Combining Machine Learning and Nanofluidic Technology to Diagnose Pancreatic Cancer Using Exosomes

Circulating exosomes contain a wealth of proteomic and genetic information, presenting an enormous opportunity in cancer diagnostics. While microfluidic approaches have been used to successfully isolate cells from complex samples, scaling these approaches for exosome isolation has been limited by the low throughput and susceptibility to clogging of nanofluidics. Moreover, the analysis of exosomal biomarkers is confounded by substantial heterogeneity between patients and within a tumor itself. To address these challenges, researchers at the University of Pennsylvania developed a multi-channel nanofluidic system to analyze crude clinical samples. Using this platform, they isolated exosomes from healthy and diseased murine and clinical cohorts, profiled the RNA cargo inside of these exosomes, and applied a machine learning algorithm to generate predictive panels that could identify samples derived from heterogeneous cancer-bearing individuals. Using this approach, the researchers classified cancer and pre-cancer mice from healthy controls, as well as pancreatic cancer patients from healthy controls, in blinded studies.

ExoTENPO-based exosome capture


a. A photograph of the ExoTENPO and the NdFeB external magnet with an SEM image of the magnetic nanopores, scale bar: 600 nm. b. A schematic of our chip-based assay. From plasma, magnetically labeled exosomes are isolated and their mRNAs are isolated and profiled using qPCR, and the signature is found using a machine learning algorithm. c. Finite element simulations of ExoTENPO. The field strength |B| is plotted on the cross-section of a 600 nm pore. An exosome passing through the pore experiences both a drag force Fd from the fluid flow and a magnetophoretic force Fm towards the pore’s edge where the magnetic field gradient is maximized. d. SEM image of exosomes captured at the edge of the ExoTENPO pores. The small round objects are unbound magnetic nanoparticles (MNPs). e. The size distribution of exosomes isolated using ExoTENPO measured by DLS. The input, cell culture media, consisted of primarily debris (d =10.1 nm) whereas the exosomes isolated on ExoTENPO consisted primarily of exosomes (a = 149.7 ± 11.5 nm). (N = 3 technical replicates, ± represents standard deviation) f. The relative RNA expression level of 12 genes were compared between ExoTENPO and centrifugation method. The PCR threshold cycles (Ct) are plotted versus one another and showed positive correlation (R2 =0.8). g. The relative expression levels of RNA cargos from cells and exosomes are quantified from 5 different human pancreatic cancer cell lines, which showed positive correlation between cells and exosomes (h). (R2 =0.88, N = 3 technical replicates, N = 2 biological replicates)

Ko J, Bhagwat N, Yee SS, Ortiz N, Sahmoud A, Black T, Aiello NM, McKenzie L, O’Hara M, Redlinger C, Romeo J, Carpenter EL, Stanger BZ, Issadore D. (2017) Combining Machine Learning and Nanofluidic Technology to Diagnose Pancreatic Cancer Using Exosomes. ACS Nano [Epub ahead of print]. [abstract]

Leave a Reply

Your email address will not be published. Required fields are marked *