We demonstrate the layer-by-layer (LbL) assembly of polyelectrolyte multilayers (PEM) on three-dimensional nanofiber scaffolds. High porosity (99%) aligned carbon nanotube (CNT) arrays are photolithographically patterned into elements that act as textured scaffolds for the creation of functionally coated (nano)porous materials. Nanometer-scale bilayers of poly(allylamine hydrochloride)/poly(styrene sulfonate) (PAH/SPS) are formed conformally on the individual nanotubes by repeated deposition from aqueous solution in microfluidic channels. Computational and experimental results show that the LbL deposition is dominated by the diffusive transport of the polymeric constituents, and we use this understanding to demonstrate spatial tailoring on the patterned nanoporous elements. A proof-of-principle application, microfluidic bioparticle capture using N-hydroxysuccinimide-biotin binding for the isolation of prostate-specific antigen (PSA), is demonstrated.
The ability to achieve nanoscale control of interfaces and surfaces has fueled new ideas and technologies in a myriad of applications, including the design of various microelectronic1,
A promising nanostructuring approach to create 3D templates is the use of textured nanoporous elements as scaffolds for surface functionalization and as building blocks for various devices. One such textured element that has received considerable attention is an aligned array of nanofibers, such as carbon nanotubes (CNTs)19,
Although in principle, various approaches exist for modifying the surfaces of CNTs contained within CNT arrays, the layer-by-layer (LbL) assembly approach34 holds the most promise for fulfilling the multiple requirements of nanoscale thickness, morphology control, conformal coating uniformity, and the ability to create a wide range of different surface chemistries and functionalities. Indeed, the greatest advantage of the LbL assembly process compared to other coating processes continues to be its ability to produce nanoscale conformal coatings from an extremely wide variety of organic, inorganic, and biological molecules and materials35,
Our approach has been to utilize LbL coatings as a means to systematically functionalize patterned CNT arrays contained within microfluidic devices designed to filter, capture, and detect low levels of biological markers of disease32,33. We demonstrate that it is possible to control the LbL assembly pattern within a microfluidic device by only varying the height of the microchannel. In this manner, we can coat and functionalize individual CNTs throughout the entire CNT array or cover just the outer surface of the array. In the former case, conformal coatings on individual nanotubes have been achieved within arrays with tube-to-tube spacings of approximately 80 nm. This level of functionalization control can be used to take full advantage of the highly porous nature of the CNT arrays, as well their ability to controllably capture and release biomarkers of various types. To demonstrate this, we used this technology to functionalize the CNT surfaces with antibodies and capture prostate-specific antigen (PSA) as an exemplary capture target. A significant challenge in microfluidic platform bioparticle isolation is to achieve sufficient interaction between the target biomolecule and the functional surfaces to promote binding. This work provides a powerful tool to functionally tailor and grade nanomaterials in three dimensions, enabling the design of devices that enhance the bioparticle–surface interaction. Additionally, owing to the nature of the spacing between the individual CNTs (∼80 nm), this platform specifically has high potential in the isolation of nano-sized bioparticles, such as viruses, exosomes, or DNA.
Materials and methods
The microfluidic chip design and assembly processes (Figure 1) were developed to favor adsorption and LbL conformal coating on the nanoporous CNT elements contained within the microfluidic device (Figure 1a). Patterned CNT elements were grown via chemical vapor deposition on Si wafers. (Fabrication details can be found in the Supplementary Information). These porous VACNT forests consist of arrays of multi-walled CNTs (OD: 7.78 ± 0.85 nm; ID: 5.12 nm ± 0.76 nm, spaced ∼80 nm apart, # walls 5 ± 1, 1 vol%, 1.59 g cm−3 51,52) (Figure 1b and c) patterned by using photolithography into macroscopic elements that have high porosity (∼99%) and accessibility to aqueous solutions. For this study, one geometric pattern of CNTs was used for simplicity: a single cylindrical pillar of 1 mm in diameter (consisting of more than 108 individual carbon nanutubes) centered in the 3 mm wide, 7 mm long microfluidic channel (Figure 1b and c). CNT pillar heights averaged 50 μm but ranged between 30 μm and 80 μm. Results were also obtained with other element geometries (not shown), such as an array of pillars or a porous “wall”, i.e., a porous rectangular feature arranged perpendicular to flow. In fact, for particle isolation, a partitioned microfluidic device containing a nanoporous wall, which acts like a filter, would be an ideal design; however, the resulting pressure drop across the dividing wall can damage the adhesion of the CNTs to the substrate. Therefore, our cylindrical geometric design resolves the trade-off between the maximum allowed pressure and maximum interception of flow streamlines with the CNT elements. Photolithography permits numerous patterns to be designed in the future for further investigation of the optimal nanoporous CNT element geometries.
Microfluidic channels were fabricated according to standard soft lithography protocols53 in conjunction with negative molds made of SU-8 photoresist (Microchem, MA, USA) onto silicon wafers. The microfluidic chip design had three inlets: one for each polyelectrolyte solution used in the LbL process and one for a water rinse required to wash out the polymer molecules that are not adsorbed in each step onto the CNTs (Figure 1a).
Polyelectrolyte multilayer (PEM) film deposition on the CNT surfaces was performed via the LbL assembly of PAH (15 kDa) and poly(styrene sulfonate) (SPS; 70 kDa) within the microfluidic devices. PAH conjugated with fluorescein (PAH-Fluor) was used to permit visualization. Integrated CNT microfluidic devices were primed with a negatively charged surfactant (sodium dodecyl benzene sulfonate, SDBS), which was selected for its propensity to bind to individual CNTs54 and its electrical charge that favors polyelectrolyte adsorption. Because CNTs have a tendency to be intrinsically hydrophobic, the SDBS served to lower the surface energy of the nanotubes, making the CNT elements more hydrophilic and more receptive to polymer coating55,56. After priming, the devices were rinsed with DI water for 5 min at 6 μL min−1, and LbL assembly was performed via the alternate flow of PAH-Fluor and SPS solutions at 6 μL min−1 for 5 min each, with intermediate water rinse flow (5 min, 6 μL min−1). PAH-Fluor was the first and last layer deposited to retain a positive charge and because it was labeled for imaging.
As previously described, PAH/SPS films assembled at pH 9.3 present primary amine groups isolated in hydrophobic pockets that can be opened via an acid treatment that will protonate the amine groups and expose them to the surface of the film57,58. The exposed amine groups can be reacted with selected chemistries, such as N-hydroxysuccinimide (NHS) groups, creating stable amide bonds to link a biotinylated surface to the film surface59. Thus, pH 9.3 was selected for the standard LbL assembly, and an acid treatment at pH 2.5 was performed after the LbL deposition and prior to the bioparticle capture protocol to generate amine-rich surfaces. LbL assembly was performed successfully with 3.5 bilayers of PAH-Fluor/SPS at pH 9.3 via continuous flow LbL assembly.
Results and discussion
We first sought to achieve full, uniform coatings throughout the microfluidic 1 mm pillar CNT device. The full CNT pillar element coating was achieved when devices were fabricated and assembled such that there was a gap (∼50 µm) between the CNT element height and the PDMS channel height. Confocal microscopy images showed a polymer coating throughout the entire volume of the CNT element (Figure 1d), and TEM micrographs suggested nanoscale conformal coating layers on individual CNTs within the element by comparing the CNT diameters before and after the LbL deposition (Figure 1e), confirming the ability to coat the individual CNTs and construct a 3D LbL-coated surface. It is notable that there tended to consistently be a higher intensity of fluorescence signal in the mid Z-plane of the device, as shown in Figure 1d, suggesting a larger amount of coating in that location. We believe this is an artifact of the rinsing methods used, which removes some polymer from the top surface of the device. Additionally, it is possible that this intensity variation is diagnostic of the CNT morphology, and perhaps the changes in the CNT density along the Z-axis give rise to small changes in polymer coating. It is also possible that this variation is due to an imaging artifact based on the use of a laser scanning confocal microscope to image through the PDMS channel, which causes scattering and difficulty in focusing far away from each boundary.
Because a uniform, conformal coating was achieved with a gap in the device geometry, we hypothesized that the presence of this gap was important in controlling the coating morphology. We designed devices such that there was no clearance gap between the CNTs and the top of the PDMS channel by slightly compressing the CNT element along the Z-axis (reference axis in Figure 1d). The results in this modified geometry showed only an annular coating of the CNTs. The thickness of the annular coating was on the order of half the height of the CNT element. This annular coating mechanism was maintained throughout the volume of the CNT element and was distinctly different from the full coating results obtained when a clearance gap was maintained (Figure 2). Thus, by changing the geometry of the device elements, we were able to exhibit spatial control of the LbL coating. Furthermore, methods have been investigated to close a PDMS gap in situ60 such that the channel could be sealed after the LbL coating but prior to any biological or chemical assay.
To describe the difference in the uniform vs. annular coating mechanisms, a 3D numerical model was developed to investigate the transport of the polymer solution through the CNT elements. The transport phenomena involved in the experiments included convection, diffusion, and adsorption. In our numerical model, we included convection and diffusion but neglected adsorption because the timescale for the adsorption is much smaller than that of diffusion61. Experimental observations suggested that as soon as the polymer solution diffuses into the CNT forest, the polymer is adsorbed onto the CNT surface. To further quantify the kinetics of the adsorption, we calculated the Damköhler number, Dk, which compares the reaction rate with the rate of diffusive transport62: Here, the reaction of interest is the adsorption of the long-chained polymer molecules with charged groups to the surface with opposite polarity. On the basis of the theoretical analysis by Stuart et al.,61 equilibrium in adsorption should be attained on a timescale of seconds in our system. In our experiments, the diffusion occurred on a timescale of minutes to hours; therefore, the Damköhler number was in the range of 100–1000, which justifies neglecting the adsorption kinetics in the numerical model.
For a quantitative comparison of the rate of the polymer transport by convection and diffusion, we determined the Péclet number, defined as: where Uc and Lc are the characteristic velocity and length scale, respectively, and D12 denotes the binary diffusion coefficient of species 1 (polymer) and solvent 2 (water). In our system, two Péclet numbers existed owing to the two different velocities as a result of the different structural length scales: one based on the height of the microchannel and the other based on the pore size in the CNT forest. Inside the nanoporous domain, the characteristic velocity can be obtained by comparing the hydrodynamic resistances of the porous CNT pillar element and the side gap between the pillar and the channel wall. The order of magnitude for the average velocity through the side gap can be estimated as , where H is the channel height, μ is the fluid viscosity, and Δp/L is the pressure drop over the length of the channel, L. Additionally, according to Darcy's law, the apparent velocity inside the porous CNT forest is , where k is the permeability of the CNT forest. Therefore, the ratio of the two velocities is (using H = 100 μm and k ∼ 10−16 m2, derived from Ref. 63). In our experiments, the channel flow velocity is U = 3 × 10−4 m s−1 (corresponding to a flow rate of 6 μL min−1 through the microchannel), and the binary diffusion coefficient is m2 s−1 64. Therefore, the Péclet number in the channel is , whereas in the CNT forest, the relevant dimensionless ratio is . The high disparity between these values clearly indicates that outside the porous CNT pillar, convection is dominant whereas inside the CNTs, diffusion controls the transport of polymer molecules in spite of the high porosity (99%) of the CNTs.
In fact, the flow permeation inside the pillar is negligible as a result of two contributing factors. The first is the low Darcy number of the nanoporous medium, Da. Da is defined as the permeability, non-dimensionalized by the porous collector diameter, (D): which yields a value of Da=10−10 for our system. The second is the low confinement ratio, the ratio of the collector diameter (1 mm) to the channel width (3 mm). The flow permeation through confined cylindrical porous collectors has been studied by Shahsavari et al. According to Shahsavari and McKinley65, the superficial velocity in the porous region relative to the free stream velocity should be on the order of 10−8 for our geometrical parameters, which is the same result that we estimated by comparing the hydrodynamic resistance of the channel and the porous CNT pillar. Ultimately, the combination of these two factors suggests that the porous CNT collector is essentially impermeable, owing to the dominance of convective effects when Pe ≥ 1. Further details regarding the numerical simulation can be found in the Supplementary Information.
The experimental findings (Figure 2a and b) were consistent with the resulting concentration distributions based on the numerical modeling (Figure 2c and d), which explain the two separate and distinct coating patterns observed experimentally. The numerical modeling confirmed that the assembly process is dominated by diffusion, predicting that for a microfluidic device with a controlled gap on top of the pillar, the polymer molecules diffuse axially through a thickness of only ∼50 μm, whereas in the device with no gap clearance, the polymer molecules can only diffuse radially. The experimental confocal microscopy results showed that this radial, annular coating penetrated radially to a distance on the order of 50 μm as well.
We used this understanding to functionalize the CNT surfaces with antibodies and capture PSA as an exemplary capture target. A modified sandwich ELISA on cylindrical CNT elements coated with 3.5 bilayers was performed to capture PSA (100 ng mL−1) by using a biotinylated secondary antibody and streptavidin-coated quantum dots (Qdots) (605 nm; Figure 3a). All capture and control experiments were performed on devices with a gap, thus using a full spatial LbL coating (further information on assay located in Supplementary Information). Confocal microscopy results indicated a high correlation between the LbL coating signal (fluorescein isothiocyanate (FITC), 488 nm) and the capture signal (Qdots, 605 nm), and capture was observed throughout the volume of the device (through Z-plane Figure 3b). Particles were advantageously captured throughout the element volume rather than solely on the outer surface, as in typical microfluidic capture elements. Intensity measurements were collected across multiple devices and Z-plane stacks with ImageJ, and the average ratio of signal to background noise was determined. The LbL functionalization approach yielded a capture rate 1.4 times higher than all the CNT element controls, which demonstrates that specific, covalent capture was achieved within the CNT elements.
The functionalization protocol59 used herein has only recently been reported and has not yet been optimized for the CNT scaffolds. The capture protocol demonstrates the broad utility of the 3D LbL coating platform on CNT elements rather than a specific ability to target PSA66. Tailoring devices with sufficient physical interactions between bioparticle targets and the functionally coated surfaces is a great challenge in microfluidic bioparticle isolation. By using this platform, we increased the surface area by a factor of 20 for an equivalent microfluidic channel volume with a planar LbL coating, providing an enhanced sensing platform that is compatible with current microfluidic platforms.
Owing to the conformality of the LbL coating, the approach reported here can also be used to control the intra-CNT spacing within the 3D element, enabling both physical and surface chemistry tailoring. Porous CNT scaffolds allow isolation of particles over three orders of magnitude in size with a single chip33: particles that are larger than the average intra-CNT spacing (∼80 nm) do not penetrate into the CNT elements and can be isolated on the features’ external surfaces via mechanical filtration and/or chemical affinity; particles that are smaller than the average intra-CNT spacing penetrate the forest and can be isolated on the forests’ internal surfaces by using biomolecular recognition, and particles that are larger than the inter-element spacing are filtered. These complementary mechanisms inherent in a single element, combined with our new ability to change the intra-CNT spacing via LbL on the nanoscale, allow for the possibility of simultaneous multiscale separation across even larger bioparticle size ranges and chemistries, toward particles ≤80 nm, such as viruses, proteins, and exosomes.
Modeling has suggested that changing the CNT elements’ geometry to optimize the flow interception at a given pressure drop and fluid volumetric flow rate enhances polymer and bioparticle interception with the CNT surfaces67. The recorded intensity via confocal imaging is likely to be conservative and may underestimate the count because of several factors: (1) Fabrication of the microfluidic devices requires inverted imaging, i.e., through the PDMS because of the opacity of silicon. (2) The PDMS cap can vary in thickness depending on soft lithography fabrication. Thus, there is higher variability across the light intensity data because of the reflection of the excitation light from the silicon substrate and light scattering through the PDMS cap. Future work should focus on changing the CNT element geometry and the substrate used for the fabrication or configuration necessary for imaging to increase the device repeatability. One possibility for enhancement is to use quartz68,69 rather than silicon as the base substrate, which would avoid optical reflection or light scattering by the PDMS.
A key future application of the materials and technique demonstrated here is to enhance capture in microfluidic devices by enhancing interaction with 3D functionalized surfaces to improve capture and detection of nanoscale bioparticles, such as viruses (<100 nm), exosomes (30–100 nm), or rare proteins (∼10 nm). Such markers contain important information about disease state and progression. As a first step toward this goal we demonstrated polyelectrolyte multilayer coating on individual CNTs in nanoporous arrays, by using PAH-Fluor and SPS. The wide-ranging functionality of assembly materials makes it possible to tailor, at the nanoscale, the physical properties including the mechanical, electrical, and optical properties of the coatings as well as chemical functionality. Our findings provide a versatile platform for functionally tailoring and grading nanoporous scaffolds for microfluidic applications. Although the work presented here demonstrates this capability by using aligned CNT (textured) scaffolds, the technique developed may also be applied to other nanoporous material scaffolds, such as carbon nanofibers, nanowires, or aerogels, to control interfaces and surfaces in both 3D and 2D interfacial configurations.
This work was supported by the MRSEC Program of the National Science Foundation under award number DMR-0819762. The authors thank the Microsystems Technology Laboratory (MTL) at MIT for their valuable time and insight into processing techniques and instruction.
This work is licensed under a Creative Commons Attribution 4.0 Unported License. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0