Identifying an ovarian cancer cell hierarchy regulated by bone morphogenetic protein 2

Significance

Significant controversy persists regarding a hierarchical vs. stochastic model of cancer. Using a microfluidic single-cell culture device, we define for the first time, to our knowledge, the differentiation capacity of primary human ovarian cancer cells. We demonstrate that ovarian cancer follows a hierarchical model with rare stochastic events. Defining the differentiation capacity allowed us to explain apparently paradoxical actions of bone morphologenetic protein 2 (BMP2); BMP2 suppresses growth in vitro by suppressing bulk cell proliferation, but promotes growth in vivo by promoting cancer stem-like cell (CSC) expansion. This work supports BMP2 signaling as a critical therapeutic target regulating ovarian CSC growth.

Abstract

Whether human cancer follows a hierarchical or stochastic model of differentiation is controversial. Furthermore, the factors that regulate cancer stem-like cell (CSC) differentiation potential are largely unknown. We used a novel microfluidic single-cell culture method to directly observe the differentiation capacity of four heterogeneous ovarian cancer cell populations defined by the expression of the CSC markers aldehyde dehydrogenase (ALDH) and CD133. We evaluated 3,692 progeny from 2,833 cells. We found that only ALDH⁺CD133⁺ cells could generate all four ALDH^+/−CD133^+/− cell populations and identified a clear branched differentiation hierarchy. We also observed a single putative stochastic event. Within the hierarchy of cells, bone morphologenetic protein 2 (BMP2) is preferentially expressed in ALDH⁻CD133⁻ cells. BMP2 promotes ALDH⁺CD133⁺ cell expansion while suppressing the proliferation of ALDH⁻CD133⁻ cells. As such, BMP2 suppressed bulk cancer cell growth in vitro but increased tumor initiation rates, tumor growth, and chemotherapy resistance in vivo whereas BMP2 knockdown reduced CSC numbers, in vivo growth, and chemoresistance. These data suggest a hierarchical differentiation pattern in which BMP2 acts as a feedback mechanism promoting ovarian CSC expansion and suppressing progenitor proliferation. These results explain why BMP2 suppresses growth in vitro and promotes growth in vivo. Together, our results support BMP2 as a therapeutic target in ovarian cancer.

The cancer stem cell (CSC) hypothesis postulates that cancers are composed of hierarchically organized cell subpopulations with distinct phenotypes and tumorigenic capacities. CSCs are rare cells within the tumor that can not only proliferate to maintain tumorigenic potential, but also asymmetrically divide to generate other, phenotypically distinct cells with diminished tumorigenic potential. CSCs have therefore been suggested as a source of disease recurrence; residual CSCs after primary therapy can proliferate and differentiate to recreate a tumor (1).

Although there is very strong support in the literature for the existence of cancer stem-like cells, the CSC hypothesis remains controversial. Supporting the CSC hypothesis, murine studies using lineage-tracing experiments identified subsets of cells with CSC-like characteristics in tumor precursor lesions (2, 3). Furthermore, in overt malignancy, a subset of rare chemotherapy-resistant CSC-like cells were found to be responsible for hierarchical tumor cell regrowth after chemotherapy treatment (4). However, other studies indicate stochastic events in which non–stem-like cells can acquire stem-like characteristics under the influence of the right environmental/genetic stresses (5–8). Studies to define a hierarchical vs. stochastic model with human tumor specimens are complicated by a small but real contamination rate associated with both magnetic- and FACS-based cell purification.

Little is known about a potential ovarian cancer cell hierarchy. Although some reports suggest that ovarian cancer may not follow a hierarchical model (9), numerous potential CSC populations, defined by various cell markers, have been reported (1, 10). Aldehyde dehydrogenase enzymatic activity (ALDH) and the stem cell marker CD133, either alone or in combination, are perhaps the best supported ovarian CSC (OvCSC) markers (11–16). We recently reported that ALDH and CD133 can be used to define distinct heterogeneous populations of ovarian cancer cells (17). We found that both ALDH⁺CD133⁺ cells and ALDH⁺CD133⁻ primary human ovarian tumor cells can act as OvCSCs, with the ability to initiate and passage tumors in mice that recapitulate the primary human tumor. ALDH⁺CD133⁺ OvCSCs demonstrated the greatest engraftment potential and generated tumors within 2–4 months whereas ALDH⁺CD133⁻ primary human ovarian cancer cells have lower engraftment potential and longer growth requirements (6–12 months). ALDH⁻CD133⁻ cells from primary samples are unable to initiate tumors. Clinical observations also support ALDH and CD133 as identifying at least one population of OvCSCs; the presence of increased numbers of ALDH⁺CD133⁺ cells in primary tumor debulking samples is associated with poor patient outcome (11); and ALDH and CD133 (and CD44) are enriched in patient tumors and human PDX tumors immediately after chemotherapy (14, 18). ALDH and CD133 (along with LGR5) were also found to identify a population of normal ovarian stem cells, suggesting a potential role of ALDH⁺CD133⁺ cells in the ovarian cancer cell of origin (19).

Based on these data, we hypothesized that ALDH and CD133 can be used to define a hierarchy of OvCSC differentiation; highly tumorigenic ALDH⁺CD133⁺ CSCs give rise to somewhat less tumorigenic ALDH⁺CD133⁻ CSC/progenitors, which in turn give rise to less tumorigenic/nontumorigenic ALDH⁻CD133⁻ cells. Here, we report a using a microfluidic device (20) to directly interrogate the differentiation potential of the distinct ALDH^+/−CD133^+/− cell populations in ovarian cancer and to define a branched differentiation lineage. We show that BMP2 is differentially expressed in the distinct ovarian cancer cell populations, with low expression in ALDH⁺CD133⁺ cells and highest expression in ALDH⁻CD133⁻ cells. Furthermore, we find that BMP2 promotes expansion of the ALDH⁺CD133⁺ CSC cell population and inhibits the proliferation of progenitors. In vivo treatment with BMP2 results in increased tumor growth and chemotherapy resistance. Inhibition of BMP2 signaling with Noggin or BMP2 knockdown is associated with a decrease in the number of ALDH+CD133+ and ALDH-CD133+ cells, reduced tumor initiation capacity, and a reduction in tumor growth.

Results

Direct Observation of the Differentiation Potential of Ovarian Cancer Cell Populations.

We sought to test the differentiation capacity of each of the four ALDH^+/−CD133^+/− cell populations (ALDH⁺CD133⁺, ALDH⁻CD133⁺, ALDH⁺CD133⁻, and ALDH⁻CD133⁻). We used a microfluidic, single-cell culture method in which FACS-associated cell contamination could be excluded (21) (Fig. S1). The four ALDH^+/−CD133^+/− cell populations from three primary patient specimens and three ovarian cancer cell lines (A2780, OVCAR8, and PEO1) were FACS isolated and loaded as single cells into separate devices. Once loaded, the identity of each cell was confirmed using in situ ALDEFLUOR and CD133 immunofluorescent microscopy within the microfluidics chamber (Fig. 1A, Captured Cell). Cells were cultured within the device for 72 h (cell lines) or 120 h (primary cells) and then relabeled and reevaluated for ALDH and CD133 expression in resultant daughter cells. Cell lines were >95% viable, and, from 1,594 evaluable cells (cells that underwent at least one cell division), we observed 2,343 resultant daughter cells (Table 1). Primary cells had lower viability (∼60%) in both traditional and microfluidic cultures. Primary cells also had reduced proliferation rates, with a large number of nondividing cells. From primary cells, there were 1,239 evaluable cells, and we evaluated 1,347 daughter cells.

Fig. S1.

Hydrodynamic single-cell capture scheme. (A) Overall microwell design consists of a microchamber with two fluidic paths A and B. In its open state, the resistance of path A, Ra, is significantly lower than the resistance of path B, Rb. (B) When a cell flows into the chamber, it preferentially travels path A due to its low resistance. However, because the width of path A is smaller than the diameter of the cells, they are sterically captured. (C and D) Subsequent cells will travel the bypass path, path B, due to the increase in Ra with capture. (E) The cell suspension can be changed to fresh media, and the cells captured in each well may proliferate within the chamber.

Fig. 1.

Differentiation potential of different ovarian cancer cell populations. (A, i–iv) Representative immunofluorescent imaging demonstrating confirmed identity of the indicated captured cell types and their resultant progeny from different observed cell divisions. CD133 is labeled in red, and ALDH activity is green; thus, ALDH⁺CD133⁺ cells are yellow, ALDH⁺CD133⁻ cells are green, ALDH⁻CD133⁺ cells are red, and ALDH⁻CD133⁻ cells are gray. (A, v) Single observed putative stochastic event where an ALDH⁺CD133⁻ cell produces a daughter cell that expressed CD133. (B) Bar graph summary of percentage of progeny from ALDH⁺CD133⁺, ALDH⁺CD133⁻, ALDH⁻CD133⁺, and for ALDH⁻CD133⁻ cells from both A2780 cells (i) and primary ovarian cancer patient cells (ii) evaluated in the microfluidic device. Results are from four replicate devices for each cell population in two independent experiments. Error bars indicate SEMs.

Table 1.

Summary of cell divisions observed in the heterogeneous cell populations from ovarian cancer cell lines and primary patient samples

Initial cell loaded	Treatment	No. of cells loaded	No. of daughter cells	Daughter cell type
				ALDH+CD133+	ALDH+CD133−	ALDH−CD133+	ALDH−CD133−
Cell line data (PEO1, OVCAR8, A2780)
ALDH+CD133+	Control	273	331	174 (52%)	95 (29%)	32 (10%)	30 (9%)
	BMP2	411	433	322 (75%)	65 (15%)	36 (8%)	10 (2%)
ALDH−CD133+	Control	122	312			207 (66%)	105 (34%)
	BMP2	290	591			434 (73%)	157 (27%)
ALDH+CD133−	Control	67	177		119 (67%)		58 (33%)
	BMP2	203	226		155 (69%)		71 (31%)
ALDH-CD133-	Control	56	130				130 (100%)
	BMP2	172	143				143 (100%)
ALDH+CD133+	Control	328	394	186 (47%)	76 (19%)	85 (22%)	47 (12%)
	BMP2	312	357	238 (66%)	49 (14%)	49 (14%)	21 (6%)
Patient data (n = 3)
ALDH−CD133+	Control	112	133			88 (66%)	45 (34%)
	BMP2	107	125			80 (64%)	45 (36%)
ALDH+CD133−	Control	100	109	1 (1%)	66 (61%)		42 (38%)
	BMP2	83	77		50 (65%)		27 (35%)
ALDH−CD133−	Control	94	91				91 (100%)
	BMP2	103	61				61 (100%)

Numbers highlighted in bold demonstrate statistically significant differences between control and BMP2 treatment. ALDH⁻CD133⁻ cells observed in ALDH⁺CD133⁺-initiated wells were the result of “daughter” cell divisions and were not observed to come directly from ALDH⁺CD133⁺ cells. Empty cells indicate that no daughter cells of the indicated type were observed.

For all samples, ALDH⁺CD133⁺ cells were capable of giving rise to all four ALDH^+/−CD133^+/− cell populations: ALDH⁺CD133⁺ cells demonstrated divisions resulting in (i) two ALDH⁺CD133⁺ cells, (ii) asymmetric divisions (relative to ALDH and CD133 expression) resulting in an ALDH⁺CD133⁺ cell and an ALDH⁺CD133⁻ cell, or (iii) an asymmetric division yielding an ALDH⁺CD133⁺ cell and an ALDH⁻CD133⁺ cell [Fig. 1 A, i and B (aggregate)] (individual sample results in Fig. S1). ALDH⁻CD133⁻ cell progenitors were observed in wells initially seeded by ALDH⁺CD133⁺ cells; however, we never observed a direct division of an ALDH⁺CD133⁺ cell into an ALDH⁻CD133⁻ cell, suggesting that these cells were the result of a division of either ALDH⁺CD133⁻ or ALDH⁻CD133⁺ progenitor cells (Fig. 1 A, ii–iii). ALDH⁻CD133⁺ cells could (i) divide to yield two ALDH-CD133+ cells or (ii) asymmetrically divide to produce ALDH⁻CD133⁺ and ALDH⁻CD133⁻ cells (Fig. 1 A, iii and B and Fig. S1). We never observed the ability of an ALDH⁻CD133⁺ cell to produce ALDH⁺ cells. Similarly, ALDH⁺CD133⁻ cells could (i) divide to yield two ALDH+CD133- cells or (ii) undergo an asymmetric division to produce ALDH⁺CD133⁻ cells and ALDH⁻CD133⁻ cells (Fig. 1 A, ii and B and Fig. S1). Interestingly, from one primary sample, we observed one putative “stochastic” event: An ALDH⁺CD133⁻ cell underwent a division that produced an ALDH⁺CD133⁺ cell and an ALDH⁺CD133⁻ cell (Fig. 1 A, v). Finally, ALDH⁻CD133⁻ cells were observed only to expand and did not produce any cells expressing ALDH or CD133 (Fig. 1 A, iv and B).

Differential Expression of BMP2 in Distinct Ovarian Cancer Cell Populations.

BMP2 is expressed by ovarian cancer cells (22) and can impact tumor “stemness” (23). Exactly how BMP2 impacts stemness is unclear. We FACS isolated the four ALDH^+/−CD133^+/− cell populations from two ovarian cancer cell lines and two primary patient samples and performed quantitative real-time PCR (qRT-PCR) to characterize BMP2 mRNA expression in the different populations of ovarian cancer cells. BMP2 mRNA was undetectable/detected at low levels in the ALDH⁺CD133⁺ cells and most highly detected in ALDH⁻CD133⁻ cells (Fig. 2 A, i and ii). Differential expression of BMP2 was also observed between ALDH⁻ and ALDH⁺ cells from cell lines in which CD133 expression is not detectable (Fig. 2 A, iii). Similarly, BMP2 protein was found to be differentially expressed using both Western blotting (Fig. 2B) and ELISA (Fig. 2C). We then used coimmunofluorescence with ALDH or CD133 and BMP2 and Cytokeratin-19 in to evaluate BMP2 expression in primary human tumor debulking specimens. BMP2 protein was generally expressed in ALDH⁻ and CD133⁻ cytokeratin⁺ tumor cells in human tumor samples (Fig. 2D). Lastly, to confirm BMP2-mediated signaling in ovarian cancer cells, we evaluated p-SMAD 1/5 levels in the presence or absence of BMP2. As previously reported, BMP2 treatment of A2780, SKOV3, and primary ovarian cancer cells was associated with a clear increase in phosphorylation of p-SMAD 1/5 (Fig. S2A).

Fig. 2.

BMP2 is differentially expressed in the different ovarian cancer cell populations. (A) qRT-PCR confirmation of differential BMP2 mRNA expression in the ALDH^+/−CD133^+/− and ALDH^+/− cell populations in both human ovarian cancer cell lines (i and iii) and patient samples (ii). (B) Western blot demonstrating the expression of BMP2 in ALDH⁺CD133⁻, ALDH⁻CD133⁺, and ALDH⁻CD133⁻ cells from both A2780 cells and primary ovarian cancer patient cells (PT224). (C) BMP2 expression from ALDH^+/−CD133^+/− cell populations as determined by ELISA. (D) Coimmunofluorescence of BMP2, ALDH and CK8 or BMP2, CD133 and CK8 demonstrating that BMP2 expression is primarily in ALDH⁻ and CD133⁻ cancer cells, respectively. *P < 0.01, from two independent assays.

Fig. S2.

Ovarian cancer cells respond to BMP2. (A) Western blot demonstrating that BMP2 treatment of the indicated ovarian cancer cells results in increased phosphorylation of SMAD 1/5. (B) Cell counts of A2780 and SKOV3 cells treated with BMP4 demonstrating that BMP4 reduces A2780 cell counts ∼25% but has no impact on SKOV3 cells. (C) FACS analysis of tumors initiated by RFP-labeled A2780 scrambled shRNA -RFP control cells and tumors initiated by WT ALDH⁺CD133⁺ cells coinjected with ALDH-CD133- BMP2shRNA-RFP, demonstrating loss of RFP in the latter group.

Differential Impact of BMP2 on the Different Ovarian Cancer Cell Populations.

Given the differential expression of BMP2 in the different ovarian cancer cell populations, we treated A2780 cells or primary patient cells with BMP2 or Noggin (BMP2/4 inhibitor). Surprisingly, both BMP2 and Noggin treatment restricted cancer cell growth in vitro (Fig. 3A). However, BMP2 treatment of primary cells in a tumor sphere assay was associated with a two- to threefold increase in the numbers of ALDH⁺CD133⁺ cells (Fig. 3B). Treatment with BMP4, a close family member, weakly impacted A2780 cells and had no clear impact on SKOV3 cells (Fig. S2B). Although these experiments do not exclude an important contribution of BMP4 to ovarian cancer cell growth, we chose to focus the remaining experiments on BMP2.

Fig. 3.

BMP2 promotes ALDH⁺CD133⁺ cell expansion and suppresses progenitor cell growth. (A) Total cell numbers of A2780 and primary patient cells after 7 d of treatment with BMP2 or Noggin. (B) FACS analysis of primary human ovarian cancer associated ascites tumor spheres cultured in the presence or absence of BMP2, demonstrating that BMP2 increases the percentage of ALDH⁺CD133⁺ cells. (C) Total number of cells generated from the isolated ALDH⁺CD133⁺, ALDH⁺CD133⁻, ALDH⁻CD133⁺, and ALDH⁻CD133⁻ cells after 7 d of growth with or without BMP2. (D) Summary of symmetric and asymmetric divisions of A2780 and primary ovarian cancer cells in the presence and absence of BMP2, demonstrating that BMP2 increases symmetric division of ALDH⁺CD133⁺ cells. For A–C, error bars represent SD from at least two independent experiments compared using Student’s t test. For D, error bars represent SEMs with all samples evaluated in four replicated devices in at least two experiments. P value determined using ANOVA.

We next FACS purified the four ALDH^+/−CD133^+/−A2780 cell populations and grew them in the presence/absence of BMP2. In the absence of BMP2, the ALDH⁻CD133⁺ and ALDH⁻CD133⁻ cells were the most proliferative whereas ALDH⁺CD133⁺ and ALDH⁺CD133⁻ grew the slowest (Fig. 3C). BMP2 treatment increased the proliferation rate of ALDH⁺CD133⁺ cells by 25% whereas the growth of ALDH⁺CD133⁻, ALDH⁻CD133⁺, and ALDH⁻CD133⁻ cells decreased 2.2-, 4.4-, and 6.9-fold, respectively.

We then evaluated the impact of BMP2 on isolated single cells from the four ALDH^+/−CD133^+/− cell populations from the same three primary patient samples and three cell lines as above. In both primary patient cells and cell lines, BMP2 treatment of ALDH⁺CD133⁺ cells led to a statistically significant 1.5-fold (range 1.2- to 2.2-fold) increase in the number of ALDH⁺CD133⁺ daughter cells (Fig. 3D and Table 1, aggregate data; and Tables S1 and S2 and Fig. S3, individual sample data). BMP2 treatment was also associated with a strong suppression of the proliferation of ALDH⁻CD133⁻ cells in cell lines and a nonstatistically significant decrease ALDH⁻CD133⁻ cell proliferation from primary samples (Table S3). These data demonstrate that BMP2 promotes expansion of the ALDH⁺CD133⁺ CSC cell population while suppressing the proliferation of bulk ALDH⁻CD133⁻ cells.

Table S1.

Summary of cell divisions observed in the heterogeneous cell populations from three ovarian cancer cell lines

Cell type loaded	Treatment	Daughter cells
		ALDH+CD133+	ALDH+CD133−	ALDH−CD133+	ALDH−CD133−
PEO1 cell line
ALDH+CD133+	Control	33.33%	19.05%	23.81%	23.81%
	BMP2	76.43%	8.57%	10.71%	4.29%
ALDH−CD133+	Control			59.63%	40.37%
	BMP2			74.39%	25.61%
ALDH+CD133−	Control		69.39%		30.61%
	BMP2		75.21%		24.79%
ALDH−CD133−	Control				100.00%
	BMP2				100.00%
OVCAR data
ALDH+CD133+	Control	58.82%	28.51%	7.24%	5.43%
	BMP2	73.91%	15.94%	8.21%	1.93%
ALDH−CD133+	Control			82.05%	17.95%
	BMP2			79.37%	20.63%
ALDH+CD133−	Control		80.56%		16.67%
	BMP2		74.36%		25.64%
ALDH−CD133−	Control				100.00%
	BMP2				100.00%
A2780 data
ALDH+CD133+	Control	44.12%	35.29%	8.82%	11.76%
	BMP2	72.09%	23.26%	4.65%	0.00%
ALDH−CD133+	Control			64.38%	35.62%
	BMP2			50.00%	50.00%
ALDH+CD133−	Control		60.22%		39.78%
	BMP2		54.29%		45.71%
ALDH−CD133−	Control				100.00%
	BMP2				100.00%

Cell population loaded into the device, treatment condition, and total number of cells loaded (evaluable cells) are indicated in the left columns. Right columns indicate absolute number of daughter cells observed and percentage of each daughter cell type observed from the initial cell population. Numbers highlighted in bold demonstrate statistically significant differences between control and BMP2 treatment. Empty cells indicate that no daughter cells of the indicated type were observed.

Table S2.

Summary of ovarian cancer cells observed in the heterogeneous cell populations from three primary ovarian cancer specimens

Cell type loaded	Treatment	Daughter cells
		ALDH+CD133+	ALDH+CD133−	ALDH−CD133+	ALDH−CD133−
Patient sample 1
ALDH+CD133+	Control	40.63%	12.50%	34.38%	12.50%
	BMP2	52.73%	12.73%	21.82%	12.73%
ALDH−CD133+	Control			50.00%	50.00%
	BMP2			48.15%	51.85%
ALDH+CD133−	Control	2.78%	72.22%		25.00%
	BMP2		66.67%		33.33%
ALDH−CD133−	Control				100.00%
	BMP2				93.33%
Patient sample 2
ALDH+CD133+	Control	49.62%	14.29%	27.82%	8.27%
	BMP2	65.18%	11.61%	18.75%	4.46%
ALDH−CD133+	Control			77.14%	22.86%
	BMP2			72.55%	27.45%
ALDH+CD133−	Control		47.22%		52.78%
	BMP2		59.26%		40.74%
ALDH−CD133−	Control				100.00%
	BMP2				100.00%
Patient sample 3
ALDH+CD133+	Control	46.72%	23.14%	16.16%	13.97%
	BMP2	71.58%	15.26%	8.42%	4.74%
ALDH−CD133+	Control			67.14%	32.86%
	BMP2			63.83%	36.17%
ALDH+CD133−	Control		62.16%		37.84%
	BMP2		69.57%		30.43%
ALDH−CD133−	Control				100.00%
	BMP2				100.00%

Cell population loaded into the device and treatment condition are indicated in the left columns. Right columns indicate percentage of each daughter cell type observed from the initial cell population. Numbers highlighted in bold demonstrate statistically significant differences between control and BMP2 treatment. Empty cells indicate that no daughter cell of the indicated type were observed.

Fig. S3.

Bar graph summary of percentage of progeny from single (i) ALDH⁺CD133⁺, (ii) ALDH⁺CD133⁻, (iii) ALDH⁻CD133⁺, and (iv) for ALDH⁻CD133⁻ cells from three cell lines (Left) and three primary ovarian cancer patient specimens (Right). Each sample was assayed in four devices in duplicate. Error bars represent an SE.

Table S3.

Percentage of control and BMP2 treated cells loaded into microfluidic culture that were viable yet demonstrated no division during the observation period

Cell type loaded	Treatment	Cell lines			Primary ovarian cancer cells
		Evaluable cells loaded	Percent nondividing cells	P value	Evaluable cells loaded	Percent nondividing cells	P value
ALDH+CD133+	Control	273	49	0.24	328	45	0.84
	BMP2	411	60		312	48
ALDH−CD133+	Control	122	32	0.10	112	53	0.93
	BMP2	290	44		107	54
ALDH+CD133−	Control	67	33	0.01	100	40	0.91
	BMP2	203	54		83	41
ALDH−CD133−	Control	56	41	0.01	94	60	0.95
	BMP2	172	63		103	61

The impact of BMP2 on Tumor Initiation and Growth.

To confirm the biologic impact of BMP2-mediated CSC expansion, we performed limiting dilution tumor initiation assays. Ten or 100 control, BMP2-treated, or noggin-treated cells were injected into the bilateral flanks of nonobese diabetic (NOD)-SCID mice (n = 5 animals per group in two independent experiments). Although BMP2 treatment was not associated with a statistically significant increase in tumor initiation rates, for both A2780 and SKOV3 cells, Noggin treatment was associated with a statistically significant decrease in tumor initiation as determined by extreme limiting dilution analysis (Fig. 4A).

Fig. 4.

BMP2 inhibition or knockdown suppresses tumor growth in vivo. (A) Number of tumors initiated from 10 or 100 A2780 or SKOV3 cells after treatment with the indicated compounds. (B) Tumor weights from A2780 (i) and SKOV3 (ii) tumors mock treated or treated with BMP2 or Noggin (n = 10 per group). (C, i) qRT-PCR and Western blot demonstrating successful knockdown of BMP2 expression in A2780 cells, (ii) FACS analysis of ALDH and CD133 expression in control and BMP2 knockdown cells, (iii) cell growth curves of A2780 and two independent BMP2-shRNA knockdown A2780 lines, (iv) tumor weights from A2780 control and independent BMP2-shRNA expressing A2780 lines (n = 16 controls, n = 8 for each BMP2 knockdown) in duplicate, and (v) tumor weights and initiation rates of tumors initiated with the indicated cell population/s (n = 10 per group) from WT and BMP2-shRNA cells.

We also examined the impact of BMP2 or the BMP2/4 inhibitor Noggin on SKOV3 and A2780 cell tumor growth in NOD-SCID mice. BMP2 treatment significantly increased tumor growth for both A2780 and SKOV3 (Fig. 4B) tumors whereas BMP2 inhibition with Noggin significantly reduced tumor growth (Fig. 4 B, i and ii). To specifically assess the contribution of tumor cell BMP2 production, we used shRNA to knockdown BMP2 production in A2780 cells. qRT-PCR evaluation of mRNA levels and Western blot analysis of protein levels confirmed four- to fivefold knockdown of BMP2 (Fig. 4 C, i). FACS evaluation of BMP2 knockdown cells demonstrated an eightfold to 10-fold decrease in ALDH⁺CD133⁺ cells and ALDH⁻CD133⁺ and a modest increase in ALDH⁺CD133⁻ cells (Fig. 4 C, ii). In vitro, BMP2shRNA cells demonstrated growth rates similar to those of controls (Fig. 4 C, iii). However, when grown in vivo, BMP2 knockdown tumors demonstrated significantly slower growth (Fig. 4 C, iv). Finally, to directly show the impact of ALDH⁻CD133⁻ cell-produced BMP2 on the growth of ALDH⁺CD133⁻ cells, we mixed 50 ALDH⁺CD133⁺ cells with either 250 ALDH⁻CD133⁻ WT or 1,000 ALDH⁻CD133^{-BMP2shRNA-RFP} cells (double sorted to increase purity) and monitored growth in vivo. The addition of WT ALDH⁻CD133⁻ cells significantly increased growth of ALDH⁺CD133⁺ cells whereas the addition of ALDH⁻CD133⁻ BMP2shRNA cells did not increase the growth of the ALDH⁺CD133⁺ cells (Fig. 4 C, v). FACS analysis of resected ALDH⁺CD133⁺/ALDH⁻CD133^{-BMP2shRNA-RFP} tumors demonstrated complete loss of RFP⁺ cells, consistent with cells being primarily derived from ALDH⁺CD133⁺ WT cells (Fig. S2C).

Impact of BMP2 on Chemotherapy Response.

Given that BMP2 increases the number of CSCs, it could impact chemotherapy response; increasing the number of CSCs could increase chemoresistance, or, alternatively, by pushing CSCs into the cell cycle, BMP2 could enhance chemosensitivity. For both A2780 and SKOV3, combined treatment with cisplatin and BMP2 demonstrated a significant reduction in cell numbers (Fig. 5 A, i). Combined treatment with Noggin and cisplatin also led to a modest reduction in cellular recovery compared with cisplatin treatment alone. However, despite the significant reduction in total cell numbers, compared with cisplatin treatment alone, treatment with BMP2 and cisplatin was associated with an increase in the absolute number of ALDH⁺CD133⁺ CSCs. In contrast, treatment with Noggin and cisplatin was associated with a decrease in absolute CSC numbers (Fig. 5 A, ii). Consistent with this observation, A2780-BMP2shRNA cells were more sensitive to cisplatin therapy (Fig. 5B).

Fig. 5.

BMP2 increases chemoresistance. (A, i) Cell growth curves of A2780 and SKOV3 cells treated with cisplatin, or cisplatin and either BMP2 or Noggin. (ii) Absolute CSC number from A2780 and SKOV3 cells 6 d after treatment with cisplatin, or cisplatin and either BMP2 or Noggin. (B) Cell growth curves of cisplatin-treated A2780 and cisplatin-treated BMP2-shRNA expressing A2780 cells. (C) Tumor weights of A2780 tumors treated with cisplatin, or cisplatin and either BMP2 or Noggin (n = 10 animals per group).

To confirm the importance of these changes in CSCs, we repeated tumor initiation studies of cisplatin, cisplatin plus Noggin, or cisplatin plus BMP2 treated cells. Cisplatin alone reduced tumor initiation; however, cisplatin plus Noggin treatment further decreased tumor initiation capacity (Fig. 4A). Finally, we assessed the impact of BMP2 and Noggin on chemotherapy response in vivo. A2780-derived mouse xenografts were treated with cisplatin, cisplatin and BMP2, or cisplatin and Noggin. Tumors treated with BMP2 and cisplatin were significantly larger whereas tumors treated with cisplatin and Noggin were smaller than cisplatin monotherapy tumors (Fig. 5C).

Discussion

Whether cancer follows a hierarchical/stem cell model or a stochastic model is highly controversial. For primary human epithelial tumors, there are minimal data, and interpretation of the literature is challenging due to the potential for cell contamination associated with FACS isolation. Using a microfluidic single-cell culture (20), we directly observed the differentiation potential of four distinct human ovarian cancer cell populations. The microfluidics device allows confirmation of cellular identity and eliminates issues of FACS-associated contamination. Using this device, we demonstrated a branched differentiation hierarchy of ovarian cancer cells. Although generally CSC marker negative cells were unable to produce CSC marker⁺ progeny, we observed one division (in ∼3,000) where an ALDH⁺CD133⁻ cell was able to give rise to an ALDH⁺CD133⁺ cell. The functional importance of this event remains to be determined. Our studies suggest that, whereas the majority of cell divisions follow a hierarchical model, the model is not rigid, such that stochastic events may also take place.

We find that BMP2 is differentially expressed in the distinct ALDH^+/−CD133^+/− ovarian cancer populations, with highest expression in ALDH⁻CD133⁻. BMP2 promotes expansion of the ALDH⁺CD133⁺ cell population, while suppressing the proliferation of ALDH⁺CD133⁻ and ALDH⁻CD133⁻ cells. Consistent with this observation, BMP2 knockdown results in an ∼10-fold reduction in ALDH⁺CD133⁺ and ALDH⁻CD133⁺ cells and expansion of the ALDH⁺CD13⁻ pool. We propose that, parallel to normal stem cell biology where progenitor cells regulate adult stem cell homeostasis (9), BMP2 produced by progenitor cells acts as a feedback mechanism promoting CSC expansion. Such a mechanism would assure maintenance of a stable CSC percentage as the progenitor pool expands and would be consistent with other reports specifically indicating a role for BMP2 in stem cell maintenance (24).

We found that BMP2 reduces cancer cell growth in vitro, yet promotes tumor growth and chemoresistance in vivo. These findings are indicative of the complexity of BMP2 in regulating the distinct ovarian cancer populations, increasing CSC expansion while suppressing the production and growth of more differentiated cells. As such, BMP2 treatment in vitro is associated with a decrease in absolute cell numbers due to a reduced proliferation of the bulk ALDH⁻CD1333⁻ cells. In contrast, while we cannot rule out other contributing factors, in vivo BMP2 contributes to growth related to the increase in OvCSC. These data point out the critical need to evaluate the impact of novel therapies on CSC rather than total cell numbers when screening for anticancer activity.

Supporting a role for BMP2 pathway activation in ovarian cancer, compared with normal ovarian surface epithelium, BMP2 is up-regulated in ovarian cancer cells (25) whereas the BMP antagonist Chordin is down-regulated (26). Importantly, increased BMP2 expression in ovarian tumors is correlated with poor patient prognosis (27). In other malignancies, both tumorigenic and tumor-suppressive roles have been reported (28). BMP2 can promote cancer growth and metastases in gastric cancer (29), and carcinogen exposure induces BMP2 expression in the breast, resulting in stem cell transformation (30). Although controversial, use of the high dose recombinant BMP2 in patients has been linked with increased cancer rates (31–33). In contrast, loss of BMP signaling via BMP receptor mutations/deletion is associated with increased tumorigenesis and metastasis (34, 35). The contradictory roles of excess BMP exposure versus receptor loss demonstrate the complexity of the pathway. Understanding which receptors mediate the disparate responses of the distinct ovarian cancer cell populations will be import to better define the BMP pathway as a therapeutic target.

In conclusion, under normal conditions, ovarian cancer cells typically differentiate following a predictable branched differentiation hierarchy. BMP2 is differentially expressed in distinct ovarian cancer cell populations, promoting the expansion of ALDH⁺C133⁺ CSCs and restricting the growth of progenitors. Our data indicate that the BMP pathway is an important therapeutic target for ovarian cancer.

Methods

Tumor Processing.

Informed consent was obtained from all patients before tissue procurement. All studies were performed with the approval of the Institutional Review Board of the University of Michigan. All tumors were stage III or IV epithelial ovarian cancer. Tumors were mechanically dissected into single-cell suspensions and isolated on a ficol gradient as previously described (36). For ascites, cell pellets were collected by centrifugation, and red cells were lysed using ACK buffer (Lonza), washed, passed through a 40-μm filter, and then passed four times through a Standard Hub Pipetting needle to isolate single cells (17).

Quantitative Real-Time PCR for Validation of PCR Array Data.

Total RNA was isolated from ALDH^+/−CD133^+/− cell populations using an RNeasy mini kit (Qiagen). Then, 2 μg of RNA was subjected to cDNA synthesis, and qRT-PCR reactions were set up in triplicate with hypoxanthine phosphoribosyltransferase as the internal control using SYBR Green. Each sample was repeated three times.

Flow Cytometric Analysis and Fluorescence-Activated Cell Sorting.

FACS was performed as previously described (17). For each sample, half of the cell/substrate mixture was treated with 50 mmol/L diethylaminobenzaldehyde to establish gating. For FACS characterization of BMPRII and ACTRIIB, we first FACS isolated CD133^+/− cells, and then cells were allowed to recover for 36 h in culture, were fixed with methanol, blocked as above, and then stained with BMPR2 (ab78422,1:20; Abcam) or ACTRIIB (ab76940, 1:285; Abcam) and analyzed with the MoFlo Astrios (Beckman Coulter).

Cell Culture.

A2780, OVCAR8, and PEO1 ovarian cancer cells were provided by S. Murphy, Duke University, Durham, NC. For all in vitro cell culture experiments, isolated cells were allowed to recover overnight after FACS and were then treated with BMP2 or Noggin, (200 ng/mL; R&D). Media was replaced with fresh medium containing BMP2 or Noggin every 48 h. Cell counts were assessed using the Cell Countess. BMP2-shRNA and control A2780 cells were created using lentiviral transduction of a scrambled shRNA control and five different BMP2 pLKO.1-puro-CMV-TagRFP constructs (Sigma). RFP⁺ cells were FACS isolated and expanded, and then BMP2 expression was assessed via qRT-PCR and Western blotting. Tumor sphere cultures were performed as previously described (17, 23) using 10,000 primary cells were cultured in mammary epithelial basal medium (MEBM) ± 200 ng/mL BMP2. Fresh media with BMP2 was added every 3 d. Cells were evaluated after 12 d in culture.

Microfluidic Culture.

Cells were FACS isolated and dispersed into the microfluidic device (37) in supplemented MEBM media (17). Cells were allowed to recover from FACS overnight and then captured cells were restained with ALDEFLUOR and anti-CD133 and photographed to confirm marker expression. ALDH⁻CD133⁻ captured cells were similarly stained and served as negative controls for background fluorescence. Wells with more than one cell were excluded from analysis. Cells were maintained in microfluidic culture for an additional 48 h (cell lines) or 96 h (primary cells) in the presence or absence of 200 ng/mL BMP2 and were again stained with ALDEFLUOR and anti-CD133 and photographed. Divisions (∼40) of each ALDH^−/+CD133^−/+ cell population were scored. An average of 120 progenitors were scored for each cell type. At the conclusion of the experiment, cells were stained with Calcein-AM to confirm cellular viability. All samples were evaluated in four replicate devices from at least two experiments. A Student’s t test was used to determine statistical significance for results for each sample. An ANOVA analysis was used to determine statistically significant findings of aggregate analyses.

Immunoblotting and ELISA.

To detect BMP2 via Western blot, tumor cells were treated with GolgiPlug (1 μL /mL; BD Science) for 4 h, washed, and then lysed in 200 μL of radioimmunoprecipitation assay buffer (Invitrogen) with complete proteinase/phosphatase inhibitor (Roche). Then, 10 μg of total protein was probed with mouse anti-BMP2 (1:800 dilution; Abcam) and detected with ECL. To detect phospho-Smad1/5 after BMP2 (R&D Systems) treatment, cancer cells were treated with BMP2 (200 ng/mL) for 30 min, and then cells were processed as above. Antibodies used for Western blot were as follows: anti-p-Smad1/5 and anti-Smad5 (1:1,000 dilution; Cell Signaling) and anti–β-actin (1:10,000 dilution; Sigma). Bands were visualized using the ECL kit (Pierce, Thermo Scientific). For ELISA, A2780 cells were FACS sorted in the four ALDH^+/−CD133^+/− cell populations and immediately plated into 24-well plates (25,000 cells per well) with serum containing RPMI media. Media was changed after 3 h into serum-free RPMI. Cells were grown for 36 h, and media from each cell population was removed and analyzed with the BMP2 Quantikine ELISA Kit (R&D Systems) per the manufacturer’s protocol.

Chemosensitization.

Cells were plated in replicate, rested overnight, and then treated with cisplatin (0.1–3 μg/mL; SICOR Pharm. Inc.) with or without BMP2 (200 ng/mL) or Noggin (200 ng/mL). After 3 d, replicates of each sample were analyzed via FACS with PI/ALDEFLUOR. The cell number was assayed in the remaining replicates using Cell Countess 3, 5, 8, and 10 d after cisplatin therapy and plotted as a percentage of initial cell input.

Animal Studies.

All studies were approved by the University of Michigan Committee for the Use and Care of Animals. To establish s.c. tumors, 1,000 ALDH⁺CD133⁺ CSCs or 5,000 total cells were injected subcutaneous (SQ) in 300 μL of Matrigel in NOD-SCID mice. Tumors were allowed to engraft for 3 d and then treated with BMP2 (200 ng) or Noggin (200 ng) via s.c. peritumoral injections every other day. Tumor growth was assessed using calipers (using length × width × width/2 calculation). At the time of euthanasia, tumors were resected and weighed. For chemoresistance studies, animals were treated as above, and, after 7 d of tumor engraftment, mice were treated with cisplatin (0.5 mg/kg) weekly for two doses. Then, 3 d after the second dose, a subset of animals were euthanized, and tumors were processed into single cells. Then, 100 cells from each tumor were then mixed with 200 μL of Matrigel and injected SQ into NOD-SCID IL2rγ null (NSG) mice as above. Similarly, for tumor initiation studies, unsorted cells were cultured in vitro in the indicated conditions. Then, 10 or 100 in vitro treated cells were injected SQ into NSG mice as above. Mice were monitored for 9 mo for tumor initiation. For mixing studies, cells were FACS sorted (ALDH⁻CD133⁻ cells were double sorted to increase purity) and mixed in the indicated ratios and injected as above (n = 10 per group). Tumors were monitored for ∼3 mo, and then tumors were resected and weighed.

Immunofluorescence.

AQUA immunofluorescence was performed as previously described (17). Primary antibodies [anti-BMP2 (AP1713A, Abgent), anti-CD133 (Abcam), and anti-ALDH (611195; BD Pharmingen)] were used at 1:100 dilutions. Goat anti-rabbit IgG conjugated to AF488 (A21424, 1:200; Molecular Probes) and goat anti-mouse IgG conjugated to AF648 (A21237; Molecular Probes) were used for fluorescent detection.