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 [32
]. 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 [33
] 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 [34
] 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 [35
] or in the setting of HR+ node-negative disease [36
]. Such patients have been shown to have low rates of response to chemotherapy and very low rates of early recurrence [37
]. Confirmation of chemotherapy benefit in molecularly low risk patients will be forthcoming from the TAILORx [38
] and MINDACT [39
] 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. ).
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 [40
] and breast cancer in particular [41
]. 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 [34
]. 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 [43
]. 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 [44
]. 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 [13
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 [45
]. 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.