SUPPLEMENTARY INFORMATION

Protein typing of circulating microvesicles allows real-time monitoring of glioblastoma therapy

Huilin Shao1,2, Jaehoon Chung1, Leonora Balaj3, Alain Charest4, Darell D. Bigner5, Bob S. Carter6, Fred H. Hochberg7, Xandra O. Breakefield3,8, Ralph Weissleder1,7,9*, Hakho Lee1*

1 Center for Systems Biology, Massachusetts General Hospital, 185 Cambridge St, Boston, MA 02114 2 Harvard Biophysics Program, Harvard Medical School, Boston, MA 021153 Neuroscience Center, Department of Neurology, Massachusetts General Hospital, Charlestown Navy Yard, Boston, MA 021294 Molecular Oncology Research Institute, Tufts University School of Medicine, Boston, MA 021115 Brain Tumor Center, Department of Pathology, Duke University Medical Center, Durham, NC 277106 Division of Neurological Surgery, UCSD School of Medicine, San Diego, CA 921037 Massachusetts General Hospital Cancer Center, Boston, MA 021148 Program in Neuroscience, Harvard Medical School, Boston MA 021149 Department of Systems Biology, Harvard Medical School, Boston, MA 02115

1

Supplementary Methods

Fabrication of microfluidic deviceThe microfluidic system was fabricated by stacking three polydimethylsiloxane (PDMS; Dow Corning) layers on a glass slide. All PDMS layers were prepared by soft lithography, replicating channels from the patterns of epoxy-based SU8 photoresist (Microchem). The bottom PDMS layer was fabricated by spin-coating PDMS on a SU8 mold. This allowed the fabrication of a thin membrane for control channels, which open and close the main fluidic channels by pneumatic valves1. The remaining layers were prepared by direct PDMS pour. A custom-built microcoil for NMR measurement was embedded within the middle layer along with the main fluid channels. The top layer contained channels connecting the coil and the outlet. Holes were punched to connect the middle and top layers. For filter incorporation, a membrane filter (pore size: 50 nm; Nuclepore, Whatman) was placed between the top and middle layer before irreversible sealing. After bonding the top and the middle layer, inlets and outlets were punched out. The bottom layer was then aligned and connected to the middle layer, and connection holes for control channels were punched. Finally, the stacked PDMS block was attached to a glass slide.

Tetrazine (TZ) modification of magnetic nanoparticlesAmine-terminated cross-linked iron oxide (CLIO) nanoparticles with a core size of 7 nm were obtained from the CSB Chemistry Core (MGH). Amino-CLIO was modified with 2,5-dioxopyrrolidin-1-yl 5- (4-(1,2,4,5-tetrazin-3-yl)benzylamino)-5-oxopentanoate (TZ-NHS) to create CLIO-TZ2. The reaction was performed in excess TZ-NHS relative to amino-CLIO, and proceeded in PBS containing 0.1M sodium bicarbonate for 3 hours at room temperature. Following conjugation, excess TZ-NHS was removed using Sephadex G-50 columns (GE Healthcare).

Antibody modification with transcyclooctene (TCO)Antibodies were modified with (E)-cyclooct-4-enyl 2,5- dioxopyrrolidin-1-yl carbonate (TCO-NHS) as previously reported2,3. Each antibody was buffer-exchanged into PBS (pH 8.0) using 2 mL Zeba desalting columns (Thermo Fisher). Purified antibodies were then reacted with TCO-NHS in 10% dimethylformamide for 3 hours at room temperature. TCO conjugated antibodies were subsequently purified by buffer exchange into PBS and their concentrations determined by absorbance measurements. The following antibodies were used in this study (see Supplementary Table 4 for details): human CD63 (BioLegend, Abgent), mouse CD63 (MBL International), human wild type EGFR (Thermo Scientific), human EGFRvIII (clone L8A4, generated and provided by Dr. Bigner), human PDGFRα (R&D Systems), mouse PDGFRα (BioLegend), human PDPN (eBioscience), mouse PDPN (R&D Systems), human and mouse EphA2 (R&D Systems), human IDH1 R132H (MBL International, Abgent), human and mouse HSP90 (AbCam), human and mouse CD41 (BioLegend), human and mouse MHCII (Santa Cruz Biotechnology), and respective isotype controls (BioLegend).

Scanning electron microscopyGBM cells (GBM20/3) were grown on a glass coverslip, fixed with half-strength Karnovsky’s fixative and washed twice with PBS. The cells were then dehydrated in a series of increasing ethanol concentrations and transferred for critical drying (Samdri, Tousimis). After coating with platinum/palladium using a sputter coater (208HR, Cressington Scientific Instruments), the sample was imaged with a scanning electron microscope (Supra55VP, Carl Zeiss).

2

Transmission electron microscopyFor transmission electron microscopy, GBM20/3 MVs were fixed with 2% paraformaldehyde and immuno-targeted with MNPs (core size of 7 nm) via CD63 antibody. The targeted MVs were then membrane-filtered similarly as in the microfluidic device, and the retentate was collected. After contrast staining with uranyl oxalate/methyl cellulose, the sample was imaged with a transmission electron microscope (JOEL 2100).

In vitro drug treatmentTemozolomide (TMZ, Temodar, Schering Plough) and geldanamycin (17AAG, Tocris) were used to investigate the effects of drug treatments. To determine the inhibitory concentrations (IC50) of these drugs, cells were seeded at a density of 2,000 cells/well in a 96-well plate overnight, and treated with drug (TMZ or geldanamycin) or vehicle (final concentration of 0.1% DMSO) for four days. Cell viability was assessed using the 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium inner salt (MTS) cell proliferation assay (Promega). For cellular and MV analyses of drug effects, T103 and GLI36vIII cells were treated with varying doses of drugs at concentrations comparable to their respective IC50 for 48 hours before subsequent analyses.

Flow cytometryAdherent cells were trypsinized and labeled with antibodies (10 µg/mL) in PBS with 0.5% bovine serum albumin (BSA, Sigma) for 45 minutes at 4 °C. Following centrifugation and aspiration of the antibody solution, cells were labeled with FITC-conjugated secondary antibodies (AbCam) and washed twice by centrifugation. For intracellular labeling (HSP90, IDH1 R132H), trypsinized cells were fixed with 2% paraformaldehyde and permeabilized with BD perm/wash buffer (BD Biosciences) before antibody incubation and washing. FITC fluorescence was assessed using a LSRII flow cytometer (Becton Dickinson). Mean fluorescence intensity (MFI) was determined using FlowJo software, and biomarker expression levels were normalized with appropriate isotype control antibodies.

Western blot analysisGBM cells and their purified MVs were lysed in radio-immunoprecipitation assay buffer and supplemented with protease inhibitors (Thermo Scientific). MV samples were collected either directly from the microfluidic device or following further fractionation into subpopulations of ≤ 100 nm (pore size: 100 nm; Anodisc, Whatman; Supplementary Fig. 4a). Protein concentration was quantified using the bicinchoninic acid assay (BCA assay kit, Thermo Scientific). Protein lysates were loaded and resolved by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE), and subjected to immunoblotting with EGFR antibody, HSP90 antibody, Integrin β1 antibody, Integrin α5 antibody (Cell Signaling), Flotillin 1 antibody, Flotillin 2 antibody (BD Biosciences), TSG101 antibody, CD9 antibody (AbCam) and CD63 antibody (Santa Cruz Biotechnology). Following incubation with horseradish peroxidase-conjugated secondary antibody (Cell Signaling), enhanced chemiluminescence was used for immunodetection.

Mouse tumor modelWe used cells derived from a previously described transgenic GBM model4. T103 cells (5×106) in Matrigel (Becton Dickinson) were implanted into immunodeficient nu/nu mice. Tumors were allowed to grow for two weeks before mice were randomized into two groups (a control and treatment group). For the treatment group, 80 mg/kg TMZ was administered intraperitoneally on a

3

daily basis. Tumor volumes in control and treated animals were measured by magnetic resonance imaging (MRI) following 1, 3, 5 and 7 days of continuous treatment; blood samples were collected on these days for µNMR measurements. All animal procedures were performed according to guidelines issued by the Committee of Animal Care of Massachusetts General Hospital.

Data analysisAll measurements were performed in triplicate, and the data is displayed as mean ± standard error of the mean. Receiver operating characteristic (ROC) curves for individual markers and the four-marker average were constructed by plotting sensitivity versus (1 − specificity), and the values of area under the curve (AUC) were computed using the trapezoidal rule. The optimal ξ-threshold for each marker and the four-marker average was obtained from the point closest to the top-left part (perfect sensitivity or specificity) of the corresponding ROC curve. The empirical ROC curves were smoothed by applying the binormal fitting model5. Detection sensitivity, specificity and accuracy were calculated using standard formulas. In evaluating drug responses of cells and MVs, an unpaired, one-sided t-test was employed, and a P value of < 0.05 was considered statistically significant. Statistical analysis was performed using the R-package (version 2.13.2).

4

Supplementary Figure 1. Microscopy analysis of MVs. (a) MVs from GBM20/3 primary human glioblastoma cell line, previously used in other MV studies6,7, were counted by nanoparticle tracking analysis (NTA). The system monitored the trajectory of MV movement (in red) to obtain MV sizes and concentrations. NTA was initially used to estimate the MV concentration in a given sample. For consistent reading, the measurement settings were optimized and maintained for a detection range of 10 - 500 nm. (b) Size distribution of MVs purified by the developed microfluidic device. Purified GBM20/3 MVs were imaged by transmission electron microscopy (inset), and the size histogram was obtained by measuring 100 MVs. The prepared MVs have a log-normal size distribution (dotted line) with a median diameter of 83 nm. Note that about 75% of the prepared MV population was less than 120 nm in diameter.

5

Supplementary Figure 2. Workflow for extra- and intravesicular biomarker detection by µNMR. MVs were isolated from biofluids via gradient centrifugation as previously described6,7. For detection of membrane proteins, MVs were directly incubated with TCO-modified antibody and coupled with TZ-modified magnetic nanoparticles (MNP-TZ) in the microfluidic device. For detection of intravesicular proteins, MVs were first lysed and then incubated with antibody-coated polymer beads (500 nm in diameter). Similar to ELISA, TCO-modified detection antibody was then added to create a sandwich assay, onto which MNP-TZ were coupled. Following on-chip filtration to remove unbound nanoparticles, NMR measurements were made with built-in microcoils to determine R2 values of samples. As a control, MV concentration-matched sample was incubated with TCO-modified isotype control antibody2. Relative R2 changes (∆R2) were then calculated as the R2 differences between the targeted and the control samples. MV biomarker expressions were normalized by CD63 expression to account for differences in MV numbers.

6

Supplementary Figure 3. Two-step operation of the microfluidic system for on-chip circulating MV detection. (a) Schematics of a microfluidic MV detection system which can load and test three samples individually. Insets show the operation of the device, visualized by flowing dye solutions. Clockwise from the bottom left: a peristaltic pump which drives fluid flow through the channels; a microcoil and an embedded membrane filter system for µNMR detection; a herring-bone mixer showing mixing of two different colors. (b) The fluid pathway in the first operation step (colored violet): a MV sample (Sample 1, yellow dot in the illustration) and magnetic nanoparticles (MNPs) are injected, and mixed while traveling across the herring-bone mixer. As the mixture moves across the membrane filter, MNP-bound MVs are retained while excess unbound MNPs pass through. After mixing and filtration, a buffer solution is injected to further wash the retained MVs and clear the channels. To prevent contamination, three filters were embedded, one for each target marker. (c) The fluid pathway in the second operation step (colored red). Buffer solution is introduced to transport the targeted MVs to the microcoil. Note that the fluid direction is opposite to that of the first operation step to enable the release of the captured MVs. Once the buffer solution containing the targeted MVs fills the bore of the microcoil, the fluid flow is halted and NMR measurements are performed. After measurement, all channels are washed by keeping the buffer solution flowing.

7

Supplementary Figure 4. Western blot analysis of purified MV samples. (a) MVs from GBM20/3 and GLI36vIII cells were harvested using the microfluidic device (left) or were fractionated to ≤ 100 nm (right). Lysates of both samples were immunoblotted for exosomal markers (HSP90, Flotillin 1, Flotillin 2, TSG101, CD9, and CD63) as well as other vesicular markers (Integrin β1 and Integrin α5)8,9 before undergoing chemiluminescence to estimate relative protein abundance. Both MV samples exhibited similar protein compositions, consisting mainly of exosome-specific markers such as HSP90, Flotillin 1, Flotillin 2, TSG101 and CD9. Importantly, CD63 expression was found to be both more consistent as well as higher than other exosomal markers; this corroborated our selection of CD63 as the normalizer for subsequent protein analyses. (b) Immunoblotting of SkMG3 MV lysates with CD63 antibody. The detection threshold for Western blotting was ~108 MVs. In comparison, using CD63-tagged MVs, the detection threshold for µNMR was ~104 MVs (corresponding to ~0.02 ng of protein).

8

Supplementary Figure 5. Protein stability of MVs. (a) GBM20/3 MVs, fixed with paraformaldehyde and non-fixed, were analyzed with µNMR for CD63 expression over time at room temperature. Fixed MVs maintained protein expression for at least 8 hours at room temperature, whereas proteins progressively degraded in the non-fixed samples. (b) Paraformaldehyde fixation has little effect on the immunoreactivity of MV proteins. When GLI36vIII MVs were targeted with the selected antibodies, both fixed and non-fixed MV preparations showed similar measurements on µNMR. Experiments were performed quickly at 4 °C to prevent potential protein degradation.

9

Supplementary Figure 6. Expression of different protein biomarkers in circulating MVs isolated from human blood samples. (a) The expression levels of four markers (EGFR, EGFRvIII, PDPN, and IDH1 R132H) are arranged by subject number. High degree of heterogeneity in MV protein expression is apparent across different subjects. (b) Receiver operating characteristic (ROC) curves for four markers as well as the numerical average of four-marker combination are shown. Raw data (dashed lines) were fitted to a linear model to generate the smoothed (solid lines) sensitivities and specificities. Note that the ξ-cutoff, area under curve (AUC), sensitivity, specificity and accuracy values were all obtained from ROC curves with raw data (Supplementary Table 2).

10

Supplementary Figure 7. Microvesicular response to temozolomide (TMZ) treatment. After treating T103 cells with varying doses of TMZ, cellular and MV responses were monitored. (a) Effects of TMZ treatment on cellular expression of key protein markers. The expression level of CD63, EGFR or EGFRvIII in cells treated with TMZ was unchanged relative to untreated cells, as determined by flow cytometry and Western blotting. (b) Following TMZ treatment, µNMR detection showed that MVs have similar EGFR and EGFRvIII profiles to that of whole cells. Normalized expressions of EGFR and EGFRvIII (ξ), with respect to CD63 expression in MVs, were unaffected by TMZ treatment, a finding that was further verified by Western blotting of MVs. (c) Overall cell and MV numbers decreased in a TMZ dose-dependent manner. Changes are statistically significant with P < 0.05. (d) Total levels of CD63, EGFR, and EGFRvIII in MV samples, as measured by µNMR. In parallel to the TMZ dose-dependent decrease in total MV number, all protein markers showed a similar decline (compared to the untreated sample; P < 0.0001). All analytical measurements were performed in triplicate, and the data is shown as mean ± s.e.m.

11

Supplementary Figure 8. Characterization of drug treatment effects on GBM MVs in the GLI36vIII model. (a) Effects of TMZ treatment on expression of key proteins. TMZ treatment did not change the expression level of CD63 or EGFRvIII in cells, as determined by flow cytometry and Western blotting. µNMR detection showed that the normalized expression of EGFRvIII in MVs, with respect to CD63 expression, was not affected by TMZ application. This was further confirmed by Western blotting on the MVs. (b) Effects of TMZ treatment on cell and MV numbers. TMZ treatment decreased overall cell and MV numbers. (c) Effects of geldanamycin treatment on expression of key protein biomarkers. Geldanamycin treatment did not change the expression level of CD63, but considerably reduced the amount of EGFRvIII in cells, as determined by flow cytometry and Western blotting. µNMR measurements indicated that the normalized expression of EGFRvIII in MVs exhibited a similar dose-dependent decrease with geldanamycin treatment; this was further confirmed by Western blotting on the MVs. (d) Effects of geldanamycin treatment on cell and MV numbers. Geldanamycin treatment decreased overall cell and MV numbers in a dose-dependent manner.

12

Supplementary Table 1. Clinical information for patient profiling study.Sample # Age Sex WBC MRI Max area (cm2) Type

1 52 M 4.1 + 75 GBM with prior GBMO2 51 F 6.5 + 164 Recurrent GBM3 61 M 4.5 + 14 GBM with prior GBMO4 74 M 13.5 + 15 Newly diagnosed GBM5 79 M 8.2 + 58 Newly diagnosed GBM6 73 M 16.2 + 19 Newly diagnosed GBM7 69 M 15.7 + 5 Newly diagnosed GBM8 85 M 12.8 + 27 Newly diagnosed GBM9 69 M 3.7 + 44 Recurrent GBM

10 62 F 16.5 + 31 Newly diagnosed GBM11 72 F 20.5 + 20 Newly diagnosed GBM12 72 F 12.2 + 25 Newly diagnosed GBM13 63 M 9.8 + 16 Newly diagnosed GBM14 50 F 7.6 + 16 Newly diagnosed GBM15 71 M 7.6 + 172 Newly diagnosed GBM16 72 F 20.2 + 21 Newly diagnosed GBM17 53 F 10.2 + 17 Newly diagnosed GBM18 51 F 12.6 + 32 Newly diagnosed GBM19 57 M 13.4 + 29 Newly diagnosed GBM20 62 F 13.5 + 17 Newly diagnosed GBM21 74 M 13.5 + 14 Newly diagnosed GBM22 33 M 13.6 + 228 Recurrent GBM23 64 M 11.8 ND ND Newly diagnosed GBM24 48 M ND + 276 GBM with prior GBMO25 39 M Normal Normal NA NA26 32 F Normal Normal NA NA27 26 M Normal Normal NA NA28 35 M Normal Normal NA NA29 30 M Normal Normal NA NA30 23 NA Normal Normal NA NA31 NA NA Normal Normal NA NA32 47 NA Normal Normal NA NA

All patients have single focus GBM (rather than multifocal), and all had surgery for 'resection with purpose of treatment' rather than just a biopsy. Normal samples were obtained from MGH blood bank (blinded clinical samples with identifiers coded) and from healthy donors. GBMO: glioblastoma multiforme with oligodendroglial component; ND: not determined; NA: not available.

13

Supplementary Table 2. Summary of µNMR detection on patient samples.

Markers ξ–cutoff* Sensitivity Specificity Accuracy AUC

EGFR 0.06 0.64 0.88 0.70 0.78

EGFRvIII 0.06 0.68 1.00 0.76 0.88

PDPN 0.08 0.68 0.88 0.73 0.82

IDH1-R132H 0.06 0.16 1.00 0.36 0.60

QUAD† 0.04 0.92 0.88 0.91 0.95

*The cutoff values for ξ were determined from the points on ROC curves that have the minimal distance to the 0% false-negative and the 100% true-positive. †QUAD, four-marker average.

14

Supplementary Table 3. Clinical information for treatment response study.

Sample # Age Sex GBM typeTreatmentTreatmentTreatmentTreatment

ResponderSample # Age Sex GBM type Radiation Surgery TMZ Other Responder

33 52 M GBM Yes Yes Yes Cediranib No34 51 F GBM (rec) Yes Yes Yes Cabozantinib Yes35 51 F GBM (rec) Yes Yes Yes Cabozantinib No36 61 M GBM (rec) Yes Yes Yes Bevacicumab Yes37 74 M GBM Yes Yes Yes Cediranib Yes38 79 M GBM Yes Yes Yes Vandetanib No39 79 M GBM Yes Yes Yes Vandetanib No40 73 M GBM (rec) Yes Yes Yes Vandetanib No41 69 M GBM Yes Yes Yes None No42 85 M GBM Yes Yes Yes None Yes43 69 M GBM (rec) Yes Yes Yes None Yes44 69 M GBM (rec) Yes Yes Yes None Yes

GBM (rec): GBM recurrent.

15

Supplementary Table 4. List of antibodies used in the study.

Target Vendor Clone Cat. #

Human CD63 Ancell AHN16.1/46-4-5 215-820Human CD63 Biolegend MEM259 312002Human CD63 Abgent RB26813 AP5333bMouse CD63 MBL International R5G2 D263-3Human WT EGFR Thermo Scientific EGFR.1 MA1-24225Human EGFRvIII Dr. Bigner Laboratory L8A4Human PDGFRα R&D Systems PRa292 MAB1264Mouse PDGFRα Biolegend APA5 135902Human PDPN eBioscience NZ-1.3 14-9381-82Mouse PDPN R&D Systems Ployclonal AF3244Human EphA2 R&D Systems 371805 MAB3035Mouse EphA2 R&D Systems 233720 MAB639Human IDH1 R132H MBL International HMab-1 D299-3Human IDH Abgent RB18237 AP7454aHuman and mouse HSP90 Abcam AC88 ab13492Human CD41 Biolegend HIP8 303702Mouse CD41 Biolegend MWReg30 133902Human and mouse MHCII Santa Cruz Biotechnology YE2/36-HLK sc-80970

All antibodies were conjugated at a concentration > 0.5 mg/ml and used at 10 µg/ml.

16

Supplementary References

1.' Unger, M. A., Chou, H. P., Thorsen, T., Scherer, A. & Quake, S. R. Monolithic microfabricated valves and pumps by multilayer soft lithography. Science 288, 113-116 (2000).

2.' 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 5, 660-665 (2010).

3.' Haun, J. B., Devaraj, N. K., Marinelli, B. S., Lee, H. & Weissleder, R. Probing Intracellular Biomarkers and Mediators of Cell Activation Using Nanosensors and Bioorthogonal Chemistry. ACS Nano (2011).

4.' Zhu, H. et al. Oncogenic EGFR signaling cooperates with loss of tumor suppressor gene functions in gliomagenesis. Proc Natl Acad Sci U S A 106, 2712-2716 (2009).

5.' Robin, X. et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 12, 77 (2011).

6.' Skog, J. et al. Glioblastoma microvesicles transport RNA and proteins that promote tumour growth and provide diagnostic biomarkers. Nat Cell Biol 10, 1470-1476 (2008).

7.' Balaj, L. et al. Tumour microvesicles contain retrotransposon elements and amplified oncogene sequences. Nat Commun 2, 180 (2011).

8.' Thery, C., Ostrowski, M. & Segura, E. Membrane vesicles as conveyors of immune responses. Nat Rev Immunol 9, 581-593 (2009).

9.' Simons, M. & Raposo, G. Exosomes--vesicular carriers for intercellular communication. Curr Opin Cell Biol 21, 575-581 (2009).

17