Engineering luminescent biosensors for point-of-care SARS-CoV-2 antibody detection thumbnail

Engineering luminescent biosensors for point-of-care SARS-CoV-2 antibody detection

Abstract

Current serology tests for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) antibodies mainly take the form of enzyme-linked immunosorbent assays, chemiluminescent microparticle immunoassays or lateral flow assays, which are either laborious, expensive or lacking sufficient sensitivity and scalability. Here we present the development and validation of a rapid, low-cost, solution-based assay to detect antibodies in serum, plasma, whole blood and to a lesser extent saliva, using rationally designed split luciferase antibody biosensors. This new assay, which generates quantitative results in 30 min, substantially reduces the complexity and improves the scalability of coronavirus disease 2019 (COVID-19) antibody tests. This assay is well-suited for point-of-care, broad population testing, and applications in low-resource settings, for monitoring host humoral responses to vaccination or viral infection.

Main

As the vaccine deployment starts worldwide for COVID-19, broad antibody testing for SARS-CoV-2 faces severe limitations. Although nucleic acid testing is critical to detecting the virus, serological antibody tests are vital tools for monitoring the dynamic human humoral response to SARS-CoV-2 vaccination and viral infection1. Population-scale, longitudinal evaluation of antibody responses is needed to determine the strength and duration of immunity to the primary virus, to the variants, and to vaccines, which is important in informing public policy and vaccination strategies2,3,4,5,6,7. In addition, antibody tests serve as a complement or an alternative to nucleic acid diagnostics for patients with a low viral load or for low-resource areas where expensive reverse transcription polymerase chain reaction (RT–PCR) testing is difficult to access8,9,10. Serological tests also support therapeutic development through identification of either individuals who could serve as donors for convalescent serum therapeutics11, or patients with potentially strong neutralizing antibodies that can be produced in vitro as new antivirals and prophylactics12,13. All these applications would be greatly accelerated with an assay that is simple, rapid and high throughput, without sacrificing accuracy and sensitivity.

Traditional serological assays are not optimal in the face of this broad pandemic. The most widely used laboratory serological tests take the form of enzyme-linked immunosorbent assays (ELISAs)3,14,15,16, which usually entail a more than 2-h protocol involving several steps of protein incubation and washes and is not readily amenable to deployment outside of a laboratory. A faster but considerably more expensive approach is a lateral flow assay17,18. However, lateral flow assays can produce less reliable results depending on the quality of the lateral flow device and different evaluation criteria18,19. In addition, lateral flow tests poorly capture the magnitude of a patient’s antibody response, as the test is qualitative and not quantitative. An alternative is a chemiluminescent microparticle immunoassay (for example, the Abbott AdviseDx and Roche Elecsys SARS-CoV-2 tests), but these assays are expensive, lacking scalability, and typically require specific integrated instrument platforms to perform the tests.

Here we describe a next-generation, simple and rapid assay to meet the mounting needs for broad antibody testing in the face of the ongoing pandemic and the population-wide vaccine deployment. The assay, which is compatible with serum, plasma, whole blood and saliva samples, uses a simple split luciferase (spLUC) antibody sensor to generate quantitative serological data in 30 min. Testing of over 150 patient serum/plasma samples across three validation cohorts demonstrates that the spLUC assay has a sensitivity of 89% for detecting anti-S protein antibodies and 98% for detecting anti-N protein antibodies and a specificity of over 99% for both.

Engineering spLUC biosensors for SARS-CoV-2 antibody detection

When envisioning a next-generation serological assay, we hypothesized that sensitive biosensors for anti-SARS-CoV-2 antibodies could be used to greatly enhance the speed and simplicity of serological testing20. We constructed anti-SARS-CoV-2 antibody biosensors by fusing split nanoluciferase (NanoLuc) fragments SmBiT and LgBiT20 to SARS-CoV-2 viral protein antigens (Fig. 1a). Because an antibody has two fragment antigen-binding (Fab) arms, incubating serum with 1:1 mixed SmBiT and LgBiT biosensors will result in half of the antiviral antibodies binding LgBiT with one Fab arm and SmBiT with the other Fab arm. This hetero-bivalent interaction localizes the LgBiT and SmBiT fragments in close proximity, resulting in reconstitution of an intact, active NanoLuc enzyme for luminescence-based detection of reactive antibodies.

Fig. 1: Engineering luminescent biosensors for rapid and quantitative detection of SARS-CoV-2 antibodies.
figure1

a, Schematic of the solution-based serology assay. Patient antibodies are incubated with SARS-CoV-2 S or N proteins fused to LgBiT/SmBiT. For the population of antibodies with one arm bound to the LgBiT sensor and the other arm bound to the SmBiT sensor, the NanoBiT luciferase enzyme is reconstituted and, thus, can produce active luciferase signal. b, Dose-dependent spLUC signals for the recombinant anti-S-RBD antibody C004 in PBST + 10% FBS. Two technical replicates are plotted from n = 1 independent experiment. c, Dose-dependent spLUC signals for an anti-N-RBD antibody (Sino Biological, cat. no. 40588-T62–50) in PBST + 8% FBS. Two technical replicates are plotted from n = 1 independent experiment. d, Comparison of assay procedures between the ELISA and the spLUC assay. While ELISA takes more than 2 h and involves multiple wash and incubation steps, the spLUC solution-based assay is simply completed in 30 min or less without the need for wash steps. e, The S (L15 + S25) sensors are able to detect antibodies in 5/5 recovered patients with COVID-19. At all dilutions tested, all five patients generated signal above the background signal of two control serum samples collected before the pandemic. Each dot represents a technical replicate. n = 2 independent experiments with three replicates each are plotted for all samples, except for those of Patient 1, Patient 7 and Control 2, which have n = 1 independent experiment plotted owing to limited reagents. f, The N (LC + SC) sensors are able to detect antibodies in 4/4 recovered patients with COVID-19. At all dilutions of serum tested, all four patients generated signal above the background signal of two control serum samples collected before the pandemic. Two technical replicates are plotted from n = 1 independent experiment. g, Patient antibodies for SARS-CoV-2 have various epitopes on the S-RBD (red). C004 and C105 have ACE2-competitive epitopes, whereas C135 and CR3022 (blue) have non-ACE2-competitive epitopes. h, S sensors can detect patient antibodies of various epitopes with similar sensitivity. C004, C105, C135 and CR3022 patient antibodies were incubated with the S sensors at ten-fold antibody dilutions from 10 nM to 0.001 nM. The average of three technical replicates from n = 2 independent experiments are plotted. i, Schematic of antibody epitope competition assay with patient serum samples. Direct signal is compared to signal generated in the presence of the pre-incubated 1 µM Fab +1 nM sensor. j, Competition assay performed with C135 Fab on 12 outpatient sera samples and recombinant C135 IgG protein. Samples were incubated with either no Fab (blue) or C135 Fab (off-white). Sera 7, 42 and 98 showed more than 50% decreases in luminescence signal, suggesting the presence of antibodies with the C135 epitope. Each dot represents a technical replicate from n = 1 independent experiment. The center of the bar represents the mean of the measurements. Direct signals (−Fab) were measured in technical duplicates, and competition signals (+C135 Fab) samples were done in technical triplicate. For b, c, e, f and h, the center of the line represents the mean of all measurements. Lines connecting the means of the samples are plotted. RLU, relative luminescence unit; RT, room temperature; NFM, non-fat milk; O/N, overnight.

Source data

We chose to develop S and N sensors for SARS-CoV-2 antibody tests because COVID-19 patient antibodies are predominantly directed against epitopes on the viral S protein, which interacts with the host receptor angiotensin-converting enzyme 2 (ACE2) and mediates viral entry21, and the N protein, which packages the viral genome into a ribonucleocapsid22. These two viral proteins are the primary antigens used in the current COVID-19 serological tests3,14,23,24,25,26.

The S sensors were constructed by fusing NanoLuc fragments to its receptor-binding domain (S-RBD), which is the primary target of neutralizing antibodies (Extended Data Fig. 1a,b)3,14,26,27. We modeled S-RBD binding to two antibodies—C105 (refs. 12,28), an ACE2-competitive binder, and CR3022 (ref. 29), an ACE2 non-competitive binder—to determine linker lengths (Supplementary Text and Extended Data Fig. 1c,d). Based on the models, we constructed SmBiT fusions to the S-RBD C terminus with 15 or 25 residue glycine/serine (GS) linkers (S15 and S25) and LgBiT fusions to S-RBD C terminus with 5, 15 or 25 residue GS linkers (L5, L15 and L25). These variants varied in expression yields (Extended Data Fig. 1e). Using recombinantly expressed S-RBD antibodies and ACE2 variants30, we determined the optimal linker variant, enzyme concentration, buffer conditions and effect of antibody–antigen binding affinity to signal strength (Supplemental Text, Extended Data Figs. 1 and 2 and Supplementary Fig. 1). The (L15 + S25) sensor pair at 1 nM enzyme concentration was identified as optimal, and these conditions were used for all subsequent assays.

In further characterizing the relationship between assay signal strength and antibody concentration/binding affinity, we performed ordinary differential equation modeling in R (Supplemental Text and Extended Data Fig. 3). The modeling predicted a linear relationship between antibody concentration and luciferase signal (Extended Data Fig. 3b), consistent with our experimental data (Fig. 1b). In addition, the results highlighted that the sensors at 1 nM are more sensitive to an antibody binder with a KD ≤ 1 nM (Extended Data Fig. 3b,c). This threshold is equivalent to the median affinity reported for polyclonal antibody repertoires31,32.

To construct the N sensors, we used the N-terminal sequence because amino acids (aa) 44–257 are found to be more immunogenic than the C-terminal dimerization domain (aa 258–419) (Extended Data Fig. 4a)33. In addition, dimerization promoted by the C-terminal domain might lead to high basal NanoLuc reconstitution levels. The atomic structures of N44–18022 showed that the N and C termini are not in close proximity, and, therefore, fusion at the N or C terminus might result in different sensor sensitivity (Extended Data Fig. 4b). Given this knowledge, three fusion sensor pairs were designed: (1) LN + SN: L/S–N44–257; (2) LC + SC: N44–180–L/S; and (3) LC2 + SC2: N44–257–L/S, where L and S represent LgBiT and SmBiT, respectively, C represents C-terminal fusion and N represents N-terminal fusion (Extended Data Fig. 4c). Testing on a commercial polyclonal anti-N protein antibody revealed that the LC + SC and LC2 + SC2 sensors generated stronger signals over LN + SN (Extended Data Fig. 4d). The LC + SC sensors generated linear, dose-dependent signals with commercial anti-N protein antibody (Fig. 1c).

We next designed a simple and rapid protocol to assay a pilot set of serum samples from convalescent patients with SARS-CoV-2 (Fig. 1d). Two healthy control sera collected before the emergence of the SARS-CoV-2 virus were also tested. Serial dilutions (1:12.5, 1:25 and 1:50) of heat-inactivated sera were measured using S or N sensors. Robust, dose-dependent luminescence signal was observed across all serum concentrations tested, with the 12.5-fold dilution showing the highest signal (Fig. 1e,f). The S (L15 + S25) sensors generated signal for all five patients tested (Fig. 1e). The N (LC + SC) sensors detected patient antibodies from all four patients tested (Fig. 1f). However, the N (LN + SN) sensors detected antibodies from only two patient sera samples that had the strongest seropositivity (Extended Data Fig. 4e), which further confirmed that a C-terminal fusion enhances NanoLuc reconstitution relative to the N-terminal fusion. Interestingly, much higher signals were generated with the N (LC + SC) sensor compared to the S sensor. This could be due to differing levels of anti-N or S antibodies in the patients34, the affinity of antibodies, or a favored geometry for split enzyme reconstitution when anti-N antibodies bind to the N (LC + SC) sensors.

Competitive spLUC assay to profile epitope classes of antibodies

In addition to a test to determine total binding antibodies, an assay that allows profiling of epitope classes of antibodies can be highly valuable. In this regard, competitive ELISA assays developed by us and others have enabled characterization of percentage of ACE2-competitive antibodies26,35. However, S-RBD is known to have multiple additional neutralization epitopes outside of the ACE2-binding site. An assay that allows rapid, unbiased profiling of those alternative epitopes could unveil further details of a patient’s humoral response to neutralize SARS-CoV-2.

We first showed that the spLUC assay can detect antibodies binding to various S-RBD epitopes (Fig. 1g). We expressed and tested four reported neutralizing antibodies against S-RBD. This includes C004 and C105 (ref. 12), which are ACE2-competitive binders; CR3022 (ref. 29), which binds at a cryptic site outside of the ACE2-binding site; and C135 (ref. 12), which does not compete with C004, C105, CR3022 or ACE2-Fc, representing a third binding epitope on S-RBD (Supplementary Fig. 2). All four IgG antibodies generated dose-dependent luminescence signals at ≥0.1 nM concentrations (Fig. 1h).

We then designed a competitive spLUC assay to determine thepresence of a specific epitope class of antibodies (Fig. 1i). Of the four antibodies tested, C135 represents an unconventional and less understood epitope class. It neutralizes very potently (IC50 = 17 ng ml−1) and might be used in combination with other ACE2-competitive binders as a cocktail therapy12. We converted C135 IgG to a single binding arm Fab binder and pre-incubated 1 µM of C135 Fab with the S sensors to generate ‘blocked sensorsʼ. We then determined how much signal from a patient’s sample corresponds to antibodies with a C135 epitope by comparing signals between the original and the ‘epitope-maskedʼ sensors (Fig. 1i). We assayed 12 patient serum samples with representative high, medium and low anti-S-RBD antibody levels at a 1:25 dilution of serum. IgG C135 served as a control for competition with Fab C135. As expected, the luminescence signal of IgG C135 was reduced by ~90% with the blocked sensors. Sera 7, 42 and 98 showed more than 50% decreases in luminescence signals, indicating that a large portion of the antibodies in these samples are C135 competitive (Fig. 1j). These results suggest that antibodies recognizing this unconventional, neutralizing S-RBD epitope are present in a substantial proportion of patient samples. Performing this competitive spLUC assay with different competitive Fab antibodies in an expanded patient cohort could further understanding of the distribution of epitopes on S-RBD and the correlation between epitopes and clinical outcomes.

Characterization of larger cohorts of serum/plasma samples using the spLUC assay

We next applied the spLUC assay in an expanded number of patients (Fig. 2). First, to determine assay cutoff values and specificity, which reflects how well an assay performs in a group of disease-negative individuals, we performed the tests on three cohorts of negative control samples (total n = 144), which included mainly healthy individual samples, 12 seasonal coronavirus patient samples and 20 flu vaccine pre- and post-vaccination samples. All controls were collected before the SARS-CoV-2 pandemic. These controls generated significantly lower luminescent signals than the COVID-19 patient sera samples (Fig. 2a,b). The range, median, mean and standard deviation values were calculated, and stringent cutoff values were determined by calculating the mean plus three standard deviations (Supplementary Table 1). With these determined cutoffs, we calculated the specificity of the S sensors (1:12.5 serum dilution) to be 100% (56/56) and the N sensors (1:12.5 serum dilution) to be 99.2% (119/120).

Fig. 2: Characterization of outpatient and inpatient serum samples using the spLUC test.
figure2

Cohort 1: samples drawn during the convalescent phase of an outpatient group; Cohort 2: samples drawn during the acute phase or the convalescent phase of a hospitalized group; and Cohort 3: samples drawn during the convalescent phase of a mixed inpatient and outpatient group. A 10-base logarithmic scale conversion was applied to all the solution assay signals for the correlation analysis, unless otherwise specified. a, SpLUC assay tested on expanded COVID-19 patient cohorts with S sensors at 1:12.5 serum dilution. Dots represent the average between two technical duplicates. Lines represent median values. The inpatient samples showed significantly higher antibody titers than the outpatient cohorts. Sample sizes are as indicated in parentheses: Control (56), Cohort 1 (57), Cohort 2 (55), Cohort 3 outpatient (47) and Cohort 3 inpatient (9). b, SpLUC assay tested on expanded COVID-19 patient cohorts with N sensors at 1:12.5 serum dilution. The inpatient samples showed significantly higher antibody titers than the outpatient cohorts. Sample sizes are as indicated in parentheses: Control (120), Cohort 1 (56), Cohort 2 (47), Cohort 3 outpatient (47) and Cohort 3 inpatient (9). c, A positive correlation (R = 0.78) was observed between S sensor signal and N sensor signal in the three cohort samples. All cohorts individually presented a similar trend (Extended Data Fig. 5). The line represents linear regression. In total, 159 patient samples are plotted. d, Correlation of spLUC signals (Cohort 1) to neutralization efficiency12. S sensor signal (blue) and N sensor signal (purple) are plotted against 50% maximal neutralization titer (NT50). Both show positive correlation (R = 0.76 for S and NT50 and R = 0.62 for N and NT50). Fifty-seven patient samples are plotted for the S sensor, and 56 samples are plotted for the N sensor. e, Inpatients show significantly higher signal than outpatients in all three cohorts (P < 0.0001). Sample sizes are as indicated in parentheses: S-outpatient (104), S-inpatient (64), N-outpatient (103) and N-inpatient (56). f, Patients from Cohort 1 who reported higher disease severity (6–10 versus 1–5) had higher antibody titer for both S and N sensors, and the difference for N sensors is statistically significant (P = 0.0049). Sample sizes are as indicated in parentheses: S-score 1–5 (24), S-score 6–9 (33), N-score 1–5 (24) and N-score 6–9 (32). g, Higher overall antibody titers were observed in patients who reported fever compared to no-fever patients for Cohort 3. This difference was statistically significant for the S sensors (P = 0.0011) but not for the N sensors. Seventeen patient samples associated with no fever and 38 patient samples associated with a fever are plotted. h, Slightly higher overall antibody titers were observed in females than in males for Cohort 3, although the differences were not statistically significant. There is a similar trend for Cohort 1 (Supplementary Fig. 3a). The difference was more obvious for S sensors. Twenty-five female patient samples and 30 male patient samples are plotted. i, For Cohort 3, there is a slightly higher level of antibodies in the 60–85 age group compared to 19–39 and 40–59. There is a similar trend for Cohort 1 (Supplementary Fig. 3b). The differences were not statistically significant. Sample sizes are as indicated in parentheses: age 19–39 (23), age 40–59 (25) and age 60–85 (8). For a and b, P values (two-tailed) of a Kruskal–Wallis test with Dunn’s multiple comparison post hoc testing to compare each clinical cohort with the healthy control group are indicated. For c and d, R values and P values (two-tailed) of a non-parametric Spearman correlation analysis are labeled in the graphs. For ei, an unpaired Mann–Whitney test is performed, and P values (two-tailed) for each comparison are labeled on top of the datasets. For all panels, dots represent the average of two technical replicates from n = 1 independent experiment. For a and b and ei, horizonal lines represent median values. For c and d, lines represent linear regression. RLU, relative luminescence unit.

Source data

We then used the spLUC assay to study three additional cohorts of patient samples (Fig. 2a,b). Cohort 1 is an outpatient cohort recruited at the Rockefeller University Hospital12. The samples were collected from individuals free of COVID-19 symptoms for ≥14 d. The S sensors showed 84.2% (48/57) sensitivity, and the N sensors showed 100% (56/56) sensitivity. Cohort 2 samples consisted of remnant sera from patients with COVID-19 at Kaiser Permanente Hospitals of Northern California. These samples were drawn in any phase of infection, including the early acute phase. A subset of these patients, who might have not fully seroconverted at the time of sampling, had lower S sensor or N sensor signals than others in the spLUC assays. The sensitivities of the assays were 89% (49/55) for S sensors and 98% (46/47) for N sensors. Cohort 3 patients were part of the LIINC (Long-term Impact of Infection with Novel Coronavirus) study from San Francisco General Hospital and included plasma of a mixture of outpatient and inpatient samples drawn in the convalescent phase of the disease. With the S sensors, we detected antibodies in 94% (44/47) of outpatient samples and 100% (9/9) of inpatient samples. With the N sensors, we detected antibodies in 96% (45/47) of outpatient samples and 100% (9/9) of inpatient samples. For all cohorts, the S and N signals show a strong correlation (Fig. 2c and Extended Data Fig. 5). Consistent with previous findings, we observed varying degrees of anti-S and N antibody seropositivity among patients (Fig. 2a,b), which reflects a wide range of patient humoral response to this virus8,2.

Notably, we observed strong correlation of spLUC assay results to anti-Fab and anti-IgG S-RBD ELISA signals (Extended Data Fig. 6a–c; R = 0.43–0.91). A base-10 logarithmic scale conversion was applied to the spLUC assay signals for the correlation analysis to ELISA signals. This non-linear correlation between the spLUC and ELISA assays is likely due to signal compression in ELISAs at high antibody concentrations (Abcam, ELISA guide). For all cohorts, the S sensor seronegative samples also had very low signals in S-RBD ELISA assays (Extended Data Fig. 6d–f), which confirmed the presence of low levels of anti-S-RBD antibodies in these sub-cohorts of patients. Interestingly, the correlations to IgM signals were much weaker (Extended Data Fig. 6a,b). It is possible that IgM was not sensitively detected by the spLUC assay owing to the weaker affinities of the individual binding arms in IgMs36 or that the IgG response dominated the signal in many of the tested patients.

One of the key uses of a highly sensitive serology assay is to grade the quality of convalescent sera to neutralize virus1. In Cohort 1, our analysis showed that the S sensor signals correlated with the half-maximal neutralizing titers reported by Robbiani et al. (Fig. 2d, left panel), consistent with previous studies4,12,14,37. Moreover, we found that the N sensor signals showed a similar correlation with half-maximal neutralizing titer (Fig. 2d, right panel). Our results indicate that determining either anti-S or anti-N seropositivity is a general means to assess the neutralization potential of sera samples.

We further analyzed our data in the context of demographic and clinical features. First, the degree of seropositivity for inpatient samples was significantly higher than that of outpatient samples (Fig. 2a,b,e). Disease severity scores and fever were also associated with a stronger antibody response (Fig. 2f,g). These results indicated a direct correlation of disease severity and adaptive immune response, consistent with previous studies2,4,8,10,12,38,39. In addition, no statistical differences were observed in antibody titers between males and females (Fig. 2h and Supplementary Fig. 3a) or for different age groups (Fig. 2i and Supplementary Fig. 3b). We expect that analyzing a larger number of samples can further unveil if infected older adults develop more antibodies than younger individuals, as suggested in previous studies18,19. These results highlight that clinical features are correlated with the antibody response of patients with COVID-19.

Collectively, our assay showed high sensitivity and specificity for all three representative cohorts of serum/plasma samples (including inpatient, outpatient, acute phase and convalescent phase samples), with an overall specificity of 100% (S sensor) and 99% (N sensor) and sensitivity of 89% (S sensor) and 98% (N sensor). These values are similar or superior to reported values for laboratory ELISA and lateral flow tests18,19.

Additionally, we sought to determine the inter-day and intra-day assay variability. To further reduce labor work, we used a liquid handler system to dispense plasma, sensors and substrate. We characterized a set of 46 convalescent plasma samples multiple times over 2 d to determine assay variation (Extended Data Fig. 7a). We observed average intra-assay coefficient of variations (CVs) of 6% for both S and N sensors, intra-day CVs of 7% for S sensors and 9% for N sensors and inter-day CVs of 8% for S sensors and 9% for N sensors (Extended Data Fig. 7b).

The simplicity of the assay can greatly improve laboratory antibody test throughput. From our experience, one individual can run four plates of 96 samples (in total, 384 assays) per hour by hand with multi-channel pipettes. The automation of the assay is also straightforward as no wash step is involved. We showed that individual steps can be assisted with liquid dispenser instruments (Extended Data Fig. 7a). We further set up a workflow for full automation of this assay using the University of California, San Francisco (UCSF) Antibiome Center robotics platform (Extended Data Fig. 8a,b)40, which will further improve assay throughput. A simulated run for 40 plates (3,840 assays) was estimated to take 3 h (Extended Data Fig. 8c). This robotic platform opens up the opportunity for characterizing hundreds of thousands of samples for population-wide epidemiology and longitudinal studies.

Adapting the assay for low-resource settings and expanded sample types

Lastly, we adapted our assays to begin to meet the clinical needs in remote and low-resources settings and for point-of-care deployment. A target assay profile was created based on a World Health Organization consensus meeting report on high-priority target product profiles for new tuberculosis diagnostics (Supplementary Table 2)41. Although the current properties of the assay meet most of the assay profile requirements, we tested to see if the reaction time (30 min), reagent format (frozen aliquots of sensors), luminescence instrument type (plate-based luminometer) and sample type (serum/plasma) could be further optimized.

We first tested if our initial reaction times (20-min sensor incubation and 10-min substrate incubation; Fig. 1d) are necessary and optimal. CR3022 (10 nM) was incubated with 1 nM S sensors for 5, 10, 15 and 20 min, followed by luciferase substrate addition and incubation for 0, 2, 4, 6, 8 and 10 min (Fig. 3a). We found that all time points resulted in similar, bright luminescence signal, suggesting that the time of the assay protocol can be further shortened.

Fig. 3: Adapting the assay for whole blood and saliva sample types.
figure3

a, spLUC assays can be accomplished in as few as 5 min. CR3022 (10 nM) was incubated with S sensors for 5, 10, 15 or 20 min. Luciferase substrates were then added and incubated with the reaction mix for 0, 2, 4, 6, 8 or 10 min. All reactions showed bright luminescence signal. Each dot represents the average of two technical duplicates from n = 1 independent experiment. Lines connecting the means of the samples are plotted. b, The spLUC assay is compatible with whole blood samples and shows similar signal in the corresponding plasma samples with both fresh and lyophilized sensors. Each dot represents the average of two technical replicates from n = 1 independent experiment. A non-parametric Spearman correlation analysis was performed, and R = 0.94 was observed for S sensors, and R = 1 and 0.98 were observed for N sensor fresh and lyophilized sensors, respectively. c, Anti-S antibodies were detected in saliva samples with moderate sensitivity (33/42, 79%). The signals from saliva samples positively correlated with corresponding serum samples. Each dot represents the average of two technical replicates from n = 1 independent experiment. A non-parametric Spearman correlation analysis was performed, and the R value (0.66) and P value (<0.0001, two-tailed) are labeled in the graphs. The line represents linear regression.

Source data

We then tested if the sensors can be lyophilized for ambient temperature storage and transportation. Although a small quantity (0–30%) of S sensors and N sensors were lost due to the lyophilization process (Extended Data Fig. 9a), both the lyophilized S and the N sensors can still robustly detect recombinant IgG or patient antibodies in serum with similar sensitivities seen for the fresh sensors (Extended Data Fig. 9b,c). Furthermore, we found that vacuum drying of the commercial liquid form of the NanoLuc substrate, furimazine, and lyophilization of the substrate dilution buffer did not affect the signals of the assay (Extended Data Fig. 9d).

As plate readers might be difficult to access at point-of-care locations or in the field, we tested a battery-supported portable luminometer (32526-11 Junior LB9509, Berthold Technologies) and found that it detected the same concentration range of CR3022 compared to a Tecan M200 infinite plate reader (Extended Data Fig. 10). The device uses a tube to read samples and, therefore, requires twice or more of the sample volume than in a 384-well plate, which (~2 µl of serum/plasma) is still compatible with the amount of sample generated from a finger prick. Together, our data indicate that both the reagents and instruments required for the spLUC assay are well suited for point-of-care or field applications.

Finally, we sought to determine if the spLUC assay could be compatible with other sample types. First, whole blood samples were collected from six convalescent patients with COVID-19, and plasma samples were prepared in parallel for comparison (Fig. 3b). Remarkably, although the overall signals were lower from whole blood samples, all six samples generated N sensor signals, and four had S sensor signals above control levels with the lyophilized sensors. In comparison, all six patients generated N sensor signals, and five had S sensor signals above cutoff values from the plasma samples. Strong correlations were observed between the whole blood signals and the plasma signals (R > 0.9). Fresh and lyophilized sensors showed very little difference in performance.

Next, we tested the potential of using saliva as an input. To determine conditions, we added varying concentrations of the CR3022 antibody into saliva from a healthy individual (Supplementary Fig. 4). We saw a significant reduction in sensitivity for undiluted saliva relative to buffer alone but, remarkably, no loss in sensitivity when the saliva was diluted 1:2 in PBS buffer. We then tested 42 saliva samples at 1:2 dilution with the S sensors. We increased the reaction volume from 20 µl to 100 µl and the luminescence signal integration time from 1,000 ms to 5,000 ms for better sensitivity, as lower antibody concentrations are expected from saliva samples42. Of the 42 samples, 33 had signals above the two healthy saliva controls, indicating a 79% assay sensitivity (Fig. 3c). A moderate correlation of saliva signal with corresponding serum signals was observed (R = 0.66), consistent with recent reports43. These results highlight the potential of using lyophilized sensors/substrates and whole blood or saliva samples for rapid and quantitative point-of-care antibody testing.

Discussion

As SARS-CoV-2 continues to spread, the need will continue to grow for serology assays to determine not only the scope of infection but also vaccine efficacy during clinical trials and after large-scale vaccine deployment. We present here spLUC, a simple, sensitive, specific, fast, low-input sample volume and quantitative solution-phase serological assay to detect antibodies against S and N proteins. We were able to test 159 patient samples across three different cohorts with varying clinical and demographic features. Our results enabled association analysis between these features and antibody titers, demonstrating the promise of this assay to generate large datasets to better understand factors that modulate the humoral response after SARS-CoV-2 infection.

The quantitative and solution-based nature of the spLUC assay allows convenient assay variations. We presented a competitive spLUC assay using epitope-masked S sensors and used it to study the prevalence of an unconventional neutralization epitope in the S-RBD domain. This competitive spLUC assay has the potential to serve as a surrogate virus neutralization assay and to unveil details of the interaction of patient antibodies to viral antigens.

Robust ELISA-based assays, such as the one developed by Krammer and colleagues, have enabled tremendous progress of COVID-19 serological studies14,24, but these assays are still laborious with multiple wash steps, which limits their broad application for population-scale sero-surveillance, point-of-care diagnostics and deployment in countries or remote areas that have limited access to analytical equipment and reagents. The spLUC assays have important features amenable to these applications. We have shown that our reagents are not only compatible with lyophilization for easy transport and storage but also can readily detect antibodies directly from whole blood samples and saliva samples. With simple pipettes and a battery-supported portable luminometer, the spLUC assay could be readily established at care centers or in the field worldwide, regardless of infrastructure. To this end, we showed that a commercially available portable luminometer is compatible with this assay, and we are currently collaborating with bioengineers to further develop luminometers that can be manufactured at lower cost but provide equal or better detection sensitivity. We have also shown that the assay is amenable to automation with standard liquid dispenser instruments, which allows for an opportunity to set up automated stations for self-testing at point-of-care locations.

Another important strength of our approach is the modularity. We expect that our strategy can be readily adapted to develop rapid serological tests for immunity against any infectious or autoimmune disease that elicits an antibody response for which the protein antigen is known. Our work identified important protein engineering details in constructing sensitive biosensors against anti-coronavirus antibodies, such as using the RBD domain of the S protein, removing the dimerization domain from the N protein and fusing split enzymes to the C terminus of the N protein. These sensor designs are instructive for a rapid response if we have to face a future pandemic caused by a new strain of coronavirus. Moreover, the spLUC constructs are also nimble for engineering variant RBD domains to assess antibody response to new viral variants that are emerging. These should be useful for determining vaccine effectiveness in communities having multiple viral strains. Future development of our spLUC assay includes exploring orthogonal split enzyme systems to allow multiplexing of assays. For instance, split ß-lactamase, used by Huang and colleagues for detecting herpes simplex virus antibodies44, can provide an orthogonal readout to luminescence. We envision that such multiplexed assays could be used to develop broad-spectrum serological assays to simultaneously detect immunity against multiple infectious diseases.

In summary, we took a structure-based protein engineering approach to design novel split enzyme-fused sensors. These biosensors enable spLUC, a next-generation SARS-CoV-2 antibody test suited for population-scale sero-surveillance, epitope mapping of patient antibody responses and testing in resource-limited areas. Future efforts will focus on continued evaluation of alternative sample sources, collaboration with industry and clinical labs for broad deployment of the assay and development of similar split enzyme-based serological approaches for a range of infectious diseases.

Methods

Plasmid construction

Plasmids were constructed by standard molecular biology methods. The DNA fragments of Spike-RBD, N protein, ACE2 and LgBiT were synthesized by IDT Technologies. The SmBiT tag was generated by overlap extension PCR. The Spike-RBD-5/15/25aa-LgBiT-12xHisTag, Spike-RBD-15/25aa-SmBiT-12xHisTag, N protein(44–180)-10aa-LgBiT-12xHisTag, N protein(44–180)-10aa-SmBiT-12xHisTag, LgBiT-10aa-N protein(44–257)-12xHisTag and SmBiT-10aa-N protein(44–257)-12xHisTag were generated by subcloning into a pFUSE-12xHisTag vector (adapted from the pFUSE-hIgG1-Fc vector from InvivoGen). The ACE2-Fc fusion plasmids were generated by subcloning the gene fragments of ACE2 and mutant into the pFUSE-hIgG1-Fc vector. The C004, C105 and C135 IgG LC and HC plasmids were a generous gift from the Nussenzweig lab (Rockefeller University). The CR3022 IgG plasmids were a generous gift from the Kim lab (Stanford University) and the Wilson lab (Scripps Research Institute). The C135 Fab was cloned by removing the Fc domain from the HC plasmid. SnapGene and ApE were used for cloning design and sequence verification.

Expression and protein purification

All proteins were expressed and purified from Expi293 BirA cells according to established protocols from the manufacturer (Thermo Fisher Scientific). Briefly, 30 μg of pFUSE (InvivoGen) vector encoding the protein of interest was transiently transfected into 75 million Expi293 BirA cells using the ExpiFectamine kit (Thermo Fisher Scientific). For the IgG and Fab proteins, 15 μg of each chain was transfected. Enhancer was added 20 h after transfection. Cells were incubated for a total of 3 d at 37 °C in an 8% CO2 environment before the supernatants were harvested by centrifugation. Fc-fusion proteins were purified by protein A affinity chromatography, and His-tagged proteins were purified by Ni-NTA affinity chromatography. Purity and integrity were assessed by SDS/PAGE. Purified protein was buffer exchanged into PBS and stored at −80 °C in aliquots. Concentration was verified by absorbance on a NanoDrop with NanoDrop 2000c software.

Solution serology protocol for in vitro, serum, blood, and saliva samples

LgBiT and SmBiT sensors for either the Spike or N protein were prepared at a final concentration of each sensor at 2 nM in PBS + 0.05% Tween-20 + 0.2% BSA (PBSTB). For in vitro IgGs or ACE2-Fc, the samples were prepared at 1:10 dilutions in PBSTB unless otherwise specified. Serum and blood samples were diluted to 1:12.5 for both the S and N sensor samples in PBSTB unless otherwise specified. Healthy individual saliva was spiked in with CR3022 and used undiluted or diluted 1:2 in PBSTB. Then, 10 μl of the 2 nM sensor mix and 10 μl of the sample were combined in a 384 Lumitrac white plate (Greiner), skipping every other well and row to avoid potential bleedover in signal. The plate was mixed on a plate shaker for 20 min. NanoLuc substrate was diluted according to protocol 1:50 in NanoLuc dilution buffer (Promega), and 15 μl was added to each well, followed by a 10-min incubation period for the signal to stabilize. Luminescence was measured on a Tecan M200 infinite plate reader with an integration time of 1,000 ms. Tecan i-control plate reader software was used for data acquisition.

Competition serology protocol for in vitro and serum samples

The competition serology assay was performed similarly to the solution serology assay except that the S sensors were individually pre-incubated at 4 nM with 4 μM of C004 Fab, C105 Fab or C135 Fab for the in vitro competition assay and C135 Fab only for the serum competition assay. The two sensors + Fab were combined 1:1 to make a 2 nM mix, and 10 μl of this mix was added to the assay as described above.

Epitope binning experiment

Biolayer interferometry data were measured using an Octet RED384 (ForteBio). Biotinylated Spike RBD protein was immobilized on the streptavidin biosensor (ForteBio). After blocking with biotin, the sensor was loaded with one IgG followed by another IgG or ACE2-Fc to determine epitope binning. PBS with 0.05% Tween-20 and 0.2% BSA was used for all diluents and buffers. ForteBio Octet Data Acquisition software was used for acquiring data, and ForteBio Octet Data Analysis software was used for analyzing data.

Spike protein ELISA assay

The Spike ELISA assay was performed as previously described. Briefly, 384 MaxiSorp plates were coated with 100 μl of 0.5 μg ml−1 NeutrAvidin for 1 h. The plate was washed three times with PBS + 0.05% Tween-20 (PBST) followed by incubation with 20 nM S-RBD for 30 min. After three washes, the plate was blocked with 3% non-fat milk in PBS for 1 h. The plate was washed three times before the addition of 1:50 dilutions of serum in 1% non-fat milk for 1 h. After three washes, secondary anti-human Fab (Jackson ImmunoResearch Laboratories, 109-036-097 (1:5,000)), anti-human IgM (Sigma-Aldrich, A6907 (1:3,000)) or anti-human IgG (Sigma-Aldrich, A0170 (1:3,000)) antibody was added and incubated for 30 min before the addition of TMB for 3 min. The reaction was quenched with 1 M phosphoric acid, and absorbance was read on a Tecan M200 infinite plate reader at 450 nm.

Inter-day and intra-day assay with liquid dispensers

Plasma was diluted 12.5-fold into a 96-well plate, and, subsequently, 10 µl was dispensed into an individual flat-bottom white plate (Greiner Lumitrac 200, 384-well plates) using a Biomek FXP Automated Workstation (Beckmann Coulter). Next, 10 µl of 1 nM biosensors was dispensed using Thermo Multidrop Combi Reagent Dispenser (Thermo Fisher Scientific) to assay plates and incubated at room temperature for 20 min. Then, 15 µl of substrate was added using the same reagent dispenser and incubated at room temperature for 10 min. Luminescence was read on a Tecan M200 infinite plate reader with an integration time of 1,000 ms.

Simulated automation of the spLUC assay on a robotic platform

To facilitate high-throughput screening of serum, a semi-automated approach was developed and simulated using the UCSF Antibiome Center robotics platform40. Serum in 96-well plates is first diluted 12.5-fold, and 10 µl is dispensed into four individual flat-bottom white plates (Greiner Lumitrac 200, 384-well plates) using a Biomek FXP Automated Workstation (Beckmann Coulter). Serum-containing assay plates are then transferred to a robotics protocol with dispensing of biosensor and substrate followed by luminescence reading. Although one iteration of 96 samples takes 40 min, each additional iteration takes an additional 3.5 min, limited by luminescence reading (1 s per well plus plate transfer). As such, it is estimated that 40 plates (3,840 assays) could be run in 3 h. The robotics run was developed and simulated using Thermo Momentum software (v5.0.6).

Lyophilization of sensors and dilution buffer

The S and N protein sensors were flash frozen in liquid nitrogen at concentrations between 10 μM and 60 μM in 10 μl. The dilution buffer was frozen in liquid nitrogen in 5-ml aliquots. A small hole was poked into the caps of the samples and left on a BenchTop K (VirTis) lyophilizer overnight. The next day, the sensors were reconstituted in 10 μl of double-distilled water (ddH2O), and concentration was verified by NanoDrop. The dilution buffer was similarly reconstituted in 5 ml of ddH2O.

Vacuum-dried centrifugation of substrate sample

Next, 20 μl of substrate was aliquoted in a dark Eppendorf tube and subjected to vacuum centrifugation overnight on a Genevac instrument. The substrate was stored in the dark for 2 d at room temperature before reconstitution in 20 μl of 100% methanol. The substrate was diluted 1:50 as normal in dilution buffer for the assay.

Serum, plasma, whole blood and saliva samples

The initial small patient cohort was a generous gift from the Wilson lab (UCSF) and heat inactivated at 56 °C for 1 h before storage at −80 °C. The first (outpatient) sample serum set (Cohort 1) was a generous gift from the Wilson lab (UCSF) and the Nussenzweig lab (Rockefeller University). These samples were heat inactivated at 56 °C for 1 h and stored at 4 °C in a 1:1 dilution in 40% glycerol, 40 mM HEPES (pH 7.3) and 0.04% NaN3 in PBS. The second (inpatient) sample serum set (Cohort 2) was a generous gift from the Wang lab (Stanford University) and was stored at −80 °C as pure serum samples. The third plasma cohort (Cohort 3) and blood samples were generous gifts from the Greenhouse lab (UCSF) and the Henrich Lab (UCSF) as part of the LIINC study. The plasma samples were stored at 4 °C in a 1:1 dilution in 40% glycerol, 40 mM HEPES (pH 7.3) and 0.04% NaN3 in PBS. The whole blood was stored undiluted at 4 °C. Healthy blood samples were purchased from Vitalant and stored undiluted at 4 °C. The saliva samples were obtained as unstimulated, unexpectorated saliva and were stored at −80 °C. Before being assayed, the samples were thawed and centrifuged at 9,000g to remove any insoluble or coagulated matter. Control saliva from November 2019 was purchased from Lee Biosolutions, stored at −20 °C and processed similarly.

Data and statistical analysis

All graphing and statistical analysis were performed in GraphPad Prism or Microsoft Excel. The unpaired multiple t-test was performed in Prism to compare conditions in each patient for the competition spLUC assay. The non-parametric Spearman correlation analysis was used in Prism to determine the correlation R value between datasets. When two groups were compared, an unpaired Mann–Whitney test was performed to determine the difference between datasets. For comparison of three or more groups, a Kruskal–Wallis test with Dunn’s multiple comparison post hoc testing was used. A two-tailed P value was used to determine statistical significance for all analysis. P < 0.05 was considered statistically significant.

Ethics declaration

All patient samples were obtained under protocols approved by the institutional review boards (IRBs) and in accordance with the Declaration of Helsinki. Samples were de-identified before delivery to the lab where all assays described here were performed. Cohort 1 samples were a kind gift from M. Nussenzweig, M. Caskey and C. Gaebler of Rockefeller University, collected under Rockefeller IRB protocol DRO-1006. Cohort 2 samples from Kaiser Permanente were collected under Stanford University IRB protocol 55718. Cohort 3 samples were collected under UCSF IRB protocol 20-30479. Influenza virus vaccination samples were from a US cohort enrolled at the Rockefeller University Hospital in New York City in 2012–2013 under a protocol approved by the IRB of Rockefeller University (protocol TWA-0804). Samples from people with seasonal coronavirus infections were collected at the University of Chicago. Samples were de-identified serums of healthcare workers who had respiratory illnesses, were swabbed and tested positive for common cold coronavirus infections in 2019 (University of Chicago protocol 09-043-A).

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author (J.A.W.) upon reasonable request. Datasets 6XCN, 6W41, 1N8Z, 5IBO and 5D6D used for modeling were downloaded from the Protein Data Bank. Source data are provided with this paper.

Code availability

R scripts used for ordinary differential equation modeling are available for download at https://github.com/alexmartinko/Serological_ODE_Model.

References

  1. 1.

    Krammer, F. & Simon, V. Serology assays to manage COVID-19. Science 368, 1060–1061 (2020).

    CAS 
    Article 

    Google Scholar
     

  2. 2.

    Lynch, K. L. et al. Magnitude and kinetics of anti-SARS-CoV-2 antibody responses and their relationship to disease severity. Clin. Infect. Dis. 72, 301–308 (2020).

  3. 3.

    Okba, N. M. A. et al. Severe acute respiratory syndrome coronavirus 2-specific antibody responses in coronavirus disease patients. Emerg. Infect. Dis. 26, 1478–1488 (2020).

    CAS 
    Article 

    Google Scholar
     

  4. 4.

    Seow, J. et al. Longitudinal observation and decline of neutralizing antibody responses in the three months following SARS-CoV-2 infection in humans. Nat. Microbiol. 5, 598–1607 (2020).

    Article 

    Google Scholar
     

  5. 5.

    Smith, T. R. F. et al. Immunogenicity of a DNA vaccine candidate for COVID-19. Nat. Commun. 11, 2601 (2020).

    CAS 
    Article 

    Google Scholar
     

  6. 6.

    Yu, J. et al. DNA vaccine protection against SARS-CoV-2 in rhesus macaques. Science 369, 806–811 (2020).

  7. 7.

    van Doremalen, N. et al. ChAdOx1 nCoV-19 vaccine prevents SARS-CoV-2 pneumonia in rhesus macaques. Nature 586, 578–582 (2020).

    CAS 
    Article 

    Google Scholar
     

  8. 8.

    Long, Q.-X. et al. Antibody responses to SARS-CoV-2 in patients with COVID-19. Nat. Med. 26, 845–848 (2020).

  9. 9.

    To, K. K.-W. et al. Temporal profiles of viral load in posterior oropharyngeal saliva samples and serum antibody responses during infection by SARS-CoV-2: an observational cohort study. Lancet Infect. Dis. 20, 565–574 (2020).

  10. 10.

    Zhao, J. et al. Antibody responses to SARS-CoV-2 in patients of novel coronavirus disease 2019. Clin. Infect. Dis. 71, 2027–2034 (2020).

  11. 11.

    Casadevall, A. & Pirofski, L.-a The convalescent sera option for containing COVID-19. J. Clin. Invest. 130, 1545–1548 (2020).

    CAS 
    Article 

    Google Scholar
     

  12. 12.

    Robbiani, D. F. et al. Convergent antibody responses to SARS-CoV-2 in convalescent individuals. Nature 584, 437–442 (2020).

  13. 13.

    Rogers, T. F. et al. Isolation of potent SARS-CoV-2 neutralizing antibodies and protection from disease in a small animal model. Science 369, 956–963 (2020).

  14. 14.

    Amanat, F. et al. A serological assay to detect SARS-CoV-2 seroconversion in humans. Nat. Med. 26, 1033–1036 (2020).

  15. 15.

    Tan, W. et al. Viral kinetics and antibody responses in patients with COVID-19. Preprint at medRxiv https://doi.org/10.1101/2020.03.24.20042382 (2020).

  16. 16.

    Xiang, J. et al. Evaluation of enzyme-linked immunoassay and colloidal gold-immunochromatographic assay kit for detection of novel coronavirus (SARS-Cov-2) causing an outbreak of pneumonia (COVID-19). Preprint at medRxiv https://doi.org/10.1101/2020.02.27.20028787 (2020).

  17. 17.

    Li, Z. et al. Development and clinical application of a rapid IgM-IgG combined antibody test for SARS-CoV-2 infection diagnosis. J. Med. Virol. 92, 1518–1524 (2020).

  18. 18.

    Lassaunière, R. et al. Evaluation of nine commercial SARS-CoV-2 immunoassays. Preprint at medRxiv https://doi.org/10.1101/2020.04.09.20056325 (2020).

  19. 19.

    Whitman, J. D. et al. Evaluation of SARS-CoV-2 serology assays reveals a range of test performance. Nat. Biotechnol. 38, 1174–1183 (2020).

    CAS 
    Article 

    Google Scholar
     

  20. 20.

    Dixon, A. S. et al. NanoLuc complementation reporter optimized for accurate measurement of protein interactions in cells. ACS Chem. Biol. 11, 400–408 (2016).

    CAS 
    Article 

    Google Scholar
     

  21. 21.

    Letko, M., Marzi, A. & Munster, V. Functional assessment of cell entry and receptor usage for SARS-CoV-2 and other lineage B betacoronaviruses. Nat. Microbiol. 5, 562–569 (2020).

    CAS 
    Article 

    Google Scholar
     

  22. 22.

    Kang, S. et al. Crystal structure of SARS-CoV-2 nucleocapsid protein RNA binding domain reveals potential unique drug targeting sites. Acta Pharm. Sin. B 10, 1228–1238 (2020).

  23. 23.

    Qu, J. et al. Profile of IgG and IgM antibodies against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Clin. Infect. Dis. 71, 2255–2258 (2020).

  24. 24.

    Stadlbauer, D. et al. SARS-CoV-2 seroconversion in humans: a detailed protocol for a serological assay, antigen production, and test setup. Curr. Protoc. Microbiol 57, e100 (2020).

    CAS 
    PubMed 
    PubMed Central 

    Google Scholar
     

  25. 25.

    Zhao, J. et al. Comparison of immunoglobulin G responses to the spike and nucleocapsid proteins of severe acute respiratory syndrome (SARS) coronavirus in patients with SARS. Clin. Vaccine Immunol. 14, 839–846 (2007).

    CAS 
    Article 

    Google Scholar
     

  26. 26.

    Byrnes, J. R. et al. Competitive SARS-CoV-2 serology reveals most antibodies targeting the spike receptor-binding domain compete for ACE2 binding. mSphere 5, e00802-20 (2020).

  27. 27.

    Rosado, J. et al. Multiplex assays for the identification of serological signatures of SARS-CoV-2 infection: an antibody-based diagnostic and machine learning study. Lancet Microbe 2, e60–e69 (2021).

    CAS 
    Article 

    Google Scholar
     

  28. 28.

    Barnes, C. O. et al. Structures of human antibodies bound to SARS-CoV-2 spike reveal common epitopes and recurrent features of antibodies. Cell 182, 828–842 (2020).

  29. 29.

    Yuan, M. et al. A highly conserved cryptic epitope in the receptor binding domains of SARS-CoV-2 and SARS-CoV. Science 368, 630–633 (2020).

    CAS 
    Article 

    Google Scholar
     

  30. 30.

    Glasgow, A. et al. Engineered ACE2 receptor traps potently neutralize SARS-CoV-2. Proc. Natl Acad. Sci. USA 117, 28046–28055 (2020).

    CAS 
    Article 

    Google Scholar
     

  31. 31.

    Poulsen, T. R., Meijer, P.-J., Jensen, A., Nielsen, L. S. & Andersen, P. S. Kinetic, affinity, and diversity limits of human polyclonal antibody responses against tetanus toxoid. J. Immunol. 179, 3841–3850 (2007).

    CAS 
    Article 

    Google Scholar
     

  32. 32.

    Reddy, S. B. et al. Differences in affinity of monoclonal and naturally acquired polyclonal antibodies against Plasmodium falciparum merozoite antigens. BMC Microbiol. 15, 1–11 (2015).

    Article 

    Google Scholar
     

  33. 33.

    Zamecnik, C. R. et al. ReScan, a multiplex diagnostic pipeline, pans human sera for SARS-CoV-2 antigens. Cell Rep. Med. 1, 100123 (2020).

  34. 34.

    Burbelo, P. D. et al. Sensitivity in detection of antibodies to nucleocapsid and spike proteins of severe acute respiratory syndrome coronavirus 2 in patients with Coronavirus Disease 2019. J. Infect. Dis. 222, 206–213 (2020).

    CAS 
    Article 

    Google Scholar
     

  35. 35.

    Tan, C. W. et al. A SARS-CoV-2 surrogate virus neutralization test based on antibody-mediated blockage of ACE2-spike protein–protein interaction. Nat. Biotechnol. 38, 1073–1078 (2020).

  36. 36.

    Mäkelä, O., Ruoslahti, E. & Seppälä, I. J. Affinity of IgM and IgG antibodies. Immunochemistry 7, 917–932 (1970).

    Article 

    Google Scholar
     

  37. 37.

    Wajnberg, A. et al. Robust neutralizing antibodies to SARS-CoV-2 infection persist for months. Science 370, 1227–1230 (2020).

    CAS 
    Article 

    Google Scholar
     

  38. 38.

    Cervia, C. et al. Systemic and mucosal antibody responses specific to SARS-CoV-2 during mild versus severe COVID-19. J. Allergy Clin. Immunol. 147, 545–557 (2020).

    Article 

    Google Scholar
     

  39. 39.

    Klein, S. L. et al. Sex, age, and hospitalization drive antibody responses in a COVID-19 convalescent plasma donor population. J. Clin. Invest. 130, 6141–6150 (2020).

    CAS 
    Article 

    Google Scholar
     

  40. 40.

    Hornsby, M. et al. A high through-put platform for recombinant antibodies to folded proteins. Mol. Cell Proteom. 14, 2833–2847 (2015).

    CAS 
    Article 

    Google Scholar
     

  41. 41.

    Denkinger, C. M. et al. Guidance for the evaluation of tuberculosis diagnostics that meet the World Health Organization (WHO) target product profiles: an introduction to who process and study design principles. J. Infect. Dis. 220, S91–S98 (2019).

    Article 

    Google Scholar
     

  42. 42.

    Pisanic, N. et al. COVID-19 serology at population scale: SARS-CoV-2-specific antibody responses in saliva. J. Clin. Microbiol. 1, e02204-20 (2021).


    Google Scholar
     

  43. 43.

    Faustini, S. E. et al. Detection of antibodies to the SARS-CoV-2 spike glycoprotein in both serum and saliva enhances detection of infection. Preprint at medRxiv https://doi.org/10.1101/2020.06.16.20133025 (2020).

  44. 44.

    Fry, S. R. et al. Detection of HSV type-1 and type-2 IgG using an in vitro PCA based homogeneous immunoassay. Biochem. Biophys. Res. Commun. 372, 542–546 (2008).

    CAS 
    Article 

    Google Scholar
     

Download references

Acknowledgements

We acknowledge the members of the Wells Lab, especially those involved in our COVID-19 research program. We also acknowledge M. Nussenzweig, M. Caskey and C. Gaebler (Rockefeller University) as well as C. Zamecnik, M. Wilson and J. DeRisi (UCSF) for providing the first cohort of convalescent sera. We thank M. Nussenzweig and D. Robbiani (Rockefeller University) for providing plasmids for C004, C105 and C135 IgGs. We thank P. Kim (Stanford University) and I. Wilson (Scripps Research) for providing the plasmid for CR3022. We additionally thank P. Wilson (University of Chicago) for the seasonal coronavirus control samples. We thank M. Lin, X. Shu and C. Chiu for helpful discussions. We thank all the patients for their participation in this study. J.A.W. is grateful for funding from the Harry and Dianna Hind Endowed Professorship in Pharmaceutical Sciences and the Chan Zuckerberg Biohub that helped support this work. The National Science Foundation Graduate Research Fellowship Program supported S.K.E. (1650113) and I.L. (1650113). Postdoctoral fellowship support included a Merck Fellowship of the Damon Runyon Cancer Research Foundation (DRG-2297-17 to X.X.Z.), a National Institutes of Health K99/R00 award (1K99EB030587-01 to X.X.Z.), a National Institutes of Health National Cancer Institute F32 (5F32CA239417 to J.R.B.), a Merck Postdoctoral Research Fellowship from the Helen Hay Whitney Foundation (S.A.L.) and a National Institutes of Health K99/R00 (1K99GM135529 to A.A.G.). The LIINC cohort study was funded by NIH/NIAID 3R01AI141003-03S1 (T.J.H.). T.T.W. was supported, in part, by Fast Grants, the CEND COVID Catalyst Fund, the NIH/NIAID (U19AI111825 and R01AI139119) and the Rockefeller University Center for Clinical and Translational Science Grant no. UL1 TR001866. We would also like to acknowledge funding from the Chan Zuckerberg Biohub, Rapid Response.

Author information

Author notes

  1. These authors contributed equally: Susanna K. Elledge, Xin X. Zhou.

Affiliations

  1. Department of Pharmaceutical Chemistry, University of California, San Francisco, San Francisco, CA, USA

    Susanna K. Elledge, Xin X. Zhou, James R. Byrnes, Irene Lui, Katarina Pance, Shion A. Lim, Jeff E. Glasgow, Kevin K. Leung & James A. Wells

  2. Soteria Biotherapeutics, San Francisco, CA, USA

    Alexander J. Martinko

  3. Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco, San Francisco, CA, USA

    Anum A. Glasgow

  4. Division of Experimental Medicine, University of California, San Francisco, San Francisco CA, USA

    Keirstinne Turcios, Nikita S. Iyer, Leonel Torres, Timothy J. Henrich & Bryan Greenhouse

  5. Division of HIV, Infectious Diseases, and Global Medicine, University of California, San Francisco, San Francisco, CA, USA

    Leonel Torres, Michael J. Peluso & Bryan Greenhouse

  6. Chan Zuckerberg Biohub, San Francisco, CA, USA

    Taia T. Wang, Cristina M. Tato & James A. Wells

  7. Departments of Medicine and of Microbiology and Immunology, Stanford University School of Medicine, Stanford, CA, USA

    Taia T. Wang

  8. Department of Cellular and Molecular Pharmacology, University of California, San Francisco, San Francisco, CA, USA

    James A. Wells

Contributions

S.K.E., X.X.Z. and J.A.W. conceived the study, designed the research and wrote the manuscript. S.K.E. and X.X.Z. designed the experiments and analyzed the data. S.K.E. performed the experiments, unless otherwise stated. X.X.Z. performed structure modeling and analysis for sensor design. J.R.B. performed the anti-Fab ELISA experiments, provided advice for the whole blood work and co-wrote the manuscript. K.K.L. designed the robotic workflow and co-wrote the manuscript. A.J.M. performed the in silico differential equation modeling and analyzed the data. I.L., K.P. and S.A.L. helped with expression and purification and performed the ACE2 epitope binning experiment. J.E.G. and A.A.G. designed, expressed and purified the higher-affinity ACE2 mutant. T.T.W. provided patient sera and control sera samples. T.J.H., M.J.P., B.G., N.S.I., L.T. and K.T. provided patient samples and oversaw LIINC sample collections, sample processing, sample maintenance and cohort design. B.G. and C.M.T. provided helpful discussions. X.X.Z. and J.A.W. supervised the research. All authors edited and approved the final manuscript version.

Corresponding author

Correspondence to
James A. Wells.

Ethics declarations

Competing interests

S.K.E., X.X.Z. and J.A.W. have filed a provisional patent on the solution-based spLUC assay. J.E.G., A.A.G., I.L., X.X.Z. and J.A.W. have filed a provisional patent on the ACE2 variants.

Additional information

Peer review information Nature Biotechnology thanks Harry Ostrer, Kevin Nichols and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Design and characterization of S sensors.

a, Annotated depiction of the SARS-CoV-2 Spike protein. The S sensors were developed using only the S-RBD domain (aa 328–533, PDB: 6W41) shown in pink. b, Structure of the S-RBD domain shows the N and C termini locate in close proximity. c, d, Modeling of c, ACE2-competitive antibody C105 (PDB: 6XCN) binding to S-RBD-SmBiT/LgBiT sensors, and d, CR3022 (PDB: 6W41) binding to S-RBD-SmBiT/LgBiT sensors. Modeling and distance measurements were performed with PDB 6XCN, 6W41, 1N8Z, 5IBO and 5D6D in PyMOL. e, Yield of the 5 Spike-NanoBiT sensor fusions. The Spike LgBiT sensors were made with 5aa, 15aa, and 25aa Glycine-Serine (GS) linkers (L5, L15 and L25). The Spike SmBiT sensors were made with 15aa, and 25aa GS linkers (S15 and S25). Because the N and C termini of the S-RBD domain locate in close proximity, only fusions to the C termini of S-RBD were constructed. f, The S sensors are most sensitive at 1 nM for detecting CR3022 in solution compared to higher or lower sensor concentrations. Two technical replicates are plotted from n=1 individual experiment. Lines connecting the means of the samples are plotted. g, S sensors with varied linker lengths resulted in very similar signal strength in detecting CR3022. Two technical replicates are plotted from n=1 individual experiment. Lines connecting the means of the samples are plotted.
Source data

Extended Data Fig. 2 The biosensors are more sensitive to high-affinity binders.

The ACE2-Fc variant which binds 10-fold tighter to S-RBD generated ~3-fold higher signal at 10 nM protein concentration comparing to WT ACE2-Fc. Two technical replicates are plotted from n=1 individual experiment. Lines connecting the means of the samples are plotted.
Source data

Extended Data Fig. 3 ODE models predict a linear, dose-dependent response and KD dependence of the luminescence signal.

a, Antibody (C) and sensor components (A and B) are in thermodynamic equilibrium with enzymatically inactive (D, E, G, and I) and active (H) sensor bound species. b, At 1 nM starting concentration of sensor ([A] and [B]), spLUC assays are predicted to generate signals linearly correlated to a broad range of antibody concentrations ([Ab]). Signal is predicted to be insensitive to antibody concentrations for antibodies with high affinity for the sensor (≤ 1nM), but weaker affinity antibodies (KD > 1 nM) will result in significantly lower levels of reconstituted enzyme. c, At KD values equivalent or higher than the sensor concentrations, the spLUC signals are predicted to drop significantly.

Extended Data Fig. 4 Design and characterization of N sensors.

a, Annotated depiction of the SARS-CoV-2 Nucleocapsid protein (protein N). All N protein fusions designed included the RNA binding domain (aa 44–180, N-RBD) and excluded the dimerization domain (aa 257–419). b, Structure of the N-RBD domain shows the N and C termini locate far from each other and fusion of the split enzyme fragments to N or C termini may result in different detection sensitivity (PDB: 6YI3). c, Yield of the six N protein-NanoBiT sensor fusions. d, The N-terminal N sensor pair (LN + SN, 44–257) was less sensitive than the LC + SC (44–180) and LC2 + SC2 (44–257) C terminal N sensor pairs when the assay was performed on a rabbit polyclonal anti-N protein antibody (Sino Biological, Cat#: 40588-T62-50). Two technical replicates are plotted from n=1 individual experiment. Lines connecting the means of the samples are plotted. e, Additionally, only patient 6 and 8 showed signals above controls in the serological assay performed with LN + SN sensors, while all four patients showed signals with the LC + SC sensors. Two technical replicates are plotted from n = 1 independent experiment. Lines connecting the means of the samples are plotted.
Source data

Extended Data Fig. 5 Individual cohorts show good correlation between S and N sensors.

Each cohort shows robust correlation with R = 0.59, 0.87, and 0.73 for a, cohort 1 (56 samples), b, cohort 2 (47 samples), and c, cohort 3 (87 samples), respectively. R values and P values (two-tail) of a non-parametric Spearman correlation analysis are labeled in the graphs. Lines represent linear regression. For all graphs, dots represent the average of two technical replicates from n=1 independent experiment.
Source data

Extended Data Fig. 6 Comparison of the ELISA and the spLUC results.

a, Signals from the S sensor spLUC assay (cohort 1, 57 samples) correlate very well with S-RBD ELISA anti-Fab signals (R = 0.91), moderately well with anti-IgG signals (R = 0.43), and poorly with anti-IgM signals for cohort 1 (R = −0.066). Line represents linear regression. b, Signals from the S sensor spLUC assay (cohort 2, 40 samples with anti-Fab detection, 47 samples with anti-IgM/IgG detection) correlate very well with S-RBD ELISA anti-Fab signals (R = 0.84) and with anti-IgG signals (R = 0.86), but poorly with anti-IgM signals for cohort 1 (R = 0.29). Line represents linear regression. c, Signals from the S sensor (cohort 3, 87 samples) correlate well with S-RBD ELISA anti-IgG signals (R = 0.88). For a-c, R values and P values (two-tail) of a non-parametric Spearman correlation analysis are labeled in the graphs. Lines represent linear regression. d, e, f, The seronegative samples in the anti-S spLUC assay also showed low anti-Fab or anti-IgG signals in ELISA serology tests for d, cohort 1 (negative samples (9), positive samples (48)), e, cohort 2, (negative samples (5), positive samples (35)), and f, cohort 3, (negative samples (3), positive samples (53)). Horizontal lines represent the median value. For all graphs, dots represent the average of two technical replicates from n=1 independent experiment.
Source data

Extended Data Fig. 7 Inter-assay, Inter-day and Intra-day variability of spLUC assay.

a, 46 plasma samples were assayed a total of five times in three independent experiments over two days for each sensor. All five replicates are plotted on the graph, with the bar representing the average. The dotted line represents the cutoff values for positive and negative samples for the S and N sensors. b, The coefficient of variation was calculated for intra-assay, intra-day, and inter-day variability. Coefficient of variation is calculated as ratio of standard deviation to the average value.
Source data

Extended Data Fig. 8 Simulated robotics-assisted spLUC assay.

a, Serum sample transfer to an assay plate using Biomek Fx Automated Workstation in ~2 minutes. b, Robotics-assisted dispensing and luminescence reading for one iteration of 96 assays takes ~35 minutes. c, Simulated run for 40 iterations (3840 assays) can be completed in 3 hours. Gantt chart generated by simulated run using Thermo Momentum software.

Extended Data Fig. 9 S and N sensors are functional after lyophilization.

a, Both the S and the N sensors can survive lyophilization. The majority of proteins (70–100%) can be reconstituted after lyophilization. The lyophilized S sensors lost 50% of signal. The lyophilized N sensors remain 100% active. b, The lyophilized S sensors detected CR3022 at ~50% signal strength compared to fresh sensors. Three technical replicates are plotted from n=1 independent experiment. Lines connecting the means of the samples are plotted. c, The lyophilized N sensors detected antibodies from patient sera at similar signal strength compared to fresh sensors. Two technical replicates are plotted from n=1 independent experiment. The bars represent the mean. d, Vacuum dried substrate and substrate stored at −20 °C (fresh) behave similarly when detecting recombinant CR3022 with S sensors. Lyophilized dilution buffer and dilution buffer stored at −20 °C (fresh) also showed similar signal. Three technical replicates are plotted from n=1 independent experiment. Lines connecting the means of the samples are plotted.
Source data

Extended Data Fig. 10 Portable luminometer.

The spLUC assay is also amenable to detection with a Berthold portable luminometer. The handheld luminometer showed similar sensitivity of recombinant CR3022 with S sensors compared to the plate reader. Due to the tube format of the handheld luminometer, the sample volume was doubled and thus the overall signals are higher than for the plate reader samples, but similar sensitivity is maintained. Two technical replicates are plotted from n=1 independent experiment. Lines connecting the means of the samples are plotted.
Source data

Supplementary information

About this article

Verify currency and authenticity via CrossMark

Cite this article

Elledge, S.K., Zhou, X.X., Byrnes, J.R. et al. Engineering luminescent biosensors for point-of-care SARS-CoV-2 antibody detection.
Nat Biotechnol (2021). https://doi.org/10.1038/s41587-021-00878-8

Download citation

Read More

Leave a Reply

Your email address will not be published. Required fields are marked *