Connectivity Map-based discovery of parbendazole reveals targetable human osteogenic pathway

Significance

Osteoporosis, a disease characterized by increased bone fragility and fracture risk, affects over 20% of the ever-growing elderly population. It is important to further our knowledge of bone cell biology so we can develop new bone anabolic treatments. By combining genomic and bioinformatic tools against the backdrop of osteogenic differentiating human mesenchymal stromal cells, we have identified a previously unidentified bone anabolic compound that induces osteoblast differentiation in a subset of the hMSC population through cytoskeletal changes and increased bone morphogenetic protein 2 activity. Through this novel approach we identified an important mechanism of lineage allocation and demonstrated the significance of cytoskeletal organization in osteogenic differentiation, providing us with a novel mechanism for bone formation to target for new osteoporosis treatments.

Abstract

Osteoporosis is a common skeletal disorder characterized by low bone mass leading to increased bone fragility and fracture susceptibility. In this study, we have identified pathways that stimulate differentiation of bone forming osteoblasts from human mesenchymal stromal cells (hMSCs). Gene expression profiling was performed in hMSCs differentiated toward osteoblasts (at 6 h). Significantly regulated genes were analyzed in silico, and the Connectivity Map (CMap) was used to identify candidate bone stimulatory compounds. The signature of parbendazole matches the expression changes observed for osteogenic hMSCs. Parbendazole stimulates osteoblast differentiation as indicated by increased alkaline phosphatase activity, mineralization, and up-regulation of bone marker genes (alkaline phosphatase/ALPL, osteopontin/SPP1, and bone sialoprotein II/IBSP) in a subset of the hMSC population resistant to the apoptotic effects of parbendazole. These osteogenic effects are independent of glucocorticoids because parbendazole does not up-regulate glucocorticoid receptor (GR) target genes and is not inhibited by the GR antagonist mifepristone. Parbendazole causes profound cytoskeletal changes including degradation of microtubules and increased focal adhesions. Stabilization of microtubules by pretreatment with Taxol inhibits osteoblast differentiation. Parbendazole up-regulates bone morphogenetic protein 2 (BMP-2) gene expression and activity. Cotreatment with the BMP-2 antagonist DMH1 limits, but does not block, parbendazole-induced mineralization. Using the CMap we have identified a previously unidentified lineage-specific, bone anabolic compound, parbendazole, which induces osteogenic differentiation through a combination of cytoskeletal changes and increased BMP-2 activity.

Osteoporosis is a common and devastating bone disease characterized by reduced bone mass and increased fragility and fracture risk. It has been estimated that an osteoporotic fracture occurs once every 8 s worldwide (1), and direct healthcare costs in Europe alone are at least €31.7 billion annually (2). Osteoporosis means porous bones and occurs when bone remodeling is disrupted. Bone remodeling is a balancing act between removal of old bone and formation of new bone, which is achieved by two distinct cells, the osteoclast and osteoblast, respectively. When uncoupling of these two processes takes place, bone resorption can overtake bone formation, resulting in osteoporosis. Most osteoporosis treatments, such as bisphosphonates, reduce bone resorption and result in modest increases in bone density; however, these treatments do not result in a true bone anabolic effect, so patients do not regain bone that has been lost at time of diagnosis. An ideal treatment would stimulate bone formation as well, to help repair the damage already done to the bone microarchitecture and strength; with this in mind, our goal was to search for previously unidentified molecules and/or mechanisms that stimulate human osteoblast differentiation and bone formation.

The connectivity map (CMap) is a web-based tool that allows for screening of compounds against a genome-wide disease or physiological gene signature (3, 4). This screening is achieved by comparing microarray data from more than 1,300 small molecules to a user’s selected gene signature of the phenotype of interest using a pattern-matching algorithm with a high level of resolution and specificity. The screening results in a list of compounds with a highly correlating gene expression pattern to that of the phenotype of interest, which has the potential to aid in finding a novel treatment for a disease or to identify novel pathways or genes involved in a complex biological process. To date, the CMap has been successfully used to identify compounds and combination therapies that show promise in the treatment of osteoarthritic pain (5), adenocarcinoma (6), kidney disease (7), gliomas (8), and NK cell neoplasms (9).

Our aim was to identify previously unidentified anabolic therapeutic targets by genomic, proteomic, and bioinformatic dissection of human mesenchymal stromal cell (hMSC)-derived osteoblasts. Therefore, we used the CMap to identify compounds with a matching gene expression profile to human mesenchymal stem cells undergoing osteogenic differentiation. By following this approach, we aimed to not only discover novel compounds that stimulate osteogenic differentiation, but also novel processes underlying this process.

Results

Parbendazole Has the Strongest Correlating Gene Signature to hMSCs Undergoing Osteoblast Differentiation.

Using a pattern-matching algorithm, the CMap links compounds with disease or physiological phenotypes by measuring similarities in gene expression (3). To identify compounds that may exert bone anabolic effects due to their ability to stimulate genes regulated during osteoblast differentiation, we performed a CMap analysis in which we searched for drugs that have a gene expression pattern positively correlating to hMSCs differentiating toward osteoblasts (Table S1). Multiple drugs were identified that have a significantly correlating gene expression pattern to that of differentiating human osteoblasts, including dexamethasone (dex) and a number of other corticosteroids (Table 1). These results demonstrate the validity of the CMap, because dex is the compound that we originally used to stimulate osteoblast differentiation and glucocorticoid (GC)-mediated activation of the glucocorticoid receptor (GR) is the classical stimulus for human osteoblast differentiation. Because our aim was to identify novel compounds/pathways that stimulate osteoblast differentiation, we excluded any corticosteroid compound for further testing. We found that the strongest correlating compound (P < 0.0001) (Table 1), and only noncorticosteroid in the top eight compounds, is the benzimidazole anthelmintic parbendazole. Based on these results, we chose to scrutinize the effects parbendazole has on hMSCs in regard to its osteogenic potential.

Table S1.

List of up-regulated and down-regulated genes at 6 h post start of osteogenic differentiation in hMSCs and their associated Affymetrix probe identifiers used for initial CMap query

No.	Up-regulated probes		Down-regulated probes
	Affymetrix probe ID HG-U133A	Gene symbol	Affymetrix probe ID HG-U133A	Gene symbol
1	205730_s_at	ABLIM3	222162_s_at	ADAMTS1
2	201963_at	ACSL1	201034_at	ADD3
3	207589_at	ADRA1B	218631_at	AVPI1
4	204174_at	ALOX5AP	204907_s_at	BCL3
5	221009_s_at	ANGPTL4	207510_at	BDKRB1
6	206029_at	ANKRD1	201169_s_at	BHLHB2
7	206176_at	BMP6	211518_s_at	BMP4
8	218723_s_at	C13orf15	216598_s_at	CCL2
9	218309_at	CAMK2N1	203666_at	CXCL12
10	219398_at	CIDEC	212977_at	CXCR7
11	221541_at	CRISPLD2	202434_s_at	CYP1B1
12	209283_at	CRYAB	204977_at	DDX10
13	214724_at	DIXDC1	218858_at	DEPDC6
14	204602_at	DKK1	201340_s_at	ENC1
15	203810_at	DNAJB4	215704_at	FLG
16	201041_s_at	DUSP1	204948_s_at	FST
17	209457_at	DUSP5	203925_at	GCLM
18	214445_at	ELL2	221576_at	GDF15
19	201324_at	EMP1	206614_at	GDF5
20	205521_at	ENDOGL1	210640_s_at	GPER
21	208962_s_at	FADS1	218468_s_at	GREM1
22	204560_at	FKBP5	206432_at	HAS2
23	206860_s_at	FLJ20323	202934_at	HK2
24	205021_s_at	FOXN3	203665_at	HMOX1
25	202723_s_at	FOXO1	209905_at	HOXA9
26	203592_s_at	FSTL3	201565_s_at	ID2
27	209892_at	FUT4	201631_s_at	IER3
28	203725_at	GADD45A	206332_s_at	IFI16
29	207574_s_at	GADD45B	214059_at	IFI44
30	201841_s_at	HSPB1	205207_at	IL6
31	206375_s_at	HSPB3	201625_s_at	INSIG1
32	218934_s_at	HSPB7	201464_x_at	JUN
33	209184_s_at	IRS2	210261_at	KCNK2
34	201389_at	ITGA5	201650_at	KRT19
35	204301_at	KBTBD11	206969_at	KRT34
36	208960_s_at	KLF6	206481_s_at	LDB2
37	212442_s_at	LASS6	206953_s_at	LPHN2
38	218574_s_at	LMCD1	218559_s_at	MAFB
39	215322_at	LONRF1	212530_at	NEK7
40	212859_x_at	MT1E	206814_at	NGF
41	204745_x_at	MT1G	220132_s_at	NPM1
42	217546_at	MT1M	203708_at	PDE4B
43	204326_x_at	MT1X	220343_at	PDE7B
44	203036_s_at	MTSS1	203131_at	PDGFRA
45	218330_s_at	NAV2	217996_at	PHLDA1
46	213012_at	NEDD4	203354_s_at	PSD3
47	201502_s_at	NFKBIA	207017_at	RAB27B
48	218786_at	NT5DC3	202677_at	RASA1
49	205960_at	PDK4	209568_s_at	RGL1
50	212239_at	PIK3R1	204337_at	RGS4
51	207290_at	PLXNA2	212724_at	RND3
52	204285_s_at	PMAIP1	213236_at	SASH1
53	206631_at	PTGER2	202656_s_at	SERTAD2
54	204748_at	PTGS2	205856_at	SLC14A1
55	206157_at	PTX3	205396_at	SMAD3
56	204916_at	RAMP1	208127_s_at	SOCS5
57	202388_at	RGS2	202935_s_at	SOX9
58	212099_at	RHOB	203217_s_at	ST3GAL5
59	202082_s_at	SEC14L1	210612_s_at	SYNJ2
60	204541_at	SEC14L2	203083_at	THBS2
61	202627_s_at	SERPINE1	209386_at	TM4SF1
62	201739_at	SGK1	206025_s_at	TNFAIP6
63	214719_at	SLC46A3	204932_at	TNFRSF11B
64	209453_at	SLC9A1	215111_s_at	TSC22D1
65	212797_at	SORT1	204881_s_at	UGCG
66	219257_s_at	SPHK1	220976_s_at	ZNF98; KRTAP1-1; ZNF49; ZNF849P
67	220983_s_at	SPRY4
68	219315_s_at	TMEM204
69	207001_x_at	TSC22D3
70	205480_s_at	UGP2
71	206796_at	WISP1
72	221029_s_at	WNT5B
73	205883_at	ZBTB16
74	212704_at	ZCCHC11
75	211962_s_at	ZFP36L1
76	204131_s_at	ZNF286C
77	212742_at	ZNF364

Table 1.

CMap permuted results showing compounds with significant positive correlation to osteogenic hMSCs gene signature

Rank	Compound name	Cell line	Mean CMap score	n	P value
1	Parbendazole	PC3	0.855	2	<0.00001
2	Dexamethasone	PC3	0.913	2	<0.00001
3	Fludrocortisone	PC3	0.768	2	0.00002
4	Halcinonide	PC3	0.788	2	0.00002
5	Fludroxycortide	PC3	0.775	2	0.00002
6	Flumetasone	PC3	0.854	2	0.00002
7	Flunisolide	PC3	0.718	2	0.00002
8	Fluocinonide	PC3	0.72	2	0.00002

Top matching compounds from the Connectivity Map based on the average of the replicates of a single compound for each cell line. Shading indicates a glucocorticoid. Top matching compound and only nonglucocorticoid, parbendazole, in white. Rank is based on the P value calculated from the CMap scores all of the replicates of a single compound in a single cell line. Score is based on the relative strength of a given signature in an instance from the total set of instances calculated upon execution of a query.

Parbendazole Induces Osteoblast Differentiation of hMSCs in Vitro.

Similar to the known stimulator of osteogenic differentiation, dex, parbendazole treatment stimulated alkaline phosphatase (ALP) activity after 1 wk of culture and mineralization after 3 wk of culture, dose dependently up to 4 μM (Fig. 1 A and B, respectively). We confirmed the dose-dependent parbendazole-induced mineralization by alizarin red staining (Fig. 1C). These results show that parbendazole induces biochemical changes in hMSCs leading to osteoblast differentiation and mineralization.

Fig. 1.

Parbendazole induces osteogenic differentiation of hMSCs. Results of ALP activity after 1 wk of culture (A) and mineralization after 3 wk of culture (B) in hMSCs treated with 1 μM parbendazole (light gray bar), 4 μM parbendazole (dark gray bar) compared with negative control (control medium; white bar) or positive control (0.1 μM dex; black bar) treated cells. (C) Dose-dependent induction of mineralization was confirmed by alizarin red staining after 3 wk of culture. mRNA expression levels of ALPL (D), IBSP (E), and SPP1 (F) 7 d after the start of treatment with control medium (white bar), 4 μM parbendazole (gray bar), or 0.1 μM dex (black bar) as assessed by quantitative PCR. For biochemistry, n = 12. For PCRs, n = 6. *P < 0.05, **P < 0.01, ***P < 0.001. Results are presented relative to control.

We also examined if the most effective dose of parbendazole, 4 μM, induces expression of well-known osteoblast marker genes by performing quantitative PCRs on hMSCs treated with parbendazole for 7 d. Parbendazole significantly increased the expression of alkaline phosphatase (ALPL) (Fig. 1D) as well as the genes encoding the matrix proteins bone sialoprotein (IBSP) and osteopontin (SPP1) (Fig. 1 E and F).

Simian virus-immortalized human fetal osteoblasts (SV-HFO) treated with parbendazole did not mineralize (Fig. S1A), and total protein was decreased by parbendazole (Fig. S1B) at the doses stimulating osteogenesis in the hMSCs.

Fig. S1.

Parbendazole does not induce osteogenic differentiation in human preosteoblast cells (SV-HFOs), but does reduce total protein. Results of mineralization (A) and total protein content (B) after 3 wk of culture in SV-HFOs treated with 1 μM parbendazole (light gray bar), 100 nM parbendazole (medium gray bar), and 10 nM parbendazole (dark gray bar) compared with negative control (control medium; black bar) or positive control (100 nM dex; darkest gray bar) treated cells. n = 3. Results are of one representative experiment.

Taken together, these results demonstrate that parbendazole induces osteogenic differentiation of hMSCs independent of the known osteogenic stimulus, dexamethasone.

Parbendazole Increases Both Apoptosis and Proliferation.

To determine how parbendazole may affect hMSC viability, we used FACS analysis to look at apoptosis and proliferation. Parbendazole increased apoptosis at day 5 and 8 of culture compared with control-treated hMSCs (44.8–58.5 vs. 15.8–17.9%, respectively; Fig. 2A). We also found that at days 5 and 8 of culture, parbendazole significantly increased proliferation compared with control-treatment (24.7–26.1 vs. 9.1–10.4% Ki67+, respectively; Fig. 2B). To verify the overall viability of hMSCs treated with parbendazole, we performed a PrestoBlue assay, a metabolism-based assay as a readout for cell viability. Parbendazole treatment led to a decrease in cell viability compared with both controls and dex-treated hMSCs starting at day 4 of culture (Fig. 2C). These findings reveal that parbendazole is capable of increasing both proliferation and apoptosis in hMSCs, and because apoptosis (i.e., increased cell death and decreased cell survival) exceeds proliferation conditions, the total accumulation of cells is reduced in parbendazole-treated cells.

Fig. 2.

Parbendazole treatment decreases cell viability by increasing apoptosis. (A) FACS assessment of hMSCs treated with control medium (white bar), 4 μM parbendazole (gray bar), or 0.1 μM dex (black bar) and stained with annexin to determine combined early and late apoptosis. (B) FACS assessment of hMSCs treated with control medium (white bar), 4 μM parbendazole (gray bar), or 0.1 μM dex (black bar) and stained with Ki67 to determine proliferation (n = 7). (C) Relative cell viability was assessed by PrestoBlue assay as represented by the relative fluorescence units (RFU). *P < 0.05, **P < 0.01, ***P < 0.001. Results are presented relative to control at each time point.

Parbendazole Induces Osteoblast Differentiation Independent of Glucocorticoid Receptor Signaling.

To determine if parbendazole induces osteogenic differentiation through direct GR–mediated stimulation of osteoblast marker genes, similar to dex, we performed quantitative gene expression analyses for known GR target genes following parbendazole treatment. Dex-induced osteoblast differentiation strongly up-regulated the GR target genes ZNF145, GILZ, and FKBP51 (up to 10,000-fold), whereas parbendazole treatment did not (Fig. 3 A–C), implicating that parbendazole acts independently of GR signaling. We then cultured dex- or parbendazole-treated hMSCs in combination with the GR antagonist mifepristone (RU486), which blocks signaling through the GR, and performed biochemical assays for osteoblast differentiation. Whereas mifepristone had no effect on induction of ALP activity (Fig. 3D) or mineralization (Fig. 3E) by parbendazole, it completely abolished dex-induc”ed increases in ALP activity (Fig. 3D) and mineralization (Fig. 3E). These results prove that parbendazole induces human osteoblast differentiation independent of GR signaling.

Fig. 3.

Parbendazole-induced osteoblast differentiation is independent of glucocorticoid receptor signaling. (A–C) Quantitative PCR results from hMSCs incubated with parbendazole and from control hMSCs either undifferentiated or differentiated with dex. Relative gene expression of direct glucocorticoid receptor signaling targets, ZNF145 (A), GILZ (B), and FKBP51 (C) of hMSCs treated with control medium (white bar), 4 μM parbendazole (gray bar), or 0.1 μM dex (black bar) for 24 h. Biochemical assays for ALP (day 6) (D) and mineralization (week 3) (E) of hMSCs treated with control medium (white bar), 4 μM parbendazole (gray bar), or 0.1 μM dex (black bar) with (patterned bars) or without (solid bars) the GR antagonist, mifepristone, throughout the culture period (n = 6). **P < 0.01, ***P < 0.001. Results are presented relative to control at each time point.

Parbendazole Inhibition of Microtubule Polymerization Is Required for Parbendazole-Induced Osteogenic Differentiation.

Parbendazole is known to be an inhibitor of microtubule formation (10), and we confirmed this in our hMSCs. In control (Fig. S2 A, D, and J) and dex-treated (Fig. S2 C, F, and L) cultures, both the actin and microtubule structures are similarly distributed throughout the cell, whereas in parbendazole-treated hMSCs, the microtubule structure is severely degraded, leaving only short strands of microtubule filaments surrounding the nucleus (Fig. S2 E and K). However, we observed numerous thick, crossing actin stress fibers present in cells treated with parbendazole (Fig. S2B), whereas in control-treated cells the actin filaments were primarily organized parallel to the cell axis (Fig. S2A). To determine whether degradation of the microtubule structure is required for parbendazole-induced osteogenic differentiation, we used the microtubule stabilizing agent paclitaxel (Taxol). Taxol completely abolished the parbendazole-induced mineralization of hMSCs (Fig. 4). These results clearly demonstrate that inhibition of microtubule formation is required for parbendazole to elicit osteogenic differentiation of hMSCs.

Fig. S2.

Parbendazole inhibits microtubule polymerization. hMSCs were incubated with control medium (A, D, G, and J), 4 μM parbendazole (B, E, H, and K), or 0.1 μM dex (C, F, I, and L) for 4 d, and then the actin and tubulin cytoskeleton were visualized by immunofluorescence microscopy using phalloidin and β-tubulin antibodies. Parbendazole treatment inhibits microtubule formation (E and K), with only small microtubules remaining perinuclear (white arrow). Crossing of the actin microfilaments can also be seen in parbendazole-treated cells (white stars in B). (Magnification: 630×.)

Fig. 4.

Parbendazole inhibition of microtubule polymerization is required for parbendazole-induced osteogenic differentiation. Mineralization in hMSCs treated with control medium (white bar) or 4 μM parbendazole (gray bar) in combination with (striped bars) or without (solid bars) the microtubule-stabilizing drug Taxol (n = 6). ***P < 0.001. Results are presented relative to control.

Parbendazole Increases Focal Adhesions.

Based on the evidence that cytoskeletal changes are induced by parbendazole during hMSC osteogenic differentiation, and previous evidence that cytoskeletal changes and integrin binding and signaling play a significant role in osteoblast differentiation (11–13), we investigated whether focal adhesions (FAs) are also affected by parbendazole. Analysis of immunofluorescent images show that FAs appear to be longer and more numerous following parbendazole treatment (Fig. S3 E, K, and N) compared with control treatment (Fig. S3 D, J, and M). Quantification of the FAs revealed that parbendazole treatment significantly increased the number of focal adhesions compared with both control and dex treatment (Fig. 5A) and increased their length compared with control (Fig. 5B) after 24 h of treatment.

Fig. S3.

Parbendazole affects focal adhesions. hMSCs were incubated with control medium (A, D, G, J, M), 4 μM parbendazole (B, E, H, K, N), or 0.1 μM dex (C, F, I, L, O) for 24 h, and actin microfilaments and FAs were visualized by immunofluorescence microscopy using phalloidin and vinculin antibodies. FAs are identified by the filled arrowheads. Crossing of the actin microfilaments can also be seen in parbendazole-treated cells (white stars in B). (Magnification 400×.) (Scale bar: 20 μM.)

Fig. 5.

Parbendazole increases number and length of focal adhesions. Quantification of the number (A) and length (B) of focal adhesions was performed on control, parbendazole, and dex-treated hMSCs (n = 28–30 cells). *P < 0.05, **P < 0.01, ***P < 0.001.

Parbendazole Increases BMP-2 Expression and Activity.

It has previously been demonstrated that microtubule inhibitors stimulate osteoblast differentiation and increase bone mass in mice through elevated levels of BMP-2 (14, 15). We show that parbendazole significantly increased BMP2 expression (Fig. 6A), whereas dex significantly inhibited BMP2 expression compared with control-treated cells. To determine BMP bioactivity stimulated by parbendazole, we used a C2C12-BRE-Luc reporter cell line (16, 17). Incubation with conditioned media from hMSCs treated with parbendazole for 48 h significantly increased luciferase activity (Fig. 6B), whereas conditioned medium from control and dex treatment had no effect. These results show that parbendazole stimulates BMP-2 signaling in hMSCs.

Fig. 6.

Parbendazole regulates BMP-2 expression and activity. (A) Gene expression of BMP2 after treating hMSCs for 24 h with control medium (white bar), 4 μM parbendazole (gray bar), or dex (black bars). (B) Luciferase reporter assay for BMP signaling reporter BRE-Luc of control medium (solid white bar), conditioned medium from hMSCs treated from control (white patterned bar), 4 μM parbendazole for 48 h (gray patterned bar), or 0.1 μM dex (black patterned bar) for 48 h, and the positive control, recombinant BMP-2 protein (black bar). (C) Pretreatment of hMSCs with the BMP inhibitor DMH1 (patterned bars) enhances mineralization. Relative mineralization in hMSCs cultures following control (white bar), DMH1 (white patterned bar), parbendazole (gray bar), and parbendazole + DMH1 (gray patterened bar) treatment. n = 6. ***P < 0.001 by one-way ANOVA. ^###P < 0.001 by two-way ANOVA. Results are presented relative to control.

Inhibiting BMP-2 Signaling Limits Parbendazole-Induced Osteogenic hMSCs Differentiation.

To determine whether BMP-2 is involved in the osteogenic effect of parbendazole, we used the BMP-specific antagonist DMH1 (18). These studies demonstrated significant interaction between parbendazole and DMH1 resulting in limitation of osteogenic differentiation following the cotreatment of parbendazole and DMH1 compared with parbendazole or DMH1 alone (Fig. 6C); this reveals that BMP-2 signaling is involved in the effect of parbendazole on osteoblast differentiation of hMSC. In the control condition, DMH1 induced mineralization in hMSCs cultures, which is consistent with the results of Rifas (19) showing that inhibition of BMP signaling by another BMP antagonist, noggin, resulted in osteogenic differentiation of hMSCs.

Discussion

Using the CMap, we were able to identify a compound with a highly significant positively correlating gene expression profile to that of differentiating human osteoblasts. As expected, the top of the list of compounds was dominated by glucocorticoids, including dexamethasone; however, the top resulting compound was found to be parbendazole. We confirmed the validity and power of the CMap approach by demonstrating that parbendazole, independent of an additional osteogenic stimulus such as dexamethasone, is able to stimulate human osteoblast differentiation as evidenced by increased ALP activity, mineralization, and up-regulation of genes known to be important in osteoblast differentiation and function. Mechanistically, we showed that the osteogenic effect of parbendazole does not occur through glucocorticoid receptor signaling, but rather via affecting microtubule formation and cytoskeletal organization. Overall, our findings demonstrate that by using the CMap, we identified a novel compound that independently stimulates human osteogenic differentiation.

The web-based resource CMap was created to identify compounds that induce a similar or opposite effect on the physiological processes or diseases of interest; as a consequence, it is a bioinformatics tool to identify novel applications for established drugs. The CMap uses several human tumor cell lines, including PC3, MCF7, and HL60, to generate gene expression profiles of more than 1,300 compounds. Interestingly, all of the top eight compound profiles most strongly correlating with our human osteoblast expression profile are derived from the PC3 cell line, a cell line originating from a bone metastasis of a prostate tumor (20). An explanation for this intriguing observation is lacking, but it is tempting to hypothesize that, based on our CMap results, there must be a resemblance in the genomic profile of both osteogenic hMSCs and the bone metastatic PC3 cells, which is potentially linked to both of their abilities to thrive in the bone environment. A prerequisite of the CMap approach for discovering bone anabolic compounds is that one has to assume similarities in how different cell types respond to the same compound regarding gene expression. However, the evidence presented here, and the growing evidence from other groups (5–9), strongly supports the concept that CMap is a potent tool for identification of compounds with medicinal benefit for a wide range of diseases.

Benzimidazoles, including parbendazole, are a class of compounds that consist of the fusion of benzene and imidazole rings and are primarily used as anthelminthics (21). The majority of these family members have a similar mechanism of action—namely, preventing tubulin polymerization through specific binding to tubulin (22, 23). Parbendazole is no exception: it binds the tubulin dimer with a mol:mol stoichiometry and microtubules withdraw from the peripheral area of the cell within 30 min of treatment with parbendazole (10). Our immunocytological findings in osteogenic hMSC are consistent with the findings by Havercroft and coworkers. Disassembly of microtubules in other cell types has been associated with blebbing of the plasma membrane, indicative of apoptosis. Our FACS results showing that parbendazole treatment increases the number of annexin V positive cells in our cultures supports the idea that microtubule inhibition leads to increased apoptosis. However, Havercroft et al. (10) also showed that after 1 d of parbendazole treatment, cells flattened out without any signs of plasma membrane blebbing, despite absent microtubules. This morphology correlates to our observation that a subset of parbendazole-treated hMSCs displays a flattened, widespread morphology. Interestingly, although we do see an increase in apoptosis, we also observe an increase in the proliferation marker Ki67 at day 5 and 8 in hMSCs treated with parbendazole, which indicates that there are at least two populations of MSC in the bone marrow with different sensitivity to parbendazole. One population, likely the true stem cells, is resistant to the apoptotic effects of parbendazole and are induced to proliferate and eventually differentiate and mineralize, and another population, the precommitted cells, undergoes apoptosis. Taking the hMSC FACs data together with the biochemical results in the SV-HFOs, we propose that parbendazole acts at the stage of lineage allocation in hMSCs.

Microtubules are highly dynamic cytoskeletal elements that undergo continuous assembly and disassembly to maintain normal function (24, 25); they play an essential role in a variety of cellular processes, including mitosis, cell motility, and intra- and intercellular trafficking (26), but are also taking part in various signaling pathways including those involving sonic hedgehog, Wnt, and MAPK (24, 27). Microtubules act as binding sites for a number of proteins and transcription factors, including RUNX2; and in the case of RUNX2, modulate transcriptional activity by acting as a shuttle between the nucleus and cytoplasm (28). It has been shown previously that Taxol treatment leads to depletion of nuclear levels of RUNX2 (28), and this could also be one explanation for our results demonstrating that Taxol pretreatment blocks the osteogenic effect of parbendazole. Chang et al. (29) observed that short-term treatment of MSCs with a related compound, nocodazole, increased cell contractility and cytoskeletal tension followed by enhanced osteoblast differentiation. Cytoskeletal tension rises, and perhaps even directly stimulates, osteoblast differentiation through changes in the actin cytoskeleton (13, 30–33). In addition, disruption of the tubulin organization also affects the kinetics of actin organization (13), indicating a regulatory role for microtubules in actin reorganization. Cytoskeletal rearrangements may also affect osteoblast differentiation through enhanced integrin signaling. FA complexes and integrins are known to directly interact with extracellular matrix proteins, including SPP1 and IBSP, which play a critical role in osteoblast survival and differentiation (11, 30, 34, 35). An increased number of FAs is proposed to increase adhesion to the extracellular matrix (35, 36), which may also contribute to osteogenic differentiation. The fact that parbendazole-treated hMSCs display a widespread, flattened morphology, prominent and rearranged actin fibers, increased focal adhesion complexes, and higher expression of SPP1 and IBSP, supports the idea that cytoskeletal rearrangements contribute to osteoblast differentiation through enhanced integrin signaling and/or matrix binding. Liu et al. (14) demonstrated a role for microtubule inhibitors as a bone anabolic drug by showing that stathmin, a ubiquitously expressed protein that inhibits microtubule assembly, promotes osteoblast differentiation and inhibits osteoclast activity; and mice lacking stathmin are osteopenic as a consequence of decreased osteoblast numbers and differentiation, and increased numbers of osteoclasts. Our results, demonstrating that changes in the balance between the tension forces of the microfilaments and the compression forces of the microtubules elicit changes in the cellular shape, function, and cell fate are in line with the cellular tensegrity model (37, 38) and with previous findings that changes in cell shape and cytoskeleton can strongly influence hMSC lineage decision-making (39).

There is growing evidence that microtubule inhibitors have a positive effect on bone formation and osteoblast differentiation through increased BMP signaling. Zhao et al. (15) showed that several types of microtubule inhibitors, including nocodazole and TN16, increased Bmp2 expression and ALP activity in murine osteoblasts; additionally, the authors showed that short-term administration of TN16 locally over the calvaria or systemically in mice lead to increased calvarial periosteal bone formation and trabecular volume in long bones, respectively. It was shown in murine osteoblasts that the osteogenic effect of the inhibition of microtubule assembly is due to increased cytoplasmic Gli2 protein concentration through disassociation of the microtubule Gli–Ci complex that would otherwise lead to proteasomal degradation of Gli2 (15). Gli2 enhances Bmp2 expression through BMP promoter binding (40). Bmp-2 is known to be a potent stimulator of osteoblast differentiation in murine cells (41, 42; reviewed by 43), but the evidence for BMP-2 effects on human osteogenic differentiation in vitro is conflicting, ranging from strongly enhancing osteogenic differentiation (44, 45) to having no positive effect (46, 47); this may be, in part, due to differences in the BMP receptor expression profiles by the various osteoblast precursors and hMSC donors (44, 47). Clinical trials looking at the efficacy of BMP-2 in fracture healing and fusions are mixed; two recent independent meta-analyses of human clinical trials comparing recombinant human BMP-2 to autologous bone grafts in spinal fusions found no significant difference in healing or pain (48, 49). In this study we found that parbendazole treatment induces both BMP2 expression and BMP activity, as evidenced by our gene expression and reporter assay data; in support of this, the BMP inhibitor DMH1 (18) seems to limit the parbendazole-induced osteoblast mineralization in our hMSC cultures but does not completely block it. Thus, our data confirms a parbendazole-induced human osteogenic pathway through inhibition of microtubule assembly and increased BMP-2 activity.

In summary, we conclude that the CMap identified compound parbendazole is a novel compound that stimulates human osteoblast differentiation in vitro. We have proven that parbendazole, independent of additional osteogenic stimulus, stimulates ALP activity and mineralization, and up-regulates genes important in osteoblast differentiation and extracellular matrix production in a subset of hMSCs resistant to its apoptotic effects. In line with previous reports of other microtubule inhibitors, the stimulatory effect of parbendazole is partly due to increased BMP-2 activity; however, additional cytoskeletal-driven mechanisms, such as cytoskeletal-associated proteins and transcription factors, and integrin signaling, are likely to be of equal or greater importance in stimulating osteogenic differentiation of hMSCs. Though we do not propose parbendazole itself to be the ideal bone anabolic drug for treatment of osteoporosis in humans, the evidence presented here significantly strengthens the concept that cytoskeletal changes strongly influence hMSC lineage allocation and osteoblast differentiation. We envisage that cytoskeletal manipulation, or the downstream processes that results from it, hold promise as novel antiosteoporotic treatments, but this will need to be studied further. We have also clearly demonstrated the power of the CMap as an effective tool to discover bone anabolic compounds.

Materials and Methods

Further details of materials and methods are included in SI Materials and Methods and Table S2.

Table S2.

Sequences of primer sets used for quantitative PCR in this study

Gene	Forward primer	Reverse primer	Pmol/reaction
GAPDH	CCGCATCTTCTTTTGCGTCG	CCCAATACGACCAAATCCGTTG	2.5
ALPL	TAAAGCAGGTCTTGGGGTGC	GGGTCTTTCTCTTTCTCTGGCA	2.5
SPP1	AGGCATCACCTGTGCCATAC	CACAGCATTCTGCTTTTCCTCA	2.5
IBSP	TGCCTTGAGCCTGCTTCC	GCAAAATTAAAGCAGTCTTCATTTTG	10
Probe FAM-TAMRA: CTCCAGGACTGCCAGAGGAAGCAATCA			2.5
ZNF145	GCGGTTCCTGGATAGTTTGC	TGATCACAGACAAAGGCTTTGG	10
Probe FAM-TAMRA: ATGCACTTACTGCGTCATTCAGCGGG			2.5
FKBP51	CGGAAAGGAGAGGGATATTCAA	TCTGCAGTCAAACATCCTTCCA	2.5
GILZ	GCACAATTTCTCCATCTCCTTCTT	TCAGATGATTCTTCACCAGATCCA	2.5
BMP2	ACGGACTGCGGTCTCCTAA	GGAAGCAGCAACGCTAGAAG	1.25

Most genes were detected using SYBR green; IBSP and ZNF145 PCRs were performed with a specific probe (FAM-TAMRA).

Cell Culture.

Human bone marrow-derived MSCs were cultured as described previously (50, 51).

Connectivity Map Query.

To find compounds that have similar gene expression patterns compared with our human osteogenic differentiation-induced gene expression patterns, we generated gene signatures from microarray gene expression analyses of bone marrow-derived hMSCs treated with dex to stimulate osteogenic differentiation. We identified the 100 most significantly (Z < 0.001) up- and down-regulated probes based on log ratio of gene expression during osteogenic differentiation at 6 h following the start of differentiation treatment, compared with time 0. We chose this time point to be able to temporally match it with the time of incubation of each compound within the CMap. Illumina probe identifiers were converted to Affymetrix probe identifiers, which are required for input into the CMap. After removal of duplicates, we ended up with 77 up-regulated genes and 66 down-regulated genes (Table S1) that we submitted simultaneously for our CMap query (build02; www.broadinstitute.org/cmap/). Each signature was queried against the CMap using the gene set enrichment analysis algorithm described by Lamb et al. (3).

Flow Cytometric Analysis of Proliferation and Apoptosis.

Apoptotic cells were determined based on staining with phycoerythrin (PE)-conjugated annexin V and 7-aminoactinomycin D (7AAD). Proliferating cells were identified as being Alexa 488-conjugated Ki67⁺.

PrestoBlue Assay for Cell Survival Analysis.

For analysis of cell survival we used the PrestoBlue cytotoxicity assay following the manufacturer’s protocol.

Immunocytochemistry.

Microtubules were visualized using mouse monoclonal anti–β-tubulin antibody. Focal adhesions were labeled with rabbit monoclonal ABfinity anti-Vinculin antibody. Quantitative morphometric analysis (length and number of focal adhesions per cell) was performed using ImageJ (NIH) as described previously (52).

BMP Reporter Assay.

We used a previously reported C2C12 cell line stably transfected with a reporter plasmid consisting of BMP-responsive elements from the Id1 promoter fused to a luciferase reporter gene (C2C12-BRE-Luc) to detect BMP activity (16, 17) following treatment with conditioned media from hMSCs exposed to parbendazole.

Statistics.

The data provided here are based on at least two independent experiments performed in at least triplicate. Values displayed are mean ± SEM. Significance was calculated using either the Student's t test, one-way ANOVA with Tukey’s or Dunn’s post hoc test, or two-way ANOVA where appropriate, using GraphPad prism 6.0. P < 0.05 was considered significant.

SI Materials and Methods

Materials.

MSCs were purchased from Lonza (PT-2501). αMEM was purchased from Gibco BRL, Life Technologies. Dexamethasone, β-glycerophosphate, mifepristone, paclitaxel (Taxol), and DMH1 were all purchased from Sigma. The calcium assay kit, alizarin Red S, ethanolamine, Triton X-100, mouse monoclonal anti–β-tubulin antibody, and oligonucleotide primer pairs were purchased from Sigma-Aldrich. TRIzol, PrestoBlue cytotoxicity assay reagent, and rabbit monoclonal ABfinity anti-Vinculin antibody were purchased from Invitrogen. Illumina Human HT-12 v3 BeadChip arrays, iScan, GenomeStudio V2010.1 (Gene Expression Module 1.6.0) were obtained from Illumina Inc. The 2100 Bioanalyzer was purchased from Agilent Technologies. Illumina TotalPrep RNA Amplification Kit was purchased form Ambion. Streptavidin Cy3 was purchased from GE Healthcare. BCA protein assay reagent A and B were purchased from Pierce. Victor2 plate reader and Steady Lite Plus Luciferase reagent was purchased from PerkinElmer Life and Analytical Science. All FACS antibodies, the Accuri C6 Personal Flow Cytometer, and BD Accuri C6 software analysis program were obtained from Beckon Dickinson. Alexa Fluor 488 goat anti-mouse IgG, rhodamine-conjugated phalloidin, and Alexa Fluor 488 goat anti-rabbit IgG were purchased from Molecular Probes. VectaShield mounting medium containing DAPI was from Vector Laboratories.

Cell Culture.

Human bone marrow-derived MSCs were cultured as described previously (50, 51). For osteogenic differentiation, MSCs were cultured in αMEM medium containing 10% (vol/vol) heat-inactivated FCS supplemented with 100 nM dexamethasone (dex) or with various concentrations (ranging from 0.1 to 4 μM) of parbendazole, both in combination with 10 mM β-glycerophosphate. Where applicable, hMSCs were treated with 10 μM mifepristone, 0.1 μM paclitaxel (Taxol), or 0.5 μM DMH1, a highly selective dorsomorphin analog, in β-glycerophosphate containing medium alone or 30 min before adding parbendazole or dex. Cell extracts were harvested at different time points during culture by scraping the cells in PBS/triton and storing at −80 °C for biochemical analyses or in TRIzol and stored at −20 °C for gene expression analyses. Alternatively, cells were fixed in 4% (vol/vol) phosphate-buffered paraformaldehyde for immunocytochemical procedures or 70% (vol/vol) ethanol for alizarin red staining.

Illumina Gene Chip-Based Gene Expression.

For analysis of whole human genome expression, we used Illumina Human HT-12 v3 BeadChip arrays. Human MSCs (in triplicate) were treated for 6 h with dexamethasone and next RNA was isolated as previously described (48). RNA integrity of isolated RNA was assessed by RNA 6000 Nano Assay on a 2100 Bioanalyzer. The Illumina TotalPrep RNA Amplification Kit was used for RNA amplification of each sample according to manufacturer’s instructions. In short, T7 oligo(dT) primer was used to generate single-stranded cDNA, followed by a second-strand synthesis to generate double-stranded cDNA. Biotin-labeled cRNA was synthesized using T7 RNA polymerase. The cRNA was column-purified and checked for quality by RNA 6000 Nano Assay. A total of 750 ng of cRNA was hybridized for each array with the standard Illumina protocol, using streptavidin-Cy3 for detection. Slides were scanned on an iScan and analyzed using GenomeStudio.

Microarray Analysis.

Background was subtracted from the raw data using GenomeStudio V2010.1 (Gene Expression Module 1.6.0), and data were processed using the Bioconductor R2.10.0 Lumi package (www.bioconductor.org) (53). Data were transformed by variance stabilization and quantile normalization. Probes that were detected at least once in the experiments (Illumina detection, P < 0.01) were considered to be expressed and were further analyzed. Differentially expressed probes were identified using Bioconductor Package Limma (www.bioconductor.org) (54) with P values adjusted to reduce the false discovery rate (P < 0.001).

Alkaline Phosphatase, Mineralization, and Protein Assays.

ALP and calcium measurements were performed as described previously (55). Briefly, ALP activity is determined by an enzymatic reaction, where the ALP-mediated conversion of para-nitrophenylphosphate (pNPP) to paranitrophenol (PNP) during 10 min at 37 °C is measured. For calcium measurements, cell lysates were incubated overnight with 0.24 M HCl at 4 °C. Calcium content was determined colorimetrically with a calcium assay kit according to the manufacturer’s description. ALP results were adjusted for protein content of the cell lysates. For protein measurement, 200 µL of working reagent was added to 10 µL of sonicated cell lysate. The mixture was incubated for 30 min at 37 °C, cooled down to room temperature (RT), and absorbance measured at 595 nm. All measurements were performed using a Victor2 plate reader. Staining for mineralization was performed as described previously (55). Briefly, cells were fixed with 70% (vol/vol) ethanol and, after washing, stained for 10–20 min with alizarin Red S solution.

Quantification of mRNA Expression.

Cultures continuously treated with control, 100 nM dex, or 4 μM parbendazole were harvested during the early or middle differentiation period (6 h, 24 h, 7 d). RNA isolation, cDNA synthesis, and PCR reactions were performed as described previously (48). Oligonucleotide primer pairs were designed to be either on exon boundaries or spanning at least one intron (Table S2). Gene expressions were corrected for the housekeeping gene GAPDH.

Flow Cytometric Analysis of Proliferation and Apoptosis.

To measure apoptosis or proliferation, hMSCs were treated with 100 nM dex or 4 μM parbendazole for 1, 5, or 8 d (n = 7).

To measure apoptosis, cells were trypsinized, collected in media, and washed once in PBS before being resuspended in 50 μL binding buffer and 2.5 μL annexin V and 2.5 7AAD. Cells were incubated in the dark for 15 min μL before an additional 75 μL binding buffer was added and cell were analyzed by flow cytometry. The percentage of apoptotic cells was measured by flow cytometry after staining with PE-conjugated annexin V and 7AAD (annexin V⁺, 7AAD⁺). We scored viable cells as those that are negative for annexin V and 7AAD. Irrelevant isotype-matched Ig was used as a control. Fold induction of apoptosis was calculated from the increase in the percentage of apoptotic cells between the treated and untreated samples. The amount of necrotic cells (annexin V^–, 7AAD⁺) was always minimal.

For analysis of proliferation, hMSCs were trypsinized, collected in media, and after centrifugation, 5 mL of ethanol was added dropwise to the cells, after which they were incubated at −20 °C for 30 min. Cells were then centrifuged and resuspended in 1 mL PBS plus 1% BSA before being centrifuged again. Supernatant was discarded and 55 μL PBS containing 5% (vol/vol) serum plus 2.5 μL Pe-Cy7–conjugated Ki67 antibody was added. Cells were incubated 30 min in the dark at RT. After incubation, 1 mL PBS containing 5% (vol/vol) serum was added and centrifuged again. Supernatant was discarded and the pellet was dissolved in 100 μL PBS containing 5% (vol/vol) serum. Proliferating cells were identified as being Ki67 positive. Stained cells were analyzed using a BD Biosciences Accuri C6 Personal Flow Cytometer and analyzed with BD Accuri C6 software analysis program. For each sample, a minimum of 10,000 events was collected.

PrestoBlue Assay for Cell Survival Analysis.

For analysis of cell survival, hMSCs were cultured in 96-well tissue culture plates. Following incubations with control media, 100 nM dex or 4 μM parbendazole, cell viability was assayed at 1, 5, 8, and 14 d (n = 6) using a PrestoBlue cytotoxicity assay following the manufacturer’s protocol. Briefly, 10 μL of PrestoBlue reagent was added to each well and after a 1-h incubation the plate fluorescence was measured at excitation 530 nm/emission 590 nm wavelength using a Victor2 plate reader. Date was corrected for background fluorescence of PrestoBlue in media alone.

Immunocytochemistry.

Cells were fixed with 4% (vol/vol) paraformaldehyde in PBS for 15 min at RT, washed in PBS, and excess aldehyde quenched with 10 mM ethanolamine in PBS for 5 min. Cells were then permeabilized with 0.5% Triton X-100 in PBS for 10 min and blocked for 30 min at RT in PBS supplemented with 1.5% (vol/vol) BSA and 0.02% Triton X-100. For visualization of microtubules, cells were labeled for 1 h with mouse monoclonal anti–β-tubulin antibody at 1:60 dilution at RT, followed by secondary Alexa Fluor 488 goat anti-mouse IgG at 1:400 dilution for a total of 1 h, with the addition of rhodamine-conjugated phalloidin at 1:100 dilution for the last 20 min. For visualization and quantification of focal adhesions, cells were labeled for 1 h with rabbit monoclonal ABfinity anti-Vinculin antibody at 1:200 dilution at RT, followed by secondary Alexa Fluor 488 goat anti-rabbit IgG at 1:400 dilution for a total of 1 h, with the addition of rhodamine-conjugated phalloidin at 1:100 dilution for the last 20 min. Slides were mounted using VectaShield mounting medium containing DAPI, and pictures were taken on a Zeiss Axiovert 200 MOT microscope. Quantitative morphometric analysis (length and number of focal adhesions per cell) was performed using image analysis software ImageJ (NIH) as described previously (52).

BMP Reporter Assay.

C2C12-BRE-Luc cells were seeded into white-walled 96-well culture plates at a density of 40,000 cells/well in 50 μL of media. After 30 min, 50 μL of conditioned media from hMSCs treated with parbendazole (4 μM), dex (0.1 μM), or control media was then added to the cells. After 24 h, the media was removed from the cells and 100 μL of the combined lysis buffer and luciferin reagent (Steady Lite Plus) was added to each well. The plates were shaken for 15 min in the dark and then luminescence was quantified using a Victor2 plate reader.

Cell Culture of SV-HFOs.

SV-HFO cells (56) were cultured as described previously (50).