair of calipers. Mice were sacrificed for humane reasons using CO₂ asphyxiation when tumors reached 20 mm in diameter. All studies were approved under protocol 1301010 of the Institutional Animal Care and Use Committee of the University of Iowa.

Using the University of Arkansas for Medical Science (UAMS) Total Therapy 2 (TT2) dataset (GSE2658) as discovery tool, we recently reported that upregulation of FOXM1 was a common feature of patients with rMM [13]. Here we confirm this finding with the help of an updated TT2 dataset that includes 127 patients with rMM and a related TT3 dataset that includes 30 patients with rMM – both available at GSE31161 (Fig. 1a). The increase in median FOXM1 mRNA levels in these datasets, 2.7-fold in case of TT2 (76/28) and 2.3-fold in case of TT3 (330/143), was very similar to the one observed in the original TT2 dataset (2.89-fold; 79.5/27.5). Furthermore, in all 3 datasets, mean FOXM1 expression at relapse was significantly elevated (p = 10^− 3) compared to baseline. Next, we went back to the TT2 / GSE2658 dataset, presented in dot plot format in Fig. 1b, to visualize the two expression values, ND vs R, for each patient. Upregulation of FOXM1 mRNA at relapse was seen in more than two thirds of patients (61 of 88, 69%; indicated by dots above the diagonal line), whereas gene message was down in 20 (23%) patients (dots below the line) and unchanged in 7 (8%) patients (dots on the line). The magnitude of the elevation at relapse (~ 2.6-fold; 226/87.8) was similar to the decline (~ 2.3-fold; 151/64.6) using mean expression levels for comparison. These results indicated that the overall increase of FOXM1 message at relapse was caused by the preponderance of myelomas with elevated gene expression (69%), not by the circumstance that the magnitude of elevation exceeded that of reduction.

To analyze the 61 myelomas that harbored elevated FOXM1 at relapse in greater depth, we partitioned these cases in two arbitrary groups defined by median baseline expression levels of 19 microarray units (n = 46) and 85 units (n = 15), respectively, in newly diagnosed (ND) disease (Fig. 1c, black boxes with whiskers). The increase at relapse in the low expresser group was somewhat higher (4.7-fold, 89.5/19) than in the high expresser group (4.1-fold, 346/85). The difference was due, in no small measure, to four myelomas in the low expresser group that exhibited > 20-fold elevated expression peaks at relapse. This is indicated by a black ellipse in Fig. 1d. The analysis of tumors featuring reduced FOXM1 expression at relapse is depicted in Fig. 1e. Partitioning of the dataset (n = 20) in two groups with median baseline expression of 27.5 units (n = 12) and 343 units (n = 8) at the time of diagnosis (ND) demonstrated that the decrease in the high expresser group was more pronounced (6.2-fold, 343/55) than in the low expresser group (1.2-fold, 27.5/23; Fig. 1e). The difference could be attributed in large part to a subset of myelomas in the high expresser group (n = 5) that exhibited > 6-fold drops in gene expression at relapse (Fig. 1f, black ellipse). These findings led us to conclude that the pattern of FOXM1 expression in rMM is heterogeneous, with outliers in both directions contributing disproportionally to the relapse-dependent shift in gene expression. To confirm the findings presented in Fig. 1 with an independent method not relying on Affymetrix arrays, we used RT-PCR to analyze sequential ND and R CD138⁺ bone marrow tumor samples from 8 patients with myeloma undergoing HDT/ASCT therapy. Although the number of clinically confirmed relapses was low (n = 3), the increase in FOXM1 expression was impressive, up to 35-fold (Additional file 1: Figure S1).

To assess the biological outcomes of elevated FOXM1 in MM, we relied on two HMCLs, XG1 and CAG, as experimental model system. The cells were manipulated to contain either elevated levels of FOXM1 (FOXM1^Hi) or normal levels of gene message and protein (FOXM1^N) due to transfection with a human FOXM1 (isoform C) expressing lentivirus or a non-coding “empty” lentivirus (used as control), respectively [12] (Fig. 2a). Flow cytometric determination of cell proliferation showed that upregulation of FOXM1 promotes cell cycle progression (Fig. 2b). FOXM1^Hi CAG and XG1 cells exhibited 14 and 13% higher growth rates, respectively, compared to FOXM1^N cells (compare columns labeled “Co”). Treatment of cells with the FOXM1-inhibiting thiazole antibiotic, thiostrepton (TS) [17], slowed the growth of both FOXM1^Hi and FOXM1^N cells. However, the low Hi-to-N ratio indicated that FOXM1^Hi cells were more sensitive to TS than FOXM1^N cells (columns labeled “TS”). Figure 2c depicts an example of the magnitude of TS-dependent growth inhibition of FOXM1^Hi cells under conditions of higher drug concentration compared to panel B. The decrease in proliferation amounted to 72% (4.72/16.9) in case of CAG cells (left panel) and 76% (6.40/26.1) in case of XG1 cells (right panel). The results presented above were in line with genetic evidence gleaned from transcriptomic studies using microarrays, indicating that FOXM1 promotes myeloma proliferation. Thus, FOXM1 expression is positively correlated (Pearson’s r = 0.712, p < 10^− 4) with myeloma cell proliferation in 244 Bz-treated patients available at GSE9782 [18], using the global gene expression-based proliferation index (GPI) of myeloma devised by Bergsagel et al. [19] as proxy of actual tumor cell proliferation (Fig. 2d). Similarly, we recently reported [13] that FOXM1 message levels in nMM and rMM paralleled the GPI of developed by Hose and his associates [20]. Next, we used eFluxx-ID Gold MDR analysis to demonstrate that upregulation of FOXM1 results in enhanced ABC-transporter efflux-pump activity in myeloma (Fig. 2d). This finding suggested that FOXM1-dependent promotion of cell cycle progression in myeloma may facilitate drug resistance by means of enhanced outflow of myeloma drugs.

Because recurrent cancers including rMM may acquire therapy resistance due to alterations in biological pathways in which FOXM1 is involved [21], we wondered whether upregulation of FOXM1 may decrease the sensitivity of myeloma cells to widely used myeloma drugs, such as bortezomib (Bz) and doxorubicin (Dox). We found that enforced expression of FOXM1 renders myeloma partially resistant to both drugs, as evidenced by half-maximal inhibitory concentrations (IC₅₀) that, in case of Bz, were 5.6-fold (CAG) and 1.9-fold (XG1) higher in FOXM1^Hi than FOXM1^N cells (Fig. 3a, left). In case of Dox, they were 2.9-fold (CAG) and 1.5-fold (XG1) higher (Fig. 3a, right). Flow cytometric measurements of annexin V, a marker of apoptosis, were in agreement with these results as FOXM1^Hi cells invariably exhibited less drug-induced death than FOXM1^N cells did. For example, there was a 2.7-fold difference in CAG cells treated with Bz (23% vs. 63%) and a 1.5-fold difference in XG1 cells treated with Dox (49% vs. 72%, Fig. 3b). Determination of myeloma growth in soft agar demonstrated the clonogenic advantage of FOXM1^Hi cells. One example, depicted in Fig. 3c, shows that FOXM1^Hi CAG cells treated with 2 nM Bz or left untreated produced ~ 50% (right panels) or ~ 25% more colonies (left panels), respectively, than FOXM1^N controls. XG1 cells demonstrated greater differences based on Hi-to-N ratios (Fig. 3d), suggesting that FOXM1 protects the clonogenic growth of these cells under conditions of selective drug pressure even more effectively than that of CAG cells. These findings were of interest in light of published reports that suppression of FOXM1 may sensitize human cancer to killing by genotoxic drugs [22].

To complement the in vitro findings described above with in vivo data, we determined whether upregulation of FOXM1 results in decreased susceptibility of myeloma to proteasome inhibition (PI). We used human-in-mouse myeloma xenografts treated with Bz or left untreated (control) as model system. FOXM1^Hi and FOXM1^N CAG cells were transferred under the skin of the left and right flank of NSG mice (n = 10), respectively. Treatment of 5 tumor-bearing hosts using daily IP injections of Bz commenced 7 days later. Five tumor-bearing hosts left untreated were used as controls (Fig. 4a). Tumor diameters were measured in 4-day intervals to compare xenograft growth rates. FOXM1^Hi tumors harvested on day 30 (study endpoint) were larger than their FOXM1^N counterparts in both the Bz treatment arm (15.2 ± 1.97 mm vs 10.4 ± 0.771 mm) and the control arm (19.0 ± 1.31 mm vs 14.0 ± 1.73 mm) arm (Fig. 4b). This was consistent with FOXM1’s growth-promoting activity described above. However, in terms of sensitivity to PI, FOXM1^Hi tumors were slightly less responsive (by 10%) than their FOXM1^N counterparts on the contralateral side of the same host. Specifically, Bz-dependent growth inhibition of FOXM1^Hi tumors at study endpoint (1.32 ± 0.434 g in treated mice vs. 2.40 ± 0.647 g in mice left untreated) amounted to 45%, whereas that of FOXM1^N tumors (0.692 ± 0.217 g treated vs 1.54 ± 0.543 g untreated) amounted to 55% (Fig. 4c). This result, however modest it might be, was consistent with published findings that lowering FOXM1 in human cancer cells leads to enhanced PI-induced killing [23].

Our recent work has implicated the cyclin-dependent kinase 6 (CDK6) / retinoblastoma (Rb) axis in the mechanism by which FOXM1 promotes myeloma [14]. To confirm the co-regulation of FOXM1 protein levels and the tumor suppressor, Rb, in myeloma, we performed triplicate Western analyses of paired FOXM1^Hi and FOXM1^N tumors (Fig. 5a, left). For sake of comparison to FOXM1^Hi, we included FOXM1^Lo samples from H929 and ARP1 HMCL cells, in which FOXM1 had been knocked down using lentivirus-delivered RNA interference (Fig. 5a, right). Densitometric analysis of Western blots showed that total Rb and phosphorylated Rb (pRb) were increased in FOXM1^Hi cells by 20–40%, with somewhat higher levels seen in XG1 than CAG cells (Fig. 5b, left). Conversely, Rb and pRb were decreased in FOXM1^Lo cells (~ 30 to 50%), with the loss of the latter somewhat exceeding that of the former (Fig. 5b, right). Inspired by a large body of evidence that connects the Rb-governed cell fate decision of senescence with important pathways of cancer relapse such as drug resistance, tumor dormancy and tumor stemness [14], we determined whether myeloma cells might express the classic senescence-associated phenotype of β-galactosidase in a FOXM1-dependent manner. Fig. 5c, left shows that FOXM1 knockdown sufficed to induce β-gal activity in myeloma. Furthermore, treatment of cells using Dox caused higher β-gal⁺ scores in FOXM1^N than FOXM1^Hi cells (Fig. 5c, right). These findings were consistent with published reports on the role of FOXM1 in suppressing cellular senescence in a variety of solid and liquid neoplasms [24].