Ultrafast Single-Cell Level Enzymatic Tumor Profiling - Analytical

Oct 26, 2018 - Several fluidic platforms were developed previously for single-cell analysis, while the trade-off between screening speed and comprehen...
0 downloads 0 Views 2MB Size
Download PDF
Subscriber access provided by UNIV OF LOUISIANA

Article

Ultrafast Single-Cell Level Enzymatic Tumor Profiling Ee Xien Ng, Guoyun Sun, Shih-Chung Wei, Miles A. Miller, Ramanuj Dasgupta, Paula Yeng Po Lam, and Chia-Hung Chen Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.8b02576 • Publication Date (Web): 26 Oct 2018 Downloaded from http://pubs.acs.org on October 28, 2018

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 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 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.

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 9 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

Analytical Chemistry

Ee Xien Ng1†, Guoyun Sun1†, Shih-Chung Wei1,2, Miles A. Miller3, DasGupta Ramanuj4, Paula Yeng Po Lam5 and Chia-Hung Chen1,2,6* 1. Department of Biomedical Engineering, National University of Singapore, 4 Engineering Drive 3, 04-08, Singapore. 2. Biomedical Institute for Global Health Research & Technology, Singapore. 3. Center for Systems Biology, Massachusetts General Hospital, Boston, USA. 4. Genome Institute of Singapore, Agency for Science, Technology and Research (A*STAR), Singapore. 5. National Cancer Centre Singapore, 11 Hospital Drive, Singapore. 6. Singapore Institute for Neurotechnology, 28 Medical Dr. 05-COR, Singapore. †Co-first author; *Corresponding author; E-mail: [email protected] ABSTRACT: In the context of tumor analysis, the implementation of precision medicine requires on-time clinical measurements, which requires rapid large-scale single-cell screening that obtains cell population distributions and functions in tumors to determine disease progression for therapeutics. In this study, a high-throughput screening (HTS) platform integrating optical fluorescence detectors and a computational method was developed as a droplet-based microfluidic flow cytometer (Droplet-µFC) to comprehensively analyze multiple proteolytic activities of a patient-derived tumor (with ~0.5-2M cells) at single-cell resolution within two hours. The data-driven analytical method identified distinct cell types and status through protease profiling with high precision. Multiple protease activities of single cells harvested from a tumor were thus determined with a throughput of ~100 cells per second. This platform was used to screen protease activities of a wide range of cell types, forming a library. With the development of advanced computational clustering and cell mapping, rapid quantitative tumor profiling with a comprehensive description of cell population distributions and functions could be obtained for clinical treatments.

KEYWORDS: Rapid Tumor Profiling, Clinical Enzyme Analysis, Single Cell Assay, Microfluidics, Continuous Flow Screening Introduction Precision medicine refers to matching the right drugs to the right patients at the right time. To reach this goal, a fluidic platform with great specificity and single-cell resolution is essential in quantitative clinical biology. Gene analysis of single cells is currently the main method used for the characterization of patient cells in tumor biopsies. For example, single-cell multiple RNA sequences are analyzed with high precision using platforms such as High-Throughput Single-Cell Labeling (Hi-SCL)1 and Drop-seq2. However, genome-based detection is limited by an incomplete understanding of the correlation between biological phenotypes and disease progression. Moreover, single-cell genome analysis usually requires time-consuming polymerase chain reactions (PCRs) and barcoding/decoding processes, which limit rapid personal tumor identification for on-time therapeutics. Indeed, highthroughput phenotypic functional characterization of single cells in patient tumor biopsies is necessary for determining the statistics of cell populations and functional characterization, offering comprehensive information of patient disease status. Several fluidic platforms have been investigated to characterize single-cell phenotypes via functional assays, including antibody binding3, western blots4, mechanical tests5 and enzyme assays6. For example, high-throughput screening of single-cell surface biomarkers or sizes are performed via flow cytometer with a throughput of ~1000 cells per second, obtaining cell population distributions in a tumor; however, surface biomarkers need to be identified by empirical tests before they are assayed, resulting the inability to characterize a tumor constituted of unknown cells. Moreover, the challenges of using surface biomarkers to indicate a cell’s real-time functions and activities, which correlate to disease progress, remain7. Continuous-flow mechanical screening of single cells via mechanical flow cytometry8 was developed to determine the

specific disease (e.g. leukaemia, malignant pleural effusions9) status but was limited by having few read-out parameters to distinguish a wide range of disease states (e.g. glioblastoma) with high precision. Single-cell force cytometry10 was developed recently to effectively determine cell’s surface elastomeric properties for physiological and pathological behavior analysis11. Microwell platforms were used to isolate single cells for comprehensive phenotype characterization using western blotting or mass spectrometry; the throughput is ~100-1000 cells per experimental run so the difficulties associated with rapidly screening large numbers of cells for tumor profiling are expected12. Droplet microfluidics is a promising platform, opening new possibilities in high-throughput single-cell measurements13. The ability to generate and monitor droplets carrying small numbers of single cells allows us to manipulate primary patient samples for quantitative biology and diagnoses. Droplet clinical enzyme assays (multiple matrix metalloproteinase (MMP) and A disintegrin and metalloproteinase (ADAMs) activity measurements) were investigated to characterize single-cell functions of migration and metastasis14. Metalloproteinases are zinc-dependent endopeptidases that are secreted as inactive proenzymes or zymogens and are activated by the cleavage of the propeptide domain. The activation of these pro-enzymes is one of the critical steps that lead to the breakdown of the extracellular matrix (ECM) and plays an important role in metastasis by facilitating tumor cell migration and spread and angiogenesis15. Protease activity measurements allow us to identify disease situations correlated to tumor microenvironments, which consist of a mixture of tumor cells and their neighboring non-cancerous stromal cells, including endothelial cells, immune cells and fibroblast cells, indicating different clinical processes of invasion. The deregulation of MMPs/ADAMs is especially important for the determination of glioblastoma (GBM) invasion, the most lethal of all brain

ACS Paragon Plus Environment

Analytical Chemistry 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

numbers (~0.5M) of single-cell protease profiles within 2 hours is required. First, although a large number of monodispersed droplets can be effectively produced by using droplet-based microfluidic methods (~4000 droplets per second), these droplets are identical and have the same chemical environments, thus limiting the performance of the multiplexed assays required for MMPs/ADAMs. Second, single-cell protease activities were determined by the kinetics of reactions between MMPs/ADAMS and fluorescence resonance energy transfer (FRET) substrates, whose measurement requires the continuous monitoring of each cell for a certain time period (~40-120 minutes) with low read-out efficiency (~100 cells per experimental run). The number of cells that can be screened is therefore very limited. In this study, a continuous flow droplet microfluidic platform integrated with an automatic optical sensor and a computational analysis method, known as Proteolytic Activity Matrix Analysis (PrAMA)23 (Supplementary-2), was developed as a functional flow cytometer (Droplet-µFC) to perform real-time multiplexed proteolytic activity measurements on single cells for fast tumor screening. The throughput of Droplet-µFC is ~100 cells per second. Compared with conventional fluidic platforms, Droplet-µFC offers a comprehensive tumor analysis by identifying cell population distributions and their functions via multiple protease enzyme activities, while accomplishing the screening within 2 hours Figure 1. Continuous flow droplet microfluidic platform for real-time single tumor (Figure 1A). Patient-derived GBM xenograft cell screening. (A) Tumor microenvironments consist of both cancerous cells and tumors were dissociated by mild trypsinization normal cells such as endothelial cells, fibroblasts and immune cells. (B) Single cells and trituration to generate single-cell suspensions. were dissociated from tumor samples and encapsulated into droplets for protease After that, FRET substrates and lysis solution were activity screening. (C) The process of cell encapsulation, cell lysis, incubation and mixed with suspended cells in the chip prior to detection within a droplet is indicated. (D) Excitation light passed through a droplet encapsulation for single-cell protease multiband band-pass filter to stimulate four different substrates simultaneously in measurements (Figure 1B). The FRET substrates the droplets to allow efficient read-out of multiple fluorescent signals by were modified to accommodate different photomultiplier tube (PMT) sensors. A data acquisition (DAQ) system was used to fluorescent pairs with distinct excitation and convert the analog voltage signals from PMTs into digital signals for real time emission wavelengths to simultaneously PrAMA. determine multiple enzyme activities, i.e., to obtain signals of multiple important proteases (MMPs and malignancies16. In MMPs part, MMP-2, MMP-3 and MMP-9 ADAMs) from single-cell encapsulated droplets (Figure 1C). were selected. MMP-2 and MMP-9 are the only two These fluorescent signals were rapidly detected by four metalloproteinases capable of degrading type IV collage, the photomultiplier tubes, then the voltage signals were converted main component of the basement membrane correlating to to digital signals for real-time computational analysis. metastasis directly17, 18. MMP-3 is an important Accordingly, six MMP/ADAM activities were inferred in metalloproteinase which cleaves cell adhesion molecule Esingle cells, and a cell population distribution was obtained cadherin19 and facilitates epithelial-mesenchymal transition based on protease activities in a tumor (Figure 1D). Different (EMT) among cancer cells20. In ADAMs part, ADAM-8, types of cells showed their own specific protease fingerprints, ADAM-10 and ADAM-17 were selected. ADAM-8 is an forming distinct clusters in data. Measuring a number of essential enzyme to be involved in the facilitation of migration different cell lines in different states allowed a comprehensive of cancer cells21. ADAM-10 and ADAM-17 are reported to library of tumor microenvironments and cancer characterization activate ligands such as epidermal growth factor (EGF) and for cell mapping to be generated using a computational method transforming growth factor alpha (TGF-α) for cancer for unknown tumor microenvironment profiling and cancer progression. The inhibition of these ADAMs in clinical trials characterization. We used this platform to focus on analyzing have shown positive effect complementing existing therapies to patient-derived GBMs. A number of different cell lines supress tumor growth22 (Supplementary-1). correlated to brain tumors (GBM12 and GBM22) and the cells In contrast to the abovementioned advantages of droplet surrounding tumors such as endothelial cells, astrocytes, clinical enzyme assays, two main challenges remain in the use fibroblasts and immune cells were identified by screening for of this droplet platform as a diagnostic tool. Screening large

ACS Paragon Plus Environment

Page 2 of 9

Page 3 of 9 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

Analytical Chemistry their distinct protease MMP/ADAM activity fingerprints. The library of the protease activities of cells was established to indicate unknown cells from a tumor with high precision via a computational method (k-nearest neighbors (k-NN) algorithm). Primary patient-derived tumor cells and drug- treated cells were analyzed for comparisons. Tumor screening (primary patient cells GBM12 & GBM22) was performed within ~2 hours to profile every single cell based on their phenotypic protease measurements and thus could be used to strategize a combinational use of drugs to address the case of an individual patient’s tumor. Materials and Methods Cell Preparation: To develop a library of different cell types’ protease activities as a data base to represent brain tumor microenvironments, different cell types were examined, including cancer cells, fibroblasts, astrocytes, endothelial cells and immune cells. Normal human astrocytes (NHAs) were purchased from Lonza Bioscience (Basel, Switzerland), whereas normal human dermal fibroblasts (NHDFs) and human umbilical vein endothelial cells (HUVEC) were purchased from Lonza group Ltd. (Walkersville, MD) (Supplementary-3). Primary GBM xenograft cell lines, GBM12 and GBM22, were purchased from the Mayo Clinic (Rochester, MN) and maintained as subcutaneous xenografts as previously described24. Tumor dissociation and preparation of patientderived primary GBM cell culture were included in Supplementary-4. Continuous Flow Microfluidic Process: For droplet encapsulation enabling single-cell multiplexed assays, a chemical sensor solution was prepared by mixing four modified FRET substrates (Supplementary-5)23, 25 with 0.5% Triton X and an EDTA-free protease inhibitor cocktail (Roche, USA). The details of optical setting are included in Supplementary-6. Fluorescence bleed-through was negligible, and the four fluorescent signals were clearly identified from their respective PMT modules (Supplementary-7). In the cell encapsulation process (Supplementary-8), the chemical sensors and lysis solution were mixed with suspended cells in the microfluidic chip prior to droplet encapsulation 25. For continuous flow screening, the protease reaction kinetics obtained by measuring the reaction rate were converted to endpoint measurements. Protease-substrate reactions were initiated immediately once single cells were encapsulated inside the droplets. The order of the droplets with encapsulated cells was determined to ensure fixed reaction time intervals to pass through the optical sensor to record fluorescence signal measurements precisely at the endpoint. Single-cell droplets with FRET substrates were collected from the outlet channel of the single-cell encapsulation chip into a long polyethylene (PE) tube with an inner diameter of 0.38 mm (Scientific Commodities, USA) to maintain the sequence of droplets (Supplementary-9). These droplets with cell encapsulations were directly stored in a tube once produced to maintain the temporal order of the droplets in the long tube. The starting reaction time point was therefore determined to regulate the durations of reactions in the droplets. Moreover, the droplets were incubated for ~1 hour to reach a stable reaction regime for measurements. These droplets were then injected into a screening microfluidic channel of 25 µm width and 40 µm height for continuous screening for another 1 hour. Spacer oil was introduced from the side of the straight channel to space

Figure 2. Real-time PrAMA and throughput characterization. (A) Fluorescent signals from single-cell droplets were detected by a set of photomultiplier tubes (PMTs), and signals were converted into digital signals for real-time PrAMA calculations. A “detect and match” approach was used to minimize the calculation time of PrAMA for real-time analyses. (B, C) Throughput of screening against spacer oil flow rate (B) and droplet flow rate (C).

droplets and maintain the cell ordering during screening. Because the order in which droplets with encapsulated cells were generated was maintained in a long tube, the last time point to measure was fixed at 1 hour after encapsulation to calibrate the increasing fluorescence slopes and accurately convert to protease activities. Identification method: We used the k-NN algorithm to identify the species of unknown single tumor cells and thus study the microenvironment of a tumor. The basic principle of the k-NN algorithm is to find the k labeled (known) samples from the library that are nearest to the unknown sample, where k is a user-defined constant. The species of that unknown sample are defined by the identification of these k samples. In this study, the library was built from six protease activities of six cell lines, including tumor cell lines (U87MG and U87MG.wtEGFR) and normal cell lines (NHA, NHDF, HUVEC and WBC). For each cell line, the protease activities from 10,000 cells were analyzed using PrAMA, and these analyses were used to form a data base library. Six protease activities were determined to identify each individual cell. This library contained 60,000 cells. To determine the species of unknown cells, we determined six protease activities of each unknown cell by PrAMA. Euclidean distances between the unknown cell and cell lines from the library were measured to address the k nearest cells. The k value used here was 1,000, which was 10% of the total number of each cell species in the library. The species of the unknown cells were determined by the member of these 1,000 cells whose contribution was

ACS Paragon Plus Environment

Analytical Chemistry 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

weighted the highest. The weight was defined as the inverse of the distance.

The correlation between droplet flow rate and screening outcomes was characterized in Figure 2C. Our results suggested that a higher droplet flow rate would shorten the time period between droplets. In this study, the throughput of droplet screening using Droplet-μFC was optimized at ~1000 Hz. For tumor analysis, tumors that were 2-4 mm in diameter contained ~0.5 million cells26. Tumors were dissociated, and the cells were suspended in a PBS solution (100 µL) with 0.5% bovine serum albumin (BSA). The cell concentration was set to 5 million cells per mL. For rapid tumor screening within an hour, the sample flow rate was 3 µL/min, and the oil flow rate was 14 µL/min, resulting in a droplet throughput of ~1000 Hz.

Results and Discussion High-throughput real-time analysis: To achieve high-speed screening, the kinetic read-out of protease reactions with their corresponding time intervals were converted to end time-point measurements (Supplementary-10). This process required that the ordering of droplets encapsulating cells/sensors was maintained in a long tube. A system with high-speed fluorescence analysis of single cells and a high rate of data transmission between modules and software was developed for continuous flow-cell analysis. Analog voltage signals from the PMTs were converted to digital form through a data acquisition (DAQ) system (National Instruments, USA) at a sampling rate of 12,500 samples/s (Figure 2A). PrAMA operates based on a linear matrix algebra calculation with a bootstrapping scheme to generate multiple iterations for each experimental calculation to achieve high accuracy (Supplementary-11). MATLAB was used to record digital signals (Supplementary-12) and perform PrAMA in real time (Supplementary-13) to infer specific activities of multiple metalloproteinases from their catalytic reactions with multiple FRET substrates. Real-time PrAMA was developed for the rapid detection and analysis of droplets. The flow rates of the aqueous droplet phase Figure 3. Cell library construction. (A) Proteolytic activity profile of various cell lines and cell types from and oil phase of the PrAMA inference. (B) Cell population distribution in terms of proteolytic activity in a 3D scatter plot. (C) system were Classification accuracy for 6 cell types through k-nearest neighbors algorithm. (D) Classification probability characterized to for every cell, where every row represents a single cell’s probability of being classified as one of the six cell examine the overall types. (E) Hierarchical unsupervised clustering of all cells based on their proteolytic activity profile. throughput for tumor screening (Figure 2B). The droplet flow rate was kept constant Cell library construction: Different cell lines were at 1 µL/min, while the oil flow rate was manipulated from 12 characterized to form a data base that allowed the identification µL/min to 20 µL/min with a 2 µL/min increments. A higher of different types of cells in a tumor via their protease profiles. spacer oil flow rate resulted in a longer time interval between For each cell type, 104 cells were screened for protease activity droplets, which lowers the throughput of screening. Moreover, profiling. PrAMA inference was performed for every single cell increasing the oil flow rate increased the droplet movement from all 6 cell types (Figure 3A, Supplementary-14), and the speed, resulting in a short time period for droplets to remain average rates of reaction of all cells from each cell type against within the optical sensing zone and thus affecting the detection sensitivity. the 4 substrates were calculated (Supplementary-15). The raw substrate cleavage profiles of various cell types were

ACS Paragon Plus Environment

Page 4 of 9

Page 5 of 9 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

Analytical Chemistry determined to distinguish the difference in proteolytic activities among different cell types (Supplementary-16). It was observed that different cell types showed distinguished proteolyticsubstrate cleavage profile. To be able to identify cancerassociated cells using proteases to represent tumor microenvironments, we screened the protease activities of endothelial cells, astrocytes, fibroblasts and immune cells through PrAMA. Different heterogeneities among cell lines were also observed (Supplementary-17). ELISA was performed to validate the presence of protease MMPs and ADAMs (Supplementary-18). To further prove the presence of active proteases, additional inhibitor experiment was also performed (Supplementary-19). The unique protease activity profiles that corresponded to specific cell types were used to rapidly indicate different cell constitutions in a tumor. Activities of three relevant MMPs and ADAMs were used to plot a 3D scatter diagram of all cells (Figure 3B). In this work, MMP-2, ADAM-8 and ADAM-17 were selected as they were essential secretions in brain tumor progression27. ADAM-8 is especially important in identifying immune cells as it was part of a cluster of differentiation (CD) for leukocytes28. In this diagram, immune cells (white blood cells) were characterized distinctly from the other adherent cell types. Astrocytes and fibroblasts were characterized to have high protease activities, while endothelial cells and cancer cell lines had lower activities. Nevertheless, U87MG.wtEGFR, which is expected to be more aggressive, has potentially higher MMP-2, ADAM-8 and ADAM-17 activities than U87MG cell lines, indicating that aggressiveness of cancer might be associated with the levels of protease activities. A leave-one-out cross-validation method was subsequently used to measure the accuracy of our cell line library. The species of each cell in the library was identified using the k-NN algorithm on all data in the library except that of the target species. This process was performed for all cells in the library. Our results suggested >70% classification accuracy for all cell types (Figure 3C). Endothelial cells, NHDFs and white blood cells were classified with the high accuracies of 100.00, 95.63 and 100.00% respectively; NHA, U87MG and U87MG.wtEGFR cells had lower classification accuracies, 89.03, 84.10 and 71.04%, respectively. We examined the probability of classifying a given cell into the 6 cell types (Figure 3D), and U87MG and U87MG.wtEGFR had a certain degree of overlap in terms of their classification as they have similar proteolytic activity profiles. In addition, another validation test of cell line was performed. 8,000 cells for each cell line (48,000 cells in total) were picked randomly from 10,000 cells for each cell line. These 48,000 cells formed a new library. The rest 2,000 cells for each cell line (12,000 in total) were used to examine the library. Accordingly, the k value used for this test is decreased to 800, which is 10% of cell number for each cell line. The similar result was showed comparing with fig. 3C. (Supplementary-20). To further validate the results, we performed a 1:1 mix of NHDF and U87MG to investigate the performance of the clustering algorithm, the results were included in Supplementary-21. In addition, hierarchical unsupervised clustering was applied to analyze the MMP-2, MMP-3, MMP-9, ADAM-8, ADAM10 and ADAM-17 proteolytic activities of all our cell lines (Figure 3E). In this figure, high activities are indicated in red, while low activities are indicated in blue; each row represents one cell, and each column shows the PrAMA calculated activity from the 6 proteases. We mainly distinguish 7 groups of cell

Figure 4. Primary cell analysis. (A) Cell morphologies of GBM12 and GBM22 under bright field microscopy. (B) Proteolytic activity profile of GBM12 and GBM22 from PrAMA inference. (C) Cell population distribution in terms of proteolytic activity in a 3D scatter plot with MMP-2, ADAM-8 and ADAM-17 activities as axes.

types. Immune cells were clustered the furthest away from all other adherent cell types because of their distinctly different proteolytic activity profile. For adherent cell types, HUVEC, NHDF, NHA and U87MG cells were clustered with high accuracy more closely, in a branch. Some U87MG cells were clustered together with U87MG.wtEGFR cells since both cell types have similar proteolytic activity profiles. A subpopulation of NHDFs with high ADAM10 activity was clustered respectively, because of cell heterogeneity of NHDF. Primary cell analysis: In this study, two sets of primary cells were used: GBM12 and GBM22. Bright field images of cell morphology were recorded by using an inverted microscope (Leica DMi8). Both spindle-shaped cells and astrocytic cells were observed, showing high cell heterogeneity among the primary cells (Figure 4A). The protease activities of primary cell lines were determined via PrAMA using microfluidics. The raw substrate cleavage data was determined (Supplementary22). The PrAMA result suggested that GBM12 exhibited significantly higher MMP-2, ADAM-8 and ADAM-17 activities than GBM22 and lower MMP-3 and MMP-9 activities (Figure 4B). As an aggressive cell type, GBM12 expressed high levels of EGFR29, MMP-2, ADAM-8 and ADAM-17, forming a distinguished cluster in Figure 4C. Plotting the twodimensional projections of this three-dimensional diagram indicated that GBM12 showed much higher ADAM-17 activity than GBM22. GBM22 exhibited significant heterogeneity in ADAM-8 and MMP-2 activities between cells.

ACS Paragon Plus Environment

Analytical Chemistry 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

expression and increase invasiveness of PTEN mutant glioma cells30. Functional cooperation of MMP-2 with PAK4 resulted in enhanced αvβ3/EGFRmediated anoikis resistance and glioma invasion31. Mostly recently, Ou Y. and colleagues reported a feedback loop between EGFR and MMP-2 that is important for glioma cell motility32. Consistent with the earlier U87MG glioma isogenic cells (Figure 3A), the primary patient-derived glioma cells also showed that the EGFRoverexpressing GBM12 exhibited higher levels of MMP-2 compared to EGFR-low/null GBM22, before or post treatment with erlotinib (Figure 5A and B). Taken together, the slight decrease in the proteases activities post erlotinib treatment also support the very modest clinical efficacy brain tumor patients treated with erlotinib33. We have demonstrated that our Droplet-µFC system provided Figure 5. Drug effect. (A) Average proteolytic activities of GBM12 and GBM22 treated with essential insights to the cell erlotinib and without erlotinib. (B) Normalized average proteolytic activity of erlotinib treated functional type constituents within cells with untreated cells for GBM12 and GBM22. (C) Single-cell proteolytic activity profiles the tumor mass. Unlike the current of GBM12 and GBM22 treated with and without erlotinib. (D) Cell population distributions of immunohistochemistry methods GBM12 and GBM22 were plotted in terms of proteolytic activity in a 3D scatter plot with MMPused by pathologists for diagnosis 2, ADAM-8 and ADAM-17 activities as axes. Single-cell analysis revealed the presence of two and/or assessment of treatment population of GBM12 cells in response to the drug treatment (~27.5% of the GBM12 population efficacy, our system analysed elevated their ADAM-8 protease activity). multiple clinical enzymes from individual cells at an ultra-fast speed, thus, overcoming the Drug effect: The drug response of the primary cell lines known biasness associated with bulk tissue sampling, and offers GBM12 and GBM22 was analyzed using single-cell protease important information of tumor heterogeneity for precision activities via Droplet-µFC to demonstrate potential applications medicine (Figure 5D). in drug selections based on individual tumor profiling for precision medicine. Erlotinib, which suppresses EGFR tyrosine Tumor analysis: Mouse xenograft tumors of primary GBM12 kinase, was used as a cancer drug to target EGFR. GBM12 and and GBM22 cells were harvested and dissociated to generate GBM22 cells were seeded on a 6 well plate. To test each type suspended single cells for Droplet-μFC protease analysis. The of cell, a control group and a drug test group with the same tumors were dissociated by using the GentleMACS tumor number of cells were prepared in standard 10% FBS containing dissociation protocol mentioned previously (Materials and DMEM. After one day of culturing, erlotinib was added at a Methods section). The average diameter of tumors was concentration 10 µM for 24 hours. The raw substrate cleavage approximately 3.5 mm, and after the dissociation, we observed data was included to determine activities of MMPs and ADAMs different cell types with a wide range of morphologies (Figure (Supplementary-23). Heat maps of protease activities were 6A). In the case of a GBM12 tumor, cancer cells and stromal generated using PrAMA and are shown in Fig. 5C. The results cells expressed different phenotypic characteristics of proteases show that there is a slight reduction in MMP-2, ADAM-10 and when interacting with other cells within a tumor ADAM-17, post erlotinib treatment in both GBM12 and microenvironment. Droplet-μFC with PrAMA showed that GBM22 primary culture (Figure 5A and B). ADAM-10 and cells harvested from a GBM12 tumor had higher MMP-2, ADAM-17 were among the metalloproteinases that have been MMP-3 and MMP-9 activities than a GBM12 primary cell line, shown to be important for the processing of EGFR ligand while ADAM-10 activity was not observed in cells from a precursors. The fact that GBM22 is also modestly responding GBM12 tumor (Supplementary-24). to erlotinib suggesting that these cells may contain low levels Single-cell protease activities from a GBM22 tumor were of endogenous EGFR or EGF-domain containing receptors. In also screened using Droplet-μFC. A three-dimensional plot HUVEC Cells, ADAM-10 and ADAM-17 are responsible for using MMP-2, ADAM-8 and ADAM-17 activities as the three the shedding of surface proteoglycan, syndecan-1. Targeted axes (Figure 6C and 6D) was constructed. Protease activities of knockdown of syndecan-1 is recently shown to inhibit glioma cells from a tumor and primary cells were similar, suggesting cell proliferation and invasion via a c-src/FAK-associated that the phenotypic changes of cells in a tumor signaling pathway. Irradiation has been shown to induce Srcmicroenvironment could be affected by the interaction between dependent EGFR activation leading to enhanced MMP-2

ACS Paragon Plus Environment

Page 6 of 9

Page 7 of 9 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

Analytical Chemistry cancer cells and stromal cells32 and were dependent on the initial activity levels of the cells. This observation was consistent with previous reports in tumor microenvironment analysis34. The phenotypic resemblance of the tumor cells to the previously characterized tumor microenvironment was investigated by using the k-NN algorithm to classify the tumor cells. This analysis showed that according to single-cell protease screening in a GBM12 tumor, 86.98% of cells phenotypically resembled the U87MG.wtEGFR cell type, 9.58% to the U87MG cell type, and 3.44% to the NHA cell type (Fig. 6C). Moreover, examining the probability of classification found that most cells are somewhat similar to NHA and U87MG.wtEGFR cells. In a GBM22 tumor, 72.41% of cells phenotypically resembled the U87MG cell type, 13.23% to the HUVEC cell type, 6.38% to the NHA cell type, and 7.93% Figure 6. Tumor analysis. (A) Tumor dissociation into single suspended cells. (B) Proteolytic to the NHDF cell type (Figure activity profile of GBM12 and GBM22 xenograft tumors from PrAMA inference. Cell population 6D). Most of the cells from a distribution of (C) GBM12 and (D) GBM22 were plotted in terms of proteolytic activity in a 3D GBM22 tumor were classified scatter plot with MMP-2, ADAM-8 and ADAM-17 activities as axes. The k-nearest neighbor’s with high probability to algorithm was used to classify GBM12 and GBM22 tumor cells into the 6 characterized cancer and U87MG GBM22 primary cells. normal cell lines with a classification probability for every cell, where every row represents a single The results were in cell’s probability of being classified as one of the six cell types. concordance with our characterized results, where the GBM12 tumor cells, which are excitation and emission bandwidths of fluorophores. New more aggressive, were mainly classified as the more aggressive design of fluorophores could potentially increase the cell type U87MG.wtEGFR and the less aggressive, GBM22, multiplexing capability of this platform. tumor cells were mainly classified as the less aggressive cell type U87MG. The protease activities rapidly indicated the Conclusions presence of stroma-like cells in the tumors. With the novel In summary, a continuous flow microfluidic device (Dropletcombination of Droplet-μFC, PrAMA and the k-NN µFC) integrated with a fast computational method was classification technique, a cell population profile (with ~0.5M developed for high-throughput single-cell multiple proteolytic cells) could be rapidly determined within 2 hours to optimize activity analysis that rapidly profiles patient-derived tumors. individual patients’ therapeutic strategies. While the presented Encapsulating single cells with lysis buffer and multicolor platform has demonstrated the capability to rapidly characterize FRET sensors in droplets that provide physical and chemical cell populations from a tumor based on its phenotype, we isolation allowed the proteolytic activities of multiple proteases acknowledge limitations of the current platform for clinical from single cells to be quantified by measuring the fluorescence translation purpose. There are two main limitations of presented generated from proteases cleaving these FRET sensors. The platform: 1. Cell encapsulation rate, and 2. Capability of proteolytic activities of different cell lines of tumor stromal multiplexing. Indeed, cell encapsulation rate is dependent on components (such as fibroblasts, endothelial cells and immune Poisson distribution. The efficiency of single cell encapsulation cells) and cancer cell lines of different aggressiveness (such as in droplets is ~1-10%. Therefore by screening ~1000 U87MG, U87MG.wtEGFR) were characterized and distinctly droplets/sec, the reading out efficiency is ~100 cells/sec. There clustered to form a data base library. This library allowed are several methods investigated before to increase cell different cell types in a tumor to be determined by using a encapsulation rate. For example, by uploading high density computational method of cell mapping, the k-NN algorithm, cells in an inertial microfluidic device, the cells could order with a throughput of ~100 cells per second. Accordingly, a along the channel for droplet single cell encapsulations with tumor with cancer cells and tumor-associated cells (~0.5 million high efficiency35. Another limitation lies in the multiplexing cells) could be fully characterized within 2 hours for on-time capability of the platform. At present, it is limited to the four tumor biopsy analysis. It is worth noting that current single-cell channel detectors (maximum excitation wavelength: 400 nm, identification using gene sequence requires several days for the 490 nm, 546 nm, and 635 nm) due to the limitations of the completion of PCRs, meaning that the enzyme phenotype

ACS Paragon Plus Environment

Analytical Chemistry 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

identification method reduced the assay time by a factor of ~100. The capability of using Droplet-µFC to rapidly obtain tumor status information allows the elucidation of single-cell activities in a tumor to be obtained within 2 hours, which allows the ontime strategizing of therapeutics for precision medicine.

The authors acknowledge Luke Pyungse Lee for the discussion, Mark Schroeder and Jann Sarkaria (Mayo Clinic, Rochester, Minnesota) for providing GBM samples.

The experimental details are included in Supporting Information.

(1) Rotem, A.; Ram, O.; Shoresh, N.; Sperling, R.; Schnall-Levin, M.; Zhang, H.; Basu, A.; Bernstein, B.; Weitz, D. PLOS ONE 2015, 10, e0116328. (2) Macosko, E.; Basu, A.; Satija, R.; Nemesh, J.; Shekhar, K.; Goldman, M.; Tirosh, I.; Bialas, A.; Kamitaki, N.; Martersteck, E.; Trombetta, J.; Weitz, D.; Sanes, J.; Shalek, A.; Regev, A.; McCarroll, S. Cell 2015, 161, 1202-1214. (3) Abbaspourrad, A.; Zhang, H.; Tao, Y.; Cui, N.; Asahara, H.; Zhou, Y.; Yue, D.; Koehler, S.; Ung, L.; Heyman, J.; Ren, Y.; Ziblat, R.; Chong, S.; Weitz, D. Sci. Rep. 2015, 5, 12756. (4) Zlotorynski, E. Nat. Rev. Mol. Cell Biol. 2014, 15, 429-429. (5) Di Carlo, D.; Lee, L. P. Anal. Chem. 2006, 78, 7918-7925. (6) Wu, L.; Claas, A.; Sarkar, A.; Lauffenburger, D. A.; Han, J. Integr. Biol. 2015, 7, 513-524. (7) Luk, H. Y.; McFarlin, B. K.; Vingren, J. L. Methods 2017, 112, 175-181. (8) Otto, O.; Rosendahl, P.; Mietke, A.; Golfier, S.; Herold, C.; Klaue, D.; Girardo, S.; Pagliara, S.; Ekpenyong, A.; Jacobi, A.; Wobus, M.; Töpfner, N.; Keyser, U. F.; Mansfeld, J.; Fischer-Friedrich, E.; Guck, J. Nat. Methods 2015, 12, 199-202. (9) Gossett, D. R.; Tse, H. T.; Lee, S. A.; Ying, Y.; Lindgren, A. G.; Yang, O. O.; Rao, J.; Clark, A. T.; Di Carlo, D. Proc. Natl. Acad. Sci. 2012, 109, 7630-7635. (10) Pushkarsky, I.; Tseng, P.; Black, D.; France, B.; Warfe, L.; Koziol-White, C. K.; Jester, W. F., Jr; Trinh, R. K.; Lin, J.; Scumpia, P. O.; Morrison, S. L.; Panettieri, R. A., Jr; Damoiseaux, R.; Di Carlo, D. Nat. Biomed. Eng. 2018, 2, 124-137. (11) Resto-Irizarry, A. M.; Fu, J. Nat. Biomed. Eng. 2018, 2, 60-61.

Page 8 of 9

(12) Islam, S.; Zeisel, A.; Joost, S.; La Manno, G.; Zajac, P.; Kasper, M.; Lönnerberg, P.; Linnarsson, S. Nat. Methods 2013, 11, 163-166. (13) Rane, T. D.; Zec, H. C.; Wang, T.-H. Anal. Chem. 2015, 87, 19501956. (14) Radisky, E. S.; Radisky, D. C. J. Mammary Gland Biol. Neoplasia 2010, 15, 201-212. (15) Deryugina, E. I.; Quigley, J. P. Matrix Biol. 2015, 44-46, 94-112. (16) Nakada, M.; Okada, Y.; Yamashita, J. Front. Biosci. 2003, 8, e261-269. (17) Moulik, L. S.; Pal, S.; Biswas, J.; Chatterjee, A. J. Tumor, 2014, 2(2), 87-98. (18) Moulik, S.; Sen, T.; Dutta, A.; Banerji, A.; Ghosh, C.; Das, S.; Chatterjee, A. J. Cancer Mol. 2008, 4, 55-60. (19) Noe, V.; Fingleton, B.; Jacobs, K.; Crawford, H. C.; Vermeulen, S.; Steelant, W.; Bruyneel, E.; Matrisian, L. M.; Mareel, M. J. Cell Sci. 2011, 114, 111-118. (20) Lochter, A.; Galosy, S.; Muschler, J.; Freedman, N.; Werb, Z.; Bissell, M. J. J. Cell Biol. 1997, 139, 1861-1872. (21) Mochizuki, S.; Okada, Y. Cancer Sci. 2007, 98(5), 621-628. (22) Duffy, M. J.; McKiernan, E.; O’Donovan, N.; McGowan, P. Clin. Cancer Res. 2007,13, 2335-2343. (23) Miller, M. A.; Barkal, L.; Jeng, K.; Herrlich, A.; Moss, M.; Griffith, L. G.; Lauffenburger, D. A. Integr. Biol. 2011, 3, 422-438. (24) Sarkaria, J. N.; Carlson, B. L.; Schroeder, M. A.; Grogan, P.; Brown, P. D.; Giannini, C.; Ballman, K. V.; Kitange, G. J.; Guha, A.; Pandita, A.; James, C. D. Clin. Cancer Res. 2006, 12, 2264-2271. (25) Ng, E. X.; Miller, M. A.; Jing, T.; Chen, C.-H. Biosens. Bioelectron. 2016, 81, 408-414. (26) Del Monte, U. Cell Cycle 2009, 8, 505-506. (27) Kessenbrock, K.; Plaks, V.; Werb, Z. Cell 2010, 141, 52-67. (28) Tissue expression of ADAM8 - Summary - The Human Protein Atlas. (accessed 25 April 2018) (29) Taylor, T. E.; Furnari, F. B.; Cavenee, W. K. Curr. Cancer Drug Targets 2012, 12, 197-209. (30) Park, C. M.; Park, M. J.; Kwak, H. J.; Lee, H. C.; Kim, M. S.; Lee, S. H.; Park, I. C.; Rhee C. H.; Hong, S. I. Cancer Res. 2006, 66, 85118519. (31) Kesanakurti, D.; Chetty, C.; Maddirela, D. R.; Gujrati, M.; Rao, J. S. Cell Death Dis. 2012, 3, e445-e445. (32) Ou, Y.; Wu, Q.; Wu, C.; Liu, X.; Song, Y.; Zhan, Q. J. Genet. Genomics 2017, 44, 557-565. (33) Addeo, R.; Zappavigna, S.; Parlato, C.; Caraglia, M. Expert Opin. Investig. Drugs 2014, 23, 1027-1037. (34) Mitra, A.; Mishra, L.; Li, S. Trends Biotechnol. 2013, 31, 347-354. (35) Edd, J. F.; Di Carlo, D.; Humphry, K. J.; Köster, S.; Irimia, D.; Weitz, D. A.; Toner, M. Lab Chip 2008, 8, 1262.

ACS Paragon Plus Environment

Page 9 of 9 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

Analytical Chemistry

For TOC Only

9 ACS Paragon Plus Environment