J Med Dent Sci 2010； 57： 165－175

Original Article

Corresponding Author: Kazumi Horiguchi, MDDepartment of Surgery, Cancer and Infectious Diseases Center of Tokyo Metropolitan Komagome Hospital, 3-18-22, Hon-komagome, Bunkyo-ku, Tokyo 113-8677, JapanTel: +81-3-3823-2101　Fax: +81-3824-1552E-mail: kazumito@cick.jpReceived February 1；Accepted March 12, 2010

Purpose: We investigated the significance of CD24 and CD44 expression for predicting responses to chemotherapy and prognosis in primary breast cancer patients.Patients and Methods: Diagnosis of breast cancer was confirmed by core needle biopsy, and immunohistochemical studies were performed. Preoperatively, patients received anthracycline-containing chemotherapy. Expression of CD44 and CD24 was assessed immunohistochemically and the relationship with chemotherapy response and with prognosis was analyzed. Results: Between 2001 and 2004, 139 women were enrolled in this study. In the correlation analysis, CD24 expression was negatively associated with pathological response to chemotherapy (p=0.0003). A machine learning technique with an alternating decision tree (ADTree) showed that four logical rules are

Predictive value of CD24 and CD44 for neoadjuvant chemotherapy response and prognosis in primary breast cancer patients

Kazumi Horiguchi*1,2, Masakazu Toi3, Shin-ichiro Horiguchi1,4, Masahiro Sugimoto5,6, Yasuhiro Naito5,6,7, Yukiko Hayashi4, Takayuki Ueno3, Shinji Ohno8, Nobuaki Funata4, Katsumasa Kuroi2, Masaru Tomita5,6,7 and Yoshinobu Eishi1

1) Department of Human Pathology, Tokyo Medical and Dental University Graduate School, Tokyo, Japan2) Department of Surgery, Cancer and Infectious Diseases Center of Tokyo Metropolitan Komagome Hospital, Tokyo, Japan3) Department of Surgery, Graduate School of Medicine, Kyoto University, Kyoto, Japan4) Department of Pathology, Cancer and Infectious Diseases Center of Tokyo Metropolitan Komagome Hospital, Tokyo, Japan5) Institute for Advanced Biosciences, Keio University, Kanagawa, Japan6) Systems Biology Program, Graduate School of Media and Governance, Keio University, Kanagawa, Japan7) Department of Environment and Information Studies, Keio University, Kanagawa, Japan8) Department of Breast Oncology, National Hospital Organization Kyusyu Cancer Center, Fukuoka, Japan

involved in predicting the response depending on the combination of CD24, HER2, tumor stage, CD44, progesterone receptor, and patient age. In the survival analysis, patients having CD44 (++) showed a significantly favorable prognosis as compared with others (p=0.0002). A multivariate analysis showed that CD44 expression had an independent prognostic value (p<0.001).Conclusion: We found a significant correlation between CD44 expression and prognosis and between CD24 expression and response to chemotherapy. CD24 and CD44 expressions would be useful predictive markers, although further studies are needed.

Key words: breast cancer, CD24, CD44, response, prognosis

Introduction

  Systemic chemotherapy, which is still commonly used for primary breast cancer patients, improves survival outcomes and increases the chance for breast conservat ion when app l ied preoperat ive ly 1-4. Furthermore, the pathological complete response (pCR) in the breast is associated with favorable prognosis, whereas non-pCR of the breast or node-positive status

166 J Med Dent SciK. Horiguchi et al.

is not, indicating the need to tailor the treatment to each individual patient. Correlative studies of tumor samples before and after treatment may provide further information on markers that could predict response or resistance to treatment5, 6.  According to recent studies, cancer-initiating cells play a role in disease progression and therapeutic sensitivity, particularly chemotherapy. Experimental investigations have characterized cancer stem cells in hematologic malignancies7, brain tumors8, and breast cancers. For breast cancers, these cells have been identified to have a CD44+/CD24-/low phenotype9 and represent a minor population (10-20%) within primary breast tumors10. Experimentally, cells with this phenotype have been shown to be tumorigenic, multipotent, and capable of generating cells into all the different lineages. In addition, such specific cancer-init iat ing cel ls may be involved in therapeutic resistance11-13.  CD44 is a cell adhesion molecule known to be expressed in most cell types14 and has been associated with stem cells in normal and malignant breast tissues15-18. CD24 is expressed in the early stages of B-cell development and is highly expressed on neutrophi ls, but is absent in normal T cel ls or monocytes19. Although CD24 is not present in healthy adult human tissues, it has been shown to be expressed in human carcinomas20, 21. Although higher CD24 expression as seen by immunohistochemistry has been reported to indicate worse prognosis 21, recent data suggest a different role for CD24 in the carcinogenesis of breast cancer 11, 12. CD24 reduces stromal cell-derived factor-1-mediated migration and signaling via CXCR4 in breast cancer cell lines, suppressing their metastatic potential. On the other hand, the metastatic potential of CD24 (-) cells is increased, as evidenced by their lesser interference with RNA inhibition22. However, the actual function of breast cancer stem cells is still unclear, and the relationship between their role and pat ient response to therapy remains large ly unknown15-18, 23.  In this study, we accessed the predictive value of CD24 and CD44 concern ing the response to preoperative multi-drug cytotoxic chemotherapy, consisting of anthracycline and taxane, and for prognosis. In addit ion, we developed accurate prediction model using data mining technique, in addition to standard statistical analysis, to explore possible rules to predict the efficacy of preoperative chemotherapy with known predictive factors and visualize the dependency of these factors. We aimed to

gain a better understanding of the relationship between the immunohistochemical results of biopsy specimens and the response to the preoperative chemotherapy and postoperative survival in this study.

Patients and methods

Patients:  One hundred thirty-nine patients with operable primary invasive ductal carcinoma were enrolled in this s t udy and were t r ea ted w i t h p reope ra t i ve , anthracyc l ine -based chemotherapy at Tokyo Metropolitan Cancer and Infectious Diseases Center Komagome Hospital or National Kyushu Cancer Center from 2001 to 2004. The median patient age was 49 years (range, 23-74 years) . The major patient characteristics are described in Table Ⅰ. All patients received a diagnosis of invasive ductal carcinoma from a provided diagnostic core needle biopsy specimen that there was sufficient paraffin-embedded tissue before therapy began. The eligibility criteria for this study were as follows: 20-75 years of age, pathological diagnosis of invasive ductal carcinoma by true core needle biopsy, no prior specific cancer treatment, no pr ior h is tory of heart d isease (conf i rmed by electrocardiography), adequate biological functions, and written informed consent. The clinical data of patients are summarized in Table Ⅰ.  This study was conducted in accordance with the Declaration of Helsinki. The protocol was reviewed and approved by the ethics committee of the institutions and written informed consent was obtained from all patients prior to the study.

Tissue Sampling:  Before commencing chemotherapy, each patient underwent a core needle tumor biopsy to evaluate the biological status. Tissues from each patient were fixed in buffered formalin and embedded in paraffin. Four-micrometer - th ick sect ions were s ta ined wi th hematoxylin-eosin and used for immunohistochemical staining.

Immunohistochemical assessment of hormone receptor and HER2:  Estrogen receptor (ER) status and progesterone r e c e p t o r ( P g R ) s t a t u s w e r e d e t e r m i n e d b y immunohistochemistry using J-score 24, 25. In general, tumors with >10% positively stained tumor cells were classified as positive for ER and PgR. HER2 status was also determined by immunohistochemistry or by

167The predictive value of CD24 and CD44

fluorescence in situ hybridization (FISH) analysis. HER2-positive tumors were defined as 3+ using ASCO/CAP guidelines as the staining criteria on immunohistochemistry staining or as positive by FISH26.  For FISH detection of HER2, HER2/CEP17 probe kits (Vysis) were distributed and used according to the manufacturer’s protocol. Signal numbers for the HER2 gene (labeled with SpectrumOrange, Vysis) and the CEP17 gene (labeled with SpectrumGreen, Vysis) were counted in more than 20tumor cells from each site, and the ratio of the HER2/ CEP17 signal numbers was calculated The results were interpreted as positive when the signal ratio of HER2/CEP17 was equal or greater than 2.2 and negative when it was less than 2.2 according to manufacturer’s protocol26.

Immunohistochemical assessment of CD24 and CD44:  Immunohistochemical staining was done with monoclonal antibodies against CD24 (clone 24C02, Neomarkers, Fremont, CA, USA; 1:100) and CD44 (clone DF1485, Novocastra, Newcastle upon Tyne, UK; 1:50). Antigen retrieval for CD24 sections was performed by pretreatment with 0.2% pepsin for 15 min, and for CD44 by heating in 1 mM EDTA at 98°C for 30 min. Immunostaining was performed using the avidin-biotin horseradish peroxidase method (Vectastain elite ABC kit, Vector, Burlingame, CA, USA). For each batch both a positive and negative control (no primary antibody) were run simultaneously. Immunostained samples were evaluated by two pathologists who were unaware of the diagnosis. Expression of CD24 and CD44 was graded both in terms of percentage of staining in each block and the intensity of staining. However, there is no common cut off value on the patterns of staining. In this study, grade was designated according to the following criteria: (–), negative; (+), expression in less than 50% of tumor cells; (++), expression in more than 50% of cells.

Treatment procedures:  Four cycles of FEC (f luorouracil 500 mg/m2, epirubicin 100 mg/m2, and cyclophosphamide 500 mg/m2) administered intravenously (i.v.) on day 1 and every 21 days thereafter were followed by four cycles of docetaxel (DOC) (i.v.) (75 mg/m2) every 21 days administered intravenously (i.v.) on day 1 every 21 days for 81 patients, or six cycles of FAC (fluorouracil 500 mg/m2, doxorubicin 50 mg/m2, and cyclophosphamide 500 mg/m2) administered intravenously (i.v.) on day 1 every 21 days for 58 patients, prior to surgery.

Following chemotherapy and clinical assessment of response, patients underwent surgery, permitting pathological assessment of the response of the breast tumor. If the tumor was too large or invasive for breast-conserving surgery, modified radical mastectomy was recommended. Sentinel lymph node biopsy (SNB) was performed before neoadjuvant chemotherapy for clinical axillary node negative patients to confirm disease stage. Most patients with negative SNB did not undergo surgical clearance of axillary nodes.

Assessment of clinical response:  Clinical response of tumor was assessed by ultrasound, mammography, and physical examination. Tumor response was assessed using the Response Evaluation Criteria in Solid Tumors (RECIST) guidelines in patients who had measurable lesions27. Clinical assessments were performed within 4 weeks prior to FEC-DOC or FAC treatment, and before surgery.

Assessment of pathological response:  After preoperative chemotherapy, patients underwent appropriate surgery. Pathologic response of tumor and dissected lymph nodes were classified according to the evaluation criteria of the Japanese Breast Cancer Society28, 29, using a 5-grade scale (Grade 0, 1a, 1b, 2, and 3) as follows: Grade 0, no response or almost no change in cancer cells after treatment; Grade 1, slight response; Grade 1a, mild response, mild change in cancer cells regardless of the area, or marked changes in cancer cells in less than one-third of total cancer cells; Grade 1b, moderate response, marked changes in one-third or more but less than two-thirds of tumor cells; Grade 2, marked response or marked changes in two-thirds or more of tumor cells; and Grade 3, pathological complete response (pCR), necrosis or disappearance of all tumor cells, or replacement of all cancer cells by granuloma-like and/or fibrous tissue.  In the case of complete disappearance of cancer cells, pretreatment pathological evidence of the presence of cancer is necessary28, 29.

Statistical analysis:  The primary endpoint was to examine the predictive factors of pathological therapeutic effect stratified by pathological responses. Kaplan-Meier analysis was used to estimate the values of relapse-free survival (RFS). RFS was compared using a log-rank test. Events for the calculation of RFS included all local, regional, or distant recurrence. In the logistic regression analyses, adjustments were made

168 J Med Dent SciK. Horiguchi et al.

for the stratification variables of menopause status, tumor stage, ER status, PgR status, HER2 status, and clinical and pathological response to treatment. Analyses were performed with JMP® (version 8, SAS Institute Inc.). Analyses of endpoint data reported here are based on information received as of September 2009.

Machine learning method with ADTree:  Machine learning technique used in this study is an alternating decision trees (ADTree) method combining boosting and decision tree algorithms to generate classification rules 30. Traditional decision tree algorithms, such as if-then type tree and classification and regression trees (CART), have been successful in generating classifiers, but these classifiers were usually boosted to obtain higher accuracy and versatility; i.e., multiple trees were simultaneously used to classify a given problem. Such conjunct structures often represent convoluted decision rules that are difficult to interpret. In contrast, an ADTree generates simpler decision-tree structures and easier-to-interpret classification rules. During the training of the model, each feature was assigned as a child node to a parent node with dependency condition and numerical score. The condition and scores were automatically calculated following to the AdaBoost algorithm31 In this study, we present the results of ADTrees applied in conjunction with immunohistochemical features. Data mining software Weka (ver. 3-6-2) 32 was used for these analyses.

Results

  One hundred thirty-nine patients were enrolled in this study. The major patient characteristics are described in Table Ⅰ. The nuclear grade of cancer cells was used in this study to assess histological tumor property. ER, PgR, and HER2 status were determined with the biopsied specimens. Eighty one patients were treated by FEC followed DOC regimen and 58 patients were treated by FAC regimen as neoadjuvant chemotherapy. The median clinical follow-up was 56 months. Disease recurrence was observed in 36 patients during the follow-up period. A total of 23 patients died, with all deaths attributed to breast cancer. The overall clinical response was 80 (57.6%), with 16 (11.5%) complete response and 64 (46.0%) partial response. The number of patients who achieved pathological grade 2 and

grade 3 responses were 23 (16.5%) and 11 (7.9%), respectively. The results of response to treatment are shown in Table Ⅰ.  Immunohistochemical patterns of CD24 and CD44 expression analyzed from core needle biopsy sam-ples taken from patients before preoperative chemo-therapy were shown in Figure 1. Patients with a worsening clinical stage showed a lower expression of CD44. The relationships between these markers and the clinical data are shown in Table Ⅰ. A signifi-cant correlation between CD24 expression and pathological tumor response was detected that pat-ents with a lower expression of CD24 showed a bet-ter response (p=0.0003, Spearman rank correlation). There was no significant association between CD44 expression and CD24 expression (χ square test, p=0.3660). No significant correlation between CD24/CD44 combined status and pathological re-sponse was demonstrated (Kruskal-Wallis rank test, p=0.3789). A multiple logistic regression analysis was performed to examine which factors among tu-mor stage, ER, HER2, nuclear grade, CD24, and CD44 were associated with pathological response. CD24 expression and tumor size were associated with pathological response grade significantly (CD24 (-) vs. (+) and (++); χ2=14.698 and p=0.0006, T1+2 vs. T3+4; χ2=12.697 and p=0.0017, respectively).  As to the identification of the rules or factors by ADTree, we planned to find the rules to search the rate of prospection possibility to archive pathologi-cal complete response; pCR. Patients with We classi-fied the pathologic response into two groups: a high-er response group (grade 3) and a lower response group (grades 0, 1, 1a, 1b, or 2), and trained the ADTree classifier to discriminate the differences be-tween these two groups. The prediction accuracy of the ADTree increases according to the number of boosting. Although a larger boosting number might have a higher discriminating capability, the obtained tree is apt to be over-fit for specific data. To prevent over-fitting and to minimize the boosting number in order to facilitate the tree’s generalization ability, we used boosting number 7, which was the simplest structure with adequate discriminating ability (Figure 2a). The probabilities calculated by the tree of the higher responder and the lower responder and re-ceiver operating characteristics (ROC) curves are shown in Figure 2b and 2c, respectively.. The area under ROC (AUC) of this decision tree for the full data was 0.981 (95% CI; 0.961-1.000) (P < 0.0001). Although some probabilities were not correctly pre-

169The predictive value of CD24 and CD44

dicted, these tended to cluster around the middle probabilities (mostly 50-60%), which indicated that no patient was clearly predicted as being in an ad-verse group. To obtain relatively unbiased estimates, we used bootstrapping test to generate 200 repeti-tions of the datasets by randomly selecting individu-als, permitting redundant selection. The mean AUC values of the boostrapped datasets were 0.987 ± 0.00919 (standard deviation).  Relapse-free survival (RFS) for patients whose tumors had CD44 (++) was significantly longer than that for patients whose tumors had CD44(-) and CD44(+) (p=0.0002 by log-rank test) (Figure 3). Ninety three percent of patients with indication of hormone therapy were treated by postoperative adjuvant hormone therapy, and anti-HER2 antibody therapy was applied to the limited number of the HER2-positive patients (Table Ⅰ). A multivariate analysis with Cox’s proportional hazard model using postoperative information showed that CD44 expression status, ER status, and nodal status were

significantly associated with RFS (p<0.0001, p=0.0001, p=0.0007, respectively) (Table Ⅱ). The examination of the CD24/CD44 expression and survival by using log-rank test whether there was any correlation between them was impossible because the survival curves crossed at many points.

Discussion

  In this study, 27% had high expression of CD24 and 36% of tumors had high CD44 positivity. No expression of CD24 was observed in 16% of tumors. We found a significant association between tumor size and CD44 or CD24 expression. The CD44 positive rate was higher in smaller sized tumors than in larger sized tumors, whereas the CD24 negative rate was the highest in T2 tumors as compared with T1 and T3/4 tumors, indicating that the breast cancer stem cell population was altered by tumor progression. Since there st i l l remains much discussion concerning the cut-off line in defining

CD44 (++) CD44 (+) CD44 (-)

CD24 (++) CD24 (+) CD24 (-)

10 m

Figure 1 : Immunohistochemical staining patterns with CD44 and CD24 on formalin-fixed, paraffin-embedded tissue sections of breast carcinoma. Expression of CD24 and CD44 was graded both in terms of percentage of staining in each block and the intensity of staining. Grade was designated according to the following criteria: (–), negative; (+), expression in less than 50% of tumor cells; (++), expression in more than 50% of cells.

170 J Med Dent SciK. Horiguchi et al.

CD44 and CD24 status, we examined different cut-off l ines for each status, but no signif icant association with any clinico-pathological parameters other than tumor s ize was detected in the correlation analysis.  Since a CD44+/CD24-/low phenotype is reported to be associated with breast cancer stem cells, the combination status of CD24 and CD44 was also examined, but no significant association with any clinico-pathological parameters was detected. Several recent studies have indicated that the CD44+/CD24-/low phenotype or CD44 expression correlate positively with basal-like features in non-treated primary breast cancers. In particular, it has been reported that primary breast cancers with BRCA1 disorders or metaplastic features are

enriched in CD44+/CD24-/low phenotype tumor cells. The proportion of these types of tumors would be small in the present study. We failed to find any association with breast cancer phenotypes; however, further details need to be examined33, 34.  In the present study, CD24 pre-treatment status was associated significantly with clinical and pathological tumor response in a univariate analysis. The pathological response of grade 3 (total disappearance of tumor cells) developed more frequently in CD24 (-) tumors than other tumors. Although the mechanisms are largely unknown, CD24 has been shown to promote tumor cell binding to the extracellular matrix, including fibronectin, through the activation of integrins 35. Since tumor-matrix interaction has been implicated in resistance

Figure 2 : (A) The ADTree structure for classifying higher responders (grade 3) and lower responders (grades 0, 1, 1a, 1b, or 2). A negative score in this tree represents a lower responder, while a positive score represents a higher responder. To calculate the score of this tree, all the scores reachable from root node are summed. The child nodes linked with dashed line are always reachable when the parent node is reachable; e.g. HER2 (node 1 and 4), CD24 (node 2), and stage (node3) are always reachable from root node. In contrast, the child nodes linked with solid lines are reachable when the condition is satisfied; e.g. stage (node 6) is reachable when CD24 (node 2) is 1 and more.

(B) The distribution of probabilities for higher responders of pathological therapeutic effect which were calculated by the ADTree scores. The red and blue plots are the predicted probabilities of higher responders and lower responders, respectively. The long black bars indicate the mean value of each group, and short black bars represent the standard deviations.

(C) The receiver operating characteristics curve of the developed ADTree.

171The predictive value of CD24 and CD44

to chemotherapy36, this interaction through integrins may explain the lower frequency of pathological response in tumor cells with CD24 expression. In addition, the analysis with the ADTree method showed that the trained tree consists of tumor stage, age, and the expression levels of CD24, CD44, PgR, and HER2, indicating the importance of CD24 in chemotherapy response. The trained tree showed a high prediction performance; however, this should be carefully validated in a larger study, since the analytical results of machine learning methods strongly depend on the distribution of the variables.  According to Li et al., the CD44+/CD24-/low subpopulation is enriched by 12 weeks of treatment with cytotoxic chemotherapy in primary breast cancer tissues. This subpopulation might be involved in chemotherapy resistance and possibly in chemotherapy sensitivity as well 37. In other types of cancers, such as colorectal and pancreatic cancer, it is known that CD44+/CD24-/low subpopulations exhibit a resistant phenotype to chemotherapy or to radiation38, 39. In this study, we did not find a correlat ion between the CD44+/CD24-/ low phenotype and chemotherapy response. A recent paper also showed that CD44+/CD24-/low tumor cell proportions are not associated with the pathological response rates by neoadjuvant chemotherapy, while aldehyde dehydrogenase (ALDH)1-positive tumors are associated with a low pCR rate40. Thus, further studies are needed in terms of chemotherapy sensitivity in association with cancer stem cell markers. In addition, ALDH1 is induced in the

stromal cells by chemotherapy and the stromal induction of ALDH1 correlates with favorable prognosis. It is necessary to further examine tumor stromal cells as well as cancer cells41.  The prognostic implications of CD24 and CD44 expression in primary breast cancer are still unknown. Our data showed that there was a significant association with increased RFS for patients whose tumors had high CD44 expression. Similar findings have been reported for breast cancer42. In the multivariate analysis, CD24 status showed no significant prognostic value, whereas CD44 status exhibited a significant value. There are several studies showing that CD24 expression is re l evan t t o poor p rognos i s43. A d i f f e ren t categorization of CD24 expression status might be responsible for the discrepancy between our study and the previous studies44. Another possibility is a difference in the type of systemic adjuvant therapies. In neoadjuvant chemotherapy, it is uncommon to see recurrence in patients who achieve a remarkable pathological response such as grade 3. It may be that the CD24 (-) group consists of a heterogeneous population, including favorable prognosis and unfavorable prognosis subgroups, because remarkable pathological responses were detected in CD24 (-) tumors.  The prognostic value of the high expression of CD44 might be independent of chemotherapy sensitivity because the multivariate analysis showed a consistently significant prognostic value of CD44 expression before and after chemotherapy.

Figure 3 : Kaplan-Meier relapse-free survival curves for patients in relation to CD44　expression. The higher CD44 expression showed the better prognosis (p=0.0002, Log- rank test).

172 J Med Dent SciK. Horiguchi et al.

CD44 CD24(-) (+) (++) (-) (+) (++)

Variables n=139 46 43 50 22 80 37

Menopausepre 68 24 20 24 12 43 13post 71 22 23 26 10 37 24P valuea) 0.856 0.147

T (1,2,3-)1 13 2 3 8 1 6 62 50 14 8 28 10 22 183- 76 30 32 14 11 52 13P valueb) <0.0001* 0.0298*

N (0,1)0 22 8 6 8 2 11 91 117 38 37 42 20 69 28P valuea) 0.905 0.222

n (0,1)0 49 11 14 24 13 17 191 90 35 29 26 9 63 18P valuea) 0.0431* 0.0003*

Clinical stage (I, II, III) I 8 2 1 5 1 5 2 II 60 16 12 32 11 27 22 III 71 28 30 13 10 48 13P valueb) 0.0004* 0.113

Estrogen receptor (ER)Positive 100 35 30 35 15 56 29Negative 39 11 13 15 7 24 8P valuea) 0.746 0.588

Progesterone receptor (PgR)Positive 72 25 21 26 9 44 19Negative 67 21 22 24 13 36 18P valuea) 0.873 0.503

HER2 statusPositive 26 11 10 5 4 18 4Negative 113 35 33 45 18 62 33P valuea) 0.142 0.320

Nuclear gradeGrade 1 28 9 7 12 3 15 10Grade 2 46 14 16 16 9 23 14Grade 3 65 23 20 22 10 42 13P valueb) 0.875 0.384

Table Ⅰ. Patient characteristics, immunohistochemical results, and therapeutic effects

173The predictive value of CD24 and CD44

However, it has been revealed that CD44-targeting therapy results in a significant reduction in the growth of human breast cancer xenografts and in preventing tumor relapse after chemotherapy-induced remission, suggesting that the biological role of CD44 expression needs to be investigated

further45.  The developed ADTree (Fig. 2a) accurately predicted the response to chemotherapy, making use of expression CD24 and CD44 with known other predictive features. The AUC of the ADTree is higher than that of logistic regression model (AUC=0.923,

Clinical therapeutic effectCR / PR 80 24 23 33 18 43 19NC 57 22 20 15 4 37 16PD 2 0 0 2 0 0 2P valueb) 0.144 0.0207*

Pathological therapeutic effectGrade 0 /1 105 35 30 40 12 65 28Grade 2 23 6 12 5 3 13 7Grade 3 11 5 1 5 7 2 2P valueb) 0.100 0.0003*

Median age (range) 50 (23-74)Median follow-up (days) 1701 (66-3310)Neoadjuvant therapy regimen FEC followed by DOC cases (%) 81 (58) FAC cases (%) 58 (42)Adjuvant therapy cases (%) 100 (72) Chemotherapy 11 (8) Hormone therapy 93 (67) Anti-HER2 antibody therapy 4(3) None 39 (28)Recurrence cases (%) 36 (26)Death cases (%) 23 (17)

Abbreviations: N, preoperative lymph node status; n, postoperative lymph node status; CR, complete response; PR, partial response; NC, no change; PD, progressive disease; FAC, fluorouracil, doxorubicin, cyclophosphamide;FEC, fluorouracil, eprubicin, cyclophosphamide; DOC, docetaxela) Mann-Whitney U Testb) Spearman rank correlation* Statistical significance was accepted when P values were less than 0.05

Before treatment After treatmentVariables Risk Ratio 95% Confidence Interval P value Risk Ratio 95% Confidence Interval P valueT 0.1439537 0.4570987-1.6134707 <.0001*N 0.5086225 -0.236567-0.828574 0.2293n 4.8881698 -1.412803-0.309347 0.0007*ER 0.2850745 0.236632-1.0340617 0.0016* 0.2036662 0.3895433-1.2246029 0.0001*HER2 1.5994566 -0.625156-0.1605299 0.2408 1.60498 -0.613037-0.1468202 0.2227NG 0.7488043 -0.231102-0.5387854 0.4548 1.1003579 -0.452785-0.369082 0.8191CD44 0.2308185 0.1896048-1.465116 0.0058* 0.1192423 0.5355242-1.7841924 <.0001*CD24 0.7636982 -0.376732-0.7699971 0.6282 0.5723285 -0.220923-0.9044212 0.292PTE 0.4473238 -0.01978-0.9062774 0.0626

Abbreviations: ER, estrogen receptor; NG, nuclear grade; PTE, pathological therapeutic effect*Statistical significance was accepted when P values were less than 0.05

Table Ⅱ. Cox proportional hazards model for relapse-free survival

174 J Med Dent SciK. Horiguchi et al.

References1. Early Breast Cancer Trialists' Collaborative Group

(EBCTCG). Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomized trials. Lancet. 2005;365:1687-1717.

95% CE: 0.845 – 1.00). This tree and also implied feature dependency. The calculated probability is always affected by HER2 (node 1 and 4) and the expression of CD24 (node 2) and tumor stage (node 3). Only when the CD24 expression level is 1 and more, the tumor stage (node 6) also affects the probability. PgR (node 5) affects the probability when HER2 (node 4) is 0 or 1, and the expression of CD44 (node 7) affects when HER2 is 2 or 3. The final score zero yielded by the three is the threshold, and positive and negative score indicate the high and low responder, respectively. As an example of interpretation, consider a patient with stage, 2 PgR, negative; HER2, 1; CD44, 3; and CD24, 0. The score is -1.187 (root node) +0.22 (node 1) + 0.692 (node 2) + 0.324 (node 3) - 0.444 (node 4) +0.409 (node 5). Thus the final score of this tree is 0.014, which indicates a higher responder. Age and ER status were not selected by the model training algorithm in the consequence of its low contribution to the prediction.

  In conclusion, CD24 expression was relevant to the sensitivity of chemotherapy with anthracycline and taxane. CD44 expression was significantly associated with the prognosis of primary breast cancer pat ients . The mul t ivar iate analys is confirmed the independent prognostic value of CD44 expression. Further studies are needed to understand the mechan isms under ly ing chemotherapy resistance and cancer recurrence in terms of CD24 and CD44 expression in greater depth.

Acknowledgements

  This study was funded by a research grant from Japan’s Ministry of Health, Labor, and Welfare for a study on constructing an algorithm for multimodality therapy with biomarkers for primary breast cancer during formulation of the decision making process, led by Masakazu Toi (H18-3JIGAN-IPPAN-007, H19-3JIGAN-IPPAN-007).

2. Kuerer HM, Newman LA, Smith TL, et al. Clinical course of breast cancer patients with complete pathologic primary tumor and axillary lymph node response to doxorubicin-based neoadjuvant chemotherapy. J Clin Oncol. 1999;17:460-469.

3. Kaufmann M, Hortobagyi GN, Goldhirsch A, et al. Recommendations from an international expert panel on the use of neoadjuvant (primary) systemic treatment of operable breast cancer: an update. J Clin Oncol. 2006;24:1940-1949.

4. Bear HD, S. Anderson S, A. Brown A, et al. The effect on tumor response of adding sequential preoperative d o c e t a x e l t o p r e o p e r a t i v e d o x o r u b i c i n a n d cyclophosphamide: preliminary results from National Surgical Adjuvant Breast and Bowel Project Protocol B-27. J Clin Oncol. 2003; 21:4165-4174.

5. McShane LM, Altman DG, Sauerbrei W, et al. REporting recommendations for tumor MARKer prognostic studies (REMARK). Breast Cancer Res Treat. 2006;100 :229-235.

6. Chang JC, Wooten EC, Tsimelzon A, et al. Patterns of resistance and incomplete response to docetaxel by gene expression profiling in breast cancer patients. J Clin Oncol. 2005; 23:1169-1177.

7. Huntly BJ, Gilliland DG. Leukaemia stem cells and the evolution of cancer-stem-cell research. Nat Rev Cancer. 2005;5:311-321.

8. Singh SK, Hawkins C, Clarke ID, et al. Identification of human brain tumour in i t iat ing cel ls . Nature. 2004;432:396-401.

9. Al-Hajj M, Wicha MS, Benito-Hernandez A, et al. Prospective identification of tumorigenic breast cancer cells. Proc Natl Acad Sci USA. 2003;100:3983-3988.

10. Abraham BK, Fritz P, McClellan M, et al. Prevalence of CD44+/CD24-/low cells in breast cancer may not be associated with clinical outcome but may favor distant metastasis. Clin Cancer Res.2005;11:1154-1159.

11. Ponti D, Costa A, Zaffaroni N, et al. Isolation and in vitro propagation of tumorigenic breast cancer cells with stem/progenitor cell properties. Cancer Res. 2005; 65: 5506-5511.

12. Al-Hajj M, Becker MW, Wicha M, et al. Therapeutic implications of cancer stem cells. Curr Opin Genet Dev. 2004;14:43-47.

13. Creighton CJ, Li X, Landis M, et al. Residual breast cancers after conventional therapy display mesenchymal as well as tumor-initiating features. Proc Natl Acad Sci U S A. 2009;106:13820-13825.

14. Goodison S, Urquidi V, Tarin D. CD44 cell adhesion molecules. Mol Pathol. 1999;52:189-196.

15. Mani SA, Guo W, Liao MJ, et al . The epithel ial -mesenchymal transition generates cells with properties of stem cells. Cell. 2008;133:704-715.

16. Dontu G. Breast cancer stem cell markers - the rocky road to clinical applications. Breast Cancer Res. 2008;10:110.

17. Bloushtain-Qimron N, Yao J, Snyder EL, et al. Cell type-specific DNA methylation patterns in the human breast.

175The predictive value of CD24 and CD44

Proc Natl Acad Sci USA. 2008;105:14076-14081.18. Hebbard L, Steffen A, Zawadzki V, et al. CD44 expression

and regulation during mammary gland development and function. J Cell Sci. 2000;113 (Pt 14):2619-2630.

19. Aigner S, Sthoeger ZM, Fogel M, et al. CD24, a mucin-type glycoprotein, is a ligand for P-selectin on human tumor cells. Blood. 1997;89:3385-3395.

20. Jackson D, Waibel R, Weber E, et al. CD24, a signal-transducing molecule expressed on human B cells, is a major surface antigen on small cell lung carcinomas. Cancer Res. 1992;52:5264-5270.

21. Kristiansen G, Schluns K, Yongwei Y, et al. CD24 is an independent prognostic marker of survival in nonsmall cell lung cancer patients. Br J Cancer. 2003;88: 231-236.

22. Schabath H, Runz S, Joumaa S, et al. CD24 affects CXCR4 function in pre-B lymphocytes and breast carcinoma cells. J Cell Sci. 2006;119:314-325.

23. Hennessy BT, Gonzalez-Angulo AM, Stemke-Hale K, et al. Characterization of a naturally occurring breast cancer subset enriched in epithelial-to-mesenchymal transition and stem cell characteristics. Cancer Res. 2009;69:4116-4124.

24. Umemura S, Kurosumi M, Moriya T, et al. Imunohistochemical evaluation for hormone receptor s in breast cancer : a practically useful evaluation system and handling protocol. Breast Cancer 2006:13:232-235.

25. Kurosumi M. Immunohistochemical assessment of hormone receptor status using a new scoring system (J-score) in breast cancer. Breast Cancer 2007:24:189-193.

26. Wolff AC, Hammond ME, Schwartz JN, et al. American Society of Clinical Oncology/College of American Pathologists guideline recommendations for human epidermal growth factor receptor 2 testing in breast cancer. Arch Pathol Lab Med 2007:131:19-26.

27. Therasse P, Arbuck SG, Eisenhauer EA, et al. Newguide-lines to evaluate the　 response to treatment in solid tumors. European Organization for Research and Treat-ment of Cancer, National Cancer Institute of the United States, National Cancer Institute of Canada. J Natl Can-cer Inst. 2000;92:205-216.

28. Kurosumi M. Significance and problems in evaluations of pathological responses to neoadjuvant therapy for breast cancer. Breast Cancer. 2006;13:254-259.

29. Kurosumi M, Akashi-Tanaka S, Akiyama F, et al . Histopathological criteria for assessment of therapeutic response in breast cancer (2007 version). Breast Cancer. 2008; 15:5-7.

30. Freund Y, Mason L. The Alternating Decision Tree Algorithm, Proceedings of the 16th International Conference on Machine Learning, pp124-133, 1999.

31. Liu KY, Lin J, Zhou X, et al. Boosting alternating decision

trees modeling of disease trait information. 2005: 6: S13232. Witten I & Frank E. Data mining: Practical Machine

Learning Tools and Techniques. Morgan Kaufmann, San Francisco, 2005.

33. Honeth G, Bendahl PO, Ringner M, et al. The CD44+/CD24- phenotype is enriched in basal-like breast tumors. Breast Cancer Res. 2008;10: R53. [Epub ahead of print]

34. Klingbeil P, Natrajan R, Everitt G, et al. CD44 is overexpressed in basal-like breast cancers but is not a driver of 11p13 amplification. Breast Cancer Res Treat. 2009; [Epub ahead of print]

35. Baumann P, Cremers N, Kroese F, et al. CD24 expression causes the acquisition of multiple cellular properties associated with tumor growth and metastasis. Cancer Res. 2005;65:10783-10793.

36. Aoudjit F, Vuori K. Integrin signaling inhibits paclitaxel-induced apoptosis in breast cancer cells. Oncogene. 2001;20:4995-5004.

37. Li X, Lewis MT, Huang J, et al. Intrinsic resistance of tumorigenic breast cancer cells to chemotherapy. J Natl Cancer Inst. 2008;100:672-679.

38. Dallas NA, Xia L, Fan F, et al. Chemoresistant colorectal cancer cells, the cancer stem cell phenotype, and increased sensitivity to insulin-like growth factor-I receptor inhibition. Cancer Res. 2009;69:1951-1957.

39. Lee CJ, Li C, Simeone DM. Human pancreatic cancer stem cells: implications for how we treat pancreatic cancer. Transl Oncol. 2008;1:14-18.

40. Tanei T, Morimoto K, Shimazu K, et al. Association of breast cancer stem cells identified by aldehyde dehydrogenase 1 expression with resistance to sequential Paclitaxel and epirubicin-based chemotherapy for breast cancers. Clin Cancer Res. 2009;15:4234-4241.

41. Resetkova E, Reis-Filho JS, Jain RK, et al. Prognostic impact of ALDH1 in breast cancer: a story of stem cells and tumor microenvironment. Breast Cancer Res Treat. 2009; [Epub ahead of print]

42. Diaz LK, Zhou X, Wright ET, et al. CD44 expression is associated with increased survival in node-negative invasive breast carcinoma. Clin Cancer Res. 2005;11:3309-3314.

43. Lee JH, Kim SH, Lee ES, et al. CD24 overexpression in cancer development and progression: a meta-analysis. Oncol Rep. 2009;22:1149-1156.

44. Aulmann S, Waldburger N, Penzel R, et al. Reduction of CD44(+)/CD24(-) breast cancer cells by conventional cytotoxic chemotherapy. Hum Pathol. 2009; [Epub ahead of print]

45. Marangoni E, Lecomte N, Durand L, et al. CD44 targeting reduces tumour growth and prevents post-chemotherapy relapse of human breast cancers xenografts. Br J Cancer. 2009;100:918-922.