CALR mutant C-terminus opens up the wild-type lectin domain
Full length CALR has three domains (Figure 1A). The N-terminal domain (residues 18-197) interacts with immature N-glycans, the P-domain (residues 198-308) is associated with the chaperone functions and the C-domain is involved in calcium buffering and contains the KDEL ER retention motif. In agreement with crystallographic structure of the N-domain of human CALR 14 and NMR structure of rat P-domain 15, the AlphaFold prediction of full length calreticulin 16 reveals a compact N-domain rich in b-sheet and a central P-domain forming a long hairpin structure comprising several anti-parallel b-sheet segments. Although it has not been solved experimentally, the structure of the C-terminus is predicted to be essentially a-helical with the exception of the last 50 residues which appear to be unstructured (Figure 1A).
The CALR mutants involved in MPNs were previously reported to interact with TpoR through generic interaction between CALR lectin or N-domain and immature N-glycans on TpoR 12,13. However, this interaction is normally only transient 3 and cannot explain the stable binding observed between TpoR and CALR mutants.
To understand how CALR mutants, represented by CALR del52, acquired the ability to stably interact with TpoR, we dissected the structural changes in full length CALR del52, which result from a frameshift that replaces the C-terminal 45 amino acids with a sequence that is rich in methionine and basic amino acids, but lacks the KDEL retention signal (Figure 1B).
To assess how the frameshift mutation in del52 influences the secondary structure and overall packing of CALR, we undertook Fourier transform infrared spectroscopy (FTIR) and hydrogen-deuterium exchange mass spectrometry (HDx-MS). We produced recombinant full length human CALR WT together with CALR del52 and different variants thereof as described in Figure 1A and Supplementary Figure 1A-B. FTIR spectroscopy reports on protein secondary structure and an analysis of the amide I vibration revealed that the overall structure of the protein was similar to that expected from the structural data of isolated domains or truncated proteins. The protein has high ß-sheet and random coil content, which is likely due to the N- and P-domains, respectively. The a-helical content was higher in CALR wild-type (13.3%) compared CALR del52 (8.4%). To address the origin of the a-helical content, we deleted the last 45 residues (as in CALRc-tail, leaving only the common regions between CALR WT and CALR del52) (Figure 1B). Strikingly this deletion resulted in an almost complete loss of -helical content, demonstrating that helicity is concentrated in the C-terminus of both wild-type and mutant CALR (Figure 1C-D). To define more precisely the region of the mutant tail that is -helical, we truncated the last segment of CALR del52 tail (labelled as CALR del52 A394*) (Figure 1B). The mutant tail of CALR mutant can indeed be separated in two segments based on amino acid composition. The first region is rich in hydrophobic (Met) and basic (Arg, Lys) residues (Figure 1B, purple) while the second region starting at A394 (Figure 1B, red) has a more heterogenous amino acid composition. Analysis of FTIR spectra showed that CALR del52 A394* harbored increased relative -helix content, demonstrating that helicity was concentrated in the first segment of the mutant tail rich in Arg and Met. This was further confirmed by in silico prediction using AlphaFold 2.0 16 (Supplementary Figure 1C).
Next, we sought to assess whether introduction of two key point mutations (Y109F/D135L) in CALR del52 N-domain led to changes in its secondary structure. These mutations are in the N-domain (Figure 1B) and abolish the binding of CALR to the immature N-glycans on TpoR 12,17, Strikingly, this mutant showed a clear decrease in C-terminal helicity and of random coil percentages with corresponding increases in turns and ß-sheets (Figure 1C-D). Because ß-sheets are concentrated in the N-domain 18 (Figure 1B) and -helix present only in the C-terminus, these changes suggest conformational coupling between the N-domain and mutant tail of CALR del52.
To further dissect changes induced by the tail of CALR mutant, we acquired structural footprints of CALR WT and mutants using HDx-MS. With this technique, a mass shift in peptides from a protein after deuterium (D) incorporation in backbone amide positions provides a readout of residue accessibility and the overall conformation of the protein. We acquired the first complete HDx-MS structural footprint of CALR WT which was in line with a compact N-domain rich in ß-sheets 14 and an elongated P-domain that is highly accessible 15 (Supplementary Figure 2A-B). The last segment of the P-domain (from residue 281) was less accessible, in line with in silico prediction 19. The accessibility of the C-domain was consistent with an -helical secondary structure as shown by our FTIR data. However, lower accessibility was observed in the KDEL ER retention signal than expected from current models (Supplementary Figure 2A-B).
We then compared the structural footprint of CALR WT with that of CALR del52 and CALR c-tail. Strikingly, the removal of the last 50 amino acids of CALR WT
did not significantly affect the conformation of the rest of the protein. Only a small fragment of the N-domain, represented by residues 52SSKGKFYGDEEKDKGLQTSQDARF74 in contact with the C-domain based on AlphaFold prediction (Figure 1E) was less protected in CALR c-tail compared to CALR WT, suggesting that the C-terminus closes up the structure, but that it involves a small region of the N-terminus. This finding also suggests that mutations that involve deletion of the C-terminus without frameshift lead to enhanced accessibility of the N-terminus, making it a potential therapeutic target. Such mutations are frequent in a variety of solid-tumor cancer and are associated with immunosuppressive activity 20. In sharp contrast, CALR del52 was globally more amenable to deuterium exchange than CALR WT, demonstrating a more flexible conformation (Figure 1F). The N-domain was mostly affected by the addition of the novel mutant tail, suggesting again that the mutant tail directly interacts with and destabilizes the N-domain (Figure 1F and Supplementary Figure 2C), in line with our FTIR results. Remarkably, key residues involved in interactions with immature glycans such as C105 and W319 12,17 were more accessible in CALR del52 compared to CALR WT (Figure 1F and Supplementary Figure 2C). This finding that the lectin binding residues are more available in CALR mutant provided a first explanation for the more stable interaction between CALR mutant and immature glycans on TpoR. Identifications of such key regions that are more accessible in CALR mutant than in CALR WT also unveils potential sites for specific therapeutic targeting.
CALR mutant tail directly interacts with TpoR extracellular domain
The results above indicated that the strong interaction between CALR mutant and TpoR could be at least partially explained by increased accessibility of key residues in the CALR
N-domain able to bind immature glycans. Yet, this did not solve the specificity of CALR del52 for TpoR versus thousands of other N-glycosylated proteins.
To investigate the basis for this specificity, we sought to assess whether other regions of CALR mutants directly interacted with TpoR independently of the presence of immature glycans on TpoR. Therefore, we used recombinant extracellular domain (ECD) of TpoR, labelled as D1D4, that we showed contains mature glycans 12. In a mature glycosylated form, TpoR ECD can only bind via other residues than those involved in binding of immature TpoR to CALR. The two proteins were incubated at 1:1 molar ratio before HDx-MS analysis. Analysis revealed significant (p<0.001) hydrogen-deuterium exchange differential in multiple peptides containing the mutant tail (Figure 2A and Supplementary Figure 3A-C), demonstrating a direct interaction with the mature TpoR extracellular domain. This differential exchange was not observed in the very last residues of the mutant tail encompassing residues 405QWGTEA411 (Supplementary Figure 3A-C), showing that this last segment is not involved in the interaction. Different fragments of the N-domain predicted to be conformationally close to the C-domain, also showed significant, albeit smaller, increase in accessibility in presence of mature TpoR ECD (Supplementary Figure 3B-C). This observation suggests that binding of the mutant tail to the receptor leads to concomitant loss of contact between the new tail and the N-domain of CALR del52, thus further opening up the mutant CALR.
To assess whether interaction between CALR mutant tail and TpoR could be observed in living cells, we created N-terminal truncations of CALR del52 (Figure 2B) lacking the N-domain and thus unable to interact with immature glycans on TpoR. We used Bioluminescence Energy Transfer (BRET) to measure direct interaction in living cells between TpoR N-terminally fused with a NanoLuciferase (NL) and N-terminal truncations of CALR del52 fused to a HaloTag at their C-terminus as we did previously for full length CALR del52 12. In line with our structural findings, complete deletion of the N-domain (denoted CALR del52 P-C domain) or both the N and P-domain of CALR del52 (CALR del52 C-domain) still allowed strong interaction with the receptor in living cells (Figure 2C and Supplementary Figure 3D). This was further validated by co-immunoprecipitation of FLAG-tagged N-terminal truncations of CALR del52 and HA-tagged TpoR (Figure 2D). Thus, CALR del52 interacts with TpoR both through its lectin domain and its mutant tail. Finally, we used a STAT5-dependent transcriptional luciferase assay to assess whether CALR del52 could induce activation of TpoR independently of the lectin binding. Unexpectedly, we found that the P-C domain alone of CALR del52 was able to induce a small but significant STAT5 transcriptional activity via TpoR, but not the erythropoietin receptor (EpoR) (Figure 2E and Supplementary Figure 3E). Together these results demonstrated that direct binding of CALR del52 to TpoR is mediated both by the mutant tail and N-domain interaction with immature sugars, the former providing specificity for TpoR against other N-glycosylated proteins.
The α -helical segment of CALR mutant tail induces TpoR dimerization and CALR oligomerization
Our data indicated that the CALR mutant tail provides specificity for TpoR through direct interaction. It is also known that the mutant tail is indispensable for activation of the receptor 7,9. Yet, exactly how activation is achieved remains unclear.
To close this gap, we sought to assess how the mutant tail could at the same time bind TpoR and induce its activation. We first determined the exact region of the mutant tail required for TpoR activation by progressively truncating the C-terminus of CALR del52 (Figure 3A). We probed the ability of these truncations to induce activation of the TpoR by measuring autonomous proliferation of cytokine-dependent hematopoietic cell lines (Ba/F3) stably expressing TpoR and either of the CALR truncations. Truncations until position 394, thus removing the non -helical segment of the mutant tail, allowed autonomous growth similar to that of full length CALR del52. In sharp contrast, further truncations led to decrease (CALR del52 M387*) or complete abolition of the activity of the mutant CALR. Thus, it is the -helical region of the mutant tail that is required for activation of TpoR. Importantly, the CALR del52 Y109F/D135L mutant, which abolishes glycan-dependent binding and disturbs the helicity of the mutant tail, did not allow Ba/F3 autonomous proliferation (Figure 3B). Then, we used cysteine crosslinking in live cells to study whether the same -helical segment of CALR mutant tail was key for TpoR homo-dimerization. We used the L508C mutant of human TpoR, a mutation homologous to murine TpoR L501C, putting the cysteine residue in a dimeric orientation that we previously established to reflect the active state of TpoR 21. The specificity of the crosslinking was achieved by using a truncated form of the TpoR devoid of intracellular cysteines which remains active 21 and by preventing crosslinking of cysteines in the ECD thanks to N-ethyl-maleimide which blocks free extracellular cysteines (Figure 3C). In line with our functional assays, we observed that truncations until M387 still enabled dimerization of the TpoR transmembrane domain in an active conformation while further truncations did not allow the formation of dimers (Figure 3D). Because oligomerisation of CALR mutants themselves is key for TpoR activation 22,23, we then used the same set of deletions to probe directly CALR oligomerization by co-immunoprecipitation of HA-tagged full length CALR del52 with FLAG-tagged CALR del52 truncations. Comparably to their effect in TpoR dimerization and activation, truncations of the mutant tail beyond its non- -helical segment led to sharp decrease in CALR mutant oligomerization. Thus, the first 28 -helical residues of the mutant tail are required for TpoR activation and dimerization of both CALR mutant and TpoR (Figure 3E). We confirmed these results by showing that recombinant CALR c-tail prevented the formation of oligomers while truncations of the non -helical segment of the mutant tail, like in CALR del52 394*, did not change the oligomeric profile in presence or not of a reducing agent (DTT) in native conditions (Supplementary Figure 3F).
Finally, we used RosettaDock 24 to model CALR del52 dimers formation with monomeric structure predicted by AlphaFold 2.0 16. The top 10 models predicted dimerization through the mutant tail through residues prior A394 and the two C-terminal cysteines. The best scoring prediction is depicted in Figure 3F and shows dimerization of CALR del52 mutant tail forming a coiled-coil like structure with interactions involving Arg (dark blue), Met (orange) and Thr (purple), but not the cysteines at the C-terminus of the mutant tail. Together with above results, this demonstrates that one face of the -helical mutant C-terminus provides specificity through direct interaction with a precise region of TpoR while the other face induces dimerization of CALR, imposing an active dimeric conformation to TpoR.
Mapping interactions in the TpoR-CALR mutant complex
Having established that the tail of CALR mutant directly interacted with the TpoR ECD, we sought to identify the regions of TpoR involved in this interaction. The TpoR ECD is composed of four sub-domains labelled D1 to D4. Our previous data showed that glycosylation of Asn117 in D1, and to a lesser extent, Asn178 in D2, was critical for the formation of a productive TpoR-CALR mutant complex 12,13.
To investigate whether the same domains of TpoR were also sufficient to mediate binding via the mutant tail, we first used co-immunoprecipitation of C-terminal truncations of the TpoR ECD with CALR del52. Truncations of the D3 and D4 domains did not impact binding while truncations of D2 led to a small but significant reduction of co-immunoprecipitation compared to the D1 domain alone in line with a minor role of glycans linked to Asn178 (Figure 4A-B). Since we showed that only the mutant tail interacted with TpoR in presence of mature glycans, we used the CALR del52 Y109F/D135L deficient for binding immature glycans 12,17. Expectedly, loss of glycan-dependent binding led to a sharp decrease in co-immunoprecipitation ratios. Remarkably, though, deletions of D3D4 or even D2D3D4 allowed similar binding to CALR del52 Y109F/D135L as D1 alone (Figure 4A-B), suggesting that the mutant tail interacted essentially with TpoR D1 domain. To confirm that interaction between the mutant tail occurred through D1, we measured BRET in living cells between CALR del52 P-C and C-domain and subdomains of TpoR ECD. In line with co-immunoprecipitation, BRET showed similar or even slightly higher interaction between the CALR mutant devoid of the lectin binding domain and TpoR D1 compared to D1D2 or D1D2D3D4 (Figure 4C).
Next, we sought to assess whether the presence of immature glycans on TpoR affected the interaction with the mutant tail and identify the binding sites of CALR del52 with immature glycans on TpoR. Knowing that TpoR D1D2 was sufficient to mediate full binding to CALR del52 through both glycans and the mutant tail, we produced the CALR del52-TpoR D1D2 complex in S2 cells (Figure 4D and Supplementary Figure 4A-B). In this complex, immature glycans are attached to Asn117. Using our HDx-MS set up, we mapped all sites of CALR del52 interaction with TpoR D1D2. This revealed that mutant calreticulin interacted with TpoR via two major domains. First, the strongest interaction was observed in the putative glycan binding site of calreticulin (Figure 4E-F and Supplementary Figure 4C-E). This region included notably C105, Y109, D135 and W319, all reported to be key for binding of sugar moieties 12,25-27. Importantly, this region was not involved in binding of mature TpoR (Figure 2B), confirming that it is specifically involved in interaction with immature glycans. Secondly, multiple peptides containing CALR del52 mutant tail equally showed a high degree of differential uptake between the CALR del52-D1D2 complex and CALR del52 alone (Figure 4E-F and Supplementary Figure 4C-E); these were similar as the ones exhibiting differential uptake also when in complex with the mature TpoR. This confirmed that the mutant tail is involved not only in binding to mature forms of the receptors, but also to immature TpoR, the interaction being further consolidated by strong interactions between the CALR glycan binding pocket and immature sugar moieties on Asn117 of TpoR D1D2 13.
CALR mutant tail interacts with acidic patches on TpoR D1 domain
To achieve a more thorough understanding of the CALR-TpoR interaction, we then aimed to identify specific residues on TpoR involved in binding the mutant tail. We first used molecular dynamics (MD) to study possible sites of interactions between CALR mutant tail and TpoR ECD. We generated the structure of TpoR D1D2 (see methods) to which glycosylation sites were attached at Asn117 (immature) and Asn178 (mature) (Supplementary Figure 5A). The final model was in agreement to the one recently published by AlphaFold 2.0 16. Sequence analysis indicated that TpoR D1D2 exhibits an unbalanced charge composition with an excess of 11 negatively charged acidic amino acids with one extensive (S1) and a second more localized (S2) negatively charged region (Supplementary Figure 5A), both particularly prone for interaction with the highly basic C-terminus tail of mutant CALR. The role of the acidic-basic interactions was confirmed by showing that mutations of hydrophobic residues to either Gly or Asn led to a small but significant decrease in activation of the TpoR in a STAT5 transcriptional assay but that mutations of basic residues to Asn or Gly completely abolished CALR del52 dependent activation of TpoR (Supplementary Figure 5B-C), in agreement with previous reports 7. Then, we used in silico simulation to probe for the most stable interacting sites of the mutant tail alone to TpoR D1D2 domains (see Methods). Three final poses were obtained after 500 ns Molecular Dynamics runs (Supplementary Video 1-3). Amongst them, two were in line with our experimental data, predicting interactions occurring essentially through either the extended S1 region of TpoR D1 (centered on 46ED47) or the more restricted S2 regions (centered on 53EEE55) and basic Arg on CALR mutant tail (Figure 5A). We further estimated the free energy (ΔG) using the Prodigy 28,29. There were very large number of microstates of the complex showing very large negative values of the free energy (ΔG< –9 kcal/mol), consistent with a high binding affinity of CALR del52 mutant tail for TpoR D1D2. The computational work indicated that CALR del52 mutant tail has the ability to engage TpoR with very high affinity in a very large number of micro-configurations that target both the continuous acidic area found mainly on D1 (and partly on D2) labelled as S1, but also the small acidic patch in the N-terminal region of D1 (S2).
To challenge our MD simulation experimentally, we used HDx-MS with the same set-up as in Figure 2A, where we showed that CALR mutants interact with mature TpoR exclusively through the mutant tail. Amongst the covered region, by far the strongest interaction was observed with the 41FSRTFEDL48 motif of TpoR D1 domain, which was statistically significant (p<0.001) for all incubation time points (Figure 5B-D and Supplementary Figure 5D). Remarkably, the very same peptide could also be detected from the CALR del52-TpoR produced as a complex containing immature glycans as depicted in Figure 4D where it was even less accessible than in the glycan-independent interaction (Figure 5D). This demonstrates that this region is involved not only in binding the mutant tail in absence of immature glycans but in complex formed in living cells with immature TpoR. Among other noticeable peptides, the 51WDEE54 from the S2 patch and 272WSLPVT276 close to the extended S1 region showed significant, albeit lower, decrease in accessibility for most time points after incubation of mature TpoR D1D4 with CALR del52 (Figure 5B-C and Supplementary Figure 5E). This suggests that, in absence of immature glycans, multiple micro-configurations of TpoR-CALR del52 can co-exist but that the S1 patch centered on 44TFED47 plays a central role in binding. Consistently, mutating this motif to alanine in full length TpoR led to complete loss of TpoR activation by both CALR del52 full length and P-C domain, as assessed by a STAT5 luciferase transcriptional assay (Figure 5E). Thus, interaction between CALR mutant tail and TpoR occurs through an extended patch of negative charges on TpoR D1 domain centered on residues 46ED47.
Comprehensive model of the TpoR-CALR mutant complex
HDx-MS and MD simulations using only the mutant tail showed that the latter was able to bind the receptor in a large variety of micro-configurations and identified the extended S1 acidic region centered on 46ED47 of TpoR as critical for the mutant tail interaction with TpoR. Yet, binding of immature TpoR to full length CALR del52 also involves strong interaction between specific residues of the N-domain and immature glycans on Asn117 (Figure 4E). We sought to create a complete model of the tetrameric TpoR-CALR mutant complex using our structural data. We used AlphaFold 2.0 16 to complete our modeling of TpoR to obtain the full extracellular domain of the receptor. TpoR monomers were dimerized through their TM domain with residue L508 in the interface as in the active configuration in presence of CALR del52 (Figure 3C-D). CALR del52 dimers (Figure 3F) were docked to dimers of TpoR through glycan binding domains and mutant tail based on experimental data and energy minimization. The final structure places the mutant tail in a configuration where the main interacting sites are located around the 46ED47 motif, in line with our HDx-MS and functional results. Likewise, immature glycans on Asn117 of the receptor interacting with the N-domain pocket containing key residues involved in glycans binding including C105, Y109, D135 and W319 (Figure 6). This glycoproteic tetramer was immersed into a full-atom representation of the environment, including a POPC lipid bilayer. This overall system consisting of ~1.5 million atoms was subjected to a complete molecular dynamics cycle at 300K for 20 ns (see methods). The complex remained very stable during this timeframe, with the exception of the very flexible P-domain (Supplementary Video S4). Contacts between TpoR and CALR molecules were analyzed over the course of the simulation. Most contacts relied on basic-acidic interactions and occurred both in cis and in trans, thereby further stabilizing TpoR dimers (Supplementary File 1). In conclusion, we provide here a model where CALR del52 interacts through two regions with TpoR D1 domain. First, the mutant tail directly interacts with multiple negatively charged residues on the inner/lateral face of TpoR D1 domain represented by the S1 negative patch. Given the ability of the mutant tail to interact with multiple acidic residues on TpoR D1 domain, it is likely that different micro-configurations co-exist in living cells, especially in absence of immature glycans to stabilize one particular configuration. When this interaction occurs between immature TpoR and CALR del52, it is further stabilized by strong interactions between CALR glycan binding domain and immature sugar moieties mainly on Asn117 of TpoR with the mutant tail providing strong specificity and stability for interaction between CALR mutants and TpoR.