The Insulin Receptor-like (IRL) family of Tyrosine Kinase receptors (TKRs) governs some of the key hormonal signalling with pleiotropic control of metabolism, growth and life span (1–3). All these receptors share very similar (αβ)₂ subunits, domains and general 3-D structures organisation, comprising of the ectodomain (ECD), transmembrane (TM) and juxta-membrane (JM) regions, leading to tyrosine kinase (TK) which ends with C-terminal tails ( Figure 1A ) (for reviews see (4, 5)). In humans, IRLs include two isoforms of Insulin Receptor (hIR-A, hIR-B (with additional 12 amino acids insert at the ends of the α-subunits)) (6) and Insulin-Like Growth Factor Receptor 1 (IGF-1R). Recent structure of insect Drosophila melanogaster IR (dmIR) ECD in complex with DILP5 - one of this insect’s seven insulin-like hormones - revealed its remarkable similar blueprint to human homologous complexes ( Figure 1B , PDB ID 8CLS) (7). These findings suggest that a general conservation of the IR-like template can be expected in other invertebrates as well, underpinning the molecular and structural bases of a remarkable conservation of the insulin signalling axis in the animal kingdom.

The last ~12 years brought a long-awaited first insight into the structures of human IRs and IGF-1R, firstly crystallographic (8), and, subsequently, by cryo-EM microscopy (9), followed by many of their apo- and holo-forms in complexes with endogenous hormones (insulin, IGF-1, IGF-2), as well as viral insulins, de-novo engineered functionally-active (insulin agonists/antagonists) peptides and monoclonal antibodies (reviewed (4, 5). They revealed a wide spectrum of hormone:receptor binding modes, stoichiometries of these complexes ( Figure 1B ), and the adaptability/dynamic character of their ECDs ( Figure 1B ).

The IR/IGF-1R structures point towards the likely key steps of the apo-holo transitions of these receptors, such as: (i) a large (~110 Å) separation of the receptors TKs in the inverted V: ᐱ-like form in hormone-free apo-IRLs that maintains their inactive states, (ii) hormone(s) binding to the receptors’ ECDs triggers their large structural rearrangements, closing the gap between the FnIII-1/2/3 domains of the ‘legs’ of the receptors, resulting in the proximity of the TKs and their activity-releasing trans cross-phosphorylation.

In one-insulin:hIR (and one-IGF-1:IGF-1R) complexes the hormones are bound to the upper arm of the hIR in T-like conformation (so-called high affinity site 1), while the other arm remains close to the stem of the receptor formed by FnIII-2/FnIII-3 domains ( Figure 1B – e.g. PDB ID 6HN5/4) (e.g (10–14). Such hormone binding mode is sufficient to bring together (~30-10 Å) the membrane-proximal FnIII-3/3’ domains, hence, possibly, translating these close contacts onto an ‘active’ engagements of the TMs, JM and TKs part of the IRs ((‘) denotes corresponding sites in the other subunit)). On the opposite range of insulin:hIR complexes, four insulins saturate the almost symmetrical T-like conformer of the receptor by binding to both sites 1/1’, and to so called low affinity sites 2/2’ that are located on the FnIII-1 (and parts of the FnIII-2 in dmIR) domains ( Figure 1B – 6PXV) (15) (16). Structures of many other conformers of the complexes have been determined with two, three and four insulins in one receptor; they usually appear in often subtle variations of asymmetrical T and T -like conformations, with different occupancies of 1/1’ and 2/2’ sites, and, in some cases, with insulin ‘bridging’ the lower arm and the stem of the IR, not being fully engaged with either site 1 or site 2’ ( Figure 1B – 7PG0) (17). Overall, site-1/site-2 binding modes correspond to a well-established two different receptor-binding epitopes on hormones’ surface (18). Interestingly, hIGF-1:IGF-1R complex has only been obtained with one IGF-1 bound, with the α−CT helix threading through the IGF-1 C-domain-forming loop that joins in this hormone equivalents of insulin’s B- and A-chains (11). The first invertebrate dmIR-ECD structure in complex with three insect DILP5 insulins also falls – in general – into the landscape of hIR/IGF-1R conformations, despite that two DILP5 hormones are sitting on the lower arm in close contacts neighbouring site-1 and site-2’, while the third DILP5 occupies the classical site-1’ on the T-like upper arm of the receptor ( Figure 1B – PDB ID 8CLS) (7).

The intracellular TK components of the IRLs have also been studied extensively, resulting in many 3-D structures of their both inactive and active conformations (19–25), allowing outlining a general catalytic and activation mechanisms of these kinases (26–29). The TKs of the ILR receptors are very homologues, following also a very similar 3-D structural blueprint, with a typical protein-kinase fold, which is also greatly maintained in monomeric classes of the TK receptors (e.g., EGFR, Ron, c-Kit etc.) which dimerise upon hormone binding (30) ( Figure 2 ). Both human hIR and hIGF-1R TKs have typical two domains - N- and C-terminal lobes – organisation (19, 22), with the C-lobe containing most of the active site residues and so-called activation loop. The activation of the ILR’s kinase involves trans-phosphorylation of three tyrosine side chains of the activation loop, and its flipping over, revealing the active, ATP- and substrate binding sites which are occluded by this loop in the kinase inactive state (28) ( Figure 2 ). TK activation relies also on a significant closure of the N- and C-lobes, that results – among other structural changes – in the shift of so-called αC helix that places then catalytically important residues close to the ATP binding site. ( Figure 2 ). Subsequently, three tyrosines in the C-terminal tails of the TKs, and two in the JM region can be phosphorylated for the recruitment of the key downstream signalling effectors (e.g., Insulin Receptor Substrate-1/2, Shc, SH2B1, SH2B2), or for the internalisation of the receptor (31–33). It is also envisaged – mostly via biochemical evidences - that the kinase-proximal part of the JM region (still very elusive in structural studies) plays a TK auto-inhibitory role, maintaining ILR basal, quiescent state (34–37); this was also confirmed in some apo-monomeric type receptors’ TKs, such as in the EGFR (38–50). Structural studies of this mechanism in the hIR showed that JM Tyr972 – conserved in all ILRs in the animal kingdom – interacts in the TK apo state with also conserved residues of kinase’s N-lobe, stabilising the αC helix in a catalytically non-productive conformation (36) ( Figure 2 ).

Insect Drosophila dmIR has a unique additional ~368 amino acids extension of each TK (with some homology to the IRS-1), which can be processed (cleaved off) in certain tissues (51)( Figure 1A ); its role is still not fully known.

The exact mechanism of activation of the whole hIR/IGF-1R, and dmIR upon hormone binding has not been clearly visualised, but structural and functional data point towards the trans-phosphorylation of their TKs being brought together upon transition from apo-to-holo states of the receptors. It seems that the TKs trans-phosphorylation of ILRs follows, to some extent, a symmetrical pattern, which is in contrast to some monomeric TK-receptors, e.g., EGFR, in which ligand-induced dimer only one kinase is active upon cis-phosphorylation, being allosterically stabilised by TK from the associated receptor subunit (aka CDK: CD-like functioning) (52).

Despite recent astonishing progress and breakthroughs in the structural/functional characterisation of ILRs, the insight into the actual structural path of signal transduction through these receptors is still rather binary, with separate focuses on the roles of ECDs and TKs in this process. This extra-/intra-cellular division of our understanding of the IRL receptors results mainly from current methodological constraints. The studies of ECD-to-TK signal transduction are hampered by the occlusion of the transmembrane, JM and TK regions by lipids necessary for the extraction and stabilisation of the whole ILRs, as well as due to an inherent dynamic character of these downstream components of the receptors (16, 53).

Such methodological challenges lead to alternative views about the nature of the ‘native’, transmembrane signal-inducing ILR ECD conformer (e.g (17, 54). However, it can be envisaged, that considering the low pM concentrations of circulating insulin (going potentially towards nM range after meals), one-insulin:IR asymmetric complex should suffice to activate the whole receptor. It may emerge in this state via a transient low-affinity site 2 complex which initiate destabilisation the apo - ᐱ-like (or dynamically similar) – IR conformer (5). However, this single-hormone bound receptor model supports well IGF-1:IGF-1R as the main signalling conformer, but it does not fully explain the overall sub-nM (~0.02 nM) high hIR affinity of insulin, not covering quite the interactions of low affinity (~0.2 nM – 2μM (hIR construct dependent)) sites 2 on insulin surface and the hIR (17, 55) Structures of two-insulins:IR complexes can correspond to some transient states towards the final single-molecule assembly; although three/four-insulins:IR complexes are rather unlikely prevailing native forms of the hIR due to a limited physiological presence of the hormone, they cannot be excluded.

Therefore, prompted by the still existing structural gap between ILRs’ ECDs and their TKs we attempted to probe whether the use of AlphaFold2 (AF2) (56) can provide more structural and functional insights into bridging the knowledge about the ECDs and TKs of these receptors, by shedding some light on possible 3-D arrangements of the TM, JM and TKs modules, in the context of different, known structures of Insulin:IR (and IGF-1:IGF-R) complexes. NMR structure of IR TM helices in lipid micelle is available (57), and the longest known structure of the JM segment was described in the IGF-1R apo-TK mutant (referred here as to IGFRTK-0P, PDB-ID: 1P4O) which starts at Pro948 (Pro958 in IR-A isoform of which numbering is used through this text) (58), while the whole hIGF-1R JM spans ~Asn934-Ile969 (app. ~Pro945-Ile984 in hIR-A, including first 12 linker regions in both receptors). However, the JM chain is stabilised in IGFRTK-0P by the crystallographic dimer interface between the kinase molecules. Similarly, the longest structurally-visible JM in the hIR – from Ser962 onwards - has been observed in the crystal structure of the holo-dimer of triple-phosphorylated TK (IRTK-3P, PDB ID: 4XLV) (24). Hence, the known conformations of some parts of the JMs of IL receptor can be biased by crystal packing-related inter-molecular contacts.

Therefore we undertook a simple exploratory modelling exercise where we employed the AlphaFold2 (56) to predict the structures of FnIII-3-TM-JM-TK parts of the hIR, hIGF-1R and dmIR, superposing subsequently the AF2-derived FnIII-3 domains of the FnIII-3-TM-JM-TK segment on their counterparts in the experimentally determined representative ECD conformers of these receptors ( Figure 3 – see details in the following chapter). We expected that the relevance of these FnIII-3-TM-JM-TK AF2-derived models can be enhanced by excellent, known 3-D templates for the flanking FnIII-3 and TK domains, for which AF2-generated folds obtained here should be very accurate, and ‘native’-like. Although AF2 predictions discussed here lacks 100% reproducibility, i.e., their repetitions always result in an overall slightly different relative spatial distribution of FnIII-3-TM-JM-TK domains, the key motifs of these models (e.g., JM engagement with the TK discussed below) are still present, and are – overall – consistently repetitive.

The key questions we tried to address here were as follows: (a) whether the AF2 can help to correlate the orientations of the FnIII-3/3’ domains observed in the structures of different IR and IGF-1R complexes with the orientations of the downstream TM-JM-TK parts of the receptors, (b) whether the AF2-driven FnIII-3-TM-JM-TK modelling, and the fusion of these models with the experimental receptors’ ECDs structures, is capable of revealing any new - potentially in vivo significant - types of interactions in the TM-JM-TK regions, especially of the ‘elusive’ JM domains, and (c) whether the full length IR and IGF-1R conformers observed in combined true-ECD-structures+AF2-models can shed any light on the receptor activation process, i.e., signal transduction through these receptors, which still awaits unambiguous, experimental structural elucidation.

The models of the FnIII-3-TM-JM-TK segments of hIR, hIGF-1R and dmIR were obtained by the AF2 programme in the CCP4 Cloud (59) for the following IRL amino acid sequences: hIR - Val801-Ser1343, hIGF-1R – Val791-Cys1337, dmIR – Tyr1191-Gln1828. The TK-following dmIR-unique C-terminal Domain (CD domain) ( Figure 1A ) was not included here as (a) it was clear after initial trials that the AF2 was not able to predict its relevant structure (as expected: no known-template for CD sequence in the databases), and (b) its occurrence is limited exclusively to dmIR. The flow of the whole modelling process in outlined schematically in Figure 3 . Firstly, the representative ECD structures of the ILR-family, with different hormone:receptor stoichiometries and variety of conformers (from one arm up/one fully down to almost symmetrical T-state ones) ( Figure 1B ) were selected ( Figure 3 – stage A. Next, the abovementioned sequences of the receptors’ FnIII-3-TM-JM-TK regions were submitted for the prediction of their structures to the AF2 programme ( Figure 3 – stage B). AF2 produced several models for each FnIII-3-TM-JM-TK sequence, and top five for each receptor (m1-m5) were used for further analysis ( Figure 3 – stage C). Subsequently, the FnIII-3 domains of the AF2 models m1-m5 were superposed on their corresponding FnIII-3 and FnIII-3’ equivalents in each representative receptor ECD in programme COOT LSQ option, on their Cα atoms (60) ( Figure 3 – stage D). The Cα target atom ranges for these superpositions was Asp813-Val905 for the hIR, except for 6HN5/6HN4 structure (10) with a different numbering where models Ala811-Thr906 were superposed on 6HN4/5 Ala792-Thr887, while for the dmIR this FnIII-3 Cα target atom range span Asp1209-Val1302 for both models and the dmIR. For the mIGF-1R:hIGF-1 structure (PDB ID 6PYH) (14) the Cα range for the FnIII-3 domains was Gly800-Phe893 that were superposed with models Gly801-Phe892; this was due to differences between the structure and model sequences here (6PYH is a highly homologous mice IGF-1R (mIGF-1R) (96% seq. identity with the hIR) complexed with hIGF-1).

Subsequently, after these superpositions, the redundant/duplicated now pairs of models’ FnIII-3/FnIII-3’ domains were deleted in COOT ( Figure 3 – stage D), yielding a ILR’s full-length ‘hybrid’ model, comprising of an experimentally-known ECD structure of a particular receptor that was followed by the AF2-modelled TM-JM-TK ( Figure 3 – stage E). Finally, these full-length IRLs models were manually placed (translate/rotate option in COOT) in a phosphocholine lipids bilayer: one of the typical models of human cell membrane - derived from the NMR structure of the hEGFR TM helices determined in these lipids (PDB ID 2M20). Lipids DMPC-to-DHPC ratio of in the 2M20 was 1:0.25, where DMPC is 1,2-ditetradecanoyl-sn-glycero-3-phosphocholine, and DHPC - 1,2-dihexanoyl-sn-glycero-3-phosphocholine (61). Finally, some lipid molecules were deleted in COOT to accommodate TM segments. Analysis of the models have been performed in COOT, and the best models, i.e., without steric clashes, were finally analysed, and presented here in figures made by the CCP4MG programme (62).

It should be noted that the conformation/proximity of the Fn-III-3 domains, and, in general, overall conformations of the ILRs’ stems, hence the structure of the full receptors’ models presented here, could be affected by different tagging of the C-termini used for the expression and purification of a particular receptor construct. Here, the C-termini of the FnIII-3 domains were as follows: PDB ID 6HN5/6HN4 – hIRA ECD (1–916) ended with 33-residue GCN4 zipper sequence (RMKQLEDKVEELLSKNYHLENEVARLKKLVGER), and with Y144H, I421T, and Q465K mutation; Fv83–7 monoclonal antibodies were used for complex stabilization (10); 7PG4 – was a full-length hIRA with human protein C-tag (17); 7PG0 – the same as for 7PG4; 6PXV - full-length hIRA with endocytosis-preventing mutations Y960F, S962A, D1120N, R1333A, I1334A, L1335A, L1337A, untagged (Tsi3 and His8 tags were removed during purification) (16); 6PYH – Mus musculus IGF-1R (mIGF-1R), full-length with C-terminal truncation (1–1262), and Y951A, D1107N kinase internalization-preventing mutations (Tsi3 tags removed upon purification) (14); 8CLS – dmIR ECD (264-1309), C-terminal StrepII-tag (7). Nevertheless, despite the C-terminal differences in these constructs, their consistently recurring structural convergences, i.e., similar proximities of their FnIII-3 domains in their full-length, or ECD, liganded states, allow to assume a considerable physiological meaning of these conformations.

Overall, the LSQs-derived rms for the corresponding structures with known-models of FnIII-3 and TK, were very low - in the range of ~0.8-0.9A - as the AF2 predictions of these domains were very accurate due to their many known 3-D templates/structures. The dmIR TK was also predicted with an excellent accuracy due to its very high homology with hIR and hIGF-1R TKs (~60% identity), and its similar AF-based model has already been analysed (25). The trans-membrane sequence peculiarity of the TM helices, together with some known NMR structures (57, 61), resulted in their typical helical models, as expected.

The AF2 predicted local difference test (pLDDT) scores (0-100, score >80 shows high confidence of the prediction) were low, usually below 50, for the TM, JM and TKs’ active loops regions, remaining well >80 for the remaining parts of the proteins. However, a fully truthful conformations of these ILRs’ regions was not really expected, but rather their possible 3-D arrangements in the context of well known, hence reliably predicted, flanking FnIII-3 and TK domains.

The most anticipated predictions here were the putative folds of the JM regions, especially in the context of their neighbouring TM and TK domains, as their available 3-D definitions are very limited. As mentioned above, the JM hIR region is structurally best defined in hIRTK-3P crystal structure (PDB ID 4XLV) (24) – from Ser962 onwards (IR-A notation), but it points away from the TK’s core, being trapped in crystal-symmetry JM-JM contacts from the other TK, and is also wedged there between the β-sheet and alpha-helix (αC) surface of the second TK molecule (residues 944–961 are disordered/invisible there). This JM conformation is also stabilised there by the interaction of its N-terminal part with the third TK crystal-symmetry related molecule. The JM region from Val966 is also visible in the crystal structure of the inactive hIR TK (PDB ID 1P14) (36), where the Val966-Glu976 JM chain – in contrast to hIR TK-3P (PDB ID 4XLV) structure (24) - runs close to the core of the TK. This work also provides an evidence for the auto-inhibitory role of Tyr972 which interacts there with several conserved residues of the TK’s N-terminal lobe, stabilising a catalytically nonproductive position of the αC helix. The best 3-D evidence for the conformation of the hIGF-1R Pro948-Tyr957 JM segment is in the apo-form of this TK, but it is an important part of the TK-dimer inter-molecular interface (58). It’s worth noting that the JM segment is shorter in the IGF-1R TK, missing the equivalent of Val966-Cys969 peptide present in the IR-A were unpaired Cys969 is frequently mutated in structural studies to avoid protein non-specific aggregating.

Here, the top five FnIII-3-TM-JM-TK models for each AF folding exercise were analysed for packing against the lipid cell membrane model, and potential clashes with kinase counterparts from another subunit of the receptor. Models with the TKs domains too much embedded into the membrane due to a very ‘parallel’ run of their TMs along the lipid bilayer, or where TKs were significantly overlapping, were discarded.

Interestingly, the activation loop was predicted in both active and inactive conformation for the dmIR TK, while it was always in the active conformation in hIGF-1R TK models, and an inactive one in the hIR models.

Also, our attempts to predict the FnIII-3-TM-JM-TK dimers in the next generation AF3 programme (63) were unsuccessful, not yielding any sensible structures. Even more surprisingly, similar AF3 attempts to generate TK dimers (i.e., TKs without any other domains) also failed, despite many excellent structural templates.

The DFA model outlined here is a very simple attempt to unify structural and biochemical evidences reported for ILRs ectodomains and their intracellular components. Its general character is striving to maintain a holistic approach to the knowledge about the components of the receptors, combining all their elements needed for an effective signal transduction. It cannot/does not pretend to be a final view of signal transduction through the ILRs, but it aims to provoke alternative thinking about this process, highlighting some of its aspects that – so far - eluded unambiguous experimental approaches.

Firstly, the DFA model underlines the importance of the already well postulated JMs domains inhibitory effect on the TKs (35–37), especially extensively investigated and evident in EGFR and other classes of the TK receptors (e.g., the EGFR’s JM is even partially helical being in very close interactions with the cell membrane) (47, 48), (84–86) (87). However, the DAF model expands the JMs’ auto-inhibitory role onto their possible stabilisation of the activated forms of the TKs’ active loops, thus giving the JMs’ dual and opposite functionalities in the activation process of the receptors.

Secondly, the DFA model highlights the likelihood of a very intimate character of the JM/TK interaction with cell membrane: possibly an important feature of receptors’ activation process that relies on the JM/TK:membrane-inner-leaf ‘friction’, hence taming the intracellular movement of the TKs in response to hormone binding in order to allow unzipping of the JMs from their in-TK-groove auto-inhibitory conformation. Different qualitative roles of the protein:protein and protein:lipid interactions for the EGFR-like receptors have already been highlighted (88).

Surprisingly, the DFA model translate molecularly very sophisticated, specific (atomic – ‘quantum-like”) receptors’ ECD:hormone mechano-sensing interactions into activation of their TKs, and the initiation of a chemical signalling, by a very inter-molecularly “crude” (“Newtonian-like”) drag-and-friction forces. However, the distinctive properties of the hormone/analogue, and its interactions with the ILRs-ECDs sites 1/2, remain invariably crucial for the nature of the functionality of the hormone, and its downstream signalling signatures as they dictate the dynamics of TM/JM/TK movements. Interestingly, the emphasis of a possible importance of ‘simple’ mechanical forces as a part of ILRs activation parallels a growing recognition of such interactions, mechanical stress concentrations, and resulting feedback loops, being crucial for such diverse life’s phenomena like formation of embryo and organogenesis, to cancer (89, 90) and geometry of rose’s petals (91).

Finally, the DFA model emphasise the emerging role of the heterogeneity of cell membrane inner/outer leaves, their different chemical composition and structural micro-/macro-diversity in modulation of activation of Insulin-Like Receptors. It also underlines the necessity of more frontal – although methodologically most challenging – attack on the phenomena that occur on outer/inner membrane leaf:receptor interfaces, without understanding of which the insight into the signal transduction through ILRs may remain incomplete.