Decoding phospho-regulation and flanking regions in autophagy-associated short linear motifs

Abstract

Short Linear Motifs (SLiMs) play a pivotal role in the interactions between intrinsically disordered proteins and their binding partners. SLiMs can undergo regulation through post-translational modifications, including phosphorylation. The flanking regions surrounding the core motifs also exert a crucial role for the interaction. While phosphorylation and flanking regions are known to influence SLiM function, the mechanistic basis of this regulation remains poorly understood. We integrate biomolecular simulations, in silico high-throughput mutational scans, and biophysical experiments to elucidate the phospho-regulation in SLiMs crucial for autophagy, i.e., the LC3 interacting regions (LIRs). We investigate the Optineurin LIR, which perfectly exemplifies a class of LIR with a complex interplay of phosphorylations and flanking regions. Here we show that specific phosphorylation events and flanking residues modulate binding to LC3 at the atomic level, and that disease-associated mutations alter these interactions in the phosphorylated context. Notably, we establish an approach based on Microfluidic Diffusional Sizing to investigate binding affinities of SLiMs to target proteins, complemented by Surface Plasmon Resonance, enabling precise measurements of dissociation constants and kinetics for a selection of variants. Our work provides a versatile toolkit to characterize phospho-regulated SLiMs, advancing the understanding of important cellular processes.

Introduction

Several proteins harbor regions that are intrinsically disordered¹, which support networks of protein–protein interactions^2,3. Intrinsically disordered proteins or regions feature diverse interaction modes with their binding partners, such as folding-upon-binding, conformational selection, formation of dynamic complexes, retained disorder in the bound state or a combination of these mechanisms^4,5,6,7,8. Thanks to their pliability, intrinsically disordered proteins are often implicated in multiple biological processes^2,9, and post-translational modifications can alter their interactome^10,11,12,13.

Disordered proteins frequently utilize short linear motifs (SLiMs) to mediate interactions with their biological partners^4,14,15,16. SLiMs are short and degenerated sequence stretches, formed by a few highly conserved residues that form the SLiM core motif, interspersed with less conserved positions¹⁵. Regions flanking the core motifs and post-translational modifications, such as phosphorylation, can play a crucial role in modulating binding specificity and affinity^17,18,19,20. The contribution of flanking regions and phosphorylation to the binding mechanisms of SLiMs are often challenging to study both experimentally and computationally due to the transient nature of the interactions. Structure-based methods have the potential to provide an accurate understanding of SLiM modulation by phosphorylation or flanking regions, accompanied by a mechanistic understanding of their interactions with biological partners^19,21.

SLiMs involved in autophagy represent an intriguing case study due to their central role in interactions with autophagy receptors and adaptors and a variety of binding modes^19,22,23. The ATG8 (LC3/GABARAP) family of proteins has a key role in both non-selective and selective degradation through autophagy. ATG8 proteins interact with other proteins through the recognition of a SLiM, namely the LC3-interacting region (LIR), which binds to two main hydrophobic pockets (HP) of LC3 or GABARAP structures, known as HP1 and HP2^19,22,23. The LIR motif is often associated with autophagy receptors that mediate the simultaneous binding of LC3/GABARAP components of the autophagy machinery and cytosolic cargo^24,25,26, therefore representing a hub in the autophagy process. LIR amino acid sequences exhibit variability between proteins and share a core motif [W/F/Y]xx[L/I/V]. Both the core motif and flanking residues can influence binding affinity and specificity toward different LC3/GABARAP family members, as demonstrated in previous structural and biochemical studies^27,28,29.

Within the spectrum of LIR-containing proteins, optineurin (OPTN) serves as a selective autophagy receptor and adapter^30,31,32. Beyond autophagy, OPTN participates in diverse cellular processes, particularly in the trafficking of membrane-delivered cargo³⁰—a characteristic feature of scaffolding proteins. OPTN is a multi-domain protein (Fig. 1A) and it harbors an LIR motif (Fig. 1B), which mediates the interaction with LC3 and GABARAP proteins^33,34. This motif has a role in mitophagy amplification rather than in selectivity, suggesting a general mechanism in which LC3 proteins recruit autophagy factors to drive the autophagosome growth and amplify selective autophagy³⁵. OPTN mutations or level alterations are linked to a variety of disorders, including glaucoma, amyotrophic lateral sclerosis, Paget's disease of bone, Crohn’s disease and cancer^36,37,38.

Fig. 1: OPTN LIR region and constructs used in the molecular dynamics (MD) simulations.

A OPTN is a multi-domain protein, and it harbors an LIR motif in a predicted unstructured region. B We summarized the constructs used for the different MD simulations. The list of simulations collected in the study can be found in Table S1. C The AlphaFold multimer model selected for our simulations of the OPTN LIR (158–191) in complex with LC3B (1–120). Key residues, including the ones occupying the HP1 (F178) and the HP2 (I183) of LC3B, and the five phosphorylation sites (S170, S171, S173, S174, and S177) residues are highlighted in the figure (left). The model is color-coded by confidence: dark blue represents very high confidence (pLDDT >90), light blue indicates high confidence (90 > pLDDT >70), yellow corresponds to low confidence (70 > pLDDT >50), and orange signifies very low confidence (pLDDT <50). On the right, the model Predicted Aligned Error (PAE) matrix is shown, illustrating the expected positional error (in Å) between residues. Lower values (dark blue) indicate higher confidence in relative positioning, while higher values (green-yellow) reflect greater uncertainty. The axes correspond to the residue indices of the interacting proteins. Although the model shown here is the truncated version used in our simulations (LC3B_1–120-OPTN_158–191), the PAE matrix corresponds to the full-length construct.

Given these multifaceted roles, the OPTN function needs to be tightly orchestrated and modulated at the post-translational level. In this context, OPTN undergoes phosphorylation, and a subset of these phosphorylations are relevant in the selective autophagy process known as xenophagy, namely the degradation of intracellular pathogens³⁶. Notably, phosphorylation of OPTN by TANK-binding kinase 1 (TBK1) at S177 within its LIR domain substantially increases its affinity to LC3B, enhancing the autophagic clearance of ubiquitin-coated Salmonella enterica^33,34.

Our study presents an integrative approach, combining AlphaFold-based models for interaction^39,40, biomolecular simulations^41,42,43 and structure-based computational approaches^44,45,46,47, with experimental biophysical data based on microfluidic diffusional sizing^48,49 and surface plasmon resonance. This approach has been effective in studying the mechanism through which phosphorylation modulates SLiMs, particularly focusing on the OPTN LIR–LC3B complex and its disease-related mutations. By bridging computational predictions with experimental validation, we offer a robust framework that can be extended to other LIRs or SLiMs regulated by phosphorylation, contributing to a more profound understanding of their role in autophagy or other cellular processes.

Results

Overview of the complex between OPTN and LC3B

The LIR (residues 169–185) of OPTN is located C-terminally to a coiled-coil region and in a disordered tract, according to our trimmed model of the LC3B and OPTN complex (LC3B_1–120-OPTN_158–191) obtained using AlphaFold Multimer³⁹(Fig. 1C). The core region of the motif is encompassed by the sequence 178-FVEI-181, where F178 and I181 occupy the HP1 and HP2 of LC3B, respectively (Fig. 1C). S177 can be phosphorylated in position -1 to the phenylalanine (Fig. 1C). Four additional serine residues (S170, S171, S173, and S174) N-terminal to the LIR core motif can also undergo phosphorylation (Fig. 1C). The structure of the complex between an OPTN LIR variant with the five phospho-serines in complex with LC3B (LC3B_1–119-p5OPTN_169–185) has been solved by NMR³⁴ and used as a starting structure for the modeling and simulations in this work (Table S1).

Moreover, we collected one-microsecond unbiased MD simulations of the unphosphorylated (LC3B_1–120-OPTN_169–185 and LC3B_1–120-OPTN_158–191), mono-phosphorylated (LC3B_1–120-p1OPTN_169–185 and LC3B_1–120-p1OPTN_158–191) and the multiple phosphorylated variants (LC3B_1–120-p5OPTN_169–185 and LC3B_1–120-p5OPTN_158–191), all without the tag sequence used for NMR, and with or without an OPTN N-terminal flanking region that exhibits helical propensity (Fig. 1B, C and Table S1).

MD simulations of LC3B in complex with phosphorylated OPTN are in good agreement with NMR chemical shifts

Before analyzing the MD ensembles, it is crucial to assess the quality of the structures collected in the simulations, namely, the conformational sampling. Hence, we compared the MD ensembles to NMR-derived chemical shift data, similar to what we did for other case studies^50,51,52,53.

For this purpose, we carried out four one-μs MD replicates of LC3B_1–119-p5OPTN_169–185. The associated NMR data includes the assignment of backbone and side-chain chemical shifts³⁴. Chemical shifts, which serve as probes of dynamics along different time scales^54,55,56, can also be predicted by an MD simulations ensemble^57,58,59. The calculated chemical shifts from our simulations agree with the experimental data, exhibiting low χ²_red values. The χ²_red values, generally, reach a plateau after 200-300 ns for most of the atom types except for hydrogens (Fig. 2A). The results for the other replicates and their corresponding plots are reported in the OSF repository associated with the publication⁶⁰ (see Materials and Methods for more details).

Fig. 2: Comparison with experimental data and changes in protein flexibility and secondary structures induced by phosphorylation.

In the following, p1OPTN and p5OPTN refer to the case of the mono-phosphorylated and multiple phosphorylated variants, respectively. A For each replicate of the LC3B_1–119-p5OPTN_169–185 construct (n = 100,000 frames), we estimated a reduced χ² (χ²_red) over the simulation time. This value provides a measurement of the agreement between the experimental chemical shifts from NMR and the calculated chemical shifts from the MD trajectories. The metric is calculated for each atom type. Low values of χ²_red indicate better agreement between experiments and simulations. We reported the example of the MD replicate 1 of LC3B_1–119-p5OPTN_169–185 for the sake of clarity. The results for the other replicates are available in the OSF repository⁶⁰ as detailed in the Materials and Methods. B, C The average Cα RMSF for each replicate (replicate 1: continuous line; replicate 2: dashed-line; replicate 3: square-dot line), calculated using 100-ns time windows (n = 10 independent values) is shown for LC3B_1–120-p5OPTN_158–191 (B) and LC3B_1–120-p1OPTN_158–191 (C) compared to the unphosphorylated construct, respectively. D, E The bar plots show the secondary structure content according to the DSSP dictionary using a concatenated trajectory with all the MD replicates for LC3B_1–120-OPTN_158–191 (D) and LC3B_1–120-p5OPTN_158–191 (E) (n = 300,000 frames for each system). It can be observed that the multiple phosphorylation slightly rearranges the propensity to β-strand formation while also rearranging the α-helix content in OPTN. F The panel shows the mono-dimensional free energy surface (FES) profiles for the parabeta collective variable employed in the metadynamic simulations (n = 506 frames) to describe the intermolecular β–sheet between OPTN and LC3B, i.e., measuring the number of segments that resemble a parallel β-sheet conformation. The wild-type LC3B_1–120-OPTN_158–-191 and penta-phosphorylated LC3B_1–120-p5OPTN_158–191 are highlighted in green and red, respectively. We observe that the phosphorylation induces an increase in intermolecular β-sheet formation.

Additionally, we determined the side-chain chemical shifts for methyl-containing residues (Ile, Val, and Leu). These residues are suitable probes of local conformations (Table 1) and correlate with preferences for different rotameric states of each residue in solution^61,62,63. The predicted values for most chemical shift classes match the experimental data, with low χ²_red values indicating that the MD simulations capture the overall dynamics observed in solution. An exception was noted for isoleucine residues, which shows a χ²_red higher than 3 for the Cδ, depending on the replicate under investigation (Table 1). We further investigated this behavior and noticed that it is mostly due to the predicted chemical shifts for the LC3B isoleucine residues 34 and 67, which are not in direct contact with the LIR-containing peptide (see OSF repository). In the NMR structure, the side chain χ₂ dihedral is mostly in the minus state, according to calculations with Sider (https://www.ucl.ac.uk/hansen-lab/sider/⁶⁴). In the MD simulation, we observed states of these isoleucines with a side chain χ₂ dihedral angle in both the minus and trans states.

Table 1 Comparison between experimental and predicted side chain chemical shifts for methyl groups

Given that the MD simulations are of reasonable quality and in agreement with the protein dynamics in solution on a microsecond timescale, it is feasible to use the MD ensembles to investigate phosphorylation effects and reveal new details of the underlying mechanism.

Phosphorylation reduces the flexibility of the OPTN LIR in the binding pocket of LC3B

Phosphorylation of LIR motifs can decrease the flexibility of the LIR peptide in complex with LC3B⁶⁴, thereby stabilizing the conformation of the LIR residues in the binding pocket of LC3B. We thus investigated whether the phosphorylation of the LIR of OPTN could exert a similar effect. Using the root mean square fluctuation (RMSF) of the Cα atoms of the LIR peptide as a metric, we assessed structural flexibility changes in the LC3B_1–120-OPTN_158–191 constructs. Our observations indicate a marked decrease in LIR flexibility upon phosphorylation of the OPTN LIR (Fig. 2B), in contrast to the unphosphorylated peptide, which displays higher RMSF values along its entire sequence. This effect is evident across various timescales, confirmed by analyses carried out within 100-ns time windows (Fig. 2B). However, phosphorylation of S177 alone is not sufficient to decrease the flexibility of the LIR in the complex (Fig. 2C).

Phosphorylation of OPTN increases the propensity to β-strand formation in the LIR

LIRs, similar to other SLiMs, can have different secondary structure propensities upon binding to an interactor. Indeed, it has been experimentally shown that several LIRs fold upon binding, inducing the formation of an intermolecular parallel β-sheet with the β2 strand of LC3B^25,34. We thus evaluated if the phosphorylation could affect the secondary structure propensity of the OPTN LIR within the complex with LC3B, employing simulations with the LC3B_1–120-OPTN_158–191 constructs (Fig. 2D, E and Fig. S1).

We only observed the two residues downstream of F178 (i.e., 179–180) in the unphosphorylated variant with a propensity for β-strands (Fig. 2D). The phosphorylation of S177 appears sufficient to locally enhance β-strand propensity, particularly at F178 (Fig. S1). However, this modification is also associated with a decrease in β-strand content at I181, while R182 shows a slight increase. Additional phosphorylation does not further increase β-strand formation compared to the mono-phosphorylated variant and instead leads to a reduction in β-strand content at both I181 and R182 in the penta-phosphorylated variant (Fig. 2E and Fig. S1). Beyond β-strand changes, we observe a marked reduction in α-helical content in the N-terminal region preceding the core LIR (residues 160–168), along with a pronounced increase in α-helical propensity between residues 172–176 upon pentaphosphorylation (Fig. 2E). The β-strand of OPTN contributes to the intermolecular β-sheet via interaction with the β-strand 2 of LC3B, as observed in the experimental structures of the multi-phosphorylated variants and phospho-mimetics³⁴.

To evaluate if limited conformational sampling in unbiased MD simulations might explain this behavior, we applied an enhanced sampling method based on metadynamics (see Materials and Methods). The collective variables used for these simulations measured the number of segments resembling a parallel β-sheet conformation. The metadynamics results confirm an increase in intermolecular β-sheet formation following phosphorylation (Fig. 2F).

Phosphorylation of OPTN does not change the solvent accessibility of the core LIR residues but influences charged residues in the LDK tripeptide and in the LC3B HP1

Next, we verified if the decreased flexibility of the peptide in the pocket, caused by the phosphorylation, was also associated with changes in solvent accessibility for the side chains of the two hydrophobic residues in the LIR core motif (F178 and I181). We cannot observe marked differences in the distribution of the solvent-accessible surface area (SASA) values of the residues occupying the HP1 and HP2 of LC3B (Figs. S2, S3). Furthermore, we evaluated if the residues lining the HP1 and HP2 of LC3B, the LDK tripeptide, or the LC3B ubiquitin patch, displayed changes in solvent accessibility. A notable change is detected for K51 in the HP1, where phosphorylation of S177 alone is sufficient to induce this effect (Fig. 3A). Moreover, multiple phosphorylations further decreased the solvent accessibility of D48 and K49 within the LDK tripeptide (Fig. 3B, C). Other LC3B residues did not show alterations in side-chain accessibility upon phosphorylation.

Fig. 3: Phosphorylated OPTN158–191 causes changes in solvent accessibility of charged residues of LC3B at the interaction interface.

A–C Relative solvent-accessible surface area (SASA) of side chain atoms for K51 (A), K49 (B), and D48 (C) is shown across three replicates (n = 100,000 frames each) per system, along with comparative simulations based on the NMR structure of the LC3B–OPTN complex (n = 100,000 frames). The main systems (LC3B₁₋₁₂₀–OPTN_158–191, LC3B_1–120–p1OPTN_158–191, and LC3B_1–120–p5OPTN_158–191) are derived from AlphaFold models, while the comparative simulations were initialized from the first conformer of the NMR structure 2LUE (residues 169–185 of OPTN and 1–120 of LC3B), with matching phosphorylation patterns. Each box represents the interquartile range (IQR, 25th to 75th percentile), with the horizontal line inside the box indicating the median. Whiskers extend to data points within 1.5×IQR from the lower and upper quartiles. Outliers are not shown. The AlphaFold-based models include a longer OPTN fragment (residues 158–191), and during simulations, the additional flanking regions can transiently shield LC3B residues such as D48 and K51, potentially reducing their solvent accessibility relative to the shorter experimental construct.

In summary, while decreased structural flexibility does not appear to alter the solvent accessibility of the LIR core motif residues, it does affect the accessibility of the interacting residues in LC3B.

Phosphorylation increases the interaction strength between the LIR and LC3B both locally and long-range

Furthermore, we evaluated whether changes in solvent accessibility and flexibility were associated with changes in the interaction strength between OPTN and LC3B. We assessed pairwise interaction strengths, calculated from the atomic-contact-based protein structure network (see Materials and Methods, Table S2 and Fig. 4A).

Fig. 4: Changes in intermolecular interactions upon phosphorylation.

A The histograms report the results of pairwise Interaction Strength as calculated using an atomic-contact-based protein structure network for the pairs more sensitive to changes upon phosphorylation for the LC3B_1–120-OPTN_158–191, LC3B_1–120-p5OPTN_158–191 and LC3B_1–120-p1OPTN_158–191 constructs (n = 3 independent values per system). In each histogram, the boxes represent the average interaction strength, and the error bars indicate the standard deviation across the three replicates. The full set of data on interaction strengths is reported in Table S2. B, D Analysis of intermolecular contacts between LC3B and OPTN during our simulations using CONAN. The interaction patterns for OPTN residues E163 (B), L166 (C), and M183 (D) are shown. Each panel displays the contact patterns for replicate 1 of LC3B_1–120-OPTN_158–191 (left, n = 100,000 frames), LC3B_1–120-p1OPTN_158–191 (center, n =100,000 frames), and LC3B_1–120-p5OPTN_158–191 (right, n = 100,000 frames). E–G The networks of salt bridges are reported on the 3D structure of the first frame of the MD simulations of LC3B_1–120-OPTN_158–191 (E), LC3B_1–120-p1OPTN_158013;191 (F), and LC3B_1–120-p5OPTN_158–191 (G) using the MD replicate 1 for the sake of clarity. The salt bridges are depicted as gray sticks, whose size is proportional to their occurrence in the simulation. The full set of data is reported in Table S3. H Correlation analysis of pairwise distances between residues involved in salt bridges (E175-K51, D176-K51, E175-K49, and D176-K49). The Pearson correlation coefficient was used as a metric.

We observe that phosphorylation at pS177—and to a greater extent, at all five serine residues— increases the interaction strength between F178 of OPTN and I23 of the HP1 of LC3B. Moreover, it also increases the strength of interaction between V179 of OPTN and F52 of the HP2 of LC3B. In contrast, the LIR core residue I181 and the C-terminal flanking region residue M183 (Table S2) exhibit similar interaction strengths across all simulations. Multiple-site phosphorylation also enhances long-range interactions, notably between E163 in the N-terminal flanking region of the OPTN LIR and R37 of LC3B (Fig. 4A). This broader effect is not replicated by phosphorylating S177 alone. The role of E163 as an interaction hotspot in the penta-phosphorylated OPTN variant is further supported by additional analysis using another atomic-contact-based method⁶⁵. This method evaluates not only contact occurrences but also the frequency of contact formation and breaking during simulations. (Fig. 4B). This analysis also reveals K42 as an additional stable intermolecular interaction partner for E163 post-phosphorylation. It also identifies L166 as another site in the N-terminal flanking region of OPTN that can be altered by phosphorylation at S177 (Fig. 4C), and interacts with other hydrophobic residues such as L71, L47, and V46. On the other hand, we did not observe any changes in the interaction patterns between the unphosphorylated and phosphorylated variants for M183 (Fig. 4D) or for the additional residues associated with disease-related mutations that are predicted to be pathogenic by EVE⁶⁶ or AlphaMissense⁶⁷ (E175, D176, F178, and V179; Fig. S4A–D). However, we did observe an increase in the contact occurrence between E175 and R11 (Fig. S4A) and between V179 and R70 in the phosphorylated variants (Fig. S4D).

Phosphorylation rewires intermolecular salt bridges in the LIR complex with local and long-range effects

To evaluate if phosphorylation of the OPTN LIR affects electrostatic interactions, we estimated salt bridges (Table S3 and Fig. 4E–G) from MD simulations of the LC3B_1–120-OPTN_158–191 variants (see Materials and Methods).

The interaction between E180 of OPTN and K30 of LC3B is the only salt bridge that is highly conserved (>90% occurrence) across all complexes. The unphosphorylated variant exhibits a varied pattern of salt bridges with occurrence exceeding 20% depending on the MD replicate, ranging from only E180-K30 in replicate 3 to sparser and lower-occurrence salt bridges involving various residues at the interaction interface in replicates 1 and 2 (Table S3). Interestingly, E175 and D176 of OPTN can alternate as interaction partners for K49 and K51 of OPTN within a fluid salt-bridge network, as evidenced by a correlation analysis of the time series of their distances (Fig. 4H). In particular, the formation of salt bridges by E175 and D176 with K51 is negatively correlated with their interaction with K49, suggesting mutual exclusivity (Fig. 4H).

Phosphorylation at S177 rewires the electrostatic interactions, leading to a more uniform and less dynamic behavior. High-occurrence salt bridges are strengthened and centralized within the interface, involving three main players: D176, pS177, and E180, which stably interact with K51, K49, and K30, respectively (Table S3 and Fig. 4E, G). The interaction between K51 and pS177 agrees with findings from NMR experiments and the observed affinity changes resulting from a K51A mutation in shorter OPTN constructs³⁴.

Similar interaction rewiring is observed in MD simulations of the penta-phosphorylated variant. Moreover, pS170 and pS171 contributed to stabilizing and expanding the network of electrostatic interactions involving R10 and R11 of LC3B. Consistent with previously observed long-range effects, multiple phosphorylations promote more stable electrostatic interactions in the N-terminal region of OPTN, with E163 and K167 participating in salt bridges, and pS174 strengthening the interaction with K49 (Table S3 and Fig. 4E–G). Other phospho-residues are not involved in stable salt bridges with LC3B. These findings corroborate and provide a mechanistic explanation for the decreased solvent accessibility of K49 and K51 in the phosphorylated variants, as observed in Fig. 3.

Interestingly, K49 also engages in a salt bridge with the residue corresponding to S177 in the LIR motif of p62 (i.e., D337), suggesting that this lysine may participate in the LIR-binding process by recruiting negatively charged residues from various positions in the N-terminal region of the LIR peptides. This is in line with previous MD simulations showing electrostatic interactions between K51 of LC3B and D1021 in a phosphorylated LIR peptide of AMBRA1, highlighting a broader role for this lysine in LIR recognition⁶⁴.

Analysis of intermolecular hydrogen bonds, involving either main chain or side chain atoms across different MD replicates, confirms this trend of interaction rewiring (Table S4).

In a related study with shorter OPTN LIR constructs, the phosphorylated structure was studied using NMR for the phosphorylated variants, and an X-ray structure was solved using a penta-phosphomimetic variant that replaced serines with glutamates³⁴. In the X-ray structure, the authors observed an inward flip of LC3B’s R11 to make hydrogen bonds with the side chain of E177 in OPTN. The amide proton of R11’s side chain, however, was not detected by NMR. We thus evaluated whether our simulations with longer constructs replicate this interaction between R11 and pS177. Monitoring the distances over time between the side chains of these residues in the penta-phosphorylated simulations and analyzing their probability distribution (see OSF repository), we observe distances larger than 10 Å, with R11 predominantly interacting with another phosphorylation site, pS171. The mechanism suggested here could still explain the observed NMR chemical shift changes in R11 upon phosphorylation. A control simulation with a shorter construct (LC3B_1–119-p5OPTN_169–185) does feature the interaction between pS177 and R11, consistent with the model derived from NMR data and deposited in PDB.

Phosphorylation does not reduce the conformational changes of the F-type LIR in the aromatic residue for binding to the HP1 of LC3B

OPTN has an F-type LIR, i.e., at the core position for binding to the HP1 of LC3 protein, the usual tryptophan is replaced by a phenylalanine. Phenylalanine could be more prone to undergo conformational changes in the binding pocket that cause a flipping of its side chain outside the binding pocket. We speculate that phosphorylation could be a mechanism to reduce the flipping and provide a stronger binding to HP1. We thus used enhanced sampling simulations to address this hypothesis based on metadynamics.

At first, we analyzed the unphosphorylated variant to evaluate if the flipping could occur, and in which population of the conformational ensemble it might be present. We observe one favored state, where the side chain of OPTN F178 is well-buried in the LC3B HP1, in contact at distances of less than 5 Å with F108 of LC3B, and its χ₁ dihedral is either in a minus or trans state (Fig. S5). There is also a minor state, characterized by a flipped conformation of F178 outside the binding pocket where the χ₁ dihedral angle is in a plus conformation and the distance from F108 is greater than 9 Å. The introduction of one phosphorylation at position S177 seems to change the distance between the F178 side chain of OPTN and F108 of LC3B, favoring a state where they are not in direct contact, whereas the five phosphorylations are not sufficient to stabilize F178 within the HP1 of LC3B (Fig. S5).

High-throughput mutational scans identify binding hotspots and residues conferring specificity for the binding mode of the phosphorylated variant

The integration of MD-derived ensembles of conformations with empirical free energy calculations^{52,53,68,69,70,71} could shed light on hotspots for binding between LIR and LC3B proteins. In the structure-based framework that we recently developed, namely the MAVISp framework⁷², we rely on a consensus approach based on FoldX and Rosetta. However, Rosetta-based free energy calculation protocols do not support phospho-residues, necessitating reliance on phosphomimetics instead. To determine the effectiveness of phosphomimetics in replicating phosphorylation effects, we used synthesized OPTN LIR peptides, introducing a phosphoserine at position 177 (pS177), as well as phosphomimicking substitutions (S177D and S177E). Subsequently, we conducted microfluidic diffusional sizing (MDS) experiments to evaluate their binding affinity to GST-LC3B (Fig. 5A). The pS177 peptide shows an affinity of 14.9 µM (Fig. 5A1), while the S177D and S177E variants exhibit significantly lower affinities of 36.7 µM (Fig. 5A2) and 40.2 µM (Fig. 5A3), respectively. The WT peptide affinity is 67.1 µM (Fig. 5A4), aligning with ITC data, which indicates an affinity range of 40–60 µM^33,34. Therefore, it appears that the single phosphomimetic mutation at position 177 might confer a modest increase in binding affinity compared to the wild-type sequence. Still, the mimetics do not reach the level observed with authentic phosphorylation. Consequently, we decided not to include Rosetta-based calculations in this study and relied on FoldX calculations, using 25 representative structures from each of the MD replicates of the constructs LC3B_1–120-OPTN_169–185 and LC3B_1–120-OPTN_158–191 from Table S1. We had three main goals: (i) to determine whether the mutational scan could predict the experimentally known stabilizing effect of phosphorylations on binding^33,34; (ii) to provide a baseline for similar future studies on other phosphorylatable LIRs; (iii) to use the data to delve deeper into the structural mechanisms facilitated by the phosphorylation.

Fig. 5: High-throughput mutational scans.

A Binding isotherms of OPTN_169–185 (1) pS177 (n = 5 independent experiments), (2) S177D (n = 3 independent experiments), (3) S177E (n = 3 independent experiments), and (4) WT (n = 3 independent experiments) FITC-labeled peptide variants to GST-LC3B by microfluidic diffusional sizing. Raw data from all technical replicates are shown as individual dots. Each biological replicate is fitted independently, and the resulting nonlinear curves are overlaid. The reported K_d, Rh_free, and Rh_complex values represent the mean ± standard deviation across the biological replicates, with each value obtained from a separate fit. R² values indicate the goodness of fit for each individual curve. B–D The heatmaps for the saturation scan in silico for LC3B_1–120-OPTN_158–191 (B), LC3B_1–120-p5OPTN_158–191 (C), and LC3B_1–120-p1OPTN_158–191 (D) are illustrated with a truncation of the ΔΔG values at 3 kcal/mol for visualization purposes. We can observe that the pentaphosphorylation activates binding hotspots in the N-terminal flanking region for the binding to LC3B (B, C). The effect cannot be recapitulated by the single phosphorylation (C, D). E, F The dotplot summarizes the results of the MAVISp assessment in the simple (E) and ensemble (F) modes. Black, white, and gray dots represent mutations predicted with a damaging, neutral, or uncertain effect for one of the MAVISp modules, respectively.

Initially, calculations were designed to evaluate the effect of introducing phosphorylations on the unphosphorylated complex’s structure. We also examined the phosphorylated complex structures, assessing the effect of dephosphorylation, which should, according to FoldX, destabilize binding. This “reverse approach” compensates for FoldX limitations in predicting stabilizing effects^73,74, and might serve as an effective alternative strategy for de novo discovery of phosphorylation-induced effects on SLiMs. Specifically, this approach entails a mutational scan with FoldX on ensembles of the phospho-variant, assessing the effect of reverting mutations to the original unphosphorylated amino acid. If a PTM/mutation shows a marginal effect when introduced from the wild-type ensemble but a clear destabilizing effect when removed from the modified variant’s ensemble, it can be more confidently deemed stabilizing for the wild-type protein.

In our comparison with experimental data, we also included the F178W variant of the OPTN LIR as a control. While this mutation has been experimentally shown to enhance binding affinity³⁴, our FoldX calculations do not reproduce a clear stabilizing effect. Only two structures from replicate 1 and one from replicate 2 showed a predicted ΔΔG below −1 kcal/mol. The lack of a consistent stabilizing signal may be attributed to limitations of the FoldX energy function in accurately capturing the effects of stabilizing mutations.

When introducing phosphorylations into the unphosphorylated complex, we find average ΔΔG values lower than 0, with pS177 and pS174 being the most significant modifications for enhancing binding affinity. While the average differences fall below the 1 kcal/mol threshold typically used to pinpoint stabilizing or destabilizing variants, some conformations from the MD simulations do feature values beyond this threshold, akin to the F178W mutation introduced in the unphosphorylated variant. With the reverse approach, namely using the phospho-variant and measuring the effect of reversion to serine, we note a slight average destabilizing effect (average ΔΔG values of 0.4–0.8 kcal/mol). These small changes may result from the limitation of the MutateX protocol to one mutation at a time⁶⁸, implying that the other phosphorylations were included in the model when assessing a single site’s shift from phosphoserine to serine. We avoided calculations using simultaneous mutations of multiple phospho-sites since empirical free energy functions with local sampling, like those applied here, often fail to accurately capture the contribution of multiple mutations, instead yielding predictions based on additive contributions of individual mutations.

Furthermore, we used the saturation scan data to identify residues that specifically contribute to binding affinity upon phosphorylation due to interaction rewiring at the binding interface (Fig. 5B–D). For this purpose, we selected residues that act as binding hotspots in the LIR (or LC3B) only in the penta-phosphorylated form and with marginal contributions in the unphosphorylated variant (Table S5 and Fig. 5B–D). We also examined whether the mono-phosphorylation at S177 was sufficient to promote similar effects (Table S5 and Fig. 5D).

Beyond confirming the importance of phosphorylation sites, we identify three sites in the N-terminal flaking region crucial only to the phosphorylated variant: E163 (when mutated to S, G, T, H, or R), various mutations at the positions L166 (especially substitutions to I, M, F, or R), and K167 when mutated to small, aromatic, or negatively charged residues. Single phosphorylation does not appear to induce the same N-terminal effects as observed with the penta-phosphorylated variant (Table S5). Conversely, E175 plays a more central role in the interaction of the unphosphorylated variant than in the penta-phosphorylated one, aligning with observed changes in salt-bridge interactions (Fig. 4E–G). D176 and M183 are sensitive to changes in binding free energies upon mutation in both the unphosphorylated and phosphorylated variants.

Different amino acid preferences upon phosphorylation are also noted for LC3B sites at the interface. For example, mutating D19 to aromatic or positively charged residues enhances complex stability across all peptide variants, suggesting this position contributes generally to binding affinity rather than showing phospho-specific effects (Fig. S6A–C). In contrast, K49 and K51 exhibit moderately stronger interactions with the phosphorylated peptides compared to the unphosphorylated form (Fig. S6A, B). Notably, R37 emerges as a distinct binding hotspot only in the LC3B–p5OPTN complex (Fig. S6A), consistent with its proposed role in forming stabilizing interactions and salt bridges with the N-terminal region of OPTN.

Effects of disease-related mutations of OPTN on the binding with LC3B

The methodologies described above can also be applied to examine the effects of disease-related variants that present uncertain significance or conflicting evidence, posing challenges to biomedical research⁷⁵.

We have identified twelve variants reported in cancer databases and ClinVar (Table 2 and Fig. 5E, F), located within the OPTN LIR or its flanking regions (residues 158–191). A structure-based framework was utilized to associate predicted pathogenic effects with the molecular mechanisms driven by the mutations⁷².

Table 2 Disease-related mutations and pathogenicity scores

None of the variants appears to impact OPTN homodimerization, and consequently, its potential oligomerization, though the effects of E175Q, F178V, and V179D remain uncertain according to the analyses (Fig. 5E).

The F178V variant is predicted to have pathogenic effects due to the disruption of interaction with LC3B (Fig. 5E, F). The predicted pathogenic variant D176Y has damaging effects on binding with LC3B for the unphosphorylated forms of OPTN, as determined by the ensemble mode of MAVISp, which considers protein dynamics (Fig. 5F and Table 2). L166V appears to negatively impact binding only for the phosphorylated form of the OPTN-LC3B complex, and this effect is not replicated by a single phosphorylation at S177 (Table 2).

The differential impact of disease-related mutations based on the phosphorylation status of the protein is intriguing. Current approaches to predict the effects of mutations often overlook the interplay between the mutation and the protein's phosphorylation status. Our results highlight the need for a new module in the MAVISp framework’s ensemble mode, capable of discerning mutation effects influenced by PTMs.

Notably, variants V179D, V161M, and E163D are listed in ClinVar as variants of uncertain significance for glaucoma and amyotrophic lateral sclerosis. While EVE and AlphaMissense suggest a potential pathogenic effect for E163D and V179D (Fig. 5E, F and Table 2), our structure-based assessment indicates they have neutral effects. Conversely, both the MAVISp assessment and all the pathogenicity scores agree on the neutral effect of V161M, which aligns with findings from a population study for the same variant⁷⁶.

Microfluidic diffusional sizing can accurately estimate changes in K _d and hydrodynamic radii upon substitutions in the phosphorylated OPTN LIR

To further evaluate the stabilizing or destabilizing effects of amino acid substitutions on the interaction between LC3B and phosphorylated optineurin, we implemented MDS^48,49,77 to examine the binding affinity between GST-tagged LC3B and peptide variants derived from the full-length OPTN. The length of the peptides was designed to maximize binding stability. While longer peptides can enhance stability in some cases, they may also introduce flexibility that destabilizes interactions or reduces binding specificity, especially when the interaction motif consists of only a few residues, as in the case of the OPTN SLiM^78,79. Therefore, peptide-based studies should carefully consider the optimal peptide length to balance stability and specificity. In our case, the longer OPTN 160–185 peptide destabilized the interaction with LC3B (Fig. S7). To address this, we designed FITC-labeled peptides corresponding to residues 169–185, which encompass the SLiM while minimizing excess residues. These were titrated against GST-LC3B, and binding was assessed via changes in hydrodynamic radius to determine K_d values. The minimal sample volume required for GST-LC3B per binding curve (8 µL) enabled us to concentrate GST-LC3B to the high micromolar levels necessary to saturate binding signals of most peptides. Due to synthesis challenges with penta-phosphorylated peptides, we introduced substitutions in the mono-phosphorylated pS177 peptide of OPTN. Luckily, our simulations and free energy scans indicated that many structural effects observed in the penta-phosphorylated variant are also reflected by this single phosphorylation. As previously discussed, the WT peptide’s affinity is 67.1 µM (Fig. 5A), approximately fivefold lower than the 14.9 μM for the mono-phosphorylated variant (Fig. 5A), corroborating earlier isothermal titration calorimetry findings³⁴. Phosphomimetic substitutions S177D and S177E fail to match the affinity of actual phosphorylation, showing significantly different K_d values (Fig. 5A). Additionally, we included the F178W substitution as a control, which results in a K_d of 19.1 μM, comparable to the pS177 variant (Fig. 6A, E) as reported previously³⁴. This F178W substitution, alongside the pS177 phosphorylation, enhances binding affinity sevenfold compared to phosphorylation alone (Fig. 6B, E).

Fig. 6: Equilibrium binding curves by microfluidic diffusional sizing.

A–D Binding isotherms of OPTN_169–185 A F178W, B pS177 F178W, C pS177 M183A, and D pS177 F178A FITC-labeled peptide variants to GST-LC3B by microfluidic diffusional sizing. Raw data from all technical replicates are shown as individual dots. Each biological replicate is fitted independently, and the resulting nonlinear curves are overlaid. The reported K_d, Rh_free, and Rh_complex values represent the mean ± standard deviation across the biological replicates (n = 3 independent experiments), where each value is obtained from an individual fit of the binding curve. R² values indicate the goodness of fit for each curve. E–G Comparison of E K_d, F Rh_free, and G Rh_complex values across the tested peptide variants. Dots represent the individually fitted values from each biological replicate, while bars indicate the mean ± standard deviation across replicates. P values are generated through t-tests or t-tests with Welch’s correction, as described in Methods, to assess differences to the WT or pS177 variant.

To examine variants in the LIR flanking region and test MDS’s sensitivity to minor changes in binding affinity, we evaluated the M183A substitution, predicted to weaken the interaction for both the phosphorylated and unphosphorylated variants of OPTN by ~2 kcal/mol. We observe similar K_d values, within the experimental error, for the combined S177 phosphorylation and M183A substitution compared to phosphorylation alone (Figs. 6C, E). Our findings suggest that the MDS binding assay can identify and validate predicted binding affinity changes exceeding 2 kcal/mol. Therefore, we did not experimentally explore variants listed in Table 2 or S4, since predicted average changes for mutations outside the core motif were mostly between 1 and 2.5 kcal/mol (Table S5).

An alanine substitution at OPTN’s position 178 (F178A) abolishes the hydrophobic residue for interaction with LC3B’s HP1. Combined with the phosphorylation at position 177, this variant leads to a sevenfold increase in K_d compared to pS177 alone (Fig. 6D, E), consistent with in silico scans indicating an average binding ΔΔG increase of ~4.5 kcal/mol. This suggests that phosphorylation cannot offset the affinity loss caused by removing the central core motif residue, echoing the potential deleterious effect of the shorter side-chain in the disease-related F178V variant (Fig. 5E, F).

One of the key advantages of MDS over other affinity measurement methods is its ability to estimate affinity based on changes in the hydrodynamic radius (Rh). This is done by measuring the transition from the Rh of the free, labeled monomer—here, the FITC-labeled peptide (Rh_free), to the Rh of the complex formed with an unlabeled partner—in this case, GST-LC3B (Rh_complex). Theoretical Rh values can be predicted⁸⁰, allowing the measured Rh_free and Rh_complex to act as controls for potential aggregation, to determine the stoichiometry of the complex, or to infer changes in the structure or hydration shell of the protein or complex.

For most FITC-labeled peptides, we anticipate an Rh_free in the range of 0.9–1.2 nm, and, upon interaction with GST-LC3B, an Rh_complex between 2.6–3.2 nm. The WT peptide has an average MDS-measured Rh_free of 1.2 nm, suggesting it likely exists as a monomer in solution. Phosphorylation at S177 significantly increases Rh_free to 1.3 nm (Fig. 6F), potentially indicative of an expanded hydration shell. However, given the instrument’s measurement of the average Rh in solution, this does not rule out the possibility of some peptide aggregation. Other phosphorylated and phosphomimetic variants also exhibit a minor increase in Rh, though this change is not statistically significant (Fig. 6F).

Upon evaluating the sizes of the complexes formed with GST-LC3B, we observe that they exceed the predicted upper limit of 3.2 nm yet align with a 1:1 stoichiometry (Fig. 6G). The WT complex measures 3.9 nm on average, suggesting a looser association. In contrast, the complex with the pS177 peptide compacts to 3.5 nm, likely reflecting a stronger interaction as inferred from the affinity data. This trend of compaction persists across all complexes containing either the pS177 or the F178W substitution, unlike the complexes formed with phosphomimetic variants S177D and S177E, which remain at about 4 nm. The smallest observed complex is with the F178A variant; however, the reliability of this size estimate is compromised due to the non-saturated binding curve within the GST-LC3B concentration range used, which prevents a precise determination of the complex's size.

Surface plasmon resonance reveals phosphorylation-driven dynamics in OPTN LIR–LC3B interactions

We then investigated quantitative kinetic constants for the interaction between the OPTN LIR and its partner LC3B^80,81,82. While direct kinetic measurements of LIR interactions remain challenging due to their rapid on-off kinetics, several studies—including those using surface plasmon resonance (SPR)^81,82,83—have investigated both steady-state and kinetic parameters for LC3B interactions, including with selective autophagy receptors^84,85. However, comprehensive kinetic characterization remains limited compared to affinity studies.

Our strategy involved immobilizing GST-LC3B onto a CM5 sensor chip and introducing increasing concentrations of OPTN LIR peptides. The sensorgrams generated were then modeled to a 1:1 binding interaction, from which we derived the association (k_on) and dissociation (k_off) rates (Fig. 7A–F). Given that the k_off values exceeded the measurement capabilities of the Biacore X100 instrument, the fits are acknowledged as suboptimal. Conversely, most k_on values are on the brink of the instrument's detection limit, yet they appear to be distinctly determinable. Although the association phases in some sensorgrams deviate from the idealized model, our global fitting approach—standard for such rapid on-off kinetics—provides a robust means of extracting the key kinetic parameters. Therefore, we leaned on the measured k_on and steady-state K_d values, corroborated by MDS measurements, to extrapolate k_off rates. Despite the apparent stability issues with GST-LC3B on the chip that precluded multiple measurements, our analysis yielded meaningful insights. The sensorgram analysis of the OPTN_169–185 unphosphorylated peptide revealed a steady-state K_d of 63.5 µM, consistent with MDS-derived and previously published values^33,34. The k_on was measured at 0.55 × 10³ M⁻¹s⁻¹, and with the steady-state K_d, the k_off was computed at 3.51 × 10⁻² s⁻¹. Interestingly, while the off-rate aligns with the affinity of the nanomolar-affinity K1 peptide for GABARAP identified through phage-display⁸⁶, the µM affinity of the natural OPTN LIR peptide is primarily modulated by a significantly reduced on-rate. Compared to the natural cysteine protease ATG4B, whose LIR binds LC3B with a much higher on-rate, OPTN shows a significantly slower association despite similar dissociation kinetics⁸⁷. This indicates that the lower affinity of the OPTN LIR arises predominantly from inefficient complex formation rather than increased dissociation. Introducing a phosphorylation at position 177 (pS177) in the OPTN_169–185 peptide notably enhances the interaction dynamics with LC3B. This modification leads to a ~30-fold increase in the on-rate to 17.78 × 10³ M⁻¹ s⁻¹ and about a tenfold increase in the off-rate to 37.73 × 10⁻² s⁻¹. However, the net effect is a threefold improvement in steady-state affinity to 21.2 µM due to the more pronounced increase in the on-rate. Phosphorylation creates new electrostatic environments that can stabilize certain conformations, which may be more “binding-competent” and enhance the likelihood of interaction, reflected in increased k_on. Once bound, the same electrostatic changes may also make the interaction energetically less favorable, leading to a quicker dissociation, reflected in increased k_off. This dual effect—stabilization of certain conformations, yet destabilization of the bound state—can lead to overall more dynamic interactions, as the binding occurs more readily but is less permanent. This dynamic shift is also observed, albeit to a lesser degree, with the phosphomimetic S177D substitution, leading to a steady-state affinity of 36.7 µM—improved from the wildtype but not as potent as the true phosphorylation.

Fig. 7: Kinetics of binding between the LIR motif of optineurin and GST-LC3B by surface plasmon resonance.

A Summary of the steady-state dissociation constants (K_d) and on-rates (k_on) for the five peptide variants tested: OPTN_169–185 WT, S177D, pS177, pS177 M183A, and pS177 F178A. The off rates (k_off) were calculated from these values. The averages and standard errors were derived from nonlinear regression fits of the binding curves. B–F Representative SPR sensorgrams showing the binding interactions of increasing concentrations of OPTN169–185 peptide variants B WT, C S177D, D pS177, E pS177 M183A, and F pS177 F178A with amine-coupled GST-LC3B on a CM5 chip. The colored traces represent experimental data, while the black lines indicate the kinetic fits based on a 1:1 binding model. Insets display the steady-state resonance values taken 10 s before the end of injection, ƒcollected in 5-s intervals, and plotted against peptide concentration to calculate steady-state K_d values (black line fits).

When evaluating the pS177 peptide, substitutions in the flanking region (M183A) and core LIR motif (F178A) had detrimental effects on the steady-state affinity, impacting both kinetic rates. The M183A substitution led to nearly a twofold drop in affinity to 36.7 µM, corroborating the in silico scans and suggesting a potential role of residue M183 of OPTN in the interaction with LC3B. However, limitations in SPR data reproducibility prevent a robust significance assessment. The F178A substitution within the pS177 peptide, on the other hand, induced a marked decrease in affinity, with a K_d exceeding 100 µM, echoing the MDS findings and underscoring that phosphorylation cannot compensate for the loss of the critical aromatic residue occupying HP1 of LC3B.

Our study not only fills a gap in the current understanding of LIR motif dynamics but also establishes a reliable experimental framework for future investigations into the kinetics of these transient but pivotal interactions.

Discussion

Phosphorylation can modulate LIR motifs and compensate for their low binding affinity. Phosphorylatable sites have been reported and experimentally verified for other LIR motifs and shown to impact binding affinity and recognition of LC3/GABARAP proteins¹⁹. It is thus of fundamental importance to investigate the role of phosphorylation on structure and function of LIR-containing proteins and their complexes.

In our study, we employed the OPTN LIR as a model system to dissect the nuances of phospho-switches LIR–ATG8 recognition. This choice was underpinned by the availability of comprehensive biophysical data pertaining to its phosphorylation and mutations, allowing for a rigorous validation of our simulation data^33,34. Moreover, we provided new structural insights by combining binding affinity and size measurements using Microfluidic Diffusional Sizing (applied here for the first time in the context of natural autophagy receptor LIR–ATG8 interactions) with kinetic analysis by SPR, extending previous SPR-based characterizations of these complexes. In addition, we shed light on the role of the flanking regions of the core LIR motif, the N-terminal as a sensor of long-range effects induced by OPTN phosphorylation, as well as the C-terminal, with residue M183 as a potential hotspot for binding LC3B.

Our findings, augmented by previous results from another LIR-containing protein⁶⁴, show that phosphorylation notably affects the flexibility of the entire LIR peptide within the formed complex. Moreover, the PTMs can induce long-range effects, stabilizing the electrostatic interactions in the flanking region of the peptide, rewiring the network of intermolecular interactions. Such rewiring modifies the binding hotspots within the flanking regions, altering their contribution to the binding with LC3B. We also observe that pentaphosphorylation of the LIR motif may stabilize the complex by inducing local structural rearrangements, including β-strand content at F178, I181 and R182 and a redistribution of α-helical propensity around the core motif. However, the multiple phosphorylations do not prevent the flipping movement of the OPTN F178 side chain out of the HP1 of LC3B. We note that the phosphorylation at residue pS177 alone is sufficient to promote changes in solvent accessibility for LC3B residues involved in interactions with the central part of the OPTN LIR motif. However, the mono-phosphorylation is not sufficient to decrease the LIR flexibility in the complex and cannot elicit long-range effects to recruit residues from the N-terminal flanking regions for complex formation.

Linking the mechanistic insights from MD simulations and salt bridge analysis to the kinetic measurements obtained from SPR, we propose that the phosphorylation-induced stabilization of salt bridges, as evidenced by MD, supports a “conformational selection” mechanism. Such a mechanism likely facilitates the increased on-rate seen in SPR, as the phosphorylated protein preferentially adopts a conformation conducive to binding. Paradoxically, this same stabilization may lead to an induced fit that, while initially fostering a robust interaction, eventually results in a destabilized bound state, reflected by an increased off-rate. This illustrates a dynamic interplay where phosphorylation promotes rapid binding but also fosters a transient interaction, characteristic of the dynamic regulatory roles that phosphorylation often plays.

Importantly, while our study focused on LC3B, it is well established that unphosphorylated OPTN preferentially interacts with GABARAP family members²⁷, suggesting a potential affinity switch upon phosphorylation. Although this work does not directly investigate the interaction of phosphorylated OPTN with GABARAP paralogs, the question remains open and highly relevant. Given that residues surrounding the core LIR influence ATG8 binding specificity, future studies incorporating GABARAP, GABARAPL1, and GABARAPL2 would be valuable to determine whether phosphorylation reshapes the paralog preference of OPTN.

Furthermore, we acknowledge that the OPTN LIR overlaps with its FIP200-interacting region (FIR), as previously reported⁸⁸. This dual functionality suggests that mutations within this segment—such as D176, S177, and F178, which were studied here—may not only affect LC3B binding but also disrupt interactions with FIP200, a key player in autophagy initiation. These additional roles should be considered when interpreting the effects of disease-related mutations within this region.

Phosphorylation of OPTN is important to regulate its cellular function, and the structural changes it induces are integral to understanding the implications of disease-related mutations. The approach used in this study to investigate the effects of disease-related variants on the phosphorylated form of the protein lays the groundwork for future studies on other disease-relevant targets with characterized phosphorylation events.

Overall, our comprehensive investigation into the phospho-modulation and flanking region contributions of LIR motifs, using OPTN as a model system, provides valuable insights into the intricate interplay between phosphorylation, structural alterations, and functional implications in LIR-containing proteins. The elucidation of long-range effects, alterations in binding hotspots and kinetics, and the impact on solvent accessibility not only enhances our understanding of the regulatory role of phosphorylation in OPTN but also underscores its significance in the broader context of disease-related mutations. Our methodology, encompassing biophysical data, simulations, and innovative experimental techniques, establishes a robust framework applicable to a broad array of targets, contributing to a deeper comprehension of the cellular processes governed by phospho-regulated Short Linear Motifs.

Materials and methods

Modeling of constructs for MD simulations

We used the NMR structure of OPTN LIR (residues 169–185, NpSpSGpSpSEDpSFVEIRMAE) in its penta-phosphorylated form (p5) in complex with LC3B (residues 1–119, PDB entry 2LUE³⁴, first conformer) as a starting structure for four MD replicates of the LC3B_1–119-p5OPTN_168–185 complex. The structure includes an N-terminal expression tag (1-GAMG-4). Thus, we mutated it into the canonical sequence of LC3B from UniProtKB (1-MPSE-4) and added the missing C-terminal residue G120 (LC3B_1–120-p5OPTN_169–185). We then model its corresponding wild-type variant (LC3B_1–120-OPTN_169–185) by retro-mutating pS170, pS171, pS173, pS174, and pS177 to non-phosphorylated serines, using the mutagenesis tool of PyMOL (http://www.pymol.org/pymol). Furthermore, we modeled a single phosphorylated variant (LC3B_1–120-p1OPTN_169–185) in which we kept only pS177.

We also used AlphaFold-multimer³⁹ to generate a model of the full-length complex of LC3B and OPTN and retained the first five ranked models (out of 25 models) characterized by low predicted aligned error (PAE) scores. We used the AlphaFold database (version available 25/01/2023). We then retained the first ranked model to generate a trimmed version of the complex spanning the OPTN region 158–191 (LC3B_1–120-OPTN_158–191) and remodel the corresponding phosphorylated variants (LC3B_1–120-p1OPTN_158–191 and LC3B_1–120-p5OPTN_158–191) with ViennaPTM⁸⁹.

Molecular dynamics (MD) simulations

We performed all-atom MD simulations in explicit solvent for each variant and construct (Table S1) using GROMACS⁹⁰. We employed the CHARMM22* protein force field⁹¹, which we recently showed as a good force field for LC3B⁵². We used the TIPS3P water model⁹² and SP2 parameters for phosphorylated serine residues adapted from ref. ⁹³. We used periodic boundary conditions and a dodecahedral box with a minimum distance between the protein surface and the box edges of 15 Å. We capped the N- and C-terminal groups of both LC3B (NH₃+ and COOH, respectively) and the LIR peptides (NH₂ and COOH). We predicted the His27, His57, and His86 tautomeric states with H++, Propka software⁹⁴, accompanied by visual inspection. We used the Nε2-H tautomeric state for all the histidine residues in the simulation. We equilibrated the systems before the production run minimization, solvent equilibration, pressurization, and thermalization. We performed productive MD simulations at 298 K with the canonical ensemble using a time step of 2 fs and applying periodic boundary conditions, mimicking a concentration of 150 mM NaCl and including counterions to neutralize the system. We used the particle-mesh Ewald summation scheme^95,96 and we truncated Van der Waals and Coulomb interactions at 9 Å.

The MD simulations of LC3B_1–119-p5OPTN_169–185 were used to compare the MD results with available NMR data to evaluate the sampling quality achieved with the selected force field, since they included the same construct used for NMR. The simulations with LC3B_1–120-OPTN_158–191 and LC3B_1–120-OPTN_168–185 were used for the analyses to understand the mechanisms induced by phosphorylations or the impact of disease-related mutations.

During pre-processing of the trajectories, we verified that the minimal distance between each protein complex and its image was sufficient (i.e., higher than 15 Å) to rule out artifacts due to contacts between the protein and the corresponding image in the periodic boundary conditions. We carried out the simulations without embedding LC3B in a lipidic membrane, considering our previous study where we showed that the majority of the properties of interest were similar, comparing MD simulations of LC3B in solution or associated with the membrane⁵². The details on the constructs used for MD simulations are reported in Fig. 1 and Table S1.

Enhanced sampling simulations

We carried out parallel bias metadynamics simulations for the LC3B_1–120-p5OPTN_158–191 variants using as starting structures the same used for the unbiased MD replicate 1 and the salt concentration and temperature. The metadynamics simulations were carried out using GROMACS and PLUMED^90,97. We collected two different sets of simulations, including different collective variables (Table S1). In one case, we included two different collective variables describing the conformational changes of F178 of OPTN in the HP1 of LC3B. They included the χ₁ side chain dihedral of F178 and the distance between Cζ side chain atoms of OPTN F178 and LC3B F108 side chains. Moreover, we monitored during the simulation the distance between Cζ F178 and the Cδ atom of LC3B I23. In the other set of simulations, we included a parabeta collective variable to describe the intermolecular β-sheet between OPTN and LC3B. We used a bias factor of 4 (flipping mechanism) and 8 (β-sheet formation) and an initial sigma of 0.1 for all the collective variables. In all the simulations, we included an upper wall at 20 Å for the distance between the Cζ atom of F178 and the Cδ atom of LC3B I23 to avoid sampling unbinding events for the peptide. Moreover, we included a lower wall at 8 Å for the distance between the N- and C-terminal atoms of the peptide, to avoid the sampling of conformations that are too bent and would not be possible to explore in the context of the full-length protein. We carried out a block analysis for the estimation of the error associated with the calculated free energy for each simulation and each collective variable. We also calculated the differences in free energy between the minima as a function of the simulation time to evaluate convergence. In the simulations using parabeta as collective variables, we verified that the final free energy profiles were comparable, extending the simulation by an additional 100 ns.

Comparison of MD replicates with NMR data

We selected 1000 frames (equally spaced in time) for the replicate 1 of the LC3B_1–119-p5OPTN_169–185 to predict chemical shifts from the MD simulations and compare them to the corresponding experimental data (Biological Magnetic Resonance Bank (BMRB), entry 18518). To this goal, we applied the delta_CS Python-based pipeline⁵⁰ (https://github.com/ELELAB/delta_cs). This includes the calculation of backbone, β-carbon chemical shifts, along with chemical shifts from side-chain hydrogens for the MD ensemble with PPM⁵⁸ and the associated squared errors between the predicted and the corresponding experimentally measured chemical shifts. Furthermore, the pipeline includes CH3Shift⁹⁸ to calculate chemical shifts of side-chain methyl carbon atoms (i.e., 13C shifts). It then compares them to the experimental side-chain chemical shifts. We used a reduced χ² (χ²_red) metric as recently applied to another case study⁵¹. The square deviation between the calculated and experimental values was normalized by the variance of the chemical shift predictor for each type of chemical shift and the total number of chemical shifts. Therefore, low χ²_red values indicate a good agreement between experimental and predicted chemical shifts. The input, outputs and code for the analysis is also reported in the OSF repository associated with this publication⁶⁰ (refer to the folder “chemical_shift_analysis”). We also provided the time evolution of the χ²_red values for each atom type for each MD replicate in OSF.

Protein flexibility

We calculated the average root mean square fluctuation (RMSF) for each Cα atom in the central part of the LIR peptides (residues 170–184). We did the calculations, averaging the RMSF over 100-ns time windows as a probe of the LIR flexibility in the LC3B LIR-binding pocket and estimated the associated standard errors.

Solvent-accessible surface

We calculated the side-chain solvent-accessible surface area (SASA) for the side-chain atoms of the residues of the LIR core motifs that bind to HP1 (F178) or HP2 (I181) of LC3B. We estimated the SASA for the LC3B residues belonging to HP1 (D19, I23, P32, I34, K51, L53, and F108) and HP2 (F52, V54, P55, L63, I66, and I67)¹⁹ from the MD simulations of the different variants. We also included in the analysis the LDK tripeptide (L47, D48, and K49) and the ubiquitin patch of LC3B (R10, R11, K49, and K50), which are located in the proximity of the HP1 and contribute LIR-specific interactions^19,25. We used NACCESS⁹⁹ to calculate the relative SASA for each residue using frames extracted every 20 ps from the MD trajectories. In addition to the AlphaFold-based models (which feature an extended OPTN construct, residues 158–191), we carried out the analysis also on simulations based on the first conformer of the NMR structure of the LC3B–OPTN complex (PDB ID: 2LUE). Due to differences in construct length, the NMR-based simulations are treated as structurally distinct rather than direct equivalents. Representative trajectory visualizations from all systems are available in the OSF repository⁶⁰ under the folder “simulations_data”.

Secondary structure propensity of the LIR peptides

We employed the DSSP dictionary¹⁰⁰ to estimate the different classes of secondary structures for each residue of the OPTN peptides in the MD simulations. DSSP defines eight classes of secondary structure elements based on an estimate of patterns of hydrogen bonds: 3.10 helices, α-helices, π-helices, β-sheets, β-bridges, turns, bends, and coil. For each residue of interest, we calculated the occurrence of each structure class in the MD simulation.

Analysis of electrostatic interactions and hydrogen bonds

We used PyInteraph2¹⁰¹ to calculate electrostatic interactions in the form of hydrogen bonds and salt bridges. For salt bridges, all the distances between atom pairs belonging to charged moieties of two oppositely charged residues were calculated, and the charged moieties were considered as interacting if at least one pair of atoms was found at a distance shorter than 5 Å. We used the atoms forming the carboxylic group for aspartate and glutamate. The NH3 and the guanidinium groups were employed for lysine and arginine, respectively. A hydrogen bond was identified when the distance between the acceptor atom and the hydrogen atom was lower than 3.5 Å, and the donor-hydrogen-acceptor atom angle was greater than 120°. We calculated main-chain, main-chain side-chain, and side-chain side-chain hydrogen bonds. We customized the configuration files used by PyInteraph2 for the identification of salt bridges and hydrogen bonds by adding the oxygens of the phosphate group to the charged groups (for salt bridges) and to the hydrogen bond acceptors (for hydrogen bonds) to include the phosphorylated serines. We retained only intermolecular interactions present in at least 20% of the simulation frames (pcrit = 20%), as previously applied to other proteins^{102,103,104,105}.

Atomic-contact-based protein structure network (PSN)

We used the PyInteraph2 implementation¹⁰¹ of the PSN method based on atomic contacts as originally formulated by Vishveshwara’s group⁴⁷ and applied to the MD ensemble of the different constructs used in this study. We retained the pairs of residues whose sequence distance is higher than 1 for the edge calculations and at a distance lower than 4.5 Å. In the calculation, normalization factors for each residue are applied as previously discussed¹⁰⁶ to calculate the interaction strength. We used an I_min value of 3. For phospho-serines, we used the normalization factor reported in a previous paper¹⁰⁷. In addition, we retained edges with occurrences higher than 50% of the frames in the MD ensemble in the final PSN. The interaction strength associated with each edge in the final PSN corresponds to the average interaction strength for the edge over all the ensemble structures, and it is used to weight the PSN.

Contact analysis using CONAN

We analyzed intermolecular contacts using the CONtact ANalysis (CONAN) software⁶⁵, which enables statistical assessment of protein contacts and their temporal evolution in MD simulations. CONAN determines inter-residue contacts based on user-defined distance cutoffs. For our analysis, we set an rcut value of 10 Å, with rinter and rhigh-inter values of 4.5 Å, and used a timestep of one nanosecond. The output data from CONAN were further processed to quantify the fraction of contact formation over the simulation, calculated as the number of frames in which a contact was detected divided by the total number of frames in the trajectory. Residues with zero contact events were excluded from the analysis; all residues shown in the heatmaps registered at least one contact during the simulations, although contact frequencies below 1% may appear visually similar to zero due to the color scale.

Changes in binding free energy

We employed Foldx5⁴⁴ to perform in silico saturation mutagenesis and estimate changes in binding free energy upon mutations using the Python wrapper MutateX⁶⁸. We estimated the average difference in binding free energy (ΔΔGs) upon mutations using 25 representative structures from the MD simulations of LC3B_1–120-OPTN_158–191, LC3B_1–120-p1OPTN_158–191, and LC3B_1–120-p5OPTN_158–191. This strategy may help mitigate the limitation due to the backbone stiffness of FoldX^68,69. As a reference, we also performed similar calculations using as starting structure the 20 NMR-derived structures of LC3B_1–119-p5OPTN_168–185 (deposited in the PDB entry 2LUE; the input, outputs and code for the analysis of the nmr conformers is reported in the OSF repository associated with this publication⁶⁰—refer to the folder “nmr_mutatex_analysis”). We calculated the average ΔΔGs for each position, and we used a cutoff of 1 kcal/mol in ΔΔG to classify the effects of the mutations⁶⁸. In particular, we defined stabilizing or destabilizing variants featuring ΔΔG values lower than −1 or higher than 1 kcal/mol, respectively.

MAVISp analyses for disease-related variants of OPTN_158–191

We used the data deposited in the MAVISp database ⁷² where variants found in COSMIC¹⁰⁸, cBioPortal¹⁰⁹, and ClinVar¹¹⁰ have been retrieved for OPTN (Uniprot ID Q96CV9, and RefSeq ID NP_068815).

We retained only mutations in the region used for the modeling, i.e., 158–191 (i.e., V161M, E163D, L166V, N169S/T, E175Q, D176Y, F178V, E187Q, V179D, and A184D/S). We then applied the MAVISp framework⁷² for the assessment of the effects of these variants on: (i) structural stability, (ii) interactions with LC3B, (iii) homodimerization, (iv) interplay with post-translational modifications, and (v) long-range effects.

We used the model downloaded from the AlphaFold database¹¹¹ on the 23rd of March 2023 for the STABILITY module of MAVISp. We used the starting structure for the simulations of the non-phosphorylated variant (LC3B_1–120-OPTN_158–191), and 25 representative frames from the MD replicate 1 for the LOCAL INTERACTION module in the simple and ensemble mode, respectively. We also used a model generated with AlphaFold multimer and experimentally validated¹¹² to assess the effects of the mutations on the homodimerization of OPTN_1-577.

In addition, we retrieved (REVEL, AlphaMissense) or calculated (DeMask, EVE) the pathogenicity scores associated with the mutations using REVEL¹¹³, DeMask¹¹⁴, EVE⁶⁶, and AlphaMissense⁶⁷ with MAVISp workflows.

Structural clustering

We calculated pairwise RMSD values for each trajectory using the main-chain atoms of the complex. We performed clustering on the RMSD matrix using the GROMOS algorithm¹¹⁵ and an RMSD cutoff equal to the average RMSD from each MD replicate. We then retrieved the average structures for the first three populated clusters for the simulations of the unphosphorylated variant LC3B_1–120-OPTN_158–191. The average structure of each cluster corresponds to the structure belonging to the cluster, which features the lowest RMSD values to its neighbors in the cluster. These have been used as starting structures for the calculations in the ensemble mode with MAVISp and the LOCAL_INTERACTION module with the flexddg protocol of RosettaDDGPrediction (see above).

Correlogram

We collected correlation statistics on the MD replicates to investigate the relationship between selected pairwise distances between representative side-chain atoms of the residues involved in salt bridges. To perform this analysis, we utilized R script adapted from our previous work¹¹⁶. We measured linear dependencies using the Pearson correlation coefficient as a metric and we calculated the associated p values. For visual representation, we constructed the correlograms with a hierarchical clustering approach to rank the measurements based on the strength of their correlation and used a significance level corresponding to a p value of 0.01.

Protein purification

The construct encoding for GST-fused LC3B was expressed in E. coli BL21 (DE3) cells (Thermo Scientific, #EC0114), and the expression of the fusion proteins was induced by adding 0.5 mM IPTG at 37 °C for 4 h. Cells were lysed by sonication in lysis buffer (50 mM Tris-HCl pH 8, 250 mM NaCl, 1 mM DTT, cOmplete™, EDTA-free Protease Inhibitor Cocktail tablets, and PhoSTOP phosphatase inhibitor tablets). Pierce™ Glutathione Magnetic Agarose beads were used to precipitate the GST-fused protein for 2 h at room temperature under shaking. After washing the beads in lysis buffer, the protein was eluted in elution buffer (50 mM Tris-HCl pH 8, 10 mM reduced glutathione). The buffer was exchanged in assay buffer (10 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.1% Tween-20) using PD10 gravity columns (GE Healthcare). The protein was concentrated using Amicon Ultra-0.5 10K Centrifugal Filter Devices (Merck Millipore). The protein concentration was measured on a Nanodrop 1000 spectrophotometer at 280 nm using the extinction coefficient of Trp and Tyr residues in the GST-LC3B sequence (48,820 cm⁻¹ M^–1).

Peptides

N-terminally labeled FITC-Ahx OPTN LIR peptides were purchased from TAGCopenhagen in >95% purity. We have used the following peptide sequences corresponding to the 169–185 residues in full-length OPTN:

FITC-Ahx-NSSGSSEDSFVEIRMAE (WT),

FITC-Ahx-NSSGSSEDpSFVEIRMAE (pS177),

FITC-Ahx-NSSGSSEDDFVEIRMAE (S177D),

FITC-Ahx-NSSGSSEDEFVEIRMAE (S177E),

FITC-Ahx-NSSGSSEDSWVEIRMAE (F178W),

FITC-Ahx-NSSGSSEDpSWVEIRMAE (pS177 F178W),

FITC-Ahx-NSSGSSEDpSFVEIRAAE (M183A), and

FITC-Ahx-NSSGSSEDpSAVEIRMAE (F178A),

where “pS” is phosphorylated Ser.

Moreover, we tested longer peptides corresponding to the 160–185 residues in full-length OPTN:

FITC-Ahx-IVSELQLKLNSSGSSEDSFVEIRMAE (WT),

FITC-Ahx-IVSELQLKLNSSGSSEDpSFVEIRMAE (pS177),

where “pS” is phosphorylated Ser.

Sequences were confirmed by mass spectrometry and purity by RP-HPLC. The peptides were dissolved in an assay buffer and their concentration was measured on a Nanodrop 1000 spectrophotometer at 280 nm using the extinction coefficient of the FITC group (18,542 cm⁻¹ M⁻¹) and Trp residue where present.

Microfluidic diffusional sizing (MDS)

All MDS experiments were performed on a fluidity one-M instrument at room temperature using commercial chips (Fluidic Analytics, Cambridge, UK). FITC-Ahx-labeled OPTN_169–185 LIR peptides at a total concentration of 1 µM, together with a varying concentration of GST-LC3B, were added and diluted in assay buffer. The samples were incubated at room temperature for at least 2 h to reach equilibrium. The change in hydrodynamic radius with GST-LC3B concentration was fitted with an in-house-developed Python script using a one-site total binding model, to obtain K_d values. The equation used for fitting the data were:

$${{\rm{Y}}}=\, {{{\rm{Rh}}}}\_{{{\rm{free}}}}+\left({{{\rm{Rh}}}}\_{{{{\rm{compl}}}{{\rm{ex}}}-{{\rm{Rh}}}}\_{{{\rm{free}}}}}\right)* \left(\left({{\rm{X}}}+{{\rm{n}}}* {{\rm{L}}}+{{{\rm{K}}}}_{{{\rm{d}}}}\right)-\right. \\ \left.({({{{\rm{X}}}+{{\rm{n}}}* {{\rm{L}}}+{{{\rm{K}}}}_{{{\rm{d}}}}})^{2-4}* {{\rm{n}}}* {{\rm{L}}}* {{\rm{X}}}})^{0.5}\right)/(2* {{\rm{n}}}* {{\rm{L}}})$$

where:

Rh_free = Hydrodynamic radius of the labeled peptide (nm)

Rh_complex = Hydrodynamic radius of the complex between the labeled peptide and unlabeled GST-LC3B (nm)

n = binding sites occupied on the labeled species at saturation (i.e., stoichiometry)

L = the labeled peptide concentration (nM)

K_d = the equilibrium dissociation constant or affinity of the complex (nM)

Each MDS experiment was performed in 3–5 biological replicates, each with 1–3 technical replicates. The differences in K_d, Rh_free, or Rh_complex between the tested peptide variants and the pS177 or WT variant were assessed by an unpaired, parametric t-test. Where the tested variant had a variance of more than fourfold of the pS177 or WT variant, we used the t-test with Welch’s correction.

Surface plasmon resonance (SPR)

SPR experiments were carried out on a Biacore X100 instrument (Cytiva). The instrument was primed with running buffer (HBS-EP+ 1x buffer, Cytiva). Following standard procedures for amine coupling, 1 μM of GST-LC3B in 10 mM sodium acetate, pH 5, was used for coupling to the carboxymethylated dextran matrix of a CM5 sensor chip surface (FC2). A reference surface was treated identically, but not subjected to GST-LC3B for immobilization (FC1). These SPR chips and HBS-EP+ running buffer were used for all experiments with GST-LC3B described in this paper, and all data collected was referenced to FC1. OPTN_169–185 peptide variants were dissolved in running buffer at concentrations between 600 nM and 300 μM. Various concentrations of all OPTN peptide constructs were injected at 30 μl/min and 25 °C, with an association time of 60 s and a dissociation time of 60 s. Resonance units (RU) were recorded as a function of time. The BIA evaluation software package was used for data analysis using a simple 1:1 model.

Statistics and reproducibility

Statistical analyses were conducted on data obtained from independent molecular dynamics simulations, enhanced sampling simulations, and biophysical experiments.

For molecular dynamics (MD) simulations, n = 3 independent simulations were performed per system, each consisting of 100,000 frames. Structural parameters, such as solvent-accessible surface area (SASA), were calculated per frame or per time window in the case of the root mean square fluctuation (RMSF).

Enhanced sampling simulations using parallel bias metadynamics were carried out for selected systems. While not repeated as independent replicates, these simulations were internally assessed for reproducibility. Block analysis was used to estimate the uncertainty in the calculated free energy, and convergence was evaluated by tracking the differences in free energy between minima over simulation time. For simulations using parabeta collective variables, the free energy profiles were validated by extending the simulations by an additional 100 ns to confirm stability.

Microfluidic diffusional sizing (MDS) experiments were performed in n = 3–5 biologically independent replicates, each including 1–3 technical replicates. Binding affinities were determined by fitting the change in hydrodynamic radius (Rh) as a function of ligand concentration. Differences between peptide variants and reference variants (pS177 or WT) were evaluated using unpaired, parametric t-tests, with Welch’s correction applied when variance differed by more than fourfold.

Surface plasmon resonance (SPR) experiments were performed once per variant, and dissociation constants and kinetic parameters were derived from nonlinear regression fits of the binding curves. Averages and standard errors reflect fitting uncertainty and do not represent replicate-based statistics.

No cellular or animal experiments were performed in this study.

Reporting summary

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