Emerging science: Prospective validation of gene expression profiling-based prediction of complete pathologic response to neoadjuvant paclitaxel/FAC chemotherapy in breast cancer

Reviewer: S. Jack Wei, MD
The Abramson Cancer Center of the University of Pennsylvania
Last Modified: June 2, 2003

Presenter: L Pusztai
Presenter's Affiliation: MD Anderson Cancer Center, Texas
Type of Session: Plenary


    Multiple different chemotherapy regimens are used in the treatment of breast cancer. One common regimen involves neoadjuvant paclitaxel + FAC (5-FU, doxorubicin, and cyclophosphamide) (T/FAC) which offers modest benefit over FAC alone at the expense of increased toxicity. The ability to predict which patients will respond to chemotherapy would be advantageous in maximizing the effectiveness of treatment. It is increasingly accepted that a number of genetic factors may predispose a patient to treatment response. The current study describes a gene expression profile that predicts for response to neoadjuvant T/FAC chemotherapy.

Materials and Methods

  • Fine needle aspirations (FNA) of 24 breast cancer patients were obtained to comprise a training set of patients. FNA of an additional 21 patients was obtained to comprise a validation set of patients.
  • RNA was isolated from the FNA cytologic samples and profiled using genetic microarrays.
  • Patients then went on to treatment with T/FAC chemotherapy (paclitaxel x 12 followed by FAC x 4) followed by surgical resection, and their surgical specimens examined for pathologic complete response (pCR).
  • Initial microarray analysis was performed on the training set using the Millineum microarray chip.
  • The most commonly overexpressed genes were identified in the training set and then applied for predictive value for pCR to the validation set.
  • The patients were also screened for genetic expression of estrogen receptor (ER) status and Her-2/neu status which was compared to their respective immunohistochemical expression.
  • Following the initial analysis, an addition 16 patients were added to the original training set and profiled for genetic expression using the Affymetrix microarray gene chip.
  • A set of second-generation genetic predictors were identified by randomingly isolating 25% of the new training set and identifying the most highly expressed genes. These gene sets were then tested upon the remainder of the training set to evaluate for predictive value. The process was repeated for several iterations, each time with a new random 25% of the training set until a consistent set of predictive genes was identified.


  • Both the Millineum chip and the Affymetrix chip were 100% accurate in predicting overexpression of ER status and Her-2/neu status.
  • The initial analysis yielded 5 genes consistently overexpressed; however, the sensitivity of these genes was low, and the confidence interval for prediction was wide.
  • The second-generation analysis identified 500 genes that were commonly overexpressed, 74 of which isolated as predictive for final pathologic response.
  • Overall specificity of the identified genetic profile was 100%.
  • Overall sensitivity was 38%.
  • Positive predictive value was 75%.
  • Negative predictive value was 73%.
  • The expected rate of prediction for pCR without genetic profiling on this patient population was 28%.

Author's Conclusions

  • Comprehensive profiling can be performed using FNA.
  • Accurate prediction of ER status and Her-2/neu status is possible using genetic profiling.
  • The genes identified as predictors of pCR were able to predict response with a high degree of specificity; however, a low sensitivity with a wide confidence interval was seen due to the small number of patients. Increase in the size of the training set may improve the sensitivity of prediction.

Clinical/Scientific Implications

    This study represents a promising step towards identifying specific patients that may respond to a particular treatment, allowing customization of treatment based on genetic profile. Although the analysis of the microarray data was rigourous, there is still great uncertainty regarding the appropriate manner in which the large amounts of data generated by these microarrays should be analyzed. Larger training sets should be used to increase the predictive value of the genes identified, and these should independently be tested on a large validation set. Based on the genetic profile identified in this study, an institutional prospective trial has been proposed at MD Anderson Cancer Center which will assign patients predicted to be sensitive to T/FAC to neoadjuvant T/FAC whereas patients predicted to be resistant to T/FAC will receive neoadjuvant FAC. In addition, futher analysis of the data set will be conducted in hopes of improving the predictive genetic profile. Future studies will address genetic profiling for prediction of response to other chemotherapeutic regimens in hopes of broadening our ability to individualize patient treatment.

Oncolink's ASCO Coverage made possible by an unrestricted Educational Grant from Bristol-Myers Squibb Oncology.