*Article* **SMILE Platform: An Innovative Microfluidic Approach for On-Chip Sample Manipulation and Analysis in Oral Cancer Diagnosis**

**Sofia Zoupanou 1,2 , Annalisa Volpe 3,4 , Elisabetta Primiceri 2 , Caterina Gaudiuso 3,4 , Antonio Ancona 3,4 , Francesco Ferrara 2,5, \* ,† and Maria Serena Chiriacò 2, \* ,†**


**Abstract:** Oral cancer belongs to the group of head and neck cancers, and, despite its large diffusion, it suffers from low consideration in terms of prevention and early diagnosis. The main objective of the SMILE platform is the development of a low-cost device for oral cancer early screening with features of high sensitivity, specificity, and ease of use, with the aim of reaching a large audience of possible users and realizing real prevention of the disease. To achieve this goal, we realized two microfluidic devices exploiting low-cost materials and processes. They can be used in combination or alone to obtain on-chip sample preparation and/or detection of circulating tumor cells, selected as biomarkers of oral cancer. The realized devices are completely transparent with plug-and-play features, obtained thanks to a highly customized architecture which enables users to easily use them, with potential for a common use among physicians or dentists with minimal training.

**Keywords:** oral cancer; circulating tumor cells; micromixers; 3D microfluidics; biodetection; plastic microfluidics; microfabrication

#### **1. Introduction**

Head and neck cancers represent the sixth most common type of cancer in Europe, accounting for 150,000 new patients per year, and 60% of patients with advanced disease at diagnosis die within 5 years [1,2]. The most widespread cancer in the head and neck region is oral squamous cell carcinoma (OSCC), occurring at the border of lips and/or at the posterior of the tongue or palates [3]. Many factors can increase the probability of the disease; tobacco and alcohol consumption [4], oncogenic viruses (e.g., papillomavirus, HPV, or Epstein–Barr virus EBV) [5] and poor oral health [6] are known risk factors. Genetic predisposition is also a key consideration when studying the development of oral cancer. As an example, the role of *NFKB1* gene polymorphisms are currently under investigation [7]. The link of germline genetics and environmental factors to pathologic phenotypes can contribute to a better understanding of the interactive role of the environment, tumor cells, immune cells and microbiome in various diseases [8,9]. Moreover, the importance of correct diet and lifestyle is another crucial aspect in preventing oral cancer, as they also modulate the oral microbiome, which has been demonstrated to play a role in cancer onset, particularly due to its influence in the modulation of immune system [10]. High

**Citation:** Zoupanou, S.; Volpe, A.; Primiceri, E.; Gaudiuso, C.; Ancona, A.; Ferrara, F.; Chiriacò, M.S. SMILE Platform: An Innovative Microfluidic Approach for On-Chip Sample Manipulation and Analysis in Oral Cancer Diagnosis. *Micromachines* **2021**, *12*, 885. https://doi.org/ 10.3390/mi12080885

Academic Editor: Nam-Trung Nguyen

Received: 16 July 2021 Accepted: 26 July 2021 Published: 27 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

levels of colonization of OSCC by facultative oral streptococci have been shown [11], and, more recently, Zhang and coworkers compared the microbiota compositions of tumor sites and normal tissues, finding bacteria significantly associated with mouth tumors [12]. These aspects fall in the gene-by-environment (G × E) interaction range, which considers a wider analytical method to study the onset of diseases, including the integration of microbiology into molecular pathology and epidemiology models, going in the direction of more personalized medicine [9].

Currently, biomarkers of oral cancer are indeed inadequate, and inflammatory molecules may have a low specificity. Moreover, the detection of precancerous lesions is not routinely carried out in clinical settings [13]. An easy-to-use, noninvasive assay is strongly needed, and the possibility to perform tests on saliva is an attractive strategy to increase patient compliance [14].

In order to improve the quality of diagnostic and prognostic early screening tests, as well as their availability for a large cohort of potential patients, a number of biomarkers from body fluids have been identified. Among these, inflammatory biomarkers [15–17] and circulating tumor cells are the most promising entities to be found in blood or saliva, even at the early stage of the disease, and this method of detection has the potential to be translated into on-chip platforms. Early detection of OSCC is indeed the only way to limit the consequences of disease, and the main challenge in prevention is large-scale screening [18], together with an appropriate diet and correct lifestyle [4]. To reach this goal, the need for a noninvasive assay is compulsory, and the possibility to perform tests on saliva is an attractive strategy to make this test suitable for all, in addition to the ambitious objective of the SMILE platform.

The main purpose of SMILE project is the development of a low-cost sensing device for oral squamous cell carcinoma (OSCC) early screening, with features of high sensitivity, portability, and ease of use, with the aim of reaching a large audience. OSCC is usually diagnosed at an advanced stage, where highly invasive surgery and chemotherapy are required, heavily compromising life quality and survival [3,19].

Breakthrough technologies of the platform include the optimization and integration of a highly customizable plastic microfluidics device able to perform some of the most common operating tools while handling samples, particularly micromixing, gradient generation [20], and the capture and detection of small objects such as circulating tumor cells [21].

Moreover, the recent spread of COVID-19 and its huge impact on clinical settings has forced healthcare systems to undergo a total rearrangement of rules and priorities. In cancer management, this has led to weighing up the risks of tumor progression due to a delay in treatments against the potential of adding to the hospital burden by increasing the risk of exposure to SARS-CoV2. The need of new tools able to maintain standards of diagnosis and control of active tumor cases, while limiting infections, through rapid sample collection is then compulsory. New instruments, based on a lab-on-chip (LoC) approach, with features of low cost and a plug-and-play setup, along with the aim of using body fluids easily collected without the need for healthcare personnel, could revolutionize the approach to periodic screening and follow-ups, allowing the possibility to perform them in a safe and distanced manner.

To meet this goal, microfluidic technologies and the use of plastic substrates in combination with rapid prototyping methods, i.e., fs laser technology and micromilling, seem to be a good alternative to standard methods in the realization of polymeric lab-on-chip, without any constraints on the substrate material [22,23].

Today, several techniques can be exploited for the rapid prototyping of polymeric LoC [24,25]. In particular, thanks to its high resolution (<50 nm), soft lithography is one of the most exploited methods for the rapid prototyping of a polymeric microfluidic device [26,27]. This technology requires the fabrication of a mold, typically by photolithography, as well as replication and assembly of the entire device. Consequently, like other similar molding techniques [28], soft lithography is a time-consuming process, which

often limits optimization in prototyping or iterative design. Thus, despite its very low cost, it does not allow an easy and direct translation into large-scale production and industrial exploitation, with the aim of reaching the market of in vitro diagnostics (IVD). A further limitation of soft lithography is the material used. Polydimethylsiloxane (PDMS) microchannels subjected to high liquid pressure undergo deformation [29].

Technologies based on the direct microstructuring of the substrate and, thus, not requiring a mold have been proven to be more suitable during the design of a new device. Among such technologies, three-dimensional (3D) printing gives the possibility of fabricating low-cost 3D microfluidic devices in a single step from a computer model [30,31]. The major concerns about this technique regard the inability to reliably print microfluidic channels with dimensions less than several hundred microns, dimensional fidelity, surface quality, optical transparency, and reduced choice of materials [32]. Conversely, ultrafast laser technology, as a non-clean room process, provides a convenient, economical, and flexible way to fabricate micrometric fluidic patterns by varying the laser parameters [33,34], avoiding the expensive and time-consuming production of masks. The fs laser enables a "cold" ablation of the irradiated volume, which allows the material to be removed by ablation from the irradiated area with negligible thermal damage to the surrounding substrate [35], ensuring high precision and up to submicrometric resolution [36]. Moreover, fs laser pulses do not pose any restriction on the substrate materials [37]. However, the principal constraints of this technique are the high costs of the laser source and them not being efficiently suitable for the fabrication of over-micrometric structures.

Mechanical micromilling is a flexible, cost-efficient, rapid prototyping technology for polymer device machining. In comparison with fs laser, it is more convenient for large features [38]. However, it results in poor surface quality and resolution [39].

The possibility to use plastic devices will significantly improve the reproducibility and stability of experimental setups compared with PDMS-based state-of-art microfluidics. PDMS, despite its features of low cost and disadvantages due to the presence of a hard master obtained by lithographic methods, often suffers from sealing leakage, poor connection stability, consequently low reproducibility of experiments. This aspect is not secondary when dealing with biological methods, as complexity can be achieved only by avoiding variability of the boundary conditions given, i.e., by the device.

Moreover, the possibility to obtain stable connections through customized inlet and outlet holes allows the ease of use necessary to achieve so-called "world-to-chip" connections [40], avoiding the use of magnetic gaskets, clamps, or glue which are usually expensive and time-consuming, and which do not allow reusing the device or capillary tubes [41].

On the other hand, traditional fabrication methods such as lithographical techniques, used to create micro and nanoscale structures, are very useful for prototyping and research experiments, but they do not translate well into mass production, in addition to the use of PDMS, thus limiting its commercial applications owing to the difficulty in upscaling manufacturing and the relatively high cost compared to polymeric alternatives [42]. More critically, intrinsic material properties of native PDMS represent another chapter of drawbacks, such as evaporation, leaching, and absorption of a flowed liquid sample which often makes PDMS unsuitable for repeatable, robust microfluidic biological and chemical analysis applications [43]. Postprocessing of the material (for example, parylene coating) can overcome these limitations but adds an additional backend processing stage and, therefore, makes it more undesirable for commercial manufacturing [44]. Although glass could be used as an alternative, thermoplastics (plastics) are better from a cost and fabrication perspective, allow easy surface treatment, and are generally transparent and biocompatible [45].

The combination of femtosecond laser-based micromanufacturing and micromilling technologies, associated with robust, disposable, plastic substrates and an innovative sealing method and dedicated surface chemical treatment of microchannels for their biological functionalization, has been recently demonstrated for the development of microfluidic

devices for biomedical applications. In particular, it has been shown that combining the micrometric precision offered by ultrashort pulsed laser ablation with the higher machining rate of mechanical micromilling is very beneficial for the rapid and flexible prototyping of polymeric lab-on-chips [46]. The as-fabricated devices can be exploited in applications ranging from the simple on-chip study of cells to the onsite and early diagnosis of diseases [47].

The degree of efficiency revealed by the hybrid microfabrication platform proposed in this work would allow producing 3D microfluidic devices, while embedding additional functionalities such as micromixers and gradient generators. This would push the on-chip platform toward the sample in/answer out concept of point-of-care devices.

In the frame of SMILE (SAW-MIP Integrated Device for Oral Cancer Early Detection) project, we explored several aspects of technology, spanning from simulation of the entire platform through finite element methods (FEM) to tests with artificial samples, as well as from a simulation with particle mixing and a gradient generator to the use of microfluidics using oral cancer cells, envisioning the entire on-chip manipulation and analysis of the sample, with a *plug-and-play* device. In this study, two devices, a micromixer developed on two-level microchannels and a serpentine pathway for the biorecognition of circulating tumor cells, which could work separately or in a subsequential manner, were described. The two devices constitute two building blocks toward the realization of a platform including both sample preparation and biodetection. We performed a completely on-chip functionalization of PMMA, demonstrating a step forward with respect to the current literature, as we obtained the desired results without using complex methods such as UV-curable functionalization or low-pressure radiofrequency (RF) air plasma [48,49]. Anti-EpCAM antibody was chosen as a capture probe to distinguish cancer from noncancer cells in a mixture obtained by keeping together oral cancer-derived cells with blood-derived cells [50], thus mimicking the presence of circulating tumor cells (CTCs) in blood samples. In principle, this antibody can be replaced with any other to identify different kind of cells or extracellular vesicles (microvesicles/exosomes), in order to obtain a liquid biopsy from other biological fluids such as saliva or urine. The investigation of CTCs as early biomarkers of cancer is one of the most promising topics in liquid biopsy, and the translation of this research into a point-of-care device is being explored for the high value which an easyto-use tool could bring in early diagnosis good practices. To this aim, many microfluidic devices for the separation of CTCs and sample enrichment have been realized, including microsieve integration [51–53], inertial microfluidics [34], serpentine path, and many other chip architectures. However, in most reported cases, microfabricated microsieves are assembled into PDMS-based lab-on-chips, with the inclusion of a membrane, resulting in a low-exploitable approach. In the case of inertial microfluidics, although demonstrated to be very effective in separation, the technique requires a complex simulation and experimental validation phase [54] to define the right geometry as a function of the properties (e.g., dimensions, Young's modulus) of the cells to be separated. As the aim of our work was the development of a low-cost easy-to-use device, with the goal of reaching a large audience of possible users and realizing real prevention of the disease, we preferred to use a simple and common serpentine microchannel, in order to lengthen the path that cells are forced to run, with the aim of maximizing the possibility to be captured by immobilized antibodies.

#### **2. Materials and Methods**

#### *2.1. Materials*

Both devices were fabricated by assembling different squared layers of transparent PMMA (Vistacryl CQ; Vista Optics, Gorsey Lane, Widnes, Cheshire, UK). Each layer was machined differently according to the design of the device. For the micromixer device, three 30 × 30 mm<sup>2</sup> PMMA layers were used. The bottom and intermediate layers hosting the micromilled channels were 1 mm thick, while the upper layer containing the inlet/outlet holes was 2.5 mm thick. For the cancer cell capturing device, two square 25 × 25 mm<sup>2</sup> layers were used. The top layer was 5 mm thick, while the bottom one was 1 mm thick. The bonding between layers was performed with pure isopropyl alcohol (Sigma-Aldrich, St. Louis, MO, USA).

Surface functionalization of the microchannels was required, following a different procedure for each device. Particle mixing required surface passivation, which included O<sup>2</sup> plasma surface treatment and incubation with 1 mg/mL bovine serum albumin (BSA) (1%) in phosphate-buffered saline (PBS) buffer (Sigma-Aldrich, St. Louis, MO, USA).

The cell capturing, instead, initially included the usage of 3-aminopropyltrirthoxysilane (APTES 5%) in ethanol, glutaraldehyde (0.05%) in water, bovine serum album (BSA) (1%), and Tween®-20 (0.05%) in phosphate-buffered saline (PBS). In addition, we used EpCAM mouse monoclonal antibodies (all reagents from Sigma-Aldrich, USA). It is worth emphasizing that the anti-EpCAM antibodies have the advantage of not being reactive with normal or neoplastic nonepithelial cells and recognize only human EpCAM expressed on the surface of the epithelial cells. A secondary labeled antibody anti-mouse IgG (whole molecule)–FITC antibody produced in goat (Sigma-Aldrich, USA) was used for the fluorescence confirmation assay.

The sealing and working principle of both devices was evaluated by performing a series of tests. For the sample injection/pumping and flow control, we used the Elveflow microfluidic setup (Elvesys, Paris, France), suitable for finely tuning flow injection in a range of 0.4–7 µL/min. For real-time acquisition, we used an Axio Zoom V16 fluorescence microscope (Zeiss, Oberkochen Germany), with an ApoZ1x objective and a numerical aperture (NA) of 0.25.

For the micromixers, we carried out two validation tests using colored inks and fluorescent polystyrene microspheres, with diameters of 200 nm (green) and 1 µm (red) (FluoSpheres *®* Fluorescent Microspheres, Invitrogen, Ltd. 3 Fountain Drive Inchinnan Business Park, Paisley, UK) in ethanol (Sigma-Aldrich, St. Louis, MO, USA). Instead, for the capturing of tumor cells, we injected cells derived from the OECM-1 human oral squamous carcinoma cell line (purchased from SCC/Sigma-Aldrich) and Jurkatt cell line (leukemic T-cell lymphoblast from ATCC).

#### *2.2. Computational Modeling*

The micromixing tool was simulated by modeling a 3D h-junction in Comsol Multiphysics 5 (COMSOL, Inc., Burlington, MA, USA), using the Microfluidics module CFD package and, specifically, "mixture model, laminar flow" for predicting the fluid flow and particle transport.

Navier–Stokes equations were used to predict the fluid flow through the 3D network of interconnected channels.

As it can be seen, we established two different microchannels with two independent inlets, sharing a common outlet. As boundary conditions, we set an equal uniform velocity/flow rate on both inlets, and zero pressure was applied at the outlet. Furthermore, from the material library, we selected pure water as the defined fluid to cover all domains inside the network of the channels, and we applied zero-flow conditions at the channel walls. The fluid temperature in the entire simulation was set to 25 ◦C for water. The fluid was considered incompressible, Newtonian, and with no gravitational effects anywhere in the device. Once the fluid flow regime was parameterized, set and tested, we created a surrogate model for particle mixing. Particle inlets and outlet were arranged following the same logic as the fluidics. Moreover, the applied particle parameters were those of the standard polystyrene beads, using two different sizes of 200 nm and 1 µm, respectively. The drag force was set in accordance with the material's properties. The type of mesh built for our geometry was free tetrahedral. These conditions were kept constant throughout the simulation. The constructing response was tested for evaluation, after setting all the necessary parameters, by checking the simulation results for the distribution of the flow, velocity, pressure, and particle distribution at all crucial domains of the design.

Selecting the proper channel design with the optimal mixing performance was one of the main steps in optimization. To this end, to prove the quality of our selected model, we examined the flow behavior when experimented with a range of shapes for one of the microchannels, including rectangular, rhomboidal, and elliptical designs for the mixing chamber. The numerical model used was validated by implementing a real system.

#### *2.3. Design, Fabrication, and Sealing of PMMA Substrates*

In order to fabricate the microchannels for the mixing module, with desired dimensions of 200 µm width and 200 µm height, and with specific holes as inlet and outlets, we utilized PMMA substrates and the Mini-Mill/GX micromilling machine (Minitech Machinery, Norcross, GA, USA) with a 200 µm two-flute carbide micro end milling tool. The microfluidic network was designed using Solidworks CAD software (SolidWorks Corporation, 300 Baker Avenue, Concord, MA, USA) and transported in machine code file for micromilling control through computer-aided manufacturing (CAM) software. A 150 mm/min feed rate was used to mill the PMMA layers at 20,000 rpm. The alignment of the inlet and outlet with the channels was achieved using an on-board camera of the micromilling machine. The geometry of the micromixer and the aspect of the final assembled device are shown in Figure 1a–d.

**Figure 1.** Design of the three layers (**a**) top; (**b**) serpentine; (**c**) bottom and aspects of the assembled device (**d**).

The serpentine channel used for the cancer cell capturing experiments was fabricated by exploiting the femtosecond laser milling process, as previously described in [21]. We used an ultrafast solid-state laser system (mod. TruMicro Femto Ed.; TRUMPF GmbH+ Co. KG, Ditzingen, Germany) based on the chirped pulse amplification technique, which delivers linearly polarized 900 fs pulses at a wavelength of 1030 nm with an almost diffraction limited beam (M2~1.3). The laser beam was circularly polarized by a quarterwave plate and then focused and moved onto the target surface through a galvo-scan head (IntelliSCANNse 14; SCAN-LAB, Puchheim, Germany) equipped with a telecentric lens of 100 mm focal length. The spot diameter at the focal plane was about 25 µm. The fs laser milling process was carried out by removing the material layer by layer, superimposing two perpendicular scanning paths.

The working parameters used for the serpentine channel fabrication are reported in Table 1.

**Table 1.** Laser micromilling parameters.


After the fs laser process, loosely attached debris was removed by ultrasonic cleaning in distilled water for 10 min. The dimensions of the fs laser-milled microfeatures were measured using an optical microscope (Nikon Eclipse ME600). Moreover, the average roughness Ra of the milled surface was measured by means of an optical ContourGT InMotion (Bruker, Billerica, MA, USA) profilometer with nanometric resolution and was estimated to be <2 µm. This value is negligible compared to the channel height; therefore, we assumed that the roughness did not affect the fluid flow. The PMMA layer with the fs laser-machined serpentine channel was coupled with a flat and smooth PMMA substrate with inlet and outlet holes drilled using the micromilling machine.

For both the devices, the next step in the fabrication process was the bonding of the PMMA layers. For the assembly of the microfluidic device in both experiments, a thermaland solvent-assisted bonding method was implemented. In a protected environment we spin-coated hot isopropyl alcohol on the surface of the substrates, aligned the wet slices, and transferred the devices into the oven by holding them in position with clamps and creating an irreversible bonding. In order to build the multilayered chip, it was necessary to reiterate this process twice, i.e., for bonding the substrates with the microchannels, for placing the substrate with the holes on top of the channels. In the end, the channels with the interconnection hole were buried, while the inlets and outlets remained on the upper layer, thus resulting in a monolithic device assembled with no need for additional glue, luer, or gaskets. The two different studies, for mixing of solutions and capturing of CTC, required a diverse functionalization of the microchannels.

The 3D micromixer underwent O<sup>2</sup> plasma treatment to improve hydrophilicity. After that, the bonding quality and the existence of any leakage were examined by connecting the device to the Elvesys micropumping system through capillary tubes and by gradually increasing the pressure from 10 to 800 mbar. An optical microscope was used for the evaluation. The last step during this process was the in-flow functionalization.

#### *2.4. Microchannel Passivation and Functionalization*

Although PMMA is a good alternative for customizing the design of microfluidic modules, it suffers from high hydrophobicity. Thus, to attain an optimum functionality of our device, it was essential to mitigate the hydrophobicity of the PMMA surface. To this end, the process could be initiated by treating the assembled PMMA slices with O<sup>2</sup> plasma, instigating an improved surface wettability and hydrophilicity for an easier flow of water-based solutions. In the subsequent phase, in order to attenuate any sticking of the particles into the channel surface, which could impact the performance, it was necessary to incubate the chip for 2 h with blocking buffer (1 mg/mL BSA in PBS). The sample's injection during the functionalization process was done directly in-flow. This was mostly obtained thanks to the perfect fitting between the holes and the capillary tubes, allowing for a plug-and-play usage of the device. The stable connections did not require additional glue, gaskets, or clamps to avoid any leakage of the solution, apart from the capillary tubes and the channels.

In the case of the device for cancer cell capturing, it was necessary to flow a sequence of solutions into the microfluidic chip to functionalize it, starting from APTES (5%) in ethanol, in order to increase the hydrophilization and the amine functionality of the PMMA surface. This step was essential both for the easier flowing of the water-based solutions and for promoting the next step. Subsequently, after cleaning the surface with pure water, we injected glutaraldehyde (0.05%) to allow antibody immobilization. Specifically, the glutaraldehyde underwent an imine coupling reaction with the amine group of the antibody, resulting in immobilization of the anti-EpCAM antibody. The ultimate step of this process was the incubation of the device with a blocking buffer (BSA-Tween®20 in PBS), to prevent any cell absorption. Moreover, during this process, the device was connected with the micropumping system through a perfect fitting between the capillary tubes and the micromilled inlets.

#### *2.5. Experimental Tests for Particles and Coloured Liquids*

In order to gain better insight into the performance of the micromixer, the mixing behavior of the device was investigated. Thus, experiments for mixing and gradient generation were performed using colored fluids and particles. We also examined the case when the outlet transmuted into an inlet, to detect the mixing capabilities of the chip and the influence of different paths, lengths, and shapes of the microchannels. Our first attempt was to mix two different colored liquids. For this, we used the capillary tubes to simultaneously inject the liquids into the serpentine and reservoir channels. As a second checkpoint, we chose two different kinds of particles with diameters of 200 nm and 1 µm. Both experiments were performed at different flow rates.

For the injection of both samples, we connected the chip with the Elveflow microfluidic set up. The setup was equipped with an OB1 base module, two MkIII+ channels for pressure control, and two microfluidic sensors, with an analogous temporal flow control. The two inlets were connected with two different vials, containing either the colored liquids or the particles samples. For evaluating the mixing quality, the microchip was placed under a microscope, enabling real-time evaluation of the flow behaviors, as well as image acquisition.

#### *2.6. Experimental Tests for Cells*

In the CTC capturing experiment, we grew cells in an incubator at 37 ◦C with 5% CO2, in suspension in RPMI 1640 complete growth medium and in adhesion in complete Dulbecco's modified Eagle's medium for Jurkat cells and OECM-1 cells, respectively. The growth medium was renewed every 2 days. A few minutes before the experiment started, cancer cells were suspended and harvested in 0.05% trypsin.

For sample injection into the device, we used the Elveflow micropumping system. We visualized the capturing with real-time image acquisition using an Axiozoom Zeiss V16 fluorescence microscope.

Finally, after detachment, the cells were washed and resuspended in DMEM medium. Jurkat cells were centrifuged, counted, and resuspended at the right dilution in order to inject them into the microfluidic chip. Flow rate was tuned at 7 µL/min.

Once the cells remained in contact with the microchannel walls, the microfluidic chip was gently washed with PBS, and adhered cells were stained with a subsequent injection of (i) anti-EpCAM antibody and (ii) secondary FITC-labeled antibody, in order to identify tumor cells blocked at the channels' surface.

#### **3. Results and Discussion**

#### *3.1. Verification of the Numerical Model*

The 3D fluidic micromixer was studied using CFD simulations, under different regimes, i.e., mixing of fluidics, and mixing and dilution of particles. The results of the interconnected microchannels for the flow velocity distribution and the pressure (Figure 2a,b) offered us the opportunity to gain better insight into the flow behavior. Specifically, Figure 2a depicts the velocity distribution before and after the mixing point, showing a maximal flow at the junction point and the minimal flow predominating at the inlets up to the point of convergence. The velocity at the side walls appeared to be infinitesimal. Concomitantly, the levels of pressure at the entire design are reported in the Figure 2b. As can be seen, the maxima and minima pressure levels were inversely proportional to the velocity. Subsequently, the flow rate was calculated using Equation (1).

$$\mathbf{Q} = \mathbf{A}\mathbf{u}^-\,,\tag{1}$$

where Q is the flow rate, A indicates the cross-sectional area, and u is the average velocity.

To reveal the particle mixing behavior, a suspension of particles with sizes 1 µm and 200 nm was simulated and injected in both inlets. Moreover, in order to find the best combination of flow rates from the two inlets, we tested a range of values ranging from 1 µL/min to 5 µL/min. The recorded pressure was varied as a consequence of this variation. The mixing of the two different populations is illustrated in Figure 2c–f in the time range of 0–67 s.

**Figure 2.** Simulated flow velocity (**a**) and pressure (**b**) in proximity of the interconnection point of **Figure 2.** Simulated flow velocity (**a**) and pressure (**b**) in proximity of the interconnection point of the micromixing channel. Mixing of the two populations of nanoparticles at different time points, from the beginning of the experiment (**c**) at time 0 to the complete mixing obtained by (**d**): 23 seconds, (**e**): 37 seconds and accomplished after 67 seconds (**f**).

Design variables, such as the final shape of the microchannels containing a mixing chamber, were investigated. Specifically, we drew and simulated microchannels in rhomboidal, rectangular, and elliptical shapes in order to find the optimal one. Figure 3 shows the comparison of the fluidic velocity for the three different geometries: rectangle (3a), rhombus (3b), and oval (3c). A shared outcome among all cases was the high velocity in the center of the device and lower velocity at the edges. This condition was enhanced for the rectangular shape. Moreover, it was proven from experimental results that angular edges (as can be identified in the rhomboidal and rectangular shapes) are more prone to accumulate bubbles than round edges [55,56]. Thus, in our final device, we incorporated the elliptical shape for realizing the microfluidic reservoir.

#### *3.2. Design and Fabrication of the LOC Devices*

The findings of our simulation study in predicting the mixing process were verified by fabricating a micromixer with the selected oval and serpentine design. It was of crucial importance to select the optimal design parameters that had the greatest influence on the mixing quality. The first step for the realization of the device was to draw the CAD file and to transfer it into machine code for its fabrication through the micromilling machine. The proposed geometry for mixing experiments, as schematically illustrated in Section 2.3, contained a 7 cm long serpentine-shaped channel in the center of the device organized in six loops, with a reservoir-shaped channel in the bottom with a total length of around 5 cm. The top layer contained three holes (diameter: 1.8 mm) which were defined as inlets and outlets for the capillary tubes. Furthermore, the layer with the serpentine channel also featured a buried hole, which served as a junction point between the channels on the bottom layer and the top layer, thereby creating a 3D microfluidic pathway. The common portion after the junction point ran for 1.5 cm. The device was constructed/assembled from the three individual levels which were separately fabricated, using a mechanical micromilling machine with a 200 µm tool.

**Figure 3.** Simulated flow behavior into three different reservoirs: (**a**) rectangle; (**b**) rhombus; (**c**) oval. The lowest flow velocity occurred at the edges of (**a**,**b**).

Regarding the device used for cancer cell capturing, the layout was based on a serpentine microchannel with a square cross-section of 100 µm per side and a total length of 180 mm. The purpose of this design was to increase the active path and proportionally increase the possibility of capturing cells. As displayed in Figure 4a, the device consisted of two PMMA substrates. In this case, the upper substrate was micromilled in both faces. For the lower substrate, we exploited fs laser technology to fabricate the serpentine-shaped channel. To drill the inlet and outlet, we again used a mechanical micromilling machine, this time with a 400 µm tool, ensuring tight connections, since it fit perfectly with the capillary tubes and gave the opportunity for plug-and-play connections. The connection between the holes and the serpentine channel was achieved using two auxiliary channels with a diameter of 600 µm and length of 5 mm, fabricated on the same substrate. Lastly, the bottom PMMA flat layer allowed the sealing of the serpentine channel. No cracks, burrs, or recast layers stemmed from the microfabrication process, thus also providing a great transparency (Figure 4b). Furthermore, the roughness of the bottom channel (Ra = 2 µm) did not affect the fluidic transport of the cells since it was negligible compared to the channel's height.

The device was assembled and functionalized as described in Sections 2.3 and 2.4, and the possibility of real-time monitoring of flow into the microchannels, combined with the serpentine shape and a slow flow rate (2 µL/min), enabled us to attain a tool for exploiting a very high surface/volume ratio in terms of active binding sites for antibodies.

**Figure 4.** Features of the cell capture device (**a**). PMMA microfabrication allowed for complete transparency of the device. Inlets and outlets were designed to perfectly fit with capillary tubes (**b**).

#### *3.3. Mixing and Gradient Generation Experiments*

We characterized the sealing of the micromixer by visualizing the mixing behavior using two different colored liquids (Figure 5a). The progressive filling of the microchannels was supervised using a microscope. As can be visualized in Figure 5b, the paths of the 3D microchannels could be observed with a single microscope frame, allowing the contemporary monitoring of the multilevel structure. It was of crucial importance to make sure that both solutions arrived simultaneously at the mixing point; hence, we initially set the flow rate at 3 mL/min, but adjusted it later on, when needed. In particular, the complete transparency of the device allowed monitoring the channels while they were progressively filled and differently tuning the parameters due to the diverse shape, resistance, and velocity of the flow in each path. We noticed that the best combination for synchronized arrival at the junction was to set the flow rate to 2.08 µL/min and the pressure to 42.80 mbar for the serpentine channel, whereas these values were set to 1.65 µL/min and 63.08 mbar for the reservoir. The distribution of liquids in the channels and how they initially flowed independently (the pink solution in the upper serpentine channel and the blue one at the reservoir), before being mixed at the meeting point and assuming a violet color, clearly demonstrated that the buried hole (indicated by an arrow) interconnected the two fluids, which became indistinguishable after turning violet. Furthermore, the chip could be used as gradient generator tool using the inlet and outlet alternatively, by tuning the flow rates of the channels and establishing dominance of either the pink or the blue solution (Figure 5c–f).

We then assessed the utility of our apparatus using two test samples, where each one contained green fluorescent particles of 200 nm size with 9.1 × 105/mL concentration and red fluorescent particles of 1 µm size and 7.2 × 105/mL concentration, and we injected them into the reservoir and serpentine channel, respectively. The flow of each solution could run separately into the device as explained in the cartoon in Figure 6a. By using the fluorescence microscope, as can be seen in Figure 6b,c, the two different populations flowed separately (the greens were detected only in the bottom channel and the reds were detected only in the upper one). The final injecting parameters we applied in this experiment to simultaneously reach the common portion were a flow rate of 2.13 µL/min and 26.42 mbar pressure for the red particles and a flow rate of 1.46 µL/min and 39.52 mbar pressure for the green particles.

**Figure 5.** (**a**) Whole device connected to micropumps under the microscope; (**b**) frame acquired after complete mixing of pink and blue ink. (**c**–**f**) Modulation of flow rates, resulting in different mixing conditions of the inks. Scale bars: 500 µm.

**Figure 6.** Solutions contained in the two channels flow separately into the device until they reach the interconnection point. (**a**) Scheme of the microchannel network filled with green and red nanoparticles. Green particles run in the bottom channel (**b**), while red ones run in the upper serpentine channel (**c**).

Successively, we performed the same experiment using a wide range of flow rates to test the performance of the chip. In all cases, to achieve a mutually proportional flow rate in both channels, every flow change in one channel was followed by a pressure adjustment in the other. Figure 7 presents the time-lapse images of particles captured soon before (Figure 7a–c) and after (Figure 7d–f) the mixing point. Here, the particles moved from their individual channels to the common one, thus verifying the expected mixing efficiency.

**Figure 7.** (**a**,**b**) Images of microchannels with green and red particles while running separately before reaching the interconnection point and (**d**,**e**) immediately after. (**c**,**f**) Merged images of green and red fluorescence acquisitions.

#### *3.4. CTC Capture Experiments*

An extensive test was also performed to evaluate the ability of the device to distinguish cancer cells from blood cells. With this aim, we used and immobilized anti-EpCAM antibodies able to recognize human EpCAM, which is a membrane biomarker typically present on the surface of tumoral epithelial cells. The immobilization procedure described in Section 3.3 resulted in the possibility of PMMA microchannels working as capture sites for oral cancer cells. Hence, as a proof of concept, we created a cell mixture composed of two different populations: Jurkat cells (blood-derived cells) and the OECM-1 human oral squamous carcinoma cell line (epithelial-like cells from human oral cancer). In this way, we were able to simulate the contents of a real complex sample. The two different samples were prepared separately and contained 1 × 10<sup>6</sup> cells/mL from the Jurkat line and 1 × 10<sup>4</sup> cells/mL from the OECM line. We injected the cell suspensions slowly through the serpentine channel with a flow rate of 7 µL/min. As Figure 8 displays, cells were recognized and blocked as long as they were expressing the EpCAM antigen on their membrane. Therefore, OECM-1 cells were captured on the inner walls of the channels, and most of them remained after the washing with PBS. To maximize the possibility of interaction of cells and wall channels, we used a very low flow rate (2 µL/min) to inject the cell suspension into the serpentine path. Moreover, PMMA device with its complete transparency provided sufficient proof of concept for the successful distinction of cancer cells from normal blood cells and their immobilization in a label-free manner. In the bright-field images of Figure 8, cells fixed to the microchannels walls are highlighted by a green spot.

**Figure 8.** (**a**,**b**) Two frames in bright-field acquisition related to two different regions of the capture device. Attached cells are clearly recognizable and are highlighted by green crosses.

In order to demonstrate that captured cells were tumoral cells, we performed some additional labeling. Once the cells were blocked at the microchannels walls, we again injected anti-EpCAM antibody solution, after which we flowed a solution of secondary FITC-labeled antibody, able to bind the Fc portion of the primary antibody. In this way, we were able to selectively label the cell membrane of oral cancer cells. As visible in Figure 9, the membrane of fixed cells was labeled with fluorescent green antibody. We can, thus, conclude that the functionalized microchannels were able to selectively capture tumor cells.

**Figure 9.** Identification of OECM fixed cells with anti-EpCAM antibody and secondary FITC-labeled antibody. (**a**) Bright-field acquisition of cells; (**b**) green fluorescent acquisition; (**c**) merged image of (**a**,**b**).

#### **4. Conclusions**

Plug-and-play devices for the rapid diagnosis of diseases are on the rise for the possibility to obtain quick results in a low-cost and easy-to-use manner. This paper, focusing on the development of a plastic disposable tool, describes the possibility of combining different functionalities, proposing a single chip able to stabilize, preserve, and prepare biological samples. The platform includes a module for sample preparation and mixing of solutions and a detection module, which was demonstrated to capture circulating tumor cells. The produced devices were fabricated via a highly customizable combination of fs laser and micromilling methods using low-cost plastic substrates, while allowing easy connections to the microfluidic system for in-flow functionalization and sample manipulation. The proposed functionalization is a proof of concept which can, in principle, be applied to the detection of other biological entities (exosomes, microvesicles,

and so on), by simply modifying the antibody immobilized on the surface of PMMA microchannels. In this case, the validity of the assays was confirmed by using fluorescent probes, which in turn identified the micromixing of nanoparticles and the selective binding of tumor cells in a mixture of normal and cancer cells. The proposed devices may also be of great importance in the case of cancer cell investigations from other body fluids, e.g., saliva, which in turn may require preliminary steps for sample manipulation/dilution or reagent addition. These features, in the era of COVID-19, are very important; for example, a recent release from the American Food and Drug Administration (FDA) authorized the use of home-collected saliva to detect SARS Cov-2. In this way, patients are allowed to self-collect samples for analysis in order to improve accessibility to COVID-19 testing and decrease the risks of infection for medical personnel. Moreover, automatic and low-cost devices, in pandemic contexts, have the possibility to minimize interactions between patients and medical personnel, thus furtherly lowering the probability of infections without affecting access to large-scale screening programs for cancer (and other diseases).

**Author Contributions:** Planning and design, M.S.C. and F.F.; experimental activity, S.Z., A.V., C.G., M.S.C. and F.F.; experimental setup, A.A., F.F. and M.S.C.; implementation of the research, main conceptual ideas, and proof, E.P., M.S.C. and F.F.; analysis of the results, M.S.C., E.P., F.F., A.V. and A.A.; writing of the manuscript, S.Z., M.S.C., A.V. and C.G.; manuscript revision, S.Z., A.V., M.S.C., C.G., E.P., A.A. and F.F.; project supervision and funding responsibilities, M.S.C. and F.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was supported by the following funding programs: "SMILE (SAW-MIP Integrated Device for Oral Cancer Early Detection) project, part of the ATTRACT program funded by the European Union's Horizon 2020 Research and Innovation program (grant agreement: 777222)"; PRIN 2017 Project "Prostate cancer: disentangling the relationships with tumor microenvironment to better model and target tumor progression" (grant number: Prot. 20174PLLYN); Progetto PON ARS01\_00906 "TITAN—Nanotecnologie per l'immunoterapia dei tumori", finanziato dal FESR nell'ambito del PON "Ricerca e Innovazione" 2014–2020—Azione II-OS 1.b).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **A New Predictive Technology for Perinatal Stem Cell Isolation Suited for Cell Therapy Approaches**

**Silvia Zia <sup>1</sup> , Giulia Martini 2 , Valeria Pizzuti 2,3 , Alessia Maggio 2 , Giuliana Simonazzi 4 , Pierluigi Reschiglian 1,5 , Laura Bonsi <sup>2</sup> , Francesco Alviano 2 , Barbara Roda 1,5 and Andrea Zattoni 1,5, \***


**Abstract:** The use of stem cells for regenerative applications and immunomodulatory effect is increasing. Amniotic epithelial cells (AECs) possess embryonic-like proliferation ability and multipotent differentiation potential. Despite the simple isolation procedure, inter-individual variability and different isolation steps can cause differences in isolation yield and cell proliferation ability, compromising reproducibility observations among centers and further applications. We investigated the use of a new technology as a diagnostic tool for quality control on stem cell isolation. The instrument label-free separates cells based on their physical characteristics and, thanks to a microcamera, generates a live fractogram, the fingerprint of the sample. Eight amniotic membranes were processed by trypsin enzymatic treatment and immediately analysed. Two types of profile were generated: a monomodal and a bimodal curve. The first one represented the unsuccessful isolation with all recovered cell not attaching to the plate; while for the second type, the isolation process was successful, but we discovered that only cells in the second peak were alive and resulted adherent. We optimized a Quality Control (QC) method to define the success of AEC isolation using the fractogram generated. This predictive outcome is an interesting tool for laboratories and cell banks that isolate and cryopreserve fetal annex stem cells for research and future clinical applications.

**Keywords:** fetal stem cells; amniotic epithelial cells; isolation protocol; quality control; label-free sorting; diagnostic tool

#### **1. Introduction**

Advanced therapy medicinal products (ATMP) are medicines based on genes, cells and tissues to treat human diseases. The somatic-cell therapy consists of cells infusion that replaces tissue functions, cures and prevents diseases. In the last decade, cell therapy approaches, and in particular stem cells (SCs) treatments, are increasing. Adult SCs are widely used to treat malignant diseases like leukemia by hematopoietic stem cells transplantation, and Graft Versus Host Disease (GVHD) by bone marrow mesenchymal stem cells (BM-MSCs) for their immunomodulatory capacity [1–5]. Among stem cell types, perinatal SCs have gained attention because they possess wide differentiation potential and tolerogenic ability [6]. Placenta is a rich source of stem cells: mesenchymal and epithelial cells with staminal characteristics can be derived and it was proven their therapeutic potential in various disease models [7]. Amniotic epithelial cells (AECs) derive from the innermost layer

**Citation:** Zia, S.; Martini, G.; Pizzuti, V.; Maggio, A.; Simonazzi, G.; Reschiglian, P.; Bonsi, L.; Alviano, F.; Roda, B.; Zattoni, A. A New Predictive Technology for Perinatal Stem Cell Isolation Suited for Cell Therapy Approaches. *Micromachines* **2021**, *12*, 782. https://doi.org/ 10.3390/mi12070782

Academic Editor: Nam-Trung Nguyen

Received: 24 May 2021 Accepted: 28 June 2021 Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

of the amniotic membrane, the one in direct contact with the amniotic fluid. They possess the ability to differentiate toward all three germ layers, they are not tumorigenic and they have immunosuppressive features. AECs are cuboidal epithelial cells firmly adherent to a thick basement membrane [8], expressing high levels of epithelial adhesion molecules, such as EpCAM (CD326) and integrin subunits (CD29 and CD49f), while lacking typical stromal markers expression [9]. AECs also express several pluripotency markers including octamer-binding protein 4 (OCT-4), SRY-related HMG-box gene 2 (SOX-2), and Nanog and showed multilineage differentiation capacity [10]. Perinatal cell populations, including AECs, are also characterized by physiological immunomodulatory and anti-inflammatory properties due to their embryonic origin [11]. Pre-clinical studies have shown AECs' therapeutic effect in neurological disorders, lung injury, liver injury, diabetes, acute kidney failure, cardiovascular diseases and wound healing among many [12–22]. Moreover, AECs have been proved safe and non-tumorigenic upon transplantation, with no expression of telomerase and limited growing potential in culture [10].

Despite their efficacy and safety, AECs expressed differences in stemness characteristics, differentiation potential and immunomodulatory activities depending on the heterogeneity of primary derived cells which lead to a variable effect based on population composition [23,24]. Thanks to their early origin and easy access, AECs could be isolated, cryopreserved in specialized cell banks for future autologous therapy approaches [25]. Despite AECs isolation method is easy and does not require expensive material, the yield of AECs can be quite variable and their characteristics appear to be dependent on the genotype, age of the donor, region of cell isolation on placenta [24], cross-contamination of amniotic epithelial and mesenchymal stromal cell, isolation protocol, epithelial-to-mesenchymal transition of AECs [26] and measuring methods that are used for characterization [27]. Few studies have focused attention on membrane microscopic observation before enzymatic treatment [28] but it is operator-dependent. Several protocols have been proposed for isolation of hAECs with a wide range of yielded cells, viability and purity [29,30].

Reproducible and accurate systems are needed to standardize the isolation protocol of primary SCs. Identification and selection procedures able to isolate stem cells are essential to most cell therapy models. Multiple methods have been developed including mechanical sorting, surface receptors or biological markers of stemness. All these processes involve the knowledge of specific marker expression, which is not always known, and cell manipulation that must be scalable and amenable to GMP procedures. These requirements may be no trivial. Therefore, development of new technologies and relevant application methods are welcomed.

Microfluidic systems are widely used to test quality of pharmaceutical compounds but the working dimensional range belongs to micro-nanoparticles [31]. In the last decade, field-flow fractionation (FFF) has proven its capacity to analyze, discriminate and separate a wide size range of cells mixture based on their physical characteristics with high resolution and throughput [32,33]. In order to work with cells that have the ability to adhere to plastic, a novel method has been developed, the Non-Equilibrium Earth Gravity Assisted Dynamic fractionation (NEEGA-DF) [34]. Adhesion and contact of cells with the fractionation device are totally avoided by in-flow injection, by the absence of stop-flow cell sedimentation, and by using elution flow rate values able to generate hydrodynamic forces that are intense enough to lift and keep cells away from the channel wall. Cells having different physical characteristics acquire different velocity inside the capillary channel and elute at different time, so it is possible to collect the subpopulations composing the biological sample. It was proven that mesenchymal and epithelial cells from different organs origin, showed specific profile outcome of the separation process meaning that this method is suitable to underlined intra-differences in the cell population that other techniques do not do [34]. We developed an automated instrument that implements the NEEGA-DF method (Celector®, Stem Sel srl, Bologna, Italy), using a micro-camera for cell detection and a specifically designed software for image acquisition, post-processing and data analysis. The output of

the instrument is a multiparametric fractogram representing number, size and shape of the eluted cells as a function of fractionation time and it is the fingerprint of the cell sample.

In this study, we used the instrument Celector® to perform the quality check of freshly isolated AECs, to compare possible differences in cells' yield and composition of amniotic membrane treated with two concentrations of trypsin, 0.1% and 0.25%. The live fractogram was used as predictive data to define successful isolation procedure and additionally, postprocessing image data were compared to biological data of cell recovery, cell vitality and adhesion ability.

#### **2. Materials and Methods**

#### *2.1. Instrumentation*

#### 2.1.1. Fractionation Principle and Procedure

The separation is obtained in a capillary device (channel) with rectangular cross section where cells suspensions are eluted through a laminar flow of physiological buffer. When a cell suspension is injected at a flow rate of 1mL min into the system, cells are transported by the flow and reach a specific position across the channel thickness during transportation due to the combined action of gravity, acting perpendicularly to the flow, and opposing lift forces that depend on the morphological features of the sample. Cells at a specific position in the channel acquire well-defined velocities and are therefore eluted at specific times [35]. The in-flow injection, the absence of a stop-flow, cell sedimentation step, and the use of elution flow rate values able to generate hydrodynamic forces that are intense enough to lift and keep cells away from the channel wall, make cells avoid any contact with the device with a consequent maintenance of native properties and high sample recovery.

The fractionation procedure involves at first the decontamination of the fractionation system by flushing with cleaning solution at 1 mL min flow rate. Next, the system was washed copiously with sterile, demineralized water at the same flow rate. Although the NEEGA-DF method is optimized to prevent contact between cells and fractionation device, to block non-specific interaction sites on the plastic walls, the fractionation system was flushed at 0.5 mL min with a sterile coating solution. Finally, it was filled with a sterile mobile phase. All solutions were provided by Stem Sel srl. The instrument is placed under a laminar flow hood to maintain the sterility of the collected cells.

#### 2.1.2. Optical Analysis

Eluted cells were monitored using a micro-camera detector (MER-U3 camera, DA-HENG IMAGING, Beijing, China) that is placed at the outlet of the fractionation channel. The imaging software (Celector Optics, Stem Sel srl, Italy) generates real-time fractogram representing the percentage of frame area covered by the cells versus recorded time. The imaging data are then post processed to obtain number and geometrical features of eluted cells as a function of time. In this work, we focused on the area/diameter and circularity of cells to obtain information on population heterogeneity and composition of possible sub-populations. These geometrical features were then visualized as scatter plot and curves using dedicated data processing (Stem Sel Analyzer), to obtain the average of all parameters in a selected time interval (cell fraction).

#### *2.2. Cell Analysis and Collection*

For every sample, cells were first analyzed to obtain a patient-specific fractogram and identify the fractions to collect. Consecutive analyses were run to increase the number of collected cells per fraction. The fractionated cells were centrifuged, and cell recovery was calculated by erythrosine dye (Sigma, St. Louis, MO, USA) to count alive and dead cells for both groups, 0.1 and 0.25% trypsin.

#### *2.3. Isolation of Human Amniotic Epithelial Cells (AECs)*

The study was approved by the Local Research Ethics Committee (EM894-2020\_68/2017/ U/Tess/AOUBo) and written informed consent was obtained from each healthy donor before specimen collection.

Amnion membranes were retrieved from term pregnancies (37–40 weeks of gestation) delivered by Caesarean section. hAECs were isolated using a modified procedure of the protocol previously reported [36], the amnion layer was mechanically peeled off the chorion layer and immediately washed in Phosphate Buffer Saline (PBS, Gibco, ThermoFisher Scientific, Waltham, MA, USA) without calcium and magnesium (HBSS, PAA Laboratories GmbH, Pasching, Austria) until blood clots were completely removed. For each sample, the amnion was minced into 4 to 6 medium size pieces (25 cm<sup>2</sup> approximately) and divided into two groups: one group was treated with 0.1% of trypsin-EDTA and the other group with a 0.25% trypsin-EDTA for all digestion steps (Gibco). Before enzymatic treatment, membrane was washed in PBS and 0.5 µM EDTA for 10 min then firstly, pieces were incubated for 10 min at 37 ◦C to exclude debris and then incubated for a second and third 40 min enzymatic digestion using the two concentrations to release the amniotic epithelial cells (AECs). Single cell suspension was washed with PBS and tested for viability with erythrosine dye (Biochrom AG, Berlin, Germany) and the number of viable and dead cells were counted. Cellular pellets were resuspended in the growth medium consisted of DMEM high glucose supplemented with 10% FBS, EGF (10 ng/mL; Sigma-Aldrich, St. Louis, MO, USA) and 1% penicillin and streptomycin (all solutions from Gibco) and plated at a density of 100,000 cells/cm<sup>2</sup> in a T25 flask for expansion. The leftover cells, at least 1.2 × 10<sup>6</sup> cells, were analysed using Celector® for quality control of the isolation process and cell sorting.

For the study AECs from the second digestion were diluted to a final concentration of 3 × 10<sup>6</sup> cells per mL and 100 µL were injected. Cells were automatically re-dispersed 3 times to homogenize the suspension and eluted at a flow rate of 1 mL min.

#### *2.4. Downstream Analysis*

Collected cells for each fraction and from the control group were seeded at a cell density of 100,000 cells/cm<sup>2</sup> to observe cell adhesion ability and morphology. In order to define the success of the isolation protocol, cells must adhere to plastic surface and show proliferative ability one week after isolation occur. If cells did not adhere or showed no ability in proliferation, isolation was defined as unsuccessful.

10,000 freshly sorted cells of each group, 0.1 and 0.25% trypsin, were seeded on a glass coverslip. After 4 days, cells were fixed in 10% formalin and stained for nuclear DAPI (Prolonged antifade, Molecular Probes). Images were taken using a fluorescent microscope and analyzed using the NII plugin to determine the number of normal, senescent and mitotic nuclei [37].

#### *2.5. Statistical Analysis*

Statistical analysis was performed using Graph Pad Prism v 8, running the *t*-test and mean and standard deviation were graphed.

#### **3. Results**

#### *3.1. Predictivity*

Amniotic epithelial cells (AECs) were isolated using different concentration of trypsin, 0.1 and 0.25%. Freshly isolated cells from the second digestion were immediately analysed by Celector® to profile populations. Two types of profiles were generated: a profile having two distinct peaks (type 1) (Figure 1A) and a profile with all cells eluted in the first minute of the analysis (type 2) (Figure 1C). Type 1 profile represents alive and proliferating cells while type 2 showed unsuccessful protocol, with no cells attaching to the plate and not able to proliferate. For both type of profile, unretained cells and debris eluted in the first minute of the analysis, then first subpopulation eluted between the 3rd and the 7th minute

of analysis (Fraction 1, F1), and when present the second sub-population eluted between the 7th and the 14th minute (Fraction 2, F2). No difference was observed between the use of different trypsin concentration; both concentrations showed the same predictivity about the isolation protocol. Cell aggregates were observed in F1 for both trypsin treatments, while single cells eluted in F2 (Figure 1B). Sample belonging to type 2-profile, the one resulting unsuccessful in the cell expansion, presented more and bigger cell aggregates in F1 compared to type 1 samples (Figure 1D) and very few cells eluted in F2.

**Figure 1.** Representative images of a successful isolation protocol of AECs (Type 1) and an unsuccessful (Type 2) protocol. Profile represents the number of cells versus time of analysis (**A**,**C**) for both trypsin treatment 0.1 and 0.25%. The time interval of cells collection is shown as a dotted line that divides the two subpopulations F1 and F2 (F1 from the 3rd to the 7th minute and F2 from the 7th to the 14th minute of analysis). Live images of eluting cells are shown for both groups for type 1 and 2 (**B**,**D**). Cell distribution between F1 and F2 based on the calculation of the area under the curve (AUC) expressed as a percentage compared to the total area of the profile. The difference was seen between the F1 and the F2 of the 0.1 and 0.25% groups (**E**); distribution was also expressed as a number of counted cells by the software for each fraction of all samples analyzed (**F**). (*t*-test: *p* < 0.05 \*).

For both treatments, most of the cells eluted in F1, showing a higher intensity of the peak confirmed by the area under the curve (AUC) (Figure 1E) which represents the number of eluted cells. The distribution of cells between F1 and F2 in the sample treated with a 0.25% trypsin was statistically different compared to 0.1% samples: 0.25% samples generated a higher AUC in F1 and a lower AUC for F2 compared to 0.1% samples. Postprocessing analysis of counted cells confirmed the difference between F1 and F2 from both groups, even though it was no significant (Figure 1F). AECs profiling showed the ability of this technology to predict the achievement of the isolation process.

#### *3.2. Quality Control (QC) of Freshly Isolated AECs*

The fractionation profile gave the immediate predictivity of the AECs isolation process while the post processing analysis of the recorded data allowed a better characterization of the examined populations. The software presents the physical characteristics of each cell eluted under the camera, becoming in this way an excellent tool to increase the information besides the fingerprint obtained by the profile. Thanks to this tool, we discovered a difference between the 0.1% and 0.25% samples. When the isolation protocol was successful, AECs derived by the 0.1% showed a more heterogenous population, with two clear subpopulations F1 and F2. Scatter plot representing the cell diameter versus the time of analysis showed the two-populations distribution, F1 showed a wider dimensional distribution composed of bigger cells and cell-aggregates and a second population dimensionally more homogenous and smaller in the F2 (Figure 2A(i)). On the contrary, AECs isolated using 0.25% trypsin showed a more homogeneous population in respect of the cell dimension, with a more compact cell cloud around the diameter of 30 µm especially in F1 (Figure 2A(ii)). When cell circularity was studied, the F1 sub-population express the same average in both trypsin treatments while F2 cells of the 0.25% showed the most circular cells at the 10th minute of analysis and the 0.1% at the 13th minute, an indication of how the two concentrations differ in the membrane treatment and consequent cells release. Compared to the 0.1% AECs, 0.25% AECs are a more homogenous population. Analysis of samples that did not retrieve proliferating cells, clearly showed in the scatter plot distribution that all injected cells eluted in F1 and no cells are present at the 10th minute of the analysis which is the highest point of the second peak in type 1 profile.

Besides the scatter plot, the average of every parameter was calculated for each fraction and compared to the general population (CTRL). The results confirmed the principles of the separation process because cells with higher diameter are in F1 and smaller one elutes later in F2 (Figure 3A,D). The same trend was observed for the circularity parameter: the most circular cells are in the F2. Stem cells have the characteristics to be circular and with well-defined counters, so there is a link between this parameter and then the ability of adhesion and proliferation of these cells. Even though there is only a small difference, F2-AECs from 0.25% trypsin treatment is slightly more circular than 0.1% AECs. One of the hypotheses is that this treatment released smaller and more circular cells compared to the 0.1% treatment. When cellular aspect/ratio was measured, F2 cells showed a lower value compared to F1 cells, which is in line with the higher circularity of these cells. Despite a significant difference between F1 and F2, the difference between F1 and F2 for 0.25% samples is less highlighted than in 0.1% samples. This result is in line with the ability to isolate a more homogenous population using the 0.25% concentration.

#### *3.3. Viability and Cell Recovery*

Membranes treated with the two concentrations of trypsin gave very heterogenous result on cell recovery. For both the concentrations the distribution is rather wide as shown in the graph (Figure 4C). Even though there was not a significant difference, the membrane treated with a 0.1% trypsin recovered almost half of the cells of 0.25% (4.35 × 10<sup>6</sup> vs. 8 × 10<sup>6</sup> cells). At the macroscopical and microscopical observation, membrane treated with 0.1% trypsin resulted whiter and cells were still present on the membrane (Figure 4B(i,ii)) while membrane treated with the 0.25% concentration showed gel-like consistence and very few cells were seen on the membrane (Figure 4C(i,ii)). The 0.25% treatment results more efficient in cell removal as seen by a higher cell recovery (Figure 4D). For the 0.1% samples, cell recovery was equally distributed between F1 and F2 while for the 0.25% treated samples, F2 was less abundant than F1 (Figure 4E). To see the effect of the digestion process, the viability of post-fractionation cell collection was measured by counting the number of alive cells compared to the total cells in each fraction. Statistical difference was seen between F1 and F2 for the 0.1% group with the F1 have been the most vital one, while for the 0.25% group no difference was observed because of the heterogeneity of the F2 group.

**Figure 2.** Representative images of the physical parameters of AECs from 0.1 and 0.25% trypsin treatments. Diameter and circularity were analyzed and compared between the two treatments. Scatter plot of diameter (**A**) and circularity (**B**) for 0.1% (**i**), 0.25% (**ii**) and the overly of the average (**iii**) for the successful protocol and for an example of an unsuccessful protocol for the cell diameter (**C**) and circularity (**D**).

**Figure 3.** Cells' geometrical features for AECs derived using 0.1% trypsin (**A**–**C**) and 0.25% trypsin (**D**–**F**) for diameter, circularity and aspect/ratio (A/R). (*t*-test: *p* < 0.0001 \*\*\*\*).

**Figure 4.** Representative images of amniotic membrane pre-treatment (CTRL) (**A**), macroscopic (**B**-**i**) and microscopic image using a 4x objective (**B**-**ii**) post-treatment using 0.1% trypsin and macroscopic (**C**-**i**) and microscopic image using a 4x objective (**C**-**ii**) post-treatment using 0.25% trypsin; the number of cells recovered after enzymatic treatment counted using the erythrosine solution to discriminate alive cells (**C**); the number of collected cells per analysis (run) for each fraction for both trypsin treatment, 0.1 and 0.25% (**D**); Percentage of viable cells for collected fractions F1 and F2 for both enzymatic treatment (**E**). (*t*-test: *p* < 0.01 \*\*).

#### *3.4. Morphology and Adhesion Properties*

Morphologically, cells from both conditions had an epithelial morphology, small in size and few cells showing long pedicles (Figure 5A). In the 0.1% condition we observed few cells with wider cytoplasm compared to 0.25% which resulted homogenous and there was no difference when adherent cell area was measured (Figure 5C).

**Figure 5.** Representative images of AECs from both enzymatic treatment, 0.1% (**A**) and 0.25% (**B**) from control population (CTRL), F1 and F2 derived AECS and it is visible that a minority of the cells in F1 attached to the plate. AECs from 0.1% and 0.25% treatment did not show difference in area size (**C**). The ability to adhere to the plastic surface was scored (from 0 to 4, min to max) for the separated AECs from F1 and F2 for both trypsin groups. F1 derived AECs have a lower ability to adhere compared to the F2 cells, which are the ones more vital and able later to proliferate (**D**). Cell diameter was measured by post-processing analysis and cell population was divided following the same time interval used for the fresh sample. AECs at passage 0 and 2 were analyzed for both trypsin treatments (**E**). AECs control from 0.1 and 0.25% were grown in culture to monitor proliferation ability (**F**). The 0.1% AECs have a higher adhesion propensity (70%) while AECs from 0.25% adhere to the culture dish only in 50% of the cases. Even with the difference in initial adhesion, AECs grow until the 4th passage in culture. (*t*-test: *p* < 0.05 \*).

AECs from both trypsin treatment, 0.1% and 0.25%, had the ability to adhere to plastic with a different grade, 70% for the 0.1% treated samples and 50% for the 0.25% samples. This difference could be explained by the different grade of adhesion of single fractions. For the 0.1% samples, F2 was mostly adherent and many cells from the F1, whereas in the 0.25% group there was a significant difference between F2 and F1 cells, with the latter showing a lower adhesion profile (Figure 5D). When 0.1 and 0.25% AECs were kept in culture, we saw a similar trend, meaning that the F2 component from 0.25% is probably contributing to the proliferation (Figure 5E).

Morphologically AECs from both groups have similar adhesion area and could proliferate till the 4th passage in culture. Despite the AECs in adhesion did not show any difference in the 2D cell dimension between the two concentration treatments, scatterplot graphs showed the heterogeneity of the two populations (Figure 6). AECs from 0.1% treatment showed a temporarily wider profile, with cells eluted already from the 4th to the 13th minute while 0.25% AECs eluted from the 7th till the 13th minute. The higher concentration of trypsin performs a selection on the population. To investigate even more the heterogeneity of these cells we investigated the shape and dimension of the nuclei (Figure 6). Imaging analysis of DAPI stained nuclei showed a similar distribution of normal nuclei in the 0.1% samples, with no difference between the F1 and F2 (Figure 7A). Interestingly, we observed a lower presence of normal nuclei in the F1 from 0.25% samples compared to F2 (63 vs. 77%, *p* < 0.0001). Cells derived by 0.1% trypsin treatment had a higher presence of cells with senescent nuclei (LR) compared to 0.25% cells and the majority resides in the F1 both in the 0.25 and 0.1% group (Figure 7C). Cell division was observed in culture and the 0.25% samples seemed to be more proliferating than the 0.1% even though the fraction with the higher presence of mitotic nuclei was the F1.

**Figure 6.** Representative images of the physical parameters of AECs from 0.1% and 0.25% trypsin treatments at passage 0 in culture. Profile was analyzed and compared between the two treatments (**A**). Scatter plot of diameter (**B**) for 0.1% (**i**), 0.25% (**ii**) and the overly of the average (**iii**).

**Figure 7.** Morphological analysis of the nuclei showed presence of normal nuclei (**A**), mitotic (**B**) and senescent (**C**). Representative images of immunofluorescence staining for DAPI to visualize shape heterogeneity in fractions samples for both enzymatic treatments (**D**,**E**). (*t*-test. *p* < 0.05 \*, *p* < 0.01 \*\*, *p* < 0.0001 \*\*\*\*).

#### **4. Discussion**

This study evidenced the potential of this technology as a quality control system to predict the outcome of stem cell isolation procedure and the quality of freshly isolated cells, specifically amniotic epithelial cells (AECs).

Stem cells act following different mechanisms to treat damages: they differentiate into specialized cells, modulate host immuno-response and release specific factors that stimulate tissue regeneration. After 2018, the number of clinical trials using MSCs is slowing down because comparison among trials is difficult due to cells heterogeneity, organ origin, different cell preparation protocols and different passage numbers. Standardization of cells is necessary to obtain more reproducible and comparable results among centers [38–40]. In the last decade a huge effort has been made to ameliorate standards for cell culture and production, using optimal reagents, industrialized bioreactors where all variables such as temperature and gas exchange are constantly controlled, in order to avoid operator's variability. Despite standardize protocols and reagents may help the reproducibility of isolation protocol and the in vitro cell expansion, the inter-patient differences are still under debate. As an example, patient derived induced pluripotent stem (iPS) cells are key platform to study the impact of human cell type-specific gene regulation but inevitably, differences between donors affect most iPSC traits, from cell morphology to DNA methylation, mRNA and protein abundance to pluripotency and differentiation [41]. Even after controlling for genotype, substantial experimental heterogeneity remains.

The therapeutic potential of AECs is well known and recently, it was proven to enhances the engraftment, viability and graft function of pancreatic islet organoids in diabetes model when co-cultured, which can be used as a cell therapy approach for diabetes [19]. The possibility to use these cells in an allogenic manner raised their interest even more in the scientific community [42]. The isolation procedure is very simple consisting in the use of trypsin to detach the epithelial layer from the membrane, leaving the mesenchymal component attached to it. In our records, AECs recovery and proliferative ability were very heterogeneous and could not be linked to a macroscopical observation of the quality of the membrane, therefore we investigate the potential of the NEEGA-DF method as a quality control system. Technically, enzymatic digestion could lead to the detachment of cells in an aggregate form that unlikely adhere to the plate and could hinder single cell attachment. Clearly, cell aggregates are heavier than single cells, so through the method could be separated from the single cells. Moreover, dead and damaged cells are usually eluted in the first minute of the analysis because they do not reach the equilibrium, inferring with the quality of the cells' suspension.

Results showed the prediction of the live fractogram to define the success of the isolation protocol: bimodal profile is linked to alive, adherent and proliferating cells, whereas fractogram having one single peak or with a low intensity second peak represents the non- or poorly- adherent cells. Cells isolated using both concentrations of trypsin, 0.1 and 0.25%, showed an equal prediction of the fractogram, which means that the general action of the enzyme on the membrane is very similar. This prognostic tool can save time, reagents and effort in culturing only proliferating cells as the verification of vitality required few days because any additional movement of the dish, to observe cell adhesion at the microscope, could affect cell attachment. The QC method developed delivered the answer to the closed-question on the isolation procedure in 15-min time. Count of alive cells is not a very reliable method to define the quality of digestion, since some samples, characterized by a majority of alive cells, showed a single-peak profile and resulted in non-adherent cells. These cells are probably senescent cells unable to adhere to plastic surfaces or damaged cells by the enzymatic treatment, with a consequent selection of the solely proliferative clones. Nuclei analysis showed the presence of senescent cells mostly in F1, both in 0.1% and 0.25% samples, even though the lower concentration treatment showed a higher number of senescent nuclei both in F1 and F2 compared to 0.25% cells. We hypothesize that 0.25% concentration performs a selection, removing senescent cells, and leaving a more homogenous sample as proved by scatter plot graph representing cells' diameter after first expansion in culture. These cells elute at the same time interval of F2 proving their initial belonging of the F2 peak and the consequent no attachment of mostly F1 cells. Even at later passage, 0.25% AECs showed a more homogeneous dimensional distribution, whereas 0.1%, especially at passage 2 in culture, showed an increase in the area of cells eluted in the F1 interval. We conclude that the 0.25% treatment applied a selection on the most proliferate clones of AECs, resulting in adherent proliferating cells.

We presented for the first time an analytical platform that analyze freshly primary cells and understand the effect of isolation process on their viability and propensity to adhere to plastic. The protocol could be implemented in quality control guidelines for scale-up cell production and cryopreservation of cells for future uses with the adding value that the methods separate cells in a label-free mode, therefore no additional manipulation occurs. The possibility to avoid antibodies-sorting method, that additionally stresses cells, are requested by the scientific community [43,44] and new methodologies are valued. We developed an innovative approach of cells' physical characteristics measurements and their visualization in relation to the time of analysis using a scatter plot graph, which allowed a better understanding of the heterogeneity of complex biological samples. The separative potential of the NEEGA-DF methodology was already proven in expanded stem cells from amniotic fluid where we identified the most staminal subpopulation based only on cells profile [24] and dimensional characterization was manual and could not cover all cells analyzed. In this work, the cell imaging process was implemented and new information

can be extrapolated, giving a general overview of the cell composition of the complex biological sample. Often geometrical features of cells are associated with differentiation potential and staminality, even though these studies involve 2D system and the use of special substrates. As an example, using high content imaging, Marklein et al. [45] demonstrated that nuclear morphological profiles of mesenchymal stromal cells had distinct morphological features that were highly predictive of MSC mineralization capabilities. In addition, time lapse imaging of 2D cell cultures in combination with morphological cell analysis are increasingly used to assess, track and even predict MSCs differentiation phenotypic outcomes. Dimension, circularity and aspect/ratio are the main geometrical characteristics that define cells. All these features can be obtained during the cell fractionation and related to functional ability after genetic and differentiation assessments. Hence, by converting the geometrical characteristics of sorted cells into a morphological fingerprint of the biological sample, this information may be used to improve cell culture, sorting of sub-population of interest and as we proved in this work as a prognostic marker to deduce vitality of primary cells.

#### **5. Conclusions**

In conclusion, we proved how the NEEGA-DF method and its technological implementation within Celector® technology are able to predict the outcome of the isolation of AECs using the fractogram, the fingerprint of the sample, in a very short time. Moreover, the data output of the post-processing imaging adds a new type of information of the complexity of the sample to better understand its composition and cell features for its use in therapy applications.

#### **6. Patents**

Celector® is based on a technology patented in Italy (no. IT1371772, "Method and Device to separate totipotent stem cells"), in USA, and in Canada (no. 8263359 US en. CA2649234, "Method and device to separate stem cells").

Stem Sel® has also an Italian patent (IT1426514, "Device for the Fractionation of Objects and Fractionation Method, allowed 2016).

**Author Contributions:** Conceptualization, S.Z. and F.A.; data curation, S.Z. and G.M.; investigation, G.M., V.P. and A.M.; methodology, S.Z., P.R., B.R. and A.Z.; resources, G.S.; supervision, F.A.; visualization, S.Z. and G.M.; writing—original draft, S.Z.; writing—review and editing, S.Z., P.R., L.B., F.A., B.R. and A.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was approved by the Local Research Ethics Committee (EM894-2020\_68/2017/U/Tess/AOUBo).

**Informed Consent Statement:** Informed consent was obtained from each healthy donor before specimen collection.

**Acknowledgments:** We would like to thank the AGD-Associazione Giovani Diabetici (Bologna, Italy).

**Conflicts of Interest:** Andrea Zattoni, Barbara Roda and Pierluigi Reschiglian are associates of the academic spinoff company Stem Sel Srl (Bologna, Italy). The company mission includes the development and production of novel technologies and methodologies for the separation and characterization of cells and bio samples. All the other authors report no conflict of interest since nobody have commercial associations that might create a conflict of interest in connection with submitted manuscripts.

#### **References**


### *Article* **Flexible Enzymatic Glucose Electrochemical Sensor Based on Polystyrene-Gold Electrodes**

**Annika Müsse 1,2,3, \*, Francesco La Malfa 1,4 , Virgilio Brunetti 1 , Francesco Rizzi 1, \* and Massimo De Vittorio 1,4**


**Abstract:** Metabolic disorders such as the highly prevalent disease diabetes require constant monitoring. The health status of patients is linked to glucose levels in blood, which are typically measured invasively, but can also be correlated to other body fluids such as sweat. Aiming at a reliable glucose biosensor, an enzymatic sensing layer was fabricated on flexible polystyrene foil, for which a versatile nanoimprinting process for microfluidics was presented. For the sensing layer, a gold electrode was modified with a cysteine layer and glutaraldehyde cross-linker for enzyme conformal immobilization. Chronoamperometric measurements were conducted in PBS buffered glucose solution at two potentials (0.65 V and 0.7 V) and demonstrated a linear range between 0.025 mM to 2mM and an operational range of 0.025 mM to 25 mM. The sensitivity was calculated as 1.76µA/mM/cm<sup>2</sup> and the limit of detection (LOD) was calculated as 0.055 mM at 0.7 V. An apparent Michaelis–Menten constant of 3.34 mM (0.7 V) and 0.445 mM (0.65 V) was computed. The wide operational range allows the application for point-of-care testing for a variety of body fluids. Yet, the linear range and low LOD make this biosensor especially suitable for non-invasive sweat sensing wearables.

**Keywords:** glucose; glucose oxidase; amperometric biosensor; body fluids; sweat; wearable sensor

#### **1. Introduction**

To prevent and optimally treat diseases, monitoring the health status of patients is of great importance. Among one of the most widespread diseases is diabetes, a metabolic disorder that affects more than 450 million people worldwide, which is characterized by persistent high blood glucose levels that cause vascular damage and affect the heart, eyes, kidneys and nerves. The number of affected people is expected to increase up to 700 million in 2045, and it is estimated that half of all people with diabetes are undiagnosed, which illustrates the large demand for glucose monitoring [1].

Typically, in a clinical setting, glucose levels are measured invasively using frequent blood sampling, either analyzed in central laboratories or directly at the patient's bedside, known as point-of-care testing. The great advantage of using point-of-care instruments in diabetic patients is that the turn-around time is much shorter, which is crucial as changes in the blood sugar level can require immediate adjustment. In addition, less blood volume is needed, which reduces the probability for anemia due to frequent sampling [2]. In a home setting, most commonly, finger-prick testing is used, which can be painful and does not allow for frequent measurements. These drawbacks of conventional invasive glucose testing have been addressed in recent years, and increasing efforts have been made to

**Citation:** Müsse, A.; La Malfa, F.; Brunetti, V.; Rizzi, F.; De Vittorio, M. Flexible Enzymatic Glucose Electrochemical Sensor Based on Polystyrene-Gold Electrodes. *Micromachines* **2021**, *12*, 805. https://doi.org/10.3390/mi12070805

Academic Editors: Maria Serena Chiriacò, Elisabetta Primiceri and Francesco Ferrara

Received: 28 May 2021 Accepted: 5 July 2021 Published: 7 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

develop minimally- and non-invasive methods [3]. Especially, wearable devices have been in focus, as they can allow for continuous measurements [4,5]. This is possible thanks to the correlation between blood glucose levels and the level in other body fluids, such as interstitial fluid, sweat and saliva (Table 1) [6–8].


**Table 1.** Typical glucose concentrations in different types of human body fluids. Data from ref. [9].

To date, a completely non-invasive glucose sensor still is, to the best of our knowledge, not widely available for commercial use despite the high attention this technology gained recently [3,10]. The availability of large amounts of fluid that do not require withdrawal techniques makes saliva an interesting target for glucose sensing and led to the development of such sensors [11]. Yet, despite the correlation between glucose levels in blood and saliva, measurements are only reliable under fasting conditions, which limits usability for patients with diabetes and thus commercialization [12]. On a commercial level, electrochemical biosensors to measure glucose levels in interstitial fluid have been developed [13,14]. A possibility to access interstitial fluid is to access the fluid internally using needles, which penetrate through the skin. These indwelling sensors are widely available and have gained growing market acceptance [12], but potential drawbacks such as the risk for microbial infection remain [15,16]. One such commercially available sensor is the FreeStyle Libre Flash Glucose Monitoring System (Abbott Diabetes Care) with a sensor needle length of 8.5 mm [17]. In addition, less invasive sensors using microneedles to access the fluid, such as the device from Arkal Medical, have been developed [18], but swelling and irritation can occur each time the sensor breaches the skin [12]. Withdrawing the fluid from the skin is an alternative approach that has been exploited in laboratory [19] and by the commercially approved GlucoWatch® system (Cygnus Inc., Redwood City, CA, USA) in the early 2000s [20], but the device had to be retracted from the market. Among the reasons was skin irritation due to the current that was necessary during the reverse iontophoresis process to induce fluid migration across the skin [3]. Other drawbacks of interstitial fluid sensing requiring extraction are the increased lag time and possible contamination with sweat [12].

Due to the possibility to access sweat non-invasively, this body fluid has been brought into focus, in recent years, as a good candidate to allow for non-invasive sensing, but it has been pointed out that, compared with the development of glucose sensors for interstitial fluid, the commercialization of sweat glucose sensors is still low [12]. Although sweat is easily accessible because the whole body is covered with sweat glands, high sweat rates are typically found only in people such as athletes or workers and, indeed, many sensors focus on people during physical activity to guarantee sufficient sample volumes [21–23]. To enable sweat sensing in resting people, often, induced sweating is employed, for example, by local application of sweat stimulants such as Carbachol or Pilocarpine using a reverse iontophoresis approach [24,25]. Further efforts have been made to develop sensors able to operate at low volume levels of 1–5 µL [26].

However, comparatively low glucose concentrations and low excretion rates in resting people represent challenges that have to be addressed by fabricating sensors with low detection limits and taking into account the suitability of microfluidic systems to collect such small volumes [27]. This means that, in order to be able to have a versatile substrate applicable also for low volume microfluidics, the material should show low absorption and low water vapor permeability, which could change the concentrations. Thus, biocompatible polymers with the aforementioned properties such as polystyrene (PS) are preferable over the widely used silicone-based elastomer polydimethylsiloxane (PDMS) to overcome some of these drawbacks [28]. In addition, faster fabrication processes are possible using thermoplastic polymers as PS [29,30].

Ever since the first electrochemical glucose sensor for blood was developed in 1962 by Clark and Lyons [31], an electrochemical approach is still being chosen for most glucose sensors. Enzyme-based sensors allow for high sensitivity and good reproducibility while production is usually possible in a low-cost range [9]. Different sensor generations are distinguished in literature based on the enzymatic reaction side [32,33]. Glucose oxidase (GOx) sensors are based on the enzymatically catalyzed oxidation of glucose to gluconolactone in the presence of oxygen. The coenzyme flavin adenine dinucleotide (FAD) is required as electron acceptor in this reaction and is then regenerated by reacting with O<sup>2</sup> to generate hydrogen peroxide (H2O2) [34].

$$\text{GOx(FAD)} + \text{Glucose} \rightarrow \text{GOx(FADH\_2)} + \text{Gluconolactone} \tag{1}$$

$$\text{GOx(FADH}\_2) + \text{O}\_2 \rightarrow \text{GOx(FAD)} + \text{H}\_2\text{O}\_2\tag{2}$$

When a sufficient potential is applied at the electrodes, H2O<sup>2</sup> is oxidized, and a current can be measured which correlates with the amount of H2O<sup>2</sup> that has been produced and thus correlates indirectly with the glucose concentration present in the fluid.

$$\text{H}\_2\text{O}\_2 \rightarrow \text{O}\_2 + 2\text{H}^+ + 2\text{e}^- \tag{3}$$

The aim of the study was to obtain a simple enzymatic glucose sensor with a range suitable for sweat glucose sensing (see Table 1) to be integrated in a microfluidics in order to obtain a wearable device with an efficient sweat collection. The sweat sensors analyzed in Table 2 usually show the sweat collecting system being fabricated on top of the functionalized electrode. Here, a fabrication process for microfluidic systems was adapted to suit a PS nanoimprinting process that allows for a versatile, scalable and cost-effective fabrication in the context for wearables as well as point-of-care testing. A combination of a nanoimprinting lithography of a microfluidic on a PS substrate followed by electrode definition by metal evaporation enables a route for a mass production of wearable sweat sensors preserving the mechanical and electrical integrity of the electrodes. This approach can be worthwhile for envisioning a fast roll-to-roll production of non-invasive wearable sensing systems for sweat and other biological fluids.

**Table 2.** Comparison of present work to other amperometric glucose biosensors.


#### **2. Materials and Methods**

#### *2.1. Materials and Reagents*

For the electrode fabrication, polystyrene (PS) foil with a thickness of 0.19 mm was purchased from GoodFellow (Prodotti, Gianni S.r.l., Milan, Italy) and adhesive foil sheets were obtained from Greiner Bio-one (platesealer EASYsealTM transparent, RS Components S.r.l., Milan, Italy). Isopropyl alcohol (IPA), L-cysteine (BioUltra, ≥98.5% (RT)), phosphate

buffered saline tablets (PBS), Bovine serum albumin (BSA, lyophilized powder, ≥96%), glycerol (≥99%), glutaraldehyde (GTA, Grade I, 70% in H2O), glucose oxidase (GOx, from *Aspergillus niger*, Type X-S, lyophilized powder, 100,000–250,000 units/g), D-(+)-Glucose (≥99.5%) and potassium hexacyanoferrate(II) trihydrate (98.5–102.0%), were provided by Sigma Aldrich (Merck Life Science S.r.l, Milan, Italy). Deionized (DI) water was taken from a Milli-Q® water system (Millipore).

For fabrication of the microfluidics, the following material was used: Si wafer, polyester film photomasks (JD Photo Data, Hitchin, UK), SU-8 2002 and SU-8 2100 photoresist, SU-8 developer (MicroChem Corp, Newton, MA, USA), UV-glue (NOA 68, Norland Products Inc, Cranbury, NJ, USA).

#### *2.2. Electrode Fabrication*

For the experiments to characterize the properties of the working electrodes (WE), these were fabricated as single electrodes on PS foil, and measurements were taken in a electrochemical cell with a separate silver/silver-chloride (Ag/AgCl) reference electrode and a separate platinum (Pt) counter electrode; whereas for the first experiments for integration with microfluidics, the WE was combined with a counter electrode (CE) and a reference electrode (RE) on the same PS foil substrate. In all three electrodes, a 5 × 8 mm contact pad and a 10 × 0.5 mm wire connection were present, leading to the electrode surface in contact with the fluid. The circular shaped WE had a diameter of 4 mm, and, in the three-electrode system on PS foil, it was centered between the CE and RE; both CE and RE were bent in half-circular shape surrounding the WE in order to allow for the CE and RE to be close to the WE, exploiting a large surface area. The PS foil was cleaned using IPA and DI water and then dried using nitrogen flux. Using a laser cutter (VLS2.30DT, Universal Laser Systems GmbH, Wien, Austria), the design was cut into adhesive foil sheets that were attached on the PS foil to serve as mask in the following metal evaporation process. First, a thin adhesion layer of chromium (about 15 nm) was thermally evaporated on the PS film followed by a gold (Au) layer (about 150 nm). Then, the adhesive mask was carefully removed and excess PS film was cut. For the RE, Ag (about 100 nm) was thermally evaporated using a new mask. The maximum deposition rates were up to 1 Å/s.

#### *2.3. Electrode Functionalization with GOx*

Prior to the functionalization of the WE electrode, the samples were washed with IPA and DI water. Then, the electrode was covered with 50 mM L-cysteine solution for 20 h at room temperature (RT) to create thiol–gold bonds, followed by washing with DI water and BSA solution (30 mg/mL BSA in PBS). After that, 30 µL drops of GTA solution (2.5 wt% GTA, 50 mg/mL BSA and 1 vol% glycerol) were applied on the electrode surface for 2–3 h at RT to immobilize GOx via cross-linking. BSA and glycerol contributed to the stabilization. The electrodes were then washed with BSA solution to which glycerol was added (30 mg/mL BSA and 1 vol% glycerol in PBS). Next, 30 µL drops of GOx solution (25 mg/mL GOx in PBS) were placed on the electrode for 2.5 h at 4 ◦C. Finally, electrodes were washed with PBS solution and stored at 4 ◦C in PBS.

#### *2.4. Electrode Characterization*

For the electrochemical characterization of the WE electrode, experiments were conducted in an electrochemical cell with external RE (Ag/AgCl in KCl) and external CE (Pt sheet) using a potentiostat (Autolab, Metrohm Autolab, The Netherlands) and the software NOVA (Metrohm Autolab). Cyclic voltammetry (CV) was performed on Au electrodes without functionalization from −0.2 V to 0.6 V at different scan rates (10, 20, 40, 50, 60, 80, 100, 140, 180, 200 mV/s) in [Fe(CN)6]3−/4<sup>−</sup> (ferro-ferricyanide) solution for confirmation of the response of the electrode. Further, Au electrodes were tested for their current response in different concentrations of H2O<sup>2</sup> (0–25 mM) at 0.7 V. Chronoamperometry (CA) was performed by placing the GOx-functionalized WE, the Ag/AgCl RE and the Pt CE in the electrochemical cell filled with 20 mL of PBS. After a stable baseline current was reached,

glucose stock solution (2 M in PBS, prepared the previous day to allow for mutarotation of the glucose) was added stepwise to the PBS solution until a maximum glucose concentration of 25 mM was reached. The applied potentials were 0.65 V and 0.7 V at a constant pH of 7.4, which is the standard physiological buffer. To test for a possible influence of the pH, the current response was measured for pH values between pH 4.5 to pH 8 in glucose solution (pH adjusted in 1 mM glucose in PBS) and at potentials ranging from 0.5 V to 0.75 V. Sensitivity was calculated as the slope for the linear range divided by the circular electrode area, and the limit of detection (LOD) was calculated as three times the standard deviation of the baseline current divided by the slope [37]. The apparent Michealis–Menten constant Km(app) was calculated using the software OriginPro 2018 (OriginLAB, USA) following the Lineweaver–Burk formula:

$$1/\text{I}\_{\text{SS}} = 1/\text{I}\_{\text{max}} + (\text{Km(app)})/\text{I}\_{\text{max}} \times 1/\text{c},\tag{4}$$

where c represents glucose concentration, ISS is the steady-state current at a certain glucose concentration and Imax describes the maximum current under saturated conditions.

#### *2.5. Microfluidics Fabrication for Further Device Integration*

A nanoimprinting approach was used to fabricate microfluidics in PS in order to be integrated with the electrodes. Photomasks for the fabrication of a SU-8 stamp were designed using the software CleWin (WieWeb software, Hengelo, The Netherlands) and printed on photomask foil. To fabricate the stamp, a photolithography process was exploited: After oxygen plasma treatment for surface activation (100 W for 5 min), a 2 µm layer of SU-8 2002 photoresist was spin coated on a clean 2" Si-wafer substrate after the following protocol: 500 rpm for 5 s, then 1800 rpm for 30 s; followed by a soft bake at 95 ◦C for 1 min 30 s; cool down period of 10 min; flood exposure under UV light (1 min 30 s at about 6 mW/cm<sup>2</sup> at 365 nm wavelength); post-exposure bake at 95 ◦C for 1 min 30 s; development for 1 min. The purpose of this thin SU-8 layer was to improve the attachment of the following thicker SU-8 layer during the imprinting process. In a second step, a layer of 110 µm of SU-8 2100 was spin coated: 1200 rpm for 90 s; soft bake for 5 min at 65 ◦C, then 45 min at 95 ◦C; UV exposition with 250 mJ/cm<sup>2</sup> using the photomask; post-exposure bake for 5 min at 65 ◦C, then 30 min at 95 ◦C; development 5 min; hard bake for 10 min at 150 ◦C. The obtained SU-8 structure is the negative of the desired microfluidic pattern to be transferred to the polymer substrate. The PS foil was rinsed with DI water and dried using nitrogen flux prior to the nanoimprinting process. The SU-8 stamp and the PS foil were stacked, and the pattern was imprinted at a temperature of 140 ◦C for 300 s using the nanoimprinting instrument (EITRE 3, Obducat). The channel height was measured using a profilometer (Bruker Dektat XT) resulting in a final value of 105 µm. At this stage, the electrode definition is conducted by metal evaporation with the same procedure as described in Section 2.2. The PS sample was then exposed to oxygen plasma at 200 W for 10 min (RFG 300, Diener) to hydrophilize the surface. A closed microfluidic system was obtained by applying UV glue at the borders of the microfluidics and irradiated for 1 to 1 min 30 s.

#### **3. Results**

#### *3.1. Electrochemical Characterization*

CV was performed on the Au electrodes in ferro-ferricyanide solution to characterize the current response to evaluate the electrode fabrication process for its suitability and reliability. Figure 1a shows an exemplary CV with both oxidation and reduction peak demonstrating the reversibility of the redox reaction. In addition, Figure 1b shows a plot of the square root of the scan rate against the maximum oxidation and reduction currents. The second plot showed increasing absolute current values for higher scan rates, which was expected for Au electrodes in ferro-ferricyanide solution. The characterization showed a good repeatability among several samples; thus, the fabricated Au electrodes were suitable for further functionalization steps.

**Figure 1.** Cyclic voltammetry of a Au electrode on polystyrene (PS) foil in ferro-ferricyanide solution. (**a**) Exemplary cyclic voltammogram. (**b**) Square root of the scan rate vs. maximum absolute values of reduction and oxidation currents (*n* = 6). *n* = number of analyzed samples.

The ability of the Au electrode to detect current changes in different H2O<sup>2</sup> concentrations was demonstrated as the measured current increased with increasing H2O<sup>2</sup> concentrations (Figure 2). This is of importance as the sensor working principle relies on a current response to the H2O<sup>2</sup> oxidation process that occurs as a result of the enzymatic glucose catalysis. As the chosen potential of 0.7 V vs. Ag/AgCl resulted in this reliable current response, further experiments were conducted at this potential.

**Figure 2.** Exemplary current response of a Au electrode in different H2O<sup>2</sup> concentrations. The trend showed an increasing current response with increasing H2O<sup>2</sup> concentrations.

Glucose sensing was achieved by the functionalization of the Au electrode with GOx, where thiol–gold bonds, thanks to the cysteine self-assembled monolayer (SAM layer), acted as a link between the enzyme and the Au surface, and the GTA/BSA/glycerol network enhanced immobilization and stability (Figure 3). Chronoamperometric measurements were conducted at 0.65 V and 0.7 V to demonstrate suitability for glucose sensing. In addition to an applied potential of 0.7 V, a slightly lower value of 0.65 V was chosen to test for the sensor performance, as lower potentials can be advantageous to reduce interference with other species [38]. A linear increase in the current was observed between concentrations of 0.025 mM and 2 mM at 0.7 V, whereas the whole operational range was observed to be between 0.025 mM and 25 mM of glucose for both 0.65 V and 0.7 V, as saturation occurred at concentrations higher than 25 mM (see Figure 3). Comparing the current between an applied potential of 0.65 V and 0.7 V, it is notable that the overall current is higher for 0.7 V. The sensitivity was calculated as 1.76 µA/mM/cm<sup>2</sup> at 0.7 V. The LOD was calculated to be 0.055 mM at 0.7 V. The linear range obtained at a potential of 0.7 V made the higher potential seem more favorable. However, it has to be noted that, at a higher potential, the baseline current was higher and showed higher noise levels, and a 5–10 min longer period of time was required at 0.7 V potential before a stable baseline current was

obtained. The apparent Michaelis–Menten constant was computed as 0.445 mM for 0.65 V and 3.34 mM for 0.7 V, which showed that GOx had a higher substrate affinity at the lower potential. However, the lower substrate affinity at 0.7 V showed better suitability for the determination of glucose concentrations typically present in sweat and other body fluids. As shown in Figure 3b, at 0.7 V the Michaelis–Menten curve shows a saturation at higher concentrations which implicates that, even at higher glucose concentrations, the enzyme activity is not limiting the reaction rate.

**Figure 3.** (**a**) Scheme of sensing mechanism of the glucose oxidase (GOx)-functionalized Au electrode. (**b**) Exemplary calibration plots of the GOx-functionalized Au electrode in different glucose concentrations for (**c**) 0.65 V (*n* = 3) and (**d**) 0.7 V (*n* = 3). A linear increasing current response for increasing glucose concentrations was found between 0.025 mM and 2 mM at 0.7 V (inset), whereas the operational range was up to 25 mM at both potentials before saturation occurred. *n* = number of analyzed samples.

To investigate the relation between the pH of the analyzed fluid and the response of the electrode, the current was measured at pH values between 4.5 and 8 for potentials between 0.5 V and 0.75 V (Figure 4). At a constant glucose concentration of 1 mM, which was chosen because this concentration is well within the linear range, the current was found to increase with increasing pH values despite the standard deviations overlapping in adjacent pH values mainly between pH 5 and pH 6. Moreover, an overall trend was found, which showed increasing current with increasing potential.

**Figure 4.** Current response of GOx functionalized Au electrode at different pH values (*n* = 3). *n* = number of analyzed samples.

#### *3.2. Microfluidics for Three-Electrode System*

The nanolithography and nanoimprinting process used for the fabrication of a simple microfluidic system with inlet, outlet and a circular chamber is described in Figure 5a. Following the design scheme of the three-electrode system and the microchannel (Figure 5b), the Au WE and CE and the Ag RE were evaporated on flexible PS foil (Figure 5c) and a closed microfluidic system was tested for the fluid dispersal using dyed water, which showed uniform fluid flow over all electrodes (Figure 5d).

**Figure 5.** (**a**) Microfluidic fabrication process using nanoimprinting. (**b**) Scheme of the three-electrode system and a simple microfluidic channel. (**c**) Reference electrode (RE), working electrode (WE) and counter electrode (CE) evaporated on PS foil. (**d**) Application of dyed water to validate the filling of the chamber.

#### **4. Discussion**

A first generation amperometric biosensor for glucose was realized by immobilizing GOx on an Au electrode that was evaporated on flexible PS polymer foil. The enzyme was attached to the Au surface via thiol-bonds created by a cysteine SAM layer, and embedded into the cross-linker GTA. BSA and glycerol were added for stabilization purposes.

The response to different glucose concentrations was evaluated by chronoamperometric measurements, and a good sensor output for the biologically relevant range in body fluids such as blood, interstitial fluid, sweat and saliva was demonstrated. Comparing the sensor performance at 0.65 V and at 0.7 V showed that an LOD of 0.055 mM glucose and sensitivity of 1.76 µA/mM/cm<sup>2</sup> were obtained for the higher potential, whereas, at 0.65 V, no linear range was shown. A higher substrate affinity was found at 0.65 V given by the apparent Michaelis–Menten constant of 0.445 mM compared with 3.34 mM at 0.7 V. The low Km(app) value at 0.65 V compared with a similar sensor fabricated with GTA for cross-linking (Km(app) of 1.15 mM [35]), could be due to the stabilizing effect of BSA and glycerol. In general, the obtained results for the sensor are in line with the previous work based on a GOx sensor; however, a comparatively large operational range was obtained (Table 2), also possibly due to the addition of BSA and glycerol as stabilizing agents [39]. To reduce possible interference effects in body fluids, an electrostatically charged and porous membrane such as a nafion layer could be added [40]. Further testing and assessing the optimal potential can help to find the best trade-off between the sensor characteristics depending on the requirements for the field of application.

The linear range of 0.025 mM to 2 mM and the large operational range of 0.025 mM to 25 mM make the sensor suitable for sensing glucose concentrations in a variety of body fluids (see Table 1). A possible field of application of the sensor is its use as a disposable strip for point-of-care measurements to detect the glucose concentration, which is to be used by medical professionals. In this context, the whole three-electrode system can be placed on the same strip, so the presented electrode fabrication process demonstrated the first steps for further integration. Wearable device market applications could be envisioned, especially sweat sensors that allow for non-invasive glucose monitoring.

Several adaptations should be considered for a reliable sweat sensor. In fact, it is known that sweat pH can vary between 4.5 and 7 [41], and the results showed that the current varies depending on the pH. Therefore, it could be of advantage to integrate a pH sensor to achieve a more complete calibration. In addition, it is noteworthy that a response maximum was expected at a slightly acidic pH, as the pH optimum of most GOx is between a pH of 5 and 6 [42]. The different behavior of GOx on the sensor surface could be due to nonspecific modifications of the enzyme surface during the fabrication process [43], and further work is necessary to gain full understanding. Another important adaptation regards the RE: When fabricating a reliable RE for sweat, the presence of chloride ions needs to be considered. In case of the Ag/AgCl RE, an additional layer is required to avoid sensing the presence of chloride ions. Such an electrode can be realized by chemically converting the evaporated Ag electrode into an Ag/AgCl electrode and subsequently adding a layer of polyvinyl butyral [44]. As secreted sweat volumes range between 0.1 and 2 µL/min/cm<sup>2</sup> [12], the miniaturization of the electrodes will be of advantage to collect and drive such small amounts of fluid. On that account, the presented fabrication technique for the microfluidics is highly versatile and allows to easily adapt to smaller structure sizes. Even other thermoplastic materials can be used for the nanoimprinting process, among these, more flexible polymers such as soft thermoplastic elastomers and specifically styrenic block copolymers that can adapt well to the human skin because their Young's modulus is more similar to that of skin [45]. Long-term-stability and testing of real body fluid samples will be of importance during the development of a wearable sweat sensing device that allows for continuous glucose measurements and, by this, paving the way for a new generation of non-invasive glucose sensors improving the quality of life of diabetic patients.

**Author Contributions:** Conceptualization, A.M., F.L.M., V.B., F.R. and M.D.V.; methodology, A.M., F.L.M. and V.B.; formal analysis, A.M. and F.L.M.; investigation, A.M. and F.L.M.; data curation, A.M. and F.L.M.; writing—original draft preparation, A.M.; writing—review and editing, A.M., F.L.M., V.B., F.R. and M.D.V.; visualization, A.M. and F.L.M.; supervision, F.R. and M.D.V.; All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

