Characterization of aldehyde dehydrogenase isozymes in ovarian cancer tissues and sphere cultures

Type-specific expression of ALDH isozymes

We first employed immunohistochemistry to investigate the expression levels of the different ALDH isozymes in archived ovarian tissues using isozyme-specific antibodies. Antibodies specific to ALDH1A1, ALDH1A3, ALDH3A1, ALDH3A2, ALDH3B1 and ALDH7A1 were used to stain a panel of healthy ovaries, benign, borderline, and invasive ovarian tumors. The clinicopathologic characteristics of the samples we used are shown in Additional file 1: Table S1. We found significantly elevated expression of ALDH1A3 (Table 1), ALDH3A2 (Table 2), and ALDH7A1 (Table 3) in the epithelial ovarian tumor tissues than healthy ovarian epithelia. Multiple comparisons using Dunnett’s method showed that there were significant differences between normal ovaries and invasive tumors for ALDH3A2 and ALDH7A1, whereas ALDH1A3 staining showed significant differences between normal ovaries and both borderline and invasive tumors. There was no significant difference in the staining for ALDH isozymes in normal, benign, and tumor stromal components. There was no positive staining from the ALDH3A1 antibody, and the staining of ALDH3B1 did not show significant differences between healthy ovaries and ovarian tumor tissues (data not shown). Differences between histologic tumor subtypes for ALDH1A3, ALDH3A2, and ALDH7A1 were not significant, partly due to the small and unbalanced sample sizes in these experiments and subtle changes may not be detected. Representative figures of the immunohistochemical staining of ALDH1A3, ALDH3A2, and ALDH7A1 antibodies to different categories of tissues are shown in Figure 1.

Characteristics	n a	epithelial			stromal
		score b	P c	C.I. d	score b	P c
Diagnosis
healthy	4	0.13 ± 0.25	0.02*	reference	0.13 ± 0.25	0.27
benign	3	3.25 ± 0.43		(-0.8, 7.0)	1.58 ± 1.38
borderline	3	4.92 ± 2.63		(0.9, 8.7)**	0.75 ± 0.66
invasive	19	4.28 ± 2.24		(1.3, 7.0)**	0.70 ± 0.98
Histology
serous	8	3.88 ± 2.84	0.47		0.91 ± 1.31	0.73
mucinous	3	5.00 ± 0.00			0.92 ± 1.16
endometrioid	4	5.56 ± 1.59			0.25 ± 0.00
clear cell	4	3.25 ± 2.06			0.56 ± 0.63

^a Number of cases.

^b Immunohistochemistry staining score expressed as mean ± SD.

^c P-value, ANOVA test of equal means among the four groups (* equal population variances assumption did not meet, corresponding p-value using Kruskal-Wallis test for equal medians was displayed).

^d Two-sided 95% confidence interval (C.I.) for the difference in means, Dunnett’s post-hoc test compared to value in healthy (in Diagnosis group) or clear cell (in Histology group) as the “reference” group. Dunnett’s test was performed only when the ANOVA or Kruskal-Wallis result was significant (p < 0.05).

** Significant based on Dunnett’s test (family error rate set to 0.05).

Characteristics	n a	epithelial			stromal
		score b	P c	C.I. d	score b	P c
Diagnosis
healthy	5	0.94 ± 0.38	0.003*	reference	1.90 ± 0.74	0.1*
benign	4	3.81 ± 1.84		(−1.0, 6.7)	2.25 ± 1.50
borderline	3	4.42 ± 0.63		(−0.7, 7.7)	1.83 ± 2.02
invasive	35	5.59 ± 2.60		(1.9, 7.4)**	1.00 ± 0.88
Histology
serous	17	5.44 ± 2.86	0.31		1.09 ± 0.87	0.74
mucinous	6	5.75 ± 1.17			0.92 ± 1.16
endometrioid	5	7.50 ± 2.60			0.65 ± 0.78
clear cell	6	4.54 ± 2.75			1.21 ± 0.87

^a Number of cases.

^b Immunohistochemistry staining score expressed as mean ± SD.

^c P-value, ANOVA test of equal means among the four groups (* equal population variances assumption did not meet, corresponding p-value using Kruskal-Wallis test for equal medians was displayed).

^d Two-sided 95% confidence interval (C.I.) for the difference in means, Dunnett’s post-hoc test compared to value in healthy (in Diagnosis group) or clear cell (in Histology group) as the “reference” group. Dunnett’s test was performed only when the ANOVA or Kruskal-Wallis result was significant (p < 0.05).

** Significant based on Dunnett’s test (family error rate set to 0.05).

Characteristics	n a	epithelial			stromal
		score b	P c	C.I. d	score b	P c
Diagnosis
healthy	5	0.64 ± 0.47	0.005*	reference	1.70 ± 0.45	0.54
benign	4	2.63 ± 2.29		(-2.2, 6.2)	3.75 ± 0.50
borderline	3	4.58 ± 1.44		(-0.6, 8.5)	3.00 ± 1.00
invasive	40	5.36 ± 2.77		(1.8, 7.7)**	3.27 ± 2.67
Histology
serous	18	4.57 ± 3.09	0.34		2.87 ± 2.53	0.08
mucinous	8	6.44 ± 2.46			5.43 ± 1.86
endometrioid	8	5.88 ± 2.58			1.70 ± 1.75
clear cell	5	6.20 ± 1.79			3.25 ± 4.03

^a Number of cases.

^b Immunohistochemistry staining score expressed as mean ± SD.

^c P-value, ANOVA test of equal means among the four groups (* equal population variances assumption did not meet, corresponding p-value using Kruskal-Wallis test for equal medians was displayed).

^d Two-sided 95% confidence interval (C.I.) for the difference in means, Dunnett’s post-hoc test compared to value in healthy (in Diagnosis group) or clear cell (in Histology group) as the “reference” group. Dunnett’s test was performed only when the ANOVA or Kruskal-Wallis result was significant (p < 0.05).

** Significant based on Dunnett’s test (family error rate set to 0.05).

Initial immunohistochemical staining of the stem cell marker ALDH1A1 in ovarian tissues yielded particularly interesting patterns not seen with the other ALDH isozymes described above. We have therefore added more cases to confirm the initial findings and the final results are presented here. The staining was not significantly different between healthy ovaries and ovarian tumors (Table 4). However, we saw significant differences in the expression between the different histologic subtypes of ovarian tumors. The endometrioid and mucinous tumors had significant overexpression in the epithelial tumor cells, whereas serous and clear cell epithelial tumor tissues showed very low ALDH1A1 expression (P < 0.001). Dunnett’s multiple comparisons showed significantly lower expression of ALDH1A1 in clear cell tumors than in mucinous and endometrioid tumor types. While the epithelial clear cell tumor cells showed lower ALDH1A1 expression than the other tumor types, ALDH1A1 expression was higher in the clear cell stromal fibroblasts than in the other stromal tumor types (P = 0.02). Figure 2 shows representative images of ALDH1A1 in the different ovarian tissues and in particular the absence of ALDH1A1 staining in the tumors but increased staining in the stromal part of eighteen clear cell ovarian tumors.

Characteristics	n a	epithelial			stromal
		score b	P c	C.I. d	score b	P c	C.I.d
Diagnosis
healthy	5	0.15 ± 0.22	0.102		0.60 ± 0.82	0.08
benign	4	1.94 ± 2.96			4.44 ± 3.14
borderline	8	3.78 ± 3.37			1.94 ± 1.92
invasive	101	1.66 ± 2.71			1.85 ± 2.25
Histology
serous	40	0.67 ± 1.52	< 0.001*	(-0.8, 1.5)	1.12 ± 1.28	0.02	(-3.1, -0.5)**
mucinous	12	5.65 ± 3.05		(3.7, 7.0)**	1.98 ± 2.17		(-2.7, 0.9)
endometrioid	19	3.38 ± 3.28		(1.6, 4.5)**	1.76 ± 2.70		(-2.7, 0.4)
clear cell	29	0.31 ± 0.45		reference	2.90 ± 2.74		reference

^a Number of cases.

^b Immunohistochemistry staining score expressed as mean ± SD.

^c P-value, ANOVA test of equal means among the four groups (* equal population variances assumption did not meet, corresponding p-value using Kruskal-Wallis test for equal medians was displayed).

^d Two-sided 95% confidence interval (C.I.) for the difference in means, Dunnett’s post-hoc test compared to value in healthy (in Diagnosis group) or clear cell (in Histology group) as the “reference” group. Dunnett’s test was performed only when the ANOVA or Kruskal-Wallis result was significant (p < 0.05).

** Significant based on Dunnett’s test (family error rate set to 0.05).

To evaluate the tumor type-specific expression of ALDH isozymes, we also performed quantitative reverse transcription-polymerase chain reaction (qRT-PCR) to measure the mRNA expression levels of ALDH1A1, ALDH1A3, ALDH1B1, ALDH3A1, ALDH3A2, ALDH3B1 and ALDH7A1 in tumor cells microdissected from a panel of frozen tumor tissues. Boxplot in Figure 3 shows that, in general, the RNA levels of all ALDH isozymes were significantly higher in both endometrioid and mucinous tumors than in clear cell and serous tumors. The RNA expression patterns resemble the protein expression of ALDH1A1, which shows higher expression in the endometrioid and mucinous tumors compared with clear cell and serous tumors. However, as ALDH1A3, ALDH3A2 and ALDH7A1 isozymes did not show particularly significant tumor-type specific protein expression, there might be other post-transcriptional mechanisms that regulate the different ALDH isoenzyme protein levels in the tumor tissues.

Western blot analysis was performed to evaluate the expression of the ALDH isozymes in ovarian cell lines. As illustrated in Figure 4A, most of ALDH isozymes except ALDH1A3 showed higher levels in the ovarian cancer cell lines relative to the normal human ovarian surface epithelial (HOSE) cell lines. Only OVCA433 and MCAS showed higher level of ALDH1A3 expression. Like the immunohistochemical staining results in tumor tissues, ALDH1A1 showed a strong tumor type-dependence in expression pattern. While endometrioid and mucinous cancer cell lines showed high protein expression, the serous and clear cell cell lines showed little, if any, detectable protein expression. It is noted that as we have only one endometrioid cancer cell line, the result may not reflect broadly this histologic subtype.

Expression and activity of ALDH1A1 in ovarian cancer cells growing as sphere suspension

The immunohistochemistry and Western blot results led us to further investigate ALDH1A1 as a potential stem cell marker by evaluating the expression of this protein in ovarian cancer spheres. The sphere assay, which demonstrates the capability of stem-like cells to form spheres when cultured in serum-free medium with growth factors [19], has been widely adopted as an in vitro model to study CSCs for human cancers [20–24]. We performed sphere assays with ovarian epithelial cancer cells by growing them as sphere suspensions in standard serum-free medium. We used Western blot analysis to compare the expression of ALDH1A1 and ALDH7A1 in ovarian cancer cells growing as a sphere suspension versus growing as monolayer in complete medium. ALDH7A1 protein expression showed a slight increase in the spheres formed by endometrioid and mucinous cancer cell lines than in the monolayer cells. For ALDH1A1 expression, the two mucinous cancer lines showed increase in expression in sphere cultures than monolayer cultures. The endometrioid cancer cell line expressed very high level of ALDH1A1 both in sphere and monolayer cultures. The clear cell cancer cell line showed some increase from the monolayer cells to the sphere cells. In contrast, serous cancer cell lines did not show any increase in ALDH1A1 expression in the spheres (Figure 4B).

In addition to the protein expression analyses, the Aldefluor® assay was used to measure specific ALDH activity in the monolayer and sphere cancer cells. Representative results, shown in Figure 5, are parallel the results obtained by Western blot analysis. The mucinous cancer cells showed robust increased activity in the sphere cells. Endometrioid cells showed strong ALDH activity under both monolayer and sphere conditions. The clear cell cancer cells showed a small increase in activity in the sphere cells, while the serous cancer cells retrieved from an ascites sample did not show any ALDH activity under either culture conditions. Although other studies have suggested that the Aldefluor® assay also measures the activity of some other ALDH isoforms such as ALDH1A3 [34, 36], our Aldefluor® assay results closely reflect the ALDH1A1 activity in the cancer cells.

Implications of ALDH isozyme expression in ovarian cancer

Ovarian cancer is heterogeneous in nature, comprising tumors with different histologic subtypes and developmental stages [4, 5]. The cancer stem cell hypothesis proposes the presence of distinct tumor-propagating cell populations that are responsible for self-renewal and multilineage differentiation into different types of tumors [12]. ALDH1A1, and recently ALDH1A3, have been described as valuable stem cell markers in different human tumors and in vitro systems [26, 28, 34, 36]. ALDH1A1 positivity has also been associated with chemoresistance in ovarian cancer [38, 39]. The present study revealed ALDH1A1 to be expressed predominantly in mucinous and endometrioid epithelial cancer cells, but not in most of the serous and clear cell cancer cells. Instead, high ALDH1A1 expression was found in the stromal fibroblasts in the latter two types of ovarian cancer. In a previous study, higher levels of ALDH1A1 expression were found in mammary stromal cells than in epithelial cells [40]. Although it might be argued that the stromal ALDH1A1 staining arose from cancer cells with mesenchymal features, as suggested in a proteomic profiling study of a panel of lung adenocarcinoma cell lines [37], the predominant stromal staining observed in our study is consistent with ALDH1A1 expression being distinctly lineage-specific in different histologic types of ovarian tumors. It is well documented that expression and activity levels of ALDH isozymes depend on cancer type and/or cell of origin [36, 40]. Penumatsa et al... reported recently reduced expression of ALDH1A1 in serous ovarian tumors [41] and Li et al... reported that ALDH1A1 expression was repressed by histone-lysine N-methyltransferase EZH2 in high-grade serous ovarian carcinomas [42]. It will be of great interest to delineate the role of ALDH1A1 in lineage differentiation and its regulation in ovarian cancer.

Moreover, it is of equal importance to evaluate the functional roles of elevated ALDH1A3, ALDH3A2 and ALDH7A1 isozymes in ovarian cancer. The ALDH1A3 isoform has been reported to be a novel CSC marker with potential clinical prognostic application in breast cancer [34]. ALDH7A1 was also found to be involved in prostate cancer bone metastasis [33]. Further analysis of these novel ALDH isozymes may have significant diagnostic and prognostic implications in ovarian cancer. A more thorough understanding of the molecular mechanisms underlying their activities in the development of ovarian cancer may pave a way for more effective treatment of ovarian cancer.

Ovarian clinical samples and ovarian cell lines

Archived formalin-fixed, paraffin-embedded normal, benign, and cancerous ovarian tissues were collected from women undergoing surgery at the Brigham and Women's Hospital for a diagnosis of primary ovarian cancer or from control subjects who were undergoing the procedure of hysterectomy or oophorectomy for benign gynecologic diseases. Additional 15 cases of clear cell ovarian carcinomas were from Department of Obstetrics and Gynecology, Osaka City General Hospital, Japan. All patient-derived biologic specimens were collected and archived under protocols approved by the Human Subjects Committee of the Brigham and Women's Hospital, USA, and Osaka City General Hospital, Japan. Samples were collected with written informed consent from patients and confirmed histologically by gynecologic pathologists. Cases were staged according to International Federation of Gynecology and Obstetrics (FIGO) system. The normal human ovarian surface epithelial (HOSE) cells and ovarian cancer cell lines have been described previously [43]. Normal HOSE cells were collected by scraping the ovarian surface of the control subjects who were undergoing hysterectomy or oophorectomy for benign diseases. Long-term HOSE cells were immortalized by a HPV E6/E7 gene introduction. All ovarian cell lines were maintained in a mixture of medium 199 and MCDB105 medium (1:1) (Sigma, St. Louis, MO) supplemented with 10% fetal calf serum (Invitrogen, Carlsbad, CA).

Sphere assay

Standard sphere assay was performed according to Dontu et al… with minor changes [19]. Single cancer cells were resuspended in NeuroBasal-A Medium (Invitrogen) supplemented with B27 (Invitrogen), 20 ng/ml EGF and 20 ng/ml bFGF (Invitrogen), and 4 μg/ml heparin (Sigma-Aldrich), in ultra-low attachment culture plates (Fisher Scientific, Pittsburgh, PA). Cells were cultured for 1 week to form spheres before harvesting.

Immunohistochemistry

Paraffin-embedded ovarian tissue blocks were sectioned at a thickness of 7 μm, mounted on Superfrost Plus microscopic slides (Fisher Scientific, Pittsburgh, PA), and dried at 50 °C for at least 3 hours. Deparaffinization was performed using xylene and rehydration with a graded ethanol series. Antigen retrieval was performed in a pressure-cooker in antigen-unmasking solution (Vector Laboratories, Burlingame, CA) for 10 min. Endogenous peroxidases were blocked using 0.3% H₂0₂ in methanol for 20 min. The sections were then blocked with normal blocking serum for 20 min and subsequently incubated overnight with ALDH isozyme-specific antibodies. Antibodies specific to ALDH1A1, ALDH1A3, ALDH3A1, ALDH3A2, and ALDH3B1 have been described [44, 45]. Antibody specific to ALDH7A1 was purchased from Epitomics, Inc (Burlingame, CA). After incubations with primary and secondary antibodies (Vector Laboratories, Burlingame, CA), the reaction was visualized using Vectastain Elite ABC Kit with diaminobenzidine chromogen as a substrate (Vector Laboratories, Burlingame, CA). Sections were counterstained lightly with hematoxylin and mounted in Permount® (Fisher Scientific, Pittsburgh, PA). The staining was quantified with a semi-quantitative scoring system. The weighted score was obtained by multiplying the staining intensity score ranging from 3+ (strongest positive) to 0 (no evidence of stain) and the score for the percentage of positive cells ranging from 3+ (100% stained) to 0 (no cells stained). Two trained observers scored the slides independently and the scores were compared for discrepancies and averaged.

RNA isolation and quantitative reverse transcription-polymerase chain reaction

Western blot analysis

Total cell lysates were prepared from growing cells using RIPA buffer (50 mM Tris–HCl pH 8, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate and 0.1% SDS) supplemented with PhosStop phosphatase inhibitor cocktail and complete protease inhibitor cocktail (Roche Applied Science, Indianapolis, IN) and protein concentration was measured with a MicroBCA protein assay kit (ThermoScientific, Rockford, IL). Ten μg of total cell lysates were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to a polyvinylidene fluoride (PVDF) membrane using a SEMI-DRY Transfer cell (Bio-Rad Laboratories, Hercules, CA). After blocking with 5% nonfat dry milk in 1X TBST buffer (10 mM Tris–HCl pH 7.5, 150 mM NaCl, 0.05% Tween-20) at room temperature for 1 hr, the membrane was incubated with the primary antibody at 4°C overnight, then washed at room temperature with 1X TBST buffer. The bound antibody was detected by the secondary antibodies conjugated with horseradish peroxidase (Pierce Biotechnology, Rockford, IL) and a Supersignal west pico kit (Pierce Biotechnology, Rockford, IL).

Aldefluor® assay

ALDH activity was detected using the Aldefluor® assay kit (STEMCELL Technologies, Vancouver, BC, Canada) as described by the manufacturer. Briefly, dissociated single cells from cell lines or spheres were resuspended in Aldefluor® assay buffer containing an ALDH substrate, bodipy-aminoacetaldehyde (BAAA), at 7.5 μM, and incubated for 1 hr at 37°C. An identical reaction was also performed in the presence of 15 mM diethylaminobenzaldehyde (DEAB), an ALDH-specific inhibitor. Fluorescence intensity of the stained cells was analyzed using an Accuri C6 Flow Cytometer (BD Accuri Cytometers, Ann Arbor, MI). The reaction with DEAB was used to define the baseline for the assay, i.e., fluorescence not associated with ALDH activity. ALDH activity of a sample was determined based on the fluorescence intensity beyond the threshold defined by the reaction with DEAB.

Statistical Analysis

All analyses were performed using MINITAB statistical software (Minitab, State College, PA). ANOVA was used to compare the mean IHC scores among different diagnostic and histologic groups. If the equal population variances assumption was not met, the non-parametric Kruskal-Wallis test was performed to compare the results obtained from ANOVA. When there was a statistically significant difference among the groups, multiple comparisons with a control group (Healthy group for Diagnosis and Clear Cell group for Histology) were performed using the Dunnett’s method. A difference was deemed significant when it reached the 5% level, i.e., P ≤ 0.05. As the normality assumption for the qRT-PCR data was not met, the Kruskal-Wallis test was used to analyze the ranked qRT-PCR data according to histologic subtypes, and the results are presented as a boxplot.