Neoadjuvant Endocrine Therapy as a Drug Development Strategy

There are at least two distinct motivations for conducting neoadjuvant studies in breast cancer: (a) to identify promising systemic therapies for additional testing in definitive adjuvant trials; and (b) to identify individuals with primary tumors that are resistant to standard treatment. Different clinical trial design considerations, measurement parameters and statistical techniques are required to address these two questions. When comparing drug regimens, for example, an aromatase inhibitor (AI) plus or minus a signal transduction inhibitor (STI), a number of different biochemical, histological, and clinical end points can be used in an analytical approach that focuses on short-term measurements of efficacy (e.g., is there evidence that overall tumor regression rates, induction of cell death, or inhibition of proliferation are more frequent on one arm of the study than the other?). In contrast, studies that attempt to capitalize on the potential of neoadjuvant therapy to predict the effectiveness of adjuvant therapy must focus on long-term outcomes (did tumor regression, reduction in tumor proliferation, or increase in cell death in response to neoadjuvant endocrine therapy translate into a higher individual chance of long-term remission from breast cancer?). In this latter instance, surrogate end points must: (a) be closely correlated with long-term outcomes; (b) be subject to a low degree of measurement error; and (c) be analyzable as a discontinuous variable (high or low risk) to simplify statistical analysis, as well as clinical decision making. Currently, there are no established surrogate end points or biomarkers that have been shown to correlate with long-term outcomes in the context of neoadjuvant endocrine therapy, except perhaps clinical response (1). A large practice setting trial must therefore eventually be conducted with this question in mind. As a prelude, smaller neoadjuvant endocrine therapy trials focusing largely on short-term end points will continue. These studies aim to improve the response to neoadjuvant endocrine therapy, either by defining new ways to identify highly responsive tumors or by adding new agents such as a STI with the aim of enhancing the efficacy of estrogen deprivation therapy or antiestrogens.

Pathological complete responses (pCRs) are uncommon with neoadjuvant endocrine therapy. For example, in a randomized comparison between 4 months of letrozole versus tamoxifen, the pathological complete response rates were only 1% (2). Other end points must therefore be examined to provide evidence that an experimental neoadjuvant endocrine strategy is more effective than the standard approach. A randomized trial that compared letrozole and tamoxifen demonstrated that measurements of clinical response, radiological response, and rates of breast conserving surgery all favored the third generation AI (Table 1), a finding that mirrored the superior efficacy of AI in the Arimidex, Tamoxifen Alone or in Combination study but achieved with a low sample size (n = 338) and with only a 4-month treatment period (2, 3, 4). These data support the concept that a neoadjuvant study could provide essential preliminary data on which to base a definitive adjuvant endocrine study. The objectives of these short-term presurgical investigations would be to confirm the superior efficacy of the new combination against primary breast cancer and to provide a context in which to conduct exploratory biomarker studies that might help define subgroups of tumors that are particularly sensitive to the experimental regimen. The precise design of these studies needs careful consideration, however, as adequate statistical power must be available so that questions regarding surrogate biomarkers and clinical endpoints have a reasonable chance of being answered.

Clinical findings do not say much about the biological processes underlying response or resistance. The best-explored biomarker to address mechanism in the context of neoadjuvant endocrine studies is a measure of proliferation, Ki67 immunohistochemistry (5). This cell cycle-regulated protein usually shows a dramatic fall with endocrine treatment and is expressed as a percentage of positive cells or the proliferation index. Although a simple measurement, statistical analysis presents a greater challenge than pathological complete response; because the raw data are not normally distributed, a paired analysis must be conducted (baseline and posttreatment), and the outcome is not binary (unlike pathological complete response, which is “yes or no”), so the data must be analyzed as a continuous variable. The data are correctly expressed as the median, range, and geometric mean before and after therapy. Nonparametric tests are applied such as the one sample Wilcoxon signed-rank test for assessing treatment-induced changes within treatment groups and the Mann-Whitney two-sample test to compare the overall degree of change for letrozole- versus tamoxifen-treated tumors. A disadvantage of the Mann-Whitney test is that there is no method to adjust the data for baseline differences between the two groups. To address this issue, one can take the natural log of the Ki67 values to simulate a normal distribution and then apply an analysis of covariance (6). Analysis of covariance is a type of regression analysis in which the slopes of the two lines generated by plotting the before and after treatment values on the two treatment arms are compared. Applying both parametric and nonparametric tests to the Ki67 analysis of the letrozole 024 trial established that the overall fall in Ki67 was more profound with letrozole than tamoxifen (Fig. 1). This result supports the conclusion that part of the superior effectiveness of letrozole stems from greater effectiveness as an antiproliferative agent (7). These data confirm that Ki67 analysis is a useful additional surrogate end point for the assessment of new combinations of endocrine agent and STI as potential treatment for primary breast cancer (8). However, lack of information regarding the relationship between changes in Ki67 (ΔKi67) and long-term outcomes means that there is no cutoff value for ΔKi67 that could be used for designing studies in which ΔKi67 is a decision-making tool. (One cannot say tumor X has a ΔKi67 of Y and therefore chemotherapy is not required because the chance of cure with local therapy and adjuvant endocrine therapy is Z).

Ki67 analysis was also used to compare the relative effects of tamoxifen and letrozole within subsets defined by HER1 (epidermal growth factor receptor) and HER2 (ErbB2/neu). In these studies, HER1 and HER2 were investigated as a combined category of HER1 and/or HER2 positive (HER1/2+) versus both negative (HER1/2-). The assessment of the clinical data established that within the group of patients with estrogen receptor-positive and HER1/2+ disease, letrozole was dramatically more effective than tamoxifen (2). The ΔKi67 analysis was consistent with this finding because the treatment-induced reduction in Ki67 was greatest on the letrozole arm regardless of HER1/2 expression status (Fig. 1). These data suggest that ΔKi67 can also be used as a tool to explore relationships between signal transduction pathways and therapeutic effects of different classes of endocrine agent (7).

Neoadjuvant endocrine therapy trials tend not to demonstrate close correlation between changes in Ki67 and clinical and radiological response (9). Preliminary analysis of the letrozole 024 trial also found no statistically significant difference in the degree of change in Ki67 between tumors responding to letrozole (partial response and complete response) and those tumors exhibiting stable or progressive disease, i. e., suppression of Ki67 was just as great in responders versus nonresponders. The biological basis for this paradox is unclear: it would seem that clinical outcomes and Ki67 both reflect the advantage of an aromatase inhibitor over tamoxifen, but these end points are measuring somewhat independent effects of estrogen deprivation. It may be that in nonresponders, letrozole successfully inhibited tumor proliferation but failed to trigger another key event critical for tumor regression, perhaps most likely cell death. One approach to improve our understanding of the underlying molecular basis for responsiveness to estrogen deprivation is to conduct global gene expression profiling on tumor biopsy samples taken before and after the initiation of AI treatment. At this point, only limited array information is available from an ongoing study in which baseline expression arrays are being compared with a sample taken at 1 month after the initiation of letrozole therapy (8). In an illustrative responding case, genes showing the greatest decrease include members of a proliferation cluster [topoisomerase (DNA) II α ribonucleotide reductase M2 polypeptide, 5-methyltetrahydrofolate-homocysteine methyltransferase reductase, and cell division cycle 2, G₁-S and G₂-M], an invasion cluster [matrix metalloproteinase 1 (interstitial collagenase), carboxypeptidase B1 (tissue), CD36 antigen (collagen type I receptor, thrombospondin receptor), and protein regulator of cytokinesis 1], and the apoptosis suppression cluster [baculoviral IAP repeat-containing 5 (survivin) and nucleolar protein 3 apoptosis repressor w/CARD domain]. The current plan is to enroll 90 patients into this multisite Phase II investigation. The primary goal is to define a gene expression cluster that predicts response to treatment. Paired or dynamic analysis of responding versus nonresponding tumors is an additional key aspect of the investigation as a way to understand resistance. Regarding resistance, however, a generic weakness of a neoadjuvant study design is that secondary resistance is not adequately addressed because the tumor is excised before progression develops in almost all cases. Nonetheless, a molecular analysis of the surviving cells at surgery may provide some clues as to how estrogen receptor-positive breast cancer cells can survive extreme estrogen deprivation. All patients in the study have also consented to additional tumor sampling should a relapse occur. A comparison between the baseline sample and advanced disease progressing on endocrine therapy could tell us a great deal about why tumors regrow after an initial response.

The National Cancer Institute of Canada has recently activated a United States Breast Intergroup adjuvant trial (MA 27) to compare the nonsteroidal AI anastrozole with the steroidal AI exemestane. In a 2-by-2 randomization, patients will also receive the cyclooxygenase-2 inhibitor celecoxib or placebo. Celecoxib appears to be additive with exemestane in both treatment and prevention preclinical models. The potential antitumor mechanisms involved include inhibition of prostaglandin E₂ induction of aromatase, antiangiogenesis effects, inhibition of tumor invasion, and inhibition of tumor-induced inflammation and growth factor production (10, 11, 12). There is, therefore, a strong biological rationale for determining the effects of exemestane in combination with celecoxib as a way to improve outcomes for women with breast cancer. In a parallel clinical trial, the American College of Surgeons Oncology Group is considering the activation of a randomized neoadjuvant trial in which patients will receive 4 months of exemestane alone or in combination with celecoxib (Z01031). This trial will provide an important new opportunity to determine whether small studies in the neoadjuvant setting consistently predict the outcomes of large adjuvant trials. The schema is presented in Fig. 2. The primary end point of the trial is clinical responses rate with adequate power to see an increase from 60 to 75%. Secondary end points include response by radiological measurements, rates of breast-conserving surgery, rates of lymph node involvement, and a series of cell biological end points that include Ki67, response by COX2 expression, and global gene expression analysis.

There is evolving evidence that the neoadjuvant setting provides an effective means to obtain preliminary data on the adjuvant potential of new endocrine approaches to treatment. The next neoadjuvant endocrine trial to report is the IMPACT (Intermediate Marker Project: Anastrozole, Combination or Tamoxifen) study, with results available in December 2003. This study was designed to mimic the three arms of the Arimidex, Tamoxifen Alone or in Combination trial. Clearly, if neoadjuvant anastrozole either produces a greater response rate than tamoxifen or Ki67 is suppressed to a greater extent and the combination is no better than the tamoxifen-alone arm, the concept of preliminary neoadjuvant endocrine studies will be additionally supported because if the IMPACT result had been available to the Arimidex, Tamoxifen Alone or in Combination investigators at the study design stage, enrollment of 3000 patients to an ineffective combination arm could have been avoided.

In our enthusiasm for neoadjuvant endocrine therapy studies as a biomarker discovery platform or as a way to test new biological agents, we should not ignore the need to conduct a practice setting investigation. There are only two widely accepted approaches to primary breast cancer outside of a clinical trial: either operate immediately or administer neoadjuvant chemotherapy. Depending in the clinical circumstances, postmenopausal patients with estrogen receptor-positive tumors could undergo neoadjuvant AI therapy versus immediate surgery or neoadjuvant AI versus neoadjuvant chemotherapy. All patients would continue AI long term postoperatively. The study could be powered, analogous to NSABP B18, to demonstrate that neoadjuvant endocrine therapy improves overall survival compared with traditional postoperative treatment.

Dr. Steven Come: It seemed to me that the second half of your talk undercut the first half because of the variability in what some of these markers meant. You could suppress Ki67 but not have response and vice versa. What is your reliable surrogate? The chemotherapy trials have pathological complete response (pCR) as the end point. You are not going to get that in the endocrine neoadjuvant trials. What do you have that’s really valid? You could exclude losers, but it might be hard to pick winners.

Dr. Ellis: It depends on what question you are asking. If you are asking the pharmaceutical question and looking statistically at response in large groups in relation to Ki67, we see the parallel between the adjuvant and the neoadjuvant data. In the absence of pCR, individual response has to be measured very consistently by another method, and it has to be linked to long-term outcomes. That’s the piece that is problematic because we don’t have something straightforward like pCR in neoadjuvant endocrine therapy studies.

Dr. Mina Bissell: There are new data coming out to show that the cells in the tumor really causing the trouble may have some characteristics of stem cells, and therefore they are a very small proportion of the total tumor, and the bulk of the tumor is actually non-tumorigenic. The danger of neoadjuvant therapy, especially if you decide on these particular markers, is that you may not be touching those stem cells.

Dr. Ellis: Either the neoadjuvant setting will continue to mimic the results of adjuvant trials or it won’t, but it is important that we keep pairing the adjuvant with the neoadjuvant studies to get at these issues.

Dr. Come: Don’t you have to validate this approach with long-term outcomes in sizeable number of patients? I know if it was easy it would have been done already.

Dr. Ellis: If Z01031 succeeds, the idea is to do a randomized study of neoadjuvant chemotherapy versus neoadjuvant endocrine followed by surgery. The only difference on the two arms is the preoperative therapy, and the trial will be powered to see a survival advantage.

Dr. Come: Wouldn’t it be cleaner to do it in patients who are only going to get endocrine therapy?

Dr. Ellis: That’s very difficult. If they go to surgery and have 20 positive nodes, the oncologist immediately wants to give them chemotherapy. Challenging standard of care turns out to be a very difficult thing to do. One approach would be to make an upfront decision regarding the administration of chemotherapy. If yes, the randomization would be neoadjuvant chemotherapy versus neoadjuvant aromatase inhibitor with chemotherapy given postoperatively. If no, chemotherapy was planned, then the approach would be neoadjuvant aromatase inhibitor versus immediate surgery.

Dr. Aman Buzdar: Medical oncologists think chemotherapy in the neoadjuvant setting causes a higher degree of cytoreduction. We looked at our experience at M. D. Anderson with a thousand patients; in ER-positive patients, pCR rate is in single digits with either combination chemotherapy or endocrine therapy. They are almost identical. It is high time to test whether there is advantage with chemotherapy, and whether some subset of patients might avoid chemotherapy. This kind of trial may answer that question.

Dr. Richard Santen: If you get an 80% reduction in tumors in 4 months, which you have seen with letrozole, that has to be apoptosis. Most of us would assume that blockade of proliferation and stimulation of apoptosis are generally linked, but you have shown us some examples where they are not linked. We have to ask ourselves what are the precise mechanisms occurring with neoadjuvant endocrine therapy. Are we causing persistent apoptosis or decreased proliferation? If there are tumors that have major cytoreduction through apoptosis but without a change in proliferation, those individuals might not benefit long term from adjuvant AI therapy. On the other hand, individuals who experience both apoptosis and blockade of proliferation might respond long term to adjuvant therapy.

Dr. Ellis: Or it could be that the reason why you get cure with endocrine therapy is you induce apoptosis in the micrometastases, and the cytostatic effect is of lesser importance.

Dr. Carlos Arteaga: The study I would like to see done in this setting is not therapeutic; it is a study to identify a molecular signature in those patients’ cancers that would give me a signal of cellular activity upon treatment with, in this case, a novel drug or a novel combination. This “signature,” which might be something as simple as ER plus something else, would allow me to select patients into informative Phase II trials with the novel drug or combination from which I can expect a realistic readout of clinical activity. It would also allow me to spare patients without such “predictive profile,” who are unlikely to derive any clinical benefit from the novel drug or combination. In these tumors there is important information about therapeutic resistance than can be studied at the molecular level. Obviously, these should not be included in Phase II trials because they are going to cloud the net effect of the novel drug or combination.