Breast Cancer Res Treat (2009) 113:275–283 DOI 10.1007/s10549-008-9939-y
PRECLINICAL STUDY
Comparison of gene expression profiles predicting progression in breast cancer patients treated with tamoxifen Marleen Kok Æ Sabine C. Linn Æ Ryan K. Van Laar Æ Maurice P. H. M. Jansen Æ Teun M. van den Berg Æ Leonie J. M. J. Delahaye Æ Annuska M. Glas Æ Johannes L. Peterse Æ Michael Hauptmann Æ John A. Foekens Æ Jan G. M. Klijn Æ Lodewyk F. A. Wessels Æ Laura J. Van’t Veer Æ Els M. J. J. Berns
Received: 6 February 2008 / Accepted: 8 February 2008 / Published online: 4 March 2008 Ó Springer Science+Business Media, LLC. 2008
Abstract Background Molecular signatures that predict outcome in tamoxifen treated breast cancer patients have been identified. For the first time, we compared these response profiles in an independent cohort of (neo)adjuvant systemic treatment naı¨ve breast cancer patients treated with first-line tamoxifen for metastatic disease. Methods From a consecutive series of 246 estrogen receptor (ER) positive primary tumors, gene expression profiling was performed on available frozen tumors using 44K oligoarrays (n = 69). A 78-gene tamoxifen response profile (formerly consisting of 81 cDNA-clones), a 21-gene set (microarraybased Recurrence Score), as well as the HOXB13-IL17BR Electronic supplementary material The online version of this article (doi:10.1007/s10549-008-9939-y) contains supplementary material, which is available to authorized users. M. Kok J. L. Peterse L. J. Van’t Veer (&) Department of Pathology, The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands e-mail:
[email protected] S. C. Linn T. M. van den Berg Department of Medical Oncology and Molecular Biology, The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands
ratio (Two-Gene-Index, RT-PCR) were analyzed. Performance of signatures in relation to time to progression (TTP) was compared with standard immunohistochemical (IHC) markers: ER, progesterone receptor (PgR) and HER2. Results In univariate analyses, the 78-gene tamoxifen response profile, 21-gene set and HOXB13-IL17BR ratio were all significantly associated with TTP with hazard ratios of 2.2 (95% CI 1.3–3.7, P = 0.005), 2.3 (95% CI 1.3–4.0, P = 0.003) and 4.2 (95% CI 1.4–12.3, P = 0.009), respectively. The concordance among the three classifiers was relatively low, they classified only 45–61% of patients in the same category. In multivariate analyses, the association remained significant for the 78-gene profile and the 21-gene set after adjusting for ER and PgR. Conclusion The 78-gene tamoxifen response profile, the 21-gene set and the HOXB13-IL17BR ratio were all significantly associated with TTP in an independent patient series treated with tamoxifen. The addition of multigene assays to ER (IHC) improves the prediction of outcome in tamoxifen treated patients and deserves incorporation in future clinical studies. Keywords Breast cancer Gene expression profiling Tamoxifen Endocrine response Estrogen receptor
R. K. Van Laar L. J. M. J. Delahaye A. M. Glas Agendia BV, Amsterdam, The Netherlands
Introduction M. P. H. M. Jansen J. A. Foekens J. G. M. Klijn E. M. J. J. Berns Department of Medical Oncology of the Erasmus MC, Josephine Nefkens Institute and Daniel den Hoed, Rotterdam, The Netherlands M. Hauptmann L. F. A. Wessels Department of Molecular Biology, The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands
Adjuvant tamoxifen treatment reduces the risk of breast cancer death by 31% in patients with ER-positive disease [1]. However, a substantial proportion of patients develops metastases despite tamoxifen treatment. Moreover, only half of the recurrences in ER-positive breast tumors responds to tamoxifen while the other half is resistant [2]. Although aromatase inhibitors do prolong disease-free
123
276
survival compared to tamoxifen [3], tamoxifen still has a major role in breast cancer treatment [4]. In addition, a survival benefit has been shown for sequential tamoxifen and an aromatase inhibitor [5, 6]. A molecular test helping clinicians to make a choice between starting with either tamoxifen, an aromatase inhibitor or rather with chemotherapy would have enormous potential for tailoring treatment. Molecular signatures that correlate with outcome of breast cancer patients who received adjuvant tamoxifen treatment, have been identified, namely a 21-gene set (Recurrence Score) and a HOXB13-IL17BR ratio [7, 8]. A disadvantage of assessing response in the adjuvant setting is that both response of tumor cells to tamoxifen as well as intrinsic aggressiveness of the malignancy are measured. Furthermore, some resistant tumors will not recur because they were already cured by surgery and radiation. The proportion of this group of patients is unknown. In contrast, studies in the metastatic disease setting provide information on decrease of tumor load after treatment and thereby on sensitivity to a specific drug. Even in cases where the interval between primary tumor and metastasis is long, gene profiles are maintained in metastatic lesions [9]. In this view, an 81-gene signature using primary tumors of patients treated with tamoxifen for metastatic disease was developed [10]. Using this signature, patients who had a short TTP could be identified. This study showed the relevance of profiling of the primary tumor for response prediction in the metastatic disease setting. Ultimately, guided decisions for adjuvant treatment are still most crucial and could result in significant survival benefit. However, therapy response can only be defined in patients with measurable disease. Therefore, validation of predictive markers in the metastatic disease setting is desirable before tests are applied to the adjuvant treatment setting. To provide a robust estimate of the likelihood of response to tamoxifen predicted by the 81-gene profile, 21gene set and HOXB13-IL17BR ratio, we compared these profiles in an independent cohort of (neo)adjuvant systemic treatment naı¨ve breast cancer patients treated with first-line tamoxifen for metastatic disease.
Methods Patients A consecutive series of 246 patients (diagnosed 1984– 1997) was selected from the medical registry of the Netherlands Cancer Institute according to following criteria: (1) invasive breast carcinomas, (2) relapse of disease before 2002, for which first-line tamoxifen had been given,
123
Breast Cancer Res Treat (2009) 113:275–283
(3) no adjuvant systemic treatment, (4) patients had received tamoxifen for at least 4 weeks, which is an adequate period to allow judgment of therapeutic effect. Patients who had bilateral cancer or had developed either contra lateral or a second primary breast tumor were excluded (n = 35). After reviewing medical charts, 10 patients who had received surgery or radiation for all metastatic lesions were excluded. In addition, 21 patients were excluded because no follow-up was available. After pathology revision, 22 patients with an ER-negative tumor were excluded. Of the remaining 158 patients, for 77 patients frozen tumors were available. Finally, 8 patients were excluded either due to lack of tumor cells in frozen material or no good quality RNA could be extracted. The analyses presented here are based on 69 patients. According to routine practice, patients attended the hospital every 4–8 weeks and were restaged (by imaging, physical examination and biochemical markers) if indicated based on presented symptoms. During tamoxifen treatment, eight women had undergone ovarian ablation. Besides bisphosphanates, no systemic treatment had been prescribed. This study received approval of the Ethical Committee. Gene expression profiling RNA isolation was performed as described previously [11]. RNA amplification, labeling and hybridization to a 44K oligoarray (Agilent Technologies) was performed as described previously [12]. Intensities were normalized using Feature Extraction software version 7.5. 78-gene tamoxifen response profile Training (data online in Supplements) Since the 81-gene profile [10] had been discovered using 19K cDNA arrays, we re-hybridized the original training set (n = 46, Erasmus MC) on the more advanced 44K oligoarray platform. For 40 tumors (22 non-responders, 18 responders) out of this training set, sufficient RNA was available. To allow comparison of the training with the validation set (Netherlands Cancer Institute) described in this paper, a median centering was applied to the genes separately for each hospital. After this correction, log-ratios were in the same range in the training and validation set. We mapped, based on Genbank accessions and Unigene identifiers (version 2005), 78 of the 81 genes to the 44K oligoarray. These 78 genes were represented by 161 oligos. A univariate t-test (BRB Array Tools, version 3.4.0) was used to select the oligo per gene that was most differentially expressed among classes. The 78 oligos are listed online in the Supplements. Hereafter the 81-gene profile is referred to as ‘78-gene profile’.
Breast Cancer Res Treat (2009) 113:275–283
277
In the previous study [10], an algorithm comparable to the compound covariate predictor had been used. Using the oligoarray data, this algorithm left 10% unclassified. To overcome this problem, a slightly different algorithm was used: the nearest centroid classifier [11, 13, 14]. A template was created containing the mean expression of the nonresponders. For all tumors the Pearson correlation with this template was calculated. To minimize the risk of misclassifying tumors with a non-responder profile, the threshold was set at a correlation of 0.0750 which resulted in a smaller than 10% false negative rate. Results of the training set are available online in the Supplements.
Immunohistochemistry (IHC) Biopsies of 600-lm were taken from each paraffinembedded tumor and arrayed in triplicate in a new paraffin block (tissue microarray). IHC for ER-alpha and HER2 and additional chromogenic in situ hybridization for HER2, was performed and scored as described previously [16, 17]. The primary antibody directed against PgR (R-1, Klinipath, Duiven, Netherlands) was diluted 1:500. Staining for ER and PgR was interpreted as positive when more than 10% of tumor cells were stained [18, 19]. Statistics
Validation (data presented in paper) For each tumor, a Pearson correlation with the ‘nonresponder’ template of the training set was calculated. When the correlation was more than 0.0750, a patient was assigned to the non-responder profile group. 21-gene set (Recurrence Score) To classify tumors using the Recurrence Score, we used microarray data for all 21 genes (mapped to 44K array based on Genbank and Unigene identifiers) and applied the algorithm described by Paik et al. [7]. Briefly, expression of 16 target genes was normalized relative to 5 reference genes. If genes were represented by more probes, values were averaged. Next, data were transformed from log-10 to log-2 and target genes were scaled as described by Paik et al. [7]. Using cut-offs described in Paik et al., we assigned patients into the Low, Intermediate or High Risk group, respectively.
Disease-Free Interval (DFI) was measured using interval between primary surgery and start of first-line tamoxifen treatment for relapse of disease. TTP was considered the primary outcome measured as time in months after start of tamoxifen until treatment was ended because of progression. Survival curves were generated using the KaplanMeier method. For analysis regarding TTP, P-values from log-rank tests were used. Since only 10% (n = 7) of the patients were at risk after 3 years of tamoxifen, patients with TTP exceeding 36 months were censored. Cox proportional-hazards analysis was used to calculate hazard ratios (HR). For multivariate analyses, three models were used. Menopausal status (pre vs. post), DFI (B2 vs. [2 years), grade (I/II vs. III) and HER2 (3+ vs. rest) were considered as possible confounders and added to the models, that tested ER, PgR and each of the profiles. Since only patients who developed a relapse were included, the distribution of DFI between classes predicted by the profiles were compared by the Mann-Whitney U- or Kruskal– Wallis-test. Analyses were performed using SPSS 14.0.1.
HOXB13-IL17BR ratio The HOXB13-IL17BR ratio was generated using RT-PCR as described by Jansen et al. [15]. The cut-off of 2.99 for tamoxifen treatment in metastatic disease setting previously defined by Jansen et al. was used [15]. Fig. 1 Flow chart for study design and endpoints used for analyses. Gene expression was performed using RNA from the primary tumor. RT = Radiotherapy
Surgery (RT)
Results Gene expression profiles of primary tumors of 69 adjuvant systemic treatment naı¨ve breast cancer patients who
No adjuvant systemic therapy
Primary breast tumor Gene expression
Tamoxifen
Relapse of disease DFI (Disease-Free Interval) PROGNOSIS
(Fig.2D-E)
Tumor progression
TTP (Time to Tumor Progression) TAMOXIFEN RESPONSE
(Fig.2A-C)
123
278
Breast Cancer Res Treat (2009) 113:275–283
Table 1 Patient characteristics (n = 69) No. of patients (%) Median age at diagnosis in years [range]
64
[37–84]
Breast conserving therapy
18
(26)
Modified radical mastectomy
51
(74)
Surgery
No. of positive nodes 0
45
(65)
1–3
16
(23)
C4
6
(9)
Unknown
2
(3)
36 33
(52) (48)
I
22
(32)
II
28
(41)
III
18
(26)
1
(1)
Ductal
58
(84)
Lobular
10
(14)
1
(1)
Tumor diameter B20 mm [20 mm Histologic gradea
Unknown WHO type carcinoma
Ductal + DCIS Menopausal status at diagnosis Premenopausal
19
(28)
Postmenopausal
50
(72)
Disease-free interval \2 years
13
(19)
56
(81)
Premenopausal
10
(14)
Postmenopausal
59
(86)
B6 months
19
(28)
[6 months
50
(72)
Responder profile
31
(45)
Non-responder profile
38
(55)
C2 years Menopausal status start tamoxifen
Time to tumor progression (TTP)
Survival analysis (TTP) Figure 2a–c shows the Kaplan–Meier curves for the profiles in relation to TTP. For the 78-gene profile, the difference in median TTP for the predicted groups was 14 months (7 vs. 21) (log-rank P = 0.004). The High-, Intermediate- and Low Risk group as determined by the 21-gene set showed a median TTP of 10, 21 and 28 months, respectively (log-rank P = 0.003). The two groups defined by HOXB13-IL17BR differed by 10 months (3 vs. 13, log-rank 0.004). In univariate Cox regression analysis (Table 2), patients classified as non-responder by the 78-gene profile had a HR of 2.16 (95% CI = 1.27–3.67, P = 0.005) compared to tumors predicted as responder. Patients classified as High Risk by the 21 genes had worse outcome while on tamoxifen compared to patients classified as Low or Intermediate Risk (HR = 2.31, 95% CI = 1.32–4.02, P = 0.003). The HR for HOXB13IL17BR was 4.20 (95% CI = 1.43–12.29, P = 0.009). The traditional markers ER, PgR and HER2 were also significantly associated with TTP.
b
78-gene profile
21-gene set Low-Intermediate Risk
28
(41)
High Risk
41
(59)
Low Risk
56
(93)
High Risk Unknown
4 9
(7)
HOXB13-IL17BR ratio
Abbreviations: no., number; DCIS, ductal carcinoma in situ a
According to Bloom and Richardson
b
While on first-line tamoxifen monotherapy for relapse of disease
123
developed a relapse of disease for which they were treated with first-line tamoxifen, were related to TTP while on tamoxifen. The study design is illustrated in Fig. 1. Clinical features are summarized in Table 1. The 78-gene tamoxifen response profile (microarray), the 21-gene set (microarray-based Recurrence Score) and the HOXB13IL17BR ratio (RT-PCR) predicted 55% (n = 38), 59% (n = 41) and 7% (n = 4) respectively, of the patients as having a ‘non-response’ or ‘High Risk’ expression profile (Table 1).
Sensitivity and specificity of 78-gene tamoxifen response profile, 21-gene set and HOXB13-IL17BR ratio Next, we evaluated the predictive capacity of the 78-gene profile, 21-gene set, HOXB13-IL17BR ratio, ER (low vs. high) and PgR (negative vs. positive) (Table 3). Patients who had a TTP of less than 6 months were considered as non-responders. The 78-gene profile and 21-gene set reached high accuracy for identifying non-responding tumors with a sensitivity of 84.2 (95% CI 60.4–96.6) and 73.7 (95% CI 53.9–93.5), respectively. On the contrary, the HOXB13-IL17BR ratio as well as the ER identified responding tumors with high specificity: 95.2% (95% CI 83.8–99.4) and 79.6% (95% CI 68.3-90.9), respectively.
Breast Cancer Res Treat (2009) 113:275–283
A 78-gene profile
279
B 21-gene set
C HOXB13-IL17BR ratio 21-gene set: Low Risk 21-gene set: Intermediate Risk 21-gene set: High Risk
78-gene profile: Response 78-gene profile: Non-response
p=0.004*
HOXB13-IL17BR: Low Risk HOXB13-IL17BR: High Risk
p=0.003**
Response profile 31 28 24 Non-response profile 38 22 15
16
11
7
5
9
6
3
2
D 78-gene profile
Low Risk 11 8 7 Intermediate Risk 17 15 13 High Risk 41 27 19
7
6
4
3
10
7
4
2
8
4
2
E 21-gene set
Low Risk 56 40 High Risk 4 2
11
6
2
12
4
1
20
14
7
6
0
0
0
0
F HOXB13-IL17BR ratio HOXB13-IL17BR: Low Risk HOXB13-IL17BR: High Risk
p=0.002 ¶¶
p=0.42 ¶
32 0
2
21-gene set: Low Risk 21-gene set: Intermediate Risk 21-gene set: High Risk
78-gene profile: Response 78-gene profile: Non-response
Response profile 26 19 31 Non-response profile 38 32 22
p=0.004*
Low Risk 11 10 9 Intermediate Risk 17 14 17 High Risk 31 18 41
p=0.038 ¶
7
4
1
7
2
0
9
4
2
Low Risk 56 47 High Risk 4 3
34
18
7
2
0
0
0
0
Fig. 2 Relation gene profiles with tamoxifen response (a–c) and prognosis (d–f). Analysis of probability of tumor progression while on tamoxifen (a–c) or of relapse of disease after no adjuvant treatment
(d–f), according to 78-gene profile (a and d), 21-gene set (b and e) and HOXB13-IL17BR ratio (c and f). * Log-rank test, ** log-rank test for trend, } Mann-Whitney U-test, }} Kruskal–Wallis test
Concordance of profiles
Multivariate Cox regression analysis (TTP)
Next, we compared the predictors using two-way contingency tables (see online Supplements). Comparison of classification according to the 78-gene profile with classification according to the 21-gene set and the HOXB13IL17BR ratio revealed a relatively low number of concordant cases of 42 (61%) and 30 (50%), respectively. Twenty-seven patients (45%) were classified in the same category by the 21-gene set and the HOXB13-IL17BR ratio.
To assess whether gene expression profiles could potentially add predictive information to the routinely used ER and PgR, three different multivariate analyses were performed. Table 4 shows three models all including ER, PgR and each gene profile. To adjust for possible confounders, all models included HER2 and traditional parameters: menopausal status, DFI and grade. Only the 78-gene profile and 21-gene set showed significant association independent of ER and PgR with HRs of 2.62 (95% CI 1.41-4.87,
123
280
Breast Cancer Res Treat (2009) 113:275–283
Table 2 Univariate Cox proportional-hazard regression analysis of the risk of tumor progression (TTP) after start of first-line tamoxifen treatment for relapse of disease No. Hazard ratio 95% CI
P-value
19
1.36
0.28
50
1
I/II
50
1
III
18
1.37
0.76–2.45
0.29
0.94–3.54
0.078
1.00–3.15
0.050
1.36–3.99
0.002
1.17–7.56
0.022
1.27–3.67
0.005
1.32–4.02
0.003
Menopausal statusa Premenopausal Postmenopausal Tumor grade
Unknown
0.78–2.38
Discussion
1
Disease-free interval \2 years
13
1.82
C2 years
56
1
ER (IHC) [10, \80%
18
1.78
C80%
49
1
Unknown
2
PgR (IHC) B10%
29
2.33
[10%
38
1
Unknown HER2 (IHC)
2
0, 1+, 2+
62
3+
5
Unknown
2
1 2.97
78-gene profile Responder profile
31
1
Non-responder profile
38
2.16
21-gene set Low-Intermediate Risk 28
1
High Risk
41
2.31
56
1
HOXB13-IL17BR ratio Low Risk High Risk
4
Unknown
9
4.20
1.43–12.29 0.009
Abbreviations: no., number of patients; yrs, years; CI, confidence interval; ER, estrogen receptor; IHC, immunohistochemistry; PgR, progesterone receptor; HER2, human epidermal growth factor receptor 2 a
At diagnosis of primary tumor
P = 0.002) and 1.94 (95% CI 1.01-3.73, P = 0.048), respectively. PgR was significantly associated with TTP in all models. Association of profiles with disease-free interval To investigate whether the 78-gene profile, 21-gene set and HOXB13-IL17BR ratio are actually predicting progression after tamoxifen rather than reporting prognosis, the association with DFI was studied. No adjuvant systemic
123
treatment was administered in this interval. For details of study design see Fig. 1. Figure 2d–f shows the KaplanMeier curves for the three gene profiles in relation to DFI. The 78-gene profile was not significantly related to the DFI. In contrast, the 21-gene set (P = 0.002) and the HOXB13-IL17BR ratio (P = 0.038) did predict DFI.
A profile of 81 genes had recently been identified in breast cancer as being associated with resistance to tamoxifen in the metastatic disease setting [10]. Other investigators had described a 21-gene set (Recurrence Score) and a HOXB13-IL17BR ratio for the prediction of outcome following tamoxifen treatment in the adjuvant setting [7, 8]. Simultaneous analysis of these gene profiles by applying them to a single dataset is necessary for direct comparison of the classifiers. For the first time, we have assessed the performance of these three profiles in an independent cohort of adjuvant systemic treatment naı¨ve breast cancer patients treated with first-line tamoxifen for relapse of disease. The 78-gene profile (78 oligos representing the formerly used 81 cDNA clones), the 21-gene set (microarray-based Recurrence Score) as well as the HOXB13IL17BR ratio (RT-PCR) were significantly associated with TTP of patients treated with first-line tamoxifen. In multivariate analyses only the 78-gene profile as well as the 21gene set remained significantly associated with TTP independent of ER and PgR and other traditional factors. Nowadays, the majority of patients with ER-positive breast cancer receives endocrine treatment in the adjuvant setting. A drawback of assessing tamoxifen response in the adjuvant setting is that both response of tumor cells to tamoxifen (drug response prediction) as well as intrinsic aggressiveness (prognosis) of the malignancy are measured. The best way to unravel this prognostic and predictive effect is analyzing the performance of a predictor in a randomized trial including an untreated control group. So far, for none of the classifiers studied in our series this has been published with regard to tamoxifen. Currently, almost no physician will withhold endocrine treatment from a patient with ER-positive disease, so for the untreated control group researchers depend on trials performed in the past. Unfortunately, during the years that those trials were carried out, most hospitals did not freeze tumors. Although RNA extraction from FFPE material is possible [7, 8], to our knowledge there are no whole genome datasets available including patients who were treated in a randomized trial to compare the prognostic and predictive value of a tamoxifen response profile. An alternative approach is to study tamoxifen response in patients who did not receive any adjuvant systemic
Breast Cancer Res Treat (2009) 113:275–283
281
Table 3 Sensitivity and specificity of tamoxifen response profiles and ER and PgR in the classification of patients as having either a short time to tumor progression (TTP) of B6 months (n = 19) or a TTP of [6 months (n = 50) SENSb
95% CI
SPEC
95% CI
44.4
21.5–67.4
79.6
68.3–90.9
61.1
38.6–83.6
63.3
49.8–76.8
84.2
60.4–96.6c
56.0
42.2–69.8
73.7
53.9–93.5
46.0
32.2–59.8
95.2
83.8–99.4c
ER (IHC) 11–79%a vs. C80% PgR (IHC) 0–10%a vs. [10% 78-gene profile Non-respondera vs. responder profile 21-gene set Higha vs. Low-Intermediate Risk HOXB13-IL17BR ratio Higha vs. Low Risk
11.1
1.4–34.7c
Abbreviations: SENS, sensitivity; SPEC, specificity; CI, confidence interval; ER, estrogen receptor; IHC, immunohistochemistry; PgR, progesterone receptor a
Defined as ‘positive’ test result (predictive for non-response)
b
Patients with a TTP B 6 months considered ‘positives’ (presence of non-response)
c
95% confidence interval calculated using the Binomial distribution
therapy and subsequently developed a relapse of their disease for which tamoxifen was prescribed. The TTP while on tamoxifen provides information on decrease or increase of tumor load and thereby to a large extent on sensitivity to tamoxifen. In contrast, the DFI in the period after primary surgery where no drugs were prescribed reflects general aggressiveness of the tumor. However, analysis of our series regarding the prognostic value of the classifiers has to be interpreted carefully. Our series was selected for having developed a relapse, i.e., patients who did not have a relapse of their disease were left out. Since a large proportion of these patients probably would have had a Low Risk profile, the data on prognosis presented here can be considered conservative estimates. In our series, the 78-gene profile, the 21-gene set as well as the HOXB13-IL17BR ratio were predictive for TTP while on tamoxifen. With regard to the DFI, the HOXB13IL17BR ratio and the 21-gene set were associated with time to relapse while the 78-gene profile was not. Our data suggest that the 78-gene profile is primarily a predictive marker. Regarding the 21-gene set, previous studies showed that the Recurrence Score captures both prognosis and tamoxifen response [7, 20, 21]. Our study confirms this mixed effect. Although the HOXB13-IL17BR ratio was significantly associated with TTP, after adjustment for other predictive markers, the association was no longer significant. Even though Jansen et al. showed a significant relation of the HOXB13-IL17BR ratio with TTP [15], our data suggest that this Two-Gene-Index might be somewhat more prognostic rather than predictive for tamoxifen response. This strong prognostic value is in line with previous studies [15, 22, 23].
Since the HOXB13-IL17BR ratio was analyzed using a predefined cut-off [15] that classified only a minority of the patients as High Risk, the association with TTP in our multivariate model was less robust probably due to analyses of small subgroups. When other previously described cut-points [8, 22, 23] were applied, no stronger predictive results were obtained (data not shown). The expression of the 21 genes was determined using array data, rather than the validated laboratory test (Oncotype DX, Genomic Health Inc.). Although this may result in an underestimation of the capacity of the 21-gene set, previous work suggests that applying the 21 genes to a microarray dataset results in accurate prediction of subgroups of patients [24, 25]. In contrast to a previous study comparing prognostic gene profiles where high concordance among classifiers was observed [24], profiles predicting outcome after tamoxifen treatment did not show high agreement in their classification. They classified only 45-61% of patients in the same category. The 78-gene profile and 21-gene set had relatively high sensitivity to identify non-responding tumors, while the HOXB13-IL17BR ratio identified responding tumors with a high specificity. These remarkable differences in sensitivity and specificity for the predictors suggest that if these tests will reach clinical application, the gene sets might be helpful in addressing different clinical questions. For example, does the profile identify tamoxifen responsive tumors with high accuracy or rather select the true resistant tumors? Recently, the update of the Intergroup Exemestane Study [6] suggests that postmenopausal, ER-positive breast cancer patients who switch to an aromatase inhibitor after 2-3 years on
123
282
Breast Cancer Res Treat (2009) 113:275–283
Table 4 Multivariate Cox proportional-hazard regression analysis of the risk of tumor progression (TTP) after start of first-line tamoxifen treatment for relapse of disease adjusted for traditional predictive clinicopathological factorsa Hazard ratio
95% CI
P-value
1.87
0.85–4.08
3.43
1.83–6.45 \0.001
2.62
1.41–4.87
ER (IHC) 11–79% vs. C80%
0.12
PgR (IHC) 0–10% vs. [10%
in tamoxifen treated patients and deserves incorporation in future clinical studies. Acknowledgments We are indepted to Maxime Look and Marion Meijer-van Gelder for advice regarding data analyses and to Sjoerd Rodenhuis, Fabien Reyal and Stella Mook for critically reading the manuscript. The authors would like to thank Renate de Groot for the construction of the tissue microarray and Donne Majoor and Jacinta Aantjes for their help with the immunohistochemistry. The authors would like to commemorate Hans Peterse, an excellent surgical pathologist and a highly estimated colleague who passed away unexpectedly during the course of this project.
78-gene profile Non-responder vs. Responder profile
0.002
References
ER (IHC) 11–79% vs. C80%
1.81
0.84–3.89
0.13
2.67
1.47–4.85
0.001
High Risk vs. Low-Intermediate 1.94 Risk
1.01–3.73
0.048
1.66
0.71–3.89
0.24
2.36
1.24–4.50
0.009
2.52
0.78–8.23
0.125
PgR (IHC) 0–10% vs. [10% 21-gene set
ER (IHC) 11–79% vs. C80% PgR (IHC) 0–10% vs. [10% HOXB13-IL17BR ratio High vs. Low Risk
Abbreviations: ER, estrogen receptor; IHC, immunohistochemistry; PgR, progesterone receptor Analysis based on three different models a
To adjust for possible confounders, all models included HER2 (3+ vs. rest) and traditional predictive clinicopathological factors: menopausal status at diagnosis (pre vs. post), disease-free interval (B2 vs. [2 years), histological grade (I/II vs. III). None of these parameters did significantly predict the TTP in any of these multivariate models
adjuvant tamoxifen will have an improvement in overall survival. The ATAC-trial showed that 5 years of adjuvant anastrozole was superior to tamoxifen in terms of diseasefree survival [3]. At this time, it is unclear whether postmenopausal patients should be treated according to the switch strategy [6], or should start with an aromatase inhibitor upfront [3]. It is unknown whether the multigene assays predict endocrine responsiveness in general or are specific for tamoxifen resistance. Further studies have been initiated to investigate the predictive value of the 78 and 21 genes for response to aromatase inhibitors. In conclusion, we have demonstrated that the 78-gene tamoxifen response profile as well as the 21-gene set and the HOXB13-IL17BR ratio were significantly associated with TTP in an independent patient series. The addition of multigene assays to ER improves the prediction of outcome
123
1. Early Breast Cancer Collaborative Trialists’ Group (EBCTCG) (2005) Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet 365:1687–1717 2. Pritchard K (2003) Endocrine therapy of advanced disease: analysis and implications of the existing data. Clin Cancer Res 9:460S–467S 3. ATAC Trialists’ Group (2005) Results of the ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial after completion of 5 years’ adjuvant treatment for breast cancer. Lancet 365:60-62 4. Goldhirsch A, Wood WC, Gelber RD, Coates AS, Thurlimann B, Senn HJ et al (2007) Progress and promise: highlights of the international expert consensus on the primary therapy of early breast cancer 2007. Ann Oncol 18:1133–1144 5. Jonat W, Gnant M, Boccardo F, Kaufmann M, Rubagotti A, Zuna I et al (2006) Effectiveness of switching from adjuvant tamoxifen to anastrozole in postmenopausal women with hormone-sensitive early-stage breast cancer: a meta-analysis. Lancet Oncol 7: 991–996 6. Coombes RC, Kilburn LS, Snowdon CF, Paridaens R, Coleman RE, Jones SE et al (2007) Survival and safety of exemestane versus tamoxifen after 2-3 years’ tamoxifen treatment (Intergroup Exemestane Study): a randomised controlled trial. Lancet 369: 559-570 7. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M et al (2004) A multigene assay to predict recurrence of tamoxifen-treated, nodenegative breast cancer. N Engl J Med 351:2817–2826 8. Ma X, Wang Z, Ryan P, Isakoff S, Barmettler A, Fuller A et al (2004) A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen. Cancer Cell 5: 607–616 9. Weigelt B, Hu Z, He X, Livasy C, Carey L, Ewend M et al (2005) Molecular portraits and 70-gene prognosis signature are preserved throughout the metastatic process of breast cancer. Cancer Res 65:9155–9158 10. Jansen M, Foekens J, van Staveren I, Dirkzwager-Kiel M, Ritstier K, Look M et al (2005) Molecular classification of tamoxifen-resistant breast carcinomas by gene expression profiling. J Clin Oncol 23:732–740 11. Van ‘t Veer L, Dai H, van de Vijver M, He Y, Hart A, Mao M et al (2002) Gene expression profiling predicts clinical outcome of breast cancer. Nature 415:530–536 12. Glas A, Floore A, Delahaye L, Witteveen A, Pover R, Bakx N et al (2006) Converting a breast cancer microarray signature into a high-throughput diagnostic test. BMC Genomics 7:278 13. van de Vijver M, He Y, ‘t Veer L, Dai H, Hart A, Voskuil D et al (2002) A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347:1999–2009
Breast Cancer Res Treat (2009) 113:275–283 14. Wessels LF, Reinders MJ, Hart AA, Veenman CJ, Dai H, He YD et al (2005) A protocol for building and evaluating predictors of disease state based on microarray data. Bioinformatics 21: 3755–3762 15. Jansen MP, Sieuwerts AM, Look MP, Ritstier K, Meijer-van Gelder ME, van Staveren IL et al (2007) HOXB13-to-IL17BR expression ratio is related with tumor aggressiveness and response to tamoxifen of recurrent breast cancer: a retrospective study. J Clin Oncol 25:662–668 16. Hannemann J, Kristel P, van Tinteren H, Bontenbal M, van Hoesel CM, Smit WM et al (2006) Molecular subtypes of breast cancer and amplification of topoisomerase IIa: predictive role in dose intensive adjuvant chemotherapy. Br J Cancer 95:1334–1341 17. van de Vijver MJ, Peterse JL, Mooi WJ, Wisman P, Lomans J, Dalesio O et al (1988) Neu-protein overexpression in breast cancer. Association with comedo-type ductal carcinoma in situ and limited prognostic value in stage II breast cancer. N Engl J Med 319:1239–1245 18. Harvey J, Clark G, Osborne CK, Allred DC (1999) Estrogen receptor status by immunohistochemistry is superior to the ligand-binding assay for predicting response to adjuvant endocrine therapy in breast cancer. J Clin Oncol 17:1474 19. Mohsin S, Weiss H, Havighurst T, Clark G, Berardo M, Roanh L et al (2004) Progesterone receptor by immunohistochemistry and clinical outcome in breast cancer: a validation study. Mod Pathol 17:1545–1554
283 20. Paik S, Shak S, Tang G, Kim C, Joo H, Baker J et al (2005) Expression of the 21 genes in the Recurrence Score assay and prediction of clinical benefit from tamoxifen in NSABP study B14 and chemotherapy in NSABP study B-20. J Clin Oncol 23:16S. Abstract 21. Habel LA, Shak S, Jacobs MK, Capra A, Alexander C, Pho M et al (2006) A population-based study of tumor gene expression and risk of breast cancer death among lymph node-negative patients. Breast Cancer Res 8:R25 22. Ma X, Hilsenbeck S, Wang W, Ding L, Sgroi D, Bender R et al (2006) The HOXB13:IL17BR expression index is a prognostic factor in early-stage breast cancer. J Clin Oncol 24:4611–4619 23. Goetz M, Suman V, Ingle J, Nibbe A, Visscher D, Reynolds C et al (2006) A two-gene expression ratio of homeobox 13 and interleukin-17B receptor for prediction of recurrence and survival in women receiving adjuvant tamoxifen. Clin Cancer Res 12:2080–2087 24. Fan C, Oh D, Wessels L, Weigelt B, Nuyten D, Nobel A et al (2006) Concordance among gene-expression-based predictors for breast cancer. N Engl J Med 355:560–569 25. Loi S, Haibe-Kains B, Desmedt C, Lallemand F, Tutt A, Gillet C et al (2007) Definition of clinically distinct molecular subtypes in estrogen receptor-positive breast carcinomas through genomic grade. J Clin Oncol 25:1239–1246
123