Polymers polyvinyl alcohol (PVA, 87%–90% hydrolyzed, Mw = 30,000–70,000, Sigma Aldrich), polyethylene oxide (PEO, Mw = 400,000, Scientific Polymer Products, Inc.,), polylactic-co-glycolic acid (PLGA, 50:50 L:G, carboxylate terminated, inherent viscosity = 0.55–0.75 dl/g, Lactel Absorbable Polymers), and poly-l-lactic acid (PLLA, ester terminated, inherent viscosity = 0.90–1.20 fl/g, Lactel Absorbable Polymers) were electrospun with either deionized water, hexofluoroisopropanol (HFIP), or a mixture of chloroform and 2,2,2-trifluoroethanol (2,2,2-TFE)/HFIP (Oakwood Chemical) and collected onto industrial brown waxed paper (24″ × 1,500′, 30 lb) as the electrospinning substrate. Antiretroviral (ARV) agents were purified as described in Jiang, et al. from tablets of Selzentry^® (300 mg, ViiV Healthcare), Insentress^® (400 mg, Merck), and Intelence^® (200 mg, Janssen) for maraviroc (MVC), raltegravir (RAL), and etravirine (ETR), respectively (Jiang et al., 2015). Sieved Tenofovir (TFV, Particle Sciences, gift from CONRAD) was used as obtained. Dimethyl sulfoxide (DMSO, BDH/VWR Analytical), acetonitrile (HPLC grade, Sigma-Aldrich), and potassium monophosphate (Fisher Chemical) were used for quantification.

Unless otherwise specified, PVA/PEO was formulated at 84:14 (w/w) PVA and PEO and 17.4% (w/v) in deionized water. Polymers were mixed using a stand mixer (Eurostar 20 digital, IKA) at 1,000 RPM for 10 min, then 100 RPM overnight. PLGA fibers were electrospun from a 15% (w/v) PLGA solution in HFIP and mixed with a stir bar overnight. For drug loaded formulations, each agent was homogenized (T25 digital Ultra-Turrax, IKA) into the solvent prior to addition of polymer, at the specified loading percentage.

All electrospinning was conducted using a Elmarco Nanospider (Liberec, Czech Republic) with an oscillating carriage electrospinning electrode. In this method, fibrous materials are generated from a polymer solution that is coated onto a wire electrode by an oscillating carriage. The carriage holds a reservoir filled with the polymer solution and coats the wire through an interchangeable orifice in which the wire passes through. Solid fibers are formed from this solution by an applied voltage gradient across the wire electrode and collection electrode. The collection electrode is covered by the collection substrate to allow for simple collection of the material. This substrate can undergo dynamic movement during the electrospinning process.

Unless specified, solutions were coated onto the wire electrode through a 0.7 mm orifice at a rate of 350 mm/s. Additionally, the collection electrode was maintained at a distance of 200 cm, and electrospun across a 100 kV voltage difference. Chamber humidity was controlled at approximately 10% using compressed air. Waxed paper was used as the collection substrate. Under static electrospinning conditions, the collection substrate remined stationary. Under dynamic electrospinning conditions, the collection substrate was translated at a rate of 155 mm/min. This movement oscillates over a defined distance, labeled to start (marker 1) and end (marker 2) directly above the wire electrospinning electrode, and continuously pass over this defined distance. A single pass is defined when the first marker labeled on the collection substrate and aligned directly above the wire electrode (Material direction (MD) axis = 0), is translated until the second marker, at specified dynamic path length (DPL) away from the first marker, is aligned with MD = 0. The next pass is performed with the reversed movement of the collection substrate until the first marker has returned to the starting position. Electrospinning was conducted until the center fiber mass reached a target basis weight of 100 gsm. This target was determined through periodic sampling of the center-most 4 × 4 cm fiber section, which was removed, weighed, and replaced for further electrospinning. Additional electrospinning time or pass number was then determined by linear interpolation.

Throughput was measured by the mass of the fibers collected on the pre-weighed substrate divided over time. Samples of the fiber mats were collected with a 1 cm arch punch tool (7.85 × 10⁻⁵ m²) at specified locations and weighed with a microbalance (Meddler Toledo). The value σ_s is determined from gaussian line fitting from n = 5 samples which were collected across the carriage direction (CD) averaged at both edges. To determine gaussian line fits across the material direction (MD) of the fiber mat, statically spun materials were sampled. Using these gaussian line fits, simulations were created to model dynamic electrospinning conditions, where the electrospinning substate translates to increase the collected surface area of fibers. In addition to gaussian line fit equations, simulations also accounted for the electrospinning substrate winding rate and approximate fiber deposition per pass. Simulations of mass deposition and correlations to measured mass deposition were conducted using MATLAB (MathWorks Inc, Natick, MA, USA, version R2019b).

Fiber samples were dissolved in DMSO at a concentration of 1 mg fiber/ml. Samples of the solutions were filtered with 0.22 µm syringe filters (Millex Durapore, Millipore) prior to quantification. High-performance liquid chromatography (HPLC, Shimadzu Prominence) was then used to measure the concentrations of the three drugs in tandem with a diode array detector (Shimadzu Prominence SPD-20A). Samples were analyzed at 35°C at a flow rate of 1 ml/min, and a gradient of 38%–73% acetonitrile (ACN) and 25 mM monopotassium phosphate in water (KH₂PO₄ buffer) was used as the mobile phase. Specifically, the mobile phases begin at a ratio of 62:38 KH₂PO₄ buffer/ACN, then ACN increases to 73% of the mobile phase over 8 min, and then is returned to 38% by the end of the 20 min sample run. A C18 column (5 μm, 100 Å, 250 × 4.6 mm, Phenomenex Kinetex) is used as the stationary phase. Detection wavelengths and approximate retention times were 193 nm and 4.3 min for MVC, 300 nm and 7.1 min for RAL, and 234 nm and 15.3 min for ETR.

Planning and analysis for the statistically designed optimization experiments was conducted using StatEase Design-Expert (Minneapolis, MN, USA, version 9.0.4.1). Linear and gaussian line fits, as well as determinations of statistical significance were made using GraphPad Prism (San Diego, CA, USA, version 9). Significant differences were determined using repeated measured two-way ANOVA, or repeated measures one-way ANOVA for overall drug loading comparisons, with Geisser-Greenhouse correction and with Tukey’s multiple comparisons test.

Free-surface electrospinning is amenable to scale-up, but the process has been used primarily in low-basis weight coatings for industrial filtration. Therefore, we first set out to optimize processing parameters for free-surface electrospinning that would achieve the high basis weight and uniformity required for pharmaceutical manufacturing of biomedical materials. Design of experiments (DOE) was used to efficiently identify process and solution parameters that would significantly affect fiber yield and distribution (Figure 1A). Factor inputs included polymer concentration (0.174–0.196 g/ml) and drug loading of a model agent tenofovir (0–28.57%, or 0–10 M). We also investigated the effect of four process variables: carriage speed coating the polymer solution onto the charged stationary wire (125–250 mm/s), distance between the charged electrospinning wire and oppositely charged electrospinning substrate (180–220 mm), applied electric field (0.35–0.45 kV/mm), and size of the orifice out of which the polymer solution flows onto the charged wire (0.6–0.8 mm). The specific polymer formulation used for parameter optimization was composed of PVA and PEO at a ratio of 84:14, which was determined independently via hill-climbing optimization (Skiena, 2008).

Electrospinning experiments with the specified factor input settings were performed on an Elmarco Nanospider (Figure 1B). The location of the collected material was defined with respect to the carriage direction (CD) and the material direction (MD) of the instrument. CD is defined as the axis parallel to the wire electrospinning electrode, where the carriage translates across to re-coat the wire with polymer electrospinning solutions (Figure 1B). MD is perpendicular to the CD-axis and defined as the direction in which the electrospinning substrate translates. We determined that a scalable method for producing medical materials will require dense and uniform accumulation of material within a reasonable time frame. As such, we focused on fiber throughput (units of g/m², or gsm) and fiber distribution measured along the CD-axis (σ_s, in cm) (Figures 1C,D).

Carriage speed, polymer concentration, orifice size, and electric field all significantly impact fiber throughput (Figure 2A; Supplementary Figure S1A). Indeed, low polymer concentration, high carriage speed, high orifice size and high electric field yielded the greatest throughput of 12.9 g/h. Fiber distribution was most significantly affected by carriage speed, electric field, and electrode distance (Figure 2B; Supplementary Figure S1B). With a high carriage speed, high electrode distance, and low electric field, the greatest σ_s value of 5.11 cm was achieved. Several two-factor interactions showed significant effects on the throughput and distribution response outputs (Supplementary Tables S1, S2). Throughput significantly increased with the interaction of increased drug loading and decreased polymer concentration (Figure 2C), decreased polymer concentration and a higher carriage speed (Figure 2D), as well as a higher carriage speed and larger orifice (Figure 2E). Several two-factor interactions were found to significantly effect σ_s (Supplementary Figure S1B; Supplementary Table S2). The σ_s value increased with interactions between low polymer concentration and high carriage speed (Figure 2F), as well as a decreased electric field and larger electrode distance (Figure 2G). We conclude that the optimal electrospinning parameters for our experiment would have a lower polymer concentration (0.174 g/ml), higher carriage speed (250 mm/s), an electrode distance of 200 mm, an applied 100 kV voltage difference, and an orifice size of 0.7 mm. Since the electric field had opposing impacts on throughput and distribution, we opted to use a higher electric field to prioritize productivity, and compensate for deposition consistency by controlling substrate movement.

Free-surface electrospinning supports greater scalability of fiber production due to the higher density of fiber jets and adaptability for roll-to-roll processing. We took advantage of the ability to control the movement of the collection substrate along the MD axis to obtain high-basis weight and uniform materials. By controlling the residence time and substrate movement relative to the electrospinning wire, we could precisely tune fiber collection in specific areas. Figures 3A,B illustrate the difference in fiber deposition under electrospinning conditions with a static or dynamic collection substrate, respectively. We hypothesized that longer residence times achieved by repeated “passes” of the collection substrate across a specified dynamic path length (DPL) could increase the basis weight or accumulation of fibers across a larger area. Here, we define the DPL as the distance of the collection substrate that repeatedly and directly passes over the wire electrospinning electrode (Figure 3B). In addition to the polymer blend previously optimized for fiber throughput, we assessed a polymer solution composed of 15% (w/v) PLGA in HFIP. We selected this material for the differential use of an organic solvent, common use of HFIP with electrospinning, and the common use of PLGA in biomedical applications. We expect the use of these different polymer solutions illustrates the resulting differences in electrospinning, as well as the general application of these simulations as a tool to account for differences in fiber deposition.

Electrospun fibers collect with variable deposition along the MD and away from the electrospinning electrode, as seen under static electrospinning conditions (Figure 3A). For a fixed position on the CD-axis, electrospun fibers were found to deposit with a Gaussian distribution across the MD-axis, centered at the wire electrospinning electrode (MD = 0 cm). Simulating material deposition can therefore inform necessary electrospinning parameters needed to achieve defined areas with uniform fiber basis weight. Therefore, the resulting fiber basis weight (A, in gsm) as a function of MD position (x_MD, in cm) under static conditions can be described by the following equation: A ( x M D ) = A T a r g e t e − 1 2 ( x M D − μ σ M D ) 2  

The maxima of fiber deposition is equivalent to the target fiber basis weight (A_target), which we set to 100 gsm, and controlled for during electrospinning by periodic sampling of the centermost 4 × 4 cm square of material. Due to this method for controlling material deposition, samples with coordinates at the edge of this testing region (MD = ±2 cm, CD = ±2 cm) could not be accurately collected. We additionally maintained center coordinates during dynamic electrospinning such that the MD position of the mean basis weight (µ) is equal to zero. Most interestingly, we found that the standard deviation of this Gaussian fiber-mass distribution (σ _MD, in cm) – or the spread of fiber deposition across the MD-axis–varies based on the polymer formulation. Under static electrospinning conditions, the observed spread of fiber deposition was found to be smaller, or narrower, for PVA/PEO (σ _MD = 3.787 cm) than PLGA (σ_MD = 6.443 cm) (Figure 3C; Supplementary Table S3).

It might be hypothesized that a formulation with a larger σ _MD value, and therefore wider fiber distribution, may yield fiber mats with a larger region of uniform fiber deposition at the target basis weight under dynamic electrospinning conditions. However, fiber deposition simulations do not reproduce this expectation. Simulated data was produced using the equation below, where p is the pass number of the substrate above the wire electrode, DPL is defined as the difference of the starting position (l_p,st) and the ending position (l_p,ed) centered at MD = 0, r_sub is the rate of substrate movement (155 mm/min), t_stat is the time to 100 gsm determined from the static run, and k is the step in MD position away from the center of the wire electrode. With a shorter DPL such as 12 cm, the PVA/PEO formulation with the lower σ_MD indeed yields a fiber sheet with a smaller uniform area than the PLGA formulation (Figure 3D). However, DPLs approximately 16 cm and larger can be seen to have a larger uniform area for formulations with a lower σ _MD compared to higher σ _MD values (Figures 3E,F). This is because the target 100 gsm basis weight is controlled for in all materials at the center of the mat. Therefore, as the path becomes increasingly large, formulations with larger σ _MD values achieve the target basis weight at the center with fewer passes, and have a shorter electrospinning time for fibers to accumulate. max p = 1 { p × ∑ k = l p , s t l p ,   e d ( 1 r s u b t s t a t ) A T a r g e t e − 1 2 ( x M D − k σ M D ) 2 } ≤ 100   g s m

Differences in fiber distribution were also observed between different polyester formulations. Using another polyester, poly L-lactic acid (PLLA), we electrospun solutions with different polymer concentrations and organic solvent properties. PLLA electrospun under 16 cm dynamic run conditions differs in its fiber distribution (σ _MD = 8.813 cm) compared to PLGA spun under similar conditions (σ _MD = 9.277 cm) (Supplementary Figures S2A, S2B). Though both are polyesters, PLLA has a slightly narrower distribution, confirming our findings that polymer type influences fiber uniformity. Moreover, in agreement with our DOE, PLLA fiber uniformity, albeit in the MD direction rather than CD direction, is significantly affected by the incorporation of TFE as a cosolvent compared to HFIP (Supplementary Figures S2B vs. S2C) and by lowering the polymer concentration in solution from 15% w/v to 10% w/v (Supplementary Figures S2B vs. S2D). This potentiates the results of our DOE to inform scale-up in dynamic run settings for a variety of polymer types and illustrates the importance of compatible solvent selection.

Simulations proved to be a useful tool for predicting fiber deposition patterns of different polymer formulas, and especially the region in which mass deposition plateaus. We found experimental data to be highly correlated to simulated dynamic run data, with high correlation coefficient (r) (Supplementary Table S3). Additionally, in plotting experimental data versus the simulated data (Supplementary Figure S3), we found the relationship to be highly linear with coefficient of variation values also near to 1 (Supplementary Table S3). Quantile-quantile (QQ) plots also indicate that the distributions of simulated and experimental data are comparable, as seen by linear relationships of plotted quantile values individually comparing PVA/PEO and PLGA formulations under 12 or 16 cm dynamic run conditions (Figures 3G,H).

While we conclude these simple simulations can be predictive of actual material deposition, some discrepancies do exist between the theoretical and observed data. These differences may be due in part to the manual method of controlling electrospinning time to achieve the target basis weight, or from the underlying inconsistency of electrospun fiber deposition in this process format (Collins et al., 2012; Tuin et al., 2016). Root mean square error (RSME) measurements between the experimental and simulated results show that the error is relatively uniform between the tested formulations and dynamic path lengths, except for 16 cm dynamic run PLGA results which have comparatively lower error (Supplementary Table S3). Despite some error, formulation-specific simulations are a valuable tool to determine the required electrospinning parameters needed to achieve the desired area of material at the target basis weight. The methods are especially informative in ensuring sufficient production of drug dosages or devices, while minimizing material waste.

While fibers deposit with gaussian distribution across the MD axis, centered at the wire electrode, the same pattern is not observed across the CD axis. Fibers are generated across the length of the wire (Figure 1B), and therefore the deposition is relatively linear across this length when measured at fixed MD positions (Figure 4). Further, the linear deposition has a near-zero slope. For all linear line fits of mass deposition versus CD position at fixed MD positions, slopes do not significantly deviate from zero with the exception at MD = −4 cm for the PVA/PEO 16 cm dynamic run and for MD = −6 cm for the PLGA static run. Inconsistencies in this linear trend are especially observed for MD cross-sections furthest from the wire, as seen by low basis weight values for samples collected for MD = 6 or 8 cm at CD = 0 for PVA/PEO 16 cm dynamic run (Figure 4B). This is likely due to sampling outside of the uniform area, where mass deposition dramatically decreases (Figure 3E). Plots of fiber surface density across the CD-axis further illustrate the differences in spread between the two formulations under static conditions, as seen by the differences in the spread of y-intercept values (Figure 4A,C). Improved consistency under dynamic electrospinning conditions can also be observed as measured by the reduction in y-intercept range in comparison to statically electrospun materials (Figures 4A–D).

Pharmaceutical solid dosage-formulation and agent delivery is one promising application for electrospun biomaterials. Scalable methods of electrospinning would be required for the generation of multiple dosages across a single fiber mat. Furthermore, materials with high basis weight could deliver larger dosages within a smaller material area. We therefore sought to assess the uniformity and encapsulation efficiency of physicochemically diverse pharmaceutical agents within these high basis weight materials. Here, we use three different pharmaceutical agents as model drugs: maraviroc (MVC), raltegravir (RAL), and etravirine (ETR). These drugs are all antiretroviral agents, but have differences in water-solubility ranging from 0.011 mg/ml (National Center for Biotechnology Information, 2022b), 53.9 mg/ml (National Center for Biotechnology Information, 2022b), and 0.00397 mg/ml (John and Liang, 2014) for MVC, RAL, and ETR, respectively. These drugs also have a range of reported partition coefficients (logP) of 4.3 for MVC (National Center for Biotechnology Information, 2022a), 0.4 for RAL (National Center for Biotechnology Information, 2022b), and >5 for ETR (John and Liang, 2014). Therefore, we expect these agents to represent an interesting challenge for formulation consistency. We prioritized using the optimized PVA/PEO formulation, which are common excipients in many drug delivery systems and where drug dissolution is nearly instantaneous following contact with water. Each drug was loaded into the polymer solution at 5% (w/w), representing a total 15% (w/w) theoretical drug amount within the fibers. Since the target basis weight is 100 gsm, full encapsulation of drug is theoretically equivalent to 5 gsm for each ARV drug. Measurements of percent drug loading approximates drug basis weight but is not necessarily equivalent. We hypothesized that dynamic, free-surface electrospinning could yield multi-drug loaded materials that had uniform basis weight across a defined target area, determined through simulation, and therefore uniform drug basis weight from a single batch process.

Using simulated data, we selected a DPL of 20 cm to create a minimum, proof-of-concept 4 cm MD by 20 cm CD area of fiber uniformity (Figure 5A). The measured mass distribution across multiple replicate fiber sheets correlates to simulated 20 cm dynamic run results (Run 1: Pearson r = 0.976; Run 2: r = 0.981; Run 3: r = 0.971) (Figure 5A). This electrospinning method also proved to be capable of efficiently formulating the three agents into a solid dosage-form, with an average loading efficiency across all measured samples of 82% for MVC, 71% for RAL, and 97% for ETR (Figure 5B). Although highly efficient, specific drug loading is significantly different for the three agents. These differences indicate that hydrophobic drugs formulate more efficiently in this particular formulation. In the predicted uniform area, there is also no significant variation in fiber mass or specific drug loading dependent on location [ex. Center (MD = 0, CD = 0 cm) versus corner (MD = 2, CD = 10 cm)], as determined by comparing averaged replicate values of coordinate defined replicate samples (all p > 0.16). This can also be visually observed as even fiber deposition across this center region of the fiber mat (Supplementary Figure S4). Therefore, dynamic free-surface electrospinning can produce uniform drug-loaded materials at the target 100 gsm basis weight.

Representative maps of fiber and drug basis weight distribution are shown for a region encompassing the defined target area and surrounding region within 10% of the simulated target 100 gsm basis weight (Figures 5C–F, all distribution maps are shown in Supplementary Figure S5). Some differences in fiber basis weight and drug loading can be observed for each measurement, however these maps show an overall homogeneity in fiber mass and in each drug across each fiber sheet (Figures 5C–F). While some regions of low drug basis weight may be due to overall reduced mass (Supplementary Figures S5A–D), mass is not the only driver of drug variation. In some locations, lower drug loading appears to be independent of overall mass deposition (Supplementary Figures S5E–F). Further, some maps indicate lower drug loading across all samples from sheet (Supplementary Figure S5D). However, all measured differences are minimal as determined by a coefficient of variation (CV) of less than 10% for all measurements in the representative fiber mat, and less than 15% for drug measurements in all other replicates (Table 1). These data show the capabilities of the electrospinning process to generate multiple, uniform and high basis weight solid dosage-forms from a single process.

As a proof-of-concept approach to create fiber formulations for sustained drug release, we electrospun MVC, RAL, and ETR at 5% (w/w) each in PLGA. In agreement with simulated and experimental data of PVA/PEO fibers, we used a 20 cm dynamic run to generate a fiber mat with a minimum 4 cm MD by 20 cm CD area of uniform fiber surface density (Supplementary Figure S6A). Unlike the PVA/PEO formulations, polyester formulations were observed to undergo throughput decay, potentially due to the evaporation of solvent in the solution reservoir, known as solution aging. Polyester fibers were also observed to accumulate on the electrospinning carriage during the run, which impeded the ability to reach 100 gsm. As such, we chose 50 gsm as the target basis weight and posit that two mats of controlled surface density could be layered to achieve a higher 100 gsm target. We found that the surface drug distribution of MVC, RAL, and ETR mirrored closely that of the fiber distribution in the MD direction (Supplementary Figure S6B). PLGA fiber uniformity was slightly offset from the 0 cm MD position, likely due to substrate alignment. All three drugs were encapsulated with high efficiency in this uniform region, with an average loading efficiency of 95%, 85%, and 99% for MVC, RAL, and ETR, respectively (Supplementary Figure S6C). This preliminary data on PLGA demonstrates a strong premise for further optimization of high basis weight, uniform, and sustained release solid dosage forms using this fabrication method.