Antibody-conjugated beads outperform antibody-coated streptavidin beads for EV IP, in simple and complex matrices
Initially, streptavidin-conjugated beads, coated with biotinylated antibodies (MACS-STV), were compared against beads directly coated with covalently-conjugated antibodies (MACS). The performance of both bead surfaces was tested in the context of EV IP by assessing the recovery of fluorescently-labelled HEK293 EVs in PBS-BSA, with anti-CD9 and negative control antibody-coated beads. Isotype antibody-coated beads served as negative control to evaluate the specificity of MACS-STV, while anti-CD61-conjugated beads (MACS-CD61) played the same role for MACS. Since CD61 represents a cluster of differentiation of the platelet lineage, and was absent from our cell lines and respective EVs, it represented an appropriate negative control for cell line-derived EV IPs. In this case, traceable EV spike-in models were generated by staining HEK293 EVs with CFSE, which resulted in 87,9% of fluorescently-labelled particles, as detected by nFCM (Sup. Fig. 1A). Bead concentration was also measured to assure that the number of beads outnumbered fluorescent EV inputs in this and in the following experiments (Sup. Table 1).
CFSE-stained HEK293 EVs were incubated with beads and fluorescent readouts acquired for direct (beads) and indirect (IP flow-through) estimations of IP recovery. MACS-STV achieved 44% of specificity, regardless of the readout, whereas with MACS, 89% of specificity was observed by direct measurement and 75% by indirect measurement (Fig. 1A). Additionally, the fluorescent signal detected in MACS-CD9 beads evidenced a 48% recovery of CFSE-HEK293 input, which coincides with an average CD9 expression of 42% in these EVs (Sup. Fig. 1B). Considering a 5,5% of non-specific pull-down signal by MACS-CD61, nearly 100% of the CD9 subpopulation was efficiently recovered. Instead, direct measurements on MACS-STV revealed that 22,1% and 12,1% of spike was recovered with anti-CD9 and isotype control-coated beads, respectively. These results indicate that MACS but not MACS-STV, enabled IP of the entire CD9-positive subpopulation in PBS-BSA.
Subsequently, beads, EVs and the IP immune complex formed between them were visually examined by Cryo-TEM. An irregular bead structure could be discerned (Fig. 1B-1), contrasting with the circular shape of EVs and their well-defined membrane (Fig. 1B-2). IP complexes revealed that several beads can decorate the EV surface, which is likely dependent on the number of epitopes available for binding, amongst other factors (Fig. 1B-3).
Next, we evaluated the suitability of covalently-conjugated and streptavidin beads for EV IP in plasma, likely the most complex human biofluid. CFSE-labelled HEK293 EVs were spiked in healthy donor plasma (donor 6) and IP reactions conducted as aforementioned. Only direct bead readouts were plotted, since plasma emitted a great deal of background fluorescence in the CFSE channel, which compromised indirect readouts. Surprisingly, more fluorescent EV spike was recovered by negative control antibodies than with anti-CD9-coated or conjugated beads, on average (Fig. 1C). Such results would imply the lack of specificity of both bead surfaces in the plasma matrix, which was considered true for MACS-STV, as negative control beads were coated with a human isotype control antibody. Nevertheless, the negative control for the covalently-conjugated MACS beads was anti-CD61. Even though our spiked EVs did not express CD61, due to its abundance in plasma we reasoned that this platelet-related marker could be interacting with fluorescent EV spikes, causing their co-IP.
To confirm that the direct fluorescence readout accurately portrayed spike recovery, we extracted RNA from MACS-pulled down material (from Fig. 1C) and performed ddPCR for GAPDH and CA9, a stably transfected marker over-expressed in our HEK293 cells that becomes incorporated in their EVs, which is undetectable in healthy human plasma [63, 64]. The expression of both markers faithfully correlated with the previously measured fluorescent signal (Fig. 1C-MACS) and CA9 reads confirmed the specific recovery of our spike (Fig. 1D), indicating that fluorescence detected on beads stemmed from CFSE-labelled HEK293 EVs rather than from potential plasma-derived contaminants. Thus, ddPCR validated direct fluorescent measurements as reliable readouts of IP recovery.
To find trustworthy negative controls for assessing IP specificity from plasma, we re-tested the IP of CFSE-labelled HEK293 EVs spiked in plasma with MACS-CD9 against anti-phycoerythrin (PE) coated MACS beads (MACS-PE). PE is a commonly employed fluorophore produced by algae, which makes PE-coated beads an ideal negative control for IPs in human plasma. Spike recovery was now on average 4.5x higher with MACS-CD9 than with MACS-PE, conferring 78% of specificity to covalently-conjugated MACS beads in this experiment (Fig. 1E) and confirming our suspicion that CD61 was promoting spike co-IP.
Intrigued by this CD61-mediated capture of CD61-negative fluorescent EV spikes in plasma, we evaluated the recovery of CFSE-HEK293 EVs from the plasma of a different donor (donor 7), using MACS-CD9, CD61 and PE. Remarkably, CD61-mediated co-IP of CFSE spikes was not observed on a different plasma source as both negative controls displayed comparable fluorescence signals (Sup. Fig. 1C), indicating that this effect is dependent on biological variation.
In summary, antibody-coated MACS-STV specifically captured EVs only in a simple matrix, though they were markedly outperformed by antibody-conjugated MACS beads, which captured the totality of the CD9 subset in PBS-BSA and maintained a substantial degree of specificity, even in complex matrices. Therefore, we confirmed that the streptavidin-biotin surface chemistry is more prone to non-specific interactions in affinity-based EV isolation strategies. Importantly, both fluorescence measurement strategies and ddPCR proved to be valuable readouts that complemented and validated each other for precise quantifications of IP recovery.
Whole EV subpopulations can be efficiently captured from plasma, while spike recovery is dependent on EV surface phenotypes and biological variation of complex matrices
Having selected covalently-conjugated MACS beads due to their superior performance, we aimed at optimizing and exploring their capability to capture specific EV subsets from plasma. When we attempted to read CFSE-stained spike inputs and IP flow-throughs in plasma, S/N ratios were too low to extract meaningful information. Upon light absorption, plasma emits plenty of blue/green autofluorescence, which ultimately masked CFSE signal. Because biomolecules absorb and emit almost no NIR light, fluorescent NIR probes are a promising tool for in vivo and ex vivo imaging [65,66,67]. For this reason, we generated endogenously-labelled NIR EVs by feeding 22RV1 cells with a NIR probe, which is internalized and stably latches on to lipidic membranes, even after EV secretion [60]. After SEC purification of 22RV1-NIR CCM, we detected 92% of NIR-fluorescent particles and observed a CD9 expression of 20% (Fig. 2A). To understand if the entirety of a single EV subpopulation could also be retrieved from complex matrices, not only NIR but also CD9-PE-labelled NIR EVs were spiked in plasma, followed by IP with anti-CD9 and anti-PE beads. As expected, NIR spikes delivered better S/N ratios in the plasma matrix, with respect to CFSE spikes, which allowed reliable input measurements and both direct and indirect IP readout reporting. In line with previous plasma experiments, we estimated 87 and 70% of specificity through indirect and direct readouts, respectively, during NIR spike IP (Fig. 2B). Moreover, both readouts evidenced a 20% recovery of NIR input, which exactly matched the proportion of CD9-positive 22RV1-NIR EVs, suggesting that the whole CD9 subpopulation of spiked EVs could be retrieved from plasma (Fig. 2B).
Interestingly, the recovery of CD9-PE-labelled NIR EV spike was comparable between CD9 and PE beads. The direct bead readout even evidenced a slightly higher mean of 18% for PE over 14% for CD9 (Fig. 2C), which could hint that CD9 epitopes may be less accessible to MACS-CD9 when anti-CD9-PE had already occupied them. These experiments confirmed the high efficacy of this IP approach for recovering distinct EV subsets from plasma, further highlighting its specificity and flexibility also by indirect capture.
IP complexes formed in plasma were monitored by Cryo-TEM, where an abundance of beads over EVs could be appreciated, while the size of captured EVs spanned over a wide range (Fig. 2D). Whether EVs were recovered from plasma (Fig. 2D-1) or from HEK293-spiked plasma (Fig. 2D-2), IP complexes greatly resembled the ones observed after IP in simple matrices (Fig. 1B), demonstrating that actual EV-like particles, with intact structure and function, could be efficiently retrieved from complex matrices.
Still, we reckoned that IP reactions would be more efficient in PBS-BSA than in plasma, due to the richness of the latter in biomolecules that can hinder affinity interactions. To assess that, we spiked 22RV1-NIR and HT29-CFSE in both matrices, conducted IP with triple-coated, anti-tetraspanin (CD9, CD63 and CD81) MACS beads and read their recovered fluorescence. On average, spike recovery was similar between buffer and plasma with 22RV1-NIR and 52% higher in buffer than in plasma with and HT29-CFSE (Fig. 3A). This observation suggested that depending on the identity of EV spikes, different interactions between EVs, matrix components and affinity reagents likely occur, affecting IP recovery.
To evaluate the impact of IP conditions on spike recovery, HT29-CFSE EVs were spiked in plasma from donor 6 (the same used in aforementioned experiments) and triple-coated MACS incubated for 10, 25 or 60 min. We confirmed a maximum average fluorescence signal at 60 min, whilst maintaining specificity (Sup. Fig. 2A). Moreover, the same HT29-CFSE spike was equally captured from donor 6 plasma increasingly diluted with PBS (Sup. Fig. 2B), showing that matrix dilution did not improve IP performance.
Having confirmed that IP conditions did not contribute to the variable recovery of different EV spikes, we further addressed this aspect by spiking CFSE-labelled EVs from three different cell lines (HT29, HEK293 and A549) in PBS-BSA and plasma from a different donor (donor 7), applying triple-coated MACS for IP. This time, fluorescent signals showed that HT29 EVs were equally recovered from both matrices. Similarly, the recovery of A549 EVs was not significantly different between the two matrices, while surprisingly, 33% more HEK293 spike was captured from plasma (Fig. 3B). In conclusion, such results demonstrate that the surface properties of distinct EV subsets can influence on how they are targeted and retrieved by affinity reagents, within a given matrix.
The complexity of plasma samples, exacerbated by wide inter-individual variation, is one of the major factors limiting clinical use of affinity-based assays. The disparity observed in HT29 EV recovery between PBS-BSA and plasma from donors 6 and 7 in two independent experiments (Fig. 3A, B), prompted us to estimate the real impact of biological variation on spike IP from complex matrices.
For this purpose, CFSE-labelled HEK293 and HT29 EVs were spiked into three different plasma sources and their recovery was assessed through direct IP fluorescence readouts. The recovery of HEK293 spike was similar between plasma samples of donors 5 and 7, although it doubled in donor 8 plasma. Plasma from donor 5 resulted in the lowest recovery of HT29 EVs, as this signal tripled in donor 7 and reached its maximum in donor 8 plasma (Fig. 3C). Intriguingly, the recovery of HEK293 EVs from plasma samples of donors 5 and 7 remained constant, while it tripled for HT29 EVs, further highlighting the weight of EV surface phenotypes in IP efficiency. Taken together, these results show how the affinity isolation of EV subpopulations depends both on their inherent surface characteristics, and on the composition of the matrix they are carried in.
Multiple surface markers can be directly detected to quantify EV subpopulations captured from simple and complex matrices
Upon optimization and characterization of this IP approach employing fluorescently-labelled EV spike-in models, we attempted instead to stain captured EVs directly on beads, using fluorescently-tagged primary antibodies. With the goal of developing a strategy to quantitatively detect EV subpopulations retrieved from plasma, we initially set out to gauge the staining of bead-bound HT29 EVs with CD9-PE, directly after IP with triple-coated MACS in PBS-BSA. S/N ratios obtained on increasing EV numbers could be faithfully represented by simple linear regression (R2 = 0,9992), from 1 × 108 down to 5 × 106 EVs, which corresponded to a S/N of 7 (Fig. 4A). As such, this strategy revealed quite robust for EV detection and quantification in simple matrices. To understand its applicability in plasma, we attempted to first deplete endogenous plasma EVs with triple-coated MACS, then HT29 spikes were added to this “EV-depleted plasma” and IP was performed, using also triple-coated beads. Detection with CD9-PE displayed a linear trend from 1 × 108 to 1 × 107 HT29 EVs (R2 = 0,9783), however at the lowest spike amount (5 × 106), an unexpected sharp increment in S/N ratios was noticed (Fig. 4B). Moreover, CD9-PE S/N ratios were substantially larger in plasma-derived bead samples (Fig. 4A, B), which suggested that either the pre-IP depletion step was incomplete, or that the majority of signal stemmed from nonspecific antibody binding.
To address the specificity of fluorescently-labelled primary antibody staining of plasma-derived material on beads, platelet-derived EVs were isolated from plasma samples of 3 independent donors, using MACS-CD61. Detection was done by targeting CD41, a platelet-related marker that forms a heterodimer with CD61 known as integrin αIIbβ3, present exclusively in the platelet lineage [68, 69]. A PE-labelled isotype-matched antibody was used as negative control. Specific CD41-PE signal was measured with different intensity across all three plasma samples, always significantly higher than respective negative controls, confirming that this bead-based sandwich immunoassay assay could specifically detect surface markers carried on EVs retrieved from plasma (Fig. 4C). Through Cryo-TEM we verified that, consistent with aforementioned images, MACS-CD61 clearly enabled the isolation of EV-like structures, suggesting that platelet-derived EVs could be efficiently captured from plasma (Fig. 4D).
Subsequently, we explored the possibility of simultaneously detecting two markers through double staining of platelet EVs, isolated from plasma with MACS-CD61. S/N ratios obtained after staining with anti-CD41-PE and anti-CD9-AF488 were comparable, regardless of their incubation being conducted in single or in combination (Fig. 5A), meaning that staining efficiency and accuracy is maintained as two markers are concomitantly detected.
Finally, we investigated if double staining could provide meaningful information in the analysis of EV subpopulations derived from complex samples, exploiting platelet-derived EVs as a paradigmatic example. To do so, 1 mL of plasma was first heated to 56 °C or treated with thrombin (2 U) for 8 min. Both procedures cause the precipitation of fibrinogen from plasma, noticeable by increased opacity or by the polymerization of an insoluble clot after 56 °C or thrombin treatment, respectively [70, 71]. For both treatments, insoluble fibrinogen was eliminated by centrifugation at 5000 g for 5 min and the resulting supernatant collected in a clean tube, to which either triple-coated MACS or MACS-CD61 were added. Since fibrinogen strongly interacts with the CD41/CD61 complex (also termed the fibrinogen receptor, required for clot formation) on platelets [72], we postulated that the effects of such treatments would mostly reflect on the detection of platelet markers, on platelet-derived EVs. To verify it, CD9-AF488 and CD41-PE double staining was conducted on recovered beads after incubation with treated and untreated plasma.
Fibrinogen-depleting treatments did not majorly impact CD9 detection on triple-coated MACS, however on MACS-CD61, a significant drop in fluorescent signal could be appreciated upon plasma pre-heating at 56 °C (Fig. 5B). On the other hand, sharp losses of CD41-PE signal were observed after treatments, on both triple-coated MACS and on MACS-CD61 (Fig. 5C). These results corroborated our hypothesis, as mostly platelet-related markers were indeed lost upon thrombin or 56 °C treatment, indicating that, not only fluorescent EV spike-ins but also endogenous plasma EVs were specifically captured from plasma samples, and that multiple EV subpopulations from complex samples can be simultaneously detected using this staining protocol. Of note, neither of the plasma pre-analytic treatments aforementioned resulted in increased overall EV recovery.
Taken together, we established that a simple incubation step with fluorescently-labelled antibodies on EV-carrying beads, recovered from simple or complex matrices, enables accurate detection and quantification of multiple surface markers expressed on EV immunoprecipitates.
Different EV subpopulations carry distinct mRNA biomarkers that can be valuable for liquid biopsy-based early-stage NSCLC detection
As any bona fide enrichment strategy depends on its’ specificity, we ultimately sought to provide definitive evidence to validate the performance of our IP protocol. To do so, the mRNA content of different plasma EV subpopulations, recovered either by triple-coated MACS or MACS-CD61, was profiled using the nCounter platform. Moreover, to inquire about the utility of this strategy in a real liquid biopsy scenario, both EV subpopulations were isolated from the plasma of two different cohorts, each composed of 14 donors. The expression of 594 transcripts was measured using the nCounter Human Immunology v2 Panel. To avoid biased conjectures and guarantee the quality of gene expression reads, a dedicated bioinformatics pipeline was developed, including the internal standard nCounter QC checks, exploratory data analysis (EDA), low-count gene filtering steps, normalization and DE analysis.
Firstly, we questioned whether different EV subpopulations, derived from the same healthy donor samples, contained distinct mRNA profiles. During EDA on the comparison of healthy donor EVs obtained with triple-coated or CD61 beads, PCA revealed that samples seemed to slightly cluster by the number of unnormalized reads and group (Sup. Fig. 3A, B, C), but not by batch (Sup. Fig. 3D). However, after examining unnormalized counts per group, we concluded that CD61+ EVs displayed a significantly higher number when compared to CD9, CD63 or CD81+ EVs (p = 0.00053, Wilcoxon; Sup. Fig. 4A), indicating that the apparent clustering by group, defined in this case by IP target, did not truly occur, as it was driven by the number of mRNA counts. RLE plots demonstrated that optimal sample normalization could be achieved with DESeq2 (Sup. Fig. 3E, left). Normalized samples were visually inspected on a PCA plot, which evidenced the two most variable samples depicted on RLE plots (Sup. Fig. 3E, right). DESeq2 output four DE genes, one upregulated and three downregulated, when comparing MACS-CD61 against the triple-coated MACS dataset (Fig. 6A). Supervised hierarchical clustering analysis was performed using the four DE genes and presented on a heatmap (Fig. 6B). Altogether, our data supports that depending on the targeted surface markers, distinct EV subsets could be effectively isolated from healthy donor plasma.
To understand the potential clinical value of each EV subset as biomarker carrier, we confronted our healthy cohort against a prospective early-stage NSCLC cohort, applying the same pipeline. Surprisingly, no DE genes were found after comparing the EV mRNA profiles obtained from healthy and cancer samples, using triple-coated MACS (Fig. 7A). On the other hand, the platelet-derived EV dataset allowed to compare healthy and early-stage cancer cohorts. As previously observed during EDA, samples only seemed to somewhat cluster by the number of unnormalized gene counts, but not by group or batch (Sup Fig. 5A-D). Despite being marginally elevated, the average number of unnormalized counts was not significantly higher in the early-stage NSCLC cohort (p = 0.43, Wilcoxon; Sup. Fig. 4B). As before, DESeq2 alone optimally normalized all samples (Sup Fig. 5E left). Similarly, PCA plots of normalized counts evidenced a separation of the most variable samples (RLE plots, Sup Fig. 5E left) from the main sample cluster (Sup Fig. 5E right). DE analysis with DESeq2 found 47 DE genes, which were visualized on a volcano plot (Fig. 7B). These results suggest that the identified mRNA expression patterns displayed by CD61-positive EVs, may allow for distinction between healthy and early-stage cancer samples. In summary, our experimental data demonstrates that different EV subpopulations can indeed be captured by targeting different surface markers, which reflected on their mRNA profiles and disclosed how distinct EVs subsets may confer differential clinical values in relevant liquid biopsy settings. In this case, the identification of 47 putative biomarkers for blood-based early-stage NSCLC detection, revealed that platelet-derived EVs represent an appealing biomarker source that warrants extended studies.