Optical spectral fingerprinting for anthracycline detection in synthetic clinical biofluids

Anthracyclines, commonly used as chemotherapies, are highly effective at inducing apoptosis in cancer cells but also affect healthy cells; cardiotoxicity being a major side effect of chemotherapy administration as anthracyclines accumulate in cardiac tissue. The inability to monitor their accumulation in vivo in a noninvasive, rapid, and continuous manner presents a challenge in early diagnosis of cardiotoxicity and therapeutic drug monitoring of anthracyclines. While there are established guidelines for dosing this class of drugs, the high interpatient variability coupled with variable pharmacology make it difficult to gauge when cardiotoxicity may begin developing. To reduce this risk and more effectively understand these drugs' pharmacology, we developed fluorescent nanosensors using single walled carbon nanotubes (SWCNT) that can rapidly and continuously monitor anthracycline accumulation in a concentration-dependent manner. SWCNT have the innate ability to fluoresce stably in the near infrared region, which is tissue-transparent. Several species of fluorescent SWCNT exist due to structural chirality which have independent excitation/emission wavelengths. SWCNT are dispersed with ssDNA for individual dispersion, which imparts hydrophilicity and fluorescence, and promotes interaction with anthracyclines. To create chemical library diversity, multi-species SWCNT were dispersed with 12 unique ssDNAs and challenged with four anthracyclines: daunorubicin, doxorubicin, epirubicin, and idarubicin, at concentrations ranging from 0.1-100 μM. Using spectral fingerprinting, spectral changes and ssDNA-SWCNT pairings were analyzed using principal component (PCA) analysis to determine which factors most contributed to anthracycline detection. PCA differentiated sensor responses above 5 μM and provided insight into which ssDNA-SWCNT species combinations were best suited for detecting each anthracycline. Further analysis using binary classification was done using machine learning models k-nearest neighbor (k-NN) and support vector machine (SVM) to determine the accuracy of classifying each anthracycline's concentration-based spectral changes in buffer and synthetic biofluids. We found that the models were strongly predictive for detection of daunorubicin and idarubicin, exhibiting 100% cross validation, test accuracy, and validation in synthetic biofluids. Multi-class classification distinguished anthracycline type by spectral fingerprint, with 100% accuracy using Decision Tree and eXtreme Gradient Boosting. Future work will use these ssDNA-SWCNT species combinations to develop a noninvasive therapeutic drug monitoring tool to monitor accumulation at the tumor and in the heart during chemotherapy. We anticipate this work leading to a tool to improve tolerance for anthracycline chemotherapy by establishing personalized pharmacological treatment windows.