Integrated biocompatible 3D printed isoporous membranes with 7 μm pores

Abstract

In this work, we present a new 3D printing technique that enables the realization of native digital micro-mirror device (DMD) resolution in negative features of a 3D printed part without improving 3D printer hardware and demonstrate the fabrication of fully integrated, biocompatible isoporous membranes with pore sizes as small as 7 μm. We utilize this technique to construct a microfluidic device that mimics an established organ-on-a-chip configuration, including an integrated isoporous membrane. Two cell populations are seeded on either side of the membrane and imaged as a proof of concept for other organ-on-a-chip applications. These 3D printed isoporous membranes can be leveraged for a wide variety of other mechanical and biological applications, creating new possibilities for seamlessly integrated, 3D printed microfluidic devices.

1. Introduction

Due to their abundant and diverse applications, porous membranes have been an active area of research over the last 30 years.^1–4 Isoporous membranes in particular (membranes with parallel, ordered pores) are ideal for many mechanical and biological applications such as bio-sensing, drug assays, drug delivery, organ-on-a-chip technologies, cell fractionations, and other biological separations.^5,6 This wide interest in porous membranes has resulted in numerous fabrication techniques including the use of commercially produced membranes,² ion track etching,⁷ hydrogels,¹ and other stereolithographic methods.² Despite numerous fabrication techniques, the scarcity of commercially available options with lower pore density and highly uniform, parallel pores remains a significant hurdle for many technologies relying on porous membranes.⁶ Additionally, commercially available membranes must be aligned and bonded or otherwise incorporated into a target device, increasing fabrication time and complexity^2,8 and requiring additional materials that may be cytotoxic.

In this paper we present a new 3D printing approach for overcoming these obstacles. We create porous membranes with controlled pore size and density, and that are fabricated in situ as part of the 3D print. No further integration processes or materials are required as the entire print is made from a single, biocompatible resin.^9,10 We begin with a treatment of what can be accomplished with traditional 3D printing techniques and then develop a new exposure approach that can realize voids with native micro-mirror dimensions in a 3D printed part, resulting in porous membranes with 7 μm pores. Pores this size are relevant for many organ-on-chip applications including lung^11,12 and pancreas¹³ models, among others. Two fluorescent cell populations are then seeded on each side of the membrane and a 3D scan is performed with a confocal fluorescence microscope to show that the cells adhere to the membrane but remain physically distinct on either side of the membrane, mimicking a popular organ-on-a-chip topology.^8,11–16

2. Materials and methods

Membranes with various pore sizes were designed and printed on a custom 3D printer reported in ref. 10 and 17 which is based on a DMD with 7.6 μm pixels and uses 1 : 1 projection optics, yielding square 7.6 μm pixels in the image plane. We used a custom photopolymerizable resin consisting of poly(ethylene glycol) diacrylate (PEGDA, MW258) with a 1% (w/w) phenylbis(2,4,6-trimethylbenzoyl)phosphine oxide (Irgacure 819) photoinitiator and a 0.38% (w/w) Avobenzone UV absorber, the same resin reported in ref. 9 and 10. The devices were printed on 25 mm square silanized glass substrates, prepared as described in ref. 18. The measured image plane irradiance was 26 mW cm⁻².

3D print files used to fabricate the isoporous membranes were constructed according to an updated version of our open-source 3D print JSON specification.¹⁹ The images used in the print files were generated using Python’s numpy library and with an open source CAD tool, OpenSCAD. When OpenSCAD was used, 3D STL models were generated and sliced with a custom slicer software²⁰ to generate the images used for individual exposures. An oblique view of an example porous membrane geometry as well as an individual image used for exposure are shown in Fig. 1. After printing, the 3D printed parts were washed with 2-propanol to remove uncured 3D printing resin, air dried, and post cured in a custom curing station for 20 minutes using a 430 nm LED (Thorlabs, Newton, NJ, USA) with a measured irradiance of 11.3 mW cm⁻² in the curing plane.

Fig. 1.

3D CAD design of a porous membrane. (a) Oblique view. (b) An example exposure pattern used to fabricate the membrane region.

To prepare the 3D printed parts for cell seeding, cold rat tail collagen (0.2 μg mL⁻¹ in 17 mM acetic acid) was flushed through the device channels and left to cure at room temperature for 24 hours. Next, excess collagen was removed with acetic acid and the channels were flushed with 70% ethanol in deionized water to sterilize the interior of the device. Sterile phosphate buffered saline (PBS) was then washed through the voids to remove the ethanol which could harm the cells if left behind. The devices were then incubated at 37 °C for 30 minutes. Prior to cell seeding the chips were flushed with DMEM/F12.

The cell lines used for seeding were the human lung fibroblast cell line HFL1 (ATCC CCL-153) and the human lung epithelial cell line A549 (ATCC CCL-185), which were previously stably transduced with fluorescent tags using lentiviral vectors, described in ref. 21. The HFL1 cell line was transduced with pLV-mCherry (red) and the A549 cell line was transduced with pLV-GFP (green). The cells were maintained in complete growth media consisting of DMEM/F12 supplemented with 10% fetal bovine serum (FBS) and anti-biotic/anti-mycotic (Carson) prior to seeding. Cells were next dissociated using 0.25% trypsin, 0.1% EDTA, and pelleted. The cell pellets were then washed twice with sterile 10% FBS in PBS to deactivate the trypsin and cell counts were taken to determine the dilution needed to obtain a concentration of 5.0 × 10⁶ to 10.0 × 10⁶ cells per mL in complete media. This high concentration cell mixture was then flushed through one side of the membrane chamber of the 3D printed device and centrifuged at 400 g for 4 minutes to pull the cells close to the membrane. The devices were then incubated with excess media at 37 °C and 5% carbon dioxide for 4 to 6 hours, then were flipped and the process repeated with the other cell type for the other side. A549 cells take longer to adhere after dissociation than HFL1 cells, so they were seeded first.

The chips were then mounted into a live cell imaging chamber (Okolab, Ottaviano, Italy) attached to a Leica TCS SP8 confocal microscope (Leica, Wetzlar, Germany). The chamber was set to 5% CO₂ at 37 °C throughout the experiment. The 10× and 20× objectives were used during imaging. The argon laser was set to 25 W power for imaging the A549 cells and the DPSS 561 laser was used to image the HFL1 cells. Individual laser powers were kept at less than 5% to prevent excessive phototoxicity while imaging. Images were captured at either 512 × 512 or 1024 × 1024 resolution, with 3D images generated through z-stack acquisition. Z-Stack step sizes were manually set and did not exceed half the size of the optical section to prevent under-sampling. Volumetric image processing was performed using the 3D analysis package in the LASX software and final images were exported as TIFF files.

3. Results and discussion

3.1. Evaluation of conventional 3D printing methods

Conventional 3D printing is usually limited to creating layers with equal thicknesses and exposure times. We evaluated the smallest possible pores achievable with this approach using 10 μm thick layers and a variety of layer exposure times ranging from 100 to 300 milliseconds. Fig. 2(a–c) shows the results for pores with a designed width of 5 pixels, or 38 μm. Note that for lower exposure times, the pores are much larger than designed (Fig. 2(a and b)). Ordinarily, one would then increase the exposure time to try to reduce the pore size until it matches the designed size. This approach works for pores as small as 38 μm, as seen in Fig. 2(c). For smaller pores however (Fig. 2(d–f)), the pores close off and fill in completely before reaching the designed size, instead forming a solid membrane (Fig. 2(f)). We found no combination of design and exposure settings that could yield pores below about 30 μm using this conventional 3D printing approach.

Fig. 2.

Limits of the conventional 3D printing approach for pore fabrication. Membranes (a–c) have designed pore widths of 38 μm (5 image pixels) while (d–f) have designed pore widths of 22.8 μm (3 image pixels). (a and d) Small exposure times result in polymerized filaments instead of membranes. The filaments can be sparsely attached and/or stretched. (b, c and e) Increasingly small pore sizes down to 30 μm can be achieved by extending exposure time, but the pores are always larger than designed. (f) Pores seal off completely when the exposure time is too high.

3.2. Zero-thickness layers

We found that for very low exposure times the exposed area would not polymerize to the full designed 10 μm depth, resulting in membranes that would tear free of the previously exposed bulk material and cause defects similar to those seen in Fig. 2(a). To mitigate this issue, we introduce the idea of fabricating zero-thickness layers. For each layer, the build platform begins in a raised position as shown in Fig. 3(a).

Fig. 3.

Illustration of a zero-thickness layer. In a normal layer (left), the build platform is lowered to the previous build position less the designed layer thickness (b) and an exposure is performed (d). In a zero-thickness layer (right), the build platform is lowered to its previous position with no extra space for the new layer (c). The next exposure is then performed, embedding the new layer in the previous one (e). In both cases, the build platform is then raised in preparation for the next layer (f and g).

For a normal 10 μm layer, it is then lowered until there is a 10 μm gap between the bottom of the 3D printed part and the resin tray, as shown in Fig. 3(b). The UV source is then activated, causing a new layer to form that is 10 μm thick, represented by the red area in Fig. 3(d). Finally, the build platform is raised so the part is ready for the next layer (Fig. 3(f)).

For a zero-thickness layer, the build platform is instead lowered until there is no gap between the 3D printed part and the resin tray (Fig. 3(c)). The subsequent UV exposure then causes further polymerization within the previous layer (Fig. 3(e)), allowing for the use of very small exposure times for the creation of membranes with very thin filaments. The image used for this exposure should overlap sufficiently with the existing bulk material to ensure there is proper adhesion between the new exposure and the existing part, else thin membranes and filaments can tear away. Similar to the normal case, the layer is finished once the build platform is raised in preparation for the next layer (Fig. 3(g)).

We used a single zero-thickness layer to fabricate membranes with pore sizes ranging from 38 μm (5 image pixels) down to 7.6 μm (1 image pixel). Typical results are shown in Fig. 4. Each row represents a different designed pore size and each column represents a different exposure time. The leftmost column, representing 100 millisecond exposures, demonstrates very sparse membranes with thin filaments that survive the printing process and remain adhered to the bulk material. Even at this very low exposure time, pores begin to close off for pore sizes of 7.6 μm, as seen in Fig. 4(m).

Fig. 4.

Limits of a single, zero-thickness layer for pore fabrication. Each row represents increasing exposure times while each column represents decreasing pore sizes. Using a zero-thickness layer instead of a normal 10 μm layer allows for the creation of membranes with very thin filaments as seen in the left column. Pores begin to close off before reaching their designed sizes (i, k and m) and eventually close off completely (l, n and o). The minimum achievable pore size remains approximately 30 μm (k).

We next increase the exposure time to try to obtain smaller pores closer to the designed dimensions. The middle column of Fig. 4 shows the effects of an intermediate exposure time of 150 milliseconds. For the membranes with designed sizes larger than 22 μm, the increased exposure time did have the effect of making the pores smaller, yet they are still much larger than designed (see Fig. 4(b, e and h)). Additionally, while the 15.2 μm pores were open at an exposure time of 100 milliseconds (Fig. 4(j)), they are now partially closed and still have not achieved their designed size (Fig. 4(k)). The 7.6 μm pores have closed off completely (Fig. 4(n)).

The right column of Fig. 4 shows the effects of a higher exposure time of 200 milliseconds. Pores with designed sizes greater than 30 μm have narrowed slightly, but have still not achieved their designed sizes (Fig. 4(c and f)). The previously open pores with a designed size of 22.8 μm have similarly narrowed slightly but have begun to close off (Fig. 4(i)). Pores with designed sizes smaller than 20 μm are now completely closed (Fig. 4(l and o)).

These results show that the minimum achievable pore size remains approximately 30 μm, and neither increased exposure time nor decreased design dimensions yield anything smaller.

3.3. Repeated zero-thickness layers

We discovered that using more zero-thickness layers enables the fabrication of much smaller pores. First, an exposure time that would normally produce one of the thin filament meshes is chosen, and then that exposure time is used for repeated zero-thickness layers. The first layer generates the base of the membrane as a filament mesh and the subsequent layers build on the thin mesh, widening each filament, resulting in a progressively wider mesh with smaller pores. The key here seems to be the motion of the build platform. We hypothesize that the motion of the 3D printed part in the liquid resin caused by the up and down motion of the build platform in between each exposure agitates the resin, moving partially polymerized resin away from the thin membrane mesh, thus providing the subsequent layer with a fresh pool of resin between the pores. The next exposure then only fully polymerizes near the existing filaments, widening them, while partially polymerized resin in the pores is moved away before the next exposure. This process is repeated until the pores reach the desired size.

Fig. 5 shows what this looks like in practice where each individual zero-thickness layer has an exposure time of 100 milliseconds. Each row represents a designed pore size and each column represents a different number of repeated zero-thickness layers. For a designed pore size of 38 μm, 5 zero-thickness layers still results in a membrane with irregular pore sizes (Fig. 5(a)), but 30 layers produce a membrane with uniform pores (Fig. 5(c)). Membranes with designed pore sizes of 22.8 μm behave similarly, as seen in Fig. 5(d–f).

Fig. 5.

Effect of repeated zero-thickness layers for various pore sizes. Each row represents a designed pore size and each column represents a different number of repeated zero-thickness layers. Actual pore sizes achieve their designed sizes when 30 repeated zero-thickness layers are used (c, f and i). Pores as small as 7 μm can be achieved (i), but the geometry for the smallest pores becomes irregular (g–i).

The smallest pores with designed sizes of 7.6 μm (a single image pixel) are less well behaved. Repeated zero-thickness layers do reduce pore size down to 7 μm in some regions of the membrane, but the membranes end up malformed, particularly near the center of the membrane (Fig. 5(g–i)).

3.4. Added support structures

While pores down to 7 μm can be realized, the resulting grid of pores can be very irregular, especially when small pores (<20 μm) with high fill factor are desired (note the defects in the center regions of the membranes in Fig. 5(g–i)). We believe this is due to the mechanical instability of the thin membrane as it begins to form within the first few layers; the thin filaments are easily displaced when moved through the liquid resin by the motion of the build platform. To mitigate this issue, underlying support posts can be added. Fig. 6(a) shows the support pillars (in red) placed beneath the membrane. Fig. 6(b) shows what the pillars look like when fabricated with the membrane omitted. Note in Fig. 6(c) that the support pillars are wider than the space between individual pores, resulting in a blockage of some of the pores.

Fig. 6.

Isoporous membrane fabricated with 30 zero-thickness layers and support pillars. (a) 3D CAD model of porous membrane and supporting pillars. (b) Supporting pillar geometry. (c) Surface of a full membrane 200 μm in diameter. (d) Close-up of 7 μm × 7 μm pores.

The supports improve the stability of the initial structure and result in uniform pores that are 7 μm × 7 μm, as shown in Fig. 6(d), with no known span limitation. Fabrication of these membranes requires very good optical focus. Once good focus is obtained and the appropriate exposure times and number of repeated layers are determined, the membranes can be fabricated repeatedly with a high success rate. The membranes begin thin but grow thicker with subsequent exposures. The average measured thickness of a completed membrane was 12.6 μm.

If an even higher pore fill factor is needed, a checkerboard pattern with 8 μm pores is also possible, shown in Fig. 7. The checkerboard pattern was fabricated with a bulk exposure time of 300 milliseconds and 25 zero-thickness layers with an exposure time of 100 milliseconds for the membrane. This checkerboard pattern is challenging to fabricate and we are currently limited to a span of about 60 μm.

Fig. 7.

Isoporous membrane fabricated with 25 zero-thickness layers demonstrating a high fill-factor checkerboard pattern with 8 μm pores.

3.5. Integration into an organ-on-a-chip inspired device

We next fabricated a fully 3D printed microfluidic device using a membrane similar to the one shown in Fig. 6(c), adding a roof to the membrane chamber and channels to allow seeding of cell cultures on either side of the membrane. CAD drawings of the device are show in Fig. 8(a and b). Cells were seeded and imaged as described above, resulting in the images shown in Fig. 8(c–f). Fig. 8(c) shows a microscope image of the membrane where both the underlying supports and pores are visible. Fig. 8(d) and (e) show a top and oblique view with both cell populations visible. Finally, Fig. 8(f) shows a side view where both cell populations adhered to the surface of the membrane yet remain physically distinct on either side of the membrane, emulating a topology common to many organ-on-chip applications.

Fig. 8.

CAD drawings and 3D confocal fluorescence imagery of a membrane device with two cell populations seeded on the membrane. The red cells are human lung fibroblast cell line HFL1 and the green cells are human lung epithelial cell line A549. (a) Oblique view of CAD drawing from the bottom showing the supporting pillar structure (blue) and porous membrane (yellow). The two cell chambers and the channels feeding them are in red and green to match the colors of the cells seeded in them. (b) Bottom view of CAD drawing. The pink dots are the pores in the membrane. (c) Bottom view microscope image of fabricated device. (d) Top view fluorescence image. (e) Oblique view fluorescence image. (f) Side view fluorescence image.

4. Conclusion

As we have previously demonstrated,¹⁸ conventional 3D printing processes can exhibit significant and unnecessary limitations. In this work we expand our repertoire of generalized 3D printing techniques by introducing the capability of creating smaller voids without requiring any modifications to the 3D printer hardware itself. Specifically, this technique facilitates the precise placement of voids within a membrane, resulting in the creation of isoporous membranes with controlled pore size, positioning, and density. This method expedites and streamlines the production of highly integrated microfluidic devices, as the membrane is seamlessly fabricated from the same material as the rest of the device, not requiring additional materials or processes. Moreover, this technique is compatible with a biocompatible resin, rendering it invaluable for applications involving live cells, such as organs-on-a-chip. There is also potential to apply this technique to other 3D printers and resins, enhancing the precision and versatility of 3D printing processes that are especially well-suited for microfluidics.