Chassis.

The machine chassis is constructed with standard 20 mm × 20 mm and 20 mm × 40 mm T-slot aluminum extrusions. The chassis is 37 cm deep, 65 cm tall, and 57 cm wide, and serves to support a front panel where most of the fluidic and electronic components are mounted, a top shelf to hold the sample and buffer bottles, and the fraction collector at the bottom. The top shelf is fitted with a polycarbonate tray that provides spill containment for the buffer bottles.

Component panel.

All fluidic components, except for back pressure valves, and all electronic components, except for the touchscreen monitor, are mounted directly to a front panel made of a 5mm thick laser-cut acetal sheet. Acetal was selected for the panel primarily for its strength, good chemical resistance to most reagents commonly used in laboratories (especially to ethanol, isopropanol and bleach, commonly used to clean and disinfect) and for its compatibility with laser cutting. Acrylic is not a good material for the panel because it is brittle and degrades when exposed to ethanol and isopropanol.

Single board computer.

A Raspberry Pi 3B+ was used as a single board computer to run all of the automation in the machine. It was selected due to its low cost, small size, widespread availability of accessories, and sufficient features for this project, including plenty of general-purpose input/output (GPIO) pins for hardware control, integrated WiFi, HDMI, USB and Ethernet ports. All peripherals are controlled by the Raspberry Pi via an I2C bus. However, the I2C bus on the Raspberry Pi operates at 3.3 V, while the I2C bus on the peripherals operates at 5 V. To interface with the peripherals and receive power from the 24 V power supply, a custom circuit board was built and mounted on the general-purpose input/output (GPIO) pins of the Pi (Supplementary Guide). This circuit includes bidirectional logic level shifters for operating the I2C bus at different voltages and a step-down voltage regulator to power the Raspberry Pi and the solenoid valve controllers with 5 V. A detailed schematic of this circuit can be found in the associated build guide. The system is running on Debian 10 installed on a 16 GB SD card.

Power supply.

The power supply takes 120 VAC or 240 VAC (selectable by a switch) as input and outputs 24 V, which is routed to all electronic peripherals (peristaltic pumps, rotary valve, fraction collector, solenoid valve controllers), and to the 5 V step-down voltage regulator on the Raspberry Pi hat. An illuminated main power switch is used to pass the 120/240 VAC input to the power supply and to a power socket for the touch screen. For safety, there is a 2.5 A fuse on the 120/240 VAC input. The 5 V step-down voltage regulator on the custom circuit board mounted to the Raspberry Pi converts the 24 V supply to 5 V, which is then passed to the power pins of the Raspberry Pi, the peripheral side of the bidirectional level shifter, and the IOREF input of the solenoid valve drivers. The 3.3 V output of the Raspberry Pi is passed to the Pi-side of the bidirectional level shifter.

I2C bus.

All electronic peripherals included in this design are capable of communicating over the I2C bus in the Raspberry Pi, which allows up to 128 devices to sit on the same bus as long as they are each given a unique address. This approach both simplifies the required wiring and the process of replacing or adding modules. While the Raspberry Pi 3B+ technically has hardware I2C capabilities, there is an error in how the silicon was designed and it is unstable when used with devices that support clock stretching [9]. There is evidence that suggests this error is still present on newer revisions, including the Raspberry Pi 4 [10]. To work around this issue, the I2C communication protocol was implemented via software with a GPIO overlay. Two arbitrary GPIO pins were assigned to serve as the I2C data and clock lines and were routed to the Raspberry Pi-side bidirectional level shifter.

Solenoid valves.

Each fluidic channel requires three 2/3 way solenoid valves for configuring the flow path. The first valve is used for selecting between buffer and sample, the second is used for purging the pre-column path to waste, and the third is used for directing buffers passing through the column to either the fraction collector or to waste. The valve material was selected to be PEEK in order to be compatible with a broad range of chemicals, buffers, and biological materials that may be run through the system. The solenoid valves are controlled by stackable 8-channel relay driver boards that are connected to the Raspberry Pi via the I2C bus.

Peristaltic pumps.

Each channel has a 6-roller peristaltic pump that is placed between the solenoid valve that controls the sample input and the solenoid valve that purges the pre-column path to waste. A peristaltic pump is ideal in this system because it offers smooth flow with minimal pulsing, a well-defined and controllable flow rate, and a closed fluidic path that prevents buffers from passively leaking by gravity through the system when not in use. The pump comes with an integrated stepper motor that is connected to a motor driver module that is in turn connected to the Raspberry Pi via the I2C bus. When used with the recommended 1 mm inner diameter tubing, the pump can achieve a minimum flow rate of 4×10⁻⁶ mL/min and a maximum flow rate of 21 mL/min. The flow rate range can be adjusted by using tubing with a different inner diameter.

Rotary valve.

The 8-to-1 rotary valve is used to select one of eight possible input buffers to be routed to all channels on the system through a 1-to-4 splitter. The valve material is PTFE, which has high chemical compatibility and low friction. The valve has integrated home and port position sensors and is driven by a NEMA 18 stepper motor. The sensors and the stepper motor are connected to a motor driver module that is connected to the Raspberry Pi via the I2C bus.

Fraction collector.

The output of the column can be redirected to either waste or an automated fraction collector that is composed of a carriage with a holder for fraction containers mounted onto a linear actuator. Given that different sized fractions are desirable for columns of different volumes, the carriage was designed to have an interchangeable fraction holder (top plate) that can be designed to hold many different arrangements of containers. We have designed two holders, with one version holding ten rows of 1.5 mL Eppendorf tubes and the other ten rows of 5 mL Eppendorf tubes. Both also have four rows of 50 mL of Falcon tubes to collect flow through. The linear actuator is attached to the machine chassis and powered by a NEMA 18 stepper motor that is connected to a motor driver module, which is in turn connected to the Raspberry Pi via the I2C bus.

Touchscreen display.

The touchscreen display provides a basic interface for generating and running purification protocols on the machine. The display is connected to the HDMI port on the Raspberry Pi and the touch capabilities are managed via USB. While the touchscreen interface does allow for basic modifications of protocol parameters, more advanced features are only accessible through the Python programmatic interface.

Structure.

The software developed here was written in Python 3 and organized as a client-server model (Fig 3) that abstracts low-level peripheral operation of the machine from the operator and allows for remote control. The server-side software includes various hardware drivers and middleware that manages the peripherals and communication with the client. The client-side software includes middleware for managing communication with the server and high-level instructions that can be translated into operations by the server. While the server-side software is always executed as a service on the Raspberry Pi in the instrument, the client-side software can be run on an ad-hoc basis either remotely or locally through a Python interpreter or a script executed from the command line. We have also designed a graphical user interface (GUI) that makes it very easy to run the system with predefined protocols without any knowledge of Python programming. However, the current GUI design can only be run locally, using the touchscreen of the system. As implemented, only a single client can interface with the server until control is relinquished by the client or a disconnect is forced. This means that when the GUI is running remote clients cannot connect to the machine.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. System software diagram.

The software used by the system follows a client-server model. While the client interface can be generated by a GUI, interactive shell, or a script, the TCP data connections only allow for one client to interface with the system at a time. When there isn’t an active connection, the server will periodically broadcast its status on the UDP port. Once a client has connected, all data will flow to the server’s device interface, which is responsible for parsing commands to be executed and providing system updates to the client. After connecting, the client will select a hardware configuration to be used by the system, which is then forwarded to the hardware controller and subsequently processed by the hardware initializer. Afterwards, additional commands can be routed to the peripherals through the hardware controller.

https://doi.org/10.1371/journal.pone.0297879.g003

If errors occur, system logs can be recorded at different levels of verbosity on both the client and server to debug potential hardware or communication faults. This feature is particularly useful when modifying the system to include new features or different hardware.

Server software.

Each type of peripheral in the instrument is managed through a driver class that defines how low-level commands are issued from the Raspberry Pi to the hardware device. All these driver classes are controlled by a hardware controller class, which facilitates communication with peripherals by defining a subset of operationally significant commands. This driver-controller implementation allows the higher-level functionality of the software to be preserved if alternative peripherals are used. To accommodate alternative hardware peripherals for the existing functionality such as a different stepper motor controller board, one could write a new driver to control the peripheral without modifying the higher-level code. Extending the feature set of the system, however, would require adding additional functionality to the higher-level code.

Given that each machine may have a different set of peripherals or configurations that can be changed between use, such as employing 1 mL or 5 mL resin columns, the connected client states what configuration should be used by the hardware controller prior to operation. Once the configuration is identified, the controller instantiates the hardware classes accordingly and preserves the configuration until the client disconnects or it is manually reset.

To allow for local and remote clients, there is a server communication interface that is perpetually run as a Linux service and performs three functions: 1) it broadcasts server availability to potential clients when not in use, 2) it receives commands from the client that are processed into instructions for the hardware controller, and 3) it reports to the client data resulting from both receiving and completing the command provided. This communication occurs over separate TCP sockets on the Raspberry Pi, which allows for local client operation on the machine and remote operation over a network.

Client software.

The client software includes middleware for managing communication with the server and high-level command wrappers for facilitating human-readable scripts. The communication middleware consists of a set of commands for connecting to and disconnecting from the server, validating server responses, and issuing peripheral commands. While the operator will need to provide the TCP address of the server and identify the name of the hardware configuration to be used, most other operations are fully defined by the command wrappers, which simplifies the scripting process. A GUI is also available for operators who are not comfortable with programming in Python and are not interested in advanced features. The GUI uses a few parameters such as the name of the configuration to be used and the volume of sample to be loaded into the columns in order to automatically generate a script that is run locally on the system. The GUI also allows operators to pause, hold on a step, or terminate the run using buttons.

Evaluation of long-term flow rate stability.

Phosphate Buffered Saline (PBS) was pumped through the buffer and load flow paths through one channel for a duration of 120 minutes and measured by weighing the accumulated output volume on a Mettler Toledo digital balance. The flow was paused every 10 minutes to allow the scale to acquire a stable reading, which was then transmitted to a PC using RSKey software (A&D Company) and an RS-232 cable. This process was repeated with both 1 mL and 5 mL IMAC columns at flow rates of one column volume per minute. Due to the 320 g mass limit of the scale, the accumulated volume in the 5 mL runs was disposed of every 40 minutes to stay within the limits of the scale.

Evaluation of fraction volume consistency.

The flow rates of all buffer flow paths were coarsely calibrated in parallel by flowing PBS through a 1 mL or 5 mL HisTrap excel (Cytiva)Immobilized Metal Affinity Chromatography (IMAC)column on each channel at an approximate flow rate of one column volume per minute. Following calibration, measurements were then taken by dispensing PBS from either buffer or load flow paths into the fraction collector through 4 channels simultaneously for one minute. This process was repeated with both 1 mL and 5 mL IMAC columns at flow rates of one column volume per minute and 30 measurements were taken for each condition.

Generation of stable Expi293 cells for the expression of soluble SARS-CoV-2 Spike (S) and Receptor Binding Domain (RBD).

cDNAs for soluble SARS-CoV-2 S and RBD (containing C-terminal His-tag) [11] were cloned into EcoRV-cut plenti-CMV-Puro-DEST (Addgene, #17452, gift from Eric Campeau & Paul Kaufman) using NEBuilder HiFi DNA Assembly Master Mix (NEB). Lentivirus was produced in HEK293FT by co-transfection of cDNA containing pLenti plasmids together with pCMV-dR8.2 dvpr (Addgene, #8455, gift from Bob Weinberg), pCMV-VSV-G (Addgene, #8454, gift from Bob Weinberg) and pAdVAntage (Promega) using FugeneHD (Promega). Supernatants were collected 48 hr post-transfection and filtered using a 0.45 μm syringe filter. For the transduction of Expi293 cells, five 6-well plates with 2 million cells per well were spin-infected with lentivirus diluted 1:2.5 in fresh Expi293 expression media under the following conditions: 2 hr, 33°C, 1,000 g. Cells were subsequently pooled together and cultured in 30 mL Expi293 expression media on a shaking incubator. 24 h post-transduction puromycin was added at 2 μg/mL and cells were selected for 7 days.

Expression of coronavirus S and RBD, and monoclonal antibodies.

For stable expression of SARS-CoV-2 S and RBD, stable suspension cells were maintained in Expi293 media (Thermo Fisher Scientific) at 37°C, 140 RPM (25 mm shaking diameter), 8% CO₂, and 80% humidity. All suspension cells were cultured in Thomson Optimum Growth Flasks. Flasks were seeded at a density of ~0.3×10⁶ live cells/mL at either 0.8 L scale (1.6 L flask) or 1.2 L scale (2.8 L flask). Cells were then grown for 6 days and harvested at a density of 8×10⁶ to 10×10⁶ live cells/mL at >90% viability. Harvested cell cultures were centrifuged at 4,000 g for 30 min, followed by filtration of the supernatant through a 0.45 μm NalGene Rapid Flow filter unit (Thermo Fisher Scientific). Filtered supernatants were adjusted to pH 7.4 using 1 M NaOH. Expression titers (purified protein per L of culture) for SARS-CoV-2 S and RBD from stable cell lines were ~3 mg/L and ~20 mg/L, respectively.

For transient expression of coronavirus S, RBD, and monoclonal antibodies expression plasmids were transiently transfected into suspension Expi293 cells following protocols as previously described [12]. For antibodies, the amount of plasmid to be transfected was split equally between the corresponding pair of heavy and light chain plasmids. Three days after transfection, cell cultures at a density of 3 to 5×10⁶ live cells/mL and ~50–75% viability were harvested, clarified, and adjusted to pH 7.4 as described above. For monoclonal antibodies, plasmids separately encoding for the heavy chain and light chains were transfected at 200 mL scale (500 mL flask), grown for 5–6 days (viability = ~98%), and harvested, clarified and adjusted to pH 7.4 as described above. Serum (rabbit) was purchased from a commercial vendor (Sigma, #R4505) and filtered through a 0.45 μm syringe filter prior to purification. Expression titers (purified protein per L of culture) for SARS-CoV-2 S and RBD from transiently transfected cells were similar to that of stable cell lines, ~1–3 mg/L and ~20 mg/L, respectively. Nevertheless, the low viability of cells transiently expressing SARS-CoV-2 S resulted in clogging of the filtration units and more time needed to clarify these supernatants.

Parallelized affinity chromatography purification of S, RBD, and antibodies.

For S and RBD, four HisTrap Excel columns (column volume = 1 mL or 5 mL) were equilibrated in 10 column volumes (CV) of 20 mM sodium phosphate, 500 mM NaCl pH 7.4. Following parallelized loading (4 × 50 CV for CV = 1 mL; 4 × 200 CV for CV = 5 mL), each column was washed with 60 CV of (20 mM sodium phosphate, 500 mM NaCl, 20 mM imidazole, pH 7.4) and eluted with (20 mM sodium phosphate, 500 mM NaCl, 500 mM imidazole, pH 7.4) into eight 1 CV fractions. For antibodies, four 1 mL HisTrap Protein A HP (Cytiva) were equilibrated in 10 CV of 20 mM sodium phosphate pH 7.0, loaded in parallel (4 × 10 CV), washed in 30 CV of 20 mM sodium phosphate pH 7.0, and each column was eluted with 0.1 M citric acid pH 3.0 into five 0.85 CV fractions, each fraction containing 0.15 CV of 1 M Tris pH 9.0. All steps were performed at a flow rate calibrated to 1 CV/min at the beginning of the run. Purified protein fractions (8 CV for HisTrap Excel, 5 CV for HisTran Protein A HP) were then buffer exchanged into PBS using either Amicon Ultra-15 centrifugal filter units (Millipore Sigma, 100 kDa cutoff for S, 30 kDa for antibodies, 10 kDa for RBD) or 10DG desalting columns (BioRad). S and RBD proteins, optionally supplemented with 10% glycerol, were filtered and stored at -80°C.

Size exclusion multi-angle scattering (SEC-MALS) analysis of purified S proteins.

Purified S proteins were analyzed on a SRT SEC-1000 column (5 μm, 4.6x300 mm, Sepax) at 0.35 mL/min in PBS running buffer. The SRT SEC-1000 column and SEC-MALS system, which includes a 1260 Infinity II HPLC (Agilent), a miniDAWN TREOS II MALS detector (Wyatt) and a Optilab T-rEX refractive index detector (Wyatt), were equilibrated in PBS at 0.1 mL/min for at least 24 hr prior to analysis. Molecular weights were determined using the Astra software package.