An interfacial self-assembling bioink for the manufacturing of capillary-like structures with tuneable and anisotropic permeability

Self-assembling bioinks offer the possibility to biofabricate with molecular precision, hierarchical control, and biofunctionality. For this to become a reality with widespread impact, it is essential to engineer these ink systems ensuring reproducibility and providing suitable standardization. We have reported a self-assembling bioink based on disorder-to-order transitions of an elastin-like recombinamer (ELR) to co-assemble with graphene oxide (GO). Here, we establish reproducible processes, optimize printing parameters for its use as a bioink, describe new advantages that the self-assembling bioink can provide, and demonstrate how to fabricate novel structures with physiological relevance. We fabricate capillary-like structures with resolutions down to ∼10 µm in diameter and ∼2 µm thick tube walls and use both experimental and finite element analysis to characterize the printing conditions, underlying interfacial diffusion-reaction mechanism of assembly, printing fidelity, and material porosity and permeability. We demonstrate the capacity to modulate the pore size and tune the permeability of the resulting structures with and without human umbilical vascular endothelial cells. Finally, the potential of the ELR-GO bioink to enable supramolecular fabrication of biomimetic structures was demonstrated by printing tubes exhibiting walls with progressively different structure and permeability.


Introduction
Bioinks are increasingly playing a leading role in the quest for recreating the structures and functions of biological systems [1]. These materials are being engineered to push biofabrication boundaries such as the development of thick vascularized structures [2], precise complex geometries [3], or multicellular environments [4]. However, in spite of these advances, important challenges remain such as the capacity to replicate the molecular diversity, structural hierarchy from the nano-to the macroscale, physiological biomechanical properties, and ultimate functionality of the extracellular matrix (ECM) [5] and corresponding tissues [6,7]. Consequently, it is essential to develop innovative bioinks based on molecular and nanoscale building blocks capable of organizing hierarchically [8] and enabling communication between cellular and extracellular components [9]. Nature has evolved to grow and regenerate tissues and organs by organizing, from the bottomup, multiple types of molecular components such as proteins, lipids, and carbohydrates [10]. Inspired by this, multicomponent self-assembly is being explored to grow materials with multiple types of molecular building-blocks exhibiting enhanced structural complexity and functionality [11]. These materials offer a unique opportunity to recreate chemical [12,13] and mechanical features [14,15] of the ECM and selectively communicate with cells [16]. The capacity to develop bioinks exhibiting such properties is an exciting possibility to immediately present cells [17] with complex, yet rationally designed, ECM-like milieus [18] and signals at cellular and multicellular levels [19]. Pioneering studies based on for example self-assembling peptides [17,20,21] or nanoparticle/ polysaccharide co-assemblies [22] are demonstrating this possibility. However, the widespread use of selfassembly in additive manufacturing has been limited in large part because of inadequate standardization ensuring sufficient structural integrity, fast enough speed of setting, and scalable manufacturing [23].
Liquid-in-liquid printing, the inoculation of one low viscosity liquid phase into another [24], offers an opportunity to design bioink materials that integrate the benefits of multicomponent self-assembly with additive manufacturing in a rationally designed manner. For example, taking advantage of the interface between oil and water, nanoclay-polymer surfactants have been self-assembled into semi-permeable membranes and perfusable capillaries [24]. Also, printing a dextran water solution into a polyethylene oxide water solution enabled rapid fabrication of biocompatible tissue-like 3D architectures such as capillary networks and cardiac structures [25]. Furthermore, while traditional extrusion-based 3D printing is defined by the printing nozzle dimension, bioinks for liquid-in-liquid printing can be used to immediately set and generate structures upon liquid mixing [26], from the bottom-up, facilitating fabrication of high resolution structures. In addition, liquid-in-liquid bioinks can also be used to trigger emerging phenomena such as compartmentalization [27] and diffusion-reaction processes [28], which are used by biological systems such as cells and organelles to regulate molecular transport and assembly into functional higher-order structures. Taking advantages of these opportunities, we have recently reported on a multicomponent self-assembly mechanism harnessing disorder-to-order transitions of an elastin-like recombinamer (ELR) to non-covalently co-assemble with graphene oxide (GO) [29]. We have demonstrated the possibility to use this system in liquid-in-liquid bioprinting to enable fabrication of perfusable fluidic devices. However, it is essential to optimize printing parameters and establish reproducible processes that can demonstrate its use as a functional bioink and its advantages to fabricate physiologically-relevant functional structures.
In this study, we perform thorough top-down/bottom-up standardization of our ELR (ELK1)-GO co-assembling system to serve as a liquid-in-liquid bioink and expand on its capabilities for the fabrication (figure 1(a)) of functional capillaries down to ∼10 µm in luminal diameter (figures 1(c) and (e)) and ∼2 µm in wall thickness ( figure 1(d)). Through experimental and cross-platform finite element analysis (COMSOL MultiPhysics® v5.5, COM-SOL Group, Sweden), we demonstrate how these perfusable structures exhibit a spectrum of structural and biofunctional properties that are not possible for most bioinks including sub 10 µm size features, the capacity to pulsate, tuneable permeability with and without cells, and permeability gradients within a single structure.

Chemicals
Rhodamine B (⩾95%, HPLC grade) and paraformaldehyde (95%) were obtained from Sigma-Aldrich (St. Louis, Germany). GO (product number 777676) was obtained from Sigma-Aldrich. Alexa Fluor ™ 488 NHS Ester (product number A20000) was obtained from Thermo Fisher Scientific (Waltham, USA). FITCdextran 40 kDa (product number FD40), 20 kDa (product number FD20s), and fluorescein sodium salt (product number 518-47-8) were obtained from Sigma-Aldrich. 24 ] molecule was provided by TP Nanobiotechnology (Valladolid, Spain), synthesised and purified by Escherichia coli recombinant expression system. The sequence and molecular weights of the polymers were verified using amino acid analysis. SDS-PAGE and MALDI-TOF SIMS were used to carry out the ELK1 characterisation.

Sample preparation (ELK1-GO system)
Aqueous suspension of GO (0.1 wt%, 100 µl) was added to a well of 96-well TCP and aqueous solution of the ELK1 (2 wt%, 18 µl) was slowly injected into the suspension of GO. The tip of the pipette was allowed to contact the bottom of the well before releasing the ELK1 solution vertically at a constant speed. All samples were prepared in MilliQ water.

Diffusion-reaction of ELK1-GO tube wall formation
The diffusion-reaction of ELK1-GO tube wall formation is critical and determine the increase of tube wall thickness over time. An imaging system was used to observe tube formation and its behaviour Figure1. Rationale, assembly, and structure of the ELK1-GO bioink. (a) (1) The bioink assembles upon liquid-in-liquid printing of an ELK1 (brown) solution comprising cells (green) into a solution of GO flakes (red), spontaneously triggering a diffusion-reaction process at the solution interface and leading to the formation of a multi-layered membrane. (2) The process enables trapping of cells within the membrane and immediate localization within the lumen of the resulting tubular structures. (b) Scanning electron microscope (SEM) image of the cross-section of an ELK1-GO tubular structure depicting the resulting multi-layered architecture. (c)-(e) SEM images of an ELK1-GO capillary with an internal diameter of ∼10 µm and a tube wall thickness of ∼2 µm. (f) Time-lapse imaging of an ∼1 mm diameter tube being perfused with a pulsatile (the pulse frequency was 1 Hz) flow at 12 ml min −1 depicting the change in tube diameter (white arrow) as a result of the material's resilient properties. (g) Time-lapse imaging of a 1 mm diameter tubular structure bent 180 • and released. over time. The system consisted of a Nikon camera with a 1:1 macro lens, a mirror to reflect the image on the lens of the camera and a light source powered at 30 V (figure S1 (available online at stacks.iop.org/BF/13/035027/mmedia)). A time lapse was used to observe the membrane self-assembly.

Confocal microscopy
The interaction and localization of ELK1 and GO was probed using laser scanning confocal and multiphoton microscopy (TCS SP2, Leica Microsystems, Germany). ELK1 (2 wt%) was dissolved in an aqueous solution of Alexa Fluor ™ 488 NHS Ester (10 −6 wt%) and GO were diluted to 0.1 wt% with an aqueous solution of Rhodamine B (10 −6 wt%). All solutions were incubated for 20 min at 30 • C and protected from light. Images were acquired at laser wavelengths of 488 nm and 543 nm which correspond to the excitation wavelength of Alexa Fluor and Rhodamine B, respectively. Images were further processed using ImageJ.
All samples were subjected to critical point drying (K850, Quorum Technologies, UK) prior imaging. The SEM micrographs were captured on Inspect F50 (FEI Comp, the Netherlands) after sputter-coating with gold (10 nm thick). All samples consisting only ELK1 or GO were prepared for SEM imaging without a prior cross-linking process.

3D printing of ELK1-GO materials, perfusing, and pulsating
A commercially available 3D printer (3D DISCOVERY ™ , RegenHU, Switzerland) was applied for the 3D printing of ELK1-GO materials. An 85 µm diameter nozzle was used to release the solution of the ELK1 (2 wt%) under 0.02-0.10 MPa pressure at a range of speed between 10 and 100 mm s −1 . The printing nozzle was merged in a container with 0.1% GO MilliQ water suspension. A peristaltic pulsatile pump was used to perfuse 1 v/v green food dye in MilliQ water. The perfusion speed was 12.5 ml min −1 and the pulse frequency was 1 Hz.

Fabrication of capillary-based permeability testing device
The device was produced by pouring a first base layer of polydimethylsiloxane (PDMS) (1 mm thickness) in a Petri dish. After curing at 60 • C for 1 h, a metal needle of 0.8 mm outer diameter was placed on the first layer and two pins for subsequent alignment of the device. A second layer of PDMS was then poured and cured in the Petri dish. The level of PDMS in the second layer had to cover the needle. Petri dish was then placed in an oven at 60 • C for 2 h to thoroughly cure both PDMS layers. Removal of tube with forceps leaves a cylindrical cavity with a diameter of 0.8 mm. The PDMS with cavity was cut from centre and divided into two chambers. A flat membrane from an ELK1-GO tubular structure was inserted into the middle of the device and divided the channels of the left and right chambers. A test solution was injected into the right chamber, while the same solvent of the test solution was injected into the left chamber ( figure S2(a)).

Permeability measurements
Two substances were used to measure permeability: fluorescein sodium salt and FITC-dextran. These two solutions were used to measure, respectively, the permeability of sodium ions and of macro molecules (20 kDa and 40 kDa). MilliQ water (phosphatebuffered saline) was injected into the left chamber and a water solution of FITC-dextran or Fluorescein sodium salt was added to the right chamber. The concentration of GO for the formation of ELK1-GO membranes was changed to observe the variations of the permeability across the membrane. The GO concentrations considered were: 0.05%, 0.10%, and 0.15%. An epi fluorescence microscope (Leica DMI 4000B, Epifluorescence Microscope with a LEICA DFC 300FXCCD camera) was used to determine the passage of molecules from one chamber of the device to another one. Data collected from the intensity analysis were described through following model: where V1 was the volume of the left chamber, C1 was the concentration of molecules inside the left chamber, V2 was the volume of the right chamber, C2 was the concentration of molecules inside the right chamber; L and A were the membrane thickness and area respectively and D was the diffusion constant. Then to determine the diffusion constant D, the experimental data were fitted with an exponential equation (equation (2)) with Matlab.
Once the diffusion constant D of the equation (equation (2)) was determined, it was compared with the argument of the exponential in model (equation (1)).
The permeability constant (P) was calculated based on equation (4).
where K is the partition constant, D is the diffusion constant, and ∆X is the thickness of the membrane.

Cell culture
GFP-Human umbilical vein endothelial cells (GFP-hUVECs) (Fisher Scientific, Angio Proteomie GFP-hUVECs, NC0601093, USA) were cultured in EGM ™ -2 Media (Lonza, CC-3156 and CC-4176). The medium was changed every 3 d until the cells reached 80% confluency. GFP-hUVECs between passages 2 and 4 were used for experiments. The tubes were first washed three times with PBS 8 h after assembly and sterilized with UV for 45 min. Then each tube was placed in a well of 48-well cell culture plate with inner or outer side facing up. The EGM ™ -2 Media (500 µl) containing different cell densities (5 × 10 3 , 1 × 10 4 , 5 × 10 4 , and 1 × 10 5 ml −1 ) were added to each well containing ELK1-GO membranes. The coated wells were incubated for 8 h prior to cell seeding. The cells were incubated at 37 • C and 5% CO 2 for 48 h for permeability test (figure S2(b)).

Finite elements analysis
Finite elements method simulations were carried out using COMSOL MultiPhysics® v5.5 software (license number is N. 13075339). All the presented models were developed using the Transfer of diluted species module.

Permeability model
The aim of the first model was to identify the diffusion coefficient of the membrane to dextran and sodium salt starting from experimental data. A twodimensional (2D) axial symmetric transient model was used. Different time points were set for dextran (0 h,1 h, 3 h, 5 h) and sodium (0 min, 20 min, 40 min, 60 min) according to experimental data previously acquired. The model consists of three domains (figure S3(a)). In each domain, transport properties are governed by the following balance equations (5) and (6): where C i is the concentration (mol m −3 ), D i the diffusion coefficient and R i the reaction rate. D i = 1 × 10 −9 m 2 s −1 for domains 1 and 3 (same as water) while it was considered as a parameter to be estimated in the membrane (domain 2, parametric sweep study). On the right side of the membrane (domain 1) the initial concentration (t = 0 h min −1 ) of salt was 1 M and 0.001 M for dextran and sodium, respectively (C = 0 M for other domains). A transient study allowed the evaluation of concentration profile in the whole chamber across the membrane. All external boundaries were set to no flux condition (equation (7)).
For each image (acquired by epifluorescence microscope), a normalized plot profile was obtained on the centre line using ImageJ® software. The comparison between the model and the real images was carried out considering a single point in the left chamber (1 mm from the left side) (figures S3(b) and (c)). The normalized value of the grey level from the image was compared with the normalized value of the concentration in the same point of the model at different diffusion constants for each time point. Best fitting was evaluated by calculating the adjusted R-squared parameter discarding values corresponding to the membrane in both profiles (from 2 to 2.5 mm). Results are shown in figure S4.

Reaction model
The aim of this model was to identify the time of reaction between two species (aqueous solutions of 2 wt% ELK1 in 0.1 wt% GO  S3(d)). The two species will react creating a third C3 species. The transport properties of the species are characterised by their diffusion constant in the different domains. We made two hypotheses: (a) Diffusion constant of ELK1 (C2) and GO (C1) decreases linearly as C3 is formed according to equation (8): where D is the diffusion at generic instant t, D f is the final value of diffusion constant with a stable membrane, D 0 is the starting diffusion constant before the membrane formation, c is the concentration of species C3 at generic instant t, c f is the concentration of the stable membrane. According to permeability model for dextran, the presence of the membrane causes a diffusivity drop of at least three orders of magnitude. We need to estimate concentration at which the membrane can be considered stable. Considering 1.5% ELK1, the protein is totally consumed forming a stable membrane, with a resulting concentration c f = 0.3 mol m −3 . The starting diffusion coefficient (D 0 ) for proteins is about 10 −9 m 2 s −1 , the final will be similar to that of dextran (D f = 1 × 10 −12 m 2 s −1 ). The final value of GO diffusion coefficient was estimated with a parametric simulation (D f = 1 × 10 −13 , 1 × 10 −15 , and 1 × 10 −17 m 2 s −1 ; starting D 0 for GO was considered as typical values for large molecules 7 × 10 −11 ). On the basis of these considerations, the diffusion constant was modelled on COMSOL MultiPhysics® v5.5 with the following relation: A reaction was introduced in this model in both domains. Reaction conditions were set as follows: The previous equations assume that: (a) reaction rate is 200 m 3 (mol · s) −1 ; (b) since stoichiometric ratio is ELK1:GO = 20:1, C2 decreases 20 times faster than C1; (c) C3 increases as fast as C1 is consumed.
On the basis of these assumptions, the reactions were modelled in COMSOL MultiPhysics® v5.5 with the following equations: where the expression [ * (C 1 > 0) * (C 2 > 0)] means that the reaction happens only if C1 and C2 are present at the same time.
A transient study was computed over 800 s time evaluating how C3 is formed at the interface between C1 and C2 domains. In order to obtain a more accurate solution a user-controlled mesh was introduced by setting the maximum element size to 0.005 mm. Results are shown in figure S5.

Gradient model
The aim of this model was to identify the gradient concentration trend over time of three different GO solutions placed in a 6 cm diameter Petri dish. The three solution were 0.05%, 0.10%, and 0.15% GO w/v respectively, corresponding to 0.25, 0.5 and 0.75 mol m −3 (MW of GO is 2043.8 g mol −1 ). A 2D transient model consisting of four domains was developed: domain 1 corresponds to water, while 2, 3 and 4 are the three different GO solutions placed with increasing concentration value ( figure S3(e)). Diffusion constant was set to D i = 1 × 10 −9 m 2 s −1 (same as water). Concentration trend over a 96 h time was evaluated on a cut line placed in the middle of the Petri dish (red line in figure S3(e)).

Rationale of the study
Our previous research demonstrated that, upon printing a solution of ELK1 into another of GO ( figure 1(a)), hydrophobic and electrostatic interactions trigger ELK1-GO co-assembly and an interfacial diffusion-reaction process that gives rise to a multilayered membrane ( figure 1(b)). In the current work, we study in more detail the diffusion-reaction process to better control the ELK1-GO co-assembly, formation of the tubular structure, and the properties of the interfacial membrane including thickness, porosity, and permeability. By modulating the way in which both solutions are mixed, it is possible to spatially control this process and generate hierarchical tubular structures by integrating bottom-up assembly with top-down manufacturing. Furthermore, by standardizing together both the printing and co-assembling parameters, the system enables control over the printability, shape fidelity, and ultimate feature resolution (i.e. tube diameter and tube wall thickness) of the bioink. Through this approach, the resulting material (i.e. tube wall) can be designed to exhibit a variety of physiologically-relevant properties such as tuneable thickness (down to ∼2 µm), stiffness, permeability, and the ability to be robust and compliant, evidenced by the capacity to pulsate under physiological fluid flow (i.e. shear stress of ∼0.26 N m −2 , which is within the range of that present in carotid arteries [30]) (figure 1(f) and movie S1) and be bent without displaying any kinks (figure 1(g) and movie S2). Consequently, we first conducted experiments and simulations to characterize, standardize, and optimize the printing parameters. We then focused on further controlling the printing fidelity and material properties by modulating the concentration of GO. Given the potential use of the bioink to fabricate perfusable fluidic systems and structures that can recreate biological ones, we then conducted a thorough characterization of tube wall permeability with and without cells as well as developed ways to fabricate structures with tuneable and anisotropic permeability.

Standardization and optimization of the printing process
Standardization and optimization experiments were conducted by integrating top-down (3D printing) and bottom-up (self-assembly) fabrication parameters to maximize printing fidelity and control over resulting geometric features.

ELK1-GO diffusion-reaction process and printing fidelity
We conducted experiments and simulations to characterize bioink shape fidelity as determined by the diffusion-reaction mechanism triggered upon ELK1in-GO printing. First, an 18 µl aqueous solution of ELK1 (2 wt%) was manually injected into an aqueous suspension of GO (0.10 wt%, 100 µl) in a 96-well plate to generate a vertical tube. The process was video recorded (figure S1) and imaged at different timepoints to investigate the kinetics of tube wall formation (figure 2(a)) and movie S2). Separately, a 'transport of diluted species' COMSOL model simulating similar experimental conditions was generated and used to validate the experimental results. Both experimental and simulation results were in agreement and revealed that the ELK1-GO tube exhibited a similar diameter as that of the starting ELK1 injected drop and the ELK1-GO tube wall thickness did not increase after ∼60 s of interaction ( figure 2(a), figure S5). These results demonstrate that the initial volume of the ELK1 injection determined the outside dimension of the final printed structure (figure 2(a)) and that this dimension was not affected by the growth of the ELK1-GO interfacial tube wall as a result of its diffusion-reaction process.

Tuneable geometry by varying printing parameters and GO concentration
The printability of the ELK1-GO bioink was characterized using a commercially available 3D printer (3D Discovery ™ , RegenHU, Switzerland) with adjustable nozzle heights and diameters, dispensing pressures, and printing speeds [31]. By fixing nozzle height and diameter, we assessed the relation between dispensing pressure and printing speed. The fidelity of fabricated ELK1-GO tubes was assessed by optically and systematically measuring tube diameters at different locations [32], which revealed increased printing fidelity with increasing printing speed (figures 2(b) and (c)). Larger diameter differences within a single tube led to larger variances, which resulted in rugged tube walls ( figure 2(b), printing speed: 10 mm s −1 ). By varying the dispensing pressure of the ELK1 solution between 0.02 and 0.10 MPa and the printing speed (PS) between 10 and 100 mm s −1 (nozzle height and diameter were fixed, figure 2(b), table), we found that tubes were generated when the dispensing pressure was between 0.04 and 0.08 MPa. When printing within this pressure range and printing speeds above 30 mm s −1 , variances in tube wall geometry within a single tube were under 5%, which reflected high fidelity tubes with smooth walls ( figure 2(b)). Utilizing a dispensing pressure of 0.06 MPa and printing speed of 30 mm s −1 , we were able to print tubes of up to 12 cm in length and 2 mm in diameter. It is important to mention, however, that the limitation on tube length was not lack of tube integrity but rather the size of the plate used to hold the GO solution during printing.
Printing fidelity was also characterized by varying the concentration of the GO suspension ( figure 2(c)). In this case, lower concentrations of GO (0.05 wt%) generated tubes with lower printing fidelity and thinner tube walls (∼40 µm), while higher GO concentrations (0.10 and 0.15 wt%) resulted in tubes with higher printing fidelity and thicker walls (∼60 µm) (figures 2(d) and (e)). We speculate that this difference in wall thickness results from interfacial membranes forming with varying levels of permeability as a result of having different concentrations of GO. We have demonstrated that upon initial ELK1-GO co-assembly, a diffusion barrier is formed, which defines its downstream diffusionreaction mechanism that generates the distinctive multilayer membrane [29]. We reason that lower concentration of GO (0.05 wt%) leads to a more permeable diffusion barrier, which allows for higher and faster ELK1 diffusion and consequently thinner multi-layered membranes. In contrast, higher concentration of GO (0.15 wt%) upon co-assembly results in the formation of a less permeable diffusion barrier, which results in thicker membranes.

Tuneable porosity by varying the concentration of GO suspension
Having characterized printing reproducibility and fidelity, we then focused on exploiting the coassembling nature of the bioink to control ELK1-GO tube wall porosity by varying the concentration of GO. Tubes were first co-assembled with varying GO concentrations (0.05 wt%, 0.10 wt%, and 0.15 wt%) and cut to extract flat membranes (the tube wall) (figure S1(a)). The luminal (inner) and abluminal (outer) sides of these membranes were then examined by SEM following standard preparation protocols [29]. As expected, increased GO concentrations led to decreased porosity, which was easily noticeable on both inner and outer surfaces of the membranes ( figure 3(a)). Upon closer inspection, significant differences in porosity between the inner and outer surfaces were observed on all membranes, independently of GO concentration ( figure 3(b)), which was also confirmed by histological sectioning (figure 3(d)). These results demonstrated the possibility to modulate the ELK1-GO tube wall pore sizes from 0.5 to 1.2 µm on the inner surface and 1.6-4.0 µm on the outer one by modifying GO concentrations ( figure 3(c)).
Together with the previous results of printing fidelity and tuneability of geometry and porosity, we can conclude that higher GO concentration and printing speed (50 mm s −1 < PS < 70 mm s −1 ) can be used to fabricate higher fidelity tubes with smooth and thicker (∼60 µm) fully formed tube walls but less porosity after 60 s of co-assembly. By contrast, lower GO concentration and printing speed (30 mm s −1 < PS < 50 mm s −1 ) would generate lower fidelity tubes with rugged and thinner (∼40 µm) tube walls exhibiting higher levels of porosity. Importantly, upon ELK1-GO co-assembly, the thickness of the resulting interfacial structures depends on the time-dependent diffusion-reaction process of formation ( figure 2(a)). Consequently, given the high ELK1-GO affinity, it is possible to stop the co-assembly process at different time points, resulting in interfacial membranes of different thicknesses. We found that it was possible to stop the coassembly process after just 15 s by removing the fabricated tube from the GO suspension, resulting in tube wall thicknesses down to ∼2 µm when printing at >70 mm s −1 . Furthermore, tube wall thicknesses were observed to reach a maximum of ∼20 µm in thickness (figure 2(c)) after ∼60 s of co-assembly, which did not seem to change after longer periods of time. We speculate that these thin tube walls result from the low ELK1 volumes released upon printing at fast printing speeds (>70 mm s −1 ), offering an opportunity to reproducibly fabricate with resolutions down to a few microns in size (figures 1(c) and (d)). In summary, these results demonstrate how bioprinting fidelity and resolution can be tuned through both printing parameters from the top-down and self-assembling events taking place from the bottom-up.

Assessment of applicability
After standardization of fabrication parameters, we then focused on demonstrating applicability and key advantages of the bioink including the capacity to tailor permeability, display pulsatility, and fabricate structures exhibiting anisotropic permeability.

Tuneable permeability by varying porosity
The possibility to control tube wall porosity opens an opportunity to fabricate membranes, capillaries, and other structures with tuneable permeability that could be designed to recreate that of biological tissues. For example, proper vascular permeability is of outmost importance in many biological processes such as regulating diffusion of nutrients and waste products, serving as a semipermeable barrier between blood contents and tissues to screen toxic molecules, and regulating the microenvironment of the ECM [33]. Furthermore, diffusion across vascular and lymphatic tissues is vital to enable appropriate filtration, protection, and delivery of key solutes, cells, and factors. The capacity to develop vascular structures or fluidic devices with tuneable permeability is critical to recreate and optimize biofunctionality. To assess the effect of tube wall porosity on its permeability, we designed a simple capillary-based permeability testing device where the membrane (cut from the ELK1-GO tube) can be positioned in the middle, leaving two free chambers on either side ( figure S2(a)). The chambers can be filled with any solution, enabling diffusion of solutes from one side to the other ( figure S2(a)). We tested the permeability of the membrane to fluorescein sodium salt and 20 kDa FITC-dextran in water solution to recreate physiological charge and size barriers, respectively. These two functions are normally observed in vascular tissue [33]. Using epifluorescence microscopy imaging, permeability constants were calculated based on the fluorescence intensity profiles recorded on the chamber where the solutes diffused to ( figure 4(a)). Specifically, exponential curves describing the trend of the average intensity within the chamber as a function of time were calculated for each ELK1-GO membrane and the permeability constant measured based on a diffusion constant using a fitting plot [34] (figure 4(b)).

Tuneable permeability after endothelialisation
In our previous study, we demonstrated how the ELK1-GO self-assembling system can be incorporated within liquid-in-liquid printing to biofabricate tubular structures comprising cells exhibiting high viability and growth for at least 7 d [29]. To demonstrate the functionality and biological relevance of the bioink, we conducted similar permeability tests but after endothelialisation of the lumen of the ELK1-GO tubes. This setup enabled us to better account for the localization of the cells and their effect on tube-wall permeability. Different densities (5 × 10 3 , 1 × 10 4 , 5 × 10 4 , or 1 × 10 5 ml −1 ) of hUVECs were seeded on the luminal side of membranes (cut from ELK1-GO tubes) (0.10 wt% GO) placed on Petri dishes ( figure S2(b), top). We chose the method of seeding hUVECs after tube formation instead of directly co-assembling with cells to more precisely account for the localization of the cells, their interaction, and the resulting effect on permeability. As expected, after 48 h of culture, membranes exhibited different hUVEC densities including ∼30%, 50%, 80%, and >95%, as measured by confocal microscopy imaging (figure 5(a), confocal images). Then, these membranes were inserted into the permeability test device with hUVECs facing one of the chambers ( figure S2(b), bottom). Then, 40 kDa FITC-dextran [39] in phosphate-buffered saline (PBS) was flowed into the chamber facing the cells while PBS alone was flowed into the other. Here, we used 40 kDa FITC-dextran instead of the 20 kDa FITC-dextran as it is widely used to test biomaterial permeability [39]. The results demonstrated that membrane permeability tended to decrease with increased cell density, with membranes without cells (control) exhibiting the highest permeability constant (2.2 × 10 −6 cm s −1 ) while those with the highest cell density exhibited the lowest permeability constant (6.5 × 10 −7 cm s −1 ) (figures 5(b) and (c)).
In conclusion, the presence of hUVECs growing on the ELK1-GO tube wall significantly influenced the transport of solutes in a way that is dependent on the level of hUVEC confluency. We speculate that lower cell densities led to lower number of endothelial cell junctions, which in turn led to higher permeabilities. These results further support the possibility to tune structure permeability and, in particular, to do it through the presence of cells. For example, the permeability constant of ELK1-GO-hUVEC membranes (∼1 × 10 −6 cm s −1 ) with confluence levels above 50% were comparable with that of tumour micro vessels (1.8 × 10 −6 cm s −1 ) [40].
The possibility to rapidly and controllably engineer living capillary-and vascular-like structures remains a major unmet challenge in biofabrication, tissue engineering, and in vitro models. Feng et al have demonstrated how liquid-in-liquid printing can be used to fabricate perfusable tubular structures [24] but the system relies on aqueous nanoclay dispersions and solutions of H2N-PDMS-NH2 in toluene, which prevents the use of cells and recreation of key anatomical and physiological features. On the other hand, perfusable channels within bulk gel structures have been reported by Skylar-Scott et al [41], but these structures fail to recapitulate capillary features such as an outside surface and thin wall thicknesses, which consequently affect biomechanics and diffusion kinetics. Our system aims to overcome some of these challenges by providing a facile way to manufacture biohybrid capillary-like structures with physiological properties.

Fabrication of tubular structure with anisotropic porosity and permeability
To further demonstrate the potential of the selfassembling ELK1-GO bioink for the biofabrication of complex structures with innovative properties, we printed tubes but now exhibiting different porosities and corresponding permeabilities within the same structure along the tube axis (figure 6(c), illustrator). These types of structures are observed in biological systems such as renal tubules [42] and arteriovenous networks [43]. To our knowledge, these structures have not been fabricated in a single printing process. We first designed an experimental setup that enabled the formation of a MilliQ water solution made with different concentrations of GO contiguously increasing from 0.15 wt% on the left side, 0.10 wt% on the middle, and 0.05 wt% on the right side ( figure 6(a)). The GO was stained with Rhodamin B prior to printing of the ELK1 to enable GO visualization in the tube wall. In this setup, we found that the gradient of the GO in solution was maintained for at least 3 h, which was confirmed by a 'transport of diluted species' COMSOL model ( figure 6(a)) and facilitated the printing process. Upon extrusion of the ELK1 solution into the GO solution from the lower to the higher concentration of GO, the tube formed immediately at the interface of both liquids with the tube wall progressively assembling with increasing concentrations of GO ( figure 6(b)). The resulting tubular structure was then characterized via epifluorescence microscopy, which revealed a tube wall with increasing content of GO and consequently a porosity gradient (figure 6(c), red). To test the effect of this porosity gradient on the permeability of the overall tube, 20 kDa FITC-dextran solution was perfused through and, as expected, its diffusion progressively decrease along the tube direction as the GO concentration increased (figures 6(c), green and (d)). These kinds of anisotropic structures are prevalent in biological systems such as in renal tubules [42] and arteriovenous networks [43]. Our bioink enables the fabrication of such anisotropic vascular structures in a simple manner.

Conclusion
Self-assembling bioinks offer unique opportunities to the field of biofabrication. However, it is essential to establish reproducible processes, optimize printing parameters, and demonstrate their ability to fabricate novel structures with anatomical and physiological relevance. Here, we have standardized and optimized a multicomponent self-assembling bioink and demonstrated its potential to biofabricate structures with innovative properties. Through experimental work and simulations, we have confirmed the possibility to use the ELK1-GO bioink to fabricate a single tubular structure with hierarchical organization and resolutions down to ∼10 µm in diameter made with ∼2 µm thick walls. These structures enable immediate perfusion after assembly and exhibit physical properties such as stiffnesses and the capacity to pulsate that resemble physiological structures. In addition, the bioink permits printing fidelity and control over structure porosity and permeability by modulating the concentration of GO and the possibility to guide this process to fabricate structures with permeability gradients. Furthermore, the bioink is biocompatible and facilitates hUVEC bioprinting, growth, and proper function, enabling further control of permeability. However, to fully exploit these advantages, it is of outmost importance to incorporate the printing/self-assembling processing and structures reported here within functional devices such as tissue engineered constructs or organ-onchip devices. Nonetheless, the current work moves us a step closer towards the scalable manufacturing of structurally complex capillary-based constructs that can be designed to recreate the structural and permeability properties of native tissues within engineered devices.