Automated microfluidic cell culture of stem cell derived dopaminergic neurons

Parkinson’s disease is a slowly progressive neurodegenerative disease characterised by dysfunction and death of selectively vulnerable midbrain dopaminergic neurons and the development of human in vitro cellular models of the disease is a major challenge in Parkinson’s disease research. We constructed an automated cell culture platform optimised for long-term maintenance and monitoring of different cells in three dimensional microfluidic cell culture devices. The system can be flexibly adapted to various experimental protocols and features time-lapse imaging microscopy for quality control and electrophysiology monitoring to assess cellular activity. Using this system, we continuously monitored the differentiation of Parkinson’s disease patient derived human neuroepithelial stem cells into midbrain specific dopaminergic neurons. Calcium imaging confirmed the electrophysiological activity of differentiated neurons and immunostaining confirmed the efficiency of the differentiation protocol. This system is the first example of an automated Organ-on-a-Chip culture and has the potential to enable a versatile array of in vitro experiments for patient-specific disease modelling.

Hardware selection. In brief, in the Pelican, a four-axis (X-Y-Zθ ) gantry robot is used in combination with a gripper, that allows many types of substrates to be handled and services all devices with substrate from above. The robot was attached to the top of the stainless steel frame support inside the housing through rails that allow the movement of the robotic arm along three axes. This maximised the modular capacity of the Pelican because the rails determine the robotically useful space of the system, in contrast to other automated systems with a fixed rotating robot arm, which often has limited reach. The useful space is the space that is available to potentially hold new devices, adding to the functions of the automated platform.
In the Pelican, a ZEUS pipetting module (Hamilton Inc.) with disposable tips combines the precision of contact dispensing with the versatility of non-contact dispensing. In addition, a liquid dispenser with only non-contact dispensing was also implemented in the system. The liquid dispenser is less precise than the liquid-handler. However, it is much faster as it can handle up to 96 wells per step compared to 4 wells per step for the pipetting modules. Despite its shortcomings, the contact dispensing function of the ZEUS is especially useful for microfluidic cell culture where very small volumes must be dispensed. In addition, the contact dispensing helps to make sure that the dispensed media is bubble free. This is very important as with low flow rate non-contact dispensing, bubbles that arise can imped the flow of fresh media, which could ultimately starve the cells in a chip. Contact dispensing is very precise for dispensing small volumes. However, this precision is dependent on dispensing at an exact location, which is not always possible as the dispensing tip cannot always physically access the well to make contact with the liquid 1 .
With contact dispensing, well cross-contamination is a risk as there is direct contact between the tip and the liquid in the destination well. Therefore, a cleaning protocol was implemented after each dispensing step. This does not promote speed and high-throughput capabilities so sought after in laboratory automation, however, the contact dispensing was only utilised for the initial loading of media to avoid the introduction of bubbles in the dry medium lane. Non-contact dispensing does not require any contact between the dispensing tip and the liquid. This helps to avoid cross-contamination, and promotes the integrity of the well. Non-contact dispensing is very popular in laboratory automation because it is versatile, and it is easy to dispense to any area of a well regardless of geometry such as undercuts, so long as there is an opening on the well 36 . As a result, the non-contact dispensing methodology was utilised for all subsequent media changes. In brief, LACS is a programming environment for laboratory automation. It incorporates three software packages: LacsDriverCore, LACS graphical user interface (GUI) and LACS Config Editor. LacsDriverCore reads the SiLA drivers of the devices generated by the DDK, hence, it connects the devices to a computer. The LACS GUI is the process management software, the interface between a device and an operator (via the computer) through a loaded SiLA driver. The running of all existing protocols through LACS GUI merely requires the identification of the required protocol and the input of the required plate name by the operator. LACS Config Editor is used to draft automated protocols. LACS does not make any process-specific decisions, nor does it analyse any data. In the occurrence of an unexpected event (a device failure or any other error), the operator is always prompted to assess and rectify the error or the event.
The safety features around the housing, the hotels and the incubator of the Pelican (sensors to read the states of the components), and the operations of the de-lidder and the stage incubator were all controlled through a digital and analogue logic module (UR20-FBC-MOD 1334930000, Weidmuller GmbH & Co. KG, Germany). A digital and analogue logic device has a binary set; a binary input and binary output. It is usually used for a maintenance device or to control a device with a simple binary command such as for a valve ON/OFF and a light switch. A single weidmuller SiLA driver was installed to control all devices connected to one logic module.
Software (Labware Manager, Fraunhofer IPA) handles all positions inside the Pelican plant and all substrates. In the first case it holds the information, which position is occupied and with which substrate and which free position could potentially hold which type of substrate. The software knows each position and status as well as each substrate in the plant including position and type. All data are stored in an open source PostgreSQL database (PostgreSQL, https://www.postgresql.org/). Like the hardware devices, this virtual device has a SiLA communication interface and a Windows 7 graphical user interface to view the stored data.
The integration of new devices requires two or three steps depending on the complexity of the device and the availability of compatible drivers. The first step requires the generation of a driver for the new devices. If a SiLA driver is already available through the device manufacturer, open source, or if a commercial software that can be referenced in Microsoft Visual basic is available, then one can directly use the existing driver and move to the second step. Otherwise, one has to use the Driver Development Kit to generate a SiLA compatible driver. The DDK already possesses the framework to easily write the driver for most laboratory devices such as a camera, a microscope or a liquid dispenser. One needs only to add the commands provided by the manufacturer of the device. The second step is completed by assigning the new device a SiLA IP/Port. Once the SiLA driver of the new device is written and tested, the final step is to integrate it into the Pelican. This is done, by opening the driver in LACS Configuration Editor and saving the device in the list of working devices in the Pelican.

Microfluidic device. A 2-lane
OrganoPlate (#9603-200B, Mimetas BV, Leiden, The Netherlands) consists of a stratified array of 96 microfluidic chips embedded in a customised 384-well microtiter plate format 28 (Fig. 2). Each chip consists of a single microfluidic chip contained between two pieces of glass: a top plate with holes corresponding to the underside of selected wells, and a bottom plate. Each chip is connected through 4 neighbouring wells and 2 lanes: one gel inlet well for loading of gel-embedded cells into the culture lane, one medium inlet well connected to one medium outlet well through a medium lane. The flow of media is driven by a pressure drop between the aforementioned 2 wells. The fourth well is used as an observation window for monitoring the quality of cells through an inverted microscope. The culture and medium lanes are separated by a phaseguide, preventing the gel-embedded cells from flowing into the medium lane. A phaseguide is a patterned pinning barrier that controls the liquid-air interface by forcing it to align with the ridge, hence, guiding the fluid flow into the appropriate lane 27 .
Automating the differentiation of human neuroepithelial stem cell into dopaminergic neurons. In order to biologically validate the automated system, and the ability of the Pelican to differentiate cells in microfluidic device, the assay developed by Reinhardt et al. 20 and adapted into the OrganoPlate by Lucumi et al. 35 was used to develop an automated pipeline to differentiate hNESCs into three dimensional networks of electrophysiologically active dopaminergic neurons in the OrganoPlate. The first step of the assay involve loading 0.8 μl of matrigel/hNESCs mixture into the culture lane.
The day after loading the gel-embedded hNESCs into the culture lane of the OrganoPlate, a semi-automated image acquisition protocol was executed through LACS to qualitatively assess the health of the cell culture. An appropriate field of view was selected manually and focussed. The subsequent automated image acquisition protocol ( Fig. 3) consisted of first setting the environmental condition of the onstage incubator to 5% CO 2 and to a temperature of 37 °C. Second, the robotic arm moved the plate from the storage incubator to the microscope. Third, the microscope scanned through all the observation windows of the OrganoPlate, and the camera took an image of each observation window. Fourth, the plate was transported back to the incubator by the robotic arm. Immediately after the automated image acquisition, the automated differentiation of hNESC into dopaminergic neurons protocol (Fig. 4)  to move the plate from the storage incubator to the de-lidder to remove the lid from the plate, before moving it to the liquid dispenser. Second, the dispenser aspirated media from each medium inlet and outlet well. Third, the dispenser replenished the media in each medium inlet and outlet well respectively. Fourth, the robotic arm moved the plate to the de-lidder to put the lid back before placing the plate inside the storage incubator (Supplementary Video 1). These two protocols were run in this order every two days for 24 days.
Human neuroepithelial stem cell culture differentiation. We utilised the Pelican and the existing protocol described by Reinhardt et al. and Lucumi et al. 20,35 to fully automate the differentiation of hNESC into dopaminergic neurons inside a stratified three-dimensional microfluidic device. After manually seeding hNESC in an OrganoPlate, a microfluidic cell culture device with 3D capabilities compatible with laboratory automation, the hNESC were distributed in three dimensions within the culture lane, but also adjacent to the meniscus (Fig. 5A). Thereafter the Pelican executed an automated differentiation protocol to start and maintain the differentiation process. Initially. this required a liquid dispenser to aspirate the maintenance medium and dispense the differentiation medium with PMA, via a dispensing cassette, into inlet and outlet wells of the OrganoPlate. Thereafter, it required incubation and regular media replenishment.  During the differentiation, cells started to form aggregates, and morphological changes started to appear, such as acquisition of cellular polarity and projection of processes representative of neuronal morphology (Fig. 5B). An automated protocol replaced the medium with PMA with medium without PMA, and differentiated neurons started a maturation process accompanied by acquisition of a more evident neuronal morphology (Fig. 5C). Differentiated neurons projecting their processes can be observed in the culture lane and in the area occupied by the meniscus, located in part of the medium lane. Neuronal activity of differentiated cells was tested using calcium imaging (Fig. 5D) and morphological and phenotypic characteristics like neuronal processes positive for TUBβIII (green) and the amount of tyrosine hydroxylase positive neurons (red) were characterised using an immunostaining assay (Fig. 5E). A top view of the culture chamber of a chip with hNESC (Fig. 6A), shows that cells were distributed in the entire culture chamber, Hoechst nuclear staining (blue). Furthermore, their processes have been projected in all directions of the culture lane, as well as on the meniscus part in the medium lane. It can also be seen that the neuronal networks and the level of connectivity of the differentiated neurons is homogeneous in the entire culture chamber, denoting an even effect of the differentiation protocol, as well as the efficiency of perfusion in the medium lane (Fig. 6A). Figure 6B enlarges the view of a specific area of one chip with differentiated neurons from hNESC. Neurons positive for TUBβIII (green) and positive for tyrosine hydroxylase (red) are located in both culture and medium lanes. However, all cells appear to migrate towards the medium lane. This is confirmed in the front view of the enlarged area (Fig. 6C), where the extent of the area occupied by the meniscus in the medium lane is clear, in addition to the degree of cells located on the meniscus on the medium lane. On average, the efficiency of differentiation for tyrosine hydroxylase positive neurons in 3 chips of the OrganoPlate with hNESC differentiated neurons was 15% of all neurons (Fig. 6D), which is in accordance with values reported previously in analogous manual, microfluidic and macroscopic cell culture systems 20,35 .
After 24 days, we were able to obtain midbrain-like, mature dopaminergic neurons in 96 three-dimensional self-contained microfluidic chips. Regular pictures were taken during differentiation for quality control and morphological study (Fig. 5A-C), and end point assays are represented in Fig. 5D,E to illustrate the fate of the differentiation.

Calcium imaging and immunostaining assays. To probe the neuronal activity of cells cultured in an
OrganoPlate within the Pelican, we used Fluo-4-based calcium imaging. We acquired time-series of representative culture chambers of WT and PINK1 p.I368N-mutated populations at day 24 of differentiation (Supplementary Video 2). Analysis of those calcium imaging data revealed spontaneous neuronal activity in differentiated control and PINK1-mutant neurons in an OrganoPlate, within the Pelican. We detected individual cells by applying an automated cell segmentation algorithm 37 to the raw calcium imaging data. Figure 6E,G illustrate segmented mean fluorescence frames of representative culture chambers of an OrganoPlate with WT and PINK1-mutated neurons respectively. Fluorescence traces were then measured for each segmented cell to assess their activity (Fig. 6F,H). Some of the fluorescence traces reveal calcium transients indicating neuronal firing events (Fig. 6F e.g. signal #2, 3, 9 and Fig. 6H e.g. signal #8, 14,27). Fluorescence traces revealed different firing patterns of the differentiated neurons. Some of these traces exhibited regular firing patterns, as opposed to other ones, corresponding probably to dopaminergic neurons, similar to what has been previously reported by Lucumi et al. 35 in adjacent manual culture.

Discussion
We assembled an automated cell culture observatory, termed the Pelican. It was optimised for the long-term cell culture maintenance of neurons inside three-dimensional microfluidic devices. The implementation followed a modular automation design, with generous space for further devices, and is flexible in terms of the automation platforms that can be implemented. For example, the system allows the automation of seeding, feeding and other cell culture processes that could initiate and maintain cell lines in any standard microtiter plate format. In this regard, we carefully selected the robotic system, the pipetting technologies and the techniques to integrate all the components of the system into a single plant. Many methods have been developed over the years to address the issue of standardisation and easy integration of new devices into existing laboratory automated plants 8,9,[38][39][40][41] . In the Pelican, Standardisation in Laboratory Automation (SiLA) was chosen for the integration of the devices. SiLA standardises the interfacing, integration and data representation in a simple single XML schema for all devices 9,42 . SiLA standardises the communication between process management software and one or more devices. SiLA defines common commands per device class and the common device states. SiLA defines the device classes to achieve these functionalities such as an incubator and a liquid-handler. Each device class will have the required commands for the core functionalities as well as optional commands for extended functionalities that are not necessarily present in every device of the same class. In this regard, a SiLA standard common command dictionary was developed for every single device class where the commands such as setParameter and getParameter and the expected return are known. This allows a process management software to automatically generate the required commands for every device class. Each component of the Pelican was chosen based on cell culture needs without taking into account the manufacturer of each device. The integration of any new device would require sufficient space within the useable area and possibly the development of a SiLA driver (tutorial in user manual available upon request). The modular hardware and software integration flexibility is a key advantage of the Pelican design compared to other automated cell culture systems [43][44][45][46][47][48] .
In order to demonstrate the biological utility of the Pelican, control and PINK1-mutant human neuroepithelial stem cell lines were automatically differentiated into midbrain specific dopaminergic neurons. Calcium imaging of spontaneously firing neurons, as well as immunostaining for neuronal markers demonstrate that, neuroepithelial stem cells could be successfully maintained and were spontaneously active within the OrganoPlate inside the Pelican. Further analysis of fluorescence traces for additional cell lines with different genetic backgrounds would be necessary to quantify any difference in phenotypic characteristics of neuronal activity between control and PINK1 mutant neurons. The differentiated neurons can be maintained inside the OrganoPlate for at least 100 days. The Pelican demonstrates proof-of-concept for automated generation of personalised in vitro neuronal models from human neuroepithelial stem cells via a microfluidic cell culture approach.
The Pelican is designed for longitudinal analysis of many personalised cellular models exposed to a few perturbations, rather than single, end-point analysis of one cellular model exposed to a large number of perturbations, as for instance is the focus in high throughput drug screening. Therefore, we envisage that such automated system be applied to stratification of patients with complex diseases. In addition, the Pelican can be used of automated cell culture to enable comprehensive phenotyping of large, parallel sets of personalised, in vitro, midbrain-specific, dopaminergic neuronal models of Parkinson's disease. By integrating the data generated with a generic mechanistic computational model of the underlying biochemical network of a dopaminergic neuron, each personalised computational model then becomes a coherent representation of our information about the cell autonomous characteristics of Parkinson's disease in each patient. Such personalised computational models, and thereby the corresponding patients, are then amenable to stratification with a range of powerful stratification tools. Stratification based on personalised computational models of data is statistically superior to stratification based on the personalised data alone, as the former explicitly incorporates the wealth of prior biochemical information known about midbrain dopaminergic neurons. Ultimately, this will accelerate the translation of basic biomedical knowledge from the laboratory to the therapies with clinical impact.
Increasing demands for reproducibility, parallelisation and longitudinal observations are driving cell culture research toward automation. We developed a novel automated cell culture observatory that enables long-term maintenance and longitudinal optical measurement of cellular parameters in Organ-on-a-Chip platforms. We demonstrate the use of this platform to successfully automate the generation of personalised in vitro neuronal models from human neuroepithelial stem cells. We demonstrate the feasibility of semi-automated image acquisition on this platform and compatibility with different real-time and end-point assays. It is the first time that an Organ-on-a-Chip platform is applied in an automated setting. It holds great promise for patient stratification by enabling comprehensive phenotyping of large, parallel sets of personalised, in vitro, models of complex diseases.

Methods
System Construction and Design. The Pelican is composed of a sterile automation enclosure that abuts a sterile manual enclosure on one side and an incubator on another. The automation enclosure contains a set of devices that may physically communicate via a four-axis gantry robot within a customised housing support. The manual enclosure is a cell culture hood, adapted for restricted communication of material with the automation enclosure. The automation enclosure currently includes a de-lidder, eight-fold and 96fold parallel dispenser, three-axis fourfold liquid handling robot (pipettor) with disposable tips, confocal microscope, and camera. The assembly (Fig. 7a,b), all of the components are described in the Supplementary Experimental Procedures.

Cell culture. All work with human iPSCs and thereof derived cells has been approved by the Ethics Review
Panel (ERP) of the University Luxembourg as well as by the Luxembourgish Comité National d'Ethique de Recherche (CNER). The CNER reference number is 201305/04. All experiments were performed in accordance with relevant guidelines and regulations. We confirm that written informed consent was obtained, by our cell line suppliers, for the establishment of stem cell lines, from all donors.
Human neuroepithelial stem cell culture. In order to demonstrate the capability of the automated system to maintain and monitor many types of cells including PD specific cell lines, we used a human neuroepithelial stem cell line from a healthy donor (hNESC K7) and a human neuroepithelial stem cell line derived from a patient carrying the Parkinson's disease related mutation p.I368N in PINK1 (40066C5N). These cells were maintained and differentiated into midbrain-specific dopaminergic neurons within an OrganoPlate, by automating an existing macroscopic cell culture protocol 20 , that we previously adapted for microfluidic cell culture 35 . In brief, to culture hNESCs in an OrganoPlate, they were harvested from wells of a 6 well plate. The harvested hNESCs were then re-suspended on Matrigel (catalogue number 354277, lot number 3318549, Discovery Labware, Inc., Two Oak Park, Bedford, MA, USA). 0.7 μL of this Matrigel-cell mix was loaded in assigned chips of the OrganoPlate at a density of 0.03 million cells/μL. After seeding the cells, the plate was loaded into position B3 in the hotel of the Pelican. Afterwards, the plate was moved by the robotic arm to the storage incubator, at 37 °C and 5% CO2.
Dopaminergic neuronal differentiation. The culture medium preparation "N2B27 medium" consisted of mixed equal amounts of Neurobasal medium (invitrogen/life technologies) and DMEM/F12 medium (invitrogen/life technologies) supplemented with 1% penicillin/streptomycin (life technologies), 2 mM L-glutamine (life technologies), 0.5 X B27 supplement without Vitamin A (life technologies) and 0.5 X N2 supplement (life technologies). The medium to maintain the hNESC in culture "maintenance medium" consisted of N2B27 medium with 0.5 μM PMA (Enzo life sciences), 3 μM CHIR (Axon Medchem) and 150 μM Ascorbic Acid (Sigma Aldrich). The differentiation medium formulation to induce the differentiation of hNESCs towards midbrain dopaminergic neurons "differentiation medium with PMA" consisted of N2B27 medium with 200 μM ascorbic acid, 0.01 ng/μL BDNF (Peprotech), 0.01 ng/μL GDNF (Peprotech), 0.001 ng/μL TGFβ3 (Peprotech), 2.5 μM dbcAMP (Sigma Aldrich) and 1 μM PMA. The function of PMA in this medium preparation was to stimulate the sonic hedghog (SHH) pathway in the cultured hNESCs. Differentiation medium with PMA was changed, every 2 days during the first 6 days of culture in the differentiation process. For the maturation of differentiated neurons, PMA was no longer added to the differentiation medium "differentiation medium without PMA" from day 7 onwards, which was changed every 2 days during 3 weeks. To monitor cellular morphology during differentiation, bright field images were acquired automatically in the Pelican using the microscopy station.  37 and fluorescence traces were generated for each segmented cell and presented as relative changes in fluorescence intensity ∆F/F. Immunofluorescence staining assay. Immunostaining for the dopaminergic neuronal markers class 3 beta tubulin (TUB β III) and tyrosine hydroxylase (TH), the penultimate enzyme in the biosynthesis of dopamine 20,49,50 , was performed on representative chips at day 24 of differentiation. Differentiated cells were fixed with 4% paraformaldehyde (PFA) in 1 × phosphate-buffered saline (PBS) for 15 min, by manually adding 70 μL in medium well inlet and 30 μL in medium well outlet followed by permeabilisation with 0.05% Triton-X 100 in 1 × PBS (3 min on ice), and blocking with 10% fetal calf serum (FCS) in 1 × PBS (1 h). After washing with 1 × PBS, the primary antibodies mouse anti-TUBβIII (1:2000, Covance) and rabbit anti-TH (1:2000, Santa cruz biotechnology), were incubated for 90 min at room temperature. After washing with 1 × PBS, the secondary antibodies Alexa Fluor 488 Goat Anti-Mouse and Alexa Fluor 568 Goat Anti-Rabbit together with a stain DNA (Hoechst 33342, Invitrogen), were incubated for 2 hours at room temperature. After washing with 1 × PBS and water, confocal images of representative culture chambers were acquired using a confocal microscope (Zeiss LSM 710).
Automating the differentiation of human neuroepithelial stem cell into dopaminergic neurons. LACS Config Editor was used to develop automated pipelines for the differentiation of hNESCs into dopaminergic neurons and time-lapse imaging microscopy. The automated pipelines were drafted according to the SiLA communication protocol and command format as previously described 9,42,51 . In brief, SiLA uses a Simple Object Access Protocol (SOAP) and a Web Service Description Language (WSDL) documentation, both of which are based on XML. A full library of commands for each device is downloaded once and stored in LACS as a configuration document of the Pelican. LACS Config Editor was used to incorporate the automated pipelines in this configuration file used herein to automate the differentiation of hNESCs into dopaminergic neurons. On the workbench, gel-embedded hNESCs were manually loaded into the culture lanes of a 2-lane OrganoPlate as described above. Then, the plate was put inside the Pelican through Hotel B, and placed inside the storage incubator by the robotic arm. The dispenser was fitted with a 5 μL cassette from Biotek and used as a dispensing medium for the media change.
Statistical Analysis. Three representative chips (n = 3) were selected to illustrate the results of this study, in which the statistical analyses were performed by determination of the mean value and the standard deviation of the proportion of dopaminergic neurons within the overall neuronal population.

Data Availability
All data used within this study are available.