Fabrication of a non-odorant VirD-GPCR array

To develop the first high-content VirD array, we decided to focus on the 370 non-odorant GPCRs. We assembled a collection of open reading frames (ORFs) encoding 337 non-odorant GPCRs that were available to us. To enable high-throughput cloning of these ORFs into the HSV-1 strain KOS genome carried in a bacterial artificial chromosome (BAC) vector10, we replaced the glycoprotein B coding sequence (UL27) with a Gateway cloning cassette. The engineered molecule also contained the sequences for the expression of a V5 epitope at the C-terminus of the cloned ORF followed by a STOP codon (Fig. 1a). Meanwhile, STOP codons of the GPCR ORFs were removed via PCR reactions and the modified ORFs subcloned into the Gateway Entry vector. Single colonies were picked, followed by Sanger sequencing to ensure error-free subcloning of each ORF. Confirmed STOP-free ORFs were then shuttled into the UL27 (gB) locus in the HSV-1 genome using LR recombination reactions and transformed into Escherichia coli by electroporation. For each bacterial transformation, at least two colony PCR reactions were performed using a primer pair that annealed to the viral sequences flanking the cloning site of the GPCR ORFs, and gel electrophoresis was employed to examine whether the amplicon was of the expected size of the GPCR cloned. To this end, we have successfully subcloned a total of 332 (98.5%) GPCR ORFs into the UL27 locus.

Fig. 1

Fig. 1

Construction of high-content VirD-GPCR array. a Subcloning of 335 human GPCR ORFs into the UL27 locus of the HSV-1 genome. After the STOP codons were removed from the 335 available GPCR ORFs, they were subcloned into the UL27 locus in the HSV-1 genome on a BAC vector, resulting in fusion with a V5 tag at their C-termini (middle panel). After bacterial transformation, colony PCR reactions were carried out and the products examined using electrophoresis to identify the correct construct (right panel). b Production of VirD-GPCR virions and VirD array fabrication. Confirmed recombinant virus constructs were individually transfected to Vero cells and the viruses were harvested ~7 days post-transfection. Anti-V5 mAb was used to examine expression of the GPCRs as a quality control. Passers were next used to infect cells for virion production. After sucrose cushion centrifugation, a fraction of purified virions was examined again with anti-V5. 315 VriD-GPCRs passed this quality control step and were spotted onto a glass slide to form VirD-GPCR array. The quality of VirD-GPCR array was examined using anti-gD mAb, followed by a Cy3-labeled anti-mouse IgG antibody. All the ViP.rD-GPCRs on the array showed positive anti-gD signals while the BSA showed the lowest signals

To produce recombinant viruses, we first transfected each GPCR::ΔUL27 BAC DNA into a Vero transformed cell line, D87, that complements the growth of mutants that do not express gB. When viral plaques became evident, 5–7 days post-transfection, low titer viral stocks were harvested for each GPCR recombinant virus. After a secondary infection, expression of a total of 317 GPCRs was detected in total cell lysates with anti-V5 antibodies (Fig. 1b). Next, high titer stocks of the 317 VirD-GPCRs were individually prepared following infection of D87 cells. Since we observed that expression levels of different GPCRs varied in different cell lines, we prepared the final VirD-GPCR virions from Vero-, HEL-, HeLa-, and HEK-293T-infected cells to maximize the production of VirD-GPCRs (Supplementary Fig. 1a, b). The VirD-GPCR virions were further purified to homogeneity via sucrose cushion and resuspended in a small volume to maintain a high virion concentration. A small fraction of each purified VirD-GPCR virion was subjected to anti-V5 immunoblot (IB) analysis, based on which 315 VirD-GPCRs passed this quality control step (Fig. 1b). Finally, the 315 virion preparations were arrayed into a 384-well titer dish and robotically printed in duplicate onto SuperEpoxy slides to form the VirD-GPCR array. The quality of the printed VirD-GPCR arrays was examined with anti-gD antibody and all of the 315 arrayed VirD-GPCRs showed significant anti-gD signals as compared with the negative controls (e.g., bovine serum albumin (BSA)) (lower left panel; Fig. 1b). Moreover, scatter plot analysis of an anti-gD assay performed in duplicate indicated a high reproducibility with a correlation coefficient of 0.92 (Supplementary Fig. 1c, d). Therefore, we successfully produced a high-content VirD array that covers 85% of the annotated non-odorant human GPCRs.

Profiling antibody specificity on VirD-GPCR array

Antibody-based biologicals are emerging as the next-generation therapeutics because of their unique properties, such as superior pharmacokinetics, simple formulation, and modular, easily engineerable format. Indeed, Erenumab (trade name Aimovig) was recently approved by the FDA as the first antibody-based drug that targets the GPCR, i.e., CGRPR, for the prevention of migraine11. However, there are serious problems with the quality and consistency of antibodies because of the absence of standardized antibody-validation criteria, a lack of transparency from commercial antibody suppliers and technical difficulties in comprehensive assessments of antibody cross-reactivity12,13,14,15,16,17.

To demonstrate that the VirD-GPCR arrays can be used to test antibody specificity, we selected 20 commercially available mAbs targeting 20 human GPCRs. The criteria used to select the 20 commercial antibodies were the following: (i) they have to be monoclonal antibodies; (ii) they should recognize ectodomains of the human GPCRs; and (iii) they have to be flow cytometry-positive when used against intact cells (Table 1). We applied each mAb individually to a pre-blocked VirD-GPCR array, measured the corresponding binding signals using fluorescently labeled secondary antibodies, and determined the Z-scores of all the VirD-GPCRs for that mAb on the basis of the standard deviation (SD) value for each assay. Using a stringent cut-off value of Z-score ≥5, only 9 of the 20 tested mAbs recognized their intended targets specifically. Two examples of anti-CXCR2 and -CCR7 are shown in Fig. 2a, b. Of note, four of them are known to have neutralization activity. One other mAb, anti-ACKR3, not only recognized VirD-ACKR3 as its top 1 target, but also showed Z-scores of 6.5 and 5.9 to VirD-CALCR and VirD-GPR61, respectively, suggesting off-target binding activity. The remainder ten mAbs, however, completely failed to show any detectable binding activities to their intended targets, and some of them bound to many unrelated VirD-GPCRs (i.e., anti-DRD1 in right panel; Fig. 2a).

Table 1 Binding specificity for 20 commercial mAbs using VirD-GPCR arrays

Fig. 2

Fig. 2

Specificity tests of commercial mAbs on VirD-GPCR arrays. a Examples of binding signals obtained with commercial mAbs. Anti-CXCR2 and -CCR7 are shown as ultra-specific; anti-ACKR3 can cross-react with CALCR and GPR61; anti-DRD1 completely failed to recognize its target while showing nonspecific binding activities to other GPCRs. b Histograms of Z-scores obtained with three mAbs. Z-scores of the two off- targets identified by anti-ACKR3 are also shown. c Immunofluorescence analysis (IFA) validation of anti-ACKR3 to its off-targets in infected Vero cells. K082-infected cells are shown as a negative control. d Immunoblot analysis also confirmed that anti-ACKR3 can recognize its two off-targets in the cell lysates of infected Vero cells under both denatures and native conditions. e, f Anti-DRD1 failed to recognize DRD1 in VirD-DRD1-infected cells using IFA (e) or immunoblot (IB) analyses (f). g Single-molecule imaging using TIRF microscopy to determine interactions between VirD-DRD1 and its canonical ligand D1 antagonist. K082 virions were used as a negative control. Quantitative analysis of TIRF imaging demonstrated that VirD-DRD1 showed significantly higher binding signals to D1 antagonist than K082. Data were analyzed with two-tailed Student’s t-test, n = 10

Although an easy explanation is that these mAbs are of poor quality and thus failed in this test, it was also possible that the 11 GPCRs were displayed in the wrong orientation/misfolding in the virions. To explore the latter possibility, we first focused on the off-target binding activity observed with anti-ACKR3 mAb. Using intact Vero cells infected with viruses carrying ACKR3, CALCR, or GPR61, we performed immunofluorescence assays (IFA) with anti-ACKR3. As shown in Fig. 2c, the mAb showed strong staining to all three infected cells as compared with the negative control cells infected with K082, a ∆UL27 HSV-1 virus, suggesting that this mAb could recognize CALCR and GPR61 embedded in intact cell membranes. Similarly, this mAb also strongly recognized all three GPCRs in IB analyses against cell lysates of these infected cells under either native or denatured conditions (Fig. 2d). Interestingly, sequence alignment analysis using the ectodomain sequences of the three GPCRs identified a highly conserve 6-mer motif of [NMF]GEL[VTG][RC], suggesting a commonly shared epitope for this mAb. Homology search did not identify any other non-odorant GPCRs that carry a 6-mer peptide with high similarity.

Next, we randomly selected 4 (i.e., anti-DRD1, -CCR2, -CCR9, and -S1PR1) of the 10 mAbs that completely failed to recognize their intended GPCRs on the VirD-GPCR arrays for the IFA and IB analyses. All of the four mAbs failed these tests; anti-DRD1 is shown in Fig. 2e, f as an example (Fig. 2e, f; Supplementary Fig. 2). The failure of these mAbs was not due to poor expression of their target GPCRs because these GPCRs could be readily detected with anti-V5 (Fig. 2f). To determine whether these failed mAbs could recognize linear epitopes, we performed mAb-binding assays on HuProt arrays, each comprised of 20,240 human proteins in full-length and denatured using 9 M urea treatment18. Except anti-S1PR1, which was not tested because S1PR1 was not available on HuProt, the rest nine mAbs failed to recognize their intended targets as the top targets. Similarly, none of them recognized their intended targets under native conditions on HuProt arrays (Supplementary Fig. 3).

Functional test of GPCR activity with canonical ligands

To demonstrate further that the ten GPCRs that were not recognized by their respective mAbs were functional/folded correctly, we decided to test their binding activities to their canonical ligands using an imaging approach. Four of the ten ligands were commercially available small molecules with fluorescent labels; the rest were peptide ligands and we labeled them with NHS-conjugated dyes. We immobilized these VirD-GPCRs on a passivated cover slip and incubated with their corresponding ligands at low nanomolar concentrations. Using single-molecule imaging on total internal reflection fluorescence (TIRF) microscopy, we recorded the resulting fluorescent images and compared the binding signals with that of the negative control, K082 virus. Except P2RY13, all the nine VirD-GPCRs that were not recognized by the commercial mAbs showed significantly higher binding signals than the K082 control, indicating that they were functional and folded correctly (Fig. 2g; Supplementary Fig. 4). As an example, D1 antagonist showed significantly higher binding signals to its canonical receptor DRD1 than K082 (Fig. 2g). P2RY13 receptor was not observed to interact with its ligand, ATP, presumably due to the weak affinity in the low micromolar range19. Taken together, the high-content non-odorant VirD-GPCR array was validated as a powerful platform to screen for high-quality mAbs against folded GPCRs.

Specificity test for GPCR ligands

The success of the above approach prompted us to determine whether VirD-GPCR arrays could be used to examine binding specificity of GPCR ligands. As a proof of concept, we chose two peptide ligands, dynorphin A and somatostatin-14 (SRIF-14), to perform specificity tests. The function of the labeled ligands was largely unchanged based on the calcium signaling measured in Gα-15 cells infected with the recombinant virus expressing the GPCR (Supplementary Fig. 5).

As illustrated in Fig. 3a, dynorphin A bound to its canonical receptor, OPRD1, as the top receptor with significant signal intensity20. To demonstrate that dynorphin A can activate OPRD1, we performed calcium imaging assays in OPRD1-infected cells and showed that dynorphin A could indeed induce Ca2+ influx, although to a lesser extent than OPRK1-infected cells (Supplementary Fig. 5). To better understand why OPRK1 did not show detectable binding signals to dynorphin A on the VirD-GPCR arrays, we measured its expression level in purified viruses using anti-V5 antibodies and found that it was present at a very low level (Supplementary Fig. 6a). To rule out the possibility that the kappa receptor was not correctly folded in the envelope of the virions, we employed a more sensitive TIRF microscopy method and demonstrated that VirD-OPRK1 showed significantly higher binding signals to Cy3-labled dynorphin A than the K082 control virions (Supplementary Fig. 6b). Taken together, this new evidence confirmed that dynorphin A could activate both kappa and delta receptors in cells and that virion-displayed kappa receptor could bind to dynorphin A.

Fig. 3

Fig. 3

Identification and cell-based validation of peptide ligand–GPCR interactions. a Commercially available Dynorphin A was Cy5-labeled and probed to a VirD-GPCR array. Quantitative analysis showed that it bound to VirD-OPRD1 with the highest Z-score followed by ADCYAP1R1 and P2RY2. b A commercially available peptide SRIF-14 was Cy3-labeled and probed to a VirD-GPCR array. Quantitative analysis revealed that it bound to several unexpected off-targets in addition to its canonical receptor, SSTR2. c Vero cells were separately infected with SSTR2, GABBR2, NTSR1, KISS1R, or K082 virus. Infected cells were then incubated with Cy5-labeled SRIF-14 at 8 µM in the absence (upper panel) or presence of cold SRIF-14 (middle panel) or cyclosomatostatin (cycloSST, lower panel). d Quantitative analysis of binding signals. Each binding assay was performed in triplicate and the obtained binding signals were normalized to those of the K082 controls. Data were analyzed by two-way ANOVA with repeated measures followed by Bonferroni post-test. *P< 0.05, comparison between VirD-GPCRs and K082 in the absence of competitor ligands; #P< 0.001, comparison between binding signals obtained in the absence and presence of the competitor ligands. n = 3, biologically independent samples

SRIF-14 bound to more than 15 VirD-GPCRs with Z-scores ≥2 (Fig. 3b), including its known receptor, SSTR2 (ref. 21). To determine whether the observed off-target binding activities of SRIF-14 were not due to an artifact caused by dye-labeling, we employed a cell-based competition assay to examined binding specificity between SRIF-14 and three randomly selected off-target GPCRs, namely GABBR2, NTSR1, and KISS1R, with Z-scores ≥2 (Fig. 3c). The three VirD-GPCR constructs were used to separately infect Vero cells and VirD-SSTR2 and K082 were also included as positive and negative controls, respectively. The same fluorescently labeled SRIF-14 was added to these infected Vero cells in the absence or presence of cold SRIF-14 (i.e., agonist) or cyclosomatostatin (i.e., antagonist). Quantitative analysis clearly showed that all cell lines infected with the three off-target GPCRs showed significantly higher binding signals than K082-infected cells (upper panel of Fig. 3c, d), comparable to those infected with SSTR2. More importantly, both cold SRIF-14 and cyclosomatostatin (cycloSST) could readily compete off the binding signals of Cy5-labeled SRIF-14 on the cells infected with the three off-target GPCRs, suggesting ligand-specific interactions (middle and lower panels of Fig. 3c, d).

Because the array- or cell-based assays can only offer an end-point measurement, we decided to obtain true binding kinetics using a previously reported virion-oscillator approach22,23,24. It is a label-free plasmonic imaging technique that can quantify the ligand-binding-induced mobility change of the virions anchored on the surface of the sensor chip via a flexible molecular linker and reveal the binding kinetics of small-molecule ligands. VirD-GABBR2, -NTSR1, and -KISS1R were separately attached to a gold-coated cover slip via a flexible polyethylene glycol (PEG) linker (Fig. 4a). Again, VirD-SSTR2 and K082 were included as the positive and negative controls, respectively. Using surface plasmon resonance (SPR) to monitor voltage-induced oscillation of the VirD-GPCRs, we recorded oscillation amplitude changes of VirD-GPCRs as a result of association and dissociation of SRIF-14 in a real-time, label-free fashion. As illustrated in Fig. 4b, f, all four tested VirD-GPCRs showed typical association and dissociation curves in a dose-dependent manner, which allowed us to determine the corresponding ka and kd values (Supplementary Table 1). As expected, K082 did not show any detectable binding kinetics even in the presence of 8 µM SIRF-14 (Fig. 4f). Binding affinity KD values were thus calculated for SSTR2, GABBR2, NTSR1, and KISS1R to be 11.2 nM, 0.4 µM, 2.6 µM, and 25.0 µM, respectively. These independent experiments confirmed the off-target-binding activities of SIRF-14 observed on VirD-GPCR arrays.

Fig. 4

Fig. 4

Binding kinetics of SRIF-14 to SSTR2, K082, and other three virions. a Principle of binding kinetics measurement using virion-oscillator device. b–f Binding kinetics of SRIF-14 to its canonical receptor, SSTR2, and three newly discovered off-target GPCRs. The red and purple arrows mark the starting time points of the association and the dissociation phases, respectively. Binding curves were fit using the first order kinetics model (solid lines). The calculated affinity values range from 11.2 nM (SSTR2) to 25.0 µM (KISS1R), while K082 virion showed no detectable binding activity to the SRIF-14. Detailed ka, kd, and KD values are listed in Supplementary Table 1

Discovery and characterization of GPCR targets for GBS

In recent years, several elegant studies reported that both Gram-negative and -positive bacterial pathogens, such as Neisseria meningitidis and Streptococcus pneumoniae, could utilize human GPCRs (e.g., ADRB2 and PTAFR) as receptors to penetrate human epithelial cells25,26. Streptococcus agalactiae (a.k.a. group B Streptococcus or GBS) is the most common Gram-positive organism causing neonatal meningitis by penetrating human blood–brain barriers (BBBs). However, its host receptor has remained elusive. To explore the possibility that GBS may exploit host GPCRs as a means to penetrate BBB, we probed the VirD-GPCR array with fluorescently labeled live GBS K79 (a strain isolated from a neonate with meningitis) in duplicate to discover potential host receptors. As a comparison, a Gram-positive non-pathogenic bacterium, Streptococcus gordonii, which does not penetrate BBB, was used as a negative control. Five VirD-GPCRs, namely GPR101, GPR148, LHCGR, CysLTR1, and LGR5, showed significantly higher binding signals to K79 than S. gordonii in a reproducible manner (Fig. 5a).

Fig. 5

Fig. 5

Discovery and validation of a new GPCR receptor for GBS. a Five VirD-GPCRs were identified as potential candidate receptors. A clinical strain K79 of GBS was Cy3-labeled and probed to a VirD-GPCR array. In parallel, S. gondonii was used as a non-pathogenic negative control (middle). CNR1 and CASR were not bound by either GBS or S. gondonii. Quantitative analysis of the binding signals from the two bacteria identified five GBS-specific GPCRs (right). n = 2, biologically independent samples. b In vitro validation of CysLTR1. Human brain microvascular endothelial cells (HBMEC) were pretreated with Montelukast to block CysLTR1, followed by incubation with GBS. After washes, antibiotics were added to kill free bacteria and HBMEC were lysed and plated onto blood agar plates. After overnight incubation, numbers of GBS colonies were counted. As compared with the DMSO-treated negative control, Montelukast showed a dose-dependent inhibition of GBS invasion into HBMEC. n = 5, biologically independent samples. c In vivo validation of CysLTR1. A group of mice was each intraperitoneally administered Montelukast (n = 6) or DMSO (n = 5). After 2 h each mouse received 1 × 108 CFU of GBS (K79) via the tail vein injection. One hour later, blood and homogenized brains were collected and plated for bacterial counts. Using the same colony formation method, administration of Montelukast reduced GBS brain infection in the mice by an average of 81% as compared to the DMSO controls (right). No significant differences in the levels of GBS counts were observed in the blood between the two groups. Data were analyzed with two-tailed Student’s t-test whereas *P < 0.05, **P < 0.01, and ***P < 0.001

Of the five identified potential GBS receptors, GPR101, GPR148, and LGR5 are orphan receptors without identified canonical ligands. LHCGR is mainly expressed in ovary and testis and binds luteinizing hormone; mutations in this gene are known to cause infertility27. Therefore, these four candidates are either difficult to pursue or less relevant to the pathogenesis of meningitis. The last candidate receptor, CysLTR1, recognizes cysteinyl leukotrienes and is an attractive candidate for several reasons. First, its activation is associated with increased permeability of BBB, as well as promotion of the movement of leukocytes from the blood into brain tissues in animal models28. Second, its activation may also increase the entry of leukocyte-borne viruses, such as HIV-1, into brain tissue29. Third, a well-established antagonist, Montelukast, specifically inhibits CysLTR1 but not its homolog CysLTR2 (ref. 30). Importantly, recent studies have demonstrated that Montelukast could protect against hippocampus injury induced by transient ischemia and reperfusion in rats31.

To validate further the discovery of CysLTR1 as a potential target of GBS, we first employed cultured human brain microvascular endothelial cells (HBMEC) to evaluate the role of endogenous CysLTR1 in GBS penetration. HBMEC are the major component of the BBB and CysLTR1 is known to be expressed in this cell line32. Monolayers of HBMEC were pretreated with Montelukast at different concentrations for 1 h, washed, and infected with GBS K79 (see Methods for more details). DMSO-treated cells were used as a vehicle control. After removal of unbound bacteria with several washes, antibiotics (gentamicin and penicillin) were added to the cells to kill extracellular bacteria. To evaluate GBS penetration to the cells under different conditions, cells were lysed, diluted, and plated onto blood agar plates. Colonies formed on the agar plates were counted and used as a proxy to evaluate GBS penetration of the BBB (see Methods). After normalization to the DMSO controls, it was clear that Montelukast significantly inhibited GBS penetration to HBMEC in a dose-dependent fashion. In fact, as much as 77% GBS penetration was inhibited by 50 μM Montelukast (Fig. 5b). These results corroborated our VirD-GPCR array results that CysLTR1 might play an important role in GBS penetration of the BBB.

To further demonstrate the role of CysLTR1 in GBS penetration into the brain in vivo, we employed a mouse model of experimental hematogenous meningitis (Fig. 5c). A group of six mice were each intraperitoneally administered Montelukast (5 mg kg−1), while another group of five mice received DMSO as a vehicle control. After 2 h each mouse received 1 × 108 CFU of GBS (K79) via the tail vein injection. One hour later, blood was collected and plated for bacteria counts. Immediately following the blood collection, mice were transcardially perfused to remove the remaining body blood and the brains were removed, weighed, homogenized, and plated for bacterial counts (see Methods for more details). We found that the administration of Montelukast reduced the GBS infection in the mouse brains by an average of 81% as compared to that of the DMSO controls. It is important to note that no significant differences of GBS counts were observed in the blood between the two groups, indicating that decreased GBS penetration into the brain was not the result of having less bacterial counts in the blood at the time of collecting the brain specimens (Fig. 5c). Taken together, these experiments demonstrated that using the VirD-GPCR array as an unbiased screening platform, we successfully identified CysLTR1 as a receptor for GBS penetration into host cells both in vitro and in vivo.