Selective suppression and biasing of chemokine receptors CCR9 and ACKR4 through targeting CCL25 with de novo miniproteins

  1. Bas de Boer
  2. Thomas D Lamme
  3. Karlijn Verdwaald
  4. Sara Santamaria Medina
  5. Csongor G Németh
  6. Elisabeth M Elfrink
  7. Martine J Smit
  8. Iwan JP de Esch
  9. Christopher T Schafer

  • Curated by eLife

    eLife logo

    eLife Assessment

    This important study describes computationally designed proteins that bind to the chemokine CCL25. The authors present evidence that some binders simply prevent chemokine binding to the CCR9 receptor, while one binder changes the downstream signaling triggered by chemokine binding. The evidence is convincing, showing a degree of selectivity between chemokine ligands that could be further improved. The study will be of interest to pharmacologists, cell biologists studying chemokine function, and protein engineers.

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Chemokines and their receptors mediate cell migration and coordinate immune responses, while dysregulation can lead to inflammation. Therapeutic modulation of the chemokine signaling axis has proven difficult. Most drug discovery efforts target the receptors, whereas natural regulatory mechanisms modulate immune responses by directly binding to or modifying the chemokine ligands. Despite this insight, development of effective chemokine-directed modulators has remained elusive. Recent advances in de novo protein design offer an unprecedented opportunity to produce high-affinity binders that efficiently block protein-protein interactions. We implemented a computational workflow leveraging the BindCraft platform to generate miniprotein binders against CCL25, the chemokine ligand for the receptors CCR9 and ACKR4 and implicated in inflammatory bowel diseases. The unbiased development results in several miniproteins designed to block the receptor N-terminus from wrapping the chemokine and prevent productive engagement. Thus, these proteins suppress CCL25-mediated effector coupling and halt MOLT-4 lymphoblast migration. Another class of miniprotein, represented by VUP25111, is predicted to bind CCL25 along the chemokine β1 strand and retained receptor binding. This complex inhibited arrestin recruitment to CCR9, but not to ACKR4, indicating receptor specificity. Additionally, G protein recruitment to CCR9 was unimpeded by VUP25111, suggesting that the miniprotein biased the native agonist towards G proteins. These results demonstrate the effectiveness of differentially targeting CCL25 to suppress CCR9 signaling and new tools to resolve the structural basis of chemokine receptor activation and bias.

Article activity feed

  1. Version published to 10.7554/elife.111190.2 on eLife
  2. Version published to 10.7554/elife.111190 on eLife
  3. eLife

    eLife Assessment

    This important study describes computationally designed proteins that bind to the chemokine CCL25. The authors present evidence that some binders simply prevent chemokine binding to the CCR9 receptor, while one binder changes the downstream signaling triggered by chemokine binding. The evidence is convincing, showing a degree of selectivity between chemokine ligands that could be further improved. The study will be of interest to pharmacologists, cell biologists studying chemokine function, and protein engineers.

  4. eLife

    Reviewer #1 (Public review):

    Summary:

    In this manuscript, the authors describe the use of BindCraft computational protein design to create a series of binders to the chemokine CCL25. This chemokine normally mediates CCR9-dependent trafficking of immune cells to the gut, making it a potential target for the treatment of inflammatory bowel disease and related conditions. Importantly, CCL25 also binds a scavenging receptor, ACKR4. The computational protein design approach used does not involve defining particular epitopes to be targeted, allowing a free search for any potential interaction surface.

    Among four designs tested, three were predicted to interact at a similar site on the chemokine, while a fourth clone, VUP25111, was predicted to bind to a different site. All four designs showed binding to CCL25, with similar high-nM KD values in all cases. The first three clones showed evidence of direct competition with the receptor for CCL25 binding, while VUP25111 showed incomplete inhibition of binding. In functional assays, all clones acted as antagonists except for VUP25111, which inhibited arrestin recruitment by CCR9, but did not affect G protein activation by CCR9 or arrestin recruitment by ACKR4 (which signals exclusively through arrestin and not G protein).

    Strengths:

    The work is completed to a high technical standard, and the functional diversity of the clones is intriguing. It is exciting to see pathway-selective modulation of signaling, and this basic paradigm is likely to generalize to other chemokine/receptor systems. The exceptional complexity of chemokine signaling makes this an excellent area to explore the development of modulators that can restrict signaling to a specific subset of receptors.

    Weaknesses:

    No major weaknesses were noted by this reviewer.

    Comments on revised version:

    The revision looks fine to me.

  5. eLife

    Reviewer #2 (Public review):

    This study from de Boer, Lamme, Verdwaald and Schafer describes the de novo AI-guided design of miniproteins that target the chemokine CCL25, with the aim to modulate the activation and signalling of the chemokine receptors CCR9 and ACKR4. The study focuses on characterising four miniproteins that all bind CCL25 with good affinity. Three designs appear to prevent CCL25 binding to both CCR9 and ACKR4, with increasing concentrations of miniproteins resulting in decreased arrestin (both receptors) and mini G protein recruitment (CCR9), less inhibition of forskolin-stimulated cAMP (CCR9), and decreased GRK3 recruitment and receptor internalisation (CCR9). One miniprotein, VUP25111, changes the properties of CCL25 rather than preventing ligand/receptor interactions, resulting in greater selectivity for CCR9 over ACKR4 and a G protein-biased signalling profile (maintenance of mini G protein recruitment, GRK3 recruitment, inhibition of cAMP and receptor internalisation, but loss of arrestin recruitment). VUP25111 also maintained chemotactic migration in MOLT-4 T lymphoblast cells, whereas this response was lost in the presence of the other three miniproteins.

    Overall, this is a very interesting application of AI-designed de novo miniproteins to modulate GPCR responses by directly binding the ligand rather than the receptor. This is a conceptually very intriguing approach that could, in principle, be extended to other GPCR systems beyond the chemokine family. The authors deploy an impressive array of assays spanning multiple signalling endpoints, providing a thorough picture of how each miniprotein influences receptor activation and downstream signalling. The presentation of concentration-response relationships for CCL25 alone and in the presence of each miniprotein is particularly informative, and the figures are very well constructed throughout. The inclusion of clear cartoons illustrating the basis of each assay is a nice touch that will help readers from outside the immediate field follow the logic of each experiment.

  6. eLife

    Reviewer #3 (Public review):

    The authors employed the BindCraft platform to develop binders targeting the chemokine CCL25, a selective activator of the chemokine receptor CCR9. They successfully generated two classes of binders: one that inhibits all CCL25-mediated CCR9 activation, and another that permits CCR9 G protein signaling while simultaneously preventing arrestin recruitment. These tools will enable the dissection of arrestin involvement in regulating cell migration.

  7. eLife

    Author response:

    The following is the authors’ response to the original reviews.

    Reviewer #1 (Public review):

    Summary:

    In this manuscript, the authors describe the use of BindCraft computational protein design to create a series of binders to the chemokine CCL25. This chemokine normally mediates CCR9-dependent trafficking of immune cells to the gut, making it a potential target for the treatment of inflammatory bowel disease and related conditions. Importantly, CCL25 also binds a scavenging receptor, ACKR4. The computational protein design approach used does not involve defining particular epitopes to be targeted, allowing a free search for any potential interaction surface.

    Among four designs tested, three were predicted to interact at a similar site on the chemokine, while a fourth clone, VUP25111, was predicted to bind to a different site. All four designs showed binding to CCL25, with similar high-nM KD values in all cases. The first three clones showed evidence of direct competition with the receptor for CCL25 binding, while VUP25111 showed incomplete inhibition of binding. In functional assays, all clones acted as antagonists except for VUP25111, which inhibited arrestin recruitment by CCR9, but did not affect G protein activation by CCR9 or arrestin recruitment by ACKR4 (which signals exclusively through arrestin and not G protein).

    Strengths:

    The work is completed to a high technical standard, and the functional diversity of the clones is intriguing. It is exciting to see pathway-selective modulation of signaling, and this basic paradigm is likely to generalize to other chemokine/receptor systems. The exceptional complexity of chemokine signaling makes this an excellent area to explore the development of modulators that can restrict signaling to a specific subset of receptors.

    Weaknesses:

    No major weaknesses were noted by this reviewer.

    We thank the reviewer for their positive response and thorough evaluation of our work. Below, we address their recommendations in detail in the private review.

    Reviewer #1 (Recommendations for the authors):

    Minor points for potential improvement:

    (1) Consider shortening the discussion section, which could benefit from a more focused discussion of the key conclusions and their implications. As written, it is rather long, and parts of it repeat material already covered elsewhere (e.g., the two-step binding mechanism is described in lines 55-64 and again in 331-335).

    We thank the reviewer for the suggestion to make the discussion more focused. We agree that parts repeat material from the introduction. As such, we have revised the discussion to reduce repetition with the introduction, including removal of the repeated description of the two-step binding mechanism, and condensing several passages.

    (2) The figures are clear, but it may be helpful to combine panels into a smaller number of figures to make direct comparisons among signaling curves more straightforward.

    We thank the reviewer for the suggestion to reduce the number of figures for easier curve comparisons. We prefer our current presentation of the figures so as not to conflate disparate themes and concepts; for example, GRK recruitment is a distinct event from receptor internalization. The pharmacological results have been compiled in Figure S8 in the supplementary to allow visual, side-by-side comparisons of the DRCs presented in the publication.

    Reviewer #2 (Public review):

    This study from de Boer, Lamme, Verdwaald and Schafer describes the de novo AI-guided design of miniproteins that target the chemokine CCL25, with the aim to modulate the activation and signalling of the chemokine receptors CCR9 and ACKR4. The study focuses on characterising four miniproteins that all bind CCL25 with good affinity. Three designs appear to prevent CCL25 binding to both CCR9 and ACKR4, with increasing concentrations of miniproteins resulting in decreased arrestin (both receptors) and mini G protein recruitment (CCR9), less inhibition of forskolin-stimulated cAMP (CCR9), and decreased GRK3 recruitment and receptor internalisation (CCR9). One miniprotein, VUP25111, changes the properties of CCL25 rather than preventing ligand/receptor interactions, resulting in greater selectivity for CCR9 over ACKR4 and a G protein-biased signalling profile (maintenance of mini G protein recruitment, GRK3 recruitment, inhibition of cAMP and receptor internalisation, but loss of arrestin recruitment). VUP25111 also maintained chemotactic migration in MOLT-4 T lymphoblast cells, whereas this response was lost in the presence of the other three miniproteins.

    Overall, this is a very interesting application of AI-designed de novo miniproteins to modulate GPCR responses by directly binding the ligand rather than the receptor. This is a conceptually very intriguing approach that could, in principle, be extended to other GPCR systems beyond the chemokine family. The authors deploy an impressive array of assays spanning multiple signalling endpoints, providing a thorough picture of how each miniprotein influences receptor activation and downstream signalling. The presentation of concentration-response relationships for CCL25 alone and in the presence of each miniprotein is particularly informative, and the figures are very well constructed throughout. The inclusion of clear cartoons illustrating the basis of each assay is a nice touch that will help readers from outside the immediate field follow the logic of each experiment.

    There are two main conclusions that are not currently as well-supported by the evidence as they might be, and that would benefit from some qualification. The first concerns the selectivity of the miniproteins for CCL25. Testing the impact of the miniproteins on CXCL12 activation of CXCR4 is an important and welcome experiment, but it addresses selectivity against only one other chemokine system, and the current claim of specificity is therefore stronger than the data allow. Additionally, at the highest concentration tested (10 µM), the more potent miniproteins (VUP25101, VUP25107) appear to show some inhibition of arrestin recruitment to CXCR4 - perhaps unsurprising given the degree of structural conservation among chemokines. The statements regarding selectivity and the lack of effect on the CXCL12/CXCR4 system would benefit from revision to more accurately reflect these observations.

    We thank the reviewer for the detailed review and suggestions for our publication. We agree with the reviewer that testing against only one additional chemokine receptor: chemokine pair is insufficient to support strong claims of selectivity. To further assess the selectivity of the miniproteins, we evaluated the effects of the VUPs against CCL20, a chemokine closely related to CCL25. CCL20 binds both CCR6 and ACKR4, providing an optimal platform to compare the effects on this non-specified target. Receptor activation was determined by β-arrestin recruitment (Fig. S6). For ACKR4, VUP25107 and VUP25112 have similar IC50 values for CCL20 and CCL25. In contrast, VUP25101 showed reduced potency towards CCL20 relative to CCL25 (CCL20 pIC50 = 6.06 ± 0.13, CCL25 pIC50 = 7.73 ± 0.08), suggesting greater selectivity. VUP25111 had no impact on activation of ACKR4 by CCL20, which may reflect a lack of engagement with the other chemokine or that it binds in a similar way as to CCL25 and does not impact chemokine-promoted ACKR4 activation. CCL20-mediated activation of CCR6 further supports the specificity of VUP25101, which showed only a modest effect at the highest concentration tested. In contrast, VUP25107 and VUP25112 showed a dose-dependent inhibition of arrestin recruitment (pIC50VUP25107 = 6.99 ± 0.15, pIC50VUP25112 = 7.49 ± 0.07).

    We agree that a reduction in β-arrestin recruitment to CXCR4 is apparent at the highest concentration tested (10 µM) for VUP25101 and VUP25107. To further address the selectivity of these miniproteins, we have included β-arrestin recruitment data for ACKR3 (Fig. S5), the other chemokine receptor that binds CXCL12. A similar suppression was observed for the atypical receptor as for CXCR4 at the highest concentrations, suggesting cross-reactivity at high concentrations.

    Based on the CCL20 and CXCL12 data, we have revised the text to more accurately reflect these observations. We now frame these results as an initial proof of concept demonstrating the feasibility of this design approach, while noting that further optimization and characterization of selectivity will be required to advance the miniproteins for additional applications.

    To reflect this, we have added the following text on Pg.9:

    “CCL25 specificity was experimentally evaluated by testing the miniproteins against CXCL12 by tracking arrestin recruitment to CXCR4 and ACKR3, the native agonist for this canonical and atypical receptor pair (Fig. S5). VUP25111 and VUP25112 showed no effect on CXCL12-mediated activation at any concentration, while VUP25101 and VUP25107 showed partial inhibition at the highest tested concentration (10 µM).

    In addition, the effects of the VUPs were evaluated against CCL20, a closely related chemokine to CCL25, by assessing arrestin recruitment to CCR6 and ACKR4 (Fig. S6). VUP25107 and VUP25112 displayed similar inhibitory potencies of CCL20 as to CCL25 for ACKR4 (VUP25107: pIC50 = 7.72 ± 0.21; VUP25112: pIC50 = 6.93 ± 1.03). In contrast, VUP25101 showed reduced inhibition of CCL20 relative to CCL25 (VUP25101: pIC50CCL20 = 6.06 ± 0.13 v pIC50CCL25 = 7.73 ± 0.08), indicating greater selectivity for the designed target. CCL20 activation of CCR6 further supports the specificity of VUP25101, which only showed a modest effect at the highest concentration tested. VUP25111 showed no inhibitory effects on CCL20 activation of either CCR6 nor ACKR4, which may indicate a lack of binding, but is also indiscernible from forming a receptor-competent complex similar to the proposed CCL25:VUP25111:ACKR4 interaction.”

    And added the following in the discussion on Pg.15:

    “The miniproteins displayed varying degrees of chemokine selectivity among CCL25, CCL20, and CXCL12, including cross-reactivity for some designs. Given that only two additional chemokines were examined, these experiments provide an initial assessment of selectivity. As chemokine selectivity was not explicitly optimized during design, broader experimental profiling and negative-design strategies may be required to establish and improve miniprotein selectivity.”

    The second concern relates to the interpretation of the preserved GRK3 recruitment, but the complete loss of arrestin recruitment observed with VUP25111. In the GRK3 recruitment experiments, 20 nM CCL25 was used, representing an EC40 concentration in this assay. VUP25111 causes a concentration-dependent reduction in CCL25-induced GRK3 recruitment, down to approximately 15% of the maximal response. It is worth considering whether this degree of reduction in GRK3 recruitment could itself be sufficient to disrupt patterns of receptor phosphorylation and thereby prevent observable arrestin recruitment. Both interpretations are complicated by the fact that the GRK3 recruitment and arrestin recruitment assays likely differ in their sensitivity and dynamic windows, making direct quantitative comparisons between them difficult. In the absence of direct measurements of CCR9 phosphorylation in the presence of VUP25111, the alternative interpretation remains open and would benefit from acknowledgement. Given recent work from the same group demonstrating that receptor internalisation is only partially dependent on arrestins (Lamme et al., 2025, J Biol Chem), further discussion of the relationship between GRK and arrestin recruitment and CCR9 internalisation would be of value to the broader GPCR audience this work is likely to attract.

    We appreciate the reviewer’s careful analysis of our results and highlighting earlier work from our group. The reviewer is correct that the discrepancy between VUP25111’s impact on GRK3 and arrestin recruitment can also be explained by the decrease in GRK3 recruitment reflecting a sufficient change in receptor phosphorylation to account for the loss of arrestin engagement and not solely by the presented interpretation of the adoption of an arrestin-incompatible conformation with retained phosphorylation. The main text has been modified on pg. 11 to reflect this potential:

    “This suggests that either activation of CCR9 by the VUP25111:CCL25 complex promotes a conformation incompatible with arrestin engagement, but still accessible for GRK phosphorylation, or the reduced GRK interaction represents the formation of a phosphorylation barcode incapable of supporting arrestin recruitment. Without direct phosphorylation measurements these possibilities cannot be separated.”

    And in the discussion on pg.14:

    “The extent of phosphorylation is not clear, and the inhibited GRK recruitment may reflect the formation of sub-phosphorylated states which cannot support arrestin recruitment nor promote efficient receptor internalization. Alternatively, the miniprotein: chemokine complex may induce a conformational state incompatible with arrestin engagement.”

    A partial phosphorylation explanation has also been added on internalization on pg.12:

    “These results suggest that the VUP25111:CCL25 complex suppresses both the arrestin-dependent and independent internalization mechanisms used by CCR9, despite likely allowing for some, albeit perhaps insufficient, GRK phosphorylation.”

    Finally, some additional justification for the use of 20 nM CCL25 across all assays would strengthen the study, as this concentration represents different points on the concentration-response curve depending on the assay and receptor in question. It ranges from an EC40 for CCR9 GRK3 recruitment and internalisation, to an EC50 for CCR9 arrestin and mini-Gi recruitment, an EC80 for CCR9 cAMP inhibition, and an Emax for ACKR4 arrestin recruitment. This has potential consequences for the interpretation and cross-assay comparison of miniprotein potency, and the authors are encouraged to acknowledge and discuss this in the context of their conclusions.

    We thank the reviewer for highlighting the need to discuss CCL25 concentration across assays. Our primary aim was to use a single CCL25 concentration to enable direct comparison of all assay outcomes against the same CCL25:miniproteins ratio, since the miniproteins are designed to antagonize the chemokine, rather than the receptor. We set the concentration at 20 nM based on the first arrestin recruitment assays to CCR9, where we initially characterized miniprotein: CCL25 activity. Importantly, CCL25 has lower potency for CCR9 than ACKR4, making it the limiting case. Selecting 20 nM ensured responses within the EC40-EC80 range for all assays. To improve clarity, we have now explicitly stated this rationale in the manuscript and noted the implications of using a single CCL25 concentration across diverse readouts on Pg.10:

    “Despite the higher potency of CCL25 for cAMP inhibition compared to arrestin recruitment, a fixed concentration of 20 nM CCL25 was used throughout this and subsequent assays to maintain a consistent CCL25:miniprotein stoichiometry. As this concentration represents different points on the respective concentration-response curves, cross-assay comparisons of apparent miniprotein potency should be interpreted qualitatively.”

    Reviewer #2 (Recommendations for the authors):

    Major points:

    (1) The conclusions regarding miniprotein selectivity for CCL25 would benefit from revision to reflect that only the CXCL12/CXCR4 system was tested as a counter-screen. The apparent inhibition of arrestin recruitment to CXCR4 by VUP25101 and VUP25107 at 10 µM should be acknowledged/discussed.

    We thank the reviewer for this comment and agree that selectivity should be more explicitly acknowledged. This point has been thoroughly addressed in our response in the public review. Briefly, we have added a similar selectivity assay but now against CCL20, which is closely related to CCL25, as well as revisiting the CXCL12 assay with ACKR3 that was originally presented in the manuscript. Statements in the text have been added to note the apparent inhibition at 10 µM for VUP25101 and VUP25107.

    (2) The conclusions regarding G protein-biased signalling driven by VUP25111 would benefit from qualification. In particular, it is worth acknowledging that a substantial reduction in GRK3 recruitment may itself be sufficient to account for the observed loss of arrestin recruitment, and that direct measurement of CCR9 phosphorylation in the presence of VUP25111 would help to resolve this. Discussion of the relationship between GRK recruitment, receptor phosphorylation, arrestin recruitment and internalisation in this context would also be of value to the broader GPCR audience this work is likely to attract.

    We agree with the reviewer that an incomplete or insufficient level of phosphorylation could also explain our arrestin and internalization results presented by the VUP25111:CCL25 complex. We have added the following to the discussion on pg. 14 and modified the results to better reflect the multiple interpretations:

    “The extent of phosphorylation is not clear and the inhibited GRK recruitment may reflect the formation of sub-phosphorylated states which cannot support arrestin recruitment, nor promote efficient receptor internalization. Alternatively, the complete inhibition of arrestin engagement cannot be easily explained by phosphorylation, and instead suggests the miniprotein: chemokine complex may induce a conformational state incompatible with arrestin engagement. Due to the clash with ECL2, the loop would need to distort to accommodate the larger ligand complex and this displacement may translate into the G protein and GRK-permissive structural state.”

    (3) The choice of 20 nM CCL25 across assays should be justified or its implications discussed, particularly given that this concentration represents different effective concentrations for each endpoint and receptor.

    The rationale for using a fixed 20nM CCL25 concentration and its implications is now discussed within the main text as well as our response in the public review.

    Modelling:

    Modelling of VUP25111 and VUP25101 binding to CCL25/receptor complexes is presented. Was the same analysis performed for VUP25112 and VUP25107, and do they similarly appear to interfere with direct receptor binding?

    We initially selected VUP25101 as a representative for VUP25112 and VUP25107 to simplify visualization. We have now extended the same modeling analysis to VUP25107 and VUP25112. These additional analyses have now been included in the revised manuscript in Fig. 3D.

    Methods:

    (1) Percentage concentrations throughout the methods are missing w/v or v/v designations.

    We thank the reviewer for their careful reading and apologize for the oversight. We have added all w/v or v/v designations in the methods.

    (2) More specific information for each BRET experiment would be useful. For example, incubation times with CCL25 and miniproteins (included for some but not all assays), and clarification of whether experiments were endpoint reads or timecourse experiments.

    We agree with the reviewer that the methods section could be clarified, as well as removing a lot of repetition. We have revised the methods.

    Specific text revisions:

    (1) Abstract, line 20: "Most drug discovery efforts target the receptors, whereas natural regulatory mechanisms focus on the chemokines". This is unclear for a non-chemokine audience without additional context clarifying that this refers to binding to or modification of the chemokine itself to modulate receptor activation.

    To clarify this distinction for a broader readership, we revised the abstract to state explicitly that natural regulatory mechanisms modulate immune responses by directly binding to or modifying chemokines, thereby altering their ability to bind receptors. We have changed the text in the abstract on Pg.2 to:

    “Most drug discovery efforts target the receptors, whereas natural regulatory mechanisms modulate immune responses by directly binding to or modifying the chemokine ligands.”

    (2) Results, line 177: "In the case of VUP25111, only the highest tested concentrations decreased CCL25 binding to either receptor, but did not completely eliminate the binding signal". This should be revised as there is a clear concentration-dependent effect of VUP25111 on CCL25 binding.

    We have revised the text to reflect that VUP25111 inhibits CCL25 binding in a concentration-dependent manner, while also making clear that this inhibition occurs with significantly lower potency than the other miniproteins. We have changed the following text on Pg.7:

    “VUP25111 shows a drastically lower inhibitory potency compared to the other binders and did not completely eliminate the binding signal.”

    (3) Line 179: "...this suggests that CCL25:VUP25111 may bind the receptors cooperatively". Please revise; the meaning is unclear.

    Our results indicate that VUP25111 does not prevent CCL25 from engaging either receptor. To avoid ambiguity, we have revised the text to explicitly state that our results are consistent with the binding of a pre-formed VUP25111:CCL25 complex to the receptor, resulting in the formation of a VUP25111:CCL25:receptor complex. We have changed the following text on Pg.7:

    “Since the binder readily complexes with the chemokine at these concentrations (Fig. 2), this suggests that the CCL25:VUP25111 complex retains the ability to bind the receptor.”

    (4) Line 216: "VUP25112 also completely impaired CCL25-mediated activation of the atypical receptor at the highest concentrations, but with significantly less potency". Please provide statistical tests to support this statement.

    We thank the reviewer for requesting clarification regarding the statistical support for this statement. Statistical analyses supporting the conclusions in line 216, as well as other figures, were included in the initial manuscript but were provided in the supplementary information. To improve clarity in the manuscript, we have added explicit references to the corresponding supplementary statistical table in the text, added appropriate significance indicators to the relevant figures, and expanded the figure legends to state the statistical test used.

    (5) Line 219: "The unique binder VUP25111, on the other hand, did not inhibit the activation of ACKR4 by CCL25 at any tested concentration". Please rephrase. While there was no concentration-dependent inhibition of arrestin recruitment to ACKR4, the addition of VUP25111 at any concentration appeared to reduce total arrestin recruitment to approximately 80% of the CCL25-alone condition.

    We would like to thank the reviewer for this careful analysis of the data. Although this may appear to be the case in the figure, there is no significant difference between the lowest concentration (2.5 nM) and vehicle condition (∞) for any miniprotein and no real difference can be claimed.

    We have specifically noted this on Pg. 9:

    “The unique binder VUP25111, on the other hand, did not significantly impact the activation of ACKR4 by CCL25 at any tested miniprotein concentration.”

    (6) Line 240: "suggest that the specific ligand interactions driving arrestin engagement with CCR9 may be more strict than for the G proteins". An alternative interpretation is that the miniprotein specifically impairs the receptor conformation important for arrestin engagement, without affecting the conformation required for G protein engagement, and this should be acknowledged.

    We agree with the reviewer that these statements would benefit from further clarification. Indeed, the observed effects may not be only due to specific ligand interactions but could also depend on specific receptor conformations required for the specific protein-protein interactions. We have adjusted the following text on Pg.10:

    “The deviations between the arrestin results (Fig. 4) and G protein (Fig. 5B) suggest that the specific ligand interactions or receptor conformations driving arrestin engagement with CCR9 may be more strict than for the G proteins.”

    Reviewer #3 (Public review):

    Summary:

    The authors employed the BindCraft platform to develop binders targeting the chemokine CCL25, a selective activator of the chemokine receptor CCR9. They successfully generated two classes of binders: one that inhibits all CCL25-mediated CCR9 activation, and another that permits CCR9 G protein signaling while simultaneously preventing arrestin recruitment. These tools will enable the dissection of arrestin involvement in regulating cell migration.

    My comments, in the order of reading:

    (1) Title: I strongly recommend removing the term "biasing" from the title. In this context, it does not convey a specific mechanistic concept. The term "biased signaling" has been used for a very broad range of mechanistically distinct pharmacological phenomena, and without a precise definition, it adds more confusion than clarity. I therefore suggest refraining from using it in the title.

    We appreciate the reviewer’s concern regarding the use of the term “biasing” given its broad and sometimes ambiguous use in the literature. Most commonly it refers to G protein versus arrestin-mediated signalling but is also used in regard to GRK specificity as well as to discrete differences in receptor trafficking (Motso et al. 2025, Cell). In our study, we use this in a specific mechanistic sense, referring to the bias VUP25111 induces upon binding CCL25. The miniprotein: CCL25 complex inhibits CCR9 arrestin recruitment, receptor internalization, and GRK3 engagement while not affecting G protein recruitment (mGi), G protein-mediated signaling (CAMYEL), and cell migration. We compare these directly to the effects of the chemokine alone, which shows robust responses across all of these receptor outputs. The use of the term “biasing” in the title is reflective of this observed change in receptor response to the modified agonist. Additionally, as noted by the reviewer below, we compare similar outputs to compare the impact on effector coupling, eg. Arrestin and mini-G protein recruitment. We strongly feel that the term “biasing” most accurately describes the phenomenon we are describing with regard to the altered receptor response to the VUP25111:CCL25 complex and thus prefer to keep the title without alteration.

    (2) Abstract, line 34: The term "bias" should be replaced. As currently used, it appears to suggest a dichotomy between G protein signaling and arrestin recruitment. However, arrestin recruitment is a consequence of G protein signaling, and it is not conceptually sound to compare a signaling event mediated by one protein family with a protein-protein interaction involving another protein family. A meaningful comparison requires experimental paradigms that differ by a single variable; in this case, there are two - distinct protein families and fundamentally different types of readouts (signaling versus protein-protein interaction).

    We agree that the term “bias” must be used with precision. One point of clarification, arrestin recruitment is not explicitly a consequence of G protein signaling. The atypical chemokine receptors, of which ACKR4 is a member, are well-established examples of GPCRs that do not bind nor activate G proteins yet still recruit arrestins (Lamme et al., 2026). While biased responses within the GPCR system are often the comparison of different events as noted by the reviewer, we compare the effects of miniproteins on the recruitment of mGi and arrestin via a BRET-based assay that assesses the direct protein: protein interactions between the receptor and effector. The lack of effect on mGi recruitment was further supported by measuring the effect of the miniproteins on downstream cAMP levels. To avoid any misunderstanding, we have revised the abstract to explicitly state that the observed effect induced by VUP25111 is evaluated through direct recruitment of effectors, rather than implying a comparison between disparate signaling outputs.

    (3) Abstract, line 34: The term "balanced agonist" should be removed. Any chosen reference ligand is, by definition, the "balanced" agonist for that analysis, regardless of its actual signaling profile. Consequently, the expression "balanced agonist" adds no mechanistic information beyond "the agonist used as reference in a particular bias calculation" and is potentially misleading, as it implies that this ligand possesses a uniquely unbiased, system‑independent signaling profile, which is not the case.

    We agree with the reviewer that the term “bias” is always relative to a specific reference ligand. Indeed, it is known that some natural chemokine:receptor pairs show some bias on its own. Therefore, we adjusted the text in the abstract on Pg.2 and removed the word “balanced”.

    (4) Abstract, line 36: I also recommend removing the term "bias" at this point. The concept of bias typically arises from experiments that quantitatively compare more than one variable. As currently written, the phrasing suggests a dichotomy between G protein- and arrestin-mediated signaling, yet the study does not assess arrestin signaling, only arrestin recruitment. Under these conditions, the use of "bias" is not appropriate. The data are clear and compelling on their own without the need for this potentially misleading terminology.

    We thank the reviewer for this comment. We would like to emphasize that our conclusions are based on the fact that we measure both G protein (mGi recruitment) and arrestin (arrestin recruitment) as protein:protein interactions. We have changed the text in the abstract on pg.2 to further distinguish between signaling and recruitment:

    “Additionally, G protein recruitment to CCR9 was unimpeded by VUP25111, suggesting that the miniprotein biased the native agonist towards G proteins.”

    (5) Introduction: This is interesting to read and generally well written, though certain statements would benefit from improved semantic precision. For example, in lines 110-111, the phrase "G protein-biased complex" should be reconsidered, as it relies on the notion of G protein- versus arrestin-mediated signaling. Arrestins themselves do not signal; what is measured here is their recruitment. Comparing G protein signaling with arrestin recruitment is therefore conceptually unsound, since arrestin engagement is a downstream consequence of G protein activation. Comparisons become meaningful only when designed to differentiate between G protein-dependent and G protein-independent arrestin recruitment, which is not the case in this study.

    Since the VUP25111:CCL25 complex suppresses arrestin recruitment to CCR9, but not G protein recruitment, we are confident representing the modified agonist as G protein-biased. Additionally, arrestin recruitment is only downstream of G protein activation in the occasion where the GPCR is phosphorylated specifically by GRK2/3 and not GRK5/6. CCR9 does not show any preference for GRKs and recruits arrestin as effectively when phosphorylated by either GRK2/3 or GRK5/6 (Lamme JBC 2025). Since G protein coupling is preserved when challenged with the VUP25111:CCL25 complex, both families of GRKs should be available for phosphorylation. We have modified the statement to better specify recruitment compared to signaling on Pg. 5:

    “Most prevent activation by impeding chemokine binding to the receptor, thus impairing all signaling and migratory responses, but one forms a G protein-biased complex with CCL25, preserving G protein recruitment while suppressing arrestin recruitment to CCR9, and selectively inhibits CCR9, but not ACKR4.

    (6) Results, 122,123: The authors should consider being more precise; possibly, the truncated CCL25 is somewhat less potent on CCR9. The authors should make a statistical test and then decide whether to rephrase or not for enhanced precision.

    The reviewer is correct that we detect a small difference in potency between full-length CCL25 and CCL25-ΔCT at CCR9. This was quantified and is reported in the supplementary Table S2, where CCL25 shows a logEC50 of -7.46 ± 0.16 compared with -7.89 ± 0.29 for CCL25-ΔCT, indicating that the truncated chemokine is slightly more potent. While the difference was determined to be statistically significant per an Extra sum-of-squares F test, we don’t consider the of the truncation to be substantially impactful, which is also in line with previous publication (Pinheiro Nat. Comm. 2025). We have therefore revised the text to note the modest increase in potency of CCL25-ΔCT compared to the native, full-length CCL25 and added statistical information in the figure legend. We have changed the text on Pg. 5 to:

    “The truncated CCL25 (CCL25_ΔCT, 1-77) activated both CCR9 and ACKR4 with similar efficacy while showing slightly enhanced potency only towards CCR9 compared to the full-length ligand (Fig. S1).”

    (7) Figure S5: This figure is currently confusing and needs clarification. The authors state in the main text that CXCR4 is stimulated with CXCL12, yet the figure legend refers to CCL25; this discrepancy should be corrected to ensure consistency. In addition, inhibition of CXCR4 by the miniprotein binders should be analyzed and presented with normalization to CXCR4 responses, not to CCL25-stimulated CCR9. To avoid misinterpretation, inhibition by the miniproteins should be quantified separately for CCR9 and CXCR4, each normalized to its own receptor-specific and functionally equivalent stimulation condition, rather than to the "other" receptor.

    We thank the reviewer for this comment and apologize for the oversight during editing. The mentions of CCL25 in Figure S5 were incorrect, and CXCL12 was indeed used in this cross-reactivity assay to stimulate CXCR4 and for subsequent normalization. We have adjusted the legend accordingly.

    (8) Results, lines 211-213: The authors should be more semantically precise. They state that no binder has any effect on arrestin recruitment to CXCR4. If I see the data, this is not really true, as 25101 and 25107 inhibit arrestin recruitment by about 50 % or more at the highest applied concentrations; only 111 and 112 are completely inactive. As already commented, normalization should be done to arrestin recruitment of CXCR4 and not CCR9.

    We thank the reviewer for this comment and agree that, at the highest tested concentrations of VUP25101 and VUP25107, substantial inhibition is observed. Additionally, we have extended our analysis of the miniprotein selectivity to include activation assays for the CCL20 receptors CCR6 and ACKR4, as well as the atypical CXCL12 receptor ACKR3. Based on these data, we have revised the text to more accurately reflect these observations. We now frame these results as an initial proof of concept demonstrating the feasibility of this design approach, while noting that further optimization and characterization of selectivity will be required to advance the miniproteins for additional applications.

    We have adjusted the text on Pg.11:

    “CCL25 specificity was experimentally evaluated by testing the miniproteins against CXCL12 by tracking arrestin recruitment to CXCR4 and ACKR3, the native agonist for this canonical and atypical receptor pair (Fig. S5). VUP25111 and VUP25112 showed no effect on CXCL12-mediated activation at any concentration, while VUP25101 and VUP25107 showed partial inhibition at the highest tested concentration (10 µM).

    In addition, the effects of the VUPs were evaluated against CCL20, a closely related chemokine to CCL25, by assessing arrestin recruitment to CCR6 and ACKR4 (Fig. S6). VUP25107 and VUP25112 displayed similar potencies towards CCL20 comparable to those observed for CCL25 (VUP25107: pIC50 = 7.72 ± 0.21; VUP25112: pIC50 = 6.93 ± 1.03) to ACKR4. In contrast, VUP25101 showed reduced inhibition of CCL20 relative to CCL25 (VUP25101: pIC50CCL20 = 6.06 ± 0.13 v pIC50CCL25 = 7.73 ± 0.08), indicating greater selectivity. CCL20 activation of CCR6 further supports the specificity of VUP25101, which only showed a modest effect at the highest concentration tested. VUP25111 showed no inhibitory effects on CCL20 activation of either CCR6 nor ACKR4, which may indicate a lack of binding, but is indiscernible from forming a receptor-competent complex similar to how it interacts with CCL25 and ACKR4.”

    Concerning the comment regarding the normalization to CCR9 rather than CXCR4, due to a mistake, the figure legend incorrectly stated CCL25, whereas the assay was performed using CXCL12. We thank the reviewer for bringing this up and have corrected the figure accordingly.

  8. Mentioned by @rridrobot.bsky.social on Bluesky

    Authors used pcDNA5-F-ELL2 from @addgene.bsky.social in their study. Including #RRIDs will make this less ambiguous. SciScore made a table with this resource, see “Automated Services” module (download as csv, xml or #jats) #STMpublishing #RRID

  9. Version published to 10.64898/2026.01.23.701245v3 on bioRxiv
  10. Version published to 10.7554/elife.111190.1 on eLife
  11. eLife

    eLife Assessment

    This important study describes computationally designed proteins that bind to the chemokine CCL25. The authors present evidence that some binders simply prevent chemokine binding to the CCR9 receptor, while one binder changes the downstream signaling triggered by chemokine binding. The evidence is solid overall, but some uncertainty remains with respect to functional selectivity due to sensitivity differences between functional assays and the degree of binder selectivity between the large family of chemokine ligands.

  12. eLife

    Reviewer #1 (Public review):

    Summary:

    In this manuscript, the authors describe the use of BindCraft computational protein design to create a series of binders to the chemokine CCL25. This chemokine normally mediates CCR9-dependent trafficking of immune cells to the gut, making it a potential target for the treatment of inflammatory bowel disease and related conditions. Importantly, CCL25 also binds a scavenging receptor, ACKR4. The computational protein design approach used does not involve defining particular epitopes to be targeted, allowing a free search for any potential interaction surface.

    Among four designs tested, three were predicted to interact at a similar site on the chemokine, while a fourth clone, VUP25111, was predicted to bind to a different site. All four designs showed binding to CCL25, with similar high-nM KD values in all cases. The first three clones showed evidence of direct competition with the receptor for CCL25 binding, while VUP25111 showed incomplete inhibition of binding. In functional assays, all clones acted as antagonists except for VUP25111, which inhibited arrestin recruitment by CCR9, but did not affect G protein activation by CCR9 or arrestin recruitment by ACKR4 (which signals exclusively through arrestin and not G protein).

    Strengths:

    The work is completed to a high technical standard, and the functional diversity of the clones is intriguing. It is exciting to see pathway-selective modulation of signaling, and this basic paradigm is likely to generalize to other chemokine/receptor systems. The exceptional complexity of chemokine signaling makes this an excellent area to explore the development of modulators that can restrict signaling to a specific subset of receptors.

    Weaknesses:

    No major weaknesses were noted by this reviewer.

  13. eLife

    Reviewer #2 (Public review):

    This study from de Boer, Lamme, Verdwaald and Schafer describes the de novo AI-guided design of miniproteins that target the chemokine CCL25, with the aim to modulate the activation and signalling of the chemokine receptors CCR9 and ACKR4. The study focuses on characterising four miniproteins that all bind CCL25 with good affinity. Three designs appear to prevent CCL25 binding to both CCR9 and ACKR4, with increasing concentrations of miniproteins resulting in decreased arrestin (both receptors) and mini G protein recruitment (CCR9), less inhibition of forskolin-stimulated cAMP (CCR9), and decreased GRK3 recruitment and receptor internalisation (CCR9). One miniprotein, VUP25111, changes the properties of CCL25 rather than preventing ligand/receptor interactions, resulting in greater selectivity for CCR9 over ACKR4 and a G protein-biased signalling profile (maintenance of mini G protein recruitment, GRK3 recruitment, inhibition of cAMP and receptor internalisation, but loss of arrestin recruitment). VUP25111 also maintained chemotactic migration in MOLT-4 T lymphoblast cells, whereas this response was lost in the presence of the other three miniproteins.

    Overall, this is a very interesting application of AI-designed de novo miniproteins to modulate GPCR responses by directly binding the ligand rather than the receptor. This is a conceptually very intriguing approach that could, in principle, be extended to other GPCR systems beyond the chemokine family. The authors deploy an impressive array of assays spanning multiple signalling endpoints, providing a thorough picture of how each miniprotein influences receptor activation and downstream signalling. The presentation of concentration-response relationships for CCL25 alone and in the presence of each miniprotein is particularly informative, and the figures are very well constructed throughout. The inclusion of clear cartoons illustrating the basis of each assay is a nice touch that will help readers from outside the immediate field follow the logic of each experiment.

    There are two main conclusions that are not currently as well-supported by the evidence as they might be, and that would benefit from some qualification. The first concerns the selectivity of the miniproteins for CCL25. Testing the impact of the miniproteins on CXCL12 activation of CXCR4 is an important and welcome experiment, but it addresses selectivity against only one other chemokine system, and the current claim of specificity is therefore stronger than the data allow. Additionally, at the highest concentration tested (10 µM), the more potent miniproteins (VUP25101, VUP25107) appear to show some inhibition of arrestin recruitment to CXCR4 - perhaps unsurprising given the degree of structural conservation among chemokines. The statements regarding selectivity and the lack of effect on the CXCL12/CXCR4 system would benefit from revision to more accurately reflect these observations.

    The second concern relates to the interpretation of the preserved GRK3 recruitment, but the complete loss of arrestin recruitment observed with VUP25111. In the GRK3 recruitment experiments, 20 nM CCL25 was used, representing an EC40 concentration in this assay. VUP25111 causes a concentration-dependent reduction in CCL25-induced GRK3 recruitment, down to approximately 15% of the maximal response. It is worth considering whether this degree of reduction in GRK3 recruitment could itself be sufficient to disrupt patterns of receptor phosphorylation and thereby prevent observable arrestin recruitment. Both interpretations are complicated by the fact that the GRK3 recruitment and arrestin recruitment assays likely differ in their sensitivity and dynamic windows, making direct quantitative comparisons between them difficult. In the absence of direct measurements of CCR9 phosphorylation in the presence of VUP25111, the alternative interpretation remains open and would benefit from acknowledgement. Given recent work from the same group demonstrating that receptor internalisation is only partially dependent on arrestins (Lamme et al., 2025, J Biol Chem), further discussion of the relationship between GRK and arrestin recruitment and CCR9 internalisation would be of value to the broader GPCR audience this work is likely to attract.

    Finally, some additional justification for the use of 20 nM CCL25 across all assays would strengthen the study, as this concentration represents different points on the concentration-response curve depending on the assay and receptor in question. It ranges from an EC40 for CCR9 GRK3 recruitment and internalisation, to an EC50 for CCR9 arrestin and mini-Gi recruitment, an EC80 for CCR9 cAMP inhibition, and an EMax for ACKR4 arrestin recruitment. This has potential consequences for the interpretation and cross-assay comparison of miniprotein potency, and the authors are encouraged to acknowledge and discuss this in the context of their conclusions.

  14. eLife

    Reviewer #3 (Public review):

    Summary:

    The authors employed the BindCraft platform to develop binders targeting the chemokine CCL25, a selective activator of the chemokine receptor CCR9. They successfully generated two classes of binders: one that inhibits all CCL25-mediated CCR9 activation, and another that permits CCR9 G protein signaling while simultaneously preventing arrestin recruitment. These tools will enable the dissection of arrestin involvement in regulating cell migration.

    My comments, in the order of reading:

    (1) Title: I strongly recommend removing the term "biasing" from the title. In this context, it does not convey a specific mechanistic concept. The term "biased signaling" has been used for a very broad range of mechanistically distinct pharmacological phenomena, and without a precise definition, it adds more confusion than clarity. I therefore suggest refraining from using it in the title.

    (2) Abstract, line 34: The term "bias" should be replaced. As currently used, it appears to suggest a dichotomy between G protein signaling and arrestin recruitment. However, arrestin recruitment is a consequence of G protein signaling, and it is not conceptually sound to compare a signaling event mediated by one protein family with a protein-protein interaction involving another protein family. A meaningful comparison requires experimental paradigms that differ by a single variable; in this case, there are two - distinct protein families and fundamentally different types of readouts (signaling versus protein-protein interaction).

    (3) Abstract, line 34: The term "balanced agonist" should be removed. Any chosen reference ligand is, by definition, the "balanced" agonist for that analysis, regardless of its actual signaling profile. Consequently, the expression "balanced agonist" adds no mechanistic information beyond "the agonist used as reference in a particular bias calculation" and is potentially misleading, as it implies that this ligand possesses a uniquely unbiased, system‑independent signaling profile, which is not the case.

    (4) Abstract, line 36: I also recommend removing the term "bias" at this point. The concept of bias typically arises from experiments that quantitatively compare more than one variable. As currently written, the phrasing suggests a dichotomy between G protein- and arrestin-mediated signaling, yet the study does not assess arrestin signaling, only arrestin recruitment. Under these conditions, the use of "bias" is not appropriate. The data are clear and compelling on their own without the need for this potentially misleading terminology.

    (5) Introduction: This is interesting to read and generally well written, though certain statements would benefit from improved semantic precision. For example, in lines 110-111, the phrase "G protein-biased complex" should be reconsidered, as it relies on the notion of G protein- versus arrestin-mediated signaling. Arrestins themselves do not signal; what is measured here is their recruitment. Comparing G protein signaling with arrestin recruitment is therefore conceptually unsound, since arrestin engagement is a downstream consequence of G protein activation. Comparisons become meaningful only when designed to differentiate between G protein-dependent and G protein-independent arrestin recruitment, which is not the case in this study.

    (6) Results, 122,123: The authors should consider being more precise; possibly, the truncated CCL25 is somewhat less potent on CCR9. The authors should make a statistical test and then decide whether to rephrase or not for enhanced precision.

    (7) Figure S5: This figure is currently confusing and needs clarification. The authors state in the main text that CXCR4 is stimulated with CXCL12, yet the figure legend refers to CCL25; this discrepancy should be corrected to ensure consistency. In addition, inhibition of CXCR4 by the miniprotein binders should be analyzed and presented with normalization to CXCR4 responses, not to CCL25-stimulated CCR9. To avoid misinterpretation, inhibition by the miniproteins should be quantified separately for CCR9 and CXCR4, each normalized to its own receptor-specific and functionally equivalent stimulation condition, rather than to the "other" receptor.

    (8) Results, lines 211-213: The authors should be more semantically precise. They state that no binder has any effect on arrestin recruitment to CXCR4. If I see the data, this is not really true, as 25101 and 25107 inhibit arrestin recruitment by about 50 % or more at the highest applied concentrations; only 111 and 112 are completely inactive. As already commented, normalization should be done to arrestin recruitment of CXCR4 and not CCR9.

  15. eLife

    Author response:

    We thank the editors and reviewers for thoroughly reviewing our manuscript and offering thoughtful and constructive feedback. We appreciate the positive reception of our work and welcome the opportunity to address the lingering concerns. In the coming revisions, we will be directly addressing the question of the miniprotein’s specificity and increase the precision in the language used to discuss our findings.

  16. Version published to 10.64898/2026.01.23.701245v2 on bioRxiv
  17. Mentioned by @biorxiv-biochem.bsky.social on Bluesky

    CCL25 targeting by de novo protein binders differentially suppress CCR9 and ACKR4 activation and transform CCL25 into a biased agonist https://www.biorxiv.org/content/10.64898/2026.01.23.701245v1

  18. Mentioned by @biorxivpreprint.bsky.social on Bluesky

    CCL25 targeting by de novo protein binders differentially suppress CCR9 and ACKR4 activation and transform CCL25 into a biased agonist https://www.biorxiv.org/content/10.64898/2026.01.23.701245v1

  19. Version published to 10.64898/2026.01.23.701245v1 on bioRxiv