Threedimensionally patterned, hierarchically and anisotropically structured bacterial cellulose

Structuring cellulosic materials is an important step towards realizing emerging technologies, such as so-called engineered living materials, and improving on established ones, such as tissue engineering. In this work, we present a route for the preparation of cellulose monoliths exhibiting a three-dimensional pattern on the macroscopic scale, together with structural anisotropy in the cellulose fiber level. This was achieved by rheotactic growth, i.e. under flowing medium, of bacterial cellulose over a 3D-printed dissolvable template. The surrounding setup was realized using commercially available components. Here, we report on and discuss structural properties of cellulose monoliths obtained by this process, such as shrinkages during processing, the strut densities of 50 mg cm−3, preferred orientations of cellulose within the struts, and the pore size distributions, which were determined from nanoscale-precision silica replica.


Introduction
In nature, bacterial cellulose (BC) serves as a water-retaining protective matrix for microbial colonies. They consist of a fiber network and form a hydrogel with water. Since the microbes are typically aerobic, access to oxygen from the naturally-occurring external convection has to be ensured via short internal diffusion distances. Hence, BC matrices usually take on the shape of a film, as in the prominent example of the vinegar mother [1]. This poses a challenge for the manufacture of bulk material: as is typical for exopolysaccharide production processes, the material density and dimensions are limited by the ability to continuously and evenly aerate the growing body [2]. For this purpose, BC bodies with thicknesses on the centimeter scale or above are often produced with pores larger than those found in native BC gels.
As an example for a three-dimensionally (3D) patterned BC structure, Rühs et al produced porous BC by growing Gluconacetobacter xylinus in air-in-water foams, initially stabilized by Cremodan and Xanthan [3]. Here, the spherical air bubbles provided oxygen, and served as templates for the pores in the final material. Somewhat conversely, Caro-Astorga et al produced BC spheroids from Komagataeibacter rhaeticus, which were then molded into 3D shapes [4]. Notably, they demonstrated that spheroids containing live bacteria may serve as building blocks for engineered living materials (ELM) capable of biological sensing, signalling and of self-repair. Both studies add a level of pore sizes to the already hierarchically porous native BC structure.
The structures formed in natively porous BC hydrogels are by themselves intricate enough to have triggered a wide range of research into their usage as structural templates for carbon nanofiber aerogels, as reviewed by Wu et al [5]. Sisson outlined BC membranes as possessing 'a well-formed, natural cellulose, crystalline structure', existing 'as a uniform membrane in a highly swollen or gel condition' and containing 'about 1 g. of cellulose to 100 g. of water' [6]. He determined that BC, when supercritically dried to an aerogel and therefore largely retaining the structure of the original water-swollen state, exhibits a random orientation of the cellulose crystallites.
Generally, in biological materials, the external properties are determined by an interplay of the structural features on all levels of hierarchy. For example, in wood, mechanical properties may vary by almost an order of magnitude solely due to changes in the cellulose microfibril angle relative to the wood fiber axis [7]. Wang et al observed changes to the mechanical properties of BC films of similar magnitude when wet-drawn [8]. Hence, in order to tailor the properties of 3D patterned BC, it is desirable to obtain comprehensive control over the structuring within the BC gel itself, as well as over the superstructure created by e.g. sacrificial templating or molding. Further, for simplicity and to enhance potential economical viability, it is desirable to obtain such control during BC growth.
In earlier works, we observed preferred orientation of fibers on the nanometer scale in compact BC films [9] and the formation of superstructural patterns on the centimeter scale in BC strands [10] when grown under flowing nutrient medium, which we termed rheotactic growth. In both cases, shear flow-inducing obstacles were used as anchoring points for the forming BC. In the compact BC films, we observed considerable changes to the directional mechanical properties, arguably most importantly to the work of fracture, which increased by almost a full order of magnitude from isotropic to anisotropic BC films of similar densities [9].
We hypothesized that it is possible to combine the two aforementioned approaches to grow 3D patterned, hierarchically and anisotropically structured BC in a single process, i.e. without post-treatment such as drawing or pressing. Such comprehensive structuring is -to our best knowledge-not achievable by current additive manufacturing technology. We selected the gyroid structure, as given by equation (1) as as suitable and desirable 3D pattern.
x y y z z x t sin cos sin cos sin cos 1 ( ) + + = Here x, y, z are the normalized carthesian coordinates of the body and t is the parameter governing the boundary placement, i.e. the filling factors of the two volumes divided by the surface. In the context of the current work, we consider a gyroid material to consist of a solid phase filling one of the two volumes, as commonly found in nature [11]. Compared to other minimal surfaces, such as the primitive cell, it shows favorable mechanical properties [12], whose anisotropy and chirality depends on t [13]. Since both sides of the surface form an interconnected network, mechanical stress can be equilibriated in the solid phase when loaded. By the same logic, pressure can be equilibriated in the liquid or gas phase when permeated. This makes the gyroid suitable as a template for rheotactic growth. It also makes grown BC gyroids potentially suitable as carriers for bioactive substances, for example as part of an ELM.
In 2015, Torres-Rendon et al demonstrated the manufacture of bioactive gyroid scaffolds by templating nanocellulose hydrogels [14]. They filled a sacrificial resin template with cellulose nanofibril suspensions to obtain stable cellulose gyroids. To test our hypothesis, we expanded on their process by applying the principle of rheotactic growth onto sacrificial templates to create BC gyroid structures, which constitutes the aforementioned 3D pattern on the macroscale. We demonstrate that, due to their manner of growth, these structures also exhibit structural anisotropy on lower levels of hierarchy.
From these, we prepared the medium of Hestrin and Schramm [16], composed of deionized water with 0.5 cg/g peptone, 0.5 cg/g yeast extract, 0.27 cg/g disodium hydrogen phosphate and 0.12 cg/g citric acid, adjusted to an initial of pH 6.5 with 1 M NaOH. 2 cg/g autoclaved sucrose was used as the carbon source, as opposed to the sterile-filtered sucrose used earlier [9].
The media and all thermally stable components were steam-sterilized for at least 20 min at 121°C. The continuous cultivation of the microorganisms were kept in Petri dishes at 30°C. For cultivation on agar plates, 1.5 cg/g agar was added to the medium. For the pre-culture, a colony of a 7 days old plate was placed in 200 ml of liquid sucrose medium and incubated statically in a wide-necked flask at room temperature.

Experimental setups
For three-dimensional BC patterning, we assembled a flow-cell bioreactor of our own design as described in the following, together with commercially available standard components, Into the flow cell housing, we inserted gyroid substrates realized in SolidWorks 2019 and 3D-printed (RepRap X400 Pro V3, Feldkirchen, Germany), or commissioned for 3D-printing (3DBAVARIA, Barbing, Germany) in poly(lactic acid). The bottles, filter, tubing and flow cell were autoclaved in moist heat for at least 20 min. The 3D-printed growth components were soaked in ethanol for 3 days before the experiment in order to achieve disinfection. All components were assembled in a sterile bench.
Scaling of the body defined by equation (1) was performed by scaling x, y, z. For t = 0, as applied in this work, the two volumes are equal, with filling factors of 0.5. In preliminary experiments, we tested different inset geometries, figure 2.
All types consisted of cylinders with conical end sections. As prepared, the cylindrical sections measured 51 mm in length and 15 mm in diameter. In Type I, the entire inset was a gyroid lattice. In Type II, the central section was gyroid; the end sections were solid covers and empty on the inside. Type III was similar to II, but covered entirely by a solid mantle, and 3-D printed as one part. The 'equation' scheme in figure 2 is intended to illustrate the difference between Type II and III.
Based on the fidelity of the obtained BC gyroid, we selected the inset Type III as best suited. In the following, we report on the properties obtained by using this type of substrate.

Cultivation
Cultivation was carried out at ambient temperature of ≈22°C. The material was rheotactically grown by passing nutrient medium through the flow cell with growth substrate via the peristaltic pump, figure 1. Hence, we refer to the direction in the obtained materials parallel to the long axis of the flow cell as the former direction of flow.  During preliminary experiments, we also systematically tested different growth periods and flow rates, grown from the same pre-culturing batches in medium. Based on the material yields, we selected a growth period of one week at a flow rate of 350 ml min −1 as the standard conditions.
We added antibiotics to counteract potential contaminations from the PLA substrates. This was achieved by the addition of 1 ml/l of a 34 g/l ethanolic solution of chloramphenicol. The oxygen supply during rheotactic growth was ensured by introducing sterile-filtered compressed air at a flow rate of 1 l/min, humidified by passing through a wash bottle containing desalinated and sterile water, figure 1. The exhaust air was also passed through a sterile filter to avoid backflow of aerosols.

Harvesting and characterization
After harvesting, the PLA substrates overgrown with BC were placed in deionized water to remove the medium. Thereafter, the gyroids were detached from the gyroid substrates by dissolving the latter in dichloromethane. The solvent was stirred, and changed 3 times. The BC gyroids were then placed in dionized water to allow it to swell and regain their shapes, figure 3. Then, they were freeze-dried (Alpha 2-4 LDplus, Christ, Osterode am Harz, DE).

Polarization photography
To determine preferred structural orientations within the struts of the gyroids, samples were embedded (LR White, London Resin, Reading, UK) and sectioned to thicknesses of 30 μm along planes with normals perpendicular to their long axes (the former direction of flow), using a diamond wire saw (3032-4, Well, Le Cocle, CH). They were then photographed while illuminated in transmission by a white backlighting plate, under crossed polarizers and using a Δλ = 525 nm retardation plate with its fast and slow axes oriented at angles of 45°to the polarization directions. The samples were oriented with their long axes parallel to the orientation of the optical axes of the retardation plate in order to visualize preferred orientations parallel or perpendicular to the former direction of flow.

Electron microscopy
To visualize the internal topography of the BC gyroids, samples were broken, then sputtered with gold/ palladium (SCD050, Bal-tec, Balzers, Lichtenstein) and imaged by scanning electron microscopy (SEM, DSM 940A, Zeiss, Oberkochen, DE) at an accelerating potential of 20 kV.

Gas adsorption porosimetry
In order to perform gas sorption analyses, two samples were replicated in silica by a process known to retain the hierarchical structure of complexly porous biomass down to the single-figure nanometer level [17,18]. In short, samples were dried in ethanolic solutions of tetraethyl orthosilicate (TEOS), then calcined at 500°C. This conversion allowed us to degas the silica replica samples at 300°C prior to the measurements.
Then, nitrogen adsorption curves were recorded, and corrected for the corresponding blank measurements (Surfer Nano, Thermo Fisher, Waltham, USA). From these, the specific surfaces of the materials were determined by the method of Brunauer, Emmett and Teller from the relative pressure range 0.05 < P/P 0 < 0.2. Pore size distribution histograms in the range 1 nm < d < 100 nm were obtained by the method of Broekhoff and de Boer, based on the theory of Barrett, Joyner and Halenda.

Sample growth
Samples were obtained as monoliths after the dissolution of the PLA substrate. During drying, the samples shrank by 10% in length and 20% in diameter, figure 4. From the sample masses and outer dimensions, and assuming that the gyroid's filling factor of 0.5 was retained, we estimated the final geometric density of dried materials within the gyroid struts as 50 mg cm −3 .

Polarization photography
In polarization imaging, the path differences Δλ of objects that are sequentially stacked are additive. The spectrum I(λ) transmitted proportional to the incoming light I 0 (λ) after the second polarizer can be calculated by When observing an thin cellulosic object in addition position to a Δλ = 525 nm retardation plate, i.e. with both fast axes parallel, the total Δλ will be somewhere between 600 nm and 700 nm, and the transmitted light blue-green. In subtraction position, i.e. when the path difference of the sample relative to the retardation plate has a negative sign, the net Δλ will be somewhere between and 350 nm and 450 nm, and the transmitted light orange-red. These statements assume that the observed interference colors are of the first order. We confirmed that this is the case by omission of the retardation plate, in which case only gray tones were observed, indicating Δλ < 300 nm from the samples alone.
In cellulosic materials with uniform preferred orientations (often fibers), the fast optical axis runs parallel along the molecular axes, and therefore the fibers [19]. This effect is very prominent in crystals, with their rigid atomic arrangements. Hence, in the images presented in figure 5, local orange or blue interference colors indicate that the molecular axes of cellulose in those regions lie predominantly parallel or perpendicular to the slow axis of the retardation plate.

Electron microscopy
Electron micrographs revealed the internal structure of the BC gyroids as hierarchically porous down to the micron scale. Channels with diameters of about 10 μm are alternatively straight, or curved, figure 6(a). Micrographs taken at higher magnifications reveal that the channel walls themselves are latticeworks of fine strands, figure 6(b).

Gas adsorption porosimetry
From the two measured silica replica, we determined their average specific surface areas A = 270 ± 11 m 2 g −1 .
Similarly, their proportional pore size distribution histograms were averaged, figure 7. They show a broad distribution of pore sizes on the submicron scale, with a notable maximum at 20 nm.

Structural anisotropy
The directional sample shrinkages during drying give a first indication on the presence of overall structural anisotropy: The gyroid structure has a cubic structural symmetry, and exhibits mechanical isotropy for t = 0 [13]. Hence, one would expect the longitudinal and diametrical shrinkages to be equal. However, the latter was  This is confirmed by polarization photographs, figure 5. Their assessment is somewhat complicated by the local presence of birefringence in the resin matrix, assumed to signify curing stresses. Nevertheless, it is apparent from the majority of orange-colored cellulose that the molecular axes are predominantly parallel to the slow axis of the retardation plate, and therefore to the former direction of flow, in agreement with our observations in earlier works [10]. Notable exceptions are the portions of the gyroid structure that connect perpendicular to the former direction of flow, as sketched in the right-hand columns of figure 5. These are parallel to the fast axis of the retardation plate and appear blue, as visible for example on the upper left of figure 5(a), throughout figures 5(c), (d), or on the lower right in figure 5(e).
The scanning electron micrographs presented in figure 6 reveal the internal structure on the micron scale: Here, the layers of micro-sized tubes or 'microtunnels' described and rationalized by Gromovykh et al and exhibiting a similar periodicity, were clearly visible, figure 6(b) [20]. We found that these exhibit curvatures, depending on the location within the sample, figure 6(a).
To determine the relation between the alignment of structural features and the substrate surfaces, we prepared additional scanning electron micrographs. Even though the observed surface has been roughened by the preparation process, in the lower portion of figure 8(a) the BC appear to flow around a section of a pore that was formerly part of the growth substrate. Since here, the direction of flow was approximately bottom-to-top, it suggests that the microtunnels follow the direction of flow. In figure 8(b), two regions of different microtunnel orientations are visible. While the details of the formation of this intersection are unknown, we speculate that it  marks the end-point of growth from two directions, in this particular image from bottom to top and from right to left.
A similar alignment to the growth substrate was observed by Klemm et al and by Bodin et al [21,22]. In Bodin's BC tubes grown within silicone tubes, the cellulose was arranged in layers parallel to the external silicone template (figure 5 in cited article) [22]. Within these layers, the cellulose appeared to exhibit a random orientation (figure 3 in cited article) [22]. Under certain growth conditions, larger BC layer regions exhibit a Bouligand structure, which is yet overall randomly oriented [20].
It is therefore likely that the preferred orientations observed by polarization photography in this work do not reflect uniaxial orientations. Instead, the materials may possess uniplanar or even selective uniplanar orientations, as described by Sisson [6], within layers parallel to the gyroid growth template surfaces. Then, within each thin section, preferred orientation within the plane of the section would be observed, since most of the volumes of the out-of-plane fibers were removed by the cutting process.
There are numerous reports in the literature on the preparation and properties of aligned BC films, as compiled in an earlier work [9]. However, those studies known to us and concerned with preparing 3D patterned BC structures have not aimed at achieving -and therefore not reported on the presence of-an overall structural alignment [3,4,14]. Equally, they have not reported on the geometric densities or specific surfaces of the finally obtained materials.

Density and hierarchical porosity
The silica replication approach used in this work conserves cellulosic nanometer-scale pore structures [17,18], which one may reasonably assume would be closed by all but the most elaborate drying procedures. This is confirmed by comparison of the specific surfaces determined in this work to the slightly lower 200 m 2 g −1 determined by Liebner et al for their supercritically dried BC aerogels [23].
The pore size distribution curves in figure 7, determined for the silica replica, are remarkable similar to those determined by Li et al for pyrolyzed freeze-cast graphene oxide and BC composites (figure 2 e in cited article) [24]. These were prepared by casting dispersions graphene oxide and BC, the latter having been prepared by high shear treatment of grown pellicles. While the structural details of the dispersions are unknown, presented transmission electron micrographs of the pyrolyzed products suggest that the individual BC fibers remained largely intact (figure 1 in cited article) [24]. This supports the intuitive notion that the similar pore size distributions in Li et al and in this work represent the BC fiber-internal porosities.
Liebner et al demonstrated the preparation of BC aerogels by supercritical drying, with (assumed linear) shrinkages during processing of 6.5% and leading to densities of ≈8 mg cm −3 [23]. The determined strut densities of our gyroid-patterned anisotropic BC translate to geometric densities of the entire material of ≈25 mg cm −3 , including the large gyroid pores. The directional shrinkages determined in this work translate to a volumetric shrinkage of 42%, while the shrinkage reported by Liebner et al, if indeed isotropic and linear, would translate to a volume shrinkage of 18%. Hence, the materials prepared in this work had a cellulose density before drying that we estimate as twice as high as those prepared by Liebner et al Here, one aim -apart from achieving overall anisotropy-was to replicate the gyroid 3D pattern as true as possible after drying, which is aided by a comparatively high cellulose density. However, since the density can be tailored via growth conditions and time, it is reasonable to assume that lower-density aerogels with structural alignment may be prepared using rheotactic growth.

Possible applications and processing details
While we preliminarily experimented with a number of flow cell inset geometries and growth conditions, we did not attempt to optimize the resulting materials for any possible application, which may include the following.
Biodegradable carriers for catalytic agents: The dynamics of flow through such carriers may be tailored not only by the macroscopic porosity created by the growth template, but also via the amount of overall structural anisotropy within the pores.
Growth substrates for engineered living materials: Here, one may aim for creating materials with low strut densities. Then, the pore volumes and shapes may allow to settle defined amounts and species of bio-engineered microorganisms to perform specific sensing or conversion tasks. A topic for future research might be whether monolithically grown substrates exhibit a notable advantage in structural cohesion over substrates prepared by fusing BC spheroids [4].
Scaffolds for bone tissue engineering: These require interconnected pores to allow for the ingrowth of-and their eventual replacement by healthy tissue. In this field of application, a balance must be found between the geometrical density and mechanical properties, such as strength, elasticity and toughness. Gradient scaffolds (as also found in nature) were shown to respond to mechanical stresses in a markedly different way than such with homogeneous pore structures [25].
Templates for conversion into inorganic monolithic separation column materials: The replication into silica carried out in this work demonstrates the feasibility of such a conversion. One may apply the extensive toolkit available, as reviewed by Paris et al [26]. In this case, it will most likely be best to aim for a material with a high strut density, yet with a high specific surface area. Since inorganic phases deposit on the organic template surfaces during the impregnation step [18], this can be expected to yield an inorganic phase also with a high strut density. The impregnation with inorganic precursors is aided by the open macroscale pores arising from the 3D patterning, in this work the gyroid channels.
Hence, we have presented a generic pathway to obtain source materials for further processing. During the processing in this work, we used poly(lactic acid) as the flow cell inset, to be dissolved by dichloromethane. Both may be substituted for by other polymers and their solvents. It is also conceivable to use a biodegradable substrate as a simultaneous source of nutrient.
Generally, this work represents a foray into a promising new method of patterning biopolymeric materials. Future research may focus on any of the parameters set in this work, the most promising of which we consider to be the structure of the growth template, the choice of exopolymer-producing organism, the manner of structural analysis, and the determination of properties relevant to potential applications. For example, it may be considered to determine the method's limit with regard to the size of the repeating unit of the -assumed periodic-3D pattern.
We assume that the amount of material, as well as the dimensions of individual monoliths can be increased by simply enlarging the flow-cells. Since the materials are interpenetrated by the network of macroscopic pores defined by the inset, material transport routes that are relevant for further processing, e.g. during the dissolution of the substrate, or mineralization, or drying, are largely independent of the outer geometries of individual monolithic bodies.
For drying, we discarded air and furnace drying for the large amount of shrinkage they incur. Since we found freeze-drying to lead to similar shrinkages as supercritical drying we opted for the former for better economy. It may be argued that supercritical drying leads to a better retention of the smallest structural details, as exemplified by the work of Liebner et al [23]. In this work, these were retained and measured by nitrogen sorption analysis via the silica replication approach.

Conclusion
In this work, we outlined a route to obtain 3D patterned, and hierarchically and anisotropically structured BC monoliths. We expect that further systematic work on their mechanical properties as functions of structural details will yield further unique properties of interest: current research into the -potentially chiral-mechanical behaviour of gyroids assumes isotropy of the materials within the struts [13]. Incorporating a texture into the struts themselves can be expected to yield exciting anisotropic behaviour.
Also, it can be expected that materials grown into a single porous monolith exhibit a larger amount of structural cohesion. This is the case in this work, and in the foam-templating approach followed by Rühs at al. [3], and in contrast to the cellulose nanofibril-or BC spheroid templating approaches applied by Torres-Rendon et al and Caro-Astorga et al [4,14]. Naturally, in any comparison to be drawn, possible advantages in structural cohesion would have to be weighted against a number of other considerations, such as ease of processing and non-mechanical properties of the final materials.