Engineering of a miniaturized, robotic clinical laboratory

Abstract The ability to perform laboratory testing near the patient and with smaller blood volumes would benefit patients and physicians alike. We describe our design of a miniaturized clinical laboratory system with three components: a hardware platform (ie, the miniLab) that performs preanalytical and analytical processing steps using miniaturized sample manipulation and detection modules, an assay‐configurable cartridge that provides consumable materials and assay reagents, and a server that communicates bidirectionally with the miniLab to manage assay‐specific protocols and analyze, store, and report results (i.e., the virtual analyzer). The miniLab can detect analytes in blood using multiple methods, including molecular diagnostics, immunoassays, clinical chemistry, and hematology. Analytical performance results show that our qualitative Zika virus assay has a limit of detection of 55 genomic copies/ml. For our anti‐herpes simplex virus type 2 immunoglobulin G, lipid panel, and lymphocyte subset panel assays, the miniLab has low imprecision, and method comparison results agree well with those from the United States Food and Drug Administration‐cleared devices. With its small footprint and versatility, the miniLab has the potential to provide testing of a range of analytes in decentralized locations.


| I N TR ODU C TI ON
Clinical laboratory tests are invaluable in diagnosing and managing diseases. 1,2 Currently, such tests are most often conducted in centralized laboratories using a variety of analytical methods executed on a multitude of large-scale analyzers. 3,4 These tests can also require clinically important amounts of blood, [5][6][7][8][9][10][11][12][13] sometimes including oversampling volumes 11,13,14 and often involve complex pre-analytical and analytical processing subject to human error. 15 In addition, samples must be transported from the patient to the laboratory, a process that may compromise sample stability and requires tracking procedures to avoid losing, mislabeling, or mishandling samples. 15 The ability to conduct a wide variety of tests near the patient on smaller volumes of blood especially benefits neonates, 16 the elderly, 17,18 those who need frequent testing, 12 and those with fragile or difficult to access veins. The decreased chance of losing or mislabeling samples helps physicians and hospitals provide better care. [5][6][7][8][9][10] Testing at the point of care is more convenient for the patient and increases patient compliance in fulfilling laboratory test orders. [19][20][21] Currently available point of care systems have made progress toward multiplexing commonly ordered tests, [22][23][24][25][26][27][28] but sometimes suffer from performance limitations when compared to centralized laboratories. 23 In addition, many point of care analyzers lack connectivity that would allow for comprehensive oversight and tracking of system performance.
When implemented with the same analytical performance specifications and relevant test menu as centralized laboratories, a connected, decentralized bench top testing platform could potentially augment standard clinical laboratory services. 29 We describe a simple-to-use miniaturized clinical laboratory system designed to test a diverse range of clinical analytes in distributed laboratory or near-patient settings. The system includes a hardware platform that performs the tests (i.e., the miniLab), an assayconfigurable cartridge containing all necessary consumable assay materials, and a server that communicates bidirectionally with the miniLab to manage assay-specific instructions and results (i.e., the virtual analyzer). We demonstrate the functionality of the system with analytical performance metrics across a variety of analyte classes, as represented by a Zika virus 30,31 molecular diagnostic assay, an anti-herpes simplex virus type 2 (HSV-2) immunoglobulin G (IgG) immunoassay, a clinical chemistry lipid panel, and a lymphocyte subset hematology panel. The Theranos technologies or assays described here have not been cleared, approved, or authorized for diagnostic use by the United States Food and Drug Administration (FDA).

| System design
The disposable assay-configurable cartridge, which contains the blood sample and all reagents and consumable vessels required to conduct one or more analytical tests, inserts into the miniLab hardware platform ( Figure 1, Supporting Information Video 1). The miniLab reads the barcode on the cartridge and sends it to the virtual analyzer server, which, in return, transmits the cartridge-specific protocol to the miniLab. The miniLab uses its robotic sample processing modules and cartridge contents to perform step-by-step specimen and reagent handling procedures as defined by the assay protocol. Following assay execution, the miniLab sends the protocol status and raw detector outputs to the virtual analyzer. The virtual analyzer analyzes the raw data and processes it into assay results for the sample and any controls. The virtual analyzer may also be used to communicate sample or quality control (QC) results for review and oversight by a laboratorian. This system design aims to allow the miniLab to operate in decentralized laboratory settings yet maintain many key qualities common to a centralized clinical laboratory.
Designing a laboratory testing platform that has the capability to measure multiple analyte classes, yet is simple enough to use by operators with minimal prior laboratory testing experience and compact enough to install on a countertop presented numerous challenges. To achieve the flexibility necessary to perform myriad clinical laboratory tests as well as maintain compatibility with potential future tests, we designed the platform to use miniaturized reagent and reaction vessels combined with a compact, versatile material-handling robot that has both pick-and-place and liquid-handling capabilities.
The design required a mechanism to process anticoagulated whole blood to plasma and multiple detector modules to measure various analyte classes. The material-handling robot needed to accurately and precisely aspirate and dispense microliter volumes of liquids with differing rheological properties, mix fluids, resuspend particles, and transport vessels and cuvettes to different areas of the platform.
Additionally, the entire platform required stable temperature control despite air exchange with variable ambient temperatures and filtering to provide protection for personnel and the environment. All of these FIG URE 1 Miniaturized clinical laboratory system overview. Materials and information interface and exchange between the cartridge, the miniLab, and the virtual analyzer. Abbreviation: QC 5 quality control NOURSE ET AL. | 59 modular elements had to be miniaturized and optimized to fit and function reliably together in a chassis with a volume similar to that of a desktop laser printer. Furthermore, careful hardware, firmware, and software integration was required to provide a comprehensive architecture for the successful operation of the miniLab while simultaneously facilitating ease-of-use.
We sought to limit the scope of system maintenance such as cleaning the device or refreshing reagents. Thus, we designed the assay-specific cartridge to contain all necessary reagents and consumable materials for each assay and to prevent the miniLab hardware from making direct contact with the sample or other liquids. The cartridge housing required fixed dimensions and configuration to interface with the miniLab, yet also necessitated a flexible layout to accommodate a broad combination of specialized vessels and reagents needed for varied test menus. The cartridge lid needed to open inside the miniLab and close again upon ejection to protect the contents and prevent exposure of the user to potentially hazardous materials.

| System architecture 2.2.1 | The miniLab
The miniLab is a bench top modular hardware platform with a small footprint (56 x 41 x 33 cm; up to 43 kg) that performs immunoassays, general chemistry, nucleic acid, and cellular characterization assays ( Figure 2, Supporting Information Video 1). The operational environment should be 20-308C with 20-80% relative noncondensing humidity and requires mains power and internet connectivity. After a simple software shutdown procedure that automatically locks the gantry in place, the miniLab can be moved and then reinstalled in a new location as required by the end users. The miniLab is equipped with an on-board computer running the Windows Embedded operating system. The user runs the miniLab through an exterior touchscreen that displays a simple graphical user interface application. The on-board computer and various modules' controller boards work collectively to carry out the assay workflow and simultaneously monitor all vital performance characteristics. Additionally, the miniLab's computer maintains connectivity to the virtual analyzer. The miniLab includes an on-board centrifuge (radius, 32.5 mm) for processing specimens with a relative centrifugal field of up to 3,000g and a sonicator for cell lysis. A thermal management system, composed of resistive cartridge heaters, controllable fans, and temperature sensors, monitors and controls the miniLab's internal temperature. Highefficiency particulate air filters prevent efflux of hazardous substances.
Cameras installed throughout the miniLab capture the cartridge barcode and record images of consumable materials to ensure protocols are executing as expected.
For the miniLab to have a small footprint, it incorporates four miniaturized detector modules optimized for analyzing low-volume samples ( Figure 2). Taken together, these detectors are designed to generate results for a diverse range of analyte classes. Each assay may use one or more of these detectors.  3. The spectrophotometer module is a crossed Czerny-Turner spectrograph featuring a broadband pulsed-Xenon lamp, allowing for simultaneous quantification of sample absorbance levels from 300 to 800 nm, a minimum spectral resolution of 10 nm, and better than 2.5 nm spectral accuracy.

The microscope module detects cells and other components in
samples by epifluorescence and dark-field microscopy with a minimum lateral resolution of 1.5 mm. The microscope includes a precision stage for scanning the sample (with 30-lm precision in the X-Y plane) and an independent Z-axis for auto-focusing (with 1-lm precision). An apochromatic objective lens magnifies objects onto a high-sensitivity image sensor. The microscope uses three laser diode light sources to excite samples, and the image sensor detects multiple spectral channels through selectable optical filters. Additionally, the module incorporates a ring light for darkfield microscopy, allowing for the visualization of elements with differential light scattering properties.
Due to space constraints within the housing, the current miniLab chassis can only contain either the thermocycler or the microscope.
Thus, the assay data presented here were collected on separate mini-Lab configurations.

| The assay-configurable cartridge
The self-contained, disposable, assay-configurable cartridges (9.2 x 12 x 3.75 cm) are designed and preloaded to contain all consumable materials required for single or multiple assay(s) on one sample ( Figure 2).
Consumable materials include sealed reagents and on-board controls, reaction and imaging cuvettes, assorted pipette tips, absorbent pads, and vessels. A lid encloses the cartridge body and contents. The sample container, which consists of two conjoined tubes that each hold up to 85 ml of sample (170 ml total), inserts into a dedicated, accessible cartridge slot. The cartridge lid uses a spring-loaded mechanism to open automatically after insertion into the miniLab, making the contents accessible to the material-handling robot. When the test is complete, the miniLab closes the lid and ejects the cartridge containing all used consumables and the sample for disposal. This design ensures that no fluids ever contact the hardware platform, thus limiting contamination and carryover and reducing the need for routine cleaning of the mini-Lab. Because the cartridge contains all necessary assay materials, the user does not need to maintain separate water, reagent, or waste tanks.
Cartridges described in this study were stored at 2-88C.
The barcode on the assay-configurable cartridge serves two functions. First, it links to the specific protocol stored in the virtual analyzer NOURSE ET AL. | 61 ( Figure 1). Second, it links to an assay calibration specific for each lot of cartridges. The virtual analyzer stores the assay calibration settings and uses the barcode information to identify the appropriate calibration parameters for data analysis.

| The virtual analyzer
The virtual analyzer, a central server, remotely manages and automates the workflow for all miniLabs (Figure 1). It is designed to conduct remote analysis and oversight of the test results processed in distributed locations by miniLabs. It communicates with each miniLab's computer using a secure communication channel (i.e., https protocol using Transport Layer Security 1.2) to transmit the assay protocols, receive and store raw signals, confirm the quality and integrity of the cartridge components, perform data and quality control analysis, and store results. The virtual analyzer also allows for changes in protocol and other system updates to be broadcast to existing miniLabs in the field.
The virtual analyzer stores all the raw signals, results, calibration, and intermediate files in a relational database that is encrypted at rest. The data exchanged between the miniLab and the virtual analyzer are encrypted using a device-specific certificate. Additionally, the virtual analyzer can become part of a laboratory information system. Laboratory personnel may review patient sample and QC results before authorizing the release of test results.

| System QC processes
At startup and specific intervals, the miniLab runs self-checks of the components' functions such as monitoring temperature and successful pickup of tips or vessels. The system can be configured to use several types of controls: on-board procedural controls, on-board assay controls, and external assay controls. Reagent-based procedural controls are designed to monitor hardware components, consumable materials, and individual protocol steps and can run alongside the assay-specific portion of the protocol. Analyte-specific assay controls may be included on the cartridge to monitor reagent loss and degradation as they are processed concurrently with each sample. Additionally, future versions of the system may use the spectrophotometer to monitor specimens for hemolysis, icterus, and lipemia during the protocol to determine assay validity for analytes sensitive to these factors (for the data presented here, these interferences were monitored externally). 32,50 External assay controls are run identically to samples ensuring that the entire analytical system is working according to specifications. Assay controls may be used to assess QC for the system. Any control that fails acceptance criteria voids the assay results. These system control processes will ultimately be applied automatically through software, preventing reporting of results at risk of compromise.

| System engineering for optimal assay performance and flexibility
The miniLab's core capability for small volume processing with the necessary flexibility to accommodate various assay types is underpinned by accurate, precise, and reliable liquid handling by the materialhandling robot, as well as small volume consumable components that are compatible with the on-board detection systems. In selecting raw materials to manufacture the consumable components of the cartridge, we balanced the requirements of small volume handling, including surface, optical, and physical properties. Furthermore, we devised and implemented methods to mitigate evaporation of small liquid volumes, such as capping aqueous reagents with oil or wax.
Due to the wide range of analyte classes and matrices the miniLab was designed to be able to test, the material-handling robot needed to be capable of aspirating, dispensing, and mixing diverse liquids with varying rheological properties (e.g., aqueous, high viscosity, volatile liquids, and complex fluids such as whole blood). This dictated that the pipettes on the material-handling robot have a wide dynamic range of speed, precise control, and simultaneous pump and Z-axis motion. In order to achieve the required liquid handling performance and reliability, we developed the hardware and control algorithm solutions described below.
We used custom-engineered canted coil spring shaft seals and centerless ground piston shafts in the pipettes. We designed a plastic bearing to constrain the piston motion. We optimized the gear ratio between the piston lead screw and the motor to achieve a sufficiently high number of encoder positions per unit of piston travel. We designed custom motor control algorithms and tuned the corresponding gains to achieve good steady state error, position overshoot, speed, and trajectory tracking performance for aspirating and dispensing fluids. In order to make the pipette robust, we used a profile rail linear guide for smooth motion and added a breakout printed circuit board assembly to the motor/encoder and vent valve so that all moving conductors would be combined in a high-flex, long-life cable. We also had custom metal gears designed so we could weld the gears to the motor and lead screw shafts. Lastly, the custom motor control algorithms used to control pump axis motion were also used to control pipette Zaxis motion and the gantry module motion in the X-Y plane.

| Analytical performance
To demonstrate the analytical performance of the miniLab, we performed analytical sensitivity, precision, and method comparison studies across four disparate assays that represent each of the major analyte classes and use each of the miniLab's detector modules. The functionality of the thermocycler and isothermal fluorescence detector was assessed with a Zika virus nucleic acid test; the photodetector was assessed with an anti-HSV-2 IgG immunoassay; the spectrophotometer was assessed with a lipid panel general chemistry assay; and the microscope was assessed with a lymphocyte subset panel hematology assay.
Details of each assay methodology and workflow can be found in the Supporting Information.

| Analytical sensitivity
We developed a qualitative nucleic acid test to detect Zika virus RNA in blood samples on the miniLab platform. We evaluated the analytical sensitivity of the assay by examining the limit of detection (LoD) using Zika virus spiked into whole blood with concentrations ranging from 0 to 3,520 genomic copies/ml. A minimum of six replicates were measured for each concentration tested, followed by 20 replicates at the putative limit of detection. The lowest Zika virus RNA concentration in whole blood at which a minimum of 95% of results were positive (55 genomic copies/ml) was confirmed as the assay LoD (Table 1). The analytical sensitivity is within the range reported by other available Zika virus nucleic acid amplification tests (13.4-18,000 copies/ml as reported by individual manufacturers). 33

| Precision
We measured repeatability (within-day and within-miniLab), betweenday, between-miniLab, and reproducibility (across miniLab and day) with standard deviations or percent coefficients of variation (CV) across three miniLabs over 5 days with five replicates per day at two medically relevant measurand concentrations for the anti-HSV-2 IgG, the lipid panel, and the lymphocyte subset panel assays (Table 2, Supporting Information Figure 1). 34 The precision results are all within established precision goals for each assay (lipid panel) 35,36 or are comparable to those of common clinical laboratory analyzers (anti-HSV-2 IgG and lymphocyte subset). 37-39

| Method comparisons
We compared results from the miniLab anti-HSV-2 IgG assay 40 for each study subject to those obtained by the reference method (Focus HerpeSelect 1 and 2 Immunoblot IgG; Table 3). 41  IgG assay performs with comparable accuracy to the reference method, with high sensitivity and specificity.
For the lipid and the lymphocyte subset panels, the first result from the miniLab was plotted against the mean of duplicate results from FDA-cleared comparator instruments for each study subject (Figure 3; Supporting Information Figure 2). 42 The data were analyzed using either Passing- Bablok 43 or weighted Deming 44 regression (Table   4). Median bias was calculated using the mean of the duplicate results (lipid panel) or the singlicate result (lymphocyte subset panel) in comparison with the mean of duplicate results from the comparator methods for each study subject (Table 3). Bland-Altman plots 45  The system is simple enough to be used by operators with minimal prior laboratory testing experience and is designed to be operated in decentralized laboratory and other near-patient settings.
We selected tests to represent common analyte classes (molecular diagnostics, immunoassay, clinical chemistry, and hematology) and to assess each of the detector modules built into the miniLab. These tests also demonstrated the coordination and functionality of the miniLab's sample handling components (material-handling robot, centrifuge, and magnetic bead manipulation), which together replicate many of the steps typically done by laboratory personnel in central laboratories.
Most pre-analytical and all analytical steps were self-contained and automated within the miniLab. Our analytical sensitivity results for the Zika virus NAA assay showed that the miniLab was able to detect Zika virus at 55 copies/ml of whole blood. While this is in the same range as other tests for Zika virus RNA, 33 it achieves this sensitivity using a sample volume of only 150 ul, meaning that the assay is able to reliably detect fewer copies of Zika virus than other tests with the same analytical sensitivity but higher sample volumes.
Precision studies for anti-HSV-2 IgG, lipid panel, and lymphocyte subset assays showed low imprecision for all three assays/panels and  The design of the system presents some limitations. As with certain point of care analyzers, the miniLab can test only one cartridge at a time, which limits the throughput of a single device. We plan to shorten the assay run times to increase the miniLab's throughput and utility. The data presented here show only one assay or panel per cartridge, though the system is designed to be capable of multiplexing disparate assays or panels together using a single cartridge and associated protocol. Finally, the current version of the platform cannot contain all the modules in the same chassis. Future versions of the platform will contain the full functionality of all modules within a single device.

| CON CL U S I ONS
The miniLab has a unique potential as a clinical laboratory testing platform, and the system is designed to fill several unmet opportunities in the field of diagnostic testing. The system requires smaller specimen volumes than many other analyzers. This makes frequent testing more feasible for populations for whom required blood volume may limit laboratory testing (e.g., the young, old, and select at-risk populations). The miniLab's compact size, coupled with the virtual analyzer's connectivity and the versatile assay-configurable cartridges, has the potential to provide high-quality testing of diverse analytes in decentralized laboratory and other near-patient locations, which could expand access to clinical services and expedite diagnoses and therapies. By augmenting access to clinical laboratory testing options, the miniaturized clinical laboratory system has the potential to complement the arsenal of technologies available to the clinical laboratory community.     Figure 7).
We purchased all anti-HSV-2 IgG specimens (202 samples) and some lipid panel specimens (21 samples analysis was performed with two-way nested analysis of variance with random effects ("day" nested within "device") to determine the components of the variance for each assay. We used Grubbs' test at the 99% level to exclude up to one outlier per miniLab per measurand and per sample pool, as recommended in the CLSI EP05-A3 guideline. 34 We estimated the repeatability (within-day and within-miniLab), betweenday variation, between-miniLab variation, and reproducibility (acrossday and across-miniLab; across three miniLabs) and their 95% CIs for these estimates. The root mean square of the repeatability, betweenday, and between-miniLab precision components equals the reproducibility (across-miniLab and across-day). We used the Satterthwaite approximation to calculate the 95% CIs on the precision terms. 34 The sensitivity, specificity, and their CIs (direct score calculation) were calculated as described in CLSI EP12-A. 41 Quantitative method comparison of the first replicate (lipid panel) or singlicate (lymphocyte subset) result from the miniLab and mean of duplicate results from the comparator methods 41  concentrations). Ninety-five percent CIs for the slope and intercept for each regression were calculated by the bootstrap method. Weighted Deming regression was used for analytes with wide dynamic ranges that were operating in the constant CV region, as recommended by CLSI guideline EP09-A3. 42  Median absolute bias or median proportional bias was calculated and the approximate 95% CI was obtained by an application of the binomial distribution. 49

| Data availability
The data and analysis scripts presented in this paper are available at https://osf.io/ur7kw/

ACKNOWLEDGMENTS
We thank all employees of Theranos for their contributions to this body of work. We thank the manufacturing teams for producing the hardware and consumable materials, the clinical studies staff for sample procurement, and the project managers for coordinating the work. We acknowledge Tammy Burd for the design of consumable materials, Mohit Goel for assay integration, Nicholas Haase for data analysis, Chi Nguyen for Zika virus assay development, Daniel Nguyen, James Wasson, and Alan Yip for hardware design, and Sean Tolibas for electronics design.

CONFLICT OF INTEREST
All authors of this paper, as either current or former Theranos