Chemotherapy response and recurrence-free survival in neoadjuvant breast cancer depends on biomarker profiles: results from the I-SPY 1 TRIAL (CALGB 150007/150012; ACRIN 6657)
- First Online:
- Cite this article as:
- Esserman, L.J., Berry, D.A., Cheang, M.C.U. et al. Breast Cancer Res Treat (2012) 132: 1049. doi:10.1007/s10549-011-1895-2
Neoadjuvant chemotherapy for breast cancer allows individual tumor response to be assessed depending on molecular subtype, and to judge the impact of response to therapy on recurrence-free survival (RFS). The multicenter I-SPY 1 TRIAL evaluated patients with ≥3 cm tumors by using early imaging and molecular signatures, with outcomes of pathologic complete response (pCR) and RFS. The current analysis was performed using data from patients who had molecular profiles and did not receive trastuzumab. The various molecular classifiers tested were highly correlated. Categorization of breast cancer by molecular signatures enhanced the ability of pCR to predict improvement in RFS compared to the population as a whole. In multivariate analysis, the molecular signatures that added to the ability of HR and HER2 receptors, clinical stage, and pCR in predicting RFS included 70-gene signature, wound healing signature, p53 mutation signature, and PAM50 risk of recurrence. The low risk signatures were associated with significantly better prognosis, and also identified additional patients with a good prognosis within the no pCR group, primarily in the hormone receptor positive, HER-2 negative subgroup. The I-SPY 1 population is enriched for tumors with a poor prognosis but is still heterogeneous in terms of rates of pCR and RFS. The ability of pCR to predict RFS is better by subset than it is for the whole group. Molecular markers improve prediction of RFS by identifying additional patients with excellent prognosis within the no pCR group.
KeywordsBreast cancer Neoadjuvant chemotherapy Molecular biomarkers Pathologic complete response
Molecular and genetic studies demonstrate that breast cancer is a heterogeneous disease. Several classifiers are available for distinguishing tumor types based on prognosis and prediction of response to chemotherapy and hormonal therapy [1, 2, 3]. Molecular features are associated with substantially different outcomes  and with wide variability in response to standard therapies [5, 6]. Symptomatic tumors that tend to be large and palpable on presentation have substantially higher risk of recurrence than tumors detected by screening . For these larger tumors, neoadjuvant, or preoperative, chemotherapy makes it possible to assess response to treatment and may provide insights to the tumor’s biology. Studies examining the degree to which pathologic complete response (pCR) to therapy is predictive of recurrence-free survival (RFS) or overall survival (OS) have given mixed results in relatively unselected populations [8, 9, 10, 11, 12].
The I-SPY 1 TRIAL (investigation of serial studies to predict your therapeutic response with imaging and molecular analysis) is a multicenter neoadjuvant study of women with histologically confirmed invasive breast cancers. This report describes associations between molecular markers assessed in pretreatment tumor biopsy samples and response to neoadjuvant chemotherapy at the time of surgery, longer-term disease outcomes, and the relationship between response and RFS.
Study design and patient selection
The I-SPY 1 TRIAL methods have been described in detail elsewhere [13, 14] and was a collaboration of the American College of Radiology Imaging Network (ACRIN), Cancer and Leukemia Group B (CALGB), and Specialized programs of research excellence (SPORE). All patients gave written consent and had histologically confirmed invasive breast cancers measuring at least 3 cm by clinical examination or imaging, with no evidence of distant metastatic disease. Patients’ clinical stage 1 by exam was eligible if tumor size was >3 cm by imaging. Patients with T4 or inflammatory disease were eligible. The regimen of neoadjuvant chemotherapy included an initial anthracycline-based regimen after which patients either underwent surgery or received a taxane-based regimen prior to surgery.
Assays were conducted in nine laboratories. Data was integrated for central accession for analysis using NCICB’s caINTEGRATOR application (https://caintegrator-stage.nci.nih.gov/ispy/index2.jsp)—I-SPY 1 data version dated February 2011.
Standard pathology biomarkers
Hormone and HER2 receptor expression were measured from pretreatment core biopsies. Estrogen and progesterone receptor status were determined by immunohistochemistry (IHC) and calculation of Allred scores  at the study sites. HER2 status was determined locally by IHC and/or fluorescence-in situ hybridization assays (FISH). HER2 testing (IHC and FISH) was also performed centrally at the University of North Carolina (UNC) [13, 16]. HER2 status was considered positive if either local or central assays were positive. Ki67 was recorded as low (<10%), medium (10–20%), or high (>20%) and described in detail in supplemental methods .
Evaluation of pathologic response
pCR is defined as no invasive tumor present in either breast or axillary lymph nodes. Residual Cancer Burden (RCB)  was assessed and included the primary tumor bed area, overall invasive cancer and in situ disease cellularity, number of positive lymph nodes and diameter of largest metastasis. I-SPY 1 TRIAL pathologists were centrally trained; all cases were re-reviewed and scored for RCB as a dichotomous outcome (0, I vs. II, III) and by class (0, I, II, III). Data was recorded using NCI’s Oracle Clinical Remote Data Capture version 4.5 electronic database.
Tissue samples immediately frozen in OCT were assayed on catalog 44,000 feature Agilent Human oligonucleotide microarrays (catalog # G4112F). Total RNA purification and microarray hybridization were done as previously described . The background was subtracted and Lowess normalized log2 ratio of Cy3 and Cy5 intensity values were calculated . The primary microarray data presented in this study are available in the GEO database under accession number GSE22226.
Intrinsic subtype classification was determined by PAM50 50-gene assay as described . The risk of recurrence score (ROR-S) classified patients as having high, medium, or low risk of relapse using predefined cut-points as described previously .
The 70-gene prognostic profile was determined using representative probes and data normalization as previously described . This profile classified patients as having high or low risk of relapse using the predefined threshold [20, 21].
Wound healing signature  was used to classify tumors as quiescent or activated. A gene-expression signature predictive of p53 genotype  was used to classify tumors as p53 wild type or mutated.
DNA copy number abnormality was assessed by a molecular inversion probe (MIP) platform with focal amplification and high resolution (~10 K bp) as previously described [24, 25, 26]. Direct p53 genotyping was performed and mutations were detected by the Roche p53 AmpliChip beta test array [27, 28] and a combination of two approaches described in supplemental methods.
The primary endpoint for the trial was RFS according to the STEEP criteria . Time to recurrence was computed from start of treatment; and RFS at 3 years was determined by Kaplan–Meier analysis.
Associations of signature classifications with pCR and RFS were assessed by logistic regression and proportional hazards modeling, respectively. Association of pCR with RFS within each risk subset was also tested. These analyses were conducted using JMP Version 8.0.1, SAS Institute Inc.
We took HR/HER2 categories to be standard and addressed the ability of other signatures to predict RFS and pCR assuming HR/HER2 category is given. We used multivariate Cox regression of RFS on molecular signature classification, adjusting for the predefined contribution of HR and HER2 (as derived from a Cox model fit of RFS on HR and HER2). Similarly, we employed multivariate logistic regression to evaluate whether other molecular signature classifications were independently predictive pCR when given HR/HER2. These analyses were conducted using Bioconductor R .
Differences in rates of pCR and rates of RCB class 0 or 1 within molecular signature-defined subgroups were assessed by χ2 tests.
Patient were categorized as low- or high-risk by each molecular signature HR+/HER2− versus others, luminal (A and B) versus others (HER2-enriched, basal, and normal-like) , p53 wild type versus mutation, 70-gene prognosis signature low versus high , and wound-quiescent versus wound-activated , and stage 3 (including inflammatory) versus earlier stage to ensure comparable degrees of freedom. The pairwise concordances of the molecular signatures were compared by Kendall’s rank correlation and Fisher’s exact test.
All statistical analyses were performed (CY) and verified by a second statistician (DB) to confirm the results.
Demographics and characteristics of patients in the I-SPY 1 TRIAL
I-SPY trial evaluable (n = 221)
Profiled with agilent microarray (n = 149)
Profiled with agilent microarray without any trastuzumab (n = 120)
Clinical tumor size (cm)
Tumor longest diameter on baseline MRI (cm)
Clinically node positive at diagnosis
Histologic grade (baseline)
Clinical stage (baseline)
Hormone receptors (baseline)
HR-positive (ER or PR)
Her-2 positive (baseline)
HR-negative/Her-2 negative (baseline) (triple negative)
AC + T
AC + T + trastuzumab
AC + T + other
Post-operative adjuvant therapy
Any hormonal therapy
Ovarian suppression or ablation
Institutions participating in the I-SPY trial
Total accrual 237
Biomarker tests performed*
Georgetown University Hospital
Memorial Sloan-Kettering Cancer Center
National Cancer Institute/George Mason University
Reverse phase phosphoprotein arrays
University of Alabama at Birmingham
University of California, San Francisco
Frozen tissue, DNA and RNA processing, cDNA arrays
University of Chicago
University of North Carolina
Paraffin-based IHC; p53 gene chip; expression profiling
University of Pennsylvania Medical Center
University of Texas, Southwestern
University of Washington, Seattle
Among the 215 patients, 20 of 67 (30%) HER2+ patients received neoadjuvant trastuzumab, as previously described . Of the 46 HER2+ patients who did not, 17 (36%) received adjuvant trastuzumab. Radiation and hormonal adjuvant therapy were also given at physician discretion as clinically indicated (Table 1). Analysis of RFS was limited to patients who did not receive trastuzumab.
Most (65%) patients had clinically or pathologically confirmed axillary lymph node involvement at diagnosis and 90% had tumors of intermediate or high histologic grade. Median follow-up for survival and RFS was 3.9 years.
Early outcomes: residual disease measured at the time of surgical resection
The overall rate of pCR was 27%, and the rate of RCB scores of 0 or I was 37%.
The rates of pCR and RCB scores of 0 or I for four gene-expression prognostic classifiers—70-gene prognosis signature, ROR-S, wound healing, and p53 mutation signatures—were all low (Table 2). In this population of patients treated with neoadjuvant chemotherapy, a minority of patients had good prognosis profiles: 9% were classified as 70-gene low risk, 27% as ROR-S low, 25% as wound healing quiescent, and 49% as p53 wild type. The respective pCR rates were 0, 6, 7, and 9%. Rates of pCR were higher for poor prognosis signatures, including 70-gene high risk (24%), wound healing signature activated (26%), ROR-S moderate risk (17%) and high risk (36%) and p53 mutation predicted by expression profile (34%). Clinical outcomes were better when pCR or an RCB score of 0 or 1 was achieved.
Recurrence-free and OS
Three-year RFS and OS for the entire cohort were 78 and 85%, respectively. When RFS for the population was stratified by molecular signatures, outcomes by subtype differed substantially (Table 2).
Correlations among the molecular signatures
Pathological complete response and recurrence-free survival by molecular subtypes
Pathological complete response (pCR)
Rate of pCR (n)
Odds ratio (P value)
Hazard ratio (95% CI)
Hazard ratio, pCR vs. no PCR (95% CI within subgroup)
All patients with surgery
Population without any trastuzumab
0.23 * (0.06–0.63)
HR+/HER2− (Yes vs. other)
Yes: 9% (8/93)
Yes: 0.00 (–)
Other: 38% (30/79)
Other: 0.17* (0.04–0.51)
p53 (Wt vs. Mut)
Wt: 9% (5/58)
Wt 0.00 (–)
Mut: 34% (20/58)
Mut: 0.22* (0.05–0.65)
70-Gene (low vs. high)
Low: 0% (0/11)
Low: 0.00 (–)
High: 24% (25/105)
High: 0.29* (0.07–0.82)
Luminal PAM50 (luminal vs. other)
Luminal: 8% (5/61)
Luminal: 0.00 (–)
Other: 36% (20/55)
Other: 0.26* (0.06–0.79)
Wound healing (quiescent vs. activated)
Quiescent: 7% (2/29)
Quiescent: 0.00 (–)
Activated: 26% (23/87)
Activated: 0.25* (0.06–0.71)
ROR-S (low vs. med/HIGH)
Low: 6% (2/32)
Low: 0.00 (–)
Med/high: 27% (23/84)
Med/high: 0.26* (0.06–0.74)
Univariate vs. multivariate Cox analyses adjusting for predefined HR/HER2 contribution
Univariate hazard ratio (95% CI)
Multivariate hazard ratio adjusting for predefined HR/HER2 contribution (95% CI)
P53 (Wt vs. Mt)
70-Gene (low vs. high)
Luminal PAM50 (luminal vs. other)
Wound healing (quiescent vs. activated)
ROR-S (low vs. med/high)
Clinical stage (<stage 3 vs. not)
Ki67 (low/med vs. high)
The I-SPY 1 collaboration demonstrates that standards for imaging, data and tissue collection can be followed and molecular profiling from small specimens is achievable. Molecular profiles were generated for over 65% of all patients (improving as the trial proceeded), and these patients are representative of the entire data set.
Patients who present with large breast tumors, as exemplified by the I-SPY 1 cohort, have biologically poor-risk cancers, as evidenced by 91% having 70-gene high risk profile and the fact that many are interval cancers . Even within this clinically high risk population, response to therapy was heterogeneous. HER2 positivity and HR negativity were associated with a greater rate of pCR, as were four poor prognosis molecular signatures: wound-activated signature, ROR-S high risk, 70-gene poor-risk, and p53 predicted mutation. Patients with good prognosis signatures had a lower chance of short-term (pCR, RCB) response to chemotherapy, but had better long-term (RFS, OS) outcomes, even when their tumors did not respond to therapy. These findings support the emerging consensus that patients with good risk signatures (wound healing quiescent, 70-gene low, and ROR-S low) have low rates of early recurrence in spite of large tumor size. The molecular profiles vary by the percent of the population they classify as low risk, the fraction that respond to therapy, and the outcomes among those without pCR, even though the data set was not sufficiently large to show a statistical difference.
The International Breast Cancer Study Group (IBCSG), NSABP, and MD Anderson Cancer Center  have found that pCR rates are much higher in patients with HR-negative tumors than in those with HR-positive tumors. These observations are consistent with our results and with adjuvant studies that show patients with HR-negative disease benefit more from chemotherapy  than do patients with HR-positive disease.
Molecular profiles may provide the opportunity to identify, beyond HR and HER2 status, what might be driving tumor behavior and outcomes. In a multivariate model, when receptor types were fixed, the factors that added to RFS included clinical stage, wound healing signature, ROR-S, and p53 predicted mutation. When pCR was also fixed, most of the dichotomized molecular markers added some additional predictive value, likely because of the ability to identify patients in the “no pCR” group who have excellent outcomes, largely the HR+/HER2− subgroup, though not exclusively. Given that the low proliferative HR+ subset is at risk for late recurrence, longer follow-up and additional studies will be required to validate this observation.
Molecular signatures are currently being used to identify low risk patients who are less likely to benefit from chemotherapy regardless of nodal status  or in the setting of HR+ node-negative disease . Such patients have been shown to have low rates of response to chemotherapy and very low rates of early recurrence . Confirmation of chemotherapy benefit in molecularly low risk patients will be forthcoming from the TAILORx  and MINDACT  trials. In the follow-on I-SPY 2 TRIAL, an adaptive-design neoadjuvant trial to test the ability of phase 2 agents in combination with chemotherapy to increase pCR, 70-gene low risk, HR-positive and HER2-negative patients are being excluded from randomization. In I-SPY 1, none of the 11 patients with a 70-gene prognosis profile had a pCR or a recurrence (Fig. 2).
In the I-SPY cohort, the wound healing signature identified the largest fraction of low risk patients (based on RFS) of any signature. The genes consistent with an activated wound environment characterize women with poor outcomes, in keeping with increasing evidence that supports targeting the inflammatory pathway in high risk cancers  and breast cancer in particular [41, 42]. The activated wound healing signature is associated with poor outcomes across multiple tumor types and may well reflect the importance of the microenvironment in tumor behavior.
Although pCR and RCB are very predictive of RFS among the poor prognosis molecular profiles, the profiles do not predict an individual patient’s response to standard chemotherapy. A substantial fraction of tumors with the highest risk features have a complete response to therapy and do well, while others with that same signature have a poor response and poor outcome. Ongoing analysis is focusing on the I-SPY 1 patients who did not have a complete response to therapy and had early recurrence, using the described biomarkers as well as phosphoprotein profiles, to explore targets for future therapeutic intervention.
Our study is limited by the short follow-up time. Patients with HR-positive tumors continue to be at risk for recurrence for many years, and early recurrence data may not reflect the overall outcome . However, in this select group of patients where almost all patients had grade 2 or 3 disease, recurrence risk is likely to be concentrated in the first 5 years . The Oxford Overview Analysis of the early breast cancer trials strongly suggests that the benefit of chemotherapy is reflected by distant disease-free survival at 5 years, where the survival curves for patients with chemotherapy versus not initially diverge but are then parallel, so any survival benefit from chemotherapy is likely to be manifest in the first 5 years . The median follow-up period of 3.9 years in the I-SPY cohort should reflect the benefits in HER-positive and triple negative disease, where the risk of recurrence is early .
Molecular and biological heterogeneity were substantial even within the high risk group of patients in the I-SPY TRIAL. In this patient cohort, HR and HER2 status were the most predictive of pCR, but the molecular signatures add to the ability of the receptors to predict RFS. The task that remains is to use current and emerging markers to identify optimal biological subsets for new therapeutic agents. Importantly, molecular marker data should be collected routinely in trials so that markers and imaging that are early predictors of outcome can be related to the target endpoint of RFS . The I-SPY 1 database, with its rich resource of genomic and protein expression data, is an important resource to explore emerging and new biomarkers associated with resistance and response to standard therapy.
Thank you to the study patients and patient advocates; Jorge Gomez for developing NCI collaborations across ACRIN/SPORE; Larry Norton for support/encouragement to submit through CALGB; Ken Buetow for developing the informatics infrastructure; Subha Madhavan for caINTEGRATOR; and Rachel Gomez for ensuring pathologic data quality.
National Cancer Institute Specialized Program of Research Excellence in Breast Cancer (CA58207), American College of Radiology Imaging Network (CA079778 & CA080098), Cancer and Leukemia Group B (CA31964 & CA33601), National Cancer Institute Center for Bioinformatics, The Breast Cancer Research Foundation, Bruce and Martha Atwater, The Terry Fox Foundation Postdoctoral Fellowship, and “Give Breast Cancer the Boot.”
Joe W. Gray (Author #8) declares a consultant/advisory role with New Leaf Ventures, Susan G. Komen for the Cure, and Kroma TiD. He has no other disclosures. Laura J. van't Veer (Author #13) declares an employment/leadership role and has stock or other ownership interests at Agendia Inc (Chief Research Officer). She has no other disclosures.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.