Histological and prognostic importance of CD44⁺/CD24⁺/EpCAM ⁺ expression in clinical pancreatic cancer

Abstract

CD44⁺/CD24⁺/EpCAM ⁺ cells have been reported to be cancer stem cells in pancreatic cancer; however, the histological and clinical importance of these cells has not yet been investigated. Here we clarified the characteristics of CD44⁺/CD24⁺/EpCAM ⁺ cells in clinical specimens of pancreatic cancer using immunohistochemical assay. We used surgical specimens of pancreatic ductal adenocarcinoma from 101 patients. In view of tumor heterogeneity, we randomly selected 10 high‐power fields per case, and triple‐positive CD44⁺/CD24⁺/EpCAM ⁺ expression was identified using our scoring system. The distribution, histological characteristics, and prognostic importance of CD44⁺/CD24⁺/EpCAM ⁺ cells were then analyzed. As a result, the distribution of CD44⁺/CD24⁺/EpCAM ⁺ cells varied widely among the 101 cases examined, and CD44⁺/CD24⁺/EpCAM ⁺ expression was correlated with poor glandular differentiation and high proliferation. Survival analysis showed that CD44⁺/CD24⁺/EpCAM ⁺ expression was not correlated with patient outcome; however, CD44⁺/CD24⁺ expression appeared to be correlated with poor prognosis. In conclusion, CD44⁺/CD24⁺/EpCAM ⁺ expression overlapped with poorly differentiated cells and possessed high proliferative potential in clinical pancreatic cancer. In particular, the presence of double‐positive CD44⁺/CD24⁺ expression seemed to have clinical relevance, associating with poor prognosis.

Pancreatic cancer is a rapidly progressive disease with a dismal prognosis. Despite recent advances in diagnosis and treatment, only about 4% of patients will live 5 years after diagnosis.1 This discouraging outcome may be due to the intrinsic biologic aggressiveness of pancreatic cancer, characterized by extensive local tumor invasion, early systemic dissemination, and relative resistance to chemotherapy and radiation. However, the molecular mechanisms underlying the malignant characteristics of pancreatic cancer have not been clarified.

There is emerging evidence to suggest that the capability of a tumor to grow and propagate is dependent on a small subset of cells within it, termed cancer stem cells (CSCs).2 The CSC hypothesis proposes that CSCs have self‐renewal, proliferative, and differentiative capacities, lying at the apex of the hierarchical organization of cells within a tumor.3 This hypothesis is an attractive one for explaining the functional heterogeneity that is commonly observed in solid tumors, suggesting that cells in a tumor exhibit distinct capacities for differentiation and proliferation.4 Furthermore, several studies have suggested the clinical relevance of CSCs in terms of their apparent relationship to therapy resistance,5, 6, 7 and poor outcome.8, 9, 10 Consequently, the CSC compartment is being increasingly recognized as a potential target for effective treatment of cancers.11, 12, 13

The majority of studies that have identified CSCs were performed using the side population assay with FACS of cancer tissues followed by the tumorigenicity assay involving xenotransplantation of the sorted cells into immunodeficient mice. Using these techniques, the existence of CSCs has also been validated in several solid cancers such as those of the breast,14 brain,15 head and neck,16 colon,17, 18 lung,19 liver,20 and pancreas.21, 22 Cancer stem cells in pancreatic cancer were also demonstrated using FACS and the xenotransplantation assay: Hermann et al. identified such cells as CD133⁺ cells,21 whereas Li et al. identified them as CD44⁺/CD24⁺/EpCAM⁺ cells (EpCAM is a synonym for ESA, which was one of the CSC markers described in the original article of Li et al.).22 In the latter study, pancreatic cancer cells with the CD44⁺/CD24⁺/EpCAM⁺ phenotype, accounting for only 0.2–0.8% of total pancreatic cancer cells, had a tumorigenic potential 100‐fold greater than that of non‐tumorgenic cancer cells.22

However, these CSC markers revealed by FACS and xenotransplantation must be interpreted with caution, in view of possible artifacts associated with the use of model systems.23 That is, the xenotransplantation assay system may select a cell subset that is more capable of surviving and generating tumors in immunodeficient mice, which is very different from tumor growth in the human body.4, 24, 25 Although FACS can reveal the true surface expression of cells objectively, the character of these CSC markers should be confirmed by other methods in order to provide reliable support for the relevance of the CSC hypothesis to cancer therapy. Several studies using immunohistochemistry have demonstrated the characteristics of CSCs.23, 26, 27 Immunohistochemical analysis of surgical specimen of pancreatic cancer have examined the prognostic value of CD44,28 CD24,29 and EpCAM30 individually; however, no such studies have yet examined the histological or clinical impact of combinations of these markers.

The purpose of the present study was to investigate the clinical significance of CD44⁺/CD24⁺/EpCAM⁺ cells in resected specimens of pancreatic cancer using the immunohistochemical assay. As CSCs were assumed to be a small population of cells with a heterogeneous intratumoral distribution, we observed 10 small fields selected randomly in each pancreatic cancer. In these fields we identified CD44⁺/CD24⁺/EpCAM⁺ expression using our own scoring system, and evaluated its distribution, as well as its relationship to histological differentiation, cell proliferation, and clinical outcome.

Materials and Methods

Patients and tissue samples

Tissues from primary ductal adenocarcinoma of the pancreas were obtained intraoperatively from 101 patients who underwent surgical operation at the Department of Surgery, University of Tsukuba Hospital, between December 2002 and January 2013 (Table 1). All patients provided informed consent for analysis of their tissue samples according to the ethics committee of University of Tsukuba Hospital. The samples were diagnosed by pathologists in accordance with the World Health Organization (WHO) classification.31 TNM staging was established according to the Union for International Cancer Control (UICC).32 Cases of ductal adenocarcinoma derived from intraductal papillary mucinous neoplasm were excluded. Macro‐ or microscopic residual tumor after operation, histological venous and lymphatic invasion were assessed. Eighty‐four patients received postoperative chemotherapy, consisting of gemcitabine and/or oral fluorouracil derivative (S‐1). Six patients with local recurrence after surgery received postoperative radiotherapy. Patients receiving neoadjuvant chemotherapy or radiotherapy were excluded. Surviving patients were followed up until March 2013.

Table 1.

Patients and clinicopathological characteristics

Characteristics	Number
Gender
Male	57
Female	44
Age
<60	22
≥60	79
(Mean: 68 years, Range 39–86 years)
TNM classification
T stage
T1	3
T2	6
T3	89
T4	3
N stage
N0	20
N1	81
M stage
M0	90
M1	11
Residual tumor
Absent	67
Present	34
Venous invasion
Absent	22
Present	79
Lymphatic invasion
Absent	23
Present	78
Treatment
Surgery alone	17
Surgery + CT	78
Surgery + CT + RT	6
Surgical Procedure
Pancreatoduodenectomy	74
Distal pancreatectomy	26
Total pancreatectomy	1

CT, postoperative chemotherapy; RT, postoperative radiotherapy.

Immunohistochemical procedures

The resected tissues were fixed in 10% formalin and embedded into paraffin blocks, and the most representative block being chosen for each case. Each block was cut into serial sections 2 μm thick for staining with H&E and immunohistochemistry for CD44, CD24, EpCAM, and Ki‐67. Immunohistochemistry was performed using the EnVision+ System‐HRP (Dako Japan, Tokyo, Japan), and the protocol was optimized for each antigen (Table 2). Briefly, sections were deparaffinized with xylene and rehydrated with ethanol. Antigens were then retrieved by heating at 100°C in a microwave oven in citrate buffer at pH 6.0. After blocking of endogenous peroxidase activity with 0.03% hydrogen peroxide, the sections were incubated with the primary antibodies, those for CD44, CD24, and Ki‐67 being diluted in 1% bovine serum albumin in PBS, and that for EpCAM in Can Get Signal Solution (Toyobo, Osaka, Japan). After washing the sections with PBS, anti‐mouse HRP‐labeled polymer (Dako Japan) was applied at room temperature for 30 min. After washing, the reaction product was visualized with diaminobenzidine chromogen (Dako Japan) applied for 7 min. The sections were counterstained with hematoxylin.

Table 2.

Primary antibodies used for immunohistochemistry

Antigen (Clone)	Antibody subtype	Manufacturer	Antigen retrieval temperature/time/solution	Dilution	Incubation temperature/time
CD44 (156‐3C11)	Mouse monoclonal	Thermo Scientific, Kanagawa, Japan	100°C/15 min/citrate buffer	1:100	Room temperature/30 min
CD24 (SN3b)	Mouse monoclonal	Thermo Scientific, Kanagawa, Japan	100°C/15 min/citrate buffer	1:50	Room temperature/30 min
EpCAM (VU‐1D9)	Mouse monoclonal	Leica Microsystems, Tokyo, Japan	100°C/15 min/citrate buffer	1:100	Room temperature/30 min
Ki‐67 (MIB‐1)	Mouse monoclonal	Dako Japan, Tokyo, Japan	100°C/30 min/citrate buffer	1:25	4°C/overnight

Selection of 10 assessment fields for each case

To take into account intratumoral heterogeneity of antigen expression, unbiased selection of 10 fields in each section was performed (Fig. 1a). Briefly, the whole image of a H&E section was obtained using the image joining system of the microscope (BIOREVO, KEYENCE, Osaka, Japan), and the location of the cancer cells was marked in order to exclude non‐cancerous parts of the specimen. Guided by this image, 10 high‐power fields were selected randomly per section using a × 20 objective and a × 10 ocular lens (field of view = 0.37 mm²). A total of 1010 fields from the 101 patients were selected.

Figure 1.

Selection of 10 assessment fields and scoring criteria used for immunohistochemical evaluation. (a) Whole images of H&E sections were obtained using the image‐joining system of the microscope, and the locations of the cancer cells were marked (dotted line) in order to exclude non‐cancerous part (asterisks) of the specimen. Guided by this image, 10 fields (#1–10) were randomly selected in a cancerous part of the specimen (small boxes). Scale bar, 5 mm. (b) The expression of markers in a field was evaluated in terms of intensity and percentage separately, and finally expressed as Score 0, 1, 2, or 3. (a) “Score 3” field was considered to represent a “positive” field.

Scoring criteria and comprehensive classification for each of the 1010 fields

Staining for each of CD44, CD24, and EpCAM was evaluated separately using our scoring system (Fig. 1b). Briefly, the intensity of membrane staining (none, weak, moderate, or strong) and the percentage (0–10%, 11–50%, 51–80% or 81–100%) of stained cancer cells per field were determined separately, followed by classifying the paired Score for intensity and percentage as 0, 1, 2 or 3. When evaluating an individual marker, a “Score 3” field was considered to represent a “positive” field. For evaluating a CD44⁺/CD24⁺/EpCAM⁺ field, a “Total Score” was obtained by summing the scores for CD44, CD24, and EpCAM. A “Total Score 9” (=CD44 Score 3 + CD24 Score 3+ EpCAM Score 3) field was considered to represent a “triple positive” field in which a cluster of CD44⁺/CD24⁺/EpCAM⁺ cells was present. For evaluation of histological differentiation, each field was classified morphologically into two types – poor differentiation or well differentiation – according to glandular formation in adenocarcinoma using H&E stained sections, with reference to the grading system for glandular differentiation established by Adsay et al.33 Briefly, glandular patterns 1 and 2 in Adsay's study were classified as well differentiation in ours, and pattern 3 as poor differentiation. For evaluation of Ki‐67 as a marker of cell proliferation, the percentage of cells showing nuclear Ki‐67 staining was assessed. Any sample containing ≥10% Ki‐67‐labelled cells was defined as “high Ki‐67”, whereas any containing <10% Ki‐67‐labelled cells was defined as “low Ki‐67”. The 1010 selected fields were classified as described above, and correlations among the expressions were analyzed.

Survival analysis

Correlations between marker expression and patient survival time from the date of surgery were analyzed. The 65 patients whose follow‐up time was more than 2 years were included in the survival analysis. Survival analysis with reference to CD44⁺/CD24⁺/EpCAM⁺ immunostaining was performed by comparison of patients whose tumors had, and did not have, triple‐positive fields. Survival analysis with reference to fields that were individually positive for CD44, CD24, or EpCAM was performed by comparison of patients whose tumors had ≥6 positive fields with those whose tumors had ≤5 positive fields. In addition, survival analysis with reference to CD44⁺/CD24⁺ immunostaining was performed by comparison of patients whose tumors had, and did not have, double‐positive fields (=CD44 Score 3 + CD24 Score 3).

Statistics

Correlations between immunohistological parameters were analyzed by Mann–Whitney U‐test. Correlations of immunohistochemical parameters with clinical characteristics were analyzed by Fisher's exact probability test. Overall survival time was calculated using the Kaplan–Meier method, and log‐rank test was used to estimate differences in survival. Differences at P < 0.05 were considered statistically significant. Statistical analysis was performed using the IBM SPSS Statistics software package (IBM Japan, Tokyo, Japan).

Results

Distribution of triple‐positive fields in 101 cases

First, we present four cases as examples of immunohistochemical scoring (Case 1: female, 52 years old, Case 11: female, 66 years old, Case 17: male, 56 years old, Case 30: male, 70 years old) (Fig. 2a, upper). None of the fields were triple‐positive field in Cases 1 and 11, one field was triple‐positive in Case 17, and nine fields were triple‐positive in Case 30. Next, we show the panel of Total Scores for the representative 30 cases from 101 cases (Fig. 2a, middle). Among the representative 30 cases, 14 cases had at least one triple‐positive field (Cases 17–30). Among these 14 cases, Case 30 had nine triple‐positive fields, whereas the other 13 cases had five such fields or less. The other 16 cases had no triple‐positive fields (Cases 1–16); Cases 2–16 had at least one field with a Total Score of 6–8, whereas Case 1 had fields with a Total Score of 2 or 3. These scores for the 30 cases were distributed into four patterns of CD44⁺/CD24⁺/EpCAM⁺ cells: “dominant” (Case 30), “clustered” (Cases 17–29), “scattered” (Cases 2–16), or “undetectable” (Case 1) (Fig. 2a, lower). The Total Scores of all 101 cases are shown in Figure S1. Among the total of 1010 assessed fields, only 70 (7%) were evaluated as triple‐positive. Among the 101 cases, 26 cases had at least one triple‐positive field. Representative immunohistochemically stained triple‐positive fields are presented in Figure 2(b).

Figure 2.

Immunohistochemical profiles of 300 selected fields. (a) Upper: Scoring panels of four representative cases of immunohistochemical evaluation (Cases 1, 11, 17, and 30). The field showing a Total Score 9 (=CD44 Score 3 + CD24 Score 3 + EpCAM Score 3) was picked out as a triple‐positive field. Middle: Total Scores for representative 30 cases. Lower: Four patterns of distribution of CD44⁺/CD24⁺/EpCAM ⁺ expression derived from the numbers of triple‐positive fields: “dominant” (Case 30), “clustered” (Cases 17–29), “scattered” (Cases 2–16), or “undetectable” (Case 1). (b) An example of immunohistochemistry of a triple‐positive field, including H&E‐ and Ki‐67‐stained sections. This field showed CD44 Score 3, CD24 Score 3, EpCAM Score 3, well differentiation, and a high Ki‐67 labeling. Scale bars, 100 μm.

Correlations of CD44, CD24, EpCAM, and Total Scores with glandular differentiation and cell proliferation in the 1010 fields

We analyzed the 1010 selected fields to demonstrate the correlations of CD44, CD24, EpCAM, and Total Scores with morphological glandular differentiation and cell proliferation assessed using Ki‐67 labeling (Table 3). For example, when fields were classified according to Total Score and differentiation, 40 (57%) of 70 Score 9 fields, 113 (48%) of 233 Score 6–8 fields, 178 (41%) of 430 Score 3–5 fields, and 55 (20%) of 277 Score 0–2 fields showed poor differentiation. Then we assessed that Total Score was markedly correlated with poor differentiation (P < 0.0001). As a result, CD44, CD24, and Total Score were significantly correlated with poor differentiation and high Ki‐67 labeling. On the other hand, EpCAM score was correlated with well differentiation and low Ki‐67 labeling.

Table 3.

Correlations of CD44, CD24, EpCAM, and Total Scores with glandular differentiation and cell proliferation (Ki‐67 labeling) in the 1010 fields

Expression	Differentiation; number (%)			P‐value	Proliferation (Ki‐67); number (%)			P‐value
	Poor 386 (38)	Well 624 (62)	Total 1010 (100)		High (≥10) 440 (44)	Low (<10) 570 (56)	Total 1010 (100)
CD44
Score 3 (positive)	277 (61)	176 (39)	453 (100)	<0.0001	239 (53)	214 (47)	453 (100)	<0.0001
Score 2	60 (37)	102 (63)	162 (100)		69 (43)	93 (57)	162 (100)
Score 1	32 (20)	131 (80)	163 (100)		50 (31)	113 (69)	163 (100)
Score 0	17 (7)	215 (93)	232 (100)		82 (35)	150 (65)	232 (100)
CD24
Score 3 (positive)	149 (42)	204 (58)	353 (100)	0.0003	172 (49)	181 (51)	353 (100)	<0.0001
Score 2	65 (52)	59 (48)	124 (100)		65 (52)	59 (48)	124 (100)
Score 1	61 (37)	106 (63)	167 (100)		70 (42)	97 (58)	167 (100)
Score 0	111 (30)	255 (70)	366 (100)		133 (36)	233 (64)	366 (100)
EpCAM
Score 3 (positive)	109 (32)	233 (68)	342 (100)	0.0022	133 (39)	209 (61)	342 (100)	<0.0001
Score 2	40 (34)	77 (66)	117 (100)		43 (37)	74 (63)	117 (100)
Score 1	66 (45)	81 (55)	147 (100)		56 (38)	91 (62)	147 (100)
Score 0	171 (42)	233 (58)	404 (100)		208 (51)	196 (49)	404 (100)
Total Score
Score 9 (triple positive)	40 (57)	30 (43)	70 (100)	<0.0001	44 (63)	26 (37)	70 (100)	<0.0001
Score 6–8	113 (48)	120 (52)	233 (100)		125 (54)	108 (46)	233 (100)
Score 3–5	178 (41)	252 (59)	430 (100)		176 (41)	254 (59)	430 (100)
Score 0–2	55 (20)	222 (80)	277 (100)		95 (34)	182 (66)	277 (100)

Correlations of triple positive fields with patient clinicopathological characteristics

Analysis in terms of triple positivity was performed by comparing patients whose tumors did (26 cases) and did not (75 cases) have triple‐positive fields. Triple positivity was not correlated with any patients' clinicopathological characteristics (Table 4).

Table 4.

Correlations of triple positive fields with patient characteristics

Variables	Triple positive fields (=Score 9)		P‐value
	Having n = 26 (%)	Not having n = 75 (%)
Gender
Male	15 (26)	42 (74)	0.53
Female	11 (25)	33 (75)
Age
<60 years	10 (45)	12 (55)	0.99
≥60 years	16 (20)	63 (80)
TNM classification
T stage
T1, T2	2 (22)	7 (78)	0.58
T3, T4	24 (26)	68 (74)
N stage
N0	6 (30)	14 (70)	0.78
N1	20 (25)	61 (75)
M stage
M0	22 (24)	68 (76)	0.30
M1	4 (36)	7 (64)
Residual tumor
Absent	9 (26)	25 (74)	0.90
Present	17 (25)	50 (75)
Venous invasion
Absent	6 (27)	16 (73)	0.68
Present	20 (25)	59 (75)
Lymphatic invasion
Absent	3 (13)	20 (87)	0.09
Present	23 (29)	55 (71)

Prognostic impact

Survival analysis was conducted for 65 patients whose follow‐up time was more than 2 years (Table 5). Survival analysis in terms of triple positivity was performed by comparing patients whose tumors did (18 cases) and did not (47 cases) have triple‐positive fields (Fig. 3a). The median survival times of these groups were 20.3 months and 23.8 months, respectively, but the difference in outcome between the two was not significant (P = 0.50). The clinical importance of double or single positive expression was also assessed. CD44⁺/CD24⁺ expression was significantly correlated with poorer outcome (P = 0.044, Fig. 3b), whereas CD44⁺/EpCAM⁺, CD24⁺/EpCAM⁺, CD44⁺, CD24⁺, or EpCAM⁺ expression were not correlated with outcome. Among the patient characteristics, T stage and N stage were significantly correlated with poorer prognosis. Multivariate analysis using Cox proportional‐hazards regression model performed on CD44⁺/CD24⁺ expression, T stage, and N stage; however, these three parameters failed to reach significance (data not shown).

Table 5.

Univariate survival analysis of conventional prognostic factors and CD44, CD24, EpCAM and their combinations

Variables	Cases n = 65	Median survival time (months)	P‐value
Gender
Male	34	22.7	0.51
Female	31	22.4
Age
≥60 years	48	23.0	0.69
<60 years	17	22.4
TNM classification
T stage
T1, T2	5	65.5	0.040
T3, T4	60	21.0
N stage
N0	12	42.6	0.048
N1	53	17.7
M stage
M0	56	22.7	0.38
M1	9	17.4
Residual tumor
Absent	45	23.0	0.29
Present	20	16.4
Venous invasion
Absent	17	29.8	0.12
Present	48	21.0
Lymphatic invasion
Absent	14	29.8	0.30
Present	51	21.0
Total Score 9 field
Having	18	20.3	0.50
Not having	47	23.8
CD44 Score 3 fields
≥6 fields	26	17.4	0.85
<5 fields	39	23.5
CD24 Score 3 fields
≥6 fields	21	21.0	0.23
<5 fields	44	23.0
EpCAM Score 3 fields
≥6 fields	21	21.5	0.83
<5 fields	44	22.7
CD44 + CD24 Score 6 fields
Having	33	17.4	0.044
Not having	32	24.5
CD44 + EpCAM Score 6 fields
Having	31	23.5	0.41
Not having	34	22.4
CD24 + EpCAM Score 6 fields
Having	27	21.5	0.46
Not having	38	22.7

Figure 3.

Survival analysis according to the presence of triple or double positive fields. (a) CD44⁺/CD24⁺/EpCAM ⁺ expression was not correlated with prognosis (P = 0.50). (b) CD44⁺/CD24⁺ expression was strikingly correlated with poorer outcome (P = 0.044).

Discussion

Although the biology of CSCs has become increasingly clearer as a result of many experimental studies, the histological or clinical implications of CD44⁺/CD24⁺/EpCAM⁺ cells, considered to be putative CSCs in pancreatic cancer, have not yet been examined. Our present immunohistochemical study of CD44⁺/CD24⁺/EpCAM⁺ expression in pancreatic cancer highlighted three points. First, the distribution of CD44⁺/CD24⁺/EpCAM⁺ cells varied widely among the 101 examined cases. Second, the expression of CD44⁺/CD24⁺/EpCAM⁺ in pancreatic cancer was correlated with poor glandular differentiation, and high proliferation. Third, triple positive of CD44⁺/CD24⁺/EpCAM⁺ expression was not related with prognosis; however, double positive of CD44⁺/CD24⁺ expression was closely associated with a poor clinical outcome.

We initiated this study to determine whether CSCs were actually present in clinical cases of pancreatic cancer and where in the tumors they were located. There was great variation in the distribution of CSCs among the 101 cases, being sorted into four types: “dominant”, “clustered”, “scattered”, or “undetectable” (Fig. 2a, S1). The differences in the CSC distribution patterns among clinical cases may be attributable to the relative proportion of the two activities of CSCs, that is, self‐renewal and differentiation. Sottoriva et al. focused on the probability of symmetrical division (self‐renewal) and asymmetrical division (differentiation), and demonstrated that the distribution of CSCs depended on the probability of symmetrical division (Ps).34 Briefly, the computer‐simulated the distribution of CSCs by setting two models, one assuming Ps = 0.1 (self‐renewal: 10%, differentiation: 90%), and the other assuming Ps = 0.03 (self‐renewal: 3%, differentiation: 97%). They reported that the Ps = 0.1 model reached a clustered distribution, whereas the Ps = 0.03 model reached a scattered distribution. On the basis of their simulation, it became clear that the cluster pattern of CD44⁺/CD24⁺/EpCAM⁺ cells in our study was attributable to a high frequency of self‐renewal and a low frequency of differentiation of CSCs. A dominant distribution of CSCs may be explained by the model in which CSCs have a marked self‐renewal capacity with little differentiation capacity. The undetectable pattern suggests that CD44⁺/CD24⁺/EpCAM⁺ cells may not have been responsible for tumor growth in these cases; however, this pattern appears to be rare. The variations in the distribution patterns evident in the present study imply that the self‐renewal capacity of CD44⁺/CD24⁺/EpCAM⁺ cells might have differed widely among the 101 cases.

Next, it seems pertinent to address the issues of histological glandular differentiation and proliferation of CSCs in pancreatic cancer. Marked CD44⁺/CD24⁺/EpCAM⁺ expression was evident in the component of pancreatic cancer showing poor differentiation and high proliferation (Table 3). Poor glandular differentiation, that is, a lack of gland formation, is one of the basic features indicative of cancer cell aggressiveness.35 Our results suggest that, morphologically, CSCs may show poorly differentiated (dedifferentiated) gland formation, and may have the capacity to differentiate into multiple types of cancer cells, being located at the apex of the differentiation hierarchy. This is consistent with a report by Pece et al. indicating that poorly differentiated breast cancers contained more CSCs than well differentiated tumors in a xenograft model.36 Ki‐67 is a nuclear antigen expressed in the G₁, G₂, and S phases of the cell cycle, but not in the G₀ phase; therefore, Ki‐67 is currently accepted as a marker of proliferative activity.37 The high Ki‐67 expression of CSCs in our study conflicts with earlier reports indicating that CSCs tend to be in the quiescent phase of the cell cycle.38, 39 However, Li et al. showed that CD44⁺/CD24⁺/EpCAM⁺ cells were not enriched at any particular stage of the cell cycle,22 and Gao et al. showed that CD24⁺ cells of ovarian cancer were enriched in S phase.40 Thus, the issue of whether CSCs have a specific cell cycle distribution appears to depend on the markers, tissues, or experimental methods used. Earlier studies demonstrated that poor glandular differentiation or high expression of Ki‐67 in human pancreatic cancer was correlated with high growth capability when tumors were xenotransplanted into mice.41, 42 Since CD44⁺/CD24⁺/EpCAM⁺ cells were also identified as a population with high tumorigenicity in the xenograft models, it seemed reasonable that CD44⁺/CD24⁺/EpCAM⁺ expression overlapped with poor differentiation or a high Ki‐67 labeling. Moreover, we have previously examined the expression of CD133, another pancreatic CSC marker identified by FACS and tumorigenicity assay,21 and evaluated it in the same 1010 fields (Fig. S2). Similarly to CD44⁺/CD24⁺/EpCAM⁺, CD133 was significantly correlated with poor differentiation. Therefore, our data imply that poor differentiation might be common characteristics of CSCs in clinical specimens.

As an abundance of CSCs should lead to high tumor propagation, we expected that CD44⁺/CD24⁺/EpCAM⁺ cells would be associated with a poor clinical outcome. However, we were unable to find any significant prognostic value of CD44⁺/CD24⁺/EpCAM⁺ expression among the 65 cases we examined. In contrast, our additional analysis demonstrated that the presence of CD44⁺/CD24⁺ (double‐positive) cells were strikingly correlated with poorer outcome (P = 0.044, Fig. 3b). These differences in the prognostic significance of CD44/CD24 and EpCAM corresponded to the differences in histological characteristics: CD44 and CD24 were significantly correlated with poor differentiation and high proliferation, whereas EpCAM was correlated with well differentiation and low proliferation (Table 3). It is generally understood that poor histological differentiation and high proliferation are indicative of poor outcome in patients with various solid tumors.43, 44 In addition, CD133 was not correlated with poorer outcome (Fig. S2). Therefore, our present findings indicate that the CD44⁺/CD24⁺ combination appears to be a more useful marker when considering targeted therapy for pancreatic cancer in a clinical setting.

In conclusion, we have demonstrated that CD44⁺/CD24⁺/EpCAM⁺ cells have histologically poor differentiation and high proliferation, indicating that they are responsible for the aggressiveness of cancers. Variations in the distribution of CD44⁺/CD24⁺/EpCAM⁺ cells may explain differences in the malignant behavior of pancreatic cancer among patients. We began this study under the assumption that CD44⁺/CD24⁺/EpCAM⁺ cells were the CSCs of pancreatic cancer; however, our survival analysis did not show their significant correlation with prognosis. Rather, the double positive of CD44⁺/CD24⁺ expression, was markedly correlated with a poor outcome, supporting the likelihood that targeting these double positive cells would be of therapeutic value in a clinical setting. Since we did not reveal the actual self‐renewal and differentiative capacities of CD44⁺/CD24⁺/EpCAM⁺ cells or CD44⁺/CD24⁺ cells in our clinical materials, the issues whether these cells we identified with these markers correctively reflect the “bona fide” CSCs remained to be unsolved. We anticipate that our present results indicating the clinical importance of CSC markers will contribute to new forms of cancer therapy as well as clarifying a number of issues related to molecular oncology.

Disclosure Statement

The authors have no conflict of interest.

Supporting information

Fig. S1. Total Scores and their distribution for all 101 cases.

Fig. S2. Immunohistochemical analysis of CD133.

(Cancer Sci 2013; 104: 1127–1134)