Optimising aerosol jet printing of collagen inks for enhanced piezoelectricity and controlled surface potential

Collagen is a highly versatile protein used in tissue engineering constructs and as a model piezoelectric biomaterial. The piezoelectricity of collagen can be enhanced through the alignment of collagen domains and fibres, although most fabrication techniques used to form dense collagenous constructs do not allow for significant collagen alignment. The use of aerosol jet printing (AJP) mitigates the limitations of using soluble collagen inks for bioprinting or extrusion-based 3D printing, particularly if microfibrillar collagen suspensions are used as a cost-effective and scalable ink source. In this work, Type I and Type II microfibrillar collagen from different anatomical sources were successfully deposited using AJP with two different atomisation methods, namely pneumatic-AJP (p-AJP) and ultrasonic-AJP (u-AJP). The printing parameters were optimised for their piezoelectric amplitude and surface potential. Fourier transform infrared spectra of the films revealed that ultrasonic atomisation did not cause notable denaturation of collagen, although the process resulted in the fractionation and preferential deposition of the oligomeric and gelatinous components within the slurry. The printed collagen samples displayed a piezoelectric response that was four times higher than the values obtained from drop-cast collagen films, and their surface potential was found to be positively correlated to the roughness of the films which can be controlled through the mode of atomisation. These results indicate the ability to enhance and control the piezoelectricity and surface potential using p-AJP and u-AJP, which offers a promising physical modulation technique to tailor cell adhesion, proliferation or differentiation for collagen-based tissue engineering constructs.


Introduction
Collagen is the most abundant protein in mammalian tissue [1] and is found in the extracellular matrix of a wide range of tissues with different physical and biochemical properties. Collagen exhibits versatility in its biological and physical attributes due to its hierarchical structure [2], whereby properties can be tuned through changes in the amino acid sequences at the molecular level, or in the arrangement of collagen fibrils and fibre bundles at the microscale. Type I collagen is the most abundant form found in tissues with high mechanical loading such as bones, tendons and muscles. This is closely followed by Type II collagen, which is found in softer tissues such as the cartilage [3]. The versatility and ubiquity of collagen in the human body offers remarkable potential for tissue engineering strategies in the form of membranes, scaffolds and gels.
Collagen is also of particular scientific interest due to its inherent piezoelectricity as a biopolymer. The piezoelectricity of collagen has been well-studied in the literature since initial investigations by Fukada and Yasuda in 1964 [4]. Fibrillar rat tail collagen in particular has been reported to have the highest shear piezoelectric constant (d 14 ) measured in a biomaterial [5]. The piezoelectric nature of collagen has been purported to aid with several biological functions including control of bone growth [6], wound healing [7] and cell stimulation [8]. The piezoelectricity of collagen is highly dependent on both its chemical composition and side groups [9], nanostructural packing [10] as well as microstructural orientation and alignment [5]. Thus, the optimisation of collagen fabrication processes for enhanced piezoelectricity at the nano-and micro-structure can improve the performance of biomedical constructs, as well as green piezoelectric devices based on collagen-like peptides [9,11].
Similarly, the surface potential of a material has a profound impact on the initial cellular and protein interactions with a surface, which is dependent on the substrates and cells of interest [12]. The physiological responses affected by surface charge include protein adsorption [13,14], the degree of cellular adhesion [15,16], proliferation [17] and differentiation [18] as well as the nature of such cellular binding events [14]. Typically surface potential is controlled through functionalisation, which modifies the type and density of charged groups at the material's surface. However, modification of functional groups may also inadvertently affect the sites involved in biochemical processes such as integrin-mediated cellular attachment [19]. Recently, mechanical strain on its own has been found to greatly influence the surface potential [20]. This was hypothesised to occur due to the structural rearrangement of ionisable groups at the fibrillar level [20]. As a result, investigating novel mechanical methods of influencing the nanostructural arrangement of collagen may enable greater control over the surface potential-dependent cellular response to collagen-based constructs.
Additive manufacturing techniques can offer high spatial resolution and precise control of the microstructure, including the alignment of crystalline domains [21] or filler particles [22] in extruded polymers. While 3D printing techniques have been employed with collagen, their success and use are limited by the optimisation of collagen inks to ensure compatibility with the technique. Polymeric fibrillar collagen has poor solubility in most aqueous suspensions suitable for biological use, requiring solubilisation in strong acids and subsequent neutralisation to recover the fibrillar structure [23]. The majority of additive manufacturing techniques instead rely on soluble collagen, a monomeric or oligomeric form of collagen molecules [24,25]. Such solutions of collagen are typically less cost-effective than their microfibrillar alternatives due to the extraction processes required to make them acid or base-soluble. Soluble collagen-based hydrogel structures also require additional curing, either in-situ or after printing, using temperature, pH, UV irradiation or chemical crosslinkers in order to facilitate gelation [26]. This is in contrast to the use of microfibrillar collagen which is comprised of a dense network of collagen fibrils and fibres. Microfibrillar collagen suspensions can be used to create stable three-dimensional structures such as ice-templated collagen scaffolds [27,28] and drop-cast or electrophoretically deposited collagen membranes [29][30][31].
Here, we consider the use of aerosol jet printing (AJP) as a viable fabrication method for aligned and highly piezoelectric collagen from microfibrillar sources. AJP is a relatively recent development in additive manufacturing which allows for the rapid production of high resolution features. AJP is most commonly used for maskless fabrication of printed electronics with nanoparticulate metallic inks [32][33][34][35], but the technique has also seen more recent use with biomaterial-based inks such as DNA [36], streptavidin [37], as well as solutions of collagen I, II [38] and III [39]. AJP offers the ability to print lines with precise control of both the width and height through the choice of tip size and flow rates. AJP is facilitated through the excitation of the ink and generation of aerosolised droplets (atomisation).
Atomisation can be achieved using various methods, including pressure waves for ultrasonic atomisation (u-AJP) and forced gas mixing for pneumatic atomisation (p-AJP). The aerosolised droplets from the ink are contained within a nitrogen carrier gas allowing transport to the deposition head. At the deposition head, the aerosol is then aerodynamically focused using another stream of inert carrier gas (sheath gas) which surrounds the atomiser gas flow. The sheath gas and atomised droplets flow coaxially to the nozzle, where they are sprayed out onto the substrate.
Prior work with biomaterials and collagen have predominantly focused on ultrasonic atomisation of the inks, which leaves the samples susceptible to structural denaturation [38,39]. In this paper, we successfully print microfibrillar insoluble collagen suspensions as a cost-effective and facile ink source for AJP. We show that both pneumatic and ultrasonic atomisation can be used, covering a wide range of high print qualities at the microstructure and nanostructure. This paper also represents the first investigation into the topographical impact of using AJP and its influence on optimising collagen piezoelectricity and surface potential through the choice of collagen source and printing parameters.

Collagen preparation
Four different collagen types, as summarised in table 1, were considered in this study, with the central study focussed on Type I insoluble microfibrillar bovine dermal collagen. Suspensions of Type I microfibrillar bovine dermal collagen (Devro, Collagen Solutions) were prepared by hydrating 1 g of collagen in 200 ml of Table 1. Summary of collagen inks printed in this work, classified by their source (Bovine, Rat, Chicken), anatomical location (Dermal, Tail, Cartilage), collagen type (I or II) and mixture type (solution or suspension). The inks are given a code based on the animal source, collagen type and solubility/mixture type.

Ink Code
Animal source Anatomical location Collagen type Solubility/mixture type Concentration (w/v) % 0.05 M acetic acid. Suspensions were blended using a commercial blender (Waring model 8011EG) at 22 000 rpm in two two-minute bursts, with one minute of rest between each burst to prevent denaturation through the heat generated from the motor. The collagen solutions considered in this study include Soluble Type I Rat Tail (0.3 w/v% Gibco), Soluble Type II Chicken Sternal Cartilage (0.5% Sigma Aldrich) and Soluble Type I Bovine Dermal (0.5% Sigma Aldrich). Solubilised collagens were either preformulated in acetic acid as prepared by supplier (rat-tail), or dissolved in 0.05 M acetic acid from a powder formulation (bovine dermal and chicken sternal cartilage) to a standard concentration of 0.5% to be consistent with typical concentrations for drop-cast films.
The base prints and optimisation of AJP parameters was performed on the suspension of microfibrillar collagen which is available in bulk and a cost-effective source of collagen.

Rheometry of collagen inks
Frequency and amplitude sweeps of the four collagen inks (Bov-Insol-I, Bov-Sol-I, Rat-Sol-I, Chi-Sol-II) were measured using an Anton Paar MCR 302 Rheometer at 20 • C, consistent with the temperature of AJP. The measurements were taken with a cone plate geometry with a 0.25 mm plate gap. Each of the inks was added within the base and measurement plates gap until the meniscus was visible at edge of cone plate. Since structural breakdown occurs at stresses above the linear viscoelastic region (LVER), an amplitude sweep was performed to identify the LVER at 25 points across a shear strain range of 0.01%-100% and at an angular frequency of 10 rad s −1 . The shear strain within the LVER was chosen for frequency sweep for each ink, measuring 50 points across an angular frequency range of 1000-0.001 rad s −1 . Data are plotted to represent the storage (G ′ ) and loss moduli (G ′′ ) in Pa along the primary axis and the magnitude of the complex viscosity (|η * |) in Pa s. Data were rounded to 2 s.f. and presented as the mean of three replicates ± standard deviation.

AJP
AJP was carried out using the AJ200 system from Optomec (Optomec Inc. New Mexico, USA) using both the ultrasonic and pneumatic atomiser systems. The platen was kept at room temperature to avoid denaturation of collagen inks. The ink was stirred with a magnetic bead in the pneumatic atomiser, and chilled at 20 • C in the ultrasonic atomiser. The ultrasonic excitation current was set to 0.600 A. The pneumatic exhaust flow rate was set to 1150 SCCM and the pneumatic atomiser flow rate was set to 1300 SCCM. All samples were allowed to dry at ambient conditions prior to subsequent measurements.

Atomic force microscopy
Atomic force microscopy (AFM) images were obtained using Bruker Multimode 8 for topography, surface potential and piezoelectric response. Micrographs were obtained under tapping mode for topography (Tap-Al G 300, Budget Sensors) with scan sizes 10 µm × 10 µm, 2 µm × 2 µm and 500 nm × 500 nm. Kelvin probe force microscopy (KPFM) was performed in interleave mode for surface potential measurements. Piezoforce response microscopy (PFM) was performed in contact mode using the surface potential from KPFM as a DC offset (MESP-RC-V2, Bruker; nominal frequency 150 kHz, top radius 35 nm and spring constant 5 N m −1 ) at a scan size of 2 µm, with a AC lock-in amplitude of 4000 mV. By applying the bias from KPFM, any electrostatic contributions in the PFM measurements are appreciably reduced [40,41]. Both KPFM and PFM were performed on samples printed on a conductive flexible substrates (indium tin oxide (ITO))-coated polyethylene terephthalate (PET)). Image processing was performed using Gwyddion. Gwyddion image processing involved aligning rows using the median of differences, followed by a polynomial background removal of order 5. Streaks were then removed, and the height profile was shifted to zero. Gwyddion was then used to measure surface roughness using the statistical quantities tool. Representative images are included for all conditions, and quantitative data are presented as the box plots of triplicates.

Scanning electron microscopy
Scanning electron microscopy (SEM) images were acquired using a Hitachi TM400Plus microscope and the AZtecOne software. For every line of collagen printed, three SEM images were acquired and analysed in 3 ImageJ to obtain the line width. The dried samples were affixed onto the stage using carbon tape. Since the sample was printed on a conductive surface (ITO coated PET, aluminium, etc), sputter coating was omitted to prevent obscuration of features whilst eliminating sample charging. The SEM micrographs were cropped to remove the excess white space, then binarised to delineate the edges of the printed lines. A profile was plotted perpendicular to the lines, from which the width was measured. The line widths are reported as the mean ± standard deviation of three measurements across three lines.

Fast Fourier transform alignment analysis
The alignment of fibres in the AFM micrographs was determined in ImageJ by using a fast Fourier transform (FFT) technique [42]. A 2D FFT was taken of each image, and then the Oval Profile plug-in was used to calculate the radial intensity around a circular section in 1 • increments. The alignment ratio (AR) metric was calculated as the ratio of the radial intensity at θ max , the angle at which the maximum intensity is obtained as compared with the intensities 90 • away from this orientation: i.e. at θ max + 90 • and θ max − 90 • . Thus for a given FFT radial intensity, two ARs (AR − and AR + ) are calculated as The results were then reported as means and box plots of both ARs (AR − and AR + ) obtained across triplicates.

Fourier transform infrared spectroscopy
Fourier transform infrared (FTIR) spectroscopy was used to probe the conformation of drop-cast collagen films, before and after sonication, as well as of the u-AJP printed collagen films (Atomiser: 28 SCCM, Sheath: 75 SCCM. For each condition, FTIR spectra of four independent areas (two regions of two replicates) were obtained using a Bruker Tensor 27 spectrometer in ATR mode with a resolution of 4 cm −1 .

Statistical analysis
• Significance Testing The following workflow was used to calculate statistical significance using a custom python script, using the scipy and statsmodels libraries. Datasets were firstly evaluated for normality using the Anderson-Darling and for homoscedasticity using the Levene's tests. If the conditions were met, a one-way ANOVA test was performed followed with a Tukey post-hoc test. Where the conditions are note met, a non-parametric Kruskal Wallis test is used with a Mann-Whitney U post-hoc. Statistical significance was concluded where the p < 0.05. • Correlation Analysis Correlation between datasets was assessed using the Pearson correlation coefficient (R) metric which assesses the linear correlation between two data sets, using the scipy stats library in python (linregress function). The correlation coefficient was used to identify positive (R > 0.3), negative (R < 0. 3) or no correlation (|R| < 0.3). A very strong correlation is concluded where 0.9 < |R|, high correlation where 0.9 > |R| > 0.7, a moderate correlation where 0.5 < |R| < 0.7 and weak correlation where 0.3 < |R| < 0.5.

Results and discussion
3.1. Ink rheometry AJP has been reported to successfully print inks of viscosities between 1-1000 cP (10 −3 -1 Pa s), unlike other techniques such as ink jet printing which has a smaller apparent viscosity working range (10-40 cP) [43]. The majority of inks printed using AJP are homogenous, with the viscosity providing a good indication of the ability to produce aerosol mists using ultrasound. However, the rheology of inhomogeneous microfibrillar mixtures such as those considered in this work are impacted by the shear strains and frequencies involved in the process. Figure 1 illustrates the rheology of the four inks used in this study with an amplitude sweep (flow resistance at varying oscillatory shear strains) to identify the linear viscoelastic range of the inks, followed by a frequency sweep (flow resistance at varying angular frequencies). From the amplitude sweep in figure 1(a), the loss modulus (G ′′ ) is shown to be consistently higher than the storage modulus (G ′ ) for all of the solution-based inks, indicating their fluid like flow behaviour throughout the shear strain regime. On the other hand, the bovine insoluble collagen presents a region of low shear strain where the material behaves as predominantly as a solid (G ′ > G ′ ′ ) and high shear strain where the material's fluid behaviour dominates (G ′ ′ > G). Such behaviour is typical of gel-based materials and consistent with the colloidal nature of the suspension, unlike the solutions. which is four orders of magnitude below the frequency regime at which ultrasound operates. Discontinuities in the figures at low shear strains arise from artefacts averaged across repeats; please refer to the supplementary dataset for individual graphs which do not consistently possess these peaks at a given shear strain. Figure 1(b) confirms the higher complex viscosity of the microfibrillar insoluble collagen (10 7 Pa s) at low angular frequencies as compared with the solutions at 10 4 Pa s for the soluble collagen Is and 10 5 Pa s for chicken sternal collagen II. The viscosity decreases linearly for all samples until a critical frequency at 10 rad s −1 for the collagen suspension and 100 rad s −1 for the solutions, at which point all collagen solutions and suspensions possess a comparable complex viscosity for a given angular frequency. Ultrasonic atomisation employs frequencies higher than 1 MHz (i.e. 6 × 10 6 rad s −1 ) in order to produce aerosol droplets from the ink. While it was not practicable to analyse the ink rheology at these frequencies, the frequency sweep shows a consistent behaviour in the storage and loss modulus as well as the complex viscosity from an angular frequency of approximately 100 rad s −1 . The viscosity behaviours are consistent for all four formulations (10 2 Pa s at 100 rad s −1 ) across this high frequency range, irrespective of their rheological behaviour in the other regimes. Thus in the fluid regime, all collagen suspensions and solutions behave similarly, validating their use as comparable inks for this study.

Line quality
A range of parameters, including the choice of atomiser and sheath flow rates have been previously shown to have an impact on the width and quality of printed silver lines [43]. Here, microfibrillar insoluble bovine collagen suspensions were printed using both p-AJP and u-AJP, using a range of atomiser and sheath flow rates. Continuous lines were formed on ITO-coated PET substrates with a range of atomiser and sheath gas flow rates as illustrated in figure 2. Visual inspection of the micrographs reveals that highest line quality was achieved at 28 SCCM, 75 SCCM with limited overspray, homogeneous ink coverage of the line and straight line edges.
On the other hand, p-AJP of collagen was only possible over a narrow window of printing parameters. An aerosol mist was only generated with the collagen ink at a pneumatic atomiser and exhaust flow rates of 1300 SCCM and 1150 SCCM respectively. At higher p-AJP atomiser flow rates, printed lines were uneven with ink beading, as shown in figure 1(f). Thus only the sheath flow could be varied within the p-AJP, with a fixed atomiser differential.
Previous examples of beading have been observed in the literature for nanoparticle inks, such as germanium [44], at poorly chosen printing parameters. Beading was also found to occur in u-AJP of MoS 2 and gold-conjugated peptide inks at sheath flow rates of 50 SCCM and atomiser flow rates of 35 SCCM [45]. Such methods of beading were mitigated by Lai by reducing their chiller temperature, thereby increasing the ink viscosity and reducing the size of the aerosolised droplets. While cooling is not built into the p-AJP in our work, the ink temperature modifications employed with u-AJP may be adopted to improve the window of operability for p-AJP of collagen and other protein-based inks.
Prints on substrates such as aluminium resulted in discontinuous lines with a fine dispersion of ink, unlike those printed on substrates such as polyimide or ITO-coated PET. The ability to form high quality continuous prints is dependent on factors including the strength of adhesion to the substrate [43]. The adhesion of the ink to the substrate can also be impacted by the choice of material, but also the substrate surface quality (including porosity) [46].  Figure 3 illustrates the relationship between the line widths and gas flow rates with p-AJP and u-AJP. Line widths decrease linearly with sheath gas flow rates with a very strong negative correlation for p-AJP (R = −1.00), with a variability of approximately 100 µm achieved across the parameter space from a 80 SCCM variation in sheath rates. Based on the windows of operability noted in section 3.2.1, the p-AJP exhaust flow rate was set to 1150 SCCM and atomiser flow rate to 1300 SCCM while varying sheath flows. For u-AJP, the atomiser flow rates was set to 28 SCCM when varying sheath flows, and the sheath flows were set to 75 SCCM when varying atomiser flows.

Line width
u-AJP has the largest variation achievable for line widths, with values ranging over approximately 200 µm (R = 0.99). Unlike p-AJP, sheath flow rates have a limited effect on u-AJP lines, within error (R = −0.50). The range of widths achievable through the variation of sheath gas flow rates in the u-AJP is also appreciably lower than the tuneability achieved with p-AJP sheath or u-AJP atomiser flow rates.
The focussing effect of sheath flow in the AJP has been well investigated by Mahajan et al and Smith et al through the concept of a focus ratio (FR), given by the sheath flow rate/atomiser flow rate [43,47]. In this work, the operability window for FR extends from 1.4 (Atomiser: 28 SCCM, Sheath: 40 SCCM) to 4.3 (Atomiser: 28 SCCM, Sheath 120 SCCM). The range of FRs is consistent with the ratios used for other peptide inks (1.4 for MoS 2 -gold conjugated peptide inks) [45] as well as the typical ratios utilised with silver inks (FR = 2-4) [43].

Surface roughness and topographical alignment
All printed collagen lines were imaged at three length scales (0.5, 2 and 10 µm) using atomic force microscopy to assess the nanostructural topography of collagen. Representative micrographs of the lines at the 2 µm scale are shown in figure 4.
While there is limited impact on the surface roughness with sheath flow rates, collagen lines produced using p-AJP are statistically significantly rougher than those fabricated with u-AJP with varying atomiser pressures at the 2 µm scale (p = 0.006), as well as both u-AJP datasets with varying atomiser pressures (p = 0.0009) and sheath (p = 0.0002) pressures at 10 µm, but not at 0.5 µm. Similarly drop-cast films are only a statistically significantly rougher than the u-AJP datasets with varying atomiser flows at 2 µm (p = 0.03), but both u-AJP datasets at 10 µm (p = 0.005 & p = 0.02), but not with the p-AJP datasets. This is likely due to the presence of wider fibres and bubble formation on the surface with p-AJP, unlike the final worm-like strands (white arrows) observed seen with u-AJP. At the 2 µm scale, there is an increase in the visible collagen fibres (black arrows), which is consistent with the positive correlation between surface roughness and flow rates and both atomiser and sheath flow rates determined in figure 5.
In prior work by Nair et al worm-like structures (∼2 nm in width) were observed in drop-cast films for soluble monomeric and oligomeric collagen, whereas large fibre bundles (∼60-300 nm wide) were seen in micrographs for insoluble collagen. Similar worm-like strands were observed in u-AJP collagen, suggesting some degree of denaturation or fibrillar unravelling of collagen fibres when sonicated [30]. This effect however, is not observed in the p-AJP collagen lines, where larger fibres are visible in the microstructure. In both cases, fabrication of thin collagen films with AJP produces a smooth surface, with the overall roughness  and topography values an order of magnitude smaller than the values achieved via drop-casting in the literature [30].
The topographical alignment was obtained from the AFM micrographs using the FFT analysis method described in section 2.2.1. The value for the AR varies with the degree of alignment of topographical features such as fibres, with values approximately equal to 0 for randomly oriented fibres, and tending to 1 for highly aligned features, irrespective of their orientation. The mean AR of drop-cast films is 0.29 ± 0.12. However, films produced using ultrasonic atomisation presented a statistically significantly higher AR than those produced using pneumatic atomisation (p = 0.0012 compared with u-AJP with varying sheath flows and p = 0.001 compared with u-AJP with varying atomiser flows) and against drop cast films (p = 0.005 against u-AJP with varying sheath flows and p = 0.0.02 against u-AJP with varying atomiser flows). The droplet generation process of u-AJP has been reported to produce finer, more uniform and monodisperse aerosols than the p-AJP counter part [48][49][50]. While the collimation and focusing effect of 8 Figure 6. Alignment of printed insoluble microfibrillar collagen lines at the 2 µm scale with print parameters (a) pneumatic sheath flows at 1150 SCCM exhaust and 1300 SCCM atomiser flow rates (b) ultrasonic atomiser flows at a fixed sheath flow rate of 75 SCCM (c) ultrasonic sheath flows at a fixed atomiser flow rate of 28 SCCM. As a control, the mean AR of drop-cast films is 0.29 ± 0.12. The alignment ratio ranges between 0 and 1 where 1 represents strong alignment, and 0 represents an isotropic system. sheath flows should be identical across the two methods, the initial polydispersity of droplet generation in p-AJP is likely to be responsible for the lower alignment, and greater roughness of these p-AJP prints [49].
The effect of AR on the electromechanical response of the inks will be discussed further in subsection 3.4.

Printer drift
As ink builds up in the connecting tubes over time, the width is typically expected to increase with printing time [43]. However, the impact of this printer drift on other parameters such as nano structure and alignment are less well characterised. For the Bov-Insol-I inks printed using u-AJP at 28 SCCM atomiser and 75 SCCM sheath gas flow rates, the line width increases with printing time from the initial print widths at 0 up to 80 min of continuous printing. At this point, the line widths tend to a constant value (approximately 100 µm), which is also noted to be defined by the initial overspray (green dotted lines) within the system as seen in figure 7. The micrographs reveal an increasing degree of larger fibres (black arrows) which are deposited with printing time, in addition to the worm-like features (white arrows) described in figure 4, although there is little impact on topographical alignment with printing time (R = 0.12).

Protein conformation
A major concern in the use of u-AJP is the denaturation of proteins upon printing, due to the energy imparted by the sonication process. FTIR spectra were obtained from drop-cast films before and after sonication, as well as from a printed film in order to identify possible sources of denaturation, if any. Denaturation of collagen into gelatin can be seen through spectral shifts in the Amide I and III bands [51,52], corresponding to the relative proportion of triple helical collagen and denatured α-helical gelatin. The FTIR spectra in figure 8 indicate no change in the Amide I or Amide III bands of drop-cast collagen films produced from the ink before and after sonication. This suggests that the sonicated ink does not undergo any notable structural denaturation. However, the u-AJP printed collagen films, reveal a shift in the amide I band from 1631 to 1658 cm −1 , which is consistent with a decrease in triple helical collagen and increase in α-helical components [52]. Similarly, the redshift in the peak around 1200 cm −1 is indicative of gelatinisation and loss of the native triple helical structure in collagen. Since this shift is not observed with the drop-cast collagen films following sonication, and only with the printed collagen films, we instead suggest that the primary mechanism involved is the fractionation of the collagen suspension with ultrasonic delivery of the aerosol, as depicted in figure 9. Insoluble collagen suspensions will contain a range of collagenous structures after homogenisation such as large collagen fibres, smaller soluble collagen fibrils, monomers and oligomers, as well as some fully or partially denatured gelatin. Since the aerosol must be carried against gravity into the deposition head, a greater than average proportion of gelatin, and soluble collagen monomers and oligomers are deposited as part of the film, as compared with the heavier collagen fibres. Thus the sonication process is not directly responsible for collagen denaturation, but instead preferentially aerosolises and deposits the lighter collagen fragments within the suspension. This proposed fractionation is also consistent with the increase in large fibres observed with increasing printing time in figure 7, where larger fibres are more likely to form a part of the aerosol at later printing times with varying ink concentration in the u-AJP vial.

Electromechanical properties
The surface potential and the piezoelectric response of the drop cast collagen films before printing, and printed collagen films were characterised using Kelvin probe force microscopy (KPFM) and piezoresponse force microscopy (PFM) respectively. The surface potential, the lateral amplitude as well as the lateral and vertical in-phase piezoelectric components are plotted in figure 10 as a function of the printing parameters. The amplitude of the signal generated using PFM is directly related to the physical deflection of the tip due to the voltage-induced strain in the material. As a result, a positive piezoresponse amplitude reflects a positive displacement in response to the applied field whereas a negative piezoresponse amplitude reflects a negative displacement, when resolved in the horizontal and vertical domains. Within error, there is generally little impact of the atomiser or sheath flow rates on the potential and piezoelectric lateral amplitudes. The notable exceptions are the impact of sheath flow rates on vertical in-phase channel (R = −0.93 for p-AJP and R = −0.59 for u-AJP).
While no strong trends are observed within error by comparing alignment with sheath or atomiser gas flow rates, the correlation plot in figure 11 indicates that AR, obtained from the FFT alignment can be  helpful as an indicator of the vertical in-phase piezoelectric amplitude (R = −0.45), but does not appear to have a correlation with the other parameters. However, looking in specific datasets, we can see that the alignment of drop-cast films (black data points) is amongst the lowest, correlated with their low piezoelectric responses. In addition, there is a moderately negative correlation between sheath flows and lateral amplitude (R = −0.66). Considering the ultrasonic sheath alone, there also is a moderately positive correlation between AR and surface potential (R = 0.64). Similarly, the roughness at 2 µm is moderately well correlated with the potential (R = 0.618) but like the AR (R = −0.03) is poorly correlated with the lateral amplitude (R = −0.08) in general. There is little evidence in the literature to suggest that the surface roughness is physically responsible for a change in the surface potential [53], nor is there any indication of this from the measurement of surface potential during KPFM [54]. It is likely that this effect predominantly arises from the distribution and presence of larger fibre bundles of collagen present a higher surface potential when compared with smoother molecular collagen. This is consistent with the higher potential previously observed for rough and fibrous drop-cast collagen with a surface potential of ∼0.7 V in the literature, which was fabricated from the same microfibrillar collagen compositions and measured under the same PFM, KPFM settings shown here [30], while the potentials measured in this paper do not exhibit a statistically significant difference in the roughness from the drop-cast films and also do not exhibit a high potential (∼0.1 V) .
The lateral piezoelectric amplitudes obtained using AJP are nearly 2-4 times larger than that of drop-cast microfibrillar collagen (median value of 0.125 V from literature [30] and ∼0.120 V from the two drop cast sets in this work). The increase in the piezoelectric response may be attributed to the alignment induced by the printing process as compared with the random orientation of collagen fibres in drop-cast films. This increase in piezoelectric amplitude is consistent with the alignment-driven increases observed in other aligned collagen systems such as membranes and tendons [5,10]

Comparison of collagen sources
Four different collagen sources were compared, including Type I collagen from different sources (soluble rat tail as well as soluble and insoluble bovine dermal) and Type II collagen from chicken sternal cartilage. The roughness of the films were characterised at three length scales using AFM, as shown in figure 12 which show similar trends across the length scales, i.e. rat tail collagen is the roughest, followed by soluble bovine dermal, chicken sternal cartilage and finally bovine dermal insoluble. While the values of roughness highly correlate with the surface potential (R = 0.88), the piezoelectric amplitude appears to be largely dependent on the intrinsic nature of the collagen.
Typically, differences in the piezoelectric response of the collagen sources may be attributed to differences in the primary structure [9], the fibrillar packing [10] and crosslink density [55] of the collagen. Here, the Type I collagens are noted to have similar piezoelectric responses, which are significantly higher than the Type II chicken cartilage. Rat tail collagen is typically found to have a larger piezoelectric response than other forms of extracted type I collagen, which is not observed here. The similarity in the piezoelectric response between the three Type I collagens may be attributed to the lack of D-banding in all three conditions. The presence of D-banding in rat tail collagen has been purported to be responsible for its stronger piezoelectric response when compared with non-D banded extracted collagen, in spite of additional macroscopic ordering observed in highly aligned tissues such as tendons [10]. Thus the differences observed here may be attributed to the processing involved in extracting soluble collagen from rat tail collagen, disrupting the alignment and fibrillar packing of the collagen molecules as compared with a pristine rat tail.

Conclusions
This paper has focussed on the optimisation of pneumatic and ultrasonic-driven aerosol jet printing of microfibrillar collagen inks. We demonstrate that the piezoelectric response can be increased by up to 4-fold as compared with drop-cast samples due to the alignment imparted by the AJP process. We also determined that microfibrillar bovine dermal collagen can be fabricated with comparable piezoelectric amplitudes to soluble bovine and rat tail collagen, providing a cost-effective alternative for the large scale production of piezoelectric collagen constructs. p-AJP was found to create rougher surfaces than u-AJP, likely due to the deposition of large fibre bundles with p-AJP. The corresponding surface potential of the prints is highly correlated with their roughness, particularly for collagen prints of different sources. Thus, this study highlights how u-AJP and p-AJP can be used to control the piezoelectric response and surface potential of collagen, which may aid in the optimisation of cellular adhesion and biological activity in collagen-based constructs.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://doi. org/10.17863/CAM.86917.