J. of Supercritical Fluids 160 (2020) 104786 Contents lists available at ScienceDirect The Journal of Supercritical Fluids j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / s u p f l u Surface morphology and drug loading characterization of 3D-printed methacrylate-based polymer facilitated by supercritical carbon dioxide Truc T. Ngo a , ∗ , Lauren Hoffman a , Gordon D. Hoople b , William Trevena a , Udeema Shakya a , Gregory Barr a a Department of Industrial and Systems Engineering, Shiley-Marcos School of Engineering, University of San Diego, 5998 Alcala Park, San Diego, CA, USA b Department of Integrated Engineering, Shiley-Marcos School of Engineering, University of San Diego, 5998 Alcala Park, San Diego, CA, USA h i g h l i g h t s • Drug loading, surface roughness are tunable with 3DP and scCO 2 process parameters. • Thicker 3DP layer settings are more desired due to lower surface rough- ness and cost. • Higher scCO 2 process temperature results in higher drug loading but rougher surface. • Drug loading and surface roughness are modeled after 3DP and CO 2 pro- cess parameters. g r a p h i c a l a b s t r a c t a r t i c l e i n f o Article history: Received 14 November 2019 Received in revised form 5 February 2020 Accepted 10 February 2020 Available online 12 February 2020 Keywords: Polymer Supercritical carbon dioxide Flurbiprofen Drug loading Surface roughness a b s t r a c t Polymers have been shown to have viable applications in the biomedical field, from controlled drug delivery systems to biological implants. The preparation and processing of polymers into bio-systems have nevertheless encountered some technical challenges in part customization and adverse effects to the human body functions and recovery. This study proposes the utilization of 3D printing technology and supercritical carbon dioxide (scCO 2 ) processing to deliver a drug-impregnated polymeric material system which can be engineered and tuned to suit a particular implantation procedure and dramatically improve patient outcomes. In this work, an acrylate-based polymer is 3D-printed using stereolithography, then impregnated with flurbiprofen drug using scCO 2 Drug loading above 24 % by mass is achievable under the tested conditions. The correlation of drug loading and material surface roughness with different process parameters, including 3D printing layer thickness, scCO 2 processing temperature, pressure and treatment time, are investigated and empirically modeled using the linear regression methods. © 2020 Elsevier B.V. All rights reserved. ∗ Corresponding author. E-mail address: tngo@sandiego.edu (T.T. Ngo). 1. Introduction The last decade has seen rapid innovation in 3D printing tech- nologies [1–4]. Originally used primarily for rapid prototyping during the design process, the technology has evolved to be a key manufacturing tool across multiple industries [2]. There are now https://doi.org/10.1016/j.supflu.2020.104786 0896-8446/© 2020 Elsevier B.V. All rights reserved. 2 T.T. Ngo, L. Hoffman, G.D. Hoople et al. / J. of Supercritical Fluids 160 (2020) 104786 hundreds of different 3D printing approaches that can be used to print metals, plastics, ceramics, and even living biological speci- mens [2,4]. One major advantage of utilizing 3D printed parts is the ability to create geometries with sharp corners, overhangs, or undercuts that would be difficult to manufacture using tradi- tional approaches such as casting, machining or casting [4]. For many 3D printed parts with such geometry, it is often necessary to print support material to prevent the layers from collapsing as they are deposited. Polymers are particularly well suited to 3D printing due to their chemical makeup [4]. Thermoplastic poly- mers can be heated and extruded through a nozzle while thermoset polymers can be polymerized in a carefully controlled manner. Many polymer-based 3D printing approaches take a layer by layer approach where parts are created by slicing the geometry into thin (0.001 – 0.100 mm) layers. The part is then “printed” by stacking these layers one on top of the other. 3D printed materials often show strong anisotropic behavior due to this layer-based approach, an ongoing challenge for these materials [2]. 3D printing has emerged as a particularly promising technology for biomedical applications [4]. The nature of many implantable health interventions is that they must be customized to the indi- vidual patient and they often have complicated geometries. 3D printing is ideally suited to satisfy this need by allowing for the creation of customized implants with complex geometries on short time scales. For example, 3D printed dental implants have made it possible for dentists to create custom parts for their patients in a matter of days instead of weeks [4]. After implantation, patients typically receive pharmacological treatments to inhibit infections, improve the integration of the implant into surrounding tissue, or reduce postoperative discom- fort [5]. A recent trend in implant design is toward controlled drug release systems (CDRS) with active pharmaceutical ingredi- ents (APIs) imbedded directly into the implantable device, allowing some of the challenges and risks associated with oral delivery meth- ods to be avoided [5–8]. APIs are typically incorporated into the polymer filament prior to 3D printing when fused deposition modeling (FDM) is used in processing [9–11]. This is done mainly by impregnation, such as in the case of the loading of a polyvinyl alcohol (PVA) filament with the drug fluorescein by swelling the polymer in a solution of ethanol [12]. This process requires that the drug be thermostable so that it does not degrade during the 3D-printing process. In addition, impregnation is based on passive diffusion and requires a high con- centration of drug, making the process time-consuming and costly [13]. Alternatively, the incorporation of therapeutic drugs into 3D- printed polymeric materials after printing is accomplished using supercritical fluid technologies. The fluid is selected so that it has a high diffusivity in the chosen polymer and does not damage or degrade the polymer [14]. Drug impregnation using supercritical fluids has been largely studied for polymeric materials, but there is limited available research for subsequent impregnation of 3D- printed materials with APIs. Supercritical CO 2 (scCO 2 ) processing has emerged as a promis- ing approach for loading drugs into implants [5,15,16] due to several advantages. First, it is a low temperature process, suitable for thermosensitive drugs that would be damaged or destroyed in traditional, high temperature drug loading processes. CO 2 has a rel- atively low critical pressure and temperature (P c =73.8 bar, T c =304 K), allowing for a low energy transition to a supercritical region. Relatively high drug solubilities can be achieved in scCO 2 In the case of flurbiprofen, solubility of 0.022 g/L to 0.90 g/L in scCO 2 are possible without the use of co-solvents for temperature range of 303–323 K and pressure range of 88–245 bar [17,18]. Drug loading of polymers in scCO 2 is made possible by the CO 2 -induced swelling of polymer matrix and the favorable partitioning of drug molecules (i.e. solutes) in the polymer phase over the fluid phase [5,19,20]. The higher the partitioning coefficient (defined as the ratio of solute concentrations between the polymer phase and the fluid phase), the higher the drug loading. Drug impregnation yield can be con- trolled and optimized by manipulating operating temperature and pressure, making the process adaptable to different polymer/API systems and suitable for a variety of material applications [5,21,22]. Secondly, scCO 2 processing leaves the implant free of any resid- ual compounds that could potentially have a negative impact on patient health and safety [5]. Additionally, in comparison to traditional impregnation processes using organic solvents, scCO 2 processing has significant sustainability and processing advan- tages. Since the Industrial Evolution, 81 % of all greenhouse gas that humanity has produced has been CO 2 [5], and greenhouse gas emis- sions are projected to increase by 50 % by 2050 with a 70 % increase in CO 2 emissions alone [23]. CO 2 concentration in the atmosphere has been linked to an increase in global temperature and other significant environmental consequences [24,25]. The beneficial uti- lization of this abundant greenhouse gas not only reduces the amount of greenhouse gas emissions into the atmosphere, but also reduces the amount of waste output from a material treatment pro- cess. ScCO 2 processing can be conducted in closed loop systems, in which the output CO 2 gas stream from the process can be captured, scrubbed, filtered, pressurized and recycled back into the process for continuous usage. In combination with the financial benefits of recycling the processing medium, utilizing scCO 2 in drug loading eliminates the need for post-processing purification steps that are necessary when processing with conventional organic solvents [5]. This study focuses on examining the ways in which a 3D printed facsimile of poly(methyl methacrylate) (or PMMA) can be loaded with the anti-inflammatory drug flurbiprofen using supercritical carbon dioxide. Potential applications for the developed materi- als include targeted drug-controlled release systems and biological implants. PMMA was selected due to its biocompatibility, low cost, lack of toxic contents, advantageous mechanical properties and success as a drug carrier [26,27]. Various applications include dental prosthetics [28], contact and intraocular lens [29], verte- bral spacers [30], scaffolding in tissue engineering [31] and bone cement composite that acts as a space-filler to provide stabiliza- tion between bone and implant [32]. More recently PMMA has also shown promise as a carrier for therapeutic drugs [33]. Many studies have explored drug loading of PMMA with APIs for specific appli- cations such as intraocular lenses [34] and PMMA microspheres [35] as well as more basic science research about the behavior of a range of drugs when impregnated in PMMA [36,37]. However, no one has reported the impregnation of drug into 3D-printed PMMA for applications in the biomedical field, especially when supercrit- ical carbon dioxide is used as a transport medium during the drug loading process. 2. Materials and methods 2.1. Materials Polymer samples used in the experiments were prepared from a proprietary Clear Resin v4 , appearing as light yellow liquid resin, supplied by Formlabs, Inc (Massachusetts, USA). Clear Resin v4 con- tains a mixture of monomers, oligomers of methacrylic acid esters and photoinitiator, with boiling point and flash point > 373 K, den- sity of 1.09–1.12 g/cm 3 and viscosity of 850–900 cps. Flurbiprofen (C 15 H 13 FO 2 , CAS 5104-49-4) was purchased from Sigma-Aldrich (USA) in white powder form with a melting temperature range of 383–385 K, and was used without further purification. ACS reagent- grade liquid ethanol (C 2 H 6 O, CAS 64-17-5) was obtained from Arcos Organics (USA), 99.5 % purity, and was used as the solvent for flurbiprofen and drug loading determination. Isopropyl alcohol T.T. Ngo, L. Hoffman, G.D. Hoople et al. / J. of Supercritical Fluids 160 (2020) 104786 3 Fig. 1. SEM for non-treated polymer samples 3D-printed with different layer thicknesses. (C 3 H 8 O, 91 % pure, purchased from Target in San Diego, Califor- nia, USA) was used to wash 3D-printed samples post printing. USP medical grade carbon dioxide, purchased from Airgas (San Diego, California, USA) with > 99.9 % purity, was used in all scCO 2 treat- ments. 2.2. Sample preparation All polymer samples were printed using a FormLabs Form 2 3D printer. This printer created parts using a process known as stereolithography where a laser was used to crosslink polymer pre- cursors in a liquid bath. Polymer crosslinking happened on a flexible polydimethylsiloxane (PDMS) layer at the bottom of the resin tank located inside the Form 2 printer. The laser was able to penetrate through the optically clear membrane. In between printed lay- ers, the part was raised on a Z-stage and was briefly peeled away from the bottom PDMS window. The part was then lowered to 0.025, 0.050 or 0.100 mm away from the window depending on the desired layer thickness. The printing process repeated until the programmed number of layers were completed. In all cases parts were printed with “Clear Resin v4 ′′ which has material properties very similar to poly(methyl methacrylate) (or PMMA) with manufacturer-cited [38], post-cured tensile strength of 65 MPa, Young’s modulus of 2.8 GPa, flexural modulus of 2.2 GPa, and 6.2 % elongation at failure (compared to PMMA’s tensile strength of 63–78 MPa, Young’s modulus of 3.2–3.4 GPa, flexural modulus of 3.4–3.5 GPa, and 2–6% elongation [39]). Solidworks was used to create a computer-aided design (CAD) model of a thin rect- angular part with dimensions of 20 mm × 10 mm × 0.300 mm (L × W × H). The CAD file was exported as an. STL format and imported to PreForm, FormLabs 3D printing control software. There the part was angled to 30 ◦ relative to the build platform. Six supports points were manually selected – four at the corners and two at the center along the longest (20-mm) side – and then the PreForm software autogenerated the support structure design and incorporated into the build file. The part design was copied 27 times across the build platform to generate 28 identical parts in one single batch. Three different versions of the part were printed by varying the layer thickness in the print settings: 0.025 mm, 0.050 mm, and 0.100 mm. The 0.025-mm parts required 300 layers to print and took 3.8 h; the 0.050-mm parts required 182 layers to print and took 3.2 h; and the 0.100-mm parts required 123 layers to print and took 2.9 h. Note that two versions of the 0.100 mm part were printed. The two batches had slight variations in thickness and are therefore reported separately in Table 1. After printing was finished the parts were physically separated from the support structure on the build platform. They were then cleaned using the Form Wash tool (also purchased from FormLabs), which consisted of an ultra- sonic bath filled with 91 % isopropyl alcohol. Parts were submerged in this bath for ten minutes, then removed and allowed to air dry while still attached to the support material. After a minimum of 12 h drying the support material was removed using a small needle nose clipping tool. No UV post cure was applied. Printed film thickness was measured at two to three different points on each sample using a ball-tip digital micrometer with mea- surement accuracy of ± 0.004 mm. Locations of measurements on each sample were chosen to avoid areas where the 3D printing manual supports had been attached. Six to ten samples were ran- domly selected from each printing batch. While the starting files to print samples were the same, varying layer thickness setting (0.025 mm, 0.050 mm, 0.100 mm) resulted in a slightly different overall part thickness for each sample and each batch. Result for sample thickness measurements is shown in Table 1, with both within-sample variation and sample-to-sample variation of less than 5%. The overall average sample thickness (combining all sam- ple batches) is calculated to be 0.307 ± 0.029 mm. 2.3. Supercritical carbon dioxide processing 3D printed polymer samples were first cut into two halves along the longest side prior to scCO 2 processing. Each sample was cleaned with low-pressure compressed air to remove any loose particles on its surface, and weighed separately prior to loading. ScCO 2 process- ing was performed in the same manner described in [17], as a batch process at a specific temperature and pressure condition. Two dif- ferent temperatures were tested (313 K and 323 K), in combination with four different pressure settings (115 bar, 125 bar, 135 bar and 148 bar). All experiments were carried out in 316 stainless steel, high-pressure reactors equipped with calcium fluoride optical win- dows for in-situ system monitoring. Temperature was measured using a type-K Omega thermocouple and controlled constantly at an appropriate set point. Processing pressure was monitored using a digital Keller LEO Record pressure transducer and indicator. Each reactor was loaded with up to three polymer samples, sep- arated sufficiently from one another inside the reactor chamber to avoid any contact during treatment. An excess amount of flurbipro- fen powder (four times the drug solubility limit under the same temperature and pressure conditions) was loaded into the reactor along with the polymer samples for each experiment to ensure that drug availability was not a limiting factor in the process. This was verified by in-situ monitoring of the process, showing flurbipro- fen UV peak absorbance at 250-nm wavelength reached saturation after approximately one hour of pressurization and remained sat- urated throughout the duration of the experiment. The referenced solubility data of flurbiprofen in scCO 2 was calculated using the model established by Duarte et al. based on Chrastil’s density-based approach, as displayed in Table 2 [18]. The reactor was purged with CO 2 gas prior to pressurization to remove all air initially inside the reactor chamber. After purging, the reactor was pressurized with CO 2 (using a SFT-10 CO 2 pump) and heated to a desired setpoint (using four 65 W Omega CSS cartridge heaters connected to an Omega DP7001 temperature controller). The processing condition was held constant for a pre-determined 4 T.T. Ngo, L. Hoffman, G.D. Hoople et al. / J. of Supercritical Fluids 160 (2020) 104786 Table 1 3D printed polymer sample thickness. 3D printing layer thickness setting 0.025 mm 0.050 mm 0.100 mm(batch i) 0.100 mm(batch ii) Average sample thickness (mm) 0.309 0.273 0.333 0.274 Within-sample standard variation (mm) 0.008 0.006 0.014 0.009 Sample-to-sample standard deviation (mm) 0.006 0.005 0.013 0.009 Table 2 Solubility of flurbiprofen in scCO 2 at various experimental conditions. Temperature(K) Pressure(bar) Flurbiprofen solubility in scCO 2 (g/L) 313 115 0.15 125 0.19 135 0.23 148 0.28 323 115 0.06 125 0.13 135 0.20 148 0.29 period of time, typically for 24 h. Three to four sample replicates were experimented for each of the processing conditions. In addi- tion to the 24-h runs, two additional experimental runs with 4-h treatment duration were performed (with three replicates per run) to examine the impact of treatment time on drug loading. During all sample processing the solution content was constantly stirred by two small magnetic stir bars and two micro stir plates. When the experiment was complete, the reactor was removed from heat and allowed to cool naturally to room temperature. CO 2 depressurization was done at room temperature and kept at a suf- ficiently slow rate to prevent foaming of the polymer samples. Once the reactor was fully depressurized, polymer samples were removed and inspected for any damage. Low-pressure compressed air was applied to each processed sample thoroughly to remove any loose drug particles from its surfaces. The treated samples were then weighed and prepared for post-processing, such as SEM imaging or ethanol soak for drug loading determination. 2.4. Material characterization 2.4.1. Drug loading After being removed from the processing chamber and air- cleaned, scCO 2 /drug-treated polymer samples were weighed and compared to pre-processing weights. Although the pre- and post- processing weight measurements did not accurately reflect the level of drug loading for each sample, the values were used for crude verification of any major sample damage or destruction that had occurred during scCO 2 processing. For more accurate quantifica- tion of drug loading, treated polymer samples were soaked in 80 mL of ethanol solution for 48–72 h at room temperature while sealed air-tight. The 48-h minimum soak duration was tested and proven to be more than adequate for complete drug extraction out of the polymer phases and dissolution into the ethanol solution phase. After drug dissolution was complete, the solution was well stirred and three separate samples were drawn from the stock solution for repeatability validation. A USB 2000+ Ocean Optics UV–vis spec- trometer with UV-transparent optical fibers was used to measure UV peak absorbance of flurbiprofen at 249-nm wavelength, with 24 scans and an integration time of one second. Dilution of each sample was performed as needed based on the level of drug concentration and UV peak absorption intensity. Flurbiprofen concentration in solution and the total drug amount extracted from each polymer sample were then calculated using a previously established cali- bration curve between UV absorbance and flurbiprofen solubility in ethanol [17]. Final drug loading is determined based on Eq. 1. This method of drug loading calculation is consistent to those pre- viously reported in similar studies with different material systems [5,17]. Both the average and standard deviation were calculated for each set of replicate measurements of each sample. Drug Loading (%) = Mass of extracted drug Pre − process mass of polymer × 100 (1) 2.4.2. S urface morphology Scanning electron microscopy (SEM) was used to characterize the surface morphology of pre- and post-processed polymer sam- ples. The samples were cut in halves or quarters so that imaging of both the internal structure and the surface was possible. Samples were prepared for imaging using sputter deposition with a 60/40 gold/palladium alloy source for two minutes at 25-mA in an argon- rich chamber. Images of the surface and edges of the samples were captured at varying magnifications (100x, 200x, 500x and 1000x) using a Hitachi S-3400 N scanning electron microscope at 15-kV setting. To characterize surface roughness of polymer samples, SEM images were analyzed using Motic Images Plus 3.0 ML software. Surface roughness here is defined as the percentage of top surface area of the elevated regions in the whole sample surface, as shown in Eq. 2: Surface Roughness (%) = Top surface area of elevated regions Total surface area of sample × 100% (2) Based on Eq. 2, a higher value of surface roughness typically corresponds to a sample with a rougher overall surface. For each sample type (either treated or non-treated), SEM images were obtained for up to three replicate samples. Within each sample, two to three separate SEM images were taken at various locations on the sample, and three different measurements were extracted from each SEM image. The reported data was derived from up to 27 measurements for each of the non-treated, scCO 2 -only treated or scCO 2 -flurbiprofen treated conditions. 3. Results and discussion 3.1. Surface morphology of 3D-printed polymer Fig. 1 shows SEM image comparison among non-treated poly- mer samples 3D-printed (3DP) under 0.025-mm, 0.050-mm and 0.100-mm layer thickness settings. Thinner layer setting resulted in more visible ridges on sample surfaces. This might be explained by the fact that the polymer samples were printed at 30 ◦ angle per manufacturer’s printing recommendation. This angle served to minimize the surface area for each layer, thereby reducing the chance of part failure during the printing process. Also, the sample surface was observed to be more pitted at lower layer thickness settings, most likely due to variations in forces associated with the peeling process of the layers as they were removed from the PDMS print window (see section 2.2). While the cause of this pitting is not fully understood, it was possible that thinner layers had resulted in less structural integrity and were more likely to pit when removed from the PDMS window. Qualitative SEM observations correlate with the calculated surface roughness for each sample type (using Eq. 2), as shown in Fig. 2. Surface roughness seemed to be relatively the same for 0.025-mm and 0.050-mm 3DP layer thicknesses. How- ever, it was significantly lower (a 44 % reduction) for 0.100-mm 3DP layer thickness. T.T. Ngo, L. Hoffman, G.D. Hoople et al. / J. of Supercritical Fluids 160 (2020) 104786 5 Fig. 2. Surface roughness for non-treated 3D-printed polymer samples. Fig. 3. Drug loading for 3DP polymer samples under various scCO 2 -flurbiprofen 24-h treatments. 3.2. Drug loading of flurbiprofen in 3DP polymer facilitated by scCO 2 Drug loading data for each scCO 2 -flurbiprofen treatment condi- tion is summarized in Fig. 3, with three to four repeated samples per each experimental setting. Input parameters included 3DP layer thickness setting, scCO 2 processing temperature, and scCO 2 pro- cessing pressure. A shorter run time was also tested for 0.050-mm 3DP samples. At 313 K and 115 bar, 4-h scCO 2 treatment resulted in 14.68 ± 0.28 % drug loading (based on 3 repeated samples) com- pared to 19.08 ± 0.46 % drug loading with 24-h treatment. Similarly, at 323 K and 115 bar, 4-h scCO 2 treatment showed an average of 15.79 ± 1.02 % drug loading (also based on 3 repeated samples) compared to 23.60 ± 0.78 % drug loading with 24-h treatment. Data suggest that drug loading required longer than 4 h of scCO 2 treat- ment time to reach stabilization, which was consistent with past observation in other similar scCO 2 processing systems [40,41]. This result also indicates that drug loading can be modulated with scCO 2 treatment time to reach desired levels of drug loading, depending on specific material applications. Results show that a drug loading of over 24 % was achiev- able under the tested experimental conditions. The level of drug loading can clearly be modulated by scCO 2 treatment time and pro- cessing conditions such as temperature and pressure. According to Champeau et al. [5], PMMA had been shown to achieve drug loading facilitated by scCO 2 of about 8% with ketoprofen (sim- ilar solubility in scCO 2 compared to flurbiprofen) at 313 K and 100 bar [42,43], approximately 20 % with triflusal (more soluble in scCO 2 than flurbiprofen) at 308 K and 200 bar [44] and about 25 % with ibuprofen (more soluble in scCO 2 than flurbiprofen) at 323 K and 138 bar [45]. It was also noted that the PMMA matrices used in these previous studies were prepared using conventional methods such as solution casting and extrusion molding. Although no other 3DP polymer/drug system similar to what was experi- mented in this work had been investigated and reported in the past, the currently observed drug loading capability shows promising results with highly flexible, potential applications in the biomedical field. An ANOVA (with replicates) was performed on the collected data to investigate the dependency of drug loading on various input parameters. At 95 % confidence, there was a strong corre- lation between drug loading and scCO 2 processing temperature (p-value = 2.58 × 10 − 8 ). Average drug loading increased by 19 %–88 % when temperature increased from 313 K to 323 K, with the largest impact observed at 135 bar of scCO 2 processing pres- sure and polymer samples printed at 0.050-mm layer thickness setting. The positive effect of temperature on drug loading has been widely reported in multiple studies [34,46–49]. Increasing temperature encouraged chain mobility inside the polymer matrix, leading to higher sorption of CO 2 inside the matrix. Moreover, temperature also has a positive effect on flurbiprofen solubil- ity in scCO 2 at pressures above 110 bar [18]. The combination of a higher presence of drug molecules in the fluid phase and higher CO 2 sorption in the polymer resulted in higher drug load- ing inside the polymer matrix. This observation was consistent with past results reported by Ngo et al. [17] with flurbiprofen- impregnated PMMA-based biocomposites facilitated by scCO 2 processing. Higher scCO 2 processing pressure (past 115 bar), however, did not always result in higher drug loading as one might have pre- dicted. Increasing pressure under constant temperature would lead to increasing flurbiprofen solubility in scCO 2 , according to data reported by Duarte et al. [18]. However, in order for the drug loading to also increase, the molecular interactions between the drug and the polymer matrix must also increase to keep the drug molecules locked into the polymer instead of diffusing back out of the poly- mer matrix. If these interactions are not improving at a similar rate as the interaction between drug and CO 2 molecules in the fluid phase, then the flurbiprofen partitioning in polymer may not follow the same positive trend as pressure increases. Consequently, drug loading is not always favorable at higher pressure ranges. Similar results were reported by others with comparable polymer-based drug delivery systems [47,50–53]. Two-factor ANOVA (while keeping pressure constant) also shows that 3DP layer thickness had a more statistically signifi- cant impact on drug loading at a lower pressure range (115 − 125 bar; p-value ≤ 0.0008) than at a higher pressure range (135 − 148 bar; p-value > 0.05). Thicker 3DP layers seemed to cause slightly higher drug loading, as seen in Fig. 3. For example, at the 313 K and 125 bar scCO 2 processing condition, drug loading increased from 12.79 ± 0.49 % to 16.67 ± 1.05 % (i.e. a 30 % change) as 3DP layer thickness increased from 0.025 mm to 0.100 mm; This could be explained by the differences in surface roughness among the polymer sample types. 0.100-mm 3DP samples showed a more uniform surface across the sample (see Fig. 1) with lower sur- face roughness. Smoother surfaces might allow more efficient CO 2 swelling of the polymer and potentially more consistent drug dif- fusion into the polymer surface. Although the effect of 3DP layer thickness on drug loading was not consistent for all scCO 2 tempera- ture and pressure conditions, data suggests that thicker 3DP layers could be most beneficial for certain material applications where less surface roughness, better surface uniformity and higher drug loading are desired. Moreover, thicker 3DP layer settings typically mean shorter printing times (2.9 h for 0.025-mm layer thickness versus 3.8 h for 0.100-mm layer thickness; a 22 % time saving 6 T.T. Ngo, L. Hoffman, G.D. Hoople et al. / J. of Supercritical Fluids 160 (2020) 104786 Table 3 Comparison between experimental and predicted drug loadings. Experimental drug loading(%) Predicted drug loading(%) Deviation(%) 16.00 16.26 1.59 16.67 15.62 − 6.30 15.63 15.34 − 1.84 13.49 15.30 13.45 19.08 16.26 − 14.80 16.99 15.62 − 8.08 12.72 15.34 20.63 16.45 15.30 − 6.97 13.30 16.26 22.22 12.79 15.62 22.09 18.45 15.34 − 16.84 14.79 15.30 3.47 22.58 23.95 6.05 23.28 21.99 − 5.57 24.08 21.22 − 11.88 21.75 21.10 − 2.96 23.60 23.95 1.49 23.05 21.99 − 4.59 23.90 21.22 − 11.24 19.57 21.10 7.83 22.66 23.95 5.68 21.18 21.99 3.83 21.88 21.22 − 3.03 18.44 21.10 14.44 in this case), thus lowering the overall cost of material prepara- tion. Based on the observed influence of temperature on drug load- ing, a linear regression model was fitted for drug loading in polymer samples as a function of CO 2 density and flurbiprofen solubility in CO 2 CO 2 density was obtained either directly or through interpola- tion based on the tested experimental conditions in this work and Anwar and Carroll reference data [54]. Flurbiprofen solubility data was used per Table 2 based on previous work by Duarte et al. [18]. As discussed previously, drug loading behavior was influenced by several competing factors: drug solubility in the fluid phase, CO 2 sorption in polymer, and the intermolecular interactions between the drug and the polymer matrix. Drug partitioning between the polymer phase and the fluid phase seems to be the main driving factor for drug loading. Literature has shown that solute partition- ing between the fluid phase and the polymer phase depends on both fluid density and solute solubility in the fluid phase [18,55]. As a result, both CO 2 density and drug solubility were selected for input parameters in the regression model instead of temperature since they best represent the combined variations of temperature and pressure during the experiments. Using average drug load- ing data in Fig. 3, the amount of drug loading in 3DP polymer samples is empirically modeled as shown in Eq. 3, with p-values < 0.05 for the intercept and all coefficients. There was no sta- tistically significant interactive impact on drug loading observed between CO 2 density and drug solubility (p-value > 0.05), thus this interaction was not included in the regression model. Despite some scattering in the experimental data (R-square = 0.76), this model helps predict drug loading in 3DP polymer processing with a known fluid density and drug solubility in the processing medium. Table 3 compares the actual, experimental drug loadings to the predicted drug loadings using Eq. 3. Again, data scattering in the model was mainly due to the complexity of multiple competing factors occurring simultaneously in the polymer/fluid/drug system. Drug Loading (%) = 51 72 − 0 05492 × CO 2 Density + 22 47 × Drug Solubility (3) Fig. 4. Surface roughness for scCO 2 -treated polymer samples (24-h treatment). 3.3. Surface morphology of scCO 2 -treated 3DP polymer in presence of flurbiprofen Surface roughness was calculated for scCO 2 -treated polymer samples using Eq. 2 and SEM data (up to 27 measurements per each treatment), summarized in Fig. 4. Surface morphology analysis was focused mostly on 3DP samples at 0.025-mm and 0.100-mm layer thickness settings, investigating the effect of 3DP layer thickness on surface roughness. Surface morphology for scCO 2 -only-treated (no drug presence) polymer 3DP at 0.025-mm layer thickness set- ting was also examined, showing a lower surface roughness value compared to non-treated samples and scCO 2 /drug-treated sam- ples. This result suggests no negative impact of scCO 2 processing on the surface integrity of the polymer. On the other hand, scCO 2 pro- cessing might have cleaned the polymer surface, washing away any loosely trapped impurities present on the surface after 3D printing. In the case of 0.025-mm 3DP samples, the polymer surface rough- ness was reduced from 36.91 % to 31.62 % (about 14 % reduction) when treated with pure scCO 2 in absence of flurbiprofen at 323 K and 148 bar. No change in the surface integrity was observed when the poly- mer was treated with scCO 2 in the presence of flurbiprofen. Fig. 5 compares SEM images of treated polymer samples under 323 K and 148 bar. The treated polymer surface appeared comparable to the original, non-treated surface shown in Fig. 1. In addition, no foam- ing was noted on any scCO 2 -treated samples, proving sufficiently controlled CO 2 depressurization was performed during the exper- iments. Unintended foaming of polymer surface due to high CO 2 depressurization rates at elevated temperatures could compromise material surface integrity, making the material less desirable for most applications [56]. Based on two-factor ANOVA (with replication and at 95 % con- fidence, while keeping 3DP layer thickness constant), temperature appeared to be the most statistically significant influence on the surface roughness of polymer samples when treated in scCO 2 in the presence of flurbiprofen (p-value < 0.05). As temperature increased, surface roughness also seemed to increase. However, pressure variation alone was determined to have no significant impact on material surface roughness. Some interactive impact of both tem- perature and pressure on surface roughness was also noted for 0.025-mm treated samples. These observations were consistent with the results shown with drug loading of the polymer. Data suggests that increasing drug loading could also lead to increasing surface roughness of the polymer. 3DP layer thickness also appeared to have an effect on the sur- face roughness of the polymer treated in scCO 2 and flurbiprofen. A single-factor ANOVA was applied for scCO 2 /drug-treated samples at 323 K and 148 bar for a 24-h treatment duration. Results show T.T. Ngo, L. Hoffman, G.D. Hoople et al. / J. of Supercritical Fluids 160 (2020) 104786 7 Fig. 5. SEM of scCO 2 -treated polymer in presence of flurbiprofen at 313 K and 148 bar, for different 3DP layer thicknesses. Table 4 Comparison between experimental and predicted surface roughness. Experimental surface roughness (%) Predicted surface roughness (%) Deviation(%) 35.90 35.47 − 1.20 36.37 37.27 2.50 39.83 38.77 − 2.67 39.41 40.58 2.96 37.80 36.93 − 2.29 24.08 24.54 1.91 27.29 26.35 − 3.44 26.81 27.84 3.86 29.92 29.65 − 0.90 that material surface roughness decreased as 3DP layer thickness increased (F = 24.7 > F critical = 3.23; p-value = 1.17 × 10 − 7 < 0.05). A 27.5 % reduction in surface roughness was noted when 3DP layer thickness increased from 0.025 mm to 0.100 mm. This was consis- tent with observations made with non-treated samples. Once again, analysis of this data suggests that drug loading has the same level of effect on polymer surface roughness, regardless of the starting state of the material. Because surface roughness appeared to depend on multiple pro- cess parameters, a multiple linear regression model was fitted to the experimental data. The influence of temperature and the inter- active effect of temperature and pressure suggest the use of CO 2 fluid density as one of the input parameters in the model. As previ- ously discussed, drug loading has an impact on surface roughness. Since drug solubility in scCO 2 has a direct effect on drug loading, and drug loading had shown some influence on surface rough- ness, the solubility of flurbiprofen in scCO 2 was also used as an independent variable in the regression model. In addition, 3DP layer thickness was included in the model due to its direct impact on surface roughness. Eq. 4 shows the multiple linear regression model obtained from experimental data (R-square = 0.98). Based on this empirical model, while drug solubility imposes a posi- tive effect on polymer surface roughness, both 3DP layer thickness and CO 2 density have negative effects on the material surface. Table 4 compares the experimental surface roughness calculated from SEM data to the predicted surface roughness using Eq. 4. Due to a high goodness of fit with the data, deviations between the experimental and predicted values maintain to be less than 4%. Surface Roughness (%) = 53 73 − 145 6 × 3 DP Layer Thickness − 0 02549 × CO 2 Density + 24 59 × Drug Solubility (4) 4. Conclusion 3D-printed acrylate-based polymer was impregnated with flurbiprofen in supercritical carbon dioxide to create a multi- component polymeric system with potential applications in the biomedical field. Drug loading and surface roughness were shown to be tunable with different process parameters, making the pro- cess viable for specific application needs, from drug delivery to biological implants. An average drug loading range of 12.72–24.08 % was achievable under the tested conditions (313–323 K and 115–148 bar). Thicker layer settings in the 3D-printing process (corresponding to shorter print time) showed better material sur- face smoothness while not negatively impacting drug loading. Temperature had the most significant effects on drug loading and surface roughness. Increasing temperature resulted in an increase in drug loading of flurbiprofen into the polymer matrix and also an increase in surface roughness of the material. Both CO 2 density and drug solubility seemed to be important influencing factors for drug loading and material surface roughness. Finally, linear regres- sion empirical models were established for predicting drug loading and surface roughness of the polymer based on processing fluid density, drug solubility, and 3DP layer thickness. All of these pr