Recombinant human CXCL1–3 and CXCL5–8 were obtained from PeproTech. hCXCR2 expression plasmid (pUNO1-kIL08RB) was purchased from InvivoGen. Poly-D-lysine (PDL; #2780) was purchased from MERCK. Nano-Glo^® Vivazine™ substrate (#N2581) was purchased from Promega. The REGA-SIGN plasmids and the NanoLux plasmids (i.e., CXCR2.mNeongreen (mNG), CXCR2.Nanoluciferase (NLuc), and β-arrestin1.Nluc) were previously described (Boon et al., 2023; Luís et al., 2022). The compounds were dissolved in DMSO.

Human embryonic kidney cells (HEK293A, American Type Culture Collection (ATCC)) were cultured in Dulbecco’s modified Eagle medium, high glucose (DMEM; #41965, Thermo Fisher Scientific), supplemented with 10% fetal bovine serum (FBS; #SV30160.03, Cytiva), which is referred to as growth medium from here onward. HEK293A cells stably expressing hCXCR2 were transfected with pcDNA3.1(+) hCXCR2 plasmid using FuGENE HD transfection reagent (Promega) according to the manufacturer’s protocol. HEK293A.hCXCR2 cells were cultured in the growth medium supplemented with 500 μg/mL geneticin. Receptors’ expression was validated by flow cytometry, as described previously (Boon et al., 2024).

Cells were trypsinized and resuspended at 3 × 10⁵ cell/mL in the growth medium, whereafter they were incubated for 2 hours at RT. Cells were transfected in suspension using FuGENE^® HD Transfection Reagent at a 3:1 reagent to DNA ratio in Opti-MEM™ I Reduced-Serum Medium, with the final DNA concentration of 1 μg/μL. The transfection mixture was incubated for 10 min at RT before adding it to the cell suspension. Transfected cells were seeded at 3.0 × 10⁴ cells/well in a white, clear, flat-bottom 96-well plate coated with 100 μg/mL poly-D-lysin (P7280, Sigma Aldrich) and incubated for 48 h at 37°C and 5% CO₂.

HEK293A.CXCR2 cells were transiently co-transfected with plasmids encoding Gα_i1 protein tagged with nanoluciferase (NLuc, donor) and Gγ₂ protein tagged with LSS-mKATE2 acceptor in a 1:10 donor to acceptor ratio, as described by Boon et al. (2023). Following the 48-h incubation period, cells were washed with the assay buffer. Compounds were serially diluted in 1:500 Nano-Glo^® Vivazine^™ working solution (#N2581, Promega), and 90 µL of compound/Vivazine mix was added, followed by incubation for 45 min at 37°C and 5% CO₂. Plates were transferred to the FLIPR Penta system (Molecular Devices) at 37°C, and all steps onward are automated and performed by the FLIPR Penta system. After a 10-min equilibration period, the baseline BRET signal was determined by five consecutive reads followed by automatic administration of 10 µL of 10× ligand (Table 1) to the cell plate. Changes in the BRET signal was measured for 10 min, with each read spaced by 2.5 s. Measurements were acquired using a 440- to 480-nm donor emission filter (#0200-6179, Molecular Devices) and a custom 615 nm AT600lp acceptor emission filter (#296420, Chroma).

HEK293A cells were transiently co-transfected with plasmids encoding β-arrestin1 protein tagged with NLuc (donor) and CXCR2 protein tagged with mNeonGreen (mNG) acceptor in a 1:10 donor to acceptor ratio, as described previously (Luís et al., 2022). All subsequent steps are identical to those in “NanoBRET-based G protein activation inhibition assay,” which are described above. Measurements were acquired using a 440- to 480-nm donor emission filter (#0200-6179, Molecular Devices) and a 515- to 575-nm acceptor emission filter (#0200-6203, Molecular Devices)

BRET ratios were extracted from ScreenWorks software (Molecular Devices) and represent the ratio between the acceptor emission of LSS-mKATE2 or mNG (G protein or β-arrestin recruitment, respectively) and NLuc donor emission. The basal BRET ratio was defined as the average BRET ratio of the five consecutive readings preceding ligand addition. Changes in BRET values post ligand administration (∆BRET) were calculated for each well as a percentage difference to the previously defined basal BRET ratio. Average ∆BRET values were baseline corrected by subtracting the average negative control BRET ratio (unstimulated condition). Subsequently, the negative area under the curve (AUC_neg) was used as the readout for G protein activation, whereas positive area under the curve (AUC_pos) was used as the readout for β-arrestin recruitment (calculation 1). AUCs were normalized to the positive control (stimulated condition); whereafter, dose–response curves were fitted to log (inhibitor) vs response (three parameters) in GraphPad 10.2.0 (GraphPad Software). Bottom values were constrained to 0 and top values were constrained to 100 in REGA-SIGN and β-arrestin recruitment to extract the IC₅₀ value.

Calculation 1: calculation of the BRET signal.

1. B R E T   r a t i o = 615 n m e m 460 n m e m .

For signaling bias calculations, constraints were set to bottom values equal to zero. The calculated top values generated by nonlinear fit log (inhibitor) vs response (three parameters) in GraphPad 10.2.1 were taken as I_max. I_max values were normalized to I_max of navarixin, which was set at 100%. The bias index values were calculated, as is shown in calculation 2. Briefly, log(I_max/IC₅₀) was calculated. Second, for each pathway, ∆log(I_max/IC₅₀) values are calculated by subtracting the log(I_max/IC₅₀) of the reference compound (i.e., navarixin) from the log(I_max/IC₅₀) of each compound. Finally, the bias index was calculated by subtracting the ∆log (I_max/IC₅₀) of the tested compound from the stated reference pathway from that of the pathway of interest. Notably, when I_max and IC₅₀ could not be determined due to low activation, log(I_max/IC₅₀) was taken as 0. The statistical significance of H1 was evaluated using a one-sample T-test with Benjamini-Hochberg correction applied for multiple comparisons, whereas H2 was assessed using a one-way ANOVA followed by Tukey’s test.

Calculation 2: calculation of the bias index.

1. log ⁡ I max I C 50 c o m p o u n d .

Bias plots are equimolar comparisons between the two pathways of interest. They are visualized by setting out the normalized concentration–response curve (normalized using the I_max of navarixin as 100%) of one pathway against the other.

Analysis of the biased inhibitory properties of small-molecule antagonists critically relies on the existence of an unbiased reference antagonist, to which the activity of the test compounds can be compared. We hypothesized that navarixin could serve as such an unbiased reference antagonist for CXCR2. To validate this, we assessed its ability to inhibit CXCR2-induced G protein dissociation and β-arrestin recruitment, respectively, using previously established NanoBRET-based assays. For the analysis of the inhibition of G protein activation, dissociation of the relevant G protein subtype Gα_i1 was quantified using the REGA-SIGN G protein biosensor assay (Boon et al., 2023). To quantify the inhibition of β-arrestin recruitment, we analyzed β-arrestin1 (βarr1) recruitment using the NanoBRET-based NanoLux assay (Luís et al., 2022). Although the CXCL8–CXCR2 signaling axis is the best characterized one, all ELR+ chemokines (CXCL1–3 and CXCL5–8) were included as agonists in both assays as they all activate CXCR2 and are associated with distinct pathophysiological functions (Cheng et al., 2019).

Briefly, HEK293A.CXCR2 or HEK293A cells were transfected with Gα_i1-91-NLuc and Gγ2-LSSmKATE2 (REGA-SIGN biosensor pair) or β-arrestin1.mNG and CXCR2.NLuc (NanoLux biosensor pair) encoding plasmids, respectively. The cells were subsequently incubated with the compound, which was serially diluted using an identical concentration range (15–0.021 µM) for 45 min. Finally, cells were stimulated with the different ELR+ chemokines (at their previously determined EC₈₀ value) for Gα_i1 or β-arrestin1 activation, respectively (Boon et al., 2024). The percentage inhibition was then calculated, and concentration–response curves were generated to determine the inhibitory efficacy (I_max) and potency (IC₅₀) values, which are shown in Table 2.

It is clear from the data in Table 2 that navarixin showed comparable efficacy and potency in inhibiting Gα_i1 activation and β-arrestin1 recruitment, irrespective of the ELR+ chemokine used. To further illustrate navarixin’s unbiased activity, a bias plot, which is an equimolar comparison plot in which the percentage inhibition of β-arrestin1 recruitment and Gα_i1 dissociation for the different CXCR2 endogenous ligands is compared, was constructed (Figure 1). It is clear that navarixin inhibits ELR+ chemokine-induced CXCR2 β-arrestin1 recruitment and Gαi1 activation in an unbiased manner for all endogenous CXCR2 ligands, with only slight deviations from the theoretical unbiased compound profile (dotted line). Hence, navarixin was considered as a balanced reference compound for further analysis.

To investigate the potential bias of the previously reported CXCR2 antagonists (Van Hoof et al., 2022), we applied both the Gα_i1 protein dissociation and β-arrestin1 recruitment assays, as described before (Table 2). Among the evaluated compounds, AZ10397767 and MVH-23, both thiazolo[4,5-d]pyrimidines carrying either an oxo or amino group at position 2, displayed similar efficacy as navarixin in inhibiting both pathways. Additionally, both the most potent compounds reported by the calcium mobilization assay (Van Hoof et al., 2022), that is, triazolo[4,5-d]pyrimidine MVH-9 and isoxazolo[4,5-d]pyrimidine MVH-24, were able to inhibit G protein activation and β-arrestin1 recruitment. However, MVH-9 and MVH-24 were found to be more potent in β-arrestin1 recruitment inhibition (with IC₅₀ values in the 0.03–0.17 µM range) than in Gα_i1 activation (IC₅₀ values in the 0.32–4.72 µM range). The other congeners were unable to inhibit Gα_i1 activation, with IC₅₀ values exceeding 2 µM, whereas they still displayed promising activity as inhibitors of β-arrestin1 recruitment, with IC₅₀ values of less than 100 nM for most derivatives. For most compounds, inhibition of CXCR2 was dependent on the chemokine ligand used.

For example, MVH-46 effectively inhibits CXCL2-, CXCL6-, and CXCL7-induced Gα_i1 protein activation at low micromolar concentrations (IC₅₀ values in the range of 2.6–3.5 µM). In contrast, when applying CXCL1 and CXCL8 as the agonists, MVH-46 inhibits Gα_i1 protein activation only at high micromolar concentrations (IC₅₀ values 12.9–14.0 µM), and it is completely unable to inhibit CXCL3-mediated Gα_i1 activation.

To further substantiate the potential biased activity of the MVH compounds, bias plots were generated for all compounds and compared with the antagonistic activity of navarixin (Figure 2). AZ10397767 and all MVH compounds consistently exhibited a bias toward the inhibition of β-arrestin1 recruitment over the inhibition of G protein activation (Gα_i1), following ELR+ chemokine-induced CXCR2 activation, compared to navarixin as the reference antagonist.

Notably, compounds AZ10397767, MVH-9, MVH-23, MVH-24, and MVH-32, while being biased toward β-arrestin1 inhibition, also demonstrated potent Gα_i1 protein activation inhibition at higher concentrations across most ligands, indicating that their biased activity is concentration-dependent. The bias profile also depends on the chemokine ligand used. For instance, MVH-9 exhibited a different inhibition profile for CXCL1-induced activation compared to other ELR+ chemokines. Similarly, MVH-24 displayed distinct bias profiles for CXCL3 and CXCL6, whereas MVH-32 showed a different bias profile for CXCL3-induced CXCR2 signaling inhibition. MVH-46 completely inhibited CXCL2-and CXCL6-induced Gα_i1 activation and β-arrestin1 recruitment at high concentrations, thus diminishing the biased profile at high concentrations. In contrast, when CXCL3 was applied as the ligand, MVH-46 did not reach 50% inhibition of the Gα_i1 protein activation pathway, maintaining a clear bias toward β-arrestin1 recruitment inhibition even at high concentrations.

To quantitatively assess biased inhibition profiles, the I_max and IC₅₀ parameters were used to calculate the bias index between the inhibition of the Gα_i1 protein activation (P2) and β-arrestin1 recruitment (P1) pathways, relative to navarixin (calculation 3). Given the difficulty in fitting curves for compounds that did not reach 50% inhibition, we assigned a log(I_max/IC₅₀) of zero to these cases. A positive bias index indicates compounds biased toward inhibiting β-arrestin1 recruitment over Gα_i1 activation using navarixin as the reference, with the absolute value indicating the extent of this bias.

The bias indexes shown in Figure 3A confirmed that the majority of the compounds displayed bias toward β-arrestin1 recruitment compared to Gα_i1 protein activation using navarixin as the reference. Moreover, variations in the bias index across the different ligands were observed. In order to definitively claim “bias,” two hypotheses were raised. The first hypothesis (H1) investigated if the bias index of the MVH compounds is significantly different from navarixin within a single ligand. For this purpose, one sample T-tests were performed for each compound independently, and the corresponding p-values are shown in Figure 3B. Interestingly, compounds MVH-9, MVH-23, MVH-24, MVH-32, MVH-46, and AZ10397767, which are the most potent CXCR2 antagonists in the calcium mobilization assay (Van Hoof et al., 2022), were nonsignificantly different from navarixin, for CXCL1-, CXCL2-, CXCL5-, CXCL6-, and CXCL8-induced responses. Additionally, MVH3 is nonsignificantly different from navarixin for CXCL2, CXCL5, and CXCL6 responses while being highly significant for CXCL1-induced responses. The remaining compounds are statistically different from navarixin across all ligands, with few exceptions. Overall, the majority of compounds behaved differently from navarixin, except for some compounds in CXCL1- and CXCL2-induced responses.

The second hypothesis (H2) aims to investigate if the observed switches between high or low bias indexes—indicative of significantly biased or unbiased β-arrestin1 recruitment inhibition—depend on the CXCR2 ligand applied. For this purpose, a one-way ANOVA, followed by Tukey’s test, was performed. Based on the p-values (Figure 3C), the observed differences for compounds MVH-3, MVH-32, MVH-33, MVH-46, MVH-52, MVH-61, and MVH-81 between the various CXCR2 ligands are statistically significant. Among these, MVH-3, MVH-46, and MVH-61 exhibited potent calcium mobilization antagonism. Altogether, this quantitative analysis confirmed that CXCR2 antagonism of these MVH compounds is generally biased toward β-arrestin1 recruitment inhibition and varied depending on the ligand stimulating CXCR2.