Combining Machine Learning and Nanofluidic Technology To


Combining Machine Learning and Nanofluidic Technology To...

0 downloads 143 Views 9MB Size

Subscriber access provided by UNIVERSITY OF ADELAIDE LIBRARIES

Article

Combining Machine Learning and Nanofluidic Technology to Diagnose Pancreatic Cancer Using Exosomes Jina Ko, Neha Bhagwat, Stephanie S Yee, Natalia Ortiz, Amine Sahmoud, Taylor Black, Nicole M. Aiello, Lydie McKenzie, Mark O'Hara, Colleen Redlinger, Janae Romeo, Erica L. Carpenter, Ben Z. Stanger, and David Issadore ACS Nano, Just Accepted Manuscript • DOI: 10.1021/acsnano.7b05503 • Publication Date (Web): 11 Oct 2017 Downloaded from http://pubs.acs.org on October 12, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

ACS Nano is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

Title: Combining Machine Learning and Nanofluidic Technology to Diagnose Pancreatic Cancer Using Exosomes Authors: Jina Ko1, Neha Bhagwat2, Stephanie S. Yee3, Natalia Ortiz1, Amine Sahmoud2, Taylor Black2, Nicole M. Aiello2, Lydie McKenzie3, Mark O’Hara3, Colleen Redlinger5, Janae Romeo5, Erica L. Carpenter3, Ben Z. Stanger2, David Issadore1,4, * Affiliations: 1. Department of Bioengineering, School of Engineering and Applied Sciences, University of Pennsylvania. Philadelphia, Pennsylvania, United States. 2. Division of Gastroenterology, Department of Medicine, Perelman School of Medicine, University of Pennsylvania. Philadelphia, Pennsylvania, United States. 3. Division of Hematology-Oncology, Department of Medicine, Perelman School of Medicine, University of Pennsylvania. Philadelphia, Pennsylvania, United States. 4. Department of Electrical and Systems Engineering, School of Engineering and Applied Sciences, University of Pennsylvania. Philadelphia, Pennsylvania, United States. 5. Perelman School of Medicine, University of Pennsylvania. Philadelphia, Pennsylvania, United States. * Corresponding author, [email protected]

1 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Abstract: 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, we developed a multichannel nanofluidic system to analyze crude clinical samples. Using this platform, we 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, we classified cancer and pre-cancer mice from healthy controls, as well as pancreatic cancer patients from healthy controls, in blinded studies. Keywords: cancer diagnostics, exosomes, nanofluidics, machine learning, pancreatic cancer

2 of 35 ACS Paragon Plus Environment

Page 2 of 35

Page 3 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

Many tumors (e.g. from the brain, pancreas, and lung) are located in regions of the body that are difficult to surgically access, and for which repeat biopsy is often impossible.1-5 Liquid biopsy, the minimally invasive detection of tumor material in patient blood samples, has generated many opportunities for the diagnosis and monitoring of cancer.6,7 Nanoscale exosomes (30-200 nm diameter) released during fusion of the multivesicular endosomes (MVEs) with the plasma membrane, and which are found circulating in the blood, have been discovered to contain molecular information of their cells of origin, relevant for disease diagnostics, disease monitoring, and drug efficacy screening.8-12 It has proven challenging to use conventional technology to establish the clinical utility of exosomal biomarkers and to utilize these biomarkers to improve patient care. Due to exosomes’ small size, conventional sizebased isolation is time consuming (> 6 hr), results in co-purification of cell debris, and cannot select specific sub-populations of exosomes or differentiate exosomes from other extracellular vesicles (e.g. microvesicles, oncosomes).7,13 While microfluidic technology can sort and detect rare cells,6,14-18 the scaling of these approaches to the nanoscale has been limited by the inherently low throughput and susceptibility to clogging of nanoscale fluid channels. To address these challenges, we have developed an approach for exosome isolation, wherein millions of nanofluidic exosome sorting components are incorporated onto a single chip and work in parallel to isolate exosomes from clinical samples. Our exosome track-etched magnetic nanopore (ExoTENPO) chip rotates conventional nanofluidic sorting by 90° to form magnetic traps at the edges of pores instead of in channels. By distributing the flow over millions of nanoscale pores, 1. flow rates (ɸ > 10 mL/hr) 106x greater than in individual nanofluidic devices can be achieved while preserving the precision of nanofluidic immunomagnetic sorting, and 2. sorting can be performed directly on unprocessed serum or plasma without risk of clogging, as the occlusion of any single nanopore results in diversion of the flow to a nearby pore.

3 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

In addition to the challenge of isolating tumor-derived exosomes, the successful diagnosis of cancer is made more challenging due to the complex nature of the disease and its heterogeneity both within the tumor itself and between patients.44 Conventional methods that rely on only a single molecular biomarker56,57 are often not sufficiently specific. Any single protein or nucleic acid biomarker can be controlled by many complex processes and will not necessarily map directly to a specific disease state that is universally true for all patients. As an alternative, a panel of molecular biomarkers can be measured to more comprehensively profile the complex state of the cancer. However, it can be difficult to make sense of multiple molecular biomarkers. To solve this challenge, we use machine learning algorithms and training sets of data to extract a set of optimized linear discriminators from a panel of RNA biomarkers. These discriminators are then evaluated using independent blinded test sets of data that are separate from the training data. By choosing relevant states of the cancer (pre-cancerous lesions, cancer, healthy) and training the machine learning algorithm to find patterns that optimally classify these states, we outperform any individual marker in blinded studies. We validated the utility of our ExoTENPO chip for cancer diagnostics by focusing on pancreatic cancer, in particular on the currently intractable problem of detecting the disease at an early stage.19,20 Using this approach, in blinded test sets, we were able to correctly classify healthy subjects from those with pancreatic cancer, in both a murine model and in a clinical cohort. Moreover, in the murine model, pancreatic intraepithelial neoplasia (PanIN) mice with premalignant pancreatic lesions were correctly identified relative to the healthy mice and mice with cancer. Results/Discussion The Exosome Track-Etched Magnetic Nanopore (ExoTENPO) Chip

4 of 35 ACS Paragon Plus Environment

Page 4 of 35

Page 5 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

The ExoTENPO isolates specific subtypes of exosomes by immunomagnetically labeling exosomal protein surface markers and subsequently capturing these targeted exosomes directly from unprocessed serum or plasma. (Fig. 1a) Both the paramagnetic material in ExoTENPO and the super-paramagnetic magnetic nanoparticles (MNPs) that label the exosomes are magnetized by an external NdFeB disc magnet (d = 1 1/2”, h = 3/4”, K&J Magnetics) placed immediately below the ExoTENPO device. The exosomes captured on the ExoTENPO are subsequently lysed on-chip, and their mRNA cargo is extracted and analyzed by quantitative PCR (qPCR). The qPCR data is analyzed using a machine learning algorithm to predict the state of the subject (i.e. Healthy, PanIN, or Cancer). (Fig. 1b) The ExoTENPO consists of a track-etched polycarbonate membrane coated with a 200 nm layer of magnetic material (Ni80Fe20, permalloy). The ExoTENPO device has an Adev = 15.2 cm2 sized membrane densely covered (ρ > 107/cm2) with magnetic nanopores (600 nm diameter). (Fig. 1a) The nanopores are designed to be big enough to not trap particles that are known to be present in serum or plasma(e.g. nucleic acid aggregates, cell debris, lipoproteins).68,69 Exosomes are magnetically labeled using two-step immunomagnetic labeling, with biotinylated antibodies and streptavidin-coated 50 nm MNPs (Miltenyi), allowing 10x less antibody usage than direct nanoparticle conjugation and more particle loading.24 Exosomes are labeled by targeting either tetraspanins (CD9, CD81), known to be expressed by exosomes, or markers that will enrich for exosomes derived from tumors of epithelial origin (e.g. EpCAM). Compared to other immunolabeling technologies where exosomes are directly captured to microbeads or on antibody-coated surfaces, here exosomes are labeled with multiple MNPs and the ExoTENPO selectively captures exosomes that have been labeled with sufficient MNPs to overcome the drag force. Therefore, background material with few non-specifically bound MNPs can be discarded by using a sufficiently high flow rate.

5 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

The key to ExoTENPO’s ultrasensitive isolation of exosomes is its magnetic nanopore design, which maximizes the magnetophoretic force Fm that isolates nanoscale exosomes while minimizing the competing drag force Fd. The two innovations that allow ExoTENPO to combine the precision of nanofluidic immunomagnetic sorting with a throughput sufficient for clinical diagnostics are detailed below: 1. Strong magnetophoretic forces (Fm↑). The magnetophoretic force Fm ~ (B⋅∇)B on targeted exosomes is maximized by increasing the strength of the applied field B, increasing its spatial changes ∇B, and by delivering the targeted object as close to the magnetic trap as possible. Large fields (0.4 T) are produced by an external NdFeB magnet. Large field gradients ∇B are created by the NiFe coated nanopores, which generate nanoscale field gradients at the pore’s edge, where there is a transition from the highly magnetically susceptible NiFe (𝜒 ≅ 105) to water (𝜒 ≅ -10-5). The nanoscale traps can be created because there is no inherent length-scale in Maxwell’s Equations for magnetostatics, in contrast to optical trapping where the size of a trap is limited by the wavelength of light.21 2. Weak drag forces (FD↓) at high volumetric flow rates (ɸ↑). To isolate an exosome, the magnetic trapping force Fm must overcome the drag force Fd. The drag force is proportional to the flow velocity of the fluid Fd ∝ v. The ExoTENPO chip consists of an Adev = 15.2 cm2 sized membrane densely covered (ρ > 106/cm2) with magnetic nanopores. Thus, even at the high volumetric flow rates (ɸ > 10 mL/hr), the flow velocity v, and thus the drag force Fd, within each pore vz = ɸ / (AdevρApore) can be kept small, where Apore is the cross sectional area of an individual pore. Finite element simulation We have constructed a finite element model to simulate the magnetic trapping capability of the ExoTENPO. (Fig. 1c) The field strength B drops rapidly in distance from the edge of the

6 of 35 ACS Paragon Plus Environment

Page 6 of 35

Page 7 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

nanopore, creating field gradients ∇B that lead to strong magnetic forces Fm. To be trapped, an exosome must first be translated radially by magnetophoretic forces Fr to the trap at the edge of the pore. The radial force Fr drops off quickly in distance from the pore’s edge, (Fig. S1) motivating the design goal of minimizing the pore diameter d to bring the exosome into close proximity to the strongest field regions. Once translated to the pore’s edge, the magnetophoretic trapping force Fz ≅ 1 nN must overcome the drag force Fd to successfully trap the exosome. (Fig. S1) In this chip, the magnetic trapping force far exceeds the drag force Fz >> Fd at the pore’s edge, for flow rates as large as ɸ =100 mL/hr. From this analysis, we extract three useful insights: 1. The capture rate decreases as flow rate ɸ increases, 2. the capture rate increases as the pore’s diameter d decreases, and 3. because the probability of capturing an exosome is a function of its initial radial position in the pore, the capture rate can be increased by placing multiple ExoTENPO membranes in series, (Fig. S2) allowing multiple, independent chances for capture. Multiple layers of ExoTENPO membranes are incorporated into a microfluidic device using laser cut laminate microfluidics. (Fig. S3) Characterization of ExoTENPO’s performance We characterized the ExoTENPO’s capture efficiency using 50 nm MNPs (Miltenyi) that have a uniform diameter and magnetization. To determine the concentration of nanoparticles, we measured the nuclear magnetic resonance T2 relaxation time (Bruker mq60 MR relaxometer) operating at 1.41T, (Fig. S4a) making use of the fact that the concentration of MNPs CMNP is proportional to 1/T2∝CMNP.23 As expected, the capture rate of MNPs decreased as flow rate increased. (Fig. S4b) To calculate capture rate, we first generated a calibration curve by measuring the NMR relaxation time T2 for various concentrations of MNPs CMNP. The capture rate ζ is defined as ζ = Cinput/Coutput, where Cinput is the concentration of MNPs before sorting and Coutput is the concentration of MNPs in the flow-through of the device. After we measured the T2

7 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

relaxation time for the input and output suspensions, we used the calibration curve to calculate the concentration of particles in the input Cinput and in the flow-through Coutput and reported the capture rate ζ. The capture rate ζ for a double filter (n = 2) decreased as a function of the flow rate ɸ as a power law, ζ = 264.6ɸm , m = -0.87 (R2 = 0.88). As predicted, the capture rate ζ could be recovered at high flow rates ɸ by stacking multiple layers n of the ExoTENPO membranes. (Fig. S4c) The capture rate ζ was found to increase exponentially with the number n where ζ = 9.1e0.64n (R2 = 0.92). This exponential increase (Fig. S4c) arises from the fact that each layer of ExoTENPO contains pores that have a random position, uncorrelated with the pores in the other layers. As an exosome passes through the multiple ExoTENPO membranes, separated by a 100 µm gap, it will pass through each pore at a random initial position and therefore have an independent probability of being captured in each layer. The probability of an exosome being captured by multiple ExoTENPOs in series compounds with the passage through each subsequent ExoTENPO. (Fig. S2) By placing n = 6 ExoTENPO in series, a high enrichment ζ > 500 was achieved at a flow rate sufficient for running clinical samples ɸ = 10 mL/hr. We verified that as the pore size was decreased from d = 3 µm to d = 600 nm, there was a significant increase in the capture rate ζ. (Fig. S4d) The above experiment was performed using a double (n = 2) filter at ɸ = 1mL/hr. To validate that the ExoTENPO isolated exosomes primarily due to magnetic trapping, we performed a negative control experiment where we turned off the magnetophoretic force by removing the external NdFeB magnet, which resulted in no particles captured ζ = 1 (Cinput = Coutput). Characterization and validation of exosome isolation using ExoTENPO To validate that our ExoTENPO can capture exosomes, we first isolated exosomes derived from cell culture media. Testing was carried out at ɸ = 10 mL/hr and with 6 layers of ExoTENPO membranes using human and mouse cell culture media as a source of exosomes (MiaPaCa2, Panc-1, AsPC1, BxPC3, PD7591) and performing isolation using either pan-exosome (CD9, 8 of 35 ACS Paragon Plus Environment

Page 8 of 35

Page 9 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

CD63, CD81) or epithelial markers (EpCAM). We performed the following characterization experiments to validate that our isolate contained material consistent with the standards laid out by the International Society for Extracellular Vesicles.65 First, we fixed the exosomes directly on the ExoTENPO nanopores after capture using pan-exosome markers and imaged them using scanning electron microscopy (SEM), (Fig. 1d, Fig. S5) and observed objects with a morphology consistent with exosomes covered with 50 nm magnetic nanoparticles. We also fixed the exosomes eluted from the ExoTENPO device and compared them to the exosomes isolated using ultracentrifugation. (Fig. S6) Both methods enriched for objects that have morphology and sizes that match with those of exosomes. However, we were able to observe more material in the isolate of ultracentrifugation that had a size larger than that consistent with exosomes d > 200 nm. We also analyzed the input and the isolate using dynamic light scattering (DLS). In the input, there was a distinct distribution of particles at d = 50.7 nm and a larger population of smaller particles (a = 10.1 nm) that are likely debris. (Fig. 1e) We eluted the exosomes captured by the ExoTENPO and found using DLS that 90% of the particles had a diameter a = 149.7 nm ±100 nm, consistent with that of exosomes. (Fig. 1e) We quantified the specificity of ExoTENPO capture by comparing the quantity of exosomal RNA capture from cell culture media using the ExoTENPO, using pan-exosome markers (CD9, CD63, CD81) versus a control wherein the antibodies were replaced with an isotype control (biotin mouse IgG1 k isotype, Biolegend). There was >10x RNA yield (P < 0.05) using the pan-exosome markers as affinity ligands compared to the isotype control, confirming the specificity of the ExoTENPO (Fig. S7a) We additionally performed a western blot on the isolate of our device, using EpCAM based capture on 500 µL healthy mouse plasma spiked with 15 mL PD7591 mouse pancreatic cancer cell cultured media, and showed that the isolate was positive for exosome markers (Alix, TSG101) and negative for Calnexin, known to not be expressed by exosomes.66 In comparison, isolates captured using ultracentrifugation were also positive for Alix and TSG101, but positive

9 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

for Calnexin, suggesting increased co-purification of non-exosomal material. We processed the material missed by our device, its flow through, using ultracentrifugation and found it to be positive for exosome markers and also positive for Calnexin, suggesting that our device did miss some exosomal material and that our device successfully discarded non-exosomal material positive for Calnexin. Additionally, we measured the platelet marker CD61, to check for background from the abundant platelet derived exosomes,67 and found that it was negative in the isolate of our device, but positive in the isolate of ultracentrifugation. (Fig. S7b) Isolation of RNA cargo from exosomes We extracted RNA from exosomes directly on the ExoTENPO membrane for downstream analysis. Total exosomal RNA & protein isolation kit (Life Technologies) was used for RNA extraction from isolated exosomes. Briefly, first denaturing solution was added to the chip, and the chip was incubated for 5 mins on ice. Then, the lysed solution was taken off-chip for acidphenol separation and washing steps using a spin column. The exosomal RNA was eluted in a small volume (~30 µL) and stored at -80°C or processed immediately for further analysis. We first compared our chip’s extraction of RNA to the yield of a conventional ultracentrifugation and a commercial kit (Total Exosome Isolation Kit, Life Technologies), which is known to result in higher yield than conventional ultracentrifugation.59 We used both cell culture media and a KPCY (Kras, p53, Pdx1-Cre, Yfp) mouse model, which is genetically engineered to develop pancreatic cancer. Using human pancreatic cancer cell cultured media, we achieved a 4-6x improvement in the quantity of RNA recovered compared to both ultracentrifugation and the commercial kit. (Fig. S7c) We also measured size of particles isolated by ultracentrifugation where we observed a broader peak (119 nm ± 113 nm) (Fig. S7d) compared to that of particles isolated by ExoTENPO (149.7 nm ± 100 nm) (Fig. 1e) The size and quality of the exosomal RNA extracted from mice plasma were characterized using a Bioanalyzer. Exosomal RNA

10 of 35 ACS Paragon Plus Environment

Page 10 of 35

Page 11 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

isolated using both the centrifugal method and our chip had the greatest occurrence at 24 bp, with a range from 22-30 bp for exosomes from a KPCY mouse (Fig. S8a, b) and 25-200 bp for exosomes from a cell culture media. (Fig. S9a, b). Using human pancreatic cancer cell cultured media and pan-exosome markers, we validated that the relative quantities of exosomal mRNA measured in exosomes isolated using the ExoTENPO matched that of the isolate acquired using a centrifugal method. We quantified 12 different genes using qPCR and measured their mRNA expression levels and found a positive correlation (R2 = 0.74) between the relative abundance of RNA isolated using ExoTENPO and a centrifugal method. (Fig. 1f) Thus, the ExoTENPO chip enables the efficient isolation of exosomes, whose RNA profile closely resembles that of the conventionally isolated material. We chose the genes in the panel because they are known to be differentially expressed in pancreatic cancer cells relative to the healthy state.28,47-55 Specifically, we first generated a list of genes known to be differentially expressed in pancreatic cancer cells,28,47-55 and then using the online database exocarta.org, selected a subset that were known to be packaged in exosomes. This list was validated in the subsequent in vitro and in vivo experiments, resulting in a panel of 8 exosomal mRNAs. We compared the mRNA expression levels of exosomes isolated using our chip with those of their cells of origin and found a positive correlation. We measured the mRNA expression using qPCR in both the cellular lysate and in the exosome isolate for 5 different human pancreatic cancer cell lines and 6 different highly expressed mRNAs from our list. (Fig. 1g) mRNA expression in this analysis was normalized to the mean expression for each sample. The mRNA expression levels from exosomes and cells were positively correlated (R2 = 0.88). (Fig. 1h)25 Diagnosis of pancreatic cancer using machine learning based mRNA signatures

11 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

To evaluate the performance of the ExoTENPO for the detection of pancreatic cancer, we first conducted a study on a cohort of KPCY mice, which are genetically engineered to develop pancreatic cancer. (Fig. 2a)27 The pancreatic cells in the KPCY mice express yellow fluorescent protein (YFP), enabling the true disease state, to be known. Thus, these mice are well suited to evaluate our chip’s capability to detect disease at different stages. The exosomal mRNA signatures of mice that are healthy, mice with pre-cancerous lesions (PanIN), and mice with tumors were measured, and from these measurements, a predictive panel of exosome-based biomarkers for pancreatic cancer was developed and tested using an independent, user blinded cohort of mice (N = 18). We selected a panel of candidate exosomal mRNA biomarkers using qPCR that are known to be differentially expressed in pancreatic cancer relative to healthy state.28,47-55 We isolated exosomes from approximately 0.5 mL of undiluted plasma from each mouse on the ExoTENPO (n = 6 membranes) based on their expression of epithelial cell adhesion molecule (EpCAM), which is known to be expressed by cancer cells of epithelial origin. We measured the exosomal mRNA profile using qPCR from a training set of N = 5 healthy, N = 5 PanIN, and N = 5 mice with tumors. Amongst these mice, by measuring only the total quantity of RNA, it was not possible to accurately classify the mice (P > 0.5). (Fig. S10a) Amongst the panel of mRNA that we measured, several genes were differentially expressed between the groups (e.g. H3F3A). (Fig. S10b) However, no single gene was able to classify individual mice into the correct groups, due to the variance in expression amongst mice within groups. We used linear discriminant analysis (LDA) to identify linear combinations of the mRNA profile that can discriminate between healthy mice, mice with PanINs, and mice with tumors. We first generated a prediction algorithm by running LDA on a training set wT where the true state of each subject is a priori known T. (Fig. 2b) The LDA algorithm generates a vector x that is used

12 of 35 ACS Paragon Plus Environment

Page 12 of 35

Page 13 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

to calculate a weighted sum w · x that maximally separates the healthy controls from either the subjects with PanIN lesions or with tumors. These LDA vectors were initially tested on the training set by performing N-1 cross validation, 29 and were then subsequently evaluated using a blinded test set wb (Fig. 2c), independent from the training set, where the true state of the mouse was not known a priori. (Fig. 2d) We used this technique to first do a pairwise differentiation between subjects that are healthy and those with either PanINs or tumors. (Fig. 2e) In both cases, there exists a threshold value ψ, above which subjects are predicted to be one class and below which they are predicted to be in the other class. N-1 cross validation on these pairwise comparisons result in correct identification of each mouse. (Fig. 2f) Subsequently, in an independent user-blinded test set, which was separate from the training set, all mice (N = 18) were correctly identified as either tumor, PanIN, or healthy. (Fig. 2g) We also validated the specificity of the predictive power of LDA analysis by comparing the performance to a data set where a training set of scrambled group labels was created. As expected, the prediction algorithm generated by the scrambled training set had no predictive value, resulting in an AUC of 0.5 and 0.53 for healthy versus PanIN and healthy versus tumor respectively. (Fig. 2h) Enrichment of specific exosome populations from complex media One key advantage of the ExoTENPO is its capability to specifically capture sub-populations of exosomes based on the expression of particular surface markers. Beyond its use in diagnostics, this capability can be used to aid in the study of the biological function of subtypes of exosomes. 26

To demonstrate this capability, we showed that subtypes of exosomes could be isolated

directly from plasma. We spiked pancreatic cancer cell cultured media (PD6910) into healthy mouse plasma to compare exosomal mRNA isolated using a cocktail of pan-exosome markers (CD9, CD81) to exosomes isolated using EpCAM. Compared to pan-exosome marker based

13 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

capture, EpCAM based isolation was able to more specifically isolate a subset of epithelial cellderived exosomes from the background of healthy plasma. (Fig. 3a) We quantified this specificity by calculating ΔCt = Ct,spiked -Ct,healthy, where Ct,spiked is the qPCR threshold cycle measured in the isolate from the spiked sample and Ct,healthy is the threshold cycle from the isolate of the healthy control. By using DLS, we confirmed that EpCAM captured exosomes had an average diameter a = 169.9 nm, (Fig. 3b) consistent with the results found using the panexosome marker. (Fig. 1e) To test the value of our ExoTENPO’s surface marker specific isolation of exosomes, we compared our device’s ability, using the same mRNA markers and the same LDA approach described in Fig. 2, to diagnose pancreatic cancer using EpCAM-based isolation versus a panexosome (CD9, CD81) isolation in a blinded test set. (Fig. 3c, d) We showed that using EpCAM isolation, all mice could be correctly discriminated. Whereas using pan-exosome (CD9, CD81) isolation, which isolates all exosomes in circulation, the performance was markedly degraded (AUC = 0.81 for tumor vs. healthy and AUC = 0.53 for PanIN vs. healthy). (Fig. 3c, d) Clinical diagnosis of pancreatic cancer with ExoTENPO To evaluate the ExoTENPO’s capability to diagnose pancreatic cancer in clinical specimens, we first characterized the performance of the chip isolating exosomes from human blood samples. We compared the recovery of exosomal RNA from healthy human plasma samples using the ExoTENPO (using pan-exosome isolation: CD63, CD9, CD81), to a centrifugal method (Total Exosome Isolation Kit, Life Technologies) and found a 1.6x improvement in recovery. (Fig. S11) We compared the recovery of exosomal RNA in a variety of common clinical sample types, including fresh plasma, fresh serum, frozen plasma, and frozen serum using pan-exosome isolation and found no significant difference in the RNA recovery from the various sample types (P = 0.23). (Fig. S12a) Next, we performed pairwise comparisons of the relative abundance of

14 of 35 ACS Paragon Plus Environment

Page 14 of 35

Page 15 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

mRNA targets (CK18, CD63, Erbb3, KRAS) and found that similar exosomal RNA cargo profiles (R2 > 0.77) were obtained from the different sample preparations. (Fig. S12b) Thus, our platform can be used on any of these available sample types and provide comparable information. To identify the affinity ligand to use for our clinical measurements, we compared using panexosome isolation versus cancer epitopes to enrich for subset of tumor-derived exosomes from plasma. We spiked human pancreatic cancer cell culture media into healthy human plasma to compare isolated exosomes using a cocktail of pan-exosome markers (CD63, CD9, CD81) to those using a candidate tumor marker, EpCAM. Compared to pan-exosome marker based capture, positive capture using EpCAM showed the greatest mRNA expression level difference ΔCt = Ct,spiked -Ct,healthy between cancer exosome spiked plasma Ct,spiked and healthy plasma Ct,healthy. (Fig. S12c) We conducted a study on an independent blinded test set of N = 34 clinical samples, including previously untreated metastatic pancreatic cancer patients and healthy controls. (Fig. S13) The same panel of 8 exosomal mRNA biomarkers identified in our mouse studies was used. For each subject, we isolated exosomes from 3 mL of undiluted plasma using EpCAM-based isolation on the ExoTENPO. We first measured the exosomal mRNA profile from a training set of N = 5 healthy controls and N = 5 patients with cancer. (Fig. 4a) Amongst the panel of mRNA that we measured, several genes were differentially expressed between the groups (e.g. CD63), but no single gene was able to classify individual patients into the correct groups due to the variance in expression amongst patients within groups. We first evaluated the diagnostic ability of this approach using N-1 cross-validation to prevent overfitting,29 and classified every patient into the correct group. (Fig. 4b) To validate this approach, we applied this technique to our independent, user-blinded test set (N = 24) where we classified each patient correctly,

15 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

demonstrating a positive predictive value (P < 0.001, Fisher’s Exact Test). (Fig. 4c) Additionally, we directly compared the performance of LDA of our panel of 8 exosomal RNA biomarkers to that of fewer biomarkers, down to only a single biomarker. (Fig. 4d) As the number of genes increased, our diagnostic was able to detect the mRNA signature associated with specific disease states, and the performance quickly increased, achieving correct classification for N ≧4 genes in a blinded test set. Conclusions We have developed an approach to isolate specific subtypes of exosomes with robust, high throughput nanofluidic sorting. We have used LDA on the exosomal RNA extracted from these exosomes to diagnose cancer in both a mouse model and human subjects and to diagnose precancerous lesions in a mouse model, all in blinded studies. The ExoTENPO was able to properly classify all patient samples and distinguish between healthy subjects from those with pancreatic cancer. The detection of pre-cancerous lesions (PanINs) in the murine model provides a potential clinical testing on using the ExoTENPO to screen for early stage pancreatic cancer. LDA was used in this study because it is a relatively simple and well understood technique that can discriminate categorical variables using a set of continuous variables. More complicated machine learning algorithms can be used to further improve performance42 but were not explored in this study due to the limited number of samples N. This work builds on the success of using microfluidic immunomagnetic sorting to isolate rare cells, achieving selectivity and sensitivity not possible using macroscale technologies.17,30-32, 70 This work also builds on previous work by our group using a technique to sort mammalian cells and bacteria.39,45,58 The ExoTENPO developed here integrates several technological advances to translate this approach to nanoscale exosomes. 1. Due to its parallelization, the ExoTENPO can precisely sort nanoscale exosomes at flow rates sufficient for clinical use (Φ = 10 mL/hr)

16 of 35 ACS Paragon Plus Environment

Page 16 of 35

Page 17 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

and is insensitive to clogging from unprocessed clinical samples, as the occlusion of any single nanopore results in diversion of the flow to a nearby pore. 2. By harnessing nanoscale feature sizes to develop a nanoscale immunomagnetic trap, the ExoTENPO sorts each exosome individually based on its labeling by a sufficient number of MNPs, analogous to the microfluidic devices used to selectively sort rare circulating tumor cells.17,30-32 This functionality distinguishes the ExoTENPO from previous microfluidic technologies and microbead systems (e.g. Dynabead, ThermoFisher) that capture exosomes onto functionalized surfaces61 or microbeads62, wherein the capture purity is set solely by the affinity ligand specificity. Because of our device’s surface marker specific isolation, it is less susceptible to co-purifying nonexosomal debris known to be present in serum or plasma (e.g. nucleic acid aggregates, cell debris, free-floating proteins).68,69 3. By using track etching to fabricate the nanopores, precise nanoscale features can be fabricated over a large area for low cost, making the chip manufacturable for medical diagnostic use. 4. The use of LDA to mathematically combine a panel of exosomal RNA biomarkers enabled the accurate diagnosis of cancer, even in the presence of variance in individual exosomal RNA across individual specimens. In comparison to recent work that has shown that pancreatic cancer specific markers can be detected directly from serum or plasma using on-chip plasmonic sensing to diagnose pancreatic cancer,61 we hypothesize that our approach’s higher sensitivity and specificity in the diagnosis of cancer versus healthy patients is because of our more specific exosome isolation and our machine learning approach that combines the measurement of N = 8 exosomal RNA biomarkers. Recent work has shown the benefit of isolating exosomes and analyzing their content of proteins, RNA, and DNA,62 and we believe our machine learning approach can be adapted to analyze protein and DNA content as well. Our prototype ExoTENPO costs approximately $15 per device (material cost: $6, fabrication expense : $6, antibody cost: $3/mL plasma) with unlimited sample volume processing at a throughput of 10mL/hr. Our device is less expensive than existing

17 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

exosome isolation kits (Qiagen: $33.15 per sample with limited sample volume of 4mL, Life Technologies: $16.40 per mL of sample). Because of the ExoTENPO’s all plastic construction, except for the NiFe coated membranes which can also be fabricated at low cost,39 it is compatible with ultra-low cost injection molded plastic production. There are several aspects of the ExoTENPO that can be further developed to expand the system’s functionality to detect a wider range of diseases. The size of the ExoTENPO used in this study (d = 600 nm) could be made smaller or larger to tailor it to alternative extracellular vesicle targets, such as microvesicles (200 nm - 1 µm) or oncosomes (≅ 1 µm). Building on the success of this study, multiple ExoTENPO modules can be integrated to isolate multiple subsets of exosomes targeted using either single surface markers or cocktails of surface markers,33 enabling an increasingly comprehensive view of a cancer disease stage to be measured. Finally, by integrating nanoscale nucleic acid detection34,35,35,36 downstream of the ExoTENPO in future work, we can create a compact, self-contained device for a point-of-care use. Moreover, although this paper has focused on diagnostics, exosomes have been demonstrated in recent years to play a role in intracellular communication.63,64 Our chip’s capability to specifically isolate exosome sub-populations can be an important tool in the study of this exosome-mediated communication. With its high throughput, small footprint, and ease of operation, the ExoTENPO is poised to offer a clinician-friendly tool to translate findings in the emerging field of exosome biomarkers to improve the health of patients. Methods/Experimental Chip fabrication The ExoTENPO membranes are fabricated by thermally evaporating (Kurt Lesker PVD-75, Singh Nanofabrication Facility, University of Pennsylvania) a 200 nm layer of permalloy (Ni80Fe20) isotropically onto the surface of a track-etched polycarbonate membrane (Whatman). A 30 nm layer of gold is subsequently deposited to prevent oxidation of the

18 of 35 ACS Paragon Plus Environment

Page 18 of 35

Page 19 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

permalloy. The ExoTENPO membranes are incorporated into a laminate sheet microfluidic device fabricated by laser micromachining (Universal Laser VLS 3.50) sheets of moistureresistant polyester film (McMaster-Carr, 0.004” thick). (Fig. S1) The multiple layers of polycarbonate membranes (N = 6) are stacked using a solvent-resistant double-sided sticky tape (McMaster-Carr). The tape is 100 µm thick, providing a gap of 100 µm between membrane layers, such that exosomes pass through the pores of multiple ExoTENPO membranes.An optically clear cast acrylic sheet (McMaster-Carr) is used to form a reservoir for the input, and the output is made using a polydimethylsiloxane (PDMS) piece, pressure-fit to tygon tubing to connect to a syringe, which collects the flow-through waste of the ExoTENPO. The syringe is connected to a syringe pump (Programmable Syringe Pump, Braintree Scientific) that pulls the sample through the device at flow rates from ɸ = 0.1 - 50 mL/hr. To prevent nonspecific binding, the ExoTENPO is coated with pluronic F-127 before use and then flushed with PhosphateBuffered Saline (PBS). During the coating process, high flow rates (> 10 mL/hr) are used to flush all air bubbles from the device. Finite element simulation Matlab and Comsol were used for finite element simulation. Because each pore is axially symmetric, we calculated and plotted the simulated magnetic field on the cross-section of the pore. The nanopore was modeled as a disc, with a diameter d = 600 nm and height h = 200 nm, with boundary conditions of zero field at large distances. The magnetophoretic force Fm on an exosome as it passes through a nanopore was calculated by combining the results from the simulation with a simplified model for the exosome. The magnetic moment of the exosome is proportional to the number of MNPs n and the moment mp of the particle (m = n * mp). The model assumes that the magnetic particles are fully magnetized by the externally applied field Bo ≈ 0.4 T ẑ. We assumed mp = 9.27×10-3 mA*µm22 and that each targeted exosome has n > 5 MNPs. The number of MNPs (a = 50 nm Fe2O3, Miltenyi) per

19 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

exosome is calculated based on the limit imposed by steric hinderance on the smallest exosomes, a = 30 nm, assuming 50% maximum loading. Sample collection (mice) All mouse work was performed in compliance with institutional and IACUC guidelines. Blood was obtained at sacrifice by cardiac puncture from the right ventricle of tumor-bearing Pdx1-cre, KrasLSL-G12D, p53L/+, RosaYFP/YFP (KPCY) mice.27 p53L/+, RosaYFP/YFP (PY) and KrasLSL-G12D, p53L/+, RosaYFP/YFP (KPY) littermates were used as controls. Blood was collected in sodium citrate coated blood collection tubes (BD Vacutainer™). Plasma was isolated by centrifuging the blood at 1600g for 10min, followed by a second spin at 3000g for 10min to remove cellular contamination. Sample collection (human) Peripheral whole blood was obtained from patients with stage IV pancreatic cancer and from healthy gender-matched controls at the University of Pennsylvania Health System. All patients and healthy donors provided written informed consent for blood donation on approved institutional protocols. Whole blood was drawn in either EDTA (Fisher Scientific), Streck Cell-Free DNA BCT® (Streck), or gel serum separation tubes (Fisher Scientific). Plasma and serum were isolated using the following procedures. Within 3 hours of blood draw for EDTA and within 12 hours of blood draw for Streck, tubes were centrifuged at 1600g for 10 minutes at room temperature with the brake off. Next, plasma was transferred to a fresh 15 mL centrifuge tube without disturbing the cellular layer and centrifuged at 3000g for 10 minutes (EDTA) or 4122g for 15 minutes (Streck) at room temperature with the brake off; this step was repeated with a fresh 15 mL centrifuge tube. After the third spin, plasma was transferred to a fresh 15 mL centrifuge tube, gently mixed, and transferred in 1mL aliquots to centrifuge tubes and either processed fresh for exosomal RNA or stored immediately at -80°C for future use. Gel serum separation tubes were stored at room temperature for 30 minutes after the blood draw. Within 2 hours of blood draw, serum tubes were centrifuged at 1000g for 15

20 of 35 ACS Paragon Plus Environment

Page 20 of 35

Page 21 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

minutes at room temperature. Last, serum was transferred in 1mL aliquots to cryovials and either processed fresh for exosomal RNA or stored immediately at -80°C for future use. Cell culture Mouse cell lines PD7591, PD483, and PD6910 were generated from pancreatic tumor tissue isolated from Pdx1-cre, KrasLSL-G12D, p53L/+, RosaYFP/YFP (KPCY) mice.27 The cell lines were cultured in media as previously described.37 The cell lines were periodically checked for mycoplasma contamination. All human cell lines were cultured in media recommended by ATCC. Western blot Protein was quantified using Qubit Protein Assay kit (ThermoFisher Scientific). 30 µg of protein was loaded per well and separated by electrophoresis using NuPage 4-12% BisTris gels (ThermoFisher Scientific). Protein was transferred on to polyvinylidene fluoride (PVDF) membrane (Bio-Rad) and blocked with 5% Milk in PBS + 0.05% Tween-20. Primary antibodies used were Purified Anti-ALIX Antibody (634502, BioLegend), Anti-TSG101 Antibody (ab125011, Abcam), Anti-Calnexin Antibody (C4731, Sigma Aldrich), and Anti-CD61 Antibody (VPA00632, Bio-Rad) followed by horseradish-peroxidase conjugated secondary antibodies (Anti-rabbit HRP (Jackson labs, 711-036-152), Anti-mouse HRP (Jackson labs, 715-035-150)). Proteins of interest were detected using chemiluminescent substrate (Thermo Fisher Scientific). Exosome isolation (ultracentrifugation) Supernatant fractions from confluent cell cultures (48-72h) were collected and centrifuged at 400g for 5 minutes to remove dead cells and debris. After removal, the supernatant was first spun at 17,000g for 40 minutes and then at 100,000 g for 2 hours using Type 70.1 Ti rotor (Optima XPN-100 Ultracentrifuge, Beckman Coulter).60 The exosome pellet was used fresh for downstream analysis. Exosome isolation (kit) Supernatant fractions from confluent cell cultures (48-72h) were collected and centrifuged at 400g for 5 minutes to remove dead cells and debris. Total exosome isolation reagents (from serum, plasma, and cell culture media) from Life Technologies were

21 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

used. The protocol was followed as suggested by the company. Isolated exosomes were stored at 4°C for short term storage or immediately processed for further analysis. Exosome isolation (ExoTENPO) All samples are vortexed prior to exosome isolation to resuspend any aggregated exosomes, which could be caused by frozen storage condition. Antibiotin ultrapure microbeads (Miltenyi Biotec) and biotinylated antibodies were used for magnetic labeling. For mouse, biotin anti-CD9 antibody (BioLegend) and biotin anti-CD81 antibody (BioLegend, 104903) were used. For human, biotin anti-human CD9 antibody (eBioscience, 13-0098-80), biotin anti-human CD63 antibody (BioLegend, 353017), and biotin anti-CD81 antibody (custom conjugated by BioLegend) were used. First, biotinylated antibodies (2.5 µL antibody per 20 mL media or per 1 mL plasma) were added to the sample and incubated for 20 min at room temperature with shaking. Then, anti-biotin ultrapure microbeads were added to the samples and incubated for 20 min at room temperature with shaking. The samples were then added to the reservoir of the ExoTENPO chip, and negative pressure was applied by a programmable syringe pump (Braintree Scientific). As the samples were pulled through the chip, magnetically labeled exosomes were captured at the edge of the pores of the chip. The total time to process a 3 mL sample, run through our device at 10 mL/hr, including sample preparation time, is 40 minutes. Exosomal RNA isolation Total exosomal RNA & protein isolation kit (Life Technologies) was used for RNA extraction from isolated exosomes. For the exosomes captured on-chip, denaturing solution was added to the chip, and the chip was incubated for 5 mins on ice. Then, the lysed solution was taken off-chip for acid-phenol separation and washing steps using a spin column. The exosomal RNA was eluted in a small volume (~30 µL) and stored at -80°C or processed immediately for further analysis.

22 of 35 ACS Paragon Plus Environment

Page 22 of 35

Page 23 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

Polymerase chain reaction (PCR) RT-PCR was first performed using exosomal RNA. PrimeScript RT Reagent Kit (Clontech) was used for RT-PCR. Using the kit, the exosomal RNA was mixed with reagents and run in a T100 Thermal Cycler (Bio-Rad) as per the manufacturer’s protocol. qPCR Master mix that consists of SsoAdvanced Universal SYBR Green Supermix (Bio-Rad), primers (Integrated DNA Technologies), and water were made at a 5:0.5:3.5 ratio and 9 µL of the master mix was added to each well, followed by 1 µL of cDNA. For every qPCR experiment, 40 cycles were run using a CFX384 Touch Real-Time PCR machine (Bio-Rad). The threshold level was determined automatically using the default settings on the CFX384 where the software sets the threshold at 10 times the standard deviation of the fluorescence value of the baseline. Samples were run in triplicate, and melting curves were checked prior to analysis. Linear Discriminant Analysis (LDA) Using Matlab (R2015b), multiple features (genes) from multiple groups (healthy, PanIN, tumor) were incorporated for classification using LDA. The code used to perform linear discriminant analysis (LDA) is included in Fig. S14. The confusion matrix and the LDA plot were made using results from Matlab. Cross validation (N-1) method was used for analysis of the training set data. An independent de novo user blinded validation set was used for the final evaluation. NMR relaxometry 50 nm magnetic nanoparticles (MNPs) (Miltenyi) were used to test the performance of the chip. Serial dilution of the MNPs was used t to generate a calibration curve (T2 relaxation time vs. bead concentration). Next, a suspension of the MNPs was run through the chip and flow through was collected, and its T2 relaxation value measured. All samples were measured using the minispec (Bruker) for T2 relaxation time. Dynamic light scattering (DLS) To obtain the size distribution of exosomes in our samples, we used DLS (Zetasizer, Malvern). For each measurement, 300-400 µL samples were measured in

23 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

triplicate. Captured exosomes were eluted after incubating in an elution buffer (System Biosciences) for 30 min at room temperature. Elution efficiency was not determined as our assay directly lyses exosomes on-chip, which does not require elution, release, and recovery of exosomes. RNA analysis The size of the exosomal RNA was measured using a BioAnalyzer. Exosomal RNA was measured in the BioAnalyzer using the Agilent RNA Pico chip at the NAPCore Facility at the Children’s Hospital of Philadelphia. The amount and concentration of the exosomal RNA were measured using the Qubit RNA HS Assay Kit (Thermo Fisher Scientific). Statistical Analyses To generate a predictive model in our analysis of both human and KPCY mouse subjects, we used a cohort of training set data and linear discriminant analysis, carried out using Matlab. We tested our predictive models using an independent, user blinded validation set to avoid the risk of overfitting our data. We evaluated the predictive value using Fisher’s Exact Test. No outlier analysis was performed. Sample size was chosen using Fisher’s Exact Test to measure the P value of our chip’s predictive value.

Supporting Information The supporting information is available online. It includes ExoTENPO characterization, mechanism schematic, layer-by-layer images, SEM images, western blot, control/specificity data, Bioanalyzer data, RNA yield/expression level comparison, demographic information of clinical samples, and a MATLAB code for machine learning. 


24 of 35 ACS Paragon Plus Environment

Page 24 of 35

Page 25 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

References and Notes: 1. 2.

3.

4. 5.

6. 7.

8. 9.

10.

11.

12.

13.

14.

Distler, M.; Aust, D.; Weitz, J.; Pilarsky, C.; Grutzmann, R. Precursor Lesions For Sporadic Pancreatic Cancer: PanIN, IPMN, and MCN. BioMed Res. Int. 2014, 2014, 474905. Morris, J. P.; Cano, D. A.; Sekine, S.; Wang, S. C.; Hebrok, M. Beta-Catenin Blocks KrasDependent Reprogramming Of Acini Into Pancreatic Cancer Precursor Lesions In Mice. J. Clin. Invest. 2010, 120, 508–520. Suram, A.; Kaplunov. J.; Patel, P. l.; Ruan, H.; Cerutti, A.; Boccardi, V.; Fumagalli, M.; Di Micco, R.; Mirani, N.; Gurung, R. L.; Hande, M. P.; d’Adda di Fagagna, F.; Herbig, U. Gurung. Oncogene-Induced Telomere Dysfunction Enforces Cellular Senescence In Human Cancer Precursor Lesions. EMBO J. 2012, 31, 2839–2851. Mullerat, J.; Deroide, F.; Winslet, M. C.; Perrett, C. W. Proliferation And P53 Expression In Anal Cancer Precursor Lesions. Anticancer Res. 2002, 23, 2995–2999. Hammes, L. S.; Tekmal, R. R.; Naud, P.; Edelweiss, M. I.; Kirma, N.; Valente, P. T.; Syrjänen, K. J.; Cunha-Filho, J. S. Up-Regulation Of Vegf, C-Fms And Cox-2 Expression Correlates With Severity Of Cervical Cancer Precursor (Cin) Lesions And Invasive Disease. Gynecol. Oncol. 2008, 110, 445–451. Lang, J. M.; Casavant, B. P.; Beebe, D. J. Circulating Tumor Cells: Getting More From Less. Sci. Transl. Med. 2012, 4, 141ps13–141ps13. Ko, J.; Carpenter, E.; Issadore, D. Detection And Isolation Of Circulating Exosomes And Microvesicles For Cancer Monitoring And Diagnostics Using Micro-/Nano-Based Devices. Analyst. 2016, 141, 450–460. Taylor, D. D.; Gercel-Taylor, C. Microrna Signatures Of Tumor-Derived Exosomes As Diagnostic Biomarkers Of Ovarian Cancer. Gynecol. Oncol. 2008,110, 13–21. Shao, H.; Chung, J.; Balaj, L.; Charest, A.; Bigner, D. D.; Carter, B. S.; Hochberg, F. H.; Breakefield, X. O.; Weissleder, R.; Lee, H. Protein Typing Of Circulating Microvesicles Allows Real-Time Monitoring Of Glioblastoma Therapy. Nat. Med. 2012,18, 1835–1840. Melo, S. A.; Luecke, L B.; Kahlert, C.; Fernandez, A. F.; Gammon, S. T.; Kaye, J.; LeBleu, V. S.; Mittendorf, E. A.; Weitz, J.; Rahbari, N.; Reissfelder, C.; Pilarsky, C.; Fraga, M. F.; Piwnica-Worms, D.; Kalluri, R. Glypican-1 Identifies Cancer Exosomes And Detects Early Pancreatic Cancer, Nature. 2015, 523, 177–182. Costa-Silva, B., Aiello, N. M.; Ocean, A. J.; Singh. S.; Zhang, H.; Thakur, B. K.; Becker, A.; Hoshino, A.; Mark, M. T.; Molina, H.; Xiang, J.; Zhang, T.; Theilen, T. M.; Garcia-Santos, G.; Williams, C.; Ararso, Y.; Huang, Y.; Rodrigues, G.; Shen, T. L.; Labori, K. J.; Lothe, I. M.; Kure, E. H.; Hernandez, J. Doussot, A.; Ebbesen, S. H. Grandgenett, P. M.; Hollingsworth, M. A.; Jain, M.; Mallya, K.; Batra, S. K.; Jarnagin, W. R.; Schwartz, R. E.; Matei, I.; Peinado, H.; Stanger, B. Z.; Bromberg, J.; Lyden, D. Pancreatic Cancer Exosomes Initiate Pre-Metastatic Niche Formation In The Liver. Nat. Cell Biol. 2015, 17, 816-26. Chen, C.; Skog, J.; Hsu, C.H.; Lessard, R.T.; Balaj, L.; Wurdinger, T.; Carter, B.S.; Breakefield, X.O.; Toner, M.; Irimia, D., Microfluidic Isolation And Transcriptome Analysis Of Serum Microvesicles, Lab Chip .2010, 10, 505-511. Thery, C.; Amigorena, S.; Raposo, G.; Clayton, A., Isolation And Characterization Of Exosomes From Cell Culture Supernatants And Biological Fluids. Curr. Protoc. Cell. Biol. 2006,DOI: 10.1002/0471143030.cb0322s30, 3–22. Stott, S. L.; Hsu, C. H.; Tsukrov, D. I.; Yu, M.; Miyamoto, D. T.; Waltman, B. A.; Rothenberg, S. M.; Shah, A. M.; Smas, M. E.; Korir, G. K.; Floyd, F. P.; Gilman, A. J.; Lord, J. B.; Winokur, D.; Springer, S.; Irimia, D. l.; Nagrath, S.; Sequist, L. V.; Lee, R. J.; Isselbacher, K. J.; Maheswaran, S.; Haber, D. A.; Toner, M. Isolation Of Circulating Tumor

25 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

15.

16.

17.

18. 19. 20.

21. 22. 23.

24.

25.

26.

27.

28.

29. 30.

Cells Using A Microvortex-Generating Herringbone-Chip. Proc. Natl. Acad. Sci. 2010, 107, 18392–18397. Issadore, D.; Chung, J.; Shao, H.; Liong, M.; Ghazani, A. A.; Castro, C. M.; Weissleder, R.; Lee, H., Ultrasensitive Clinical Enumeration Of Rare Cells Ex Vivo Using A Micro-Hall Detector. Sci. Transl. Med. 2012, 4, 141ra92–141ra92. Hu, X.; Bessette, P. H.; Qian, J.; Meinhart, C. D.; Daugherty, P. S.; Soh, H. T., MarkerSpecific Sorting Of Rare Cells Using Dielectrophoresis. Proc. Natl. Acad. Sci. U. S. A. 2005, 102, 15757–15761. Xia, N.; Hunt, T. P.; Mayers, B. T.; Alsberg, E.; Whitesides, G. M.; Westervelt, R. M.; Ingber, D. E., Combined Microfluidic-Micromagnetic Separation Of Living Cells In Continuous Flow. Biomed. Microdevices. 2006, 8, 299–308. Skelley, A. M.; Kirak, O.; Suh, H.; Jaenisch, R.; Voldman, J., Microfluidic Control Of Cell Pairing And Fusion. Nat. methods. 2009, 6, 147–152. Siegel, R. L.; Miller, K. D.; Jemal, A., Cancer Statistics, 2015. Ca-Cancer J. Clin. 2015, 65, 5–29. Ghatnekar, O.; Andersson, R.; and, S., Marianne; Persson, U.; Ringdahl, U.; Zeilon, P.; Borrebaeck, C. A. K., Modelling The Benefits Of Early Diagnosis Of Pancreatic Cancer Using A Biomarker Signature. Int. J. Cancer. 2013, 133, 2392–2397. Ashkin, A.; Dziedzic, J. M.; Bjorkholm, J. E.; Chu, S. Observation Of A Single-Beam Gradient Force Optical Trap For Dielectric Particles. Opt. Lett. 1986, 11, 288–290. Tchikov, V.; Fritsch, J.; Kabelitz, D.; Schutze, S. 2-Immunomagnetic Isolation Of Subcellular Compartments. Methods Microbiol. 2010, 37, 21–33. Brown, K. A.; Vassiliou, C. C.; Issadore, D.; Berezovsky, J.; Cima, M. J.; Westervelt, R. M. Scaling Of Transverse Nuclear Magnetic Relaxation Due To Magnetic Nanoparticle Aggregation, J. Magn. Magn. Mater. 2010, 322, 3122–3126. Haun, J. B.; Devaraj, N. K.; Hilderbrand, S. A.; Lee, H.; Weissleder, R. Bioorthogonal Chemistry Amplifies Nanoparticle Binding And Enhances The Sensitivity Of Cell Detection. Nat. Nanotechnol. 2010, 5, 660–665. Vlassov, A. V.; Magdaleno, S.; Setterquist, R.; Conrad, R.; Exosomes: Current Knowledge Of Their Composition, Biological Functions, And Diagnostic And Therapeutic Potentials. Biochim. Biophys. Acta. 2012, 1820, 940–948. Peinado, H.; Alečković, M.; Lavotshkin, S.; Matei, I.; Costa-Silva, B.; Moreno-Bueno, G.; Hergueta-Redondo, M.; Williams, C.; Garcia-Santos, G.; Nitadori-Hoshino, A.; Hoffman, C.; Badal, K.; Garcia, B. A.; Callahan, M. K.; Yuan, J.; Martins, V. R.; Skog, J.; Kaplan, R. N.; Brady, M. S.; Wolchok, J. D.; Chapman, P. B.; Kang, Y.; Bromberg, J.; Lyden, D. Melanoma Exosomes Educate Bone Marrow Progenitor Cells Toward A Pro-Metastatic Phenotype Through Met. Nat. Med. 2012, 18, 883–891. Rhim, A. D.; Mirek, E. T.; Aiello, N. M.; Maitra, A.; Bailey, J. M.; McAllister, F.; Reichert, M.; Beatty, G. L.; Rustgi, A. K.; Vonderheide, R. H.; Leach, S. D.; Stanger, B. Z. EMT And Dissemination Precede Pancreatic Tumor Formation, Cell. 2012, 148, 349–361. Yu, M.; Ting, D. T.; Stott, S. L.; Wittner, B. S.; Ozsolak, F.; Paul, S.; Ciciliano, J. C.; Smas, M. E.; Winokur, D.; Gilman, A. J. RNA Sequencing Of Pancreatic Circulating Tumour Cells Implicates WNT Signalling In Metastasis. Nature. 2012, 487, 510–513. Efron, B.; Gong, G. A Leisurely Look At The Bootstrap, The Jackknife, And CrossValidation, Am. Stat. 1983, 37, 36–48. Adams, J. D.; Kim, U.; Soh, H. T.; Multitarget Magnetic Activated Cell Sorter, Proc. Natl. Acad. Sci. 2008, 105, 18165–18170.

26 of 35 ACS Paragon Plus Environment

Page 26 of 35

Page 27 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

31.

32.

33. 34. 35. 36. 37.

38. 39.

40.

41. 42. 43.

44.

45. 46.

47.

48.

Issadore, D.; Shao, H.; Chung, J.; Newton, A.; Pittet, M.; Weissleder, R.; Lee, H. SelfAssembled Magnetic Filter For Highly Efficient Immunomagnetic Separation, Lab Chip. 2011, 11, 147–151. Ozkumur, E.; Shah, A. M.; Ciciliano, J. C.; Emmink, B. L.; Miyamoto, D. T.; Brachtel, E.; Yu, M.; Chen, P.; Morgan, B.; Trautwein, J. Inertial Focusing For Tumor Antigen-Dependent And--Independent Sorting Of Rare Circulating Tumor Cells, Sci. Transl. Med. 2013, 5, 179ra47–179ra47. Kalluri, R. The Biology And Function Of Exosomes In Cancer, J. Clin. Invest. 2016, 126, 1208. Veigas, B.; Fortunato, E.; Baptista, P. V. Field Effect Sensors For Nucleic Acid Detection: Recent Advances And Future Perspectives, Sensors. 2015, 15, 10380–10398. Zhang, A.; Zheng, G.; Lieber, C. M.; Nanowire Field-Effect Transistor Sensors. Nanowires. 2016, pp. 255-275. Issadore, D.; Min, C.; Liong, M.; Chung, J.; Weissleder, R.; Lee, H. Miniature Magnetic Resonance System For Point-Of-Care Diagnostics, Lab Chip. 2011, 11, 2282–2287. Schreiber, F. S.; Deramaudt, T. B.; Brunner, T. B.; Boretti, M. I.; Gooch, K. J.; Stoffers, D. A.; Bernhard, E. J.; Rustgi, A. K. Successful Growth And Characterization Of Mouse Pancreatic Ductal Cells: Functional Properties Of The Ki-RASG12V Oncogene, Gastroenterol. 2004, 127, 250–260. Kalluri, R. The Biology And Function Of Exosomes In Cancer, J. Clin. Invest. 2016, 126.4. Ko, J.; Yelleswarapu, V.; Singh, A.; Shah, N.; Issadore, D. Magnetic Nickel Iron Electroformed Trap (MagNET): A Master/replica Fabrication Strategy for Ultra-high Throughput (100 ML/hr) Immunomagnetic Sorting, Lab Chip. 2016, 16.16 3049-057. Santos, P.; Zanetti, J. S.; Ribeiro-Silva, A.; Beltrão, E. Beta 1 Integrin Predicts Survival in Breast Cancer: A Clinicopathological and Immunohistochemical Study, Diagn. Pathol. 2010, 7.1:104. Caccavari, F.; Valdembri, D.; Sandri, C.; Bussolino, F.; Serini, G. Integrin Signaling And Lung Cancer. Cell Adh Migr. 2010, 4: 124–129. Ethem, A., Introduction To Machine Learning. MIT press. 2014. Shao, H.; Chung, J.; Balaj, L.; Charest, A.; Bigner, D.D.; Carter, B.S.; Hochberg, F.H.; Breakefield, X.O.; Weissleder, R.; Lee, H. Protein Typing Of Circulating Microvesicles Allows Real-Time Monitoring Of Glioblastoma Therapy. Nat. Med. 2012, 18, 1835–1840. Salami, S.S.; Schmidt, F.; Laxman, B.; Regan, M.M.; Rickman, D.S.; Scherr, D.; Bueti, G.; Siddiqui, J.; Tomlins, S.A.; Wei, J.T.; Chinnaiyan, A.M., Combining Urinary Detection Of TMPRSS2:ERG And PCA3 With Serum PSA To Predict Diagnosis Of Prostate Cancer. Urol. Oncol. 2013, 31, 566-571. Muluneh, M.; Wu, S.; and Issadore, D. Track‐Etched Magnetic Micropores For Immunomagnetic Isolation Of Pathogens. Adv. Healthcare Mater. 2014, 3.7,1078-1085. Lane, R. E.; Korbie, D.; Anderson, W.; Vaidyanathan, R.; Trau, M. Analysis Of Exosome Purification Methods Using A Model Liposome System And Tunable-Resistive Pulse Sensing. Sci. Rep. 2015, 5, 7639. Aguirre, A.J.; Bardeesy, N.; Sinha, M.; Lopez, L.; Tuveson, D.A.; Horner, J.; Redston, M.S.; DePinho, R.A., Activated Kras And Ink4A/Arf Deficiency Cooperate To Produce Metastatic Pancreatic Ductal Adenocarcinoma. Genes Dev. 2003, 17, 3112–3126. Best, M.G.; Sol, N.; Kooi, I.; Tannous, J.; Westerman, B.A.; Rustenburg, F.; Schellen, P.; Verschueren, H.; Post, E.; Koster, J.; Ylstra, B., RNA-Seq of Tumor-Educated Platelets Enables Blood-Based Pan-Cancer, Multiclass, and Molecular Pathway Cancer Diagnostics. Cancer Cell. 2015, 28, 666–676.

27 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

49.

50.

51.

52.

53.

54.

55.

56. 57.

58.

59.

60.

61.

62.

63.

Nakatsura, T.; Senju, S.; Yamada, K.; Jotsuka, T.; Ogawa, M.; Nishimura, Y., Gene Cloning Of Immunogenic Antigens Overexpressed In Pancreatic Cancer. Biochem. Biophys. Res. Commun. 2001, 281, 936–944. Campagna, D.; Cope, L.; Lakkur, S.S.; Henderson, C.; Laheru, D.; LAcobuzio-Donahue, C.A., Gene Expression Profiles Associated With Advanced Pancreatic Cancer. Int. J. Clin. Exp. Pathol. 2008, 1, 32–43. Kolb, A.; Kleeff, J.; Arnold, N.; Giese, N.A.; Giese, T.; Korc, M; Friess, H., Expression And Differential Signaling Of Heregulins In Pancreatic Cancer Cells. Int. J. Cancer. 2007, 120, 514–523. Iacobuzio-Donahue, C.A.; Maitra, A.; Olsen, M.; Lowe, A.W.; Van Heek, N.T.; Rosty, C.; Walter, K.; Sato, N.; Parker, A.; Ashfaq, R.; Jaffee, E., Exploration Of Global Gene Expression Patterns In Pancreatic Adenocarcinoma Using cDNA Microarrays. Am. J. Pathol. 2003, 162, 1151–1162. Schek, N.; Hall, B. L.; Finn, O. J. Increased Glyceraldehyde-3-Phosphate Dehydrogenase Gene Expression In Human Pancreatic Adenocarcinoma. Cancer Res. 1988, 48, 6354– 6359. Shi, C.; Washington, M.K.; Chaturvedi, R.; Drosos, Y.; Revetta, F.L.; Weaver, C.J.; Buzhardt, E.; Yull, F.E.; Blackwell, T.S.; Sosa-Pineda, B.; Whitehead, R.H., Fibrogenesis In Pancreatic Cancer Is A Dynamic Process Regulated By Macrophage-Stellate Cell Interaction. Lab. Invest. 2014, 94, 409–21. Lane, R. E.; Korbie, D.; Anderson, W.; Vaidyanathan, R.; Trau, M. Analysis Of Exosome Purification Methods Using A Model Liposome System And Tunable-Resistive Pulse Sensing. Sci. Rep. 2015, 5, 7639. Ballehaninna, U. K.; Chamberlain, R. S. Serum CA 19-9 as a Biomarker for Pancreatic Cancer-A Comprehensive Review. Indian J. Surg. Oncol. 2011, 2, 88–100. Singh, S.; Tang, S. J.; Sreenarasimhaiah, J.; Lara, L. F.; Siddiqui, A. The Clinical Utility And Limitations Of Serum Carbohydrate Antigen (Ca19-9) As A Diagnostic Tool For Pancreatic Cancer And Cholangiocarcinoma. Dig. Dis. Sci. 2011, 56, 2491–2496. Earhart, C.M.; Hughes, C.E.; Gaster, R.S.; Ooi, C.C.; Wilson, R.J.; Zhou, L.Y.; Humke, E.W.; Xu, L.; Wong, D.J.; Willingham, S.B.; Schwartz, E.J. Isolation And Mutational Analysis Of Circulating Tumor Cells From Lung Cancer Patients With Magnetic Sifters And Biochips. Lab Chip. 2014, 14.1, 78-88. Helwa, I.; Cai, J.; Drewry, M.D.; Zimmerman, A.; Dinkins, M.B.; Khaled, M.L.; Seremwe, M.; Dismuke, W.M.; Bieberich, E.; Stamer, W.D.; Hamrick, M.W. A Comparative Study of Serum Exosome Isolation Using Differential Ultracentrifugation and Three Commercial Reagents. PLoS One. 2017, 12, e0170628. Crescitelli, R.; Lässer, C.; Szabo, T.G.; Kittel, A.; Eldh, M.; Dianzani, I.; Buzás, E.I.; Lötvall, J. Distinct RNA Profiles In Subpopulations Of Extracellular Vesicles: Apoptotic Bodies, Microvesicles And Exosomes. J Extracell Vesicles. 2013, 2: 10.3402. Liang, K.; Liu, F.; Fan, J.; Sun, D.; Liu, C.; Lyon, C.J.; Bernard, D.W.; Li, Y.; Yokoi, K.; Katz, M.H.; Koay, E.J. Nanoplasmonic Quantification Of Tumour-Derived Extracellular Vesicles In Plasma Microsamples For Diagnosis And Treatment Monitoring. Nat. Biomed. Eng. 2017, 1, 0021. Wan, Y.; Cheng, G.; Liu, X.; Hao, S.J.; Nisic, M.; Zhu, C.D.; Xia, Y.Q.; Li, W.Q.; Wang, Z.G.; Zhang, W.L.; Rice, S.J. Rapid Magnetic Isolation Of Extracellular Vesicles Via LipidBased Nanoprobes. Nat. Biomed. Eng. 2017, 1, 0058. Costa-Silva, B.; Aiello, N.M.; Ocean, A.J.; Singh, S.; Zhang, H.; kumar Thakur, B.; Becker, A.; Hoshino, A.; Mark, M.T.; Molina, H.; Xiang, J. Pancreatic Cancer Exosomes Initiate Pre-Metastatic Niche Formation In The Liver. Nat. Cell Biol. 2015, 17.6, 816.

28 of 35 ACS Paragon Plus Environment

Page 28 of 35

Page 29 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

64. 65.

66.

67.

68. 69.

70.

B. N. Hannafon; Wei-Qun, D. Intercellular Communication By Exosome-Derived Micrornas In Cancer. Int. J. Mol. Sci. 2013, 14.7,14240. Lötvall, J.; Hill, A.F.; Hochberg, F.; Buzás, E.I.; Di Vizio, D.; Gardiner, C.; Gho, Y.S.; Kurochkin, I.V.; Mathivanan, S. Minimal Experimental Requirements For Definition Of Extracellular Vesicles And Their Functions: A Position Statement From The International Society For Extracellular Vesicles. J Extracell Vesicles. 2014, 3, 26913. Gastpar, R.; Gehrmann, M.; Bausero, M.A.; Asea, A.; Gross, C.; Schroeder, J.A.Multhoff, G. Heat Shock Protein 70 Surface-Positive Tumor Exosomes Stimulate Migratory And Cytolytic Activity Of Natural Killer Cells. Cancer Res. 2005, 2, 5238-5247. Aatonen, M.T.; Öhman, T.; Nyman, T.A.; Laitinen, S.; Grönholm, M.; Siljander, P.R.M. ISolation And Characterization Of Platelet-Derived Extracellular Vesicles. J. Extracell. Vesicles. 2014, 3.1, 24692. Muller, L.; Hong, C.S.; Stolz, D.B.; Watkins, S.C.; Whiteside, T. Isolation Of BiologicallyActive Exosomes From Human Plasma. J. Immunol. Methods. 2014, 411, 55-65. Yuana, Y.; Koning, R.I.; Kuil, M.E.; Rensen, P.C.; Koster, A.J.; Bertina, R.M.; Osanto, S. Cryo-Electron Microscopy Of Extracellular Vesicles In Fresh Plasma. J. Extracell. Vesicles. 2013, 2, 21494. Ko, J.; Bhagwat, N.; Yee, S. S.; Redlinger, C;. Romeo, J.; O’Hara, M.; Raj, A.; Carpenter, E. L.; Stanger, B. Z.; Issadore, D. A Magnetic Micropore Chip For Rapid (< 1 Hour) Unbiased Circulating Tumor Cell Isolation And In-Situ RNA Analysis, Lab Chip, 2017, 17(18):3086-3096.

29 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Acknowledgements: JK helped conceive and perform all experiments in this study, as well as prepare the manuscript and figures. N.B., S.Y., N.O., T.B, N.A., A.S., N.M.A, L.M., M.O., C.R., and J.R. helped conceive and perform experiments on murine and clinical testing of our device. N.O, A.S. helped fabricated the ExoTENPO devices. B.S. and E.C. helped conceive experiments in this study, as well as prepare the manuscripts and figures. DI conceived and oversaw all aspects of this study, and prepared the manuscript. All authors reviewed the manuscript. Funding was provided by the Pennsylvania Department of Health Commonwealth Universal Research Enhancement Program, the National Institute of Health: 1R21CA182336-01A1 to DI and JK, R01-CA169123 to B.Z.S, F31-CA177163-01A1 to N.M.A., F32CA196120 to N.B., R01CA207643, Pancreatic Cancer Action Network Translational Research Award, Abramson Cancer Center Translational Centers of Excellence to E.C., and in part by the Penn Center for Molecular Studies in Digestive and Liver Diseases (P30-DK050306) from the National Institute of Diabetes and Digestive and Kidney Diseases. DI was supported by an American Cancer Society - CEOs Against Cancer - CA Division Research Scholar Grant, (RSG-15-227-01-CSM). DI is one of the founders of Chip Diagnostics and holds equity in the company.

30 of 35 ACS Paragon Plus Environment

Page 30 of 35

Page 31 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

Fig. 1. 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)


31 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Fig. 2. Pancreatic cancer diagnosis for mice using a machine learning based mRNA signature. a. The KPCY pancreatic cancer mouse model includes a fluorescent (YFP) lineage tracer to stage the mice. b. Pipeline of our machine learning based diagnosis. Training set is fed into LDA to generate LDA vectors, which are applied to the blinded test set for classification (c). After threshold detection, predicted labels are compared to the true labels to evaluate the pe rformance (d). e. Using the training set, we generated LDA scores (WT · x) for individual mice. LDA scores classify mice into the correct categories. f. N-1 cross validation was performed on the training set to evaluate the performance, where all mice were classified into the right groups. g. In an independent blinded test set all mice were classified correctly. h. The performance of our training set (AUC=1) was compared to a scrambled training set (AUC=0.53).


32 of 35 ACS Paragon Plus Environment

Page 32 of 35

Page 33 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

Fig. 3. Exosomes captured using a specific surface marker (EpCAM). a. The exosomal mRNA expression level difference between healthy plasma and spiked plasma (mouse cancer cell cultured media spiked into healthy mouse plasma) was measured from pan-based (CD9, CD81) isolation and EpCAM-based isolation, where EpCAM-based isolation showed higher difference. (N = 3 technical replicates, N = 3 biological replicates) b. EpCAM+ exosomes captured using ExoTENPO were measured using DLS for size distribution. The peak is at 169.9nm. c. Comparison of diagnostic performance between EpCAM capture and Pan capture for heathy versus tumor mice. d. Comparison of diagnostic performance between EpCAM capture and Pan capture for heathy versus PanIN mice. 


33 of 35 ACS Paragon Plus Environment

ACS Nano

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Fig. 4. Diagnosis of human pancreatic cancer using a machine learning based mRNA signature. a. Linear discriminant analysis (LDA) scores were algorithmically generated based on the RNA expression levels to separate healthy (green) from cancer patients (red). b. The capability to discriminate healthy versus cancer was first tested using N-1 cross-validation of the training set, and was able to correctly discriminate every patient, summarized in a confusion matrix. c. An independent, user blinded test set was created and was able to discriminate all patients with cancer vs. healthy donors, summarized in a confusion matrix. d. ROC curves were generated based on varying numbers of exosomal mRNA. The inset shows that AUC is a function of the number of exosomal RNA biomarkers. The color coding matches the points in the inset to the ROC curves.

34 of 35 ACS Paragon Plus Environment

Page 34 of 35

Page 35 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Nano

For Table of Contents Only:

35 of 35 ACS Paragon Plus Environment