Breast cancer screening relies upon mammography, but for women with dense breast tissue this method is often uninformative. Routine screening identifies suspicious breast lesions in some women, but the pain and risk associated with follow-up biopsies along with the poor accuracy of traditional histopathology urgently call for improved approaches to breast cancer screening. This is especially important for those high-risk patients for whom mammography is of limited value. We describe a non-invasive liquid biopsy method of profiling plasma exosome preps designed to improve the accuracy and safety of breast cancer screening for women with dense breast tissue.


We incubated plasma samples (300 microliters per sample) from breast cancer patients (n=60) and a control cohort (n=60) with a high-complexity DNA aptamer library using a modified SELEX scheme, termed “adaptive dynamic artificial poly-ligand targeting (ADAPTTM)”. Differentially bound (cancer vs. non-cancer) aptamers were recovered from precipitated exosomes and were identified by deep sequencing. Two thousand aptamer sequences were resynthesized and used to probe a larger set of 500 plasma samples from a patient cohort (n=206) and a control cohort comprised of self-reported healthy volunteers (n=117) and patients whose biopsies led to a diagnosis of non-cancer (n=177). We employed several statistical models to build a cancer/non-cancer predictor, including a Random Generalized Linear Model (RGLM) and a Random Forest Model (RFM). Both models yielded an equivalent classification performance with areas under the receiver-operator characteristic curve (ROC AUC) of 0.7. Testing the prediction performance by 100 Out-of-Bag permutations or by pre-filtered (read cutoff and estimated sample size) cross-validation (CV) resulted in ROC AUC values of 0.66 and 0.62, respectively. When samples were randomly assigned to groups, the aptamers were no longer able to distinguish the groups (ROC AUC = 0.54), indicating that the underlying information driving the model is truly specific to cancer. Importantly, incorporation of BIRAD results as a clinical covariate did not influence model performance, signifying that predictions by ADAPTTM were independent of breast tissue density.


We have identified a set of 2000 DNA aptamers that distinguish women with breast cancer from women without breast cancer. Our liquid biopsy approach requires only 300 microliters of plasma and is amenable to high-throughput processing. By employing a number of statistical approaches including rigorous cross-validation, we consistently achieve cross validation ROC AUC values approaching 0.7. The performance of the predictor was not affected by BIRAD scores, supporting its potential utility in difficult cases where imaging is insufficient, such as in women with dense breast tissue. Further optimization of the aptamer library and testing on additional samples should improve performance. Upon complete validation, an ADAPTTM – derived breast cancer test may serve as a vital diagnostic adjunct that can be easily incorporated into standard clinical practice.

Citation Format: Domenyuk V, Zhong Z, Wang J, Stark A, Chen W, Xiao N, Miglarese MR, Famulok M, Mayer G, Spetzler DB. Adaptive dynamic artificial poly-ligand targeting: Aptamer-based profiling of liquid biopsies to improve the accuracy of breast cancer diagnoses in women with dense breast tissue. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P2-01-08.