Effect of Phosphorylation Barcodes on Arrestin Binding to a Chemokine Receptor

Summary

Unique phosphorylation “barcodes” installed in different regions of an active seven-transmembrane receptor by different G protein-coupled receptor kinases (GRKs) have been proposed to promote distinct cellular outcomes¹ but it is unclear if or how arrestins differentially engage these barcodes. To address this, we developed an antigen binding fragment (Fab7) that recognizes both active arrestin2 (β-arrestin1) and arrestin3 (β-arrestin2) without interacting with bound receptor polypeptides, and employed it to determine structures of both arrestins in complex with atypical chemokine receptor 3 (ACKR3) phosphorylated in different regions of its C-terminal tail by either GRK2 or GRK5². The GRK2-phosphorylated ACKR3 resulted in more heterogenous “tail-mode” assemblies whereas phosphorylation by GRK5 resulted in more rigid “ACKR3-adjacent” assemblies. Unexpectedly, the finger loops of both arrestins engaged the micelle surface rather than the receptor intracellular pocket, with arrestin3 being more dynamic, in part due to its lack of a membrane anchoring motif. Thus, both the region of the barcode and arrestin isoform can alter the structure and dynamics of GPCR–arrestin complexes, providing a possible mechanistic basis for unique downstream cellular effects such as efficiency of chemokine scavenging and robustness of arrestin binding in ACKR3.

Introduction

Distinct phosphorylation patterns, or “barcodes”, installed by G protein-coupled receptor (GPCR) kinases (GRKs) have been invoked to explain how different GRK isoforms can lead to divergent cellular outcomes following GPCR activation. These barcodes have been suggested to modulate the arrestin responses of GPCRs, switching between effector activation and signaling versus receptor internalization^3,4,5. For example, phosphorylation of the β₂ adrenergic receptor (β₂AR) by either GRK2 or GRK6 initiates the desensitization of G protein signaling, but only GRK6 phosphorylation leads to ERK1/2 activation¹. The arrestin isoform involved has also been proposed to play a role in selecting downstream events as shown by depletion of arrestin2 (Arr2, also known as β-arrestin1) and arrestin3 (Arr3, also known as β-arrestin2) in smooth muscle cells, which results in distinct signaling and desensitization of the β₂AR⁶, the adenosine A_2B receptor⁶, the nucleotide receptor⁷, and the endothelin type A receptor⁷. SiRNA knockdown of Arr2 potentiates ERK signaling following activation of angiotensin II type 1A receptor (AT_1AR), whereas knockdown of Arr3 attenuates it⁸. Arr3 also has stronger influences than Arr2 in addiction and substance use disorders^9,10. Indirect evidence from simulations³, BRET assays^1,11,12, ¹⁹F-NMR¹³, and arrestin conformational sensors³ suggest that unique barcodes could drive different internal arrestin conformations that could lead to unique functional effects. However, the molecular basis for how different GRK barcodes differentially regulate the function of arrestin isoforms recruited to a given receptor remains poorly understood.

Atypical chemokine receptor 3 (ACKR3) works together with CXC chemokine receptor type 4 (CXCR4) to control the migration and localization of leukocytes and non-immune cells in processes ranging from immune homeostasis and development to cancer progression^14,15. In contrast to CXCR4, which is a canonical GPCR, ACKR3 does not activate heterotrimeric G proteins¹⁶. Instead, it is phosphorylated by GRKs and robustly recruits arrestins following activation by CXCL12 ^16–19. One of its primary roles is to scavenge CXCL12 by internalizing and trafficking the chemokine to lysosomes for degradation^20,21, a process that is crucial for maintaining CXCR4 responsiveness by preventing overstimulation and downregulation^22,23. We previously showed that GRK2 and GRK5 phosphorylate ACKR3 at distinct, yet partially overlapping, C-terminal positions depending on conditions leading to functionally different cellular outcomes in chemokine scavenging and the persistence of arrestin complexes². We have also determined atomic structures of activated ACKR3 with small molecule and chemokine agonists²⁴. Thus, ACKR3 is well suited for our efforts to better understand the molecular consequences of barcoding.

Here we present cryo-EM reconstructions of ACKR3 in complex with Arr2 or Arr3 after its in vitro phosphorylation by either GRK2 or GRK5, which are broadly expressed representatives of the two major GRK subfamilies. We show that although Arr2 and Arr3 bind to each phosphorylation barcode similarly, the structural and dynamic properties of the resulting receptor–arrestin assemblies are distinct and could underlie the functional outcomes instigated by instillation of GRK2 versus GRK5 barcodes in ACKR3. By extension, the results may be applicable to many other seven-transmembrane (7TM) receptors harboring regional, GRK-specific barcodes.

Results

GRKs Are Required for Arrestin Coupling.

We and others have shown that both GRK2 and GRK5 efficiently phosphorylate ACKR3^2,22 but that they phosphorylate distinct but overlapping regions of the ACKR3 C-terminus (Fig. 1a)². GRK2 phosphorylates Ser and Thr residues distal to the transmembrane (TM) core^2,22,25, whereas GRK5 installs phosphates at proximal sites, as well as sites overlapping those of GRK2 under some experimental conditions^2,22. Thus, this system enables assessment of whether GRK-specific barcodes in ACKR3 lead to distinct complexes with Arr2 and Arr3. To this end, we first assessed how phosphorylation affects arrestin binding to ACKR3 using pulldown assays with wildtype (WT) and preactivated forms of Arr2 and 3: Arr2* (I386A, V387A, F388A mutations in its C-tail), and Arr3* (truncated after residue 392) (Fig. 1b, c). To activate ACKR3, we used the chemokine variant, CXCL12_LRHQ^26,27. GRK5-phosphorylated ACKR3·CXCL12_LRHQ (pACKR3_GRK5) pulled down ~30% more Arr2* and ~40% more Arr3* than GRK2-phosphorylated ACKR3·CXCL12_LRHQ (pACKR3_GRK2) (Fig. 1b, c), suggesting that pACKR3_GRK5 binds tighter to arrestins than pACKR3_GRK2. Unphosphorylated ACKR3 failed to pull-down Arr2* or Arr3* (Fig. 1b, c), indicating that GRK phosphorylation, regardless of isoform, is required for efficient arrestin binding to activated ACKR3².

Fig. 1. Arrestin binding to ACKR3 and Fab7 depends on GRK2/5 phosphorylation.

a) Sequence of the ACKR3 C-tail with the observed GRK2 phosphorylation sites highlighted in orange and the unique sites observed for GRK5 in blue. GRK5 also phosphorylates the C-terminal GRK2 sites in cells²² or when ACKR3 is embedded in LMNG micelles but not in nanodiscs². b, c) A FLAG pulldown assay shows that pACKR3_GRK5 pulls down more Arr2*(b) and Arr3*(c) than pACKR3_GRK2. The interaction is abolished when ACKR3 is not phosphorylated (-ATP) and is weaker with WT Arr2 or Arr3. Arr3* is truncated and thus has a lower molecular weight than WT. WT Arr3 also robustly binds to pACKR3_GRK5. Two-sided one-way ANOVA followed by a Tukey multiple comparison test without adjustment was used to compare bound arrestin to pACKR3 ratios. Data are presented as mean ± S. D. from three technical replicates.

Analysis of ACKR3–Arr2/3*–Fab7 Complexes.

In most prior cryo-EM studies of GPCR–Arr complexes, a Fab optimized to bind to activated Arr2 and a hyper-phosphorylated C-tail peptide from the vasopressin receptor 2 (V₂R_pp) known as Fab30 was included to facilitate particle alignment^12,28–33. However, Fab30 binding requires a specific phosphorylation site found in V₂Rpp³⁴ and furthermore does not bind arrestins in complex with pACKR3^4,35. Therefore, we used a phage display selection to screen for Fabs that selectively bind to active arrestins regardless of the sequence or the phosphorylated receptor element. For this purpose, we used Arr3* in complex with the small molecule activator inositol hexakisphosphate (IP₆). IP₆ promotes trimerization of Arr3* via its receptor-binding interfaces³⁶ and competes with the receptor phosphate binding sites, which decreases the possibility of selecting Fabs that directly compete with receptor binding. We identified one (Fab7) that binds selectively to both Arr2* and Arr3* in complex with ACKR3 phosphorylated by either GRK2 or GRK5 (Extended Data Fig. 1).

We then determined cryo-EM structures for a series of pACKR3–Arr2/3*–Fab7 complexes (Fig. 2a, b, Extended Data Table 1, Figs. 2, 3, and 4). The main comparative set consists of four structures: pACKR3_GRK5 and pACKR3_GRK2 solubilized in LMNG/CHS micelles, each in complex with either Arr2* or Arr3* (Fig. 2a). The resolution of the resulting maps varied among the four reconstructions, indicating different levels of dynamics or heterogeneity within individual proteins and/or intermolecular interfaces. Based on resolution and map quality, pACKR3_GRK5 promoted a more ordered, less heterogeneous complex than pACKR3_GRK2 for both the Arr2* and Arr3* complexes, suggesting that the barcodes differentially regulate how strongly arrestins interact with the receptor. Arr2* complexes also had overall higher resolution than those of Arr3*. Because all complexes were isolated using FLAG pulldowns of ACKR3, and because the phosphorylated tail of ACKR3 is observed in all reconstructions, the presence of arrestin-micelle-only complexes (i.e. lacking ACKR3) cannot explain the observed heterogeneity or the lack of well-defined receptor details in some of the structures.

Fig. 2. Fab7, a new arrestin conformational sensor, enables structure determination of pACKR3–Arr2*/3* complexes.

a) Sharpened maps of the pACKR3_GRK5–Arr2*–Fab7, pACKR3_GRK2–Arr2*–Fab7, pACKR3_GRK5–Arr3*–Fab7, and pACKR3_GRK2–Arr3*–Fab7 complexes. ACKR3 in all four assemblies is solubilized in LMNG/CHS detergent micelles. b) Sharpened map and model of the 3.0 Å pACKR3_GRK2–Arr2*–Fab7–V_HH complex in POPC/POPS nanodiscs (PDB entry 8TII). Densities for the ACKR3 TM domain and the nanodisc were not evident. The CDR-H2 and -H3 loop involved in arrestin hinge binding is highlighted in yellow and magenta, respectively. a, b) GRK2 or GRK5 phosphorylated ACKR3 C-terminus (pC-tail_GRK2 or pC-tail_GRK5, respectively) bound to the arrestin N-lobe is highlighted in red and pink, respectively. c) ELISA analysis of Fab7 competition assays reveals that Arr2* (IC₅₀ ~35 nM) competes with Arr3*·IP₆ for Fab7 binding more efficiently than WT Arr2 (IC₅₀ > 5 μM). Data are presented as mean ± S. D. from three technical replicates.

Fig. 3. The pACKR3_GRK5–Arr2* assembly features a unique “TM7-H8 Linker-mode” interaction and a novel finger loop-micelle interaction.

a) Sharpened maps (top, EMD-47700) and models (bottom, PDB entry 9E82) of the pACKR3_GRK5–Arr2*–Fab7 complex with Fab7 density omitted. The micelle boundary is indicated by black traces surrounding the receptor. b) Interaction of ACKR3 residues in the TM7-H8 linker region with the inter-lobe region of Arr2* (left, PDB entry 9E82) resembles that formed with the Glucagon receptor (middle, PDB entry 8JRU). The inter-lobe hydrophobic pocket of Arr2* in receptor-bound states is distinct from the Arr2 basal state (right, PDB entry 1G4M). c) The time-dependent BRET response following CXCL12 stimulation shows reduced coupling of Arr2 to ACKR3_F317/I318G compared to ACKR3 in both HEK293A and HEK293T cells. Data are presented as mean ± S. D. from three technical replicates. d) The diverse ways arrestins interact with receptors shown by comparison of Arr2* from the pACKR3_GRK5 structure (PDB entry 9E82) with preactivated Arr2 in the NTSR₁ (PDB entry 6UP7) and M₂R (PDB entry 6U1N) complexes after alignment of the receptor TM cores. Dotted lines indicate the approximate membrane boundary. e) Arr2 structure (PDB entry 1G4M) with the Cα atoms of four bimane-labeled sites indicated with pink spheres. Fluorescence spectra of Arr2-V70B, -S193B, -T246B, -L338B and Arr3-V71B, -S194B, and T247B alone (black), or in the presence of unphosphorylated ACKR3 (grey), pACKR3_GRK2 (orange) or pACKR3_GRK5 (blue). Data are presented as mean ± S. D. from three technical replicates.

Fig. 4. The role of the Arr2 finger loop in ACKR3 binding is influenced by the ACKR3 phosphorylation barcode.

a) A FLAG pulldown assay shows that titrating CID24 from a 2.5- to 20-fold molar excess relative to pACKR3 reduces Arr2* binding by 10-20% and a consistent amount of CID24 is pulled down with the pACKR3 and Arr2* complex. CID24 blocks access to the cytoplasmic cleft of activated ACKR3 (PDB entry 7SK6). b) A FLAG pulldown assay shows that pACKR3_GRK5 pulls down a comparable amount of Arr2*_FL4A, whereas pACKR3_GRK2 pulls down 30% less Arr2*_FL4A compared to Arr2*. a, b) Two-sided one-way ANOVA followed by a Tukey multiple comparison test with adjustment was used to compare bound arrestin to pACKR3 ratios. c) Fluorescence spectra of Arr2_FL4A-L338B alone (black), or in the presence of unphosphorylated ACKR3 (grey), pACKR3_GRK2 (orange) or pACKR3_GRK5 (blue). d) Binding curve of Arr2 and Arr2_FL4A to pACKR3_GRK5 based on the L338B fluorescence intensity. The estimated K_d for Arr2 is 0.20±0.1 μM and the estimated K_d for Arr2_FL4A is 0.30±0.1 μM. e) The time-dependent BRET response following CXCL12 stimulation shows reduced coupling of Arr2_ΔFL to ACKR3 compared to Arr2 at early time points. f) The BRET dose-response curve shows an approximately 50% reduction in Emax of Arr2_ΔFL compared to Arr2 at 2 minutes following CXCL12 stimulation. a-f) data are presented as mean ± S. D. from three technical replicates.

In parallel we also determined the structure of Arr2* in complex with pACKR3_GRK2 in POPC/POPS nanodiscs (Fig. 2b), wherein Fab7 was further stabilized with a variable heavy-chain (V_HH) domain that binds to the hinge of the Fab light chain³⁷. ACKR3 was not evident in the resulting map except for the phosphopeptide region (pC-Tail_GRK2, residues 351-358; Fig. 2b), suggestive of a “hanging” or “tail-mode” complex. The 3 Å resolution of this map enabled modeling of the structure of Fab7 and its interface with Arr2* at near atomic resolution (Fig. 2b, Extended Data Fig 5a). Importantly, Fab7 does not interact with the phosphorylated C-tail or TM domain of ACKR3, suggesting that it will interact with arrestin complexes irrespective of receptor identity, barcoding, and arrestin isoform.

The Fab7 interface is not compatible with the basal conformation of Arr2 (Extended Data Fig. 5b), suggesting that it is selective for the activated states of arrestins. To test this, we performed assays where the binding of Fab7 to Arr3*·IP₆ was competed with increasing concentrations of either WT Arr2 or preactivated Arr2* (Fig. 2c). Preactivated Arr2* displaced Fab7 with an IC₅₀ of ~35 nM, whereas WT Arr2 competed with an IC₅₀ > 5 μM. Similar results were obtained for WT Arr3 vs. preactivated Arr3* (Extended Data Fig. 5c). Further characterization revealed that Fab7 selectively traps Arr2 in an active conformation (Extended Data Fig. 5d) and enhances the binding of WT Arr2 to pACKR3 (Extended Data Fig. 5e).

Novel Features of ACKR3–Arr complexes.

The pACKR3_GRK5–Arr2* complex yielded our highest resolution micelle complex (3.4 Å overall but varying between 3.2 and 8 Å locally), and focused refinements on the receptor produced the strongest density, including bulges for conserved receptor residues such as Trp265^6.48 and Tyr268^6.51(superscript refers to the Ballesteros-Weinstein numbering system) (Extended Data Fig. 6a, b). Use of a novel conditional diffusion model³⁸ suggested the path of the helical backbone of the receptor, including conserved kinks in the TM spans, allowing for more confident placement of the receptor into the density (Extended Data Fig. 6c). The 7TM helices of ACKR3 are the most well-resolved elements of the receptor and most closely resemble those in the active complex of ACKR3·CXCL12_LRHQ²⁴ (Fig. 3a, Extended Data Fig. 6d).

Residues Ser316-Tyr322 in ACKR3, which link TM7 with Helix 8 (H8), are modeled with an extended conformation in the complex (Extended Data Fig. 6e) wherein the side chain of ACKR3-Phe317 packs with the side chains of Ile61, Ala62 and Val64 in TM1, and that of ACKR3-Ile318 with Arr2*-Leu129 on the β8 strand of the N-lobe and Arr2*-Ile241 and -Tyr249 on the C-loop (Fig. 3b). The interface bears some similarities with that observed in the glucagon receptor–Arr2 complex²⁸, in that two hydrophobic residues (Leu420 and Leu424 in H8) pack into the inter-lobe hydrophobic network of Arr2 formed by the β5 and β8 strands of the N-lobe and the C-loop (Fig. 3b). The resulting rearrangement of the hydrophobic network of arrestin (Fig. 3b) can potentially stabilize a specific rotational angle between the two lobes and, consequently, a specific active state of Arr2. The high conservation of the hydrophobic network (Extended Data Fig. 6f) suggests that many GPCRs might gain affinity through interactions with the inter-lobe region of arrestins. In addition to the TM7-H8 linker interaction, the side chains of Phe244 and Asp245 in the C-loop and the subsequent β17 strand of Arr2* also form direct contacts with TM1, 6 and 7 of ACKR3 (Extended Data Fig. 6g). These contacts appear to contribute to an ~7° bend in TM1 that begins near ACKR3-Ile60 and averts a collision between the C-terminal end of TM1 and the C-lobe of arrestin (Extended Data Fig. 6h).

To validate the observed interactions, we mutated ACKR3-F317 and -I318 to glycine (ACKR3_F317/I318G) and monitored the BRET between N-terminal GFP10-tagged Arr2 and C-terminal luciferase-tagged ACKR3. In all tested cell-lines, ACKR3_F317/I318G showed reduced Arr2 coupling, with the effect being more pronounced in HEK293A cells relative to HEK293T cells (Fig. 3c). Notably, ACKR3_F317/I318G lacked the rapid Arr2 coupling observed in the first five minutes of stimulation, which is dependent on GRK5/6. We also mutated three hydrophobic residues of Arr2* modeled in contact with ACKR3 to alanine (Arr2*_3A: L129A, I241A, and Y249A). Pulldown assays showed that Arr2*_3A exhibited ~40% reduced binding to both pACKR3_GRK5 and pACKR3_GRK2 (Extended Data Fig. 7a). However, because these residues are involved in the inter-lobe interactions of Arr2*, the mutations seem to also alter stability and folding (Extended Data Fig. 7b–d). Consequently, the decrease in pulldown efficiency cannot be attributed solely to the loss of contact with ACKR3.

Another striking feature of the pACKR3_GRK5–Arr2* complex is that the Arr2* finger loop engages the micelle and not the receptor cytoplasmic cleft (Fig. 3a). This is in stark contrast to most other reported GPCR–arrestin complex structures (i.e., those involving rhodopsin, NTSR₁, M₂R, β₁AR, V₂R, and 5-HT_2A serotonin receptor)^{12,29–33,39,40}. The finger loop, which could not be accurately modeled due to its dynamics and proximity to the micelle surface, contains several highly conserved hydrophobic residues (Leu68, Val70, Leu71, and Leu73) (Extended Data Fig. 6f) whose side chains could interact with the hydrophobic phase of the micelle. As anticipated, the 338-loop of Arr2* (residues 334-341, part of the C-edge) also engaged the micellar boundary (Fig. 3a) similar to what has been observed in other GPCR–preactivated Arr2 complexes^{12,28–32,39}. The 193-loop of Arr2* (residues 191-196) are also in position to interact with the micelle. Regardless, the overall arrangement of Arr2* with respect to the TM core of ACKR3 and the micelle surface is distinct from prior structures (Fig. 3d).

To confirm the observed Arr2* interfaces with the micelle, we labeled Arr2 variants V70C, T246C, S193C, and L338C with monobromobimane (B), which installs an environmentally sensitive fluorophore into the finger loop, the C-loop and the C-edge, respectively. We then assessed whether these regions were involved in micelle binding based on fluorescence changes (Fig. 3e). In the presence of pACKR3_GRK5, all sites exhibited increases in intensity and/or changes in peak wavelength (Fig. 3e). The largest increases were observed for the V70B and L338B sites, consistent with the finger loop and C-edge moving into more hydrophobic environments. In comparison, pACKR3_GRK2 induced smaller changes in the finger loop (V70B) and C-edge (L338B) and no changes in the other two regions, indicating either a loosening or lower occupancy of these interfaces. The overall reduced responses are consistent with Arr2* interacting with pACKR3_GRK2 incorporated micelles either less tightly or often in a tail-mode configuration.

The Role of the Arr2 Finger Loop.

We first examined whether CID24, a Fab that fully occupies the cytoplasmic cleft of activated ACKR3 (Fig. 4a)²⁴, can displace Arr2*. In-vitro assays showed that the amount of Arr2* pulled-down with the receptor decreased by approximately 10-20% but that a consistent amount of CID24 was pulled down with the pACKR3–Arr2* complex even in the presence of a 20-fold excess of CID24 (Fig. 4a). This suggests that either CID24 competes with a small percentage of complexes that do access the cytoplasmic cleft, or that it interacts with the pACKR3–Arr2* complex in a manner that affects Arr2* binding allosterically. Consistent with the latter interpretation, pACKR3, Arr2* and CID24 coeluted via size exclusion chromatography (Extended Data Fig. 7e). In contrast, CID24 efficiently blocked GRK2 and GRK5 phosphorylation of ACKR3 (Extended Data Fig. 7f). Thus, GRKs, but not Arr2*, are strongly dependent on access to the cytoplasmic cleft of ACKR3.

We next mutated the four conserved hydrophobic sites in the Arr2 finger loop to create the FL4A variant (L68A, V70A, L71A, and L73A) and tested its impact on arrestin binding to pACKR3 via pull-down assays. A comparable amount of Arr2*_FL4A and Arr2* were pulled down with pACKR3_GRK5, whereas Arr2*_FL4A binding to pACKR3_GRK2 decreased by ~30% (Fig. 4b). We also introduced the FL4A mutations into the Arr2-L338B variant and measured the fluorescence change upon interaction with pACKR3. Arr2_FL4A-L338B exhibited fluorescence changes similar to those of Arr2-L338B in response to pACKR3_GRK5 but showed markedly reduced changes in response to pACKR3_GRK2 compared to Arr2-L338B (Fig. 4c versus Fig. 3e). Altogether, the results suggest that the contribution of the Arr2 finger loop to the overall binding affinity depends on the ACKR3 phosphorylation barcode and is greater for pACKR3_GRK2 than for pACKR3_GRK5. We then titrated pACKR3_GRK5 against Arr2-L338B or Arr2_FL4A-L338B and estimated the K_d from changes in the fluorescence intensity (Fig. 4d). The affinity of Arr2 and Arr2_FL4A for pACKR3_GRK5 were comparable, indicating that the finger loop residues are dispensable for pACKR3_GRK5–Arr2* complex formation. We could not perform the parallel experiment for pACKR3_GRK2 due to the lack of fluorescence response of Arr2_FL4A-L338B to pACKR3_GRK2.

To assess the impact of the finger loop on Arr2 recruitment to ACKR3 in cells, we monitored the BRET signal between N-terminal GFP10-tagged Arr2 and C-terminal luciferase-tagged ACKR3 in HEK293T cells where ACKR3 phosphorylation is dominated by GRK5². Interestingly, the deletion of the finger loop motif (residues 63-77, Arr2_ΔFL) reduced Arr2 recruitment to 50% at early stages but was indistinguishable from WT Arr2 at later time points (Fig. 4e, Extended Data Fig. 7g). At 2 minutes following CXCL12 stimulation, the E_max of Arr2_ΔFL was reduced to 50% of that of WT Arr2 (Fig. 4f), whereas at 20 minutes, Arr2_ΔFL showed the same response as WT Arr2 (Extended Data Fig. 7g). This suggests that the Arr2 finger loop is important for the initial, rapid coupling with ACKR3, but dispensable after complex formation.

Effects of Barcode and its TM Proximity.

The density for the phosphorylated region of the receptor C-tail is well resolved in the pACKR3_GRK5–Arr2* complex and is similar to that of the pACKR3_GRK2–Arr2* complex in nanodiscs, despite different barcodes (Fig. 5a, b). Both densities are consistent with the binding of a 7-residue polypeptide that contains two well-ordered phosphosites at the second and fifth position. There are three sequences in the C-tail of ACKR3 that fit this motif: 335-YpSAKpTGL-341, 337-KpTGLpTKL-343, and 351-EpTEYpSAL-357. The 351-EpTEYpSAL-357 sequence fits into the density of pACKR3_GRK2–Arr2* with good confidence, while 337-KpTGLpTKL-343 fits into that of pACKR3_GRK5–Arr2* (Fig. 5a, b, Supplementary Data Fig. 1). Both phosphopeptides form similar interactions with Arr2*. The first phosphosite (pThr338 or pThr352) is engaged by Arr2*-Lys11 and -Arg25 on the N-lobe, and Arr2*-Lys294 in the gate loop (Fig. 5a, b), whereas the second phosphosite (pThr341 or pSer355) is coordinated by the side chains of Arr2*-Lys10 and -Lys107. In addition, side chains of residues at the 4^th and 7^th positions make van der Waals contacts with residues in β1 and the αN helix of Arr2*. The residue in the 7^th position is conserved among the three phospho-motifs. Thus the consensus sequence for the ACKR3 barcodes is X(pT)XX(pS/T)Xϕ (where X is any amino acid, and ϕ is a medium sized hydrophobic residue).

Fig 5. GRK barcoding dictates distinct arrestin binding modes to ACKR3.

a) Interactions of the C-tail of pACKR3_GRK5 with the Arr2* N-lobe in the pACKR3_GRK5–Arr2*–Fab7 complex (PDB entry 9E82). b) Interactions of the C-tail of pACKR3_GRK2 with the Arr2* N-lobe in the pACKR3_GRK2–Arr2*–Fab7 complex in nanodiscs (PDB entry 8TII). a, b) The electron density of the ACKR3 phosphopeptide is shown as a wire cage contoured at 10σ. Phosphate contacts within 4 Å are shown as dashed lines. c) Twelve glycine residues were inserted between H8 and the C-tail of ACKR3 (ACKR3_+12G) to extend the GRK5 barcode to approximately where the GRK2 barcode begins. GRK2 phosphorylation sites are colored orange and GRK5 unique phosphorylation sites are colored blue. d) Cartoon showing how the distance between the GRK barcode and ACKR3 TM core dictates Arr2* binding modes. Cartoon is created with BioRender.com. e) Sharpened map of the pACKR3_{+12G GRK5}–Arr2*–Fab7 complex (EMD-41297) indicates a more heterogenous assembly.

Overall, the similarity of the bound GRK2 and GRK5 barcodes and the fact that Arr2 does not dramatically change its conformation when binding to either (RMSD 0.67 Å for 302 Cα atoms at moderate resolution) suggest that the sequence of the phosphorylation barcode itself is not what induces differences in the strength of the interaction with Arr2* or the distinct assemblies revealed by cryo-EM for the pACKR3_GRK5 and pACKR3_GRK2 complexes with Arr2*. The GRK5 barcode is, however, located 14 residues closer to the TM core of ACKR3 than the GRK2 barcode (Fig. 5c). The shorter distance in the pACKR3_GRK5 complex brings Arr2* closer to the TM domain, likely reinforcing its interactions with both the TM spans and TM7-H8 linker region of the receptor and the micelle surface (Fig. 5d). To test this idea, we inserted a flexible linker containing twelve glycine residues after ACKR3-Tyr334 (ACKR3_+12G) to extend the GRK5 barcode to approximately where the GRK2 barcode begins (Fig. 5c). This variant is still efficiently phosphorylated by GRK5 and forms a stable complex with Arr2*, but the resulting 3D reconstruction is more heterogeneous (Fig. 5e, Extended Data Fig. 8a, b), resembling that of the pACKR3_GRK2–Arr2* complex (Fig. 2a) and suggesting loss of the TM7-H8 linker interaction because of the 12G insertion (Fig. 5d). Density for the ACKR3 phospho-tail in this reconstruction is too ambiguous to discern whether the GRK5 unique phosphosite, the sites common to GRK2 and GRK5 (or a mixture of both) are bound.

Distinct Binding Modes of Arr2/3.

To test if there are notable differences in how Arr2* and Arr3* bind to pACKR3, we determined reconstructions of pACKR3_GRK5–Arr3* and pACKR3_GRK2–Arr3* (Fig. 2a, Extended Data Fig. 4). In the reconstruction of pACKR3_GRK5–Arr3*, the receptor helices are not as clearly resolved as in the analogous Arr2* complex, and the C-lobe of Arr3* is less resolved (Fig. 2a, 6a). This is not surprising because Arr3* lacks the membrane-anchoring C-edge found in Arr2* (Extended Data Fig. 6f) and consequently the C-lobe of Arr3* is not associated with the micelle (Fig. 6a). However, the finger loop of Arr3* still engages the micelle (Fig. 6a). The micellar boundary thus shifts from bracketing the length of the C-lobe in the Arr2* complex to being centered on the finger loop of Arr3* (Fig. 3a versus Fig. 6a). To test whether lack of the 338-loop found in Arr2* is responsible for this difference, we inserted the 338-loop of Arr2* into Arr3* and determined the cryo-EM reconstruction of this chimera in complex with pACKR3_GRK5 (Fig. 6b, Extended Data Fig. 8c, d). The chimeric C-lobe of Arr3* became more ordered and anchored to the detergent micelle, and the helices of ACKR3 also became more defined (Fig. 6a versus Fig. 6b).

Fig. 6. Arr3* binds to ACKR3 in a unique way compared to Arr2*, but with a similar response to barcoding by different GRK isoforms.

a, b) Sharpened map of the pACKR3_GRK5–Arr3*–Fab7 complex (a) and the pACKR3_GRK5–Arr3*_+C-edge–Fab7 complex (b) with Fab7 omitted. The surface of the LMNG detergent micelle is transparent to highlight the relative position between Arr3* and the micelle. c) Cartoon showing how the distance between the GRK barcode and ACKR3 TM core dictates Arr3* binding modes. Cartoon is created with BioRender.com.

To further compare the identified receptor and micelle interfaces of Arr2 and Arr3 as a function of barcoding, we labeled the Arr3 variants V71C, T246C and S194C with monobromobimane (B), thereby introducing the fluorophore into the finger loop, the C-loop and its 194-loop (residues 191–196), respectively. pACKR3_GRK5 induced comparable changes in Arr3-V71B, smaller changes in Arr3-S194B and distinct changes in T247B relative to Arr2 (Fig. 3e). These results suggest that the finger loops of Arr2 and Arr3 serve similar roles in both assemblies, whereas the C-lobes adopt distinct configurations in Arr2 and Arr3. Interestingly, pACKR3_GRK2 only induced subtle changes in the fluorescence of Arr3-V71B, much less so than those observed for the analogous Arr2-V70B (Fig. 3e). This is also consistent with the less extensive interactions between Arr3* and pACKR3_GRK2-incorporated micelles observed in our reconstruction (Fig. 2a).

Despite more heterogeneity, we were able to isolate a class of particles centered on Arr3* and Fab7 for examination of the C-tail interactions of Arr3* with the GRK2 barcode of ACKR3 (Extended Data Fig. 9a). The GRK2 and GRK5 phosphorylated C-tails adopt similar configurations when bound to Arr3* and engage residues analogous to Arr2* (Extended Data Fig. 9a), which is not surprising because these residues are highly conserved between Arr2 and Arr3⁴¹. We note that although the conformation and interactions of Arr3* with the C-tails of pACKR3_GRK2 and pACKR3_GRK5 both exhibit the expected active conformation for Arr3*³⁶ and are similar binding for each barcode (Extended Data Fig. 9a), a prior crystal structure of Arr3* in complex with ACKR3 phosphopeptides (PDB entry 6K3F) exhibited large conformational differences among the six copies of the complex in the model⁴². However, the presence of many structural implausibilities in 6K3F suggests that its structure determination is unreliable.

Conformational Landscape of Arrestins.

Although the differences in conformation and dynamics of the pACKR3–Arr2/3* assemblies due to the distinct GRK barcodes were profound, differences in internal arrestin conformation as a function of barcode were more subtle: the RMSD between the pACKR3_GRK2 vs pACKR3_GRK5 complexes was 0.67 Å (based on 302 Cα atoms) and 0.45 Å (based on 282 Cα atoms) for Arr2* and Arr3*, respectively (Extended Data Fig. 9b). To better understand how our structures map in the conformational space of reported arrestin structures, we used principal component analysis (Extended Data Fig. 9c, d, Supplementary Data Table 1). The two largest conformational variances among the arrestins were PC1 (~79% of the conformational variance), which corresponds to the well-known inter-lobe twist observed upon arrestin activation (Supplementary Data Video 1), and PC2 (~8.9% of the variance), which corresponds to a “wag” of the C-lobe relative to the N-lobe (Supplementary Data Video 2). Structures with PC1 values over 15 generally correspond to what has been described as activated states.

However, comparison of arrestin conformations is complicated by potential biases introduced by use of preactivated forms of Arr2/3, stabilizing Fabs (Fab7 in this study, Fab30 in others), or crystal lattice contacts in X-ray structures. Receptor complexes with Fab30 tightly cluster around PC1 values of 31-40 (Extended Data Fig. 9c), likely due to the fact that many of these structures involve the same V₂R_pp–Arr2 interaction⁴¹. In the structure of NTSR₁–Arr2 complex determined without Fab30³⁹, the PC1 and 2 values are markedly different from those of the structure of NTSR₁–Arr2 complex determined with Fab30³² (Extended Data Fig. 9c), suggesting possible bias due to Fab30. Our structures show a tight distribution (PC1 values 18-28) with a degree of twist more consistent with that of the NTSR₁–Arr2 complex determined without Fab30 (PC1 value 16)³⁹. It is clear from our work that Arr2* and Arr3* themselves have distinct activated conformations, even when bound to the same receptor, with Arr2* having ~6° more twist that Arr3* (PC1) and different degrees of wag (PC2). This does not seem to be due to the additional C-edge interaction of Arr2* because Arr3*_+C-edge exhibits the same conformational difference (Extended Data Fig. 9c).

Discussion

Herein we explored the molecular consequences of GRK barcodes installed in distinct regions of the ACKR3 C-tail on the binding of Arr2 and Arr3. The development of Fab7 was key to this study because it i) facilitated our cryo-EM reconstructions, ii) stabilized active states of both Arr2 and Arr3, and iii) unlike commonly used Fab30, does not directly bind to a phosphosite in the receptor tail and thus avoids any sequence bias. Unexpectedly, neither Arr2 nor Arr3 engaged the cytoplasmic cleft of activated ACRK3. In the highest resolution case where Arr2 contacts the TM spans, the region of the receptor involved is not known to undergo large conformational change during activation. Thus, “arrestin” bias of ACKR3 may really be “GRK” bias because the key activation event triggering arrestin binding is the phosphorylation barcode itself. We note that the cytoplasmic cleft of the Class B glucagon receptor also does not interact with Arr2²⁸, implying that there may turn out to be a large degree of heterogeneity in the way arrestins engage 7TM receptors. This idea is supported by cell-based crosslinking studies⁴³ and the fact that such heterogeneity would allow arrestins to bind to a much wider variety of receptors. We further showed that the regional position of the barcode can influence the overall configuration of the receptor–arrestin complex by constraining the position of the arrestin relative to the receptor and micelle surface. Alterations in the internal conformation of the bound arrestin to each barcode in comparison seem subtle, consistent with a recent finding that Arr3 does not undergo discernable conformational changes in response to β₂AR with different barcodes in cells⁴⁴. Because Fab7 binds to both Arr2 and Arr3, we were able to compare their binding modes to ACKR3 and observed that the most marked difference is the way they engage the micelle surface and the higher dynamics of Arr3 itself. Arr3 has been reported to be intrinsically more flexible than Arr2⁴⁵, which may contribute to the higher heterogeneity in the Arr3 complexes, but in addition the C-lobe of Arr3* does not anchor to the micelle/membrane via its C-edge, affording Arr3* greater flexibility in its orientation and perhaps the ability to recruit more effectors than Arr2*⁴⁶. These isoform dependent differences could also underly distinct downstream signaling outcomes. To this point, Arr2 has a short splice variant that lacks the C-edge and seems to play different roles than full length Arr2 in cells⁴⁷.

The fact that Fab7 is agnostic to the phosphopeptide bound to arrestin allowed us to expand upon what is known about arrestin binding motifs. The bound X(pT)XX(pS/T)Xϕ peptides of ACKR3 is similar to the proposed (pS/T)X_1-2(pS/T)XX(pS/T) motif⁴⁰ and a generalization of the (pS/T)X(pS/T)(pS/T) motif⁴ (bold letters indicating common arrestin-binding phosphosites). Residues 1-2 amino acids N-terminal to the GRK2 and GRK5 barcodes of ACKR3 are also phosphorylated, although not observed in our structures, and thus would be equivalent to the first P in the (pS/T)X_1-2(pS/T)XX(pS/T) motif. A search of the sequences of Class A GPCRs in GPCRdb.org shows that a simple PXXP motif (where P can be pSer, pThr, or the phosphomimetic residues Glu or Asp) can be found in 91% of their C-tails. A terminal hydrophobic anchor like that observed in ACKR3, or PXXPXϕ, is found in 63% of the C-tails, indicating that it is a common feature of potential barcodes. The PXPP motif is found in 68% of the C-tails, far more than a PXPPXϕ motif (29%). Based on this simple analysis, we speculate that for tighter binding, arrestins need a minimum of two closely spaced phosphorylated/negatively charged residues (as in PXXP), plus either a hydrophobic anchor or a third phosphorylated residue (as in PXXPXϕ or PXPP, respectively).

GRK5 is the dominant kinase when ACKR3 is expressed alone in HEK293 cells. GRK2 becomes important when Gβγ subunits are released by co-activation of CXCR4 by CXCL12, which augments the efficiency of chemokine scavenging and alters the stability of ACKR3–arrestin complexes². Use of two different barcodes in ACKR3 may thereby serve as a sensor of the activation state of CXCR4 and regulate how much CXCL12 to scavenge². Interestingly, CXCR4 also has a proximal barcode installed by GRK6 (a close homolog of GRK5), and a distal barcode installed by GRK2/3 and distinct cellular outcomes have been assigned to each⁴⁸. We speculate the distinct configurations we observed for Arr2 and Arr3 binding to the barcodes ACKR3 might be a large part of the molecular basis for not only the differential responses in ACKR3, but also for the signaling behaviors of CXCR4 and other receptors such as the β₂AR¹, AT_1AR^44,49, parathyroid hormone 1 receptor¹¹, and μ opioid receptor⁵⁰ that have regional differences in the location of their GRK2 and GRK5 phosphosites.

Methods

Expression and purification of ACKR3 variants.

Human ACKR3 was co-expressed with CXCL12_LRHQ in Sf9 cells as previously described²⁴. Sf9 cell lines were purchased from ATCC and have not been authenticated and have not been tested for mycoplasma contamination. CXCL12_LRHQ has a four-residue substitution of the first three N-terminal residues (Leu-Arg-His-Gln, starting with Leu0) that prolongs its residence time on the receptor relative to WT CXCL12. Briefly, Sf9 cells were infected (multiplicity of infection of 6 for each virus) with separate baculoviruses (prepared using the Bac-to-Bac Baculovirus Expression System, Invitrogen) bearing the CXCL12_LRHQ and the ACKR3 genes. The ACKR3 coding sequence spanned residues 2-362 with an N-terminal HA signal sequence and tandem C-terminal 10xHis and FLAG purification tags. After 48 hours, the infected cells were harvested by centrifugation and the membranes prepared by four rounds of dounce homogenization, first in hypotonic buffer containing 10 mM HEPES (pH 7.5), 10 mM MgCl₂, and 20 mM KCl, followed by three more washes with hypotonic buffer plus 1 M NaCl. The membranes were spun down at 50,000 x g for 30 min and resuspended between each round of douncing. The samples were then solubilized in 50 mM HEPES pH 7.5, 400 mM NaCl, 0.75/0.15% lauryl maltose neopentyl glycol/cholesteryl hemisuccinate (LMNG/CHS) with a protease inhibitor tablet (Roche) for 4 hours. Insoluble material was removed by centrifugation at 50,000 x g for 30 min. Talon resin (Clontech) with 20 mM imidazole was added to the soluble fraction and incubated overnight at 4 °C. The resin was then transferred to a plastic purification column and washed with washing buffer 1 containing 50 mM HEPES (pH 7.5), 400 mM NaCl, 10% glycerol, and 20 mM imidazole (pH 8.0), plus 0.1/0.02% LMNG/CHS, followed by washing buffer 1 plus 0.025/0.005% LMNG/CHS and finally eluted with washing buffer 1 plus 0.025/0.005% LMNG/CHS and 250 mM imidazole (pH 8.0). Imidazole was removed with a desalting column (PD MiniTrap G-25, GE Healthcare). The final protein concentration was determined by A₂₈₀ using an extinction coefficient of 85000 M⁻¹cm⁻¹, snap frozen in liquid nitrogen, and stored at −80 °C for later use. A similar strategy was used to prepare complexes of ACKR3_+12G.

Nanodisc reconstitution of ACKR3.

ACKR3 purified in LMNG/CHS was reconstituted into nanodiscs using MSP1D1 or MSP1E3 as scaffolding proteins. POPC and POPS lipids (Avanti) were mixed at a 6:4 ratio, dried under a nitrogen air stream, and placed in a desiccator for at least 30 minutes to remove any residual solvent. Sodium cholate was added to dissolve the lipids. The lipids, scaffolding protein, and ACKR3 were then combined at a molar ratio of 240:6:1 and incubated on ice for 30 minutes. Bio-Beads were added to the mixture, which was incubated on a shaker in a cold room overnight. The ACKR3 in nanodiscs was stored at 4 °C for later use.

Expression and purification of GRK5.

Human GRK5 was expressed and purified as previously described⁵¹. Briefly, a pMAL plasmid containing human full-length GRK5 with a C-terminal 6xHis tag was transformed into E. coli Rosetta cells. The expression of GRK5 was induced with 200 μM IPTG at an OD around 0.6-0.8 and the cultures were incubated with shaking at 18 °C overnight. Harvested cell pellets were resuspended and homogenized in lysis buffer containing 20 mM HEPES (pH 8.0), 400 mM NaCl, 0.1% Triton-X (v/v), 2 mM DTT, DNase, 0.1 mM PMSF, leupeptin, and lima bean trypsin protease inhibitor, then lysed using an Avestin C3 emulsifier and centrifuged at 18,000 rpm for 30 min. The supernatant was combined and loaded onto a 3 ml home-packed Ni²⁺-NTA column pre-equilibrated with buffer A containing 20 mM HEPES (pH 8.0), 400 mM NaCl and 0.5 mM DTT. The column was then washed with 50 ml buffer A, followed by 100 ml buffer B containing 20 mM HEPES (pH 8.0), 100 mM NaCl and 0.5 mM DTT plus 20 mM imidazole (pH 8.0). The bound protein was eluted with buffer B plus 200 mM imidazole (pH 8.0) and then loaded onto a linked 1 ml HiTrap Q HP (Cytiva) and 1 ml HiTrap SP HP (Cytiva) columns. A linear NaCl gradient (0.1-0.6 M) was used to elute GRK5 from the SP column. GRK5 elutes at ~0.3-0.5 M NaCl. The fractions containing GRK5 were combined, concentrated with a 50 kDa cutoff Amicon concentrator to ~500 μl, then further purified using a Superdex 200 Increase 10/300 GL column equilibrated with 20 mM HEPES (pH 8.0), 100 mM NaCl, and 0.5 mM TCEP. The peak fractions were collected, concentrated with a 50 kDa cutoff Amicon concentrator, and stored at −80 °C.

Expression and purification of GRK2.

Human GRK2-S670A (otherwise referred to in this paper as just GRK2) was expressed and purified from Sf9 cells as previously described². Briefly, GRK2 with a C-terminal 6xHIS tag was expressed using the Bac-to-Bac insect cell expression system (Life Technologies). The insect cells were harvested 48 hours post-infection and homogenized with buffer containing 20 mM HEPES (pH 8.0), 400 mM NaCl, 2 mM DTT, 1 mM PMSF, leupeptin, and lima bean trypsin protease inhibitor. The cells were lysed using an Avestin C3 emulsifier and clarified by centrifugation at 35,000 g for 60 min. The supernatant was then passed over an immobilized metal ion affinity chromatography as described above for GRK5. The purity of GRK2 after this step was ~90%. Fractions containing GRK2 were pooled and further purified on a Superdex 200 Increase 10/300 GL column equilibrated with 20 mM HEPES (pH 8.0), 100 mM NaCl, and 0.5 mM TCEP. The peak fractions were collected, concentrated with a 50 kDa cutoff Amicon concentrator, and stored at −80°C.

Expression and purification of arrestin variants.

Expression and purification of bovine Arr2 and Arr3 were as described previously⁵². Briefly, the pTrcHisB plasmid containing bovine Arr2 or Arr3 (and variants thereof) was transformed into E. coli Rosetta cells and protein expression was induced with 25 μM (Arr2) or 37.5 μM (Arr3) IPTG for 4 hours at 30 °C. The cell pellets were resuspended and homogenized in buffer containing 20 mM MOPS (pH 7.5), 400 mM NaCl, 5 mM EDTA, 2 mM DTT, 1 mM PMSF, leupeptin, and lima bean trypsin protease inhibitor. Cells were lysed using an Avestin C3 emulsifier and the lysate was clarified by centrifugation at 18,000 rpm for 30 min. The supernatant was collected and arrestin was precipitated by the addition of (NH4)2SO4 to a final concentration 0.32 mg/ml. Precipitated arrestin was collected by centrifugation at 18,000 rpm for 90 min. The pellet was then dissolved in buffer containing 20 mM MOPS (pH 7.5), 2 mM EDTA, and 1 mM DTT, and then centrifuged at 18,000 rpm for 60 min to pellet insoluble components. The supernatant containing soluble arrestin was applied to a heparin column and eluted with a linear NaCl gradient (0.2-1 M). Fractions containing arrestin were identified by SDS-PAGE and combined. For Arr2, the salt concentration of the pooled fractions was adjusted to 50 mM, loaded onto a 5 ml HiTrap Q HP column (Cytiva), and eluted with a linear NaCl gradient. For Arr3, the salt concentration of the pooled fractions was adjusted to 100 mM, and the solution was loaded onto a linked 1 ml HiTrap Q HP followed by a 1 ml HiTrap SP HP column. The columns were then uncoupled and a linear NaCl gradient (0.2-1 M) was used to elute Arr3 from the SP column. The fractions containing arrestin were concentrated with a 30 kDa cutoff Amicon concentrator to ~500 μl, then further purified using a Superdex 200 Increase 10/300 GL column (Cytiva) equilibrated with 20 mM MOPS (pH 7.5), 150 mM NaCl, and 0.5 mM TCEP. The peak fractions were collected, concentrated using a 30 kDa cutoff Amicon concentrator, snap frozen in liquid N₂ and stored at −80 °C.

Phosphorylation of ACKR3.

Phosphorylation conditions were optimized to achieve maximum phosphorylation of ACKR3. For GRK5, the phosphorylation reaction contained 2 μM ACKR3, 1 μM GRK5, 20 μM c8-PIP₂ and 200 μM ATP in phosphorylation buffer containing 20 mM HEPES (pH 8.0), 10 mM MgCl₂, and 0.002% LMNG. Everything except ATP was mixed first and incubated at room temperature for 20 min. ATP then added to initiate phosphorylation and the reaction allowed to proceed for one hour at room temperature. For GRK2, the phosphorylation reaction contained 2 μM ACKR3, 3 μM GRK2 and 200 μM ATP in phosphorylation buffer containing 20 mM HEPES (pH 8.0), 10 mM MgCl₂ and 0.02/0.002% LMNG/CHS. The phosphorylation reaction was incubated overnight at room temperature.

Arrestin pulldown.

ACKR3 (2 μM) was phosphorylated by GRK5 or GRK2 using the procedures described above. Arrestin variants (2.4 μM) and Fab7 (2.4 μM) were added to the ACKR3 phosphorylation reaction and incubated for one hour at room temperature. Anti-FLAG M2 magnetic beads (Sigma) were washed with buffer A containing 20 mM HEPES (pH 8.0), 100 mM NaCl, 0.01/0.001% LMNG/CHS and then added to the mixture. The sample tubes were incubated in a rotator for an hour at room temperature. Anti-Flag M2 beads were washed five times with 1 ml buffer A and then eluted with buffer A supplemented with 3xFLAG peptide (Sigma). Unphosphorylated ACKR3, used as a control, was prepared by omitting ATP in the first phosphorylation step. The eluted samples were analyzed by SDS-PAGE gel electrophoresis with Coomassie staining. The densities of bound arrestin and ACKR3 were quantified using Image Lab (Biorad) and the ratios between different samples were compared.

Trypsin digestion of arrestin.

WT Arr2 (10 μM) was incubated with 2-fold serial dilutions of 32 μM Fab7 at room temperature for 20 min. Trypsin (0.0015 mg/ml) was added to the reaction and the digestion was incubated for 10 min at room temperature. SDS loading buffer was used to quench the reaction. The samples were analyzed by SDS-PAGE gel electrophoresis with Coomassie staining.

Monobromobimane labeling of arrestin.

Arr2-V70C, -L338C and Arr3-V71C with all native cysteines mutated were prepared as described above in buffer containing 20 mM MOPS (pH 8.0) and 150 mM NaCl. A twenty-molar excess of freshly prepared mBrB (Sigma) was added to the reaction and incubated on ice overnight. The sample was loaded on a Superdex 200 Increase 10/300 GL column equilibrated with buffer containing 20 mM MOPS (pH 8.0) and 150 mM NaCl to get rid of excess mBrB. The fractions containing mBrB-labelled arrestins were collected, concentrated with a 30 kDa cutoff Amicon concentrator, and stored at −80 °C. The mBrB labeling level was estimated to be ~75% based on the mBrB fluorescence absorption at 383 nm and the protein absorption at 280 nm.

Fluorescence measurements.

The mBrB-labelled Arr2-V70C, -L338C or Arr3-V71C (2 μM) was incubated with 2 μM ACKR3 and 2 μM ACKR3 phosphorylated by GRK2 or GRK5 at room temperature for 20 min. The reactions were transferred to a 384 black plate (Corning) for fluorescence measurement using a plate reader (BioTek). The excitation wavelength was set to 375 nm and the absorption was monitored from 420 nm to 700 nm.

Preparation of biotinylated Arr3.

Arr3* at 1 mg/ml was incubated with 200 μM IP₆ for 20 min in buffer containing 20 mM MOPS (pH 8.0), 150 mM NaCl, and 0.5 mM TCEP. A twenty-molar excess of freshly prepared EZ-link sulfo-NHS-Biotin (ThermoFisher) was added to the reaction and incubated on ice for 2 hours. The sample was loaded onto a Superdex 200 Increase 10/300 GL column equilibrated with buffer containing 20 mM MOPS (pH 8.0), 150 mM NaCl, 0.3 mM TCEP and 0.2 mM IP₆ to remove excess sulfo-NHS-Biotin. Arr3* biotinylated in the presence of IP₆ eluted at the retention volume corresponding to a trimer and the peak fractions were collected for Fab selection. The biotinylation level was estimated to be >95% based on pulldown assays using avidin agarose beads.

Phage display selections.

Biotinylated Arr3* loaded with IP₆ was used for phage display selection. Phage display selection was performed at 4 °C according to published protocols⁵³. The selection buffer contained 20 mM MOPS (pH 8.0), 150 mM NaCl, 0.01% LMNG, 0.5% BSA and 0.2 mM IP₆. In brief, for the first round of selection, 200 nM of biotinylated Arr3*·IP₆ complex was immobilized on 250 μl streptavidin magnetic beads (Promega, Cat No: Z5482) and incubated with 100 μl of a phage library⁵⁴ containing 10¹² phage for 30 min. The resuspended beads containing bound virions were washed extensively and then used to infect freshly grown log phase E. coli XL1-Blue cells. Phages were amplified overnight in 2xYT media with 50 μg/ml ampicillin and 10⁹ p.f.u./ml of M13-KO7 helper phage. To increase the stringency of selection, four additional rounds of sorting were performed with the following concentration of biotinylated Arr3* in each round (second round, 50 nM; third round, 50 nM; fourth and fifth round, 10 nM) using the amplified pool of phage from the preceding round as the input. Selection from the second to fifth rounds was done on a Kingfisher automated purification instrument (Thermo Scientific) where the target was premixed with the amplified phage pool and then streptavidin beads were added to the mixture. From the second round onwards, the bound phages were eluted using 0.1 M glycine (pH 2.7). To eliminate the non-specific and streptavidin binders, the precipitated phage pool from the second round onwards were negatively selected against 100 μl of streptavidin beads before adding to the target. The pre-cleared phage was then used as an input for the selection.

Single-point phage ELISA.

All ELISA experiments were performed at 4°C in 96-well plates coated with 50 μl of 2 μg/ml neutravidin in Na₂CO₃ buffer (pH 9.6) and subsequently blocked by 1% BSA in PBS. A single-point phage ELISA was used to rapidly screen the binding of the obtained clones. Colonies of E. coli XL1-Blue harboring phagemids from 4^th and 5^th rounds of selection were inoculated directly into 500 μl of 2xYT broth supplemented with 100 μg/ml ampicillin and M13-KO7 helper phage. The cultures were grown overnight at 37 °C in a 96-deep-well block plate. The phage display selection buffer contained 20 mM MOPS (pH 8.0), 150 mM NaCl, 0.01% LMNG, 0.5% BSA and 0.2 mM IP₆. Culture supernatants containing Fab phage were diluted tenfold in the selection buffer. After 15 min of incubation, the mixtures were transferred to ELISA plates previously incubated with 40 nM biotinylated Arr3_ΔC in experimental wells and with buffer in control wells for 15 min. The ELISA plates were incubated with the phage for another 15 min and then washed with ELISA buffer. The washed ELISA plates were incubated with a 1:1 mixture of mouse anti-M13 monoclonal antibody (cat: 27-9420-01, GE, 1:5,000 dilution in ELISA buffer) and peroxidase conjugated goat anti-mouse IgG (cat: 115-035-003, Jackson Immunoresearch, 1:5000 dilution in ELISA buffer) for 30 min. The plates were washed again, developed with 3,3’,5,5’-tetramethyl-benzidine/H₂O₂ peroxidase substrate (TMB) (Thermo Scientific, Cat No: 34021) and then quenched with 1.0 M HCl, and the absorbance at 450 nm was read on a plate reader. Phagemid DNA from the clones from wells with a high signal/noise ratio were sequenced to identify the unique binders.

Sequencing, cloning, overexpression and purification of Fab fragments.

The sequencing, cloning, overexpression and purification of the Fab fragments were performed according to published protocols⁵⁵.

Multipoint protein ELISA for EC₅₀ determination.

A multipoint ELISA was performed at 4°C to estimate the affinity of the Fabs to Arr3*. The phage display selection buffer (20 mM MOPS; pH:8.0, 150 mM NaCl, 0.01% LMNG, 0.5% BSA) supplemented with 10 μM IP₆ was used as ELISA buffer. 40 nM of biotinylated target immobilized on a neutravidin coated ELISA plate was incubated with 3-fold serial dilutions of the purified Fabs starting from 4 μM for 20 min. The plates were washed, and the bound target-Fab complexes were incubated with a secondary HRP-conjugated Pierce recombinant protein L (cat: 32420, ThermoFisher, 1:5000 dilution in ELISA buffer) for 30 min. The plates were again washed, developed with TMB and quenched with 1.0 M HCl, and absorbance (A₄₅₀) was determined. To determine the affinities, the data were fitted in a dose response sigmoidal function in GraphPad PRISM and EC₅₀ values were calculated. EC₅₀ of Fab7 to Arr3*.IP6 is ~ 10 nM.

Fab7 competition assays.

An ELISA assay was used to determine that 50 nM Fab7 reached 50–70% of maximum binding to 25 nM biotinylated Arr3*·IP_6; thus, 50 nM was subsequently used in competition assays. For these experiments, 50 nM Fab7 was incubated separately with 3-fold serial dilutions of 3 μM competitors (Arr2, Arr2*, Arr3, or Arr3*) for 30 min. The samples were then transferred to ELISA plates containing 25 nM biotinylated Arr3*·IP₆ and incubated for 15 min to capture free Fab7. The plates were then washed with the ELISA buffer, and the bound Arr3*·IP₆–Fab7 complexes incubated with HRP-conjugated Pierce recombinant protein L (cat: 32420, Thermo Fisher, 1:5000 dilution in ELISA buffer) for 30 min. The plates were again washed, developed with TMB and quenched with 1.0 M HCl, and Fabs quantified by A₄₅₀. IC₅₀ values were calculated using GraphPad PRISM.

BRET assay to assess Arr2 recruitment to ACKR3.

HEK293T cells were purchased from ATCC. GRK knockout cell lines⁴⁹ and their parental HEK293A cells were generous gifts from the Inoue lab at Tohoku University. The cell lines were not authenticated but were regularly tested for mycoplasma contamination using the MycoAlert assay. The BRET assay measuring the signal between GFP10-tagged Arr2 and C-terminal luciferase-tagged ACKR3 was performed as described². Briefly, cells in suspension were transfected with 50 ng FLAG-ACKR3-RlucII, 1 μg of GFP10-Arr2, and 950 ng of pcDNA3.1 per 1x10⁶ cells using either polyethylenimine or the TransIT-LT1 transfection system (MirusBio). Transfected cells were plated at 3 x10⁴ cells per well in a white 96-well plate (Greiner) and cultured in Dulbecco’s Modified Eagle Media supplemented with 10% fetal bovine serum. After 48 hours, cells were washed once with PBS and maintained in either Hank’s Balanced Salt Solution supplemented with 0.1% bovine serum albumin or in Tyrode’s Buffer (25 mM HEPES, 140 mM NaCl, 2.7 mM KCl, 12 mM NaHCO₃, 5.6 mM glucose, 0.5 mM MgCl₂, and 0.37 mM NaH₂PO₄, pH 7.5). Cells were incubated with 5 μM of Prolume Purple (Prolume Ltd.) for 3-5 minutes before 2-3 baseline readings were taken. For the time curve, 100 nM of CXCL12 was used to stimulate ACKR3, whereas increasing concentrations of CXCL12 were used to generate a dose-dependent curve. Bioluminescence was measured at 410-80 nm and 515-30 nm using a PHERAstar plate reader (BMG LABTECH) or a TECAN Spark luminometer (Tecan Life Sciences) for 0.5-1 hours at 37 °C. BRET values were determined as the ratio of the red and blue luminescence readings (515-30nm/410-80nm). Results from three independent experiments were normalized to WT E_max and fit by a sigmoidal dose-response model using GraphPad PRISM.

Cryo-EM sample preparation and image acquisition.

FLAG pulldown assays described above were used to prepare all ACKR3–arrestin–Fab7 complexes reconstituted in either detergent or nanodiscs for cryo-EM. Quantifoil R1.2/1.3 300-mesh Cu grids were glow-discharged using EasiGlow at 25 mA for 60 seconds. Purified ACKR3–arrestin–Fab7 complexes (3.3 μl at ~0.6 mg/ml) were applied to the grids and the grids blotted with filter paper for 3.5 seconds before being plunge-frozen in liquid ethane using a Vitrobot MK IV (Thermo Fisher Scientific). Data were collected on a Titan Krios G4 electron microscope (FEI) equipped with a post-GIF K3 direct electron detector (Gatan) and a Quantum GIF energy filter (Gatan) in the Purdue Life Sciences Cryo-EM Facility. Micrographs were collected in super-resolution mode with a pixel size of 0.527 Å, at a defocus range of 0.6 to 2.5 μm using EPU, and 40 frames were recorded for each movie stack at a frame rate of 78 milliseconds per frame and a total dose of 53.8 electrons/Ų.

Data Processing for the pACKR3_GRK5–Arr2*–Fab7 complex.

All data processing steps were performed using Relion 5.0-beta2^56,57 unless otherwise noted. The image processing workflow is detailed in Extended Data Fig. 2. Briefly, after motion correction in Relion and CTF estimation using CTFFind 4.1.14⁵⁸, exposures were curated based on motion and CTF statistics outliers and the presence of non-vitreous ice contamination, resulting in 7,396 micrographs used for downstream processing. A random subset of 750 micrographs were processed first. Particle positions were picked from this subset using the general (pre-trained) model distributed with Topaz⁵⁹. These picks were filtered through iterative 2D classification, 3D initial model generation, and 3D classification. The resulting 3D consensus map was used as a reference to re-pick the micrograph subset, and these picks were similarly filtered, then combined (with duplicate removal) with the Topaz particles. To address orientation bias, approximately 30% of particles were removed by inspecting the results of a 2D classification and removing classes/particles from the dominant orientation. The remaining particles were used to train a Topaz model, which was used to pick the micrograph subset again. This process (dominant orientation removal, Topaz training) was repeated a second time, and the resulting trained Topaz model was used to pick the full micrograph set. The resulting picks were once again filtered by iterative 2D and 3D classification. After this converged, particles were subject to cycles of CTF refinement (per-particle defocus, per-micrograph astigmatism, whole-dataset beam tilt), fixed-pose 3D classification, and Bayesian polishing, until resolution failed to improve. Throughout this process, 3D auto-refinements utilized Blush regularization⁶⁰. To mitigate the effects of preferred orientation in the resulting map, the resulting particle set was then refined using spIsonet regularization⁶¹. Finally, focused classification on the region in and around ACKR3 was performed using signal subtraction coupled to fixed-pose 3D classification. A single class with the most contiguous density for ACKR3 (9.2% of input particles) was selected and refined using spIsonet regularization to produce the final map used for model refinement.

Cryo-EM data processing for remaining complexes.

Cryo-EM movies were imported to cryoSPARC and processed using the standard workflow⁶². Beam-induced motion was corrected and binned two-fold using Patch Motion in cryoSPARC. The contrast transfer function (CTF) parameters were estimated using the Patch CTF module. Blob picker was used to pick particles on a small set of micrographs to generate class averages as templates for subsequent autopicking using template picker. Several rounds of 2D classification were performed to exclude bad particles that fell into 2D averages with poor features. Particles from different views were selected to generate three initial models using ab initio reconstruction. The resulting 3D models were used for heterogeneous refinement in cryoSPARC. Another round of heterogenous refinement was performed to select 3D classes showing the highest-resolution features for pACKR3_GRK5–Arr3*–Fab7, pACKR3_{+12G GRK5}–Arr2*–Fab7 and pACKR3_GRK5–Arr3*_+C-edge–Fab7. The selected classes were then refined using homogeneous refinement and nonuniform refinement⁶³. The image processing flowcharts for each dataset are shown in Extended Data Figs. 3, 4, 8.

Conditional Diffusion Model.

To enhance the backbone positions in the cryo-EM map, we calculated a diffusion model as described in the DiffModeler protocol³⁸. DiffModeler uses a diffusion model, a type of deep learning network, to enhance structural features, including backbone positions, in the map, which facilitate downstream structure fitting and modeling tasks. We ran DiffModeler with a contour level of 0.1 for the pACKR3_GRK5–Arr2*–Fab7 map (EMD-47700).

Model building and refinement.

A homology model of Fab7 generated using the SWISS-MODEL server (http://swissmodel.expasy.org), the crystal structure of active Arr2 (PDB entry 4JQI) and the anti-Fab hinge-binding nanobody (PDB entry 6WW2) were docked into the highest resolution cryo-EM map, which was of the pACKR3_GRK2–Arr2*–Fab7 nanodisc complex (Extended Data Fig. 2) using Phenix⁶⁴. The CDR regions of the Fab7 heavy chain were rebuilt manually in COOT⁶⁵. The resulting model was further improved using several rounds of real space refinement in Phenix and manual adjustment in COOT. The same strategy was employed to build and refine the rest of the models except that different initial models were used for different maps: the Fab7 model from the cryo-EM structure described above, the crystal structure of active Arr2 (PDB entry 4JQI), the crystal structure of active Arr3 (PDB entry 5TV1), and the CID24-CXCL12_LRHQ-ACKR3 complex (PDB entry 7SK6). For the pACKR3_GRK5–Arr2*–Fab7 structure, we also use the diffusion map to guide the model building. Figures were prepared using PyMOL and ChimeraX⁶⁶.

Principal component analysis (PCA).

PCA was performed on previously reported and the current experimental arrestin structures using Bio3D^67–69. New structures not used for PCA were projected (along with the previous structures) onto the PC1-PC2 plane for conformational comparison. A total of 114 structures of Arr1-4 were collected from the PDB (Supplementary Data Table 1). Sequences of these structures were aligned using MUSCLE ⁷⁰. Prior to PCA, structurally invariant “core” residues were identified through iterated rounds of structural superimposition as previously described ⁷¹. These core residues, which were all from the N-lobe, were used as the reference for the superimposition of structures. The new models from this work and additional PDB structures released recently (and hence were not used for PCA) were then aligned and superimposed in the same way as the base structures (Supplementary Data Table 1). PCA was calculated for aligned positions where no sequence gap was found for any of the structures in the base set and our models. The same positions were used to project additional PDB structures, where short gaps (1-2 residues due to unresolved structures) were found in two of these structures. These gaps were filled with the Cartesian coordinates of equivalent Cα atoms from the structures showing the minimal overall Cα root-mean-square deviation from the target structures. Movies showing morphs for the PC1 and PC2 motions (Supplementary Data Video 1 and 2) were rendered by PyMOL and Adobe Photoshop 2023. Because PDB entry 6K3F⁴² (ACKR3 phosphopeptide complexes with Arr2) has profound modeling issues, we omitted this entry from the PCA analysis (Extended Data Figure 9c, d), but we provide the mapped PC1 and PC2 values for its 6 unique complexes in the supplementary information for general reference (Supplementary Data Table 1).

Extended Data

Extended Data Figure 1. Fab7 binds to Arr2* and Arr3* in complex with ACKR3 phosphorylated by either GRK2 or GRK5.

a) FLAG pulldown assays show that Fab7 binds to pACKR3_GRK2–Arr2* and pACKR3_GRK5–Arr3*. This interaction is dependent on GRK activity because unphosphorylated ACKR3 (-ATP) does not pull down Arr2*, Arr3* or Fab7. pACKR3_GRK5 pulls down more Arr2*, Arr3* and Fab7 than pACKR3_GRK2. Two-sided one-way ANOVA followed by a Tukey multiple comparison test without adjustment was used to compare bound arrestin to pACKR3 ratios. Data are presented as mean ± S. D. from three technical replicates. b-e) Size exclusion chromatography shows that Fab7 coelutes with Arr2* bound to pACKR3_GRK5 (b) or pACKR3_GRK2 (c) and that Fab7 coelutes with Arr3* bound to pACKR3_GRK5 (d) or pACKR3_GRK2 (e). SEC was performed once to confirm the complex formation and was not used for preparing samples for cryo-EM studies.

Extended Data Figure 2. Workflow of cryo-EM data processing of pACKR3_GRK5–Arr2*–Fab7.

a) The cryo-EM data processing workflow from motion correction to CTF estimation to particle picking to 2D classification to 3D classification to 3D refinement is shown. Local resolution estimation is calculated by Relion. b) FSC curves calculated by cryoSPARC with 0.143 as a cutoff for pACKR3_GRK5–Arr2*–Fab7 complex.

Extended Data Figure 3. Workflow of cryo-EM data processing and resolution analysis of pACKR3_GRK2–Arr2*–Fab7 reconstituted in detergent micelles (a, b) or nanodiscs (c, d).

a) Representative micrograph shows well-distributed complexes reconstituted in LMNG/CHS detergent micelles. The cryo-EM data processing workflow is shown. Local resolution estimation is calculated by cryoSPARC. b) FSC curves calculated by cryoSPARC with 0.143 as a cutoff for pACKR3_GRK2–Arr2*–Fab7 complex in detergent micelles. c) Representative micrograph shows well-distributed complexes reconstituted in POPS/POPC nanodiscs. An anti-Fab nanobody was included to further rigidify the hinge of Fab7. The cryo-EM data processing workflow is shown. The receptor density is not evident. Local resolution estimation is calculated by cryoSPARC. d) FSC curves calculated by cryoSPARC with 0.143 as a cutoff for pACKR3_GRK2–Arr2*–Fab7–V_HH complex in nanodiscs.

Extended Data Figure 4. Workflow of cryo-EM data processing and resolution analysis of pACKR3_GRK5–Arr3*–Fab7 (a-c) and pACKR3_GRK2–Arr3*–Fab7 (d-f).

a) Representative micrograph of pACKR3_GRK5–Arr3*–Fab7 shows well-distributed complexes. The cryo-EM data processing workflow is shown. Local resolution estimation is calculated by cryoSPARC. b, c) FSC curves calculated by cryoSPARC with 0.143 as a cutoff for pACKR3_GRK5–Arr3*–Fab7 with (b) or without (c) the ACKR3 TM core. d) Representative micrograph of pACKR3_GRK2–Arr3*–Fab7 sample shows well-distributed complexes. The cryo-EM data processing workflow is shown. Local resolution estimation is calculated by cryoSPARC. e, f) FSC curves calculated by cryoSPARC with 0.143 as a cutoff for pACKR3_GRK2–Arr3*–Fab7 with (b) or without (c) the ACKR3 TM core.

Extended Data Figure 5. The interface between Arr2* and Fab7 and characterization of Fab7.

a) The Fab7 CDR-H3 contains an 18-residue loop that binds in the hinge of Arr2*. Fab7-Trp109 interacts with His353 on the C-lobe and Val171 and Tyr173 on the N-lobe of Arr2*, while Fab7-Arg107 forms a salt bridge with Glu296 on the C-lobe. Another cluster of contacts outside the hinge is provided by Fab7-Tyr57 in CDR-H2, which forms nonpolar contacts with Arr2*-Phe277 and -Leu278, and hydrogen bonds with the backbone nitrogens of Arr2*-Ala279 and -Leu300. These same interactions are also formed by Fab30-Tyr57, which has an identical CDR-H2. However, Fab7 forms more extensive interactions with Arr2* than Fab30 (buried surface areas of ~900 vs. ~700 Ų, respectively). b) Superposition of basal Arr2 (purple, PDB entry 1G4M) with activated Arr2* from the pACKR3_GRK2–Arr2*–Fab7–V_HH complex in nanodiscs (blue, PDB entry 8TII) aligned on their C-lobes. Residues that would clash with basal Arr2 are shown as spheres. c) ELISA analysis of Fab7 competition assay reveals that Arr3* (IC₅₀ ~90 nM) competes for Fab7 binding more efficiently than Arr3 (IC₅₀ ~35 μM). d) A cartoon illustrating that the C-tail of Arr2, released from the N-lobe upon activation, becomes susceptible to trypsin digestion. Fab7 facilitated digestion, producing C-terminally truncated Arr2 (Arr2_ΔC), suggesting that Fab7 traps Arr2 in the activated states when its C-tail is dissociated. The experiment was performed three times. e) A FLAG pulldown assay shows that Fab7 significantly increased WT Arr2 binding to pACKR3 but did not promote the binding of either WT or preactivated Arr2 to unphosphorylated ACKR3. The ratios of bound Arr2 or Arr2* to ACKR3 were compared using two-sided one-way ANOVA followed by Dunnett’s multiple comparison test without adjustment. c, e) Data are presented as mean ± S. D. from three technical replicates.

Extended Data Figure 6. Structure of pACKR3_GRK5–Arr2*–Fab7 reveal novel interfaces.

a) The electron density of TM7 is shown as a wire cage, contoured at 12σ. A close-up view of the side chain of Trp265^6.48 and Tyr268^6.51 is provided. b) The electron density of the TM1-6 is shown as a wire cage, contoured at 10-12σ. c) An overview of the diffusion process. It starts from the original cryo-EM map of pACKR3_GRK5–Arr2*–Fab7 (EMD-47700) as a condition and random Gaussian noise as input. The iterative reverse diffusion process is used to trace protein backbone using a pretrained diffusion model. The final target is the ground-truth protein backbone density. d) Electron density of the N-terminus of CXCL12_LRHQ is shown as a wire cage, contoured at 12σ. e) Residue Ser316-Tyr322 of ACKR3 adopts an extended conformation in the pACKR3_GRK5–Arr2*–Fab7 structure (left, PDB entry 9E82) but links H8 to TM7 in the ACKR3–CID24 structure (right, PDB entry 7SK6). The electron density of residues Ser316-Tyr322 is shown as a wire cage, contoured at 12σ. f) Sequence alignment of the arrestin finger loop, β5 and β6 flanking the finger loop, β8, C-loop and C-edge (193-loop and 338-loop). Conserved hydrophobic residues involved in micelle binding (and likely membrane binding) in the finger loop, as well as those involved in the interaction with TM7-H8 linker region of ACKR3 and the inter-lobe hydrophobic network, are highlighted in green. Arr3 lacks one of the membrane anchor sequences in the C-edge, highlighted in pink. g) The C-loop and subsequent β17 strand of Arr2* interacts with the TM1, TM6 and TM7 of ACKR3. h) Alignment of ACKR3 from the pACKR3_GRK5–Arr2*–Fab7 structure (PDB ID 9E82) and from the ACKR3–CID25 structure (PDB ID 7SK7) highlights a ~7° bend in the cytoplasmic end of TM1 which seems to be imposed by Arr2*.

Extended Data Figure 7. Inter-lobe hydrophobic interactions are important for Arr2 folding and stability, and access to the ACKR3 cytoplasmic pocket is not required for Arr2* engagement.

a) A FLAG pulldown shows that pACKR3_GRK5 and pACKR3_GRK2 pulls down 40% less Arr2*_3A than Arr2*. b) The 3A mutation in WT Arr2 does not express but introducing it into Arr2-L338B (with native cysteines replaced) enables sufficient expression for purification. Comparing Arr2_3A-L338B to Arr2-L338B, a FLAG pulldown assay shows that pACKR3_GRK5 pulls down 40% more Arr2_3A-L338B, suggesting enhanced constitutive activity of Arr2_3A. a, b) Two-sided one-way ANOVA followed by a Tukey multiple comparison without adjustment was used to compare bound arrestin to ACKR3 ratios. c) Trypsin digestion (Extended Data Fig. 5d) of Arr2-L338B in the presence of phosphorylated active rhodopsin (pRho*) produces Arr2_ΔC. Arr2* with a pre-released C-tail, produces Arr2_ΔC with or without pRho*. Both Arr2_3A-L338B and Arr2*_3A degrade into small fragments without pRho*, suggesting disrupted 3D packing and increased constitutive activity. pRho* binding partially protects Arr2_3A-L338B and Arr2*_3A from digestion. d) BRET response upon CXCL12 stimulation over time shows unchanged or slightly elevated coupling of Arr2_3A to ACKR3, consistent with its increased constitutive activity. The combination of ACKR3_F317/I318G and Arr2_3A reduces Arr2 coupling to ACKR3 in HEK293A cells and has no effect in HEK293T cell. e) Size exclusion chromatography indicates that CID24 coelutes with the pACKR3_GRK5–Arr2* complex. f) CID24 and CID25 bind to the intercellular and extracellular regions of ACKR3, respectively (PDB entry 7SK4). CID24 is more potent in inhibiting GRK2 or GRK5 phosphorylation of ACKR3, compared to CID25. g) The dose-dependent BRET response curve shows comparable binding between Arr2_ΔFL and Arr2, 20 minutes after CXCL12 stimulation. c, e) Experiment was performed three times. a, b, d, f, g) Data are presented as mean ± S. D. from three technical replicates.

Extended Data Figure 8. Workflow of cryo-EM data processing and resolution analysis of pACKR3_{+12G GRK5}–Arr2*–Fab7 and pACKR3_GRK5–Arr3*_+C-edge–Fab7.

a) Representative micrograph of the pACKR3_{+12G GRK5}–Arr2*–Fab7 shows well-distributed complexes. The cryo-EM data processing workflow is shown. Local resolution estimation is calculated by cryoSPARC. b) FSC curves calculated by cryoSPARC with 0.143 as a cutoff for pACKR3_{+12G GRK5}–Arr2*–Fab7. c) Representative micrograph of the pACKR3_GRK5–Arr3*_+C-edge–Fab7 shows well-distributed complexes. The cryo-EM data processing workflow is shown. Local resolution estimation is calculated by cryoSPARC. d) FSC curves calculated by cryoSPARC with 0.143 as a cutoff for pACKR3_GRK5–Arr3*_+C-edge–Fab7.

Extended Data Figure 9. ACKR3 phospho-tail interaction with Arr3*, comparison of arrestin conformations as a function of barcode, and the conformational landscape of arrestins and their complexes with receptors.

a) Sharpened map of pACKR3_GRK5–Arr3*–Fab7 and pACKR3_GRK2–Arr3*–Fab7 without the ACKR3 TM core. Interactions of the ACKR3 C-tail phosphorylated by GRK5 or GRK2 with the Arr3* N-lobe. Electron density of the pACKR3 phospho-peptide is shown as a wire cage contoured at 10σ and 12σ, respectively. Distances below 4 Å are shown as a black dashed line. b) Alignment of pACKR3_GRK5 bound and pACKR3_GRK2 bound Arr2* or Arr3* on the N-lobe suggests at most subtle changes in arrestin conformation. c) Conformational map derived from previously deposited arrestin structures with new structures of arrestin–Fab7 complexes from this study (light blue and purple circles) superposed. Blue, red, green, and black circles correspond to structures that include Arr1, Arr2, Arr3, and Arr4, respectively. Detailed information on the models used for PCA is provided in Supplementary Data Table1. The PC1 axis corresponds to the well-established twist between the N- and C-lobes of arrestin that is characteristic of activation (Supplementary Data Video1), whereas the PC2 axis corresponds to an activation-independent “wag” of the C-lobe relative to the N-lobe (Supplementary Data Video2). ND, nanodisc; NA, not applicable. d) The eigenvalues measure the conformational variance along corresponding PC axes, which usually decrease rapidly after the top few components, as occurs here. PC1 and PC2 account for 89% of the total variance among the structures.

Extended Data Table1.

Cryo-EM data collection, refinement and validation statistics

No.	EMD-	PDB entry	ACKR3	Arrestin	Phosphorylated by	Model membrane system	Resolution (Å)	Tail interaction?	Finger loop membrane insertion?	C-edge loop membrane insertion?
1	47700	9E82	WT	Arr2*	GRK5	LMNG/CHS	3.4	Y	Y	Y
2	49564		WT	Arr2*	GRK2	LMNG/CHS	4.1	Y	Y	Y
3	41290		WT	Arr3*	GRK5	LMNG/CHS	3.9	Y	Y	N
4	41291		WT	Arr3*	GRK2	LMNG/CHS	6.9	Y	Y	N
5	41292		+ 12G	Arr2*	GRK5	LMNG/CHS	3.8	Y	Y	Y
6	43277	8VJ9	WT	Arr3*+C edge	GRK5	LMNG/CHS	3.3	Y	Y	Y
7	41295	8TIL	WT	Arr3*	GRK5	LMNG/CHS	3.8	Y	NA	NA
8	41296	8TIN	WT	Arr3*	GRK2	LMNG/CHS	4.0	Y	NA	NA
9	41289	8TII	WT	Arr2*	GRK2	POPC/POPS ND	3.0	Y	NA	NA
10	41297	8TIO	+ 12G	Arr2*	GRK5	LMNG/CHS	3.6	Y	NA	NA
Number in Table S1 EMD- PDB-	#1 47700 9E82	#2 49562	#3 41290	#4 41291	#5 41292	#6 43277 8VJ9	#7 41295 8TIL	#8 41296 8TIN	#9 41289 8TII	#10 41297 8TIO
Data collection and processing
Magnification	81,000	81,000	81,000	81,000	81,000	81,000	81,000	81,000	81,000	81,000
Voltage (kV)	300	300	300	300	300	300	300	300	300	300
Electron exposure (e–/Å2)	~56	~56	~56	~56	~56	~56	~56	~56	~56	~56
Defocus range (μm)	1.2-2.5	1.2-2.5	1.2-2.5	1.2-2.5	1.2-2.5	1.2-2.5	1.2-2.5	1.2-2.5	1.2-2.5	1.2-2.5
Pixel size (Å)	0.527	0.527	0.527	0.527	0.527	0.527	0.527	0.527	0.527	0.527
Symmetry imposed	no	no	no	no	no	no	no	no	no	no
Initial particle images (no.)	7,221,036	24,544,952	25,445,177	21,549,923	14,696,260	21,005,500	25,445,177	21,549,923	6,491913	14,696,260
Final particle images (no.)	104,555	398,823	345,391	142,889	459,377	214,953	296,617	174,800	698,693	285,430
Map resolution (Å)	3.4	4.1	3.9	6.9	3.8	3.3	3.8	4.0	3.0	3.6
FSC threshold	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143
Map resolution range (Å)	3.2-9.0	2.8-53.3	2.7-28.7	5.3-12.8	2.8-38.7	2.7-29.6	2.7-28.6	3.0-34.5	2.4-38.9	2.7-46.1
Refinement
Initial model used (PDB code)	4JQI, 7SK6					5TV1	5TV1	5TV1	4JQI	4JQI
Model resolution (Å)	3.4					3.3	3.8	4.0	3.0	3.6
FSC threshold	0.143					0.143	0.143	0.143	0.143	0.143
Model resolution range (Å)	3.2-9.0					2.7-29.6	2.7-28.6	3.0-34.5	2.4-38.9	2.7-46.1
Map sharpening B factor (Å2)	45.0					107.9	148.9	124.5	149.5	100.0
Model composition
Non-hydrogen atoms	8200					10983	10983	10971	12953	11165
Protein residues	1034					705	705	705	838	714
Ligands	CLR:1					0	0	0	0	0
B factors (Å2)
Protein	120					76.5	47.0	22.4	68.1	32.2
Ligand	115					-	-	-	-	-
R.m.s. deviations
Bond lengths (Å)	0.004					0.003	0.003	0.002	0.003	0.003
Bond angles (°)	0.951					0.538	0.546	0.534	0.513	0.521
Validation
MolProbity score	2.36					2.66	2.68	2.61	2.14	2.6
Clashscore	10.34					12.19	12.65	12.30	5.59	13.00
Poor rotamers (%)	3.04					4.98	5.78	5.94	3.17	3.8
Ramachandran plot
Favored (%)	92.8					90.7	93.4	93.7	93.1	90.7
Allowed (%)	7.2					9.3	6.6	6.2	6.9	9.3
Disallowed (%)	0					0	0	0.1	0	0

Data and material availability:

All data needed to evaluate the conclusions in the paper are available in the main text or the supplementary materials. Additional data related to this paper are available upon reasonable request from the authors. The maps of pACKR3_GRK5–Arr2*–Fab7, pACKR3_GRK2–Arr2*–Fab7, pACKR3_GRK5–Arr3*–Fab7, pACKR3_GRK2–Arr3*–Fab7, pACKR3_{+12G GRK5}–Arr2*–Fab7, and pACKR3_GRK5–Arr3*_{+C edge} –Fab7 have been deposited into the Electron Microscopy Data Bank under accession codes EMD-47700, EMD-49564, EMD-41290, EMD-41291, EMD-41292, and EMD-43277, respectively. These maps contain density for ACKR3 in micelles but the ACKR3 chain was only modeled for pACKR3_GRK5–Arr2*–Fab7. pACKR3_GRK5–Arr2*–Fab7 and pACKR3_GRK5–Arr3*_{+C edge} –Fab7 were deposited into the Protein Data Bank under accession codes 9E82 and 8VJ9, respectively. The tail-mode maps and structures of pACKR3_GRK5–Arr3*–Fab7, pACKR3_GRK2–Arr3*–Fab7, pACKR3_GRK2–Arr2*–Fab7–V_HH and pACKR3_{+12G GRK5}–Arr2*–Fab7 have been deposited into the Electron Microscopy Data Bank under accession codes EMD-41295, EMD-41296, EMD-41289, and EMD-41297 and the Protein Data Bank under accession codes 8TIL, 8TIN, 8TII, and 8TIO, respectively.