Parallel multiphoton excited fabrication of tissue engineering scaffolds using a diffractive optical element

: Multiphoton excited photochemistry is a powerful technique for freeform nano/microfabrication. However, the construction of large and complex structures using single point scanning is slow, where this is a signiﬁcant limitation for biological investigations. We demonstrate increased throughput via parallel fabrication using a diﬀractive optical element. To implement an approach with large ﬁeld of view and near-theoretical resolution, a scan lens was designed that is optimized for using low-magniﬁcation high NA objective lenses. We demonstrate that with this approach it is possible to synthesize large scaﬀolds at speeds several times faster than by single point scanning.

relative to the non-biological applications discussed above. For example, biomimetic models of the extracellular matrix (ECM) to investigate cell dynamics have great potential value for studying a wide range of diseases, including cancers and fibroses [38][39][40]. However, scaffolds ideally should be at least a few millimeters in diameter to study multiple cells at the same time. We have previously shown that structures fabricated by MPE with different concentrations of ECM components (e.g. laminin and fibronectin) are able to influence cell behavior (e.g., migration, proliferation and cytoskeletal alignment) in different cell lines [40,41] but the fabrication was highly time consuming.
MPE fabrication is historically performed on a laser scanning microscope with single point excitation. With conventional galvo scanning it is not possible to increase the scan speed beyond ∼1 frame per second. Moreover, it is also not possible to simply use increased power due to saturation of the photo-activator, resulting in photodamage (especially for soft biological materials) and loss of resolution due to absorption outside the point spread function (PSF). As a consequence of these limitations, the fabrication of a protein structure 1 cm × 1 cm × 100 microns requires over a week using single point excitation. Thus, low fabrication rates limit both the scaffold size and numbers of copies that can be created to perform cell biology studies.
Parallel approaches are attractive considerations to increase the fabrication speed. Several groups have utilized spatial light modulators (SLMs) to shape the beam and fabricate structures such as a one dimensional (1D) line or a two dimensional (2D) area at once [42][43][44]. This scheme however requires high energy laser pulses. An alternative approach is to use an SLM to split the incident beam into multiple beams and fabricate at several foci simultaneously [45,46]. This is plausible as 3-4 W average power titanium sapphire (ti:sapphire) oscillators are commercially available. However, the inefficiency of SLMs in creating beams with large diffraction angles and the financial cost of these devices can be a limiting factor in their use. Microlens arrays for parallel fabrication have also been reported [47,48], however it can be difficult to obtain uniform illumination across all elements without significant beam expansion (and loss of laser intensity). While detrimental for fluorescence imaging, uneven excitation is more problematic for fabrication as this will result in spatial differences in crosslinking and physical properties.
As an alternative, diffractive optical elements (DOEs) offer high efficiency in splitting a beam to multiple spots. Dong et al. [49] and Kelemen et al. [45] demonstrated the application of DOE for multi-beam, two-photon excited photo polymerization several years ago. However, these reports focused on optimizing resolution and creating multiple copies of the same small objects within a limited field of view (FOV), rather than increased throughput when greater speed is beneficial. There is untapped potential in using these devices to create larger scale objects, such as tissue engineering scaffolds, as the instrument can be enabled to create multiple copies of the same structure at once. However, proper lensing mechanisms are needed to enable diffraction limited resolution across a large FOV with minimum field curvature. Indeed, the scan lens is a main limiting factor in achieving a large FOV for many custom-made laser scanning microscopes.
In this paper, we report the design and demonstration of a large field of view DOE-based MPE fabrication system appropriate for creating tissue engineered scaffolds. We designed and tested a custom-assembled scan lens to achieve near diffraction limited fabrication resolution across a large field. The empirical resolution and diffraction limited FOV are measured, and compared with the theoretical and simulation results. An integral part of the system is the control software to implement the parallel fabrication of image-based (e.g. Second Harmonic Generation; SHG) scaffolds. We previously reported a scanning approach we dubbed modulated raster scanning that is in practice a hybrid of raster and vector scanning [31]. In this approach, a fast (>10 MHz) electro-optic modulator (EOM) precisely controls the laser exposure time in each pixel to control the concentration of cross-linked proteins [31]. This scheme is now implemented with a 2×2 DOE to achieve greater fabrication speed of image-based scaffolds. Cell biocompatibility of the resulting scaffolds is further demonstrated.

Optical setup and instrument control
The system uses a femtosecond Ti:sapphire laser (Chameleon, Coherent, Santa Clara, CA) at 740 nm with an average power of 3 W and pulse width of 140 fs. An optical isolator after the laser source blocks retro-reflections from the optical train backs toward the laser. The laser power going into the laser scanner is controlled by a zero-order half-waveplate (QWPO-780-05-2, CVI, NM) followed by a Glan laser polarizer (MGLA-SW-8, Karl Lambrecht Corp., IL, USA) (Fig. 1). A high speed (100 MHz) electro-optic modulator (EOM; 350-160-02 Conoptics, Danbury, CT) is used to modulate the laser exposure at each pixel to control the density of crosslinking. With this mechanism we previously dubbed modulated raster scanning, we are able to fabricate structures with resulting crosslinked concentration that maps the intensity of each pixel in grayscale images [31]. To maximize the efficiency of two-photon excitation, a prism pair (06SF10, Newport, Irvine, CA) compensates for dispersion caused by optical components. After beam expansion using a telescope lens mechanism (50 mm and 100 mm focal length plano-convex lenses) the beam is split to four (2×2 array) beams at a separation angle of 1.93 o using a DOE (MS-724-1-Y-A, HOLO/OR, Ness Tziona, Israel). A telescope lens mechanism (bi-convex lenses with focal lengths of 100 mm and 60 mm) projects the beams on to a pair of 3 mm galvanometric mirrors (Cambridge Technologies, Bedford, MA) of the purpose-built laser scanning microscope. This microscope itself consists of a RAMM system (ASI, Eugene, OR), a custom scan lens, a Nikon tube lens (200 mm focal length (FL)), and a 10×/0.5NA or 20×/0.75NA objective lens. A long-pass dichroic beam splitter (FF593-Di03-25 × 36, Semrock, Rochester, NY) reflects wavelengths shorter than 593 nm toward a photomultiplier tube (H7421-50, Hamamatsu, Bridgewater, NJ) for fluorescence imaging. A second dichroic beam splitter (t700lpxxr-uf2, Chroma Technology Corp., Bellows Falls, VT) reflects wavelengths between 593 nm and 700 nm toward a camera (SCE-B013-U, Mightex Systems, Toronto, Canada) for visual inspection and locating the area of interest. Galvanometric mirrors, sample stage, objective height, and EOM are all controlled via a field programmable gate array (FPGA) module (PXIe-7972R FlexRIO, National Instruments, Austin, Texas) under OpenScan software. The key elements relative to a conventional laser scanning microscope are the fast EOM affording image-based fabrication, the DOE which creates four spots and the custom designed scan lens.

Scan lens design
For many years objective lenses were the bottle neck in constructing a high resolution and large FOV laser scanner device. However, with the advent of high-performance objective lenses in the past decades, improvements in the optical relay mechanisms including the scan lens are necessary to fully utilize their capacity [50]. Here we delineate the optimal scan lens properties and then describe the optimal design for our application. An ideal scan lens provides a flat focal plane, where the focal spot position on this plane linearly depends on the deflection angle of the input beam, known as f-θ condition. This can be achieved by designing the lens mechanism with a precisely engineered amount of barrel distortion. Telecentricity is another important feature that ensures the paraxial beam is perpendicular to the focal plane. A lens with this feature creates a constant magnification at different depths. Without telecentricity the focal spot elongates and changes its orientation for off-axis beams, which affects resolution and brightness (and here uneven fabrication) across the FOV. In addition, due to the oblique angle of the off-axis beams, there is a significant lateral shift and beam loss at the back aperture of the objective lens at large scan angles. In addition, the scan lens provides the optimal beam expansion in combination with the tube lens such that the laser beam properly fills back aperture of the objective.
In practice, the overall performance of a laser scanner deviates from the ideal conditions due to optical aberrations and distortions. For example, a simple lens introduces significant deviations when used as scan lens and more complex scan lens mechanisms are thus necessary to construct a large FOV high-resolution microscope [50,51]. For example, Bumstead et al. [52] used optical invariant analysis to find an optimal combination of commercially available scan, tube, and objective lenses along with a galvo system to achieve a large FOV laser scanning device. Mostly due to the long focal length of their scan lens (115 mm), the microscope requires a large input beam diameter of 10 mm or larger to operate at the nominal resolution of their objectives. This geometry dictates using galvo mirrors of sizes equal or larger than 10 mm, which have slow scan speeds compared to the more common ∼3 mm diameter versions. Negrean and Mansvelder reported custom-designed scan and tube lenses to enable near diffraction-limit resolution across the field of view of a 25× objective (XLPLN25×WMP). Their scan lens had a focal length of 50 mm in combination with a tube lens of focal length of 250 mm, allowing expansion from a beam diameter of 3 mm to fill the back aperture of the objective [50]. However, this expansion is not sufficient for a lower magnification objective (e.g. 10×/0.5NA Nikon with back aperture of 20 mm in diameter) when using a small galvo system. Moreover, in many commercial microscopes replacing the tube lens with a 250 mm lens proves to be challenging. Therefore, there is a need for a high-performance scan lens with short focal length that allows the use of small galvo systems, low magnification high NA objectives along with conventional tube lenses.
To alleviate these difficulties and provide an optimized solution for our large FOV, DOE-based fabrication microscope, we designed a compound scan lens consisting of multiple commercially available lenses (2×AC254-100-B, LBF254-200-B, and LBF254-100-B) ( Table 1). This scan lens is designed to have small spherical, astigmatism and curvature aberrations while maximizing the FOV when paired with an achromatic (AC508-200-B, Thorlabs, Newton, NJ) tube lens (Fig. 2). The lens mechanism is specifically designed to have a short focal length of 32.1 mm which is suitable when using a low magnification, high NA objective. This combination of these attributes is necessary for our fabrication application demanding both large FOV and high resolution.
The short focal length of this scan lens provides a beam expansion greater than 6x in combination with the 200 mm FL tube lens. This expansion ratio is sufficient to expand a 2.4 mm beam to fill the back aperture of a 20× 0.75 NA objective or equivalently to expand a 3.2 mm beam to fill the back aperture of a 10× 0.5 NA lens when paired with a tube lens with focal length matching the objective. The design is comprised of two achromatic doublet lenses, each 100 mm FL, and two best form lenses with focal lengths of 100 mm and 200 mm ( Fig. 2(a); Table 1)). Since most microscope manufacturers use a tube lens focal length of ∼200 mm, we  We found that the design strikes a good balance between beam expansion and spherical and astigmatism aberrations when paired with the selected tube lens. The overall spherical aberration in this lens configuration is negligible (Fig. 2(c)), and the root mean square wavefront error remains under the diffraction limit (0.072) across a FOV larger than 600 µm (Fig. 2(d)). The field curvature at the edge of this FOV is about 0.03 µm and 0.83 µm toward the objective lens for sagittal and tangential components, respectively ( Fig. 2(b)). The separation between the sagittal and tangential components is about 0.80 µm which is smaller than the axial extent of the excitation volume, and its effect on the PSF of the optical system can be neglected. Therefore, the astigmatism will not degrade the quality of fabricated structures within a FOV of diameter 600 µm. This was further investigated by using the physical optics propagation (POP) tool in the OpticStudio software package. By fitting a Gaussian function to the intensity profile obtained from POP at wavelength of 840 nm the full width at half maximum (FWHM) at the center of FOV was estimated to be 0.45µm. At the maximum scan angle of 11 degrees the FWHM was increased to 0.57µm (Fig. 2(e)). Our analysis shows that the f-θ distortion in this setup is less than 0.5% ( Fig. 2(b)), while the paraxial beam angle deviates less than 4.2 degrees across the FOV (data not shown), demonstrating a satisfactory level of telecentricity. The maximum intensity loss due to vignetting was measured to be less than 4% which is negligible for our purposes.
This lens mechanism is not compensated for chromatic aberrations. This is not a limitation for this system as the excitation wavelength is nominally constant and fluorescence imaging is only used as an online diagnostic of the fabrication. Still, if a chromatically corrected scan and tube lens combination is desired it is possible to use the lens mechanisms designed by Negrean and Mansvelder [50]. However, these lenses need to be custom manufactured, while our design uses off the shelf lenses. Moreover, the short focal length of our lens mechanism permits a greater beam expansion and use of small galvo mirrors (3 mm beam diameter) to achieve fast scan rates.

Sample preparation
The scaffolds are created from a mixture of 75% gelatin methacrylate (GelMA) [53] and 25% rat tail collagen I (v/v). 100% collagen is not compatible with available photochemistries and GelMA is widely used in collagen scaffolds [54,55]. The slides were prepared first with a rubber hybridization chamber secured to a silanized microscope slide [56]. The collagen solution used was a 1:4 ratio (25 µL collagen and 75 µL GelMA). The GelMA and collagen was kept strictly below 40°C. Scaffolds were kept in phosphate buffer solution (PBS) until cell seeding.

Sodium 4-[2-(4-morpholino)benzoyl-2 dimethylamino]butylbenzenesulfonate (MBS) was chosen
as the photo-initiator because it is non-cytotoxic and water-soluble. Two-photon excitation of MBS at 740 nm drives a photochemical reaction that creates benzoyl and alpha amino alkyl radicals [57]. At the focal volume, reactive radicals interact with the surrounding proteins targeting oxidizable amino acid residues. The active radicals react with monomers, inducing a chain growth crosslinking and eventually termination [58].

System resolution
We obtained the lateral and axial PSFs of the system by removing the DOE and obtaining 2-photon excited (840 nm excitation wavelength) images of sub-resolution (200 nm) fluorescent microbeads (Polysciences Inc., Warrington, PA). Figure 3 represents the 3D PSF of this system acquired using a 20×/0.75NA objective (CFI60 Plan Apochromat Lambda, Nikon). Each PSF is obtained by averaging 3D images of a minimum of 15 beads. Following deconvolution with the 200 nm bead, the lateral and axial resolutions (FWHM) at the center of the FOV were approximately 0.51 ± 0.052 µm (mean ± s.d.) and 2.4 ± 0.143 µm (mean ± s.d.), respectively. The theoretical two-photon PSFs in the lateral and axial planes are 0.48 µm and 1.9 µm, respectively [59]. While similar to predicted values, the small deviations from theory can be due to multiple factors, including imperfections in the input beam, intermediate lens systems, or objective lens. To further explore the resolution, we measured the PSFs at different locations across the 600 µm FOV, where Fig. 3(b) shows the coronal, sagittal, and axial views of these measurements along the x and y directions. Figure 3(c) summarizes the resulting radial (before deconvolution) and axial dimensions of the PSF. We observed a slight increase in the PSF size across the FOV in the direction of fast galvo mirror (x), whereas there was negligible change along slow galvo axis (y). This can be understood in terms of the geometry of the scanning mirrors. The mirror assembly is placed such that the fast mirror is 3 mm further away from the back-focal-plane of the scan lens while the slow mirror is 3 mm closer. This can introduce more significant astigmatic aberration at large angles of the fast mirror, and consequently a slight enlargement of the PSF. Moreover, the excitation beam is linearly polarized in the x direction which resulted in slightly larger PSF when compared with the perpendicular axis (0.56 µm vs 0.51 µm before deconvolution), which has been previously reported [60] and [61]. We should also emphasize that we are using a Nikon tube lens in this system. However, the simulations are performed assuming an achromatic doublet tube lens from Thorlabs because we do not have access to the Nikon design data. Overall, the lateral and axial PSFs are in good agreement with predicted values and only display small differences across the FOV of the objective.

Parallel fabrication of scaffolds with the DOE
For multi-beam MPE fabrication using the modulated raster scanning approach, we incorporated the DOE in the beam path (Fig. 1). The DOE splits the beam to a 2×2 array with a separation angle of 1.93 o . We used a Nikon CFI Super Fluor 10×/0.5NA objective with a working distance of 1.2 mm for fabrication purposes. The separation between the beams is measured to be ∼180 µm at the focal plane of the objective, which allows a fabrication area of 360 µm × 360 µm in each scan, consisting of four (2×2) identical copies of same structure. The pixel dwell time for both single and multi-beam fabrication was set to 33 µs. In the scanning approach, the fast galvo mirror is retraced to the beginning of the line after each raster scan to avoid possible damages due to high accelerations at the end of each line. The slow galvo was also gradually retraced back to the start position after each frame. Therefore, the fabrication time for each frame takes 9.5 sec. Then there is a wait time of 4.3 sec between the frames for stage movement and uploading the data for the next frame on to the FPGA module, resulting in an overall fabrication speed of 4.3 frames per minute. The z-step size was set to 1 µm. The input power for single and multi-beam fabrication was set to 200 mW and 1000 mW, respectively. We first compared the quality of fabrication achieved by the single point and DOE scanning approaches. Here we define fidelity as the co-localization between the computer-generated design and the fabricated construct, where the correlation coefficient in pixel overlap and grayscale value is determined with an ImageJ plugin. Figure 4 shows the design and fabrication of a grid like structure that we have previously used as a part of a model for repair of myocardium infarction [62]. The top row in this figure shows the design (Fig. 4(a)), and single optical sections created by single point (Fig. 4(b)) and DOE (Fig. 4(c)) fabrication. Figures 4(d) and 4(e) are the corresponding respective 3D renderings of data shown in Figs. 4(b) and 4(c). We found correlation values between the designs and fabricated constructs of 0.74 and 0.90 between the two approaches. These values are comparable to our previous work [31], showing good fidelity in all cases. We have previously created these type structures of 1 cm × 1 cm × 100 microns of a grid pattern on a solid base by conventional single point scanning which required about 1 week of fabrication time (100% duty cycle). In contrast, using the DOE this was reduced to about 2 days.
Our primary interest lies in fabricating true image-based scaffolds. The modulated raster scanning approach is ideal for this purpose as complex morphology can be produced with the sub-micron feature sizes that mimic collagen fibers in the ECM of most tissues. As example, the images in Fig. 5 demonstrate the approach for fabricating SHG image-based models from the collagen in human ovarian cancer. We have previously used machine learning analysis approaches of SHG images to classify several classes of ovarian tissues and showed they could be differentiated with high accuracy based on their respective collagen morphology [63]. Figures 5(a) and 5(b) show representative SHG images of normal stroma, a benign serous tumor and a high grade serous ovarian cancer and the corresponding image-based model in each case, respectively. Due to the overlapping fibers within the focal volume, we use standard image processing tools in ImageJ (e.g. the eigenvalues of the Hessian matrix) to discretize the image. We have previously fabricated these models by single point modulated raster scanning to investigate cell migration [64,65]. Now we compare the fidelity of the single point (Fig. 5(c)) and DOE (Fig. 5(d)) fabricated scaffolds and perform the same fidelity analysis as in Fig. 4. Here we found correlation coefficients of 0.7 relative to the design for both single and parallel fabrication and 0.9 between the two approaches, showing comparably good quality fabrication from both methods.
To demonstrate the capacity of this approach for creating larger image-based scaffolds, we fabricated a structure of size 1.4 × 1.4 × 0.1 mm 3 (xyz), where the single SHG optical section and design used in the repeating pattern are shown in Figs. 6(a) and 6(b), respectively. The large-scale structure consists of blocks of size 360 µm × 360 µm × 100 µm which are tiled at a separation of 350 µm. Each of the copies represents a 512 × 512 pixel structure, resulting in a structure of 1024 × 1024 pixels. The blocks are then repeated 16 times (4×4) in the x and y directions with an overlap of 10 µm to cover an area of 1.4 mm × 1.4 mm. Each block consists of 100 layers which are stacked on top of each other with a step size of 1 µm, satisfying the Nyquist criteria for this NA. Using the 2 × 2 DOE, a total of 1600 frames were needed to make structure and required 6.2 hours of fabrication time. By conventional single point scanning this structure required about 24 hours. To investigate bio-compatibility of the resulting scaffold, HEY ovarian cancer cells were seeded on the scaffolds at density of 50K cell/mL and incubated for 48 hrs. Subsequently, the slide was fixed and cells were stained with the membrane dye DI-8-ANEPPS. Two-photon excited fluorescence images of the scaffolds and cells were acquired with 780 nm excitation wavelength. A single fluorescence optical section and 3D rendering of the entire scaffold and cells are shown in Figs. 6(c) and 6(d), respectively. Expanded views in Figs. 6(e) and 6(f) show HEY cells at depths of 40 µm and 70 µm, respectively, indicating migration on and into the scaffold, demonstrating the bio-compatibility of the resulting structure. Scattering from the scaffold prevented high resolution imaging at depths beyond 70 µm.

Discussion and conclusions
In this work we presented the application of a 2×2 DOE for parallel MPE fabrication of scaffolds that mimic the ECM of native tissues, where these can be used for cell biological investigations. Previous reports using either DOE or SLMs were limited to small fields of view and were not appropriate for creating scaffolds of sufficient size to study cell migration. Specifically, we require low magnification (10-20×) with sufficient NA (>0.5) to reproduce the fibrillar structure of native ECMs while limiting aberrations across the large FOV. This requires significant beam expansion while still using small galvos (∼3 mm) that have sufficient speed. To meet all these requirements, we designed and used a custom-made scan lens with a focal length of 32.1 mm. Zemax simulations predicted that this design when paired with an achromatic doublet tube lens and an ideal 20× objective lens would yield near theoretical resolution across a FOV of 600 µm in diameter. Our empirical measurements of the 3D PSF are in good agreement with the simulations.
Notably, the parallel DOE approach yielded comparable fabrication fidelity to that of our previous efforts using single point scanning [31]. The approach indeed increased the speed by approximately the number of focal spots, demonstrating the increased throughput of the process which is advantageous for creating tissue engineered scaffolds. Moreover, ovarian cancer cells retained good viability on and within the scaffolds.
While we used a 2×2 array of beams that are 180 µm apart, the same process can be used for patterns of different sizes. We note that this approach yielded much larger FOVs than those previously reported spacing (20-40 µm). We anticipate that with this scan lens separations larger than 500 µm can be used to create larger structures with a 10× objective. This separation can be achieved by replacing the current DOE or modifying the telescope lens following the DOE. Given sufficiently large power a DOE with a larger array size can also be used to further increase the fabrication speed. Additional speed improvements should be achievable through software optimization. We suggest that designing and constructing an optimized scan lens consisting of off-the-shelf lens components can greatly promote utilization of DOEs for microfabrication applications.

Disclosures
The authors declare no conflicts of interest.