SUPPLEMENTARY INFORMATION

Transcription

1 SUPPLEMENTARY INFORMATION Multiplexed Profiling Of Single Extracellular Vesicles Kyungheon Lee 1#, Kyle Fraser 1#, Bassel Ghaddar 1, Katy Yang 1, Eunha Kim 1, Leonora Balaj 2, E. Antonio Chiocca 3, Xandra O. Breakefield 2, Hakho Lee 1 *, Ralph Weissleder 1,2,4 * 1 Center for Systems Biology, Massachusetts General Hospital, 185 Cambridge St, CPZN 5206, Boston, MA 02114, United States 2 Department of Neurology, Massachusetts General Hospital, Boston, MA 02114, United States 3 Department of Neurosurgery, Brigham and Women's Hospital, Boston, MA 02114, United States 4 Department of Systems Biology, Harvard Medical School, 200 Longwood Ave, Boston, MA 02115, United States # equal contribution Corresponding authors: *R. Weissleder, MD, PhD or Hakho Lee, PhD Center for Systems Biology Massachusetts General Hospital 185 Cambridge St, CPZN 5206 Boston, MA, rweissleder@mgh.harvard.edu; hlee@mgh.harvard.edu

2 Supplementary Table 1. Typical signal measurements. Type Mean diameter Range (nm) SNR Comment Reference beads 250 nm Commercial beads used for quantitation/referencing EV 120 nm Not used for diagnosis if SNR < 2 EV clusters >300 nm nm > 10 Typically absent in SEA after filtration Protein complexes < 20 nm < 50 nm >10 or <2 Typically non-detectable in SEA Supplementary Table 2. Typical variations (20 objective). Type Experiment Variations (s.d.) 1. Imaging different FOV 4% Physical 2. Repeat experiment over long time period 4% 3. Spatial stability of captured EV (loss of captured EV) ~1% Experimental 4. Duplicate procedure on same batch of EV (antibody staining) 10%

3 Supplementary Table 3. Antibodies used. Antibody Company Catalog Number Species CD9 Abcam ab92726 Rabbit IgG CD63 Ancell Mouse IgG1 CD81 Santa Cruz Biotechnology sc Mouse IgG2b EGFR EGFRvIII Cell Signaling Technology Cell Signaling Technology 5108 Rabbit IgG Rabbit IgG IDH1 BioLegend Mouse IgG1,k IDH1R132H Millipore MABC171 Mouse IgG1,k PD-L1 Cell Signaling Technology 15005S Rabbit IgG PD-L2 BioLegend Mouse IgG1,k PDGFRa Cell Signaling Technology 3174 Rabbit IgG Podoplanin Abcam ab10288 Mouse IgG1

4 Supplementary Table 4. Breakdown of the assay time. Note that actual hands-on time is approximately 2 hours for an entire measurement set. Sample prepration Capture/staining/imaging Image analysis Sub-steps Ultracentrifugation or izon column Density calibration (Nanosight) Biotinylation Capture Fix/perm Blocking A Staining Quenching Blocking B Capture Image analysis tsne pipeline Approximate time 2.5 hr / 1 hr 30 min 60 min 30 min 10 min 30 min 30 min/cycle 15 min/cycle 15 min/cycle 1 min/cycle 5 min with labtop 30 min Supplementary Table 5. Number of EV in the clusters shown in Fig. 5. Cluster number Gli36-WT Gli36-EGFRvIII Gli36-IDH1R132 Total Total

5 ! Supplementary Fig. 1. Validation experiments. (a) Temporal stability of the measurement system. Fluorescent beads (diameter, 250 nm) were monitored over time by the SEA system. The signal variations were <4% (shaded region). Each data point is the intensity average from 465 beads (mean ± s.e.m). (b) Aliquots from a single parent sample were subject to repeated quenching steps before staining for CD63. The observed MFIs (from ~500 EV) were statistically identical (p = 0.242, one way ANOVA). Data is displayed as mean ± s.e.m. (c) The staining order also does not affect quantitative results. In this example, the order of CD63 staining was varied: 1st, 2nd (after EGFR), 3rd (after EGFR, CD81) and 4th (after EGFR, CD81,IDH1). The curve shows the density function of CD63-positive population (~500 EV) and was normalized with the area under the curve equal to 100%.

6 Supplementary Fig. 2. Effect of EV biotinylation. (Top) Biotinylated EV were captured, and their expression level of EGFR was measured, MFI = ± 389 (s.d). (Bottom) Nonbiotinylated EV were captured based on CD63 expression, and EGFR was profiled, MFI = ± 1470 (s.d.). The capture method does not influence the measurements of target expression (EGFR in this example) but has major effects on capture efficiency. In this example, nearly one half of all EGFR vesicles would have been missed by CD63 immunocapture.

7 Captured EV counts Nominal EV counts Supplementary Fig. 3. EV capture rate. We changed vesicle concentrations, and counted the number of captured EV. The estimated EV counts were calculated as EV concentration fluidic volume in the FOV. The average capture rate was ~4.5%. Data are from five experimental replicates, and displayed as mean ± s.e.m.

8 Supplementary Fig. 4. Spiked EV measurement with 6 markers. EVs from two cell lines (Gli36-EGFRvIII and Gli36-IDH1R132H) were spiked into human serum. Spots with circles indicate individual EV. The sample specimen did not affect measurement. As with native samples, serum spiked samples showed similar levels of EGFRvIII positive vesicles (6.1% when spiked in human serum vs 6.5% in native sample) and IDH1R132H positive vesicles (4.8% when spiked in human serum vs 4.0% in native sample).

9 ! Supplementary Fig. 5. Comparison of EV expression levels among the three cell clones: Gli36- WT, Gli36-EGFRvIII and Gli36-IDH1R132. The density functions were normalized with the area under the curve equal to 100%.

10 ! Supplementary Fig. 6. Comparison between SEA and bulk ELISA. From SEA results, the mean fluorescent intensities (MFIs) were calculated from population density distributions. Aliquots of the same samples were measured by ELISA to quantify target markers in bulk EVs. MFIs were then compared with signals from ELISA measurements. The results showed high correlation (R 2 = 0.891). ELISA data are from three technical replicas, and are displayed as mean ± s.e.m.

11 Supplementary Fig. 7. tsne analyses. tsne maps with different perplexity (P) values were generated

12 Supplementary Fig. 8. Selecting the optimal number of clusters and tsne plot. A consensus clustering algorithm was employed in which 50 k-means clustering runs were used to produce a consensus matrix that quantifies how often two data points were clustered together. This was done testing between 10 and 30 clusters on each plot, and was done for all the tsne plots. (a) Representative cumulative density function (CDF) for all cluster numbers of the consensus matrix for the tsne plot (perplexity, 100). This plot is used to calculate the proportion of ambiguous clustering (PAC). The optimal number of clusters for a given perplexity value minimizes the difference in the CDF function values at the two dashed thresholds. (b) Representative PAC plot showing how a number of clusters was determined. (c) Plot of Davies- Bouldin indices for each tsne plot clustered at its optimal number of clusters found using the PAC method. The best data representation and clustering minimizes the Davies-Bouldin Index.

13 Supplementary Note 1. Optical collection efficiency (ε) of the microscope used Parts Description Efficiency Objective lens 0.45 N.A. with 85% transmittance Attenuation from optical path a dichromatic mirror 0.9 an emission filter 0.9 Overall ( ) Photon flux to a single pixel from a fluorescent antibody Quantum (Q) yield of fluorochrome: 0.1 (Alexa 555) Absorption cross-section (A) of fluorochrome: cm 2 Estimated number (N) of fluorochromes on an antibody: 2 Incident light photon flux (I): 1 mw (input) / cm 2 (area) = 2.5 W/cm photons/cm 2 /sec Emitted photon flux (F): F = N A Q I = 360 photons/sec Total photons flux to camera (T): T = ε F = 27 photons/sec In our system, typically three pixels are used to detect a single EV spot. Therefore, Photon flux per pixel (P): P = T / 3 = 9 photons/sec 3. Signal-to-noise ratio (SNR) per pixel The SNR is estimated as SNR = P Qe t / (P Qe t + D t + Nr 2 ) 0.5 where Qe is the quantum efficiency, D is the dark current, Nr is read noise of the camera, and t is the acquisition time. With the camera system used (Andor Zyla), Qe = 0.55, D = 0.1 e /sec, Nr = 1.7 e. The acquisition time was ~1 sec to avoid photobleaching. Therefore, the expected SNR from a single antibody is SNR ~1.8.