Effects of packing density and adsorption conditions on extracellular vesicle dynamic binding capacities for capillary-channeled polymer (C-CP) fiber columns

Abstract

Extracellular vesicles (EVs) are membrane-bound nanoparticles (50–1000 nm) secreted by all cell types and play critical roles in various biological processes. Among these, exosomes, a smaller subset of EVs, have attracted considerable interest due to their potential applications in diagnostics and therapeutics. However, conventional EV isolation methods are often limited by inefficiencies in processing time, recovery, and scalability. Hydrophobic interaction chromatography utilizing capillary-channeled polymer (C–CP) fiber stationary phases offers a promising alternative, enabling rapid (<15 min), cost-effective (~$5 per column) EV isolation with high loading capacities (~10¹⁰–10¹² particles) and minimal sample pre-processing. Despite these advantages, achieving high-throughput EV isolation for larger-scale applications using the C–CP fiber platform is the present challenge. To this end, further optimization of stationary phase packing and adsorption conditions is necessary to maximize the available binding surface area in the current microbore column format. This study systematically investigates the influence of interstitial fraction (i.e. packing density) in polyester (PET) C–CP fiber columns on the dynamic binding capacity (DBC) of EVs isolated from human urine using a high-performance liquid chromatography platform. Microbore columns (0.76 mm i.d. × 300 mm) packed with PET C–CP fibers in both an eight-channel (PET-8) and a novel trilobal (PET-Y) configuration were evaluated using breakthrough curves and frontal analysis. The results reveal that lower packing densities correlate with higher mass- and surface area-based EV binding capacities, with a maximum DBCs of 2.86 × 10¹³ EVs g⁻¹ fiber and 1.22 × 10¹⁴ EVs m⁻² fiber achieved in <2 min of sample loading. Under optimum conditions, surface utilization of >50 % is realized. These results establish a framework for optimizing C–CP fiber-based platforms to enhance EV capture efficiency, facilitating the development of scalable EV isolation techniques for biomedical research and therapeutic applications.

1. Introduction

Extracellular vesicles (EVs) are nanometer-scale vesicles secreted by all living cells, being composed of a phospholipid bilayer membrane which encloses lipid, protein, and genetic (DNA, mRNA, miRNA) cargos from the cells of origin [1,2]. Exhibiting heterogeneity in size (ranging from 30 to 1000 nm), composition, and function, EV populations reflect their original microenvironment and mode of formation [3,4]. Among EVs, exosomes, physically-characterized by diameters ranging from 30 to 150 nm, emerge as pivotal agents in intercellular communication [5]. These small vesicles, originating from diverse cell types, bear specific protein biomarkers on their membrane surfaces, indicative of their cells of origin [6]. Recognized for their potential as biomarkers, exosomes hold promise in disease diagnostics, progression monitoring, and treatment response assessment. Moreover, their stable and rich cargo of proteins, RNA, lipids, and metabolites along with the innate cell-targeting characteristics have opened the way for exosomes to be developed as a drug delivery platform (i.e., vectors) [7]. The stability of exosomes in circulation, biocompatibility/non-immunogenic properties, and target specificity are key advantages for overcoming the constraints associated with standard drug delivery systems [8,9].

To fully harness the diagnostic and therapeutic potential of exosomes, a significant challenge lies in the efficient isolation of varied EV subpopulations from diverse/complex biological fluids. One of the primary limitations to the broader use of exosomes as drug delivery agents is the extremely low throughput of the current isolation methods, compounded by their low overall purities. Traditional EV isolation techniques include ultracentrifugation (UC), gravity-fed size-exclusion chromatography (SEC), ultrafiltration (UF), immunoaffinity (IAF), and polymer precipitation (PP)-based methods [10–12]. Among these, UC is widely regarded as the “gold standard” for EV isolation [11,13]. However, this technique, like others, suffers from several limitations, including low yield recovery, co-isolation of protein and lipoprotein aggregates, limited throughput, high operational costs, and potential disruption of EV membrane integrity due to the intense centrifugal forces applied [10,11,13]. Furthermore, the inherent challenges in exosome production for therapeutic applications are exacerbated by the increasing demand for larger quantities, with an estimated requirement of 0.5–1.4 × 10¹¹ exosomes per patient treatment [14]. Exosomal proteins, as key components of exosome cargo, serve as promising source of markers for cancer diagnosis, prognosis, and therapeutic applications [15]. However, the yield of exosomal proteins from cell culture media is typically low, often amounting to <1 μg mL⁻¹ [16]. In contrast, in vivo applications require substantially higher quantities, with effective doses ranging from 10 to 500 μg of exosomal protein per mouse [17,18]. The considerable gap between production yields and application requirements underscores the urgent need for scalable, efficient, and high-throughput EV isolation strategies.

To overcome the limitations associated with existing EV isolation techniques, Marcus and colleagues [19,20] developed a hydrophobic interaction chromatography (HIC) method for the isolation and purification of EVs from diverse biological matrices. This method utilizes polyester (PET) capillary-channeled polymer (C–CP) fiber stationary phases in both conventional HPLC column formats [19] and solid-phase extraction (SPE) spin-down tips [21]. Using these approaches, EVs have been effectively isolated from cell culture media and a wide array of biofluids, such as plasma, serum, urine, unpasteurized goat milk, cervical mucus, and saliva [20,22]. Notably, HIC methods employing C–CP fiber stationary phases have demonstrated high throughput and efficiency, achieving concentrated EV recoveries ranging from 10¹⁰ to 10¹² particles mL⁻¹ with purities as high as 95 % from highly complex matrices [22,23]. Additionally, the method enables EV isolation on practical timescales (<15 min), yielding populations devoid of low-density lipoprotein contamination [24]. Importantly, the biological integrity and purity of the recovered EVs have been confirmed through immunoassays, protein assays, and electron microscopy, further validating the robustness of this technique.

Central to the efficacy of C–CP fiber-based EV purification methodology are the distinctive properties of native PET fiber stationary phases. The micro- and macro-structural attributes of the fibers, combined with their parallel orientation within the column, enable high linear velocity operation (~100 mm s⁻¹) without excessive backpressure [25]. When packed, the fibers interdigitate to create collinear, capillary-like channels, resembling the behavior of thousands of single-micron, open tubular columns. This configuration enhances hydrodynamic properties and mass transfer efficiency [26], facilitating efficient fluid transport with minimal flow resistance. Furthermore, the nonporous surface of the fibers relative to the size of proteins and EVs eliminates intrafiber diffusion [26], thereby improving mass transfer kinetics and enabling rapid separations. In many respects, these characteristics are akin to those of monolithic chromatographic columns, with the performance of the two compared for protein separations in previous efforts [27]. Prior to this point, methacrylate-based monoliths have been employed for targeted EV capture following surface modification to affect immunoaffinity separations [12], but very recently Wall and co-workers [28] have described the use of anion exchange (QA) and hydrophobic interaction (HIC) monoliths as an EV polishing step following initial processing via hollow fiber tangential flow fractionation (TFF).

Building upon these advantageous properties noted to date, optimization of the column’s dynamic binding capacity is crucial for maximizing EV throughput and recovery. The novelty of the present effort is the first detailed investigation critical factors affecting binding kinetics and capacities for the previously-described eight-legged C–CP fiber shape as well a newly-developed trilobal geometry [29]. This requires a systematic evaluation of stationary phase packing and EV adsorption conditions, as trade-offs may exist between the fiber surface area for adsorption determined by the packing density (i.e., number of fibers) and the accessibility to that surface area. Accordingly, this study examines the influence of column interstitial fraction (i.e. fiber packing density) on the dynamic binding capacity of EVs within a microbore column (0.76 mm i.d. × 300 mm long) format using an HPLC platform. Dynamic loading characteristics were assessed as a function of column interstitial fraction and mobile phase linear velocity using frontal analysis (FA) as the quantitative approach [30,31]. The study compares PET C–CP fibers with an eight-channel configuration (PET-8) to a novel trilobal fiber geometry (PET-Y) [29], which was evaluated towards achieving more uniform packing and improved column performance. Scanning electron microscopy (SEM) images (Figs. 1a–d) highlight the structural differences between PET-8 and PET-Y fibers, offering insights into their packing behavior and potential effects on EV binding efficiency. By quantifying the EV loading and throughput characteristics from a key biological matrix, human urine, this study provides a foundation for scaling up EV isolation processes using C–CP fiber columns. These findings have implications for advancing fundamental EV research, clinical diagnostics, and the development of therapeutic delivery systems.

Fig. 1.

SEM images of C–CP fiber-packed column cross sections showing a) PET-8 fibers with an ε_i of 0.61, b) PET-Y fibers with an ε_i of 0.62, and the cross-sectional morphology of individual fibers for c) PET-8 and d) PET-Y. e) Micrographs showing EVs captured on the PET-Y fiber surface.

2. Materials and methods

2.1. Chemicals and reagents

Ultrapure-grade ammonium sulfate ((NH₄)₂SO₄) was obtained from Thermo Scientific (Waltham, MA), and HPLC-grade acetonitrile (ACN) was sourced from VWR Chemicals (Radnor, PA). Gibco phosphate-buffered saline (PBS) 10X solution (pH 7.4) from Thermo Scientific (Waltham, MA) was diluted to 1X using deionized water produced by an Elga PURELAB flex water purification system (18.2 MΩ·cm resistivity) from Veolia Water Technologies (High Wycombe, England). Glucose oxidase was obtained from Sigma-Aldrich (St. Louis, MO) and employed as the molecular probe to measure the interstitial fraction of the columns. Lyophilized exosomes derived from the urine of healthy donors (HansaBioMed Life Sciences, Tallinn, Estonia) were rehydrated following the manufacturer’s instructions using deionized water, yielding a final concentration of ~ 6.4 × 10¹¹ particles mL⁻¹. These exosomes were used as “standards” for quantification of the EVs present in the primary human urine sample in the study, with the understanding that they are not certified reference materials and lack detailed information on purity or classification. For the human urine matrix, first-morning urine was collected from a healthy, consenting donor, as it is known to contain a higher concentration of EVs in comparison to random spot urine [32]. Following collection, the urine was filtered through 0.22 μm polyether sulfone (PES) membrane syringe filters to remove macroscopic debris and impurities. The filtered urine was then stored in a cold room to prevent microbial growth during the study. No further processing steps were applied to the urine samples prior to loading onto the columns.

2.2. Construction and evaluation of C–CP fiber columns

The assembly of polyester (PET) capillary-channeled polymer (C–CP) fiber columns was performed using melt-extruded fibers obtained from Universal Fibers (Bristol, VA). The extrusion process utilized spinnerets, producing yarns with 28 fibers for the eight-channel configuration and 33 fibers for the trilobal configuration. Column packing followed the previously described protocol [33,34], wherein fibers were wound onto a spool mounted on an axle assembly equipped with a rotary counter to track the number of rotations. Once the desired number of fibers (equivalent to twice the number of rotations) was achieved, the fibers were removed from the spool, looped with a monofilament (8 lb-test), and pulled through polyether ether ketone (PEEK) tubing (0.76 mm internal diameter, Cole-Parmer). All columns were packed to a standard length of 300 mm. Variation of the number of fibers packed into the tubing allows the variation of packing density/interstitial fraction of the columns. A total of six PET-8 and five PET-Y columns with various packing densities were prepared. Following packing, each column was flushed sequentially with deionized water (DI-H₂O), acetonitrile (ACN), and DI-H₂O again at a flow rate of 0.5 mL min⁻¹ until a stable absorbance baseline was obtained using UV–Vis detection at 216 nm. Cleaned columns were stored at ambient conditions for subsequent use.

To determine the fiber mass in each column, the corresponding fiber loop was weighed using a precision balance before being pulled through the PEEK tubing. Each mass measurement was performed in triplicate to ensure accuracy and reproducibility. The fiber packing density (ρ), expressed in g cm⁻³, was calculated by dividing the fiber mass (m) by the total column volume (V). The maximum achievable packing density was limited by the tensile strength of the fibers and the monofilament, as excess friction with the column walls caused the fibers to tear. The minimum packing density ensured sufficient surface/interfiber friction to prevent fiber displacement during mobile solvent flow. Table 1 summarizes the physical characteristics of the columns packed, including the number of fibers, the corresponding fiber mass, packing densities, and equivalent fiber surface areas.

Table 1.

Physical characteristics of PET-8 and PET-Y C–CP fiber columns.

Fiber type	Number of rotations on rotary counter	Number of fibers	Mass of fiber ± SD (mg)	Fiber packing density (g cm−3)	Total fiber surface area (μm2) × 1010
PET-8	5	280	51.7 ± 0.45	0.38	2.0
	6	336	63.1 ± 0.26	0.46	2.4
	7	392	74.5 ± 0.43	0.54	2.8
	8	448	84.8 ± 0.16	0.62	3.2
	9	504	93.6 ± 0.31	0.69	3.6
	10	560	106.2 ± 0.2	0.78	4.0
PET-Y	4	264	58.5 ± 0.17	0.43	1.1
	5	330	72.5 ± 0.12	0.53	1.3
	6	396	85.0 ± 0.43	0.63	1.6
	7	462	102.8 ± 0.05	0.75	1.9
	8	528	116.8 ± 0.19	0.86	2.2

Fiber morphology, packing uniformity, and the verification of EVs captured on the fiber surface were assessed via scanning electron microscopy (SEM) using a Hitachi Regulus 8230 (High Technologies America, Inc., USA) instrument. Imaging was performed with an accelerating voltage of 10 kV, with representative images presented in Fig. 1. For SEM imaging of column cross-sections (Figs. 1a– d), the columns were frozen using liquid nitrogen, sectioned into 1.5 cm segments, affixed to the SEM holder using double-sided tape, and sputter-coated with platinum at 80 mTorr of argon for ~3 min using a Hummer 6.2 Sputtering System (Anatech USA, Union City, CA). To verify EV capture on the fiber surface (Fig. 1e), a PET-Y column was loaded with urine-derived EVs, and the process was stopped following the rinsing step, but prior to the elution event (depicted in Fig. 2). The fibers were then extracted from the tubing, affixed to the SEM holder, and sputter-coated to image EVs. ImageJ software was employed to measure the perimeter of the different single fiber geometries.

Fig. 2.

Elution program and temporal responses of EV load and elution experiments depicting the quantitative points of measure of dynamic binding capacity of EVs.

Columns with varying interstitial fractions (ε_i) were achieved by varying the number of fibers packed into the PEEK tubing as described above. The ε_i for each column was determined experimentally using glucose oxidase as a probe species and calculated using the elution time of an injection of glucose oxidase under non-retaining conditions, as defined by Eq. 1:

$${\epsilon }_i=\frac{F\left(t_R-t_0\right)}{v_c}$$ (1)

where, F represents the mobile phase flow rate, t_R is the elution time for glucose oxidase injection in the column, t₀ is the elution time for injection without the column, and v_C is the volume of the empty column.

To calculate the column permeability, the viscosity of the EV loading medium (1:1 mixture 2 M ammonium sulfate and 40 % acetonitrile in PBS) was determined using an Anton Paar MCR 302e rheometer.

2.3. Chromatographic measurements and EV quantification –

Chromatographic analyses were conducted using a Dionex Ultimate 3000 HPLC system equipped with an LPG-3400SD quaternary pump and an MWD-3000 UV–Vis absorbance detector (Thermo Fisher Scientific, Sunnyvale, CA, USA), operated via Chromeleon 7 software. Detection was primarily performed at 216 nm, with additional wavelengths (280, 203, and 254 nm) monitored based on their established relevance for EV detection [19,20].

The quantification of EVs in human urine samples was performed using a standard addition method previously validated for complex biological matrices, including urine [22,23]. Known volumes and concentrations of EV standards (utilizing the stock solution concentration of 6.4 × 10¹¹ particles mL⁻¹) were spiked into a urine sample matrix diluted 1:100 in 1× PBS. Three 40 μL sample aliquots were spiked once, twice, and three times to achieve EV concentrations ranging from 1.28 × 10¹⁰ to 3.84 × 10¹⁰ particles mL⁻¹. The final volume was adjusted to 300 μL with 1× PBS, and absorbance measurements at 216 nm were obtained using 100 μL injection at the column bypass position on the HPLC platform (n = 3). Linear regression of the data yielded an R² value of 0.9732 and extrapolated to estimate an EV concentration of 5.28 ± 0.02 × 10¹⁰ particles mL⁻¹, representing an EV concentration of 5.28 ± 0.02 × 10¹² in the primary human urine sample.

2.4. Dynamic binding capacity and hydrophobic interaction chromatography (HIC) conditions –

Frontal loading and elution recovery experiments were performed under hydrophobic interaction chromatography (HIC) conditions to evaluate the dynamic binding capacity towards EVs, as illustrated in Fig. 2. However, given that human urine was used as the loading sample rather than purified EVs, the continuous presence of background species (e.g., small molecules, proteins) that absorb at the monitoring wavelength renders a direct breakthrough curve unsuitable. A previously established and optimized method that allows for selective EV retention while permitting those other matrix components to pass through unretained was employed [23]. To complement the frontal analysis, a recovery assessment was performed to characterize the efficacy of the entire process.

Buffer A, comprising 2.0 M (NH₄)₂SO₄ in 1× PBS, was used in the loading buffer, while buffer B, containing 40 % v/v ACN in 1× PBS, served as the EV elution buffer. During the sample loading phase (t = 5 min to t = 10 min), a 50:50 mixture of buffer A and buffer B (resulting in 1.0 M (NH₄)₂SO₄ and 20 % v/v ACN in PBS) was introduced to alleviate the retention of salts, small organic molecules, and proteins while promoting EV retention on the column. As the sample passes over the fiber surfaces, EVs are progressively adsorbed until the binding sites are saturated, at which point unbound EVs appear in the column effluent, resulting in a notable increase in optical absorbance (t ≈ 6 min in this example). The sample loading step was followed by a 10-minute rinse (t = 10 min to t = 20 min) with the same buffer mixture to further reduce protein carryover, as evidenced by the absence of significant protein peaks in the chromatograms. EV elution was initiated at t = 20 min with 100 % of buffer B applied for 5 min, to ensure complete recovery of retained EVs and a return to baseline absorbance levels. Column re-equilibration was achieved through two 5-minute cycles of the 50:50 buffer mixture performed both before and after each experiment. The total test gradient program spanned approximately 30 min, balancing operational efficiency with effective EV separation. Further refinements in the various wash elution steps could be affected to appreciably enhance throughput and scalability for broader applications.

The loading process and DBC of EVs on the PET C–CP fiber column were assessed through breakthrough curves and frontal analysis. A plateau in the absorbance response signifies surface saturation, as illustrated in Fig. 2. The methodologies for evaluating dynamic binding capacity from chromatographic data have been described previously [23,35]. The DBC value was determined based on the EV concentration (2.64 × 10¹² particles mL⁻¹), flow rate (0.5 mL min⁻¹), and loading time (t min). The loading time, t, was defined as the point at which absorbance reached at 50 % of the feed concentration (DBC₅₀), measured at 216 nm. Although dynamic binding capacity at 10 % breakthrough (DBC₁₀) is commonly reported, assessing DBC at 50 % breakthrough (DBC₅₀) offers a more representative basis for comparison across columns with varying packing densities, particularly when accounting for differences in mass transfer properties and non-ideal breakthrough behavior. The transient shape and integrated peak areas of the elution step were used to assess any variations in the process recoveries.

3. Results and discussion

The C–CP fiber stationary phases have demonstrated superior efficiency and purity in EV isolation in direct comparison to conventional methods using the spin-down tip format [21]. Key advantages include reduced processing time (<15 min), enhanced workflow efficiency, preservation of EV integrity as confirmed by immunoassays for surface proteins (CD9, CD81), and significantly higher yields [21,22]. These qualities extend as well to the higher-volume column format processing. Additionally, C–CP fibers offer a cost advantage, being over two orders of magnitude less expensive than traditional chromatographic media relative to their capture capacity. Therefore, the premise for the present effort is to set the stage for column transformation to larger-scale formats. Towards that goal, it is critical to develop a deeper understanding of the factors influencing EV capture and processing efficiency. This study systematically investigates the relationship between fiber packing density, affecting both the interstitial fraction and available surface area, and the dynamic binding capacity (DBC) of EVs in microbore C–CP fiber columns for the two C–CP fiber geometries (PET-8 and PET-Y), providing insights relevant to scale-up for preparative applications.

3.1. Interstitial fraction determinations –

The interstitial fraction (ε_i) of the column bed is a key physical parameter in chromatographic performance as it influences hydrodynamic properties and dictates the available solid phase surface area for solute adsorption. In general, ε_i is defined as the ratio of mobile phase volume to total column volume (ε_i = V^m/V_c), inversely correlating with the fiber surface area. While higher fiber packing densities increase the potential capture surface area, overpacking can cause crimping or closure of inter-fiber channels, reducing accessibility to certain regions along the column. Additionally, ε_i impacts separation operations by influencing backpressure, diffusion path lengths, and linear velocities of the mobile phase. High ε_i values reduce backpressure at a given flow rate, whereas low ε_i values shorten diffusion path lengths which should improve solvent-to-surface mass transport. In addition, low ε_i leads to increased mobile phase linear velocities at the same volume flow rates, which affect solute residence times. Therefore, optimizing fiber packing density to balance these hydrodynamic and kinetic factors is crucial for maximizing the DBC.

To evaluate the impact of ε_i on EV DBC, columns of various fiber packing densities were prepared, as presented in Table 1. Six columns containing PET fibers with eight-channel geometries (PET-8) were packed with fiber counts ranging from 280 to 560, while five columns with trilobal fiber geometries (PET-Y) were packed with fiber counts ranging from 268 to 568. Although uracil (MW = 112 g mol⁻¹) is commonly used as a probe for ε_i determinations, it has been shown to overestimate ε_i in the case of separations on C–CP fibers as that ‘small’ molecule penetrates the 4 nm-radius pores of the fibers, whereas macromolecules and vesicles do not [26]. Therefore, glucose oxidase (MW = 160,000 g mol⁻¹) was selected as a non-retaining probe in this study using a 90:10 ACN:H₂O mobile phase to provide a more practical estimation of ε_i, consistent with previous investigations [35].

As shown in Fig. 3, the determined ε_i values for PET-8 columns decreased linearly from 0.77 to 0.49 as fiber count increased from 280 to 560. Similarly, the ε_i values for PET-Y columns decreased from 0.69 to 0.34 as fiber count increased from 264 to 528. These linear trends confirm that ε_i is directly influenced by packing density and validate the robustness of the experimental approach. Most importantly, the linear trends observed, with the data representing triplicate column preparations, suggest that the column assembly approach is reproducible and does not appear to impart appreciable non-ideal hydrodynamic effects (e.g., crimping, kinking, etc.) across this range of fiber packing densities. The parallel nature of the trends suggest that there are no shape-based biases regarding how the columns respond to changes in packing density.

Fig. 3.

Interstitial fractions of C–CP fiber columns used in this study as measured by glucose oxidase probe species.

Comparative analysis of the two fiber geometries reveals that, for the same number of fibers, the PET-8 columns exhibit higher ε_i values than PET-Y columns, indicating greater mobile phase solvent access to PET-8 surfaces. This can be attributed to the higher unit surface area of PET-8 fibers relative to PET-Y fibers. Table 1 summarizes the calculated fiber surface areas for the packed columns, showing that PET-8 provides approximately twice the surface area as PET-Y for the same fiber count. This situation is apparent in the SEM cross-sectional images in Fig. 1, where both fiber types are prepared at the same interstitial fractions (~0.61), composed of 448 PET-8 fibers and 330 PET-Y. It should be noted that the packing observed in the SEM micrographs is not likely true representation of the case where the columns are pressurized with mobile phase, at which point the fibers likely separate to fill the column more uniformly. The PET-8 micrograph suggests greater levels of interfiber digitation, albeit with greater variability across the cross-section. Indeed, previous comparisons between the two fiber types reflected the fact that there is greater packing uniformity for the Y-shaped fibers as judged by their respective van Deemter A-terms [29].

3.2. Column permeability characteristics –

One of the most practical considerations in the development of preparative-scale separations is the required pumping capacity (pressure) to affect an efficient separation [36]. A primary advantage of C–CP fiber-based columns is their ability to operate at high mobile phase linear velocities (~100 mm s⁻¹) while yielding high-fidelity protein separations [25]. As illustrated above, an increase in fiber count within the column, corresponds to higher packing density and a reduced interstitial fraction, thus decreased permeability and the required amount of operating pressures. In general, columns with higher permeability offer several advantages, including increased throughput, shorter analysis times, and reduced susceptibility to clogging, particularly when processing viscous or particle/protein-laden samples [36].

To elucidate the pressure-flow relationships in columns packed with different fiber geometries and interstitial fractions, column backpressures were measured across flow rates ranging from 0.1 to 0.9 mL min⁻¹ using the typical EV loading medium, composed of 1 M ammonium sulfate and 20 % ACN in PBS. As illustrated in Fig. 4, the pressure-flow responses remain linear across all columns with increasing volumetric flow rate, indicating the physical stability of the packed beds, with no evidence of bed structure perturbations, with Darcy’s Law being obeyed [37]. The linearity of these responses further validates the structural integrity of the columns, as any compression or deformation of the packing material, common in some polymer bead-based systems [36], would result in non-linear pressure-flow behavior. This uniformity of the responses portends well in terms of applying standard scaling laws towards preparative column applications [36].

Fig. 4.

Column back pressures measured at various mobile phase flow rates (0.1 – 0.9 mL min⁻¹) using the EV loading medium consisting of 1 M ammonium sulfate and 20 % acetonitrile (ACN) in PBS for a) PET-8 columns, and b) PET-Y columns.

As suggested in the SEM micrographs of the column cross sections of Fig. 1, the uniformity of column packing surely would impact the column permeability. Comparatively, the PET-Y fiber-packed columns exhibited slightly lower backpressures than PET-8 fiber-packed columns at similar interstitial fractions. For instance, a PET-8 column with an interstitial fraction of 0.55 exhibited backpressures of 8.6, 14.2, and 19.6 psi cm⁻¹ (equivalent to 59.3, 97.9, and 135.1 kPa cm⁻¹) at flow rates of 0.3, 0.5, and 0.7 mL min⁻¹ (corresponding to approximately 7,214, 12,024, and 16,833 cm h⁻¹), respectively. In contrast, a PET-Y column with a comparable interstitial fraction (0.53) displayed backpressures of 5.4, 9.8, and 15 psi cm⁻¹ (37.2, 67.6, and 103.4 kPa cm⁻¹) under the same flow conditions, respectively.

Column permeabilities were calculated according to Darcy’s equation [37].

$$B_0=\frac{F\eta L}{A\Delta P}$$ (2)

where B₀ represents the column permeability, F is the volume flow rate, η is the viscosity of the mobile phase, L is the column length, A is the cross-sectional area of the column, and ΔP is the column backpressure. For the aforementioned columns, the calculated specific permeability is 3.14 × 10⁻¹² m² for PET-8 (ε_i = 0.55) and 4.44 × 10⁻¹² m² for PET-Y (ε_i = 0.53), indicating higher permeability for the trilobal fiber geometry at similar interstitial fraction. In general PET-Y is found to have a superior specific permeability than the PET-8 at similar interstitial fraction columns, due to lower packing density and probable better packing homogeneity. This observation is consistent with the pressure-flow relationship (Fig. 4) and SEM images (Figs. 1a and b). A comparative analysis of specific permeability across an interstitial fraction range of ε_i = ~0.5 to ε_i = ~0.7, revealed that PET-8 columns exhibit permeability values between the range 1.9 × 10⁻¹² and 4.5 × 10⁻¹² m², whereas PET-Y columns demonstrate permeability values ranging from 4.4 × 10⁻¹² to 9.2 × 10⁻¹² m². First, the overall high permeability of columns underscores the high fluid transport efficiency of C–CP fibers, in agreement with previous studies conducted by this laboratory [29,33]. Additionally, the specific permeability values observed for the C–CP fiber columns evaluated in this study are significantly higher than those reported for polymeric monolithic columns. In comparison, the permeability values obtained in this work are one to two orders of magnitude higher than those reported for both microporous polymer monoliths (~10⁻¹³ m²) and highly cross-linked monolithic capillary columns (~10⁻¹⁴ m²) [38,39]. However, Qu et al. reported a permeability of 4.93 × 10⁻¹² m² for a bicontinuous polystyrene monolith [40]. Furthermore, comparing the two fiber types emphasizes the important role of fiber geometry and packing density in governing column hydrodynamics and overall chromatographic performance. Thus, advantages are seen for the PET-Y fibers with respect to chromatographic performance [29] and the practical hydrodynamics evaluated here.

3.3. Role of fiber packing density on EV dynamic binding capacity –

Dynamic binding capacity (DBC) is influenced by both thermodynamic and kinetic factors [37,41]. Thermodynamically, the equilibrium distribution of solutes between the mobile and stationary phases dictates binding efficiency, while kinetic factors, including solute transport and adsorption/desorption rates, determine the extent of surface coverage within a given separation time frame (i.e., residence time). With regards to protein processing on the C–CP fiber phases, with chromatographic integrity maintained up to linear velocities of up to 100 mm s⁻¹ [42], with DBC₅₀ values shown to be fairly invariant with flow rates/linear velocities of up to 5 mL min⁻¹/12 mm s⁻¹ [35]. These characteristics are a direct result of the lack of intra-fiber diffusion and well as convective flow mass transfer characteristics.

In this study, the human urine sample (diluted 1:1 with the loading medium resulting in a loading EV concentration of ~2.64 × 10¹² particles mL⁻¹) was loaded onto fiber-packed columns at a moderate flow rate of 0.5 mL min⁻¹ using the program depicted in Fig. 2, with the sample introduction begun at t = 5 min. Fig. 5 presents representative load/elute profiles for the PET-8 and PET-Y columns with various interstitial fractions. Each curve represents the average of triplicate load/elute experiments conducted for each column with different interstitial fractions to ensure data accuracy and reliability. The breakthrough curves in Fig. 5a and b reach a plateau at approximately 6 min, indicating column saturation with EVs, while the elution profiles between 20 and 25 min correspond to the desorption and recovery of bound EVs. Qualitatively, the loading breakthrough curves exhibit uniformity across all of the columns, suggesting consistency in capture performance. However, differences in the elution profiles (i.e. temporal responses) are observed, with PET-Y fibers displaying more well-behaved (Gaussian) and consistent elution profiles than PET-8 fibers. These traits further suggest that the trilobal fiber geometry yields greater packing homogeneity in comparison to the eight-pronged design, as demonstrated in protein chromatograms [29].

Fig. 5.

Frontal load and recovery responses for the dynamic loading of EVs derived from human urine on a) PET-8 and b) PET-Y C–CP columns (0.76 mm i.d. × 300 mm) varying the fiber packing densities at the flow rate of 0.5 mL min⁻¹ (12,024 cm h⁻¹). Each curve is the average of triplicate load/elute experiments. Comparison of the transition zone profiles for the PET-8 and PET-Y columns at c) higher ε_i, and d) lower ε_i.

Quantitative analysis of EV loading, summarized in Table 2, reveals no clear trend between column interstitial fraction and the amount of EVs loaded at saturation for the PET-8 columns. Counterintuitively, the highest particle loading, 1.48 × 10¹² EVs, was observed in the column having the highest interstitial fraction (0.77). Conversely, the lowest particle loading, 1.32 × 10¹² EVs, was determined for the column having an intermediate-level interstitial fraction of 0.72. In practice, the difference in particle loading between minimum and maximum was only 12 %, demonstrating that a wide variation in packing density (ε_i = 0.49 – 0.77) has minimal effect on the EV loading for the PET-8 columns. Interestingly, the lowest particle load observed was not associated with the lowest interstitial fraction but rather occurred at an intermediate value (0.72), suggesting a lack of systematic correlation. As shown in Table 1, particle loading decreases to a minimum at ε_i = 0.72 and then increases again as the ε_i value decreases to 0.49. The absence of a consistent trend of EVs loaded within the ε_i range applied in this study may be attributed to the irregularity of the accessible surface area in PET-8 fiber packed columns; quite literally a greater amount of packing inhomogeneity. Of interest, the observed particle numbers (~10¹² EVs) align with previous studies on C–CP fiber-based columns [23,43].

Table 2.

Dynamic binding capacity of EVs for the columns as a function of interstitial fraction, illustrating the relationship between column packing characteristics and EV binding efficiency. (n = 3 columns of each packing density).

Fiber type	Number of fibers	Interstitial fraction (εi)	EVs loaded (Particles) ×1012 ± SD	DBC @50 % breakthrough (particles g−1 fiber) × 1013 ± SD	DBC @50 % breakthrough (particles m−2 fiber) × 1013 ± SD
PET-8	280	0.77 ± 0.07	1.48 ± 0.04	2.86 ± 0.08	7.3 ± 0.03
	336	0.72 ± 0.05	1.32 ± 0.03	2.09 ± 0.04	5.4 ± 0.02
	392	0.66 ± 0.07	1.33 ± 0.02	1.78 ± 0.03	4.7 ± 0.01
	448	0.61 ± 0.03	1.35 ± 0.01	1.60 ± 0.01	4.2 ± 0.01
	504	0.55 ± 0.02	1.44 ± 0.01	1.53 ± 0.01	3.9 ± 0.03
	560	0.49 ± 0.02	1.45 ± 0.01	1.36 ± 0.01	3.6 ± 0.05
PET-Y	264	0.69 ± 0.03	1.32 ± 0.02	2.25 ± 0.04	12.2 ± 0.02
	330	0.62 ± 0.02	1.33 ± 0.01	1.83 ± 0.01	9.9 ± 0.04
	396	0.53 ± 0.02	1.19 ± 0.01	1.40 ± 0.01	7.4 ± 0.01
	462	0.42 ± 0.03	1.13 ± 0.01	1.10 ± 0.01	6.0 ± 0.01
	528	0.34 ± 0.03	1.12 ± 0.02	0.95 ± 0.02	5.2 ± 0.02

A more effective assessment of column performance can be achieved by evaluating dynamic binding capacity (DBC) per unit fiber mass and total fiber surface area. Fig. 6a and b depict the determined DBC₅₀ per unit mass and surface area, respectively, as a function of column interstitial fraction, with the numerical data presented in Table 2. The highest DBC per unit fiber mass (2.86 × 10¹³ EVs g⁻¹ fiber at DBC₅₀, and 2.68 × 10¹³ EVs g⁻¹ at DBC₁₀) among the PET-8 columns was obtained in the column with the highest ε_i (0.77). The same column presents the maximum DBC per unit fiber surface area (7.3 × 10¹³ EVs m⁻² fiber at DBC₅₀, and 6.98 × 10¹³ EVs m⁻² fiber at DBC₁₀). Reducing the ε_i (i.e., increasing fiber count) led to a decrease in the mass and surface area-based loading capacities, suggesting that the presence of additional sorbent phase was not fully utilized. Very simply, higher interstitial fractions provided greater access to the available surface area, with the greater space between the fibers compensated by the basic hydrodynamic efficiencies of convective flow. Conversely, lower interstitial fractions result in more compact packing, perhaps leading to crimped (perhaps touching) channels that reduce fiber surface accessibility and, consequently, lower loading efficiencies.

Fig. 6.

Dynamic binding capacities (DBC) of EVs isolated from human urine as a function of interstitial fraction in PET C–CP fiber-packed columns. a) DBC₅₀ expressed per mass unit of fiber. b) DBC₅₀ expressed per unit fiber surface area.

A similar but more pronounced set of trends was observed in the loading characteristics of the PET-Y columns as depicted in Table 2 and Fig. 6. Higher ε_i values correlate with increased EV capture, whereas lower ε_i led to reduced loading capacity. The highest number of loaded particles, 1.33 × 10¹² EVs, was observed for the column with an ε_i of 0.69, while the lowest loaded particles, 1.12 × 10¹² EVs, was recorded for the column with an ε_i of 0.34. As with PET-8, a two-fold increase in packing density (ε_i = 0.69 to 0.34) had a minimal effect on the total number of EVs captured, yielding an 18 % difference between the highest and lowest loaded particle values for PET-Y. In terms of DBC₅₀ per unit fiber mass and surface area, PET-Y exhibited clear trends in both cases, where increasing ε_i resulted in higher DBC₅₀ values. The maximum DBC₅₀ per unit mass and surface area were observed at 2.25 × 10¹³ EVs g⁻¹ fiber (2.25 × 10¹³ EVs g⁻¹ fiber at DBC₁₀) and 1.22 × 10¹⁴ EVs m⁻² (9.72 × 10¹³ EVs m⁻² at DBC₁₀) fiber respectively, at ε_i = 0.69. The consistency of DBC₅₀ per unit fiber mass and surface area in PET-Y columns underscores the advantages of trilobal fibers in maintaining uniform packing characteristics. As shown in Fig. 6a, at ε_i values greater than ~0.53, PET-Y consistently demonstrated higher DBC₅₀ per unit fiber mass than PET-8 packed at the same interstitial fraction. At tighter packing (ε_i < ~0.6), PET-8 columns exhibited inconsistencies, with some cases showing higher DBC₅₀ per mass compared to PET-Y at the same ε_i. This can be attributed to the fact that tighter packing impedes flow paths, causing potential collapse of channels, which results irregular capture performance. In terms of surface utilization for EV capture, PET-Y fibers outperformed PET-8 fibers, as evidenced in Fig. 6b The DBC₅₀ of EVs per unit surface area for PET-Y columns was greater than two-fold that of PET-8, at similar fiber packing densities (i.e., ε_i). This superior performance may be attributed to the simpler geometry of the PET-Y fibers, which ensures better channel homogeneity and, consequently, improved analyte access for surface coverage. These findings further emphasize the superiority of PET-Y in terms of column permeability, packing efficiency, and EV capture performance. While it remains to be verified by electron or optical microscopy, use of a nominal exosome diameter of 100 nm and the above surface area-based loading capacities, suggests surface area utilization efficiencies of ~50 and 90 % for the two fiber types.

While adsorption kinetics (breakthrough profiles) remained relatively consistent across columns with varying packing densities, a closer examination of the breakthrough curve profiles at different interstitial fractions (ε_i) revealed distinct mass transport behaviors. Figs. 5c and 5d compare the transition zone profiles of loading phase for PET-8 and PET-Y columns at higher and lower ε_i, respectively. As shown in Fig. 5c, columns with lower packing densities (higher ε_i) exhibited slower adsorption kinetics versus the lower ε_i (Fig. 5d), as evidenced by more gradual transition regions in the breakthrough curves and a slower approach to saturation for the lower-density PET-8 (ε_i = 0.72) and PET-Y (ε_i = 0.69) columns. Non-idealities near saturation were more pronounced at lower packing densities, as expected. The onset of breakthrough for PET-8 (ε_i = 0.72) occurred slightly earlier than for PET-Y (ε_i = 0.69), however, the latter exhibited a steeper breakthrough profile and a shorter time to reach saturation, with a ~20 % higher slope value in the transition region, indicating faster adsorption kinetics. Conversely, at the lower ε_i, kinetic differences between the two fiber types were minimized, and adsorption kinetics were slightly increased. Fig. 5d illustrates the breakthrough curves for PET-8 and PET-Y at lower ε_i, where steeper breakthrough responses were observed in comparison to higher ε_i cases, though in this case the PET-8 shows an overall higher capacity. The quantitative slope difference at the transition region was only ~3.7 % between PET-8 (ε_i = 0.49) and PET-Y (ε_i = 0.42), suggesting that at lower interstitial fractions, diffusion constraints are more pronounced. These kinetic variations arise from differences in intra-fiber separation distances, which decrease as packing density increases. Consequently, columns with shorter interfiber diffusion distances exhibit sharper transient responses, closely resembling step-function behavior.

3.4. Role of loading flow rate on EV dynamic binding capacity –

Another important parameter to consider with respect to process throughput is the on-column loading time required for EV saturation. Notably, achieving microbore column saturation, with a binding capacity (~10¹³ EVs g⁻¹ fiber) sufficient for most any fundamental biochemical study, requires only ~1 min, demonstrating the rapid transport and adsorption kinetics of EVs onto C–CP fibers. This rapid capture rate highlights the superiority of C–CP fiber-based isolation over conventional methodologies, as achieving similar isolation metrics typically requires significantly longer processing times (e.g., ~4 h for ultracentrifugation and ~3 h for polymer precipitation) [44,45]. Although rapid EV isolation methods, including affinity-based capture, also exhibit fast kinetics, C- CP fibers emerge as a promising choice for large-scale production, offering a significant reduction in processing time while maintaining high EV throughput.

As the dynamic adsorption of EVs (or any solute for that matter) onto a sorbent surface is a kinetically limited process, analyzing breakthrough curves in relation to the loading velocity provides significant fundamental insights and projections to practical implementations. Characterization of the breakthrough curves and binding capacities as a function of solution flow rate allows for the assessment of potential kinetic limitations versus the on-column residence time; critical parameters in evaluating separation efficiency and process throughput. Fig. 7a and b present breakthrough profiles at varying volume flow rates with the equivalent linear velocities noted in the figure caption for the two columns. As highlighted in the previous section, the highest DBC₅₀ per unit mass and surface area were achieved at an ε_i of 0.77 for the PET-8 column and ε_i of 0.69 for the PET-Y column. Therefore, these two columns, identified as the best-performing in terms of DBC₅₀, were selected to study the impact of loading flow rate. Specifically, Fig. 7a depicts the response of the PET-8 column with an ε_i of 0.77, while Fig. 7b shows the response for the PET-Y column with an ε_i of 0.69. Notably, variations in flow rate/residence times had minimal impact on the C/C₀ = 0.5 inflection points, indicating that mass transfer limitations were negligible under these conditions. The volumes to reach breakthrough (ranging between 0.45 and 0.5 mL) remained consistent across different flow rates, suggesting that changes in linear velocity did not significantly influence mass transfer rates. For instance, across a 2.3-fold variation in volumetric flow rate (from 0.3 to 0.7 mL min⁻¹, corresponding to linear velocities of 5153 to 12,023 cm h⁻¹), the percentage standard deviation (%RSD) of loaded EVs was only 2.8 % among the PET-8 columns. A similar trend was observed for PET-Y columns, where the variability of the loaded EV values was 1.6 %RSD over the same flow rate range. Similar trends have been reported in protein adsorption studies using PET C–CP fiber phases [46], reinforcing the advantage of the fibers in minimizing mass transfer resistance in comparison to conventional porous stationary phases. Additionally, the breakthrough curves exhibit steep slopes near saturation, regardless of loading speed, indicating efficient utilization of the stationary phase surface. This uniform adsorption behavior further highlights the superior mass transfer efficiency of PET-based C–CP fibers, enabling high-capacity EV capture even at elevated flow rates.

Fig. 7.

Influences of mobile phase flow rates on the dynamic loading characteristics of EVs isolated from human urine. a) Frontal load responses at various loading flow rates for PET-8 columns packed at an ε_i of 0.77. b) Frontal load responses at various flow rates for PET-Y columns packed at an ε_i of 0.69. c) Effect of column residence time on the loading of EVs for the PET-8 and PET-Y columns.

From a process efficiency perspective, on-column residence time is a key factor in determining overall throughput. Fig. 7c illustrates the relationship between the amount of EVs loaded and column residence time, with volume flow rates fixed at 0.3, 0.5, and 0.7 mL min⁻¹. Residence time, which depends on the void volume, flow rate, and column length, can influence the interaction between EVs and the fiber surface. In theory, longer residence times may provide greater opportunities for adsorption onto the fiber surface. It is important to note that differences in interstitial fractions result in varying residence times at identical flow rates. For example, increasing the flow rate from 0.3 to 0.7 mL min⁻¹ reduces the residence time from 19.7 s to 8.45 s for the PET-8 column (ε_i = 0.77) and from 18.16 s to 7.78 s for the PET-Y column (ε_i = 0.69). As depicted in Fig. 7c, despite the variations in column residence time in the same column, the amount of captured EVs remained consistent, showing no significant increase or decrease with longer residence times, regardless of fiber geometry. From a practical standpoint, this observation is significant, as comparable loading capacities can be achieved within a shorter residence time; less than half the duration required by the lowest flow rate applied (0.3 mL min⁻¹). Notably, achieving a capture rate of approximately ~10¹² EVs within a 7–8 s residence time corresponds to a throughput of ~10¹³ EVs per minute, representing a substantial potential of C–CP fiber platforms over traditional isolation techniques in large scale production.

4. Conclusions

This study systematically investigated the influence of column interstitial fraction (fiber packing density) on the DBC of EVs using a microbore format C–CP fiber column (0.76 mm × 300 mm) within an HPLC platform. Frontal analysis was employed to quantitatively assess dynamic loading characteristics as a function of interstitial fraction and linear velocity. In general, columns with higher interstitial fraction (lower packing densities) resulted in higher EV binding capacities, where DBC₅₀ in both unit mass and unit surface area were taken as a metric of comparison. Although a consistent increase in DBC₅₀ was observed with decreasing fiber packing density, variations in flow rate had minimal impact on binding performance. Furthermore, the columns exhibited stable adsorption kinetics across different on-column residence times, irrespective of fiber geometry. Comparative analysis of PET C–CP fibers with eight-channel (PET-8) and trilobal (PET-Y) geometries highlighted the generally-superior performance of trilobal fibers, demonstrating enhanced column permeability, packing efficiency, EV capture capacity, and elution profile homogeneity.

The effectiveness of EV isolation and purification is ultimately determined by the ability to achieve the required vesicle concentration for downstream analytical applications. Notably, most RNA sequencing studies require 10⁹–10¹⁰ EVs for accurate profiling [47,48], while LC-MS-based proteomic analyses typically require 10¹⁰–10¹¹ EVs [49,50]. The findings of this study demonstrate that microbore-scale C–CP fiber-based columns significantly exceed these requirements (10¹³ EVs min⁻¹), highlighting their potential for high-yield EV enrichment. As a comparison, studies by Riekkola et al. [12] have described the use of antibody-labeled, commercial monolithic disk to isolate EVs from small volumes (250 μL) of plasma diluted to 5 mL of H₂O to yield EV number densities of ~10¹² particles in ~10 min, with high purity. More detailed, practical comparisons to phases such as these are certainly in order. From a preparative and industrial perspective, considerations such as throughput, yield, and operational efficiency are critical metrics in evaluating the scalability of C–CP fiber-based separation platforms. Given their high binding capacities, reproducibility, robust performance, and lack of further surface derivatization, C–CP fibers hold strong potential and justification for further development for large-scale EV purification. Future research will focus on scaling up column formats to increased EV capture capacities towards levels needed towards therapeutic vector production, leveraging the insights gained in this study to advance the application of C–CP fibers in high-throughput bioprocessing. The well-behaved hydrodynamic attributes of the columns demonstrated here suggest that these developments will likely following standard scaling laws [36,41,51], though practical considerations as to the final approach in terms of column proportions and methods of packing will surely require a good deal of effort.

Data Availability Statement

The data that supports the findings of this study are available from the corresponding authors upon reasonable request.