Purification of HCC-specific extracellular vesicles on nanosubstrates for early HCC detection by digital scoring

Purification of HCC-specific extracellular vesicles on nanosubstrates for early HCC detection by digital scoring

The design and preparation of an EV click chip

An EV Click Chip (Fig. 1) is composed of two functional components: (i) Tz-grafted SiNWS: a patterned SiNWS32 covalently functionalized with disulfide bonds that link to terminal Tz motifs40, and (ii) an overlaid PDMS chaotic mixer41 (Supplementary Fig. 1), housed in a custom-designed microfluidic chip holder. The fabrication of Tz-grafted SiNWS began with introducing 10–15 µm densely packed Si nanowires (diameter = 100–200 nm) onto SiNWS, offering ~30 times more surface area (in contrast to a flat substrate) for facilitating click chemistry-mediated HCC EV capture. The incorporation of disulfide bonds and terminal Tz motifs onto SiNWS was carried out via a 3-step procedure40 (Supplementary Fig. 2). To confirm successful preparation of Tz-grafted SiNWS, X-ray photoelectron spectroscopy (XPS) was employed to monitor functional group transformation at each step9,40. The passive mixing behavior of the flow-through EVs in EV Click Chips was simulated (Supplementary Fig. 3) via the combined use of computational fluid dynamics (CFD) and dissipative particle dynamics (DPD) models24, offering a theoretical explanation on how the configuration of the EV Click Chip results in the enhanced physical contact42 between TCO-grafted HCC EVs and Tz-grafted SiNWS.

Preparation of artificial plasma samples

To allow accurate evaluation of the performance of EV Click Chips throughout the optimization process, artificial plasma samples were prepared by spiking 10-µL aliquoted HepG2 cell-derived EVs (harvested by ultracentrifugation43,44) into 90-µL plasma from a female healthy donor. As shown in Fig. 2a, the presence of male HepG2 cell-line-derived EVs in female plasma allows exploitation of the sex-determining region Y (SRY) gene for reliable quantification of HepG2-derived HCC EVs in purified EV samples since the SRY gene is absent in female healthy donor’s plasma.

Fig. 2: Optimization of EV Click Chips using artificial plasma samples.
figure2

a A quantitative method was developed for evaluating the performance of EV Click Chips using artificial plasma samples prepared by spiking HepG2 EVs into the plasma from a female healthy donor (HD). A RT-ddPCR assay was employed to quantify the copy numbers of the SRY and C1orf101 transcripts in the purified EV samples to calculate the recovery yield and recovery purity. *n is the ratio between C1orf101 and SRY transcripts in HepG2 EVs. b The recovery yields observed for EV Click Chips at different TCO-to-anti-EpCAM mole ratios. Data are presented as means ± SD of three independent assays. c–f The recovery yields obtained in the presence of individual and combined antibody capture agents, i.e., c anti-EpCAM, d anti-ASGPR1, e anti-CD147, and f combination of the three capture agents. Data are presented as means ± SD of three independent assays. g The recovery yields with different flow rates. Data are presented as means ± SD of three independent assays. h Dynamic ranges of EV recovery yields observed for EV Click Chips using artificial sample containing 0–9000 copies of SRY transcripts. Data are presented as means ± SD of three independent assays. i HepG2 EV recovery performance observed for (i) optimized EV Click Chips, devices without embedded silicon nanowires in SiNWS, and devices without herringbone features in the PDMS chaotic mixer, (ii) devices based on immunoaffinity EV capture (NanoVilli Chips) using the antibody cocktail concentration optimized for EV Click Chips, and (iii) ultracentrifugation (UC) approach. Data are presented as means ± SD of three independent assays. j General applicability of EV Click Chips for HCC EV recovery performance was validated using six artificial samples prepared by spiking three different HCC EVs (collected from HCC cell lines, i.e., HepG2, SNU387, and Hep3B) into two types of plasma samples (collected from either HD or liver cirrhotic patients). Data are presented as means ± SD of three independent assays.

RT-ddPCR assay for quantification of EVs

A RT-ddPCR assay in Fig. 2a was used to quantify the copy numbers of SRY and C1orf101 transcripts (encoded on Chromosome Y and Chromosome 1, respectively) in the artificial plasma samples before and after purification by EV Click Chips. The results can be used to calculate the recovery yield and recovery purity throughout the optimization process. We denoted the copy numbers of SRY transcripts in the original 10-µL aliquoted HepG2 EVs and the EV Click Chip-recovered HepG2 EVs as SRY transcriptsori-EV and SRY transcriptsrec-EV, respectively. The EV recovery yield obtained by EV Click Chips under a given condition can be obtained from the following equation:

$${mathbf{HepG}}2,{mathbf{EV}},{mathbf{recovery}},{mathbf{yield}} = frac{{{boldsymbol{SRY}},{mathbf{transcripts}}_{{mathbf{rec}} – {mathbf{EV}}}}}{{{boldsymbol{SRY}},{mathbf{transcripts}}_{{mathbf{ori}} – {mathbf{EV}}}}}$$

(1)

In order to obtain the recovery purity of the EVs recovered by EV Click Chips, we first measured the intrinsic ratios between C1orf101 and SRY transcripts in aliquoted HepG2 EVs across a wide range of concentrations. As shown in Supplementary Fig. 4a, the ratios between C1orf101 and SRY transcripts in HepG2 EVs exhibited a consistent linear correlation (y = 1.95 x, R2 = 0.999). With the C1orf101-to-SRY ratio determined as 1.95, we then calculated the recovery purity of the HepG2 EVs harvested from EV Click Chips as the ratio of the recovered SRY transcripts (contributed by recovered HepG2 EVs only) to the C1orf101 transcripts (contributed by both recovered HepG2 EVs and the nonspecifically captured background plasma-derived EVs, denoted as C1orf101 gene rec-EV) using the following equation:

$$begin{array}{l}{mathrm{HCC}},{mathrm{EV}},{mathrm{recover}},{mathrm{purity}} = frac{{SRY,{mathrm{transcripts}}_{{mathrm{rec}} – {mathrm{EV}}}}}{{C1orf,101,{mathrm{transcripts}}_{{mathrm{rec}} – {mathrm{EV}}}}} times 1.95 ast \ ast 1.95,{mathrm{is}},{mathrm{specific}},{mathrm{to}},{mathrm{HepG2}},{mathrm{EVs}},.end{array}$$

(2)

For HCC cell lines without SRY transcripts, cancer-cell-derived EVs were spiked into plasma from male donors, and the EV recovery yield and recovery purity can be calculated using equations shown in Supplementary Methods and Supplementary Fig. 4b, c. A reproducibility study on the C1orf101/SRY transcript quantification methods used in the equations was conducted and the results are summarized in Supplementary Table 1.

HCC EV purification with EV click chips

Prior to conducting HCC EV purification (capture/release) studies, TCO motif was covalently conjugated onto each antibody agent (Fig. 1a), and the TCO-conjugated antibody agents were incubated with the artificial or clinical plasma samples for 30 min at room temperature. In each study (Fig. 2a), a 100-µL artificial plasma sample was introduced into an EV Click Chip, in which the click chemistry-mediated rapid and irreversible immobilization of HCC EVs on SiNWS. Next, 100 µL DTT (50 mM) was introduced into the EV Click Chips to achieve disulfide cleavage-driven EV release. The DTT was removed in the subsequent RNA extraction process.

A multimarker cocktail optimization for HCC EV capture

Using published data from our group31 and others45,46, we identified surface markers that are highly expressed in HCC EVs, HCC CTCs, HCC cell lines, and primary tumor tissues of HCC patients, but virtually absent in white blood cells. Four candidate antibodies, i.e., anti-EpCAM, anti-ASGPR1, anti-CD147, and anti-GPC-3, against the corresponding surface markers were selected to achieve desired sensitivity and specificity for recognizing and capturing HCC EVs. The aforementioned RT-ddPCR assay was employed to assess the EV recovery yield of EV Click Chips in the presence of the individual antibodies and their cocktail mixtures. Figure 2b summarizes the recovery yields obtained by EV Click Chip at different TCO-to-anti-EpCAM mole ratios, and an optimal recovery yield was achieved at the TCO-to-anti-EpCAM ratio of 4:1. Under this TCO-to-antibody ratio, we suggest that the optimal amounts of individual candidate antibodies, i.e., anti-EpCAM (Fig. 2c), anti-ASGPR1 (Fig. 2d), anti-CD147 (Fig. 2e), and anti-GPC-3 (Supplementary Fig. 5a) are 50, 25, 25, and 50 ng, respectively. Using these optimized conditions, we compared the HCC EV recovery yields with different antibody cocktails. The data is summarized in Fig. 2f and Supplementary Fig. 5b and shows that the combination of anti-EpCAM, anti-CD147, and anti-ASGPR1 outperformed any single antibodies or other combinations.

Optimization of EV Click Chips for HCC EV purification

With the optimal antibody cocktail, flow rates of samples into EV Click Chips were studied, and >85% average recovery yields were observed at the flow rates of 0.2–1.0 mL h−1 (Fig. 2g). To allow for a faster turnaround time for clinical samples, the flow rate of 1.0 mL h−1 was selected. We then checked the dynamic range of EV Click Chips using artificial plasma samples spiked with different concentrations of EVs containing 0–9000 copies of SRY transcripts per 100-μL volume and confirmed the consistency of recovery yields (y = 0.827x, R2 = 0.998) (Fig. 2h). To understand the crucial roles of the embedded silicon nanowires in SiNWS, the herringbone features in a PDMS chaotic mixer, and click chemistry-mediated EV capture, we carried out control experiments (Supplementary Fig. 6) using (i) the devices without embedded silicon nanowires in SiNWS or herringbone features in the PDMS chaotic mixer, and (ii) the devices based on immunoaffinity EV capture24 (NanoVilli Chips), in parallel with EV Click Chips and the ultracentrifugation approach44. EV Click Chips exhibited a recovery yield of 82.7 ± 1.34% and recovery purity of 90.2 ± 6.2%, which were significantly higher than those observed for the controls (Fig. 2i). The reproducibility of the EV Click Chips was evaluated by calculating the percent coefficient of variation (%CV) for recovery yields. The observed %CVs were calculated to be 1.12–12.65% for the intra-assay variability and 3.88 % for the inter-assay variability of the EV Click Chips (Supplementary Table 2). To test the general applicability of EV Click Chips and the optimized EV purification condition, the performance of EV Click Chips was further tested using six artificial samples prepared by spiking three different HCC EVs (collected from HCC cell lines, i.e., HepG2, SNU387, and Hep3B) into two types of plasma samples (collected from either healthy donors or liver cirrhotic patients). Detailed calculations of the reproducibility, recovery yields, and recovery purities for these artificial samples are described in Supplementary Table 3 and Supplementary Fig. 4. Overall, EV Click Chips achieved recovery yields ranging from 81.2 to 94.6% and purities ranging from 85.9 to 99.1% (Fig. 2j).

Characterization of HCC EVs purified by EV click chips

To better understand the working mechanisms of the click chemistry-mediated EV capture and disulfide cleavage-driven EV release, fluorescence microscopy, transmission electron microscopy (TEM), dynamic light scattering (DLS), and/or scanning electron microscopy (SEM) were employed to characterize the EV sizes and EV/SiNWS interfaces during the EV purification process, in which freshly harvested HepG2 EVs in PBS and healthy donors’ plasma (Supplementary Fig. 7a–c) were used as a model system. To allow direct tracking of the capture and release processes of HCC EVs in EV Click Chips, HepG2 EVs were first labeled (Fig. 3a) with PKH26 dye (Sigma–Aldrich). The micrographs in Fig. 3b unveiled fluorescent signals on the SiNWS after EV capture and a dramatic signal reduction when the captured EVs were released by DTT. Figure 3c shows a representative TEM image of freshly harvested HepG2 EVs after uranyl acetate negative staining. These HepG2 EVs exhibited cup- or spherical-shaped morphologies with sizes ranging between 30 and 500 nm in diameter measured by TEM (inset of Fig. 3c). The size distributions of EVs measured by TEM were consistent with those observed by DLS (Supplementary Fig. 7d, e). Figure 3d shows a cross-sectional SEM image of Si nanowires with HepG2 EVs captured onto both the sidewalls (left) and the tops of the nanowires (right). After being released from EV Click Chips, the purified HepG2 EVs retained intact morphologies (Fig. 3e) with a similar size distribution (inset of Fig. 3e) to the freshly harvested HepG2 EVs. The purified HepG2 EVs from EV Click Chips were further verified by immunogold labeling with anti-CD63 (Supplementary Fig. 7f).

Fig. 3: Characterization of HepG2 EVs purified by EV Click Chips.
figure3

a Fluorescent labeling of HepG2 EVs by PKH26 dye, followed by incubation with TCO-grafted antibody cocktail, giving PKH26-labeled TCO-grafted HepG2 EVs. b Tracking the purification (capture/release) process of HepG2 EVs in EV Click Chips using fluorescent microscopy. After click chemistry-mediated capture, PKH26-labeled HepG2 EVs were immobilized on SiNWS, as confirmed by the fluorescence micrograph (upper). Upon exposure to DTT, the surface linkers that anchored the PKH26-labeled HepG2 EVs onto SiNWS were cleaved, leading to the release of PKH26-labeled HepG2 EVs, as confirmed by fluorescence micrograph (lower). Data are representatives of three independent assays. c, Representative transmission electron microscopy (TEM) images of HepG2 EVs in bulk solution before capture. Inset: Size distribution (n = 338, diameters = 30–500 nm) of HepG2 EVs, measured by TEM. d Scanning electron microscopy (SEM) images of HepG2 EVs (colored in pink) on the sidewall (left) and tops (right) of the SiNWS. Data are representatives of three independent assays. e Representative TEM images of HepG2 EVs after being released from the chip. Inset: Size distribution (n = 363, diameters = 40–500 nm) of HepG2 EVs, measured by TEM.

Quantification of 10 HCC-specific genes using purified HCC EVs

By adopting the optimal HCC EV purification conditions, a workflow (Fig. 4a) for a streamlined HCC EV-based mRNA assay was developed by coupling EV Click Chips and RT-ddPCR for quantification of 10 well-validated HCC-specific mRNA transcripts39 using clinical plasma samples. We collected 158 plasma samples from five cohorts, including (i) HCC cohort: newly diagnosed, treatment-naive HCC patients (n = 46, mean age = 66 y); (ii) cirrhosis cohort: patients with liver cirrhosis covering the etiology of hepatitis B virus (HBV), hepatitis C virus (HCV), alcoholic liver disease (ALD), and non-alcoholic steatohepatitis (NASH) (n = 26, mean age = 61 y). We confirmed that the cirrhosis cohort did not have HCC at the time of blood draw based on (1) negative multiphasic CT/MRI results, or (2) negative liver ultrasound results at the time of blood draw and a 6 month follow-up, or (3) observing no evidence of HCC on liver explant. (iii) hepatitis cohort: patients with chronic hepatitis B/C without liver cirrhosis (n = 25, mean age = 57 y); (iv) healthy donors (n = 23, mean age = 52 y); (v) other cancer cohort: patients with primary malignancies other than HCC, with or without liver metastases (n = 38, mean age = 58 y). The clinical characteristics of these cohorts are provided in Supplementary Tables 4–8. Clinical annotation of all the plasma samples was performed by a clinician blinded to the assay. For each clinical sample, 0.5 mL of aliquoted plasma was introduced into an EV Click Chip to obtain purified HCC EVs. After RNA extraction, RNA concentrations were evaluated by Bioanalyzer 2100, (Supplementary Table 9), then RT-ddPCR was carried out to quantify the 10 HCC-specific genes, i.e., alpha-fetoprotein (AFP), glypican 3 (GPC3), albumin (ALB), apolipoprotein H (APOH), fatty acid binding protein 1 (FABP1), fibrinogen beta chain (FGB), fibrinogen gamma chain (FGG), alpha 2-HS glycoprotein (AHSG), retinol binding protein 4 (RBP4), and transferrin (TF)39. We confirmed that these 10 mRNA markers are detectable in pure HepG2 EVs (Supplementary Fig. 8b). In addition, the publicly available EV databases, i.e. ExoCarta47, Vesiclepedia48, and exoRBase49 also supported that these 10 mRNA markers are detectable in EVs. Considering EV-resident RNAs can be full-length or sometimes fragmented50, the primers and probes of the 10 genes are specially designed to amplify short amplicons located 3’-most. To ensure the reproducibility of the ddPCR assay, we validated the PCR primers and probes using cDNA obtained from HepG2 cells, HepG2 EVs, and HCC EVs purified from five HCC patients’ plasma samples by random priming reverse transcription (Supplementary Fig. 8). Finally, we summarized the HCC EV-derived 10-gene signatures obtained from 158 individual subjects in heatmaps (Fig. 4b); the primary copy numbers are log2-transformed for each gene across all disease states. As depicted in the heatmaps, higher signals were observed in the HCC cohort, compared with those from the noncancer cohorts (i.e., cirrhosis, hepatitis, and healthy donors) and other cancer cohort, including intrahepatic cholangiocarcinoma (ICC), breast cancer, lung cancer, prostate cancer, midgut neuroendocrine tumor (NET), and cancers of nonhepatic origin metastatic to the liver (MET). Furthermore, signal differences between early-stage and advanced-stage HCC patients defined by the Barcelona Clinic Liver Cancer (BCLC) staging system51 allow for the separation of these two subgroups. Both Milan criteria52 and United Network for Organ Sharing down-staging (UNOS DS)53 criteria were also adopted to separate the HCC cohort into the respective early and advanced stages, and the results can be found in Supplementary Figure 9.

Fig. 4: RT-ddPCR assay for quantification of 10 HCC-specific mRNA transcripts in purified HCC EVs.
figure4

a A general workflow developed for conducting HCC EV purification, followed by quantification of 10 HCC-specific mRNA transcripts in the purified HCC EVs. b Heatmaps depicting relative signal intensities for each gene expression of the 10 HCC-specific genes across different patient cohorts. (upper) Patients with newly diagnosed HCC (n = 46) are grouped according to Barcelona Clinic Liver Cancer (BCLC) staging system from early stages to advanced stages. (middle) Noncancer cohorts, including patients with liver cirrhosis (n = 26), chronic hepatitis (n = 25), and healthy donors (n = 23). (lower) Patients with cancers other than HCC (n = 38): cancers of nonhepatic origin metastatic to the liver (MET, n = 12); other primary cancers (n = 26), including intrahepatic cholangiocarcinoma (ICC), prostate cancer, midgut neuroendocrine tumor (NET), breast cancer, and lung cancer. Primary copy numbers are log2-transformed for each gene across all disease states. Clinical characteristics for each cohort are listed in Supplementary Tables 4–8. HBV hepatitis B virus, HCV hepatitis C virus, ALD alcoholic liver disease, NASH non-alcoholic steatohepatitis.

HCC EV Z Scores for digital scoring

We computed HCC EV Z Scores for each sample based on its 10-gene signatures in purified HCC EVs using the weighted Z-score method54. The copy numbers of the 10 genes were combined into the single HCC EV Z Scores. As depicted in the box plot (Fig. 5a), the HCC EV Z Score of the HCC cohorts (both early and advanced stages) are significantly higher (****P < 0.0001) than the noncancer cohorts (i.e., cirrhosis, hepatitis, and healthy donors) and other cancer cohort. HCC largely occurs in the setting of pre-existing chronic liver diseases55. However, it can also develop in the absence of such conditions. We thus performed receiver operator characteristic (ROC) analysis to test the potential of HCC EV Z Score for distinguishing HCC patients from noncancer patients (i.e., cirrhosis, hepatitis, and healthy donors). The area under the ROC curve (AUC) for distinguishing HCC from noncancer was 0.87 (95% CI, 0.80–0.94; sensitivity = 93.8%, specificity = 74.5%, Fig. 5b). Similarly, the potential of HCC EV Z Score for distinguishing HCC patients from primary malignancies other than HCC with or without liver metastases was then explored, and the AUC was 0.95 (95% CI, 0.90–1.00; sensitivity = 95.7%, specificity = 89.5%, Fig. 5c).

Fig. 5: Statistical analysis on HCC EV Z Scores in different cohorts.
figure5

a Box plots representing the HCC EV Z Scores for different patient cohorts including early-stages HCC (n = 36), advanced-stage HCC (n = 10), cirrhosis (n = 26), hepatitis (n = 25), healthy donors (n = 23), and other cancers (n = 38). Whiskers ranging from minima to maxima, median and 25–75% IQR shown by box plots. Significant differences between different groups were evaluated using one-way ANOVA. b, c ROC curves for HCC EV Z Scores in b HCC versus noncancer (i.e., cirrhosis, hepatitis, and healthy donors) (AUC = 0.87, P = 9.64E-12, 95% CI, 0.80–0.94), c HCC versus other cancer (AUC = 0.95, P = 1.79E-12, 95% CI, 0.90–1.00). d ROC curves comparing HCC EV Z Scores (AUC = 0.93, P = 1.02E-8, 95% CI, 0.86–1.00) with the serum biomarker alpha-fetoprotein (AFP) level (AUC = 0.69, P = 0.013, 95% CI, 0.55–0.83) for differentiating early-stage HCC (BCLC, stage 0-A) vs. at-risk cirrhosis. Barcelona Clinic Liver Cancer (BCLC); ROC receiver operator characteristic.

HCC EV Z Scores for early HCC detection

Finally, we examined the potential of HCC EV Z Score to distinguish early-stage HCC (BCLC stage 0-A) from at-risk liver cirrhosis. The AUC was 0.93 (95% CI, 0.86–1.00; sensitivity = 94.4%, specificity = 88.5%, Fig. 5d), which outperformed the most widely used serum biomarker AFP (AUCs of 0.69, 95% CI, 0.55–0.83; sensitivity = 66.7%, specificity = 72.0%) for early-stage HCC detection. The ROC curves in Supplementary Fig. 10 also demonstrated that HCC EV Z Score outperformed serum AFP testing in distinguishing early-stage HCC (defined by Milan52 and UNOS DS53 criteria) from at-risk liver cirrhosis with AUC of 0.91 versus 0.68, and 0.92 versus 0.70, respectively.

Source Article