Externally-coupled optical crosstalk.

With the goal of matching the sensitivity of high-end benchtop luminometers, we investigated the performance of the instrument in very low light conditions to gain a deeper understanding of baseline noise and factors limiting sensitivity. With a sample of pure water, we noticed that the sensor current was higher when the shutter was open than when it was closed. Stray ambient light gated by the shutter was ruled out as the cause by recording this effect with the device exposed to ambient room light vs. inside a secondary sealed metal box, which resulted in identical results. Electronic noise emitted from the motors was ruled out by conducting measurements in the absence of shutter flags, which showed no modulation effects on the signal. Finally, shutter-induced stray electric fields were ruled out by grounding and un-grounding the metallic shutter flag to the PCB.

Instead, the modulation was found to be directly correlated with the reflectivity of the shutter flag and that of the sample cavity. Upon further investigation, the residual gated signal was also found to be directly proportional to the raw, temperature-dependent dark current of the sensor. Our findings are consistent with spontaneous optical emission originating from the sensor that is externally reflected back via the sample cavity. We refer to this phenomenon as externally-coupled crosstalk (ECC).

Such spontaneous optical emission is the cause of optical crosstalk, as explained by the sensor datasheet [28] and the primary literature on SiPM. Specifically, charge carriers accelerated by the applied external bias field, called hot carriers, randomly emit photons [23,24,28] which in turn cause secondary electron avalanches within the device. Although the optical crosstalk effect has been discussed exclusively in the context of internal reflections within the silicon and glass cover window [23], we reasoned that the residual shutter-mediated dark signal we observed was likely caused by some fraction of the crosstalk photons escaping the sensor and reflecting back from our system. The magnitude of the effect was found to range from 0.5–0.8% of the raw dark current. Without correction, this effect is responsible for an anomalous offset of up to 75 fW equivalent optical power and a residual temperature drift of about 2 fW/°C.

We developed a built-in push-button correction to remove the error caused by ECC (S1 Fig). By sampling the effect across a wide variety of temperatures (4°C to 40°C), the effect was found to be highly linear with respect to the shutter-closed dark current. The effect could be modeled and subtracted in real time from the raw data. In Fig 3 we demonstrate the robustness of the correction over a range of temperatures.

Limit of detection.

We benchmarked performance using recombinant firefly luciferase in both Promega Bright-Glo (BG) and Luciferase Assay System (LAS) conditions, representing commonly-used luminescence reference standards comparable between labs and across devices (Fig 2A and S2 Data). We additionally titrated purified nanoluciferase in the presence of whole blood, serving as a means to normalize device performance across our fleet of instruments in the context of a real serology application (Fig 2B and S3 Data). Our limit of detection for firefly luciferase in Promega Bright-Glo assay buffer was 6.4E-19 moles of firefly luciferase, as defined by LOD = 3σ/m, where σ is the standard deviation in RLU across baseline measurements, and m is the measured sensitivity, in units of RLU per mole of enzyme. By the same metric, our limit of detection using LAS was 1.6E-19 moles of firefly luciferase.

We characterized the performance of the luminometer using a home-built radiometric test bench (S2 Fig) in order to validate the performance over a wide range of luminescence intensities. By using a closed-loop stabilized lamp, radiometric measurements provided data on absolute hardware limitations of the device (Fig 2C and S4 Data). The limit of detection is a function of integration time, which we defined as the number of independent 1s samples: each 1s sample consists of a 1s shutter-open period and a 1s shutter-closed period. With a 150 sample integration, our instrument achieved a limit of detection of approximately 1 fW of radiometric optical power incident on the sensor, or ~2300 photons/s over a 36 mm² active area. The device is limited in practice by the diminishing return of longer integration times, which become impractical beyond this point, as the SNR improves only with the square root of the number of samples. In Fig 2, we report the limits of detection with both enzyme titration and radiometric testing, and in Fig 3 we report the background noise as a function of temperature, for both raw and gated output.

Performance under ambient temperature ramp. Since remote operation in tropical regions is a central application of the device, rejection of temperature-induced drift is central to retaining accuracy. Our temperature compensation scheme consists of several factors: first, the shutter system permits continuous subtraction of the sensor’s dark current. Second, our empirical calibration of the ECC effect has proven to be robust to within the system’s measurement accuracy for all integration times, and for all ambient temperatures that were explored (S1 Fig). Third, to prevent buildup of excess heat inside the device, a serpentine airflow path was designed into the mechanical layout of the luminometer, such that ambient air flowed over the analog sensor board prior to the main heat-producing elements such as the computer, shutter motors, and the power management board. Despite cooling, we observed shutter-generated changes in the internal temperature of the sensors, although there was no detectable change in the gated signal due to temperature effects.

Further evidence of complete DC temperature drift rejection is presented in Fig 3 (raw data provided in S5 Data), which displays raw amplifier time traces at three different temperatures, after a 1 hr pre-incubation at each temperature. The ECC calibration uses the raw, shutter-closed dark signal to compute the expected measurement offset due to optical crosstalk, which is continuously subtracted from each sampled datapoint. Gated and compensated traces are plotted in Fig 3A, scaled by 100× in order to match the scale of the plot, and exhibit only residual shot noise in the sensor dark current.

In Fig 3B, the Allan deviation of the dark signal reports on the stability of the system across different timescales, by subsampling the data and computing the pairwise deviation of nearest neighbor samples. The raw data exhibits a monotonic increase in deviation with longer integration, meaning that no amount of signal averaging is beneficial without shutter gating. In other words, the raw signal exhibits a random walk. In contrast, the Allan deviation of the shutter-gated and ECC-compensated data monotonically decreases, indicating that the residual error is uncorrelated and converges toward zero with increasing integration time at a slope of -½, consistent with Eq (1). It should be noted that the effect of elevated temperature is not entirely avoided as the amplitude of the thermal dark current increases, and with it the residual stochastic shot noise. The amplitude of the shot noise for a five sample measurement is 1.4 RLU at 4°C, 2.3 RLU at 20°C, and 3.3 RLU at 30°C. Relative shot noise scaling vs. temperature was consistent over the sampled range of integration times.

Fleet calibration and enzyme titration. A total of ten luminometers were fabricated and tested. The devices were nominally identical, but required individual testing to verify their performance, as electronic component values can vary within their expected tolerances, and minor differences in assembly may affect the sensitivity. The devices’ sensitivities were normalized by preparing a master mix of nanoluciferase diluted to 1 pM (130 μL volume per sample) and measuring it on all devices, each exactly 60 seconds after the final pipetting step when substrate was introduced. Since each luminometer employs a scaling factor used to generate RLU values, the fleet was normalized to approximately 10 RLU per attomole (1E-18 moles) of nanoluciferase. After normalization, the RLU values measured on all ten devices using samples prepared from the same master mix had a coefficient of variation of 2.45%, demonstrating DC stability and accuracy of the measurements across devices, even in the absence of temperature stabilization.

Using the device for a luminescent antibody detection assay.

With the aim of validating device performance on clinical samples, the luminometer was tested with the spLUC assay for detecting antibodies to SARS-CoV-2, which relies on the fusion of the split nanoluciferase domains to SARS-CoV-2 proteins [8]. The assay is simple and requires no wash steps, making it amenable to non-laboratory or low-resource settings. Additionally, the assay exhibits a large dynamic range and is quantitative across a broad range of antibody concentrations, compared to other serology assays. For example, the ELISA method is a common quantitative lab-based antibody assay but is limited by its multiple wash steps, calibration standard curves, and extensive incubation time (running time 2–4 hours) that makes it impractical in point-of-care applications. The more common lateral flow assay (LFA, running time <30 min) is a very simple nanoparticle method for detecting antibodies but lacks quantitative readout and can produce heterogenous results depending on variations in the manufacturing process [29]. Another assay common in clinical settings is the chemiluminescent microparticle immunoassay (CMIA), however this requires complex antibody functionalized microparticles and specialized instrumentation and that, similar to the ELISA assay, cannot currently be implemented outside of a laboratory. The spLUC assay has the advantage that it is rapid (running time 30 min) and simple, but also delivers a quantitative value for antibody level. However, it requires a highly sensitive luminometer, which is not commonly found in non-laboratory settings. The development of this SiPM luminometer allows for the spLUC assay to be used in rural or low resource settings to obtain accurate and quantitative measurement of antibody levels.

The laboratory-based spLUC assay was previously validated as a point-of-care based assay (described in [8] and illustrated in Fig 4A). To render the sensors stable at room temperature, they were lyophilized in clear, thin-walled PCR tubes. Phosphate-buffered saline (PBS) was dispensed into dropper bottles and estimated that on average 3 drops corresponded to adding 120 μL for reconstituting the sensors. Microsafe capillary pipettes (Safe-Tec) were used to collect 10 μL of fingertip blood and deposit it into the reconstituted sensor tube. At this point, the sample was incubated for 20 minutes to allow the sensors to bind antibodies present in the blood. Ahead of time, nanoluciferase live cell substrate (Promega) was dried by vacuum centrifugation in glass vials to produce a dried substrate. Upon performing the assay, the substrate is reconstituted in water and transferred to a dropper bottle with a transfer pipette. After the incubation of blood and sensors, one drop of reconstituted substrate was added to the sample. While luminescence can be detected in this whole blood suspension, we found that the signal was more than 5-fold higher if the red blood cells were pelleted by centrifugation in order to clarify the sample (S3 Fig). We therefore used a small, battery-powered centrifuge to pellet the red blood cells before detecting luminescence with the SiPM luminometer. Together, this reformatted point-of-care spLUC assay allowed for antibody detection with limited laboratory supplies and simplified steps. We demonstrated that this assay worked by measuring its ability to detect antibodies using the International Serology Standard (ISS) for SARS-CoV-2 developed by the WHO [30]. The point-of-care spLUC assay was able to detect antibodies from the ISS sample over 64-fold dilution for the S sensor and 100-fold dilution for the N sensor, where 1000 International Standard (IS) units represents the undiluted serum standard (Fig 4B). For serum, the threshold values were determined to be 22 RLU for the S sensor, and 32 RLU for the N sensor by using pre-pandemic negative control serum. The sensitivity with the handheld luminometer was similar to the levels seen on a plate-based benchtop luminometer (see [9], Fig 3).

To demonstrate that the luminometer could be used to detect antibody levels via the point-of-care spLUC assay, we tested capillary blood from 25 vaccinated adult volunteers recruited from the UCSF campus on July 9^th, 2021 (S6 Data). We measured antibodies in two replicates via spLUC for both the Spike (S) and Nucleocapsid (N) sensors. The same individuals were tested in parallel via an FDA-approved lateral flow assay for IgG Spike antibodies (Nirmidas MidaSpot), as well as by dried-blood spot Spike ELISA assay [31]. In order to assess correlation between the methods, LFA IgG band intensities were quantified by image analysis and normalized to the intensity of their control bands. Both the LFA and ELISA assay had strong correlation to the spLUC assay signal with R = 0.84 and R = 0.91, respectively (Fig 4C).

Recording serological prevalence of SARS-Cov-2 in Bangladesh.

To demonstrate the ability of the device to support global health applications, a pilot SARS-CoV-2 seroprevalence study was implemented using the spLUC assay in the Mirzapur district, 60 km outside Dhaka, Bangladesh, in partnership with the Child Health Research Foundation (CHRF). Eight portable luminometers and spLUC assay reagents were shipped to CHRF and training of personnel to conduct the assay was performed remotely over video chat. A cross-sectional cohort of 204 volunteer participants was assembled from a combination of Mirzapur community members and CHRF staff throughout the period from May-July 2022. Volunteer CHRF staff members were tested from May to through June before rolling the assay out to larger community field camps June 20^th and 27^th, 2022. Capillary blood samples were collected from each participant by CHRF staff, in addition to vaccination and previous infection history (self-reported, if known). Fingertip capillary blood was used for the spLUC assay to quantify antibodies to both S and N proteins. All raw data for the study is provided in S7 Data.

Our findings indicate that the majority of the 204 participants in the cohort had positive signals for both S and N sensors, with 156 (76.5%) positive for S, 142 (69.6%) positive for N, 185 (90.7%) positive for at least one sensor, and 113 (55.3%) positive for both sensors, defined as signal greater than 20 RLU for S, 25 RLU for N (Fig 5A). These thresholds were determined by using mock negative control blood containing washed red blood cells from a recent donor, reconstituted with pre-pandemic negative control serum. Note that positivity thresholds differ slightly between experiment with serum (Fig 4B) and whole blood (Figs 4C and 5A–5C) due to differences in optical absorption of the sample, and are determined separately (Methods). Although seroprevalence across regional populations may exhibit complex spatiotemporal dynamics, our results are in approximate concordance with previous studies of seroprevalence in Bangladesh residents [32,33].

Although our sample size was small for this pilot study, we were able to detect a robust distribution of antibody titers across the population (Fig 5A). The slope of the numerically-ascending sorted data in Fig 5A reflects the density of points at a given signal level, from which we identify three distinct regions for each sensor: a negative signal region with rapid dropoff (a large relative signal change over a small number of participants), a broad mid-level positive band with signal varying uniformly across a wide range from ~20–1,000 RLU, and a high positive signal band comprising a smaller number of individuals with exceptionally high antibody levels (> 1,000 RLU). Note that due to the nature of the luminometer’s strong coupling between its sample cavity and the sensor, the low signal region may be susceptible to small variations caused by differences in blood volume or hemoglobin levels across individuals. In particular, samples lacking SARS-CoV-2-specific antibodies will not exhibit enzymatic emission of light, but will still incur variable absorption of sensor ECC emission by the hemoglobin in the blood as compared to the calibration sample. Although we mitigate this effect by centrifugation, the pelleted cells nonetheless remain present in the detection volume of the device, but at a reduced cross sectional area. Combined with residual stochastic noise, variation in sampled blood volume and participant hemoglobin density may explain the presence of negative data points in the study, which have been clipped for logarithmic display in Fig 5.

Leveraging participant survey data, we plotted spLUC distributions stratified by the number of vaccine doses received (Fig 5B and 5C). The data reflect a robust response to vaccine boosters for spike antibodies, with the largest change occurring after the third dose. Although the median spLUC value for spike (S) response increases uniformly from one to three doses, the largest observed changes occurred in the 90th percentile of the cohort, which increases from 100 RLU (one dose), to 443 RLU (two doses), and 1417 RLU (three doses). In contrast, the nucleocapsid (N) response is relatively independent of the number doses received, with the median response approximately constant at 127 RLU (one dose), 99 RLU (two doses), and 105 RLU (three doses), and the 90th percentile exhibiting non-monotonic behavior and a high coefficient of variance across all groups. The observed increase in anti-Spike antibodies with vaccine doses may be expected due to the onset of mRNA vaccine administration in the Tangail region in October 2021 [34], which are known to elicit robust responses against spike protein alone. However, earlier vaccination campaigns in the region utilized other vaccines including the Sinovac vaccine which is an inactivated, whole virus formulation that is likely to elicit anti-nucleocapsid antibodies. Indeed, studies of other inactivated SARS-CoV-2 vaccines have been shown to boost anti-nucleocapsid responses in both previously infected and vaccinated individuals, albeit to a lower level than anti-Spike responses [35]. Furthermore, nucleocapsid antibodies have been shown to wane faster than spike-specific antibodies after infection and may explain the lack of correlation between nucleocapsid antibody titers with dose of vaccine [36]. Since there were only five unvaccinated participants in the cohort, there were not enough data to draw conclusions on the effect of a single vaccine dose versus none. Furthermore, the average elapsed time since the most recent dose for those with one and two doses was 150 and 180 days, respectively, whereas those with three received their most recent dose on average 90 days prior to the study date, suggesting titers from the third dose were still high due to the booster’s recency.

We recorded self-reported infection status: 28.4% of participants reported having at least one infection, 11.8% having at least two, and 2.9% having three infections. We explore basic statistics stratified by the survey data, in the context of inherent challenges associated with asymptomatic, undetected, or incorrectly-assumed infections, as well as a widely-varying duration and magnitude in individual titers resulting from natural infection [37,38]. For example, a study in the United States [39] found that 43.7% of adults with “serology results indicative of past infection…reported never having had COVID-19, possibly representing asymptomatic infection”, suggesting that self-reported infection data in general may not correlate well with serological findings. Similarly, previous studies in Bangladesh measured majority seropositivity fractions concordant with our own data [32,33], both of which suggest much higher rates of exposure than indicated by self-reporting.

Nonetheless, the survey data exhibits monotonically-increasing median spLUC values for both S and N sensors, as a function of the number of reported infections (S4 Fig). A Wilcox ranksum test was performed, comparing stratified distributions as a function of dose and infection number, confirming the statistical significance of self-reported infection status, with the greatest significance emerging between zero and any number of reported infections for S signal, as well as the effect of multiple infections on N signal (S5 Fig). In contrast, the N signal was found to be independent of the number of vaccine doses (Fig 5C and S5 Fig), which could be explained by the latter transition to mRNA vaccines with strong anti-Spike responses only.

Our results suggest that the split luciferase assay in combination with handheld luminometers allows for quantitative detection of anti-Spike and anti-Nucleocapsid antibodies in capillary blood, serving as a tool for global health surveillance in pandemic or endemic settings, where acquisition of geospatial health data would otherwise be impractical or too costly to obtain. The pilot study demonstrates the utility of quantitative serological monitoring, permitting downstream analysis and leading to actionable public health information in geographically remote, underserved populations.

UCSF study.

The University of California San Francisco (UCSF) cohort samples were collected under IRB protocol 20–33062. Samples were de-identified before delivery to the lab where all assays described here were performed. Advertisements including flyers, emails, and word of mouth with information about the study was shared with individuals who were employed at University of California San Francisco (UCSF) and who had received a COVID-19 vaccine. Those interested in the study contacted the study team for more details. All participants were ≥18 years and provided informed written consent.

CHRF study.

The study was approved by the ethical review board of the Bangladesh Shishu Hospital and Institute. Participants were enrolled in this study based on their interest in participating and on informed written consent. All participants ≥18 years provided informed written consent. Parental permission and informed written consent were obtained for interested participants aged 16–18 years. Enrolment procedures: Prior announcements were made about the objectives of the study and the need for volunteers. Enthusiastic participants, who communicated with the study staff, and willingly consented to join the study, were contacted for recruitment. Participants who were not CHRF staff received gifts (hand sanitizer and biscuits) for their voluntary participation in the study. Identifying information was not recorded or transmitted to research staff performing assays.

Electronics design.

PCBs were designed using KiCad and the full design is available here: https://github.com/czbiohub-sf/portable-luminometer/tree/master/Electronics%20design. The PCBs themselves were fabricated by a commercial vendor (OSH Park standard four layer). Surface mount components were soldered to the boards by using a commercial reflow oven (LPKF ProtoFlow S N2). Solder paste was applied using stainless steel stencils (OSH Stencil) and a manual stencil printer (Neodem FP2636). A high temperature solder paste was used on one side of the PCB and low temperature solder paste on the other in order to populate the sides sequentially without component adhesives. Through-hole components were hand-soldered after reflow was completed. SD cards containing the Raspberry Pi OS and custom configurations were cloned from an original disk image. A 4400 mA-hr lithium polymer battery (Adafruit #354) was installed in the enclosure along with a power management board (Adafruit Powerboost 1000C, #2465) that was positioned near the fan-driven airflow exhaust to minimize thermal load. An electronics workstation equipped with a grounded, conductive mat and a stereo microscope was used for all electronics assembly. All analog PCBs were functionally tested prior to soldering of the SiPM sensors. Schematic diagrams of the analog and digital PCBs are available in S6 and S7 Figs, respectively. Electronic background noise in the absence of a sensor (no dark current) was recorded and found to be substantially lower than any electro-optical effects in the system (S8 Fig).

Mechanical assembly.

All part numbers and manufacturers are listed in the supplementary information (S1 File). Luminometer enclosures and a majority of the custom mechanical parts were 3D printed using multi-jet fusion technology (Hawkridge technologies, HP MJF 5200 series printer, PA12 material). Individual part files for 3D printing are provided as supplementary material (S5 File). Enclosure parts were painted using a black conductive spray paint (Total Ground paint, MG Chemicals) to reduce stray light and provide electrical shielding. Internal surfaces on potential stray light paths were additionally painted with Black 3.0 (Culture Hustle USA), a highly-absorbent black paint. Tube holders were painted with both Total Ground and Black 3.0, and aluminized mylar film glued into the sample cavities. Shutter flags were custom designed and laser-cut from 0.008” thick stainless steel, and painted with both black paints. PCBs were assembled together using board-to-board interconnects, standoffs, and a custom machined aluminum board spacer containing miniature roller bearings to position and support the shutter driveshafts (M2 machine screws) protruding through both PCBs. The machined spacer additionally served to ground the shutter system electrically, and had machined bosses making contact with ground pads on the digital PCB. Coreless DC drone motors were used to drive shutter actuation (HxChen micro-DC motor). To alleviate space constraints the DC motors were offset laterally and coupled via 3D-printed spur gears with angular cutouts serving as motion stops. Laser-cut rectangular patches of electrical tape were applied to the surfaces of the motion stops to eliminate shutter bouncing at the stops. The drive spur gears were press-fit to the 0.8 mm shaft of the motors and glued with loctite 420. An additional spacer offset the motors axially and served as a motion-limiting stop. The shutter follower gear was threaded onto the drive shaft and secured with loctite 420, and a pair of stainless steel washers coated in teflon dry lube used to reduce friction and improve light shielding against the PCB. Drive shafts were inserted through the assembly, shutter flags attached on the sensor side of the assembly, and clocked rotationally using a custom laser-cut alignment tool. Shutter flags were secured using thin brass M2 hex nuts, tightened with miniature wrenches, and secured using loctite 420. Silicone gaskets were laser-cut and glued into the enclosures to improve light shielding near the button interface, and an open-cell foam gasket was used to pad the display contact area and improve light shielding. The lithium polymer battery was connected to the power management board and secured in place with a custom 3D-printed flex clip.

Software.

Luminometer control software was written in python and freely available here: https://github.com/czbiohub-sf/portable-luminometer. In brief, a button-operated user menu system was created by connecting callback functions to edge-detection events on the Raspberry Pi GPIO pins, allowing the user to navigate between calibrations, device self-diagnostics, and measurement menus. In the measurement menu, the user is able to select from a range of measurement durations. The software was generally built around a low-latency event loop, which assigned time-consuming I/O-bound hardware interactions to worker threads without delay to the main loop, allowing the button menu to remain responsive throughout long-duration measurements. Data was acquired from the 24-bit ADC via an SPI interface using a custom hardware adapter. To compute each measurement sample, raw data points from the ADC within the same shutter cycle were averaged and gated by subtraction of the mean neighboring shutter-closed points, and further compensated by ECC calibration based on the value of the shutter-closed dark signal. Values from each sample were averaged to produce a final output, with error bars defined by the standard error of the mean (sem) of all the samples. ECC calibrations and other device constants were stored on the device’s SD card.

Radiometric calibration.

A radiometric test station (S2 Fig) was built to characterize the luminometer’s optical response in the absence of any variability potentially introduced by changes in enzyme activity levels, liquid settling, the motion of bubbles, or any other effects specific to liquid samples. A closed-loop, aperture-stabilized mercury vapor lamp was used as a light source (Exacte, Excelitas Technologies). The lamp was equipped with a 3 mm Liquid Light Guide (LLG), whose tip served as a stable, homogeneous emission source. The tip of the LLG was imaged onto the luminometer SiPM sensor via a unity magnification 4-f lens relay containing a 2 mm iris at the shared Fourier plane of both lenses, reducing the numerical aperture and intensity of the light. The LLG was additionally filtered using a fluorescence excitation filter (Semrock FF02-475/50-25) in order to restrict the bandwidth of the lamp to a similar emission spectrum as the nanoluciferase enzyme. Finally, a total value of OD = 7.48 in neutral density filters were placed between the second lens and the luminometer to further reduce the light intensity (measured, as opposed to nominal, OD values were used for each filter, by testing with a commercial power meter, Thorlabs P100D). The intensity of the source could then be varied from 0–100% power using the control panel on the mercury lamp. A custom luminometer radiometric rear access lid was 3D printed, into which an optical tube was threaded, providing light-tight coupling of the metered source into the device. The sample holder with reflective cavity was removed from the luminometer to allow the lamp light to reach the sensor. The shutters remained in operation to provide drift compensation. To achieve a range of incident optical powers, the lamp’s power setting was adjusted between 0–100% power, and individual neutral density filters were added or removed.

Preparing Split luciferase sensors, nanoluciferase control, and dried substrate.

The S and N sensors were expressed and purified as previously described in [8] and more detailed in [9]. In brief, all four constructs in the pFUSE vector (S-smbit, S-lgbit, N-smbit, N-lgbit) were transfected into Expi293 cells using Expifectamine according to the manufacturer (Thermo Fisher Scientific). Following purification using Ni-NTA affinity chromatography, the proteins were flash-frozen in liquid nitrogen and stored at -80°C. For lyophilization, each sensor pair (i.e. S: RBD-LgBit + RBD-SmBiT or N protein-LgBiT + N protein-SmBiT) was prepared combined at 48 nM each sensor in PBS + 4.8% BSA + 1.2% Tween pH 7.4. Nanoluciferase enzyme was also expressed via Expi293 cells and purified with Ni-NTA affinity chromatography. Nanoluciferase was prepared at 500 pM in PBS + 4.8% BSA + 1.2% Tween. 5 μL of each mixture was aliquoted into separate PCR tubes, flash frozen in liquid nitrogen, and lyophilized with a batch lyophilizer (SP VirTis BenchTop Pro with Omnitronics, BTP-8ZGEVX, with Crylic Drum 377 Manifold and Bulk Shelf Rack). Live cell nanoluciferase substrate (Promega, N2012) was used to prevent cell lysis during the assay. For the small UCSF cohort study, 57.5 μL of the substrate was vacuum dried (GeneVac) in a black dropper bottle (US Plastic Corps, #66543) and stored at -20°C. In addition, 1192.5 μL of the substrate diluent (from Promega kit) was added to a 3 mL clear dropper bottle. One diluent bottle was prepared per black substrate bottle. For both studies, the lyophilized sensor, control tubes, and diluent bottles were stored at room temperature for at least one month prior to being used. The lyophilized substrate was stored at -20°C until the day of the study.

Firefly luciferase assays. In order to compare two different kinetic rates, the enzyme was prepared in both Bright-Glo reagents (Promega #E2610) as well as Luciferase Assay System (LAS, Promega #E1500). Both assays used Promega QuantiLum Recombinant Luciferase (Promega #E1701), and were each performed closely following the manufacturer’s respective recommendations. In the former case, the luminescence was stable enough to prepare all conditions and measure sequentially, with an expected signal decay time of about 30 minutes. In the latter case, special care was taken in order to ensure consistent timing to avoid inconsistent signal decay, which was expected to occur on the timescale of 1–2 minutes. Substrate was only combined, mixed, and immediately measured in a highly consistent manner across all titration points. In both assays, enzyme was serially diluted from stock (as reported previously [8]) into a mixture of phosphate-buffered saline (PBS), 0.2% bovine serum albumin (BSA), and 0.05% tween-20 (PBSTB). A total assay volume of 30 μL was used. For Bright-Glo assays, each measured tube contained equal parts of diluted enzyme in PBSTB (15 μL) and reconstituted substrate in Bright-Glo assay buffer (15 μL), while for LAS, 5 μL of reconstituted enzyme was combined with 25 μL of reconstituted substrate in LAS buffer. All tubes were gently centrifuged for 3–5 seconds prior to measurement to ensure the liquid was at the bottom of the tube and without air bubbles.

Nanoluciferase titration assay in whole blood. Nanoluciferase (nanoluc) enzyme was expressed and purified as reported previously [8]. To emulate the spLUC assay performed in whole blood and normalize instrument sensitivity across the fleet of devices, nanoluc enzyme was serially-diluted into a mixture of whole blood and PBSTB (PBSTB-WB, or 12 parts PBSTB, 1 part whole blood) for titration analysis. Whole blood samples were purchased through Vitalant. For each titration point, 280 μL of substrate diluted in Nano-Glo dilution buffer (Promega #N206A) was combined with 1050 μL of diluted enzyme in PBSTB-WB, and divided into a strip of eight PCR tubes, each containing 165 μL of final reaction. Each set of eight tubes (one set per titration point) were measured in parallel across the fleet of portable luminometers.

Split luciferase assay validation with International Serology Standard (ISS).

The lyophilized sensors were reconstituted with 3 drops of PBS from a clear 3 mL dropper bottle (US Plastic Corps, #66529, approximately 120 μL). After reconstitution, 10 μL of various dilutions of the ISS (WHO Serology Standard (Mattiuzzo et al. 2020) were incubated for 20 minutes with the reconstituted sensors. Dried nanoluciferase substrate was reconstituted with 1 drop of methanol from a 3 mL clear dropper bottle (US Plastic Corps, #66529). After 5 minutes, all of the previously prepared substrate diluent bottle was added to the reconstituted substrate in the black dropper bottle. Once the sensor and antibody sample incubated for 20 minutes, one drop of the reconstituted luciferase substrate + diluent was added to the sensor sample. The samples were incubated for 5 minutes before measurement on the SiPM luminometer. Negative control serum (Sigma Aldrich, pre 2020) was used to determine the negative cutoff value for serum.

Split Luciferase assay test on UCSF volunteers.

The study size was limited by the number of researchers available to process samples, with the number of participants enrolled being sufficient to assess the correlation of spLUC with ELISA and LFA.

For each participant, two lyophilized S sensor tubes and two lyophilized N sensor tubes were tested. In addition, two control nanoluciferase tubes were tested in parallel every hour to account for drift in substrate signal. Two researchers were needed to efficiently run the assay with participants scheduled every 20 minutes. First, one researcher reconstituted the sensors for the upcoming participant (i.e. two S and two N tubes) by adding 3 drops of PBS from the 3 mL dropper bottle to each sensor tube, including the control if applicable (US Plastic Corps, #66529, approximately 120 μL). The tube was flicked down to ensure the liquid reached the bottom and reconstituted the lyophilized pellet. Next, the participants’ middle finger was sanitized with an alcohol wipe and allowed to air dry. The finger was pricked with the lancet and kept facing down to let the blood pool. The microsafe capillary pipette (10 μL) gently touched the pooled blood until it reached the filled line. The pipette was then carefully placed into one of the sensor tubes and added by depressing the bulb of the pipette. This was repeated another three times so that all four tubes contained 10 μL of blood. A new microsafe capillary pipette was used each time. Blood was not added to the control samples. After the addition of capillary blood, all tubes were mixed by rotating up and down. A final flick was done to reduce the amount of blood on the cap of the tube, followed by a 20 minute incubation. For this study, dried substrate was used. During the sensor incubation time, the dried substrate in the black dropper bottle was rehydrated with one drop of methanol (3 mL bottle, approx. 40 μL). After 5 minutes of rehydration, all contents of the substrate diluent (stored in a separate 3 mL bottle) was added to the black dropper bottle with rehydrated substrate. This bottle was rotated up and down to mix. Once the 20 minute sensor incubation completed, one drop of the diluted substrate was added to each sensor tube (and applicable control tubes) and mixed. Sensor samples were then centrifuged with a table-top battery powered centrifuge (TOMY, MPN # 00101244) for 30 seconds to pellet the red blood cells. The samples were then read on the SiPM luminometer with a read time of 30 samples.

Lateral Flow assay.

The commercial lateral flow assays were purchased from Nirmidas (MidaSpot COVID-19 Combo Kit). The assays were performed according to the manufacturer’s instructions. The package was opened containing a later flow device, capillary tube, sample diluent bottle, and lancet. The middle finger of the patient was cleaned with an alcohol wipe and left to air dry. A lancet was used to prick the finger and the blood was collected with the capillary tube, which automatically stopped at a finite volume. The capillary with blood was then touched onto the white spot labeled “Sample Well” on the lateral flow device until all the blood was absorbed. Then four drops of the sample diluent buffer were added to the sample well membrane. A picture was taken of the result after 20 minutes and the band intensity was analyzed by FIJI image analysis software.

Dried blood spot anti-Spike antibody ELISA assay.

Dried blood spots (DBS) were collected from volunteers via capillary blood previously described. The blood was collected into an EDTA tube (BD, #365974). 10 μL was pipetted onto cellulose filter paper (VWR, 05-713-336) and let air dry for at least 1 hour. Afterwards, the DBS were stored at -20°C. Before the ELISA assay, the dried blood spots were cut into a circle only containing the blood portion, then cut in half, submerged into an eppendorf tube with 100 μL PBS + 0.05% Tween, and left to gently shake at 4°C overnight (making sure all of the DBS submerged). The next day, the liquid was transferred to a new tube. The ELISA assay was performed as previously described with some modifications [42]. The assay was performed in a 384-well Nunc MaxiSorp flat-bottom plate (Thermo Fisher Scientific), and each sample was run in duplicate. First, plates were coated with 50 μl of 0.5 μg/ml NeutrAvidin mixed in PBS overnight at 4°C. Plates were then washed three times with PBS containing 0.05% Tween 20 (PBST) and were washed similarly for each of the following steps. Next, 20 μl of biotinylated RBD antigen was added to NeutrAvidin-coated wells and allowed to bind for 30 min at room temperature. After washing, plates were then blocked for 60 min with 80 μl 3% nonfat milk (Lab Scientific)–PBST–10 μM biotin. The reconstituted DBS samples were diluted 1:2 with 1% non-fat milk and 20 μl were added to each well for 1 hour at room temperature. The plates were again washed, and antibodies bound to the coated antigens were detected using 20 μl of protein L -HRP (Thermo Fisher Scientific 32420 [1:5,000]) as indicated for 30 min at room temperature. Following a final wash, the plate was developed for 3 to 10 min at room temperature using 20 μl of 50/50 3,3′,5,5′-tetramethylbenzidine (TMB)/solution B (VWR International). Reactions were quenched with 20 μl 1 M phosphoric acid, and absorbance was measured at 450 nm using a Tecan Infinite M200 Pro spectrophotometer.

Determining negative cutoff values for point-of-care assay with whole blood. To determine the cutoff value for S and N sensors, mock negative control blood was constructed. Venous human blood was purchased from Vitalant and 1 mL of blood was washed three times with PBS. The final wash of PBS was removed and pre-pandemic negative control human serum (Sigma Aldrich, before 2020) was added to bring the volume back to 1 mL. This was done for two blood donors and the sample was run on the point-of-care spLUC assay in combination with the handheld luminometer. The cutoff was drawn above the higher of the two negative control values.