## Introduction

Balance sensing in vertebrates depends on signals received in the inner ear, specifically in the semicircular canals and in the vestibule. The semicircular canals allow the detection of angular acceleration thanks to the inertial flow of the endolymph, a fluid that fills the inner ear, which deflects the cupula and thus stimulates mechanoreceptors in the sensory epithelium. The vestibule consists of an utricle, a saccule, and in all vertebrates besides mammals, of a lagena, which are collectively called vestibular organs. The vestibular organs contain biominerals called “ear stones” anchored to the sensory epithelium through a gelatinous membrane1,2. They contain calcium carbonate (90–97% by mass), biological macromolecules including glycoproteins and proteoglycans and trace mineral additives, especially Sr2+ and Ba2+3,4,5,6. The ear stones have a higher density than the endolymph, and during movement of the head, they move relative to the sensory epithelium, which allows to sense linear acceleration. They also contribute to spatial orientation due to the sensing of the gravitational field2. The ear stones differ among vertebrates. In general, in fish, they form large continuous crystals called otoliths; in terrestrial vertebrates, they appear as tiny multiple otoliths, specifically named otoconia, which are embedded in a gelatinous membrane3,7. These biominerals contain calcium carbonate of different polymorphs. Most often, teleost fish otoliths contain aragonite or vaterite8. Otoconia can contain aragonite (amphibians), calcite (birds and mammals), or a mixture of both polymorphs (reptiles)7. More detailed reviews of the evolution of otoliths and otoconia can be found elsewhere3,9. The mechanism of biomineralization of otoliths and otoconia, especially the mechanism of polymorph selection, is poorly understood10,11.

Numerous studies have shown that biomineralization of otoliths and otoconia is controlled by proteins, many of which are acidic and bind calcium ions (Ca2+)11. For example, intrinsically disordered Starmaker protein (Stm) is necessary to form properly shaped aragonite otoconia in zebrafish Danio rerio12. The biomineralization activity of Stm and its analogues from other fish was thoroughly analysed in vitro13,14,15,16, but although these proteins strongly influenced the size and shape of calcium carbonate crystals grown in vitro, they were not sufficient to induce the formation of aragonite instead of calcite, the most stable polymorph of calcium carbonate. In the case of otolith matrix macromolecule-64 (OMM-64) from rainbow trout Oncorhynchus mykiss, an analogue of Stm, aggregated otolith matrix proteins were indeed shown to be required together with OMM-64 to form aragonite in vitro17. Recently, transcriptomic and proteomic studies have allowed to identify proteins that constitute complex protein mixtures of the endolymph and otolith matrix16,18. Remarkably, a major constituent of the aggregate is otolin-1—a short chain collagen-like protein, which is strongly conserved in evolution16,18,19. Proteolytic cleavage of otolin-1 has been suggested to be a key step in temporally regulated deposition of aragonite in otoliths18.

The term “otolin” was first used in the 1960s for a component of the otolith matrix, which maintained similar amino acid compositions among different species of fish20. In the 2000s, genes encoding otolin-1 from various organisms were cloned21,22,23,24, and many more otolin-1 genes since then have been identified on the basis of homology. Otolin-1 belongs to the C1q superfamily of proteins, which includes short chain collagenous proteins involved in a wide array of biological processes such as innate immune response (C1q)25, insulin sensitivity (adiponectin)26, cell adhesion (collagen VIII, CTRP5)27,28 and biomineralization (collagen X, otolin-1)29,30. In addition to a collagen-like domain, otolin-1 contains a short N-terminal domain rich in cysteine residues and a globular C1q-like (gC1q) domain at the C-terminus. Function of the N-terminal domain is not known. In adiponectin, a serum protein hormone similar to otolin-1, the N-terminal fragment is responsible for the aggregation of adiponectin trimers to biologically active high molecular weight aggregates26,31,32. The gC1q domain is inherently able to form trimers that, together with the ability of the collagen-like domain to form a triple helix, enable it to form fibrils, which could be interconnected by N-terminal domains and gC1q domains. Such a fibrillary network could provide a scaffold for biomineralization of otoliths and otoconia11,30,33. Moreover, murine otolin-1 has been shown to interact with otoconin-90 (Oc90), a major otoconial matrix protein, through the collagen-like and gC1q domains24,34. This interaction enhances the biomineralization activity of otolin-1 and Oc9030, indicating that otolin-1 has two critical functions in the biomineralization of otoliths and otoconia, namely, structural function by providing a scaffold for biomineralization and regulatory function by interacting with other matrix proteins, which allows biomineralization to be controlled. The affinity for Ca2+ also suggests the direct involvement of otolin-1 in the localized precipitation of calcium carbonate through interactions with the mineral phase33.

The molecular mechanism of otolin-1 trimerization is not known. Similar proteins from the C1q superfamily such as collagen X, adiponectin or C1q contain Ca2+ in their gC1q domains near the axes of the trimers29,35,36. In this work, we substituted two conserved acidic residues of the gC1q domain of human and zebrafish otolin-1 (abbreviated as hOtolC1q and dOtolC1q, respectively), which may bind Ca2+, with neutral alanine, and we analysed the influence of mutations on the effect of Ca2+ on the secondary structure, oligomerization propensity and stability of gC1q otolin-1. We also checked whether an abundant natural missense polymorphism in the human otolin-1 gene impacts the structure of the protein. Single mutations of the putative Ca2+-binding residues caused mixed effects. Surprisingly, when both residues were mutated, trimers of gC1q domains were highly stable even in the absence of Ca2+. Nevertheless, from all studied variants and conditions, native forms of the gC1q domain of otolin-1 saturated with Ca2+ were the most stable. We observed significant differences between hOtolC1q and dOtolC1q. A single mutation in the calcium binding site was enough to diminish the ability of dOtolC1q to bind Ca2+, which was not the case for hOtolC1q. Moreover, dOtolC1q and its mutants had lower thermal stability than all variants of hOtolC1q in all tested conditions. These observations allow us to propose a mechanism of oligomerization of otolin-1, in which the neutralization of the negative charge by Ca2+ is required at the axis of the gC1q trimer.

## Results

### Predictions indicate detrimental effects of mutations in the gC1q domain of otolin-1

We used data acquired from the Ensembl database and predictions in silico to screen currently known missense single nucleotide polymorphisms (SNPs) of hOtolC1q and dOtolC1q for their abundance and possible effect on protein function and stability. To be qualified as an SNP, a variant must be found in at least 1% of the population37. Following this definition, we found a single SNP in a region coding hOtolC1q, in which glutamate-470 is replaced by alanine (E470A) due to a substitution of adenine with cytosine (variant ID: rs3921595). The prevalence of this variant is close to 50%. Such a widespread genetic variant should not have any adverse effect on human health. However, in the model of hOtolC1q obtained through ensemble optimization method (EOM) analysis of small angle X-ray scattering (SAXS) data33, E470 is predicted to be near the trimerization interface (Fig. 1a,b). Therefore, E470 could be engaged in the stabilization of a gC1q trimer through ionic interactions. The substitution of glutamate with alanine at this position might impair the folding of otolin-1 into a trimer or decrease its stability by inducing aggregation. The importance of E470 of otolin-1 is highlighted by a strong conservation of this residue among different organisms (Fig. 1c, purple asterisk). The damage potential of the E470A substitution is indicated by SIFT and PolyPhen2; however, the model-based SNP MuSiC tool predicts this variant to be neutral (Table 1). We decided to analyse how E470A impacts the solution structure of hOtolC1q and its sensitivity towards Ca2+.

Acidic residues that contribute to the binding of Ca2+ by the gC1q domains of collagen X and adiponectin are also strongly conserved in otolin-1 (Fig. 1c, red and blue asterisks). Therefore, in addition to testing the effect of the E470A substitution in hOtolC1q, we decided to analyse how substitution of putative Ca2+-binding aspartate (hOtolC1q D425, dOtolC1q D432 and D440) and glutamate (hOtolC1q E417) residues with alanine would affect hOtolC1q and dOtolC1q. Mutations of the putative Ca2+ binding sites of hOtolC1q and dOtolC1q are clearly predicted to be damaging to the protein by all tools used (Table 1).

### Tb3+ as a LRET probe for binding of Ca2+

We attempted to measure Ca2+ binding properties of native and mutated hOtolC1q and dOtolC1q. However, upon the removal of Ca2+ by overnight incubation with buffered Chelex 100 resin (Bio-Rad, Hercules, CA, USA)38, native hOtolC1q and dOtolC1q precipitated completely. Dialysis against Ca2+-depleted buffer did not result in satisfactory removal of bound Ca2+. Upon concluding that the proteins were not obtainable in a Ca2+-free form without the addition of EDTA or EGTA, we decided to indirectly estimate the affinity of the proteins to Ca2+ by measuring the binding of terbium(III) ions (Tb3+), as Ca2+-binding proteins can also bind lanthanide ions due to similar effective ionic radii and positive charges of Ca2+ and lanthanide ions39. The binding of lanthanide ions by Ca2+-binding proteins is usually stronger than the binding of Ca2+, due mainly to a higher positive charge40,41,42. Therefore, protein-bound Ca2+ should be substituted by Tb3+. Specifically, the use of Tb3+ in such studies gives a significant advantage, as when bound to proteins, Tb3+ can be detected thanks to luminescence resonance energy transfer (LRET) from aromatic side chains (phenylalanine, tryptophan and tyrosine) to the bound Tb3+. Upon excitation of the proteins at the aromatic absorption band (280 nm), sensitized emission from the bound Tb3+ can be observed with a maximum at 545 nm39,43. Both hOtolC1q and dOtolC1q contain numerous aromatic residues, which are a prerequisite for LRET to occur. The results of the titration experiments are shown in Fig. 2.

When the samples of hOtolC1q and dOtolC1q contained Tb3+, fluorescence emission in the visible part of the spectrum with a maximum at 545 nm was observed upon excitation at 280 nm. The intensity of fluorescence emission increased when subsequent portions of TbCl3 were added (Fig. 2a,c). Apparent dissociation constants (Kd) calculated using a model of a single binding site per protein monomer were 2.0 µM for hOtolC1q and 18.2 µM for dOtolC1q (Fig. 2b,d). Mutants of hOtolC1q with at least one putative Ca2+ binding residue mutated bound Tb3+ weaker than the native protein, whereas the double mutant hOtolC1q E417A D425A (Kd = 4.5 µM) seemed to bind Tb3+ stronger than E417A (Kd = 20.0 µM) or D425A (Kd = 24.6 µM) mutants. E470A substitution is in this case indistinguishable from native hOtolC1q. In the case of dOtolC1q, the D440A mutant bound Tb3+ weaker than dOtolC1q (Kd = 32.3 µM). Surprisingly, the D432A mutant (Kd = 2.4 µM) seemed to have a higher affinity for Tb3+ than native dOtolC1q (Kd = 18.2 µM). A double mutant dOtolC1q D432A D440A had the lowest affinity to Tb3+ of all tested forms of dOtolC1q (Kd = 85.9 µM).

Mutations of the putative Ca2+-binding residues did not abolish the ability of OtolC1q to bind Tb3+. For all mutants except dOtolC1q D432A and hOtolC1q E470A, the affinity was decreased compared to the affinity of the native proteins, as indicated by an increase in Kd. The E470 residue of hOtolC1q did not seem to bind Tb3+, as substitution with alanine did not affect the affinity of hOtolC1q towards Tb3+. In the case of dOtolC1q D432A, we noted an increase in affinity (decrease in Kd). Notably, the emission intensity of dOtolC1q D432A was the highest of all tested forms of OtolC1q (Fig. 2d), which indicated that the mutation caused a structural change, which displaced the aromatic residues closer to the Ca2+ binding site, as the efficiency of LRET increased. The structural change also seems to facilitate the binding of Tb3+. In further experiments, we analysed how the mutations affected the secondary structure, oligomerization propensity and thermal stability of the gC1q domain of otolin-1. The experiments were conducted in the presence of EDTA, Ca2+ and Tb3+ to expand the conclusions drawn from the Tb3+ binding experiment.

### Mutations affected the secondary structure of hOtolC1q and dOtolC1q

Circular dichroism spectroscopy (CD) is a useful, sensitive technique that allows us to estimate the secondary structure of proteins and to detect changes in the structure induced by denaturation, stabilization or ligand binding44. We have previously used it to analyse Ca2+-induced structural changes in hOtolC1q and dOtolC1q33. Here, we used CD to identify structural changes induced by the mutations. CD also allowed us to assess secondary structure changes accompanied by binding of Ca2+ and Tb3+ to mutated hOtolC1q and dOtolC1q (Fig. 3, Suppl. Fig. 1). Concentrations of TbCl3 chosen for the experiments were saturating for native hOtolC1q and dOtolC1q. Higher concentrations, but still as low as 1 mM TbCl3 (85-fold molar excess), caused visible precipitation of both hOtolC1q and dOtolC1q (not shown).

In the case of hOtolC1q, spectra collected for proteins depleted from Ca2+ by 1 mM EDTA indicate that mutations involving the Ca2+ binding site led to changes in β-structure content (Fig. 3a). Native hOtolC1q has a negative ellipticity band at approximately 215 nm, which is attributed to β-sheets, and a decreasing ellipticity below 215 nm. The 215-nm band became deeper for hOtolC1q E417A, shallower for D425A and shallowest for E417A D425A. The spectra also changed at 233 nm in the positive aromatic side chain band45,46. E470A substitution did not affect the CD spectrum of hOtolC1q. Spectra collected for proteins in the presence of 100 mM CaCl2 indicate that only the native hOtolC1q and the E470A variant significantly changed their structures upon binding of Ca2+ (Fig. 3c, Suppl. Fig. 1a,b,d–f). For these proteins, in the presence of 100 mM CaCl2, the ellipticity below 215 nm increased instead of slightly decreasing, as was observed for 1 mM EDTA. This effect can be attributed to an increase in β-structure content with a subsequent decrease in protein disorder33. The addition of 100 mM CaCl2 to hOtolC1q D425A led to the shallowing of the 215 nm negative ellipticity band, indicating a minor structural change (Suppl. Fig. 1e). The spectra collected in the presence of a sevenfold excess of Tb3+ are similar to the spectra collected in the presence of Ca2+ (Fig. 3e, Suppl. Fig. 1a,b,d–f). Slight differences in ellipticity at 215 nm were observed for the E417A, D425A and E470A variants (Suppl. Fig. 1b,d,e). It seems that the E470A variant, despite binding Tb3+ with similar affinity to native hOtolC1q, has a slightly different secondary structure than native hOtolC1q when Tb3+ ions are bound. We conclude that in terms of the secondary structure of hOtolC1q, the binding of Tb3+ has a similar effect as the binding of Ca2+.

In the case of dOtolC1q, all spectra collected in the presence of 1 mM EDTA, despite that for the D432A D440A mutant, are similar (Fig. 3b). For dOtolC1q, the negative band at approximately 215 nm indicates the presence of β-sheets. The D432A and D440A mutants can be distinguished from the native form by an increased ellipticity at the 233-nm band, which is attributed to aromatic residues. This difference supports the conclusion that aromatic residues were displaced due to the mutations, and thus, the LRET signal was strongly increased for D432A and slightly decreased for D440A. Weakening of the CD signal of dOtolC1q D432A D440A when compared to other forms suggests that the double mutation of the Ca2+ binding site interrupted folding of dOtolC1q. When the spectra collected in the presence of 1 mM EDTA were compared with those obtained for samples with 100 mM CaCl2, it became apparent that the D440A and D432A D440A mutants were not sensitive to the presence of Ca2+ (Suppl. Fig. 1 h,i). As observed previously33, dOtolC1q underwent a major structural change when Ca2+ was added (Fig. 3, Suppl. Fig. 1c). The ellipticity minimum at 215 nm becomes deeper, and the signal strongly increases below 215 nm. As for hOtolC1q, this is interpreted as induction of β-sheets from the disordered portions of the protein. The structural change is more pronounced for dOtolC1q than for hOtolC1q. A weaker but significant change in the spectrum was also observed for dOtolC1q D432A (Suppl. Fig. 1g). This indicates that the D432A mutant retained some affinity towards Ca2+. The spectrum of dOtolC1q collected in the presence of a 14-fold molar excess of Tb3+ is much shallower below 230 nm than in the presence of 1 mM EDTA or 100 mM CaCl2. We interpret such a major change in the spectrum as a manifestation of destabilization of the protein, as at this concentration of protein (0.2 mg/ml, 11.7 µM) and Tb3+ (163.8 µM), precipitation became visible after prolonged incubation. The protein precipitated immediately at higher concentrations of dOtolC1q and Tb3+. For the mutants, the spectra collected in the presence of Tb3+ were similar to those collected in the presence of Ca2+. We conclude that, contrary to hOtolC1q, the binding of Ca2+ to dOtolC1q cannot be reliably simulated using Tb3+. Nevertheless, we can conclude that the double mutants (hOtolC1q E417A D425A and dOtolC1q D432A D440A) are completely unable to bind Ca2+, as their secondary structure is the same in the presence of 1 mM EDTA, 100 mM CaCl2 and an appropriate excess of TbCl3.

### Mutations affected Ca2+-dependent trimerization of hOtolC1q and dOtolC1q

The CD measurements showed significant secondary structure changes induced by mutations in the Ca2+ binding sites of hOtolC1q and dOtolC1q. However, analysis of the influence of Tb3+ on the secondary structure of hOtolC1q and dOtolC1q and comparison with the effect of Ca2+ gave inconclusive results. It seems that Tb3+ can be used to simulate the binding of Ca2+ to hOtolC1q but not to native dOtolC1q, which aggregates in the presence of Tb3+. We analysed the tertiary structures of hOtolC1q, dOtolC1q and their mutants using sedimentation velocity analytical ultracentrifugation (SV AUC) to see how mutations of Ca2+ binding sites affected the oligomerization propensity of gC1q otolin-1. We hypothesized that as Ca2+ stabilizes trimers of hOtolC1q and dOtolC1q33, alanine mutations in the Ca2+ binding site could affect the formation of gC1q trimers. We also checked the influence of Ca2+ and Tb3+ on the structure of the mutants to see how the eventual changes correspond to the results of LRET and CD measurements. The results of SV AUC are summarized in Fig. 4, Suppl. Fig. 2, Suppl. Fig. 3 and Suppl. Table 1.

Introduced mutations had a significant impact on the oligomerization propensity of hOtolC1q and dOtolC1q. (Fig. 4a,b). In the presence of 1 mM EDTA, hOtolC1q and dOtolC1q formed unstable oligomeric forms, which sedimented slower than expected trimers due to the fast association and dissociation events occurring during the overnight ultracentrifugation experiment33. Alanine mutations in the Ca2+ binding site induced stable oligomerization even in the absence of Ca2+, as evidenced by the apparent molecular weight values (MWapp) derived from SEDFIT analysis (Suppl. Table 1), which indicated that hOtolC1q D425A, hOtolC1q E417A D425A and dOtolC1q D432A D440A form stable trimers even in the presence of 1 mM EDTA (monomer MW is 17.0 kDa for hOtolC1q and 17.1 kDa for dOtolC1q). This observation was unexpected, as a decrease in trimerization propensity was anticipated to occur due to the introduced mutations. hOtolC1q E417A also seems to form trimers in the absence of Ca2+, but a wider c(s) distribution than for D425A and E417A D425A mutants (Fig. 4a) suggests the lower stability of the oligomer. dOtolC1q D440A seems to form trimers, but a trace of lower molecular weight forms is present (Fig. 4b). dOtolC1q and dOtolC1q D432A did not form stable oligomers, but the c(s) distribution of dOtolC1q D432A compared to native dOtolC1q seems to be skewed towards heavier forms. Analysis of the concentration dependence of c(s) distributions (Suppl. Fig. 3, Suppl. Table 1) shows that the trimers formed by hOtolC1q with any of the Ca2+-binding residues mutated are stable, which is also true for dOtolC1q D432A D440A, with both Ca2+-binding residues being mutated. Both the D432A and D440A mutants of dOtolC1q show concentration dependence of c(s) distributions, but it is not as strong as in the case of dOtolC1q. When the proteins were analysed in the presence of 10 mM CaCl2, it became apparent that the oligomers formed by the dOtolC1q D432A mutant were not as stable as the oligomers formed by native dOtolC1q (Fig. 4c,d). Under these conditions, dOtolC1q formed dimers and trimers. The proportion of trimers relative to dimers increased with increasing concentrations of dOtolC1q (Suppl. Fig. 3c, Suppl. Table 1). In contrast, the D432A mutant still did not form stable trimers. The c(s) distributions reveal a small concentration dependence (Suppl. Fig. 3g). Mutating the D432 residue of dOtolC1q seemed to disturb the formation of stable trimers. Neither dOtolC1q D440A nor the double mutant seemed to be affected by Ca2+ (Suppl. Fig. 2h,i; Suppl. Fig. 3h,i). Sedimentation profiles of hOtolC1q E417A, D425A and a double mutant also did not change upon the addition of Ca2+.

When Tb3+ ions were added to the proteins instead of Ca2+, hOtolC1q and dOtolC1q did not respond in the same way (Fig. 4e,f). hOtolC1q and all its mutants formed stable trimers in the presence of Tb3+ (Fig. 4e). This is different for dOtolC1q because its c(s) distribution shows signs of aggregation at 0.25 mg/ml protein and 14-fold molar excess of TbCl3 (Fig. 4f), amplified by the observation from preliminary experiments, in which 0.5 mg/ml dOtolC1q mixed with 1 mM TbCl3 precipitated completely. (i.e., the absorbance signal collected during SV AUC was zero, data not shown). In the presence of Tb3+, dOtolC1q D432A finally formed stable trimers. c(s) distributions of dOtolC1q D440A and dOtolC1q D432A D440A did not change compared to the c(s) distributions calculated for proteins with EDTA or Ca2+ (Suppl. Fig. 2h,i). The mutated forms of dOtolC1q seem to be stable in the presence of Tb3+ in a trimeric form, even though the native protein was not.

Compared to native hOtolC1q, the c(s) distribution of the natural variant hOtolC1q E470A in the absence of Ca2+ has a maximum at a lower sedimentation coefficient and a tail at a higher sedimentation coefficient (Fig. 4 a, Suppl. Fig. 3a,b), which may suggest decreased trimerization propensity and increased tendency to form aberrant aggregates, but similar sedimentation coefficients for both forms indicate that the effect is weak (Suppl. Table 1). In the presence of Ca2+ and Tb3+, both forms have highly similar c(s) distributions. Under these conditions, both the hOtolC1q and E470A variants formed stable trimers without any trace of unbound subunits or heavier aggregates (Fig. 4c,e).

SV AUC provided interesting information about the effects of mutations of the Ca2+ binding site of gC1q otolin-1. The substitution of conserved Ca2+ binding aspartate and glutamate residues with alanine seems to diminish the ability to bind Ca2+, which is normally found at the axis of the gC1q trimer29,47,48. However, instead of inhibiting Ca2+-dependent formation of the trimers, this substitution led to the formation of stable trimers even in the absence of Ca2+. This effect may be explained by the neutralization of the potentially repulsive electrostatic charge at the axis of the gC1q trimer by substitution of negatively charged carboxyl groups of aspartate and glutamate with uncharged methyl groups of alanine. Only in the case of dOtolC1q, mutation of D432 with D440 left intact clearly disturbed the Ca2+-dependent trimerization of the protein. Interestingly, Tb3+ allowed this mutant to form trimers. The modified Ca2+ binding site of dOtolC1q D432A seems to be accidentally optimized for the binding of Tb3+.

### Mutations decreased the stabilizing effect of Ca2+ on hOtolC1q and dOtolC1q

hOtolC1q and dOtolC1q were stabilized by Ca2+, which was shown using CD and differential scanning fluorimetry (DSF) with SYPRO Orange as a probe33. In this work, we used DSF to analyse how the mutations affected the thermal stability of hOtolC1q and dOtolC1q. We also analysed the effect of Ca2+ and Tb3+ on the stability of the mutants. Both hOtolC1q and dOtolC1q were strongly stabilized by Ca2+ (Fig. 5, Table 2), and their transition temperature (Tm) increased from 41 to 44 °C in the absence of Ca2+ to 90 °C or more at 100 mM CaCl2 (Fig. 5, Table 2). Interestingly, mutants of hOtolC1q with substituted Ca2+-binding residues were more stable than native hOtolC1q in the absence of Ca2+; their Tm values were 62–65 °C for E417A and D425A and as high as 83 °C for the double mutant. hOtolC1q E470A had a denaturation profile very similar to hOtolC1q under all tested conditions. Ca2+ stabilized the E417A and D425A mutants, as their Tm was above 80 °C in 100 mM CaCl2. The double mutant, E417A D425A, did not seem to be affected by Ca2+ up to 10 mM CaCl2. In 100 mM CaCl2, the double mutant was slightly destabilized. Although hOtolC1q with any of the Ca2+ binding residues mutated was more stable than the native form in the absence of Ca2+, native hOtolC1q supplemented with 100 mM CaCl2 was the most stable of all samples measured, also true for the E470A variant, in which the Ca2+ binding site was unaffected. Although this concentration of Ca2+ seems to be very high compared to the physiological Ca2+ concentration in the mammalian endolymph (approximately 0.1 mM)49, it simulates intracrystalline conditions of otoconia. Tb3+ ions stabilized hOtolC1q and all its mutants, but the effect was not the same as the effect for Ca2+.

In the case of dOtolC1q, the D432A and D440A mutants were slightly more stable than dOtolC1q, as their Tm was approximately 45 °C in the absence of Ca2+. The double mutant D432A D440A was very stable under these conditions, as its Tm reached 80 °C. The addition of Ca2+ stabilized dOtolC1q and, to a lesser extent, the D432A mutant. Native dOtolC1q and the D432A mutant were also stabilized by Tb3+, but the effect was different than for 100 mM CaCl2: weaker for dOtolC1q and stronger for D432A. Stabilization of dOtolC1q D432A by Tb3+ relative to Ca2+ is another indication that the mutation changed the structure of the protein, resulting in preferential binding of Tb3+. Strikingly, D440A and the double mutant D432A D440A were not stabilized by Ca2+. Similar to hOtolC1q E417A D425A, they were destabilized by 100 mM CaCl2. These mutants were also not stabilized by Tb3+. We can therefore conclude that hOtolC1q E417A D425A, dOtolC1q D440A and dOtolC1q D432A D440A are completely unable to bind Ca2+ or Tb3+ at the mutated binding site. The fluorescence signal observed in Tb3+ binding experiments for these mutants may come from nonspecific binding of Tb3+ at adventitious sites40, seemingly without an effect on their thermal stability. Mutation of a single residue, D440, seems to be sufficient to diminish the ability of dOtolC1q to bind Ca2+. An analogous mutant of hOtolC1q, D425A, seems to retain minor affinity to Ca2+.

## Discussion

In this work, a set of biophysical techniques was used to assess the mechanism of trimerization and the influence of mutations of chosen acidic residues on the structure and Ca2+ binding properties of hOtolC1q and dOtolC1q. The results are summarized in Table 3. We observed interesting differences between human and zebrafish forms of the gC1q domain of otolin-1, as a single mutation (D440A) was enough to completely diminish the ability of dOtolC1q to bind calcium ions. Although the single mutations at the Ca2+ binding site of hOtolC1q (E417A and D425A) strongly decreased its affinity to Ca2+, the mutants were still stabilized by Ca2+ and Tb3+, as evidenced by DSF. The double mutant (hOtolC1q E417A D425A) was not stabilized by these ions. All tested forms of hOtolC1q had notably higher thermal stability than the respective forms of dOtolC1q under all tested conditions. Moreover, hOtolC1q was stabilized at a lower concentration of Ca2+ than dOtolC1q, clearly showing that during evolution, the gC1q domain of otolin-1 became optimized to function at lower concentrations of Ca2+. Endolymph Ca2+ concentration in teleosts varies, as it was reported to be 1.0 mM in turbot50, 1.0–1.6 mM in trout50,51,52 and as high as 2.9 mM in oyster toadfish53. In mammals, endolymph Ca2+ concentration was determined to be much lower, between 92 and 133 µM, depending on the compartment (guinea pig)49.

In silico predictions for the prevalent natural polymorphism E470A, involving a residue that was modelled to be away from the Ca2+ binding site of hOtolC1q, were ambiguous in terms of the effect on protein stability. Thermal stability in the presence of Ca2+, binding of Ca2+ or Tb3+ and solution structure of hOtolC1q were not significantly affected by this substitution. The only notable effects were decreased thermal stability of the E470A mutant below 1 mM Ca2+ and increased tendency to aggregate in the absence of Ca2+. Normally, this effect would have been negligible in vivo. Otolin-1, as a secreted protein, is after translation exposed to the ER lumen, which contains 100–800 µM Ca2+54. This concentration should be sufficient to stabilize the gC1q domain of the E470A variant. In the utricle and saccule, otolin-1 could be protected from low concentrations of Ca2+ by being bound within otoconia and cross-linked with other proteins in the otoconial membrane. However, a decrease in local Ca2+ caused by dysregulated calcium homeostasis or degeneration of otoconia might destabilize the protein, possibly accelerating the degradation of otoconial matrix and/or otoconial membrane. Therefore, it would be interesting to examine whether the presence of the E470A allele in the genome is associated with the prevalence of benign paroxysmal positional vertigo (BPPV). BPPV is a balance disorder caused by displacement of degraded otoconial membrane fragments to the semicircular canals55, where they disrupt the flow of endolymph or movement of cupula, which occur normally due to the movement of the head. This effect is felt by the affected person as sudden attacks of vertigo, which may lead to falls and bone fractures and, consequently, to disability, especially later in life56,57. The hypothesized lability at low Ca2+ is consistent with our observation that hOtolC1q E470A is particularly difficult to produce in Escherichia coli. The E470A variant is the sole form of hOtolC1q studied here that required the addition of a detergent to remain soluble after cell lysis, and still the purification yield was much lower than the purification yield for native hOtolC1q. The concentration of Ca2+ in the cytoplasm of E. coli, which was shown to be in the range of 0.3–1.6 µM58, is too low to stabilize hOtolC1q E470A. In the context of recombinant protein production, an increased tendency to form aggregates at low Ca2+ would be especially detrimental, as it would facilitate the precipitation of the protein in the form of inclusion bodies59. E470 of human otolin-1 is conserved in fish. If such a mutation would have a similar effect on piscine otolin-1, the continuous growth of the otoliths could be disrupted in the event of catastrophic loss of endolymph Ca2+. However, as native dOtolC1q was stabilized at higher Ca2+ than hOtolC1q, such an event would likely also affect the fish not having this mutation.

Mutations of the conserved Ca2+-binding residues of hOtolC1q and dOtolC1q were, however, consistently predicted to be deleterious and functionally damaging for the protein. Our results show that the predictions were not fully correct. All forms of hOtolC1q and dOtolC1q with any residue from the Ca2+ binding site mutated formed more stable oligomers in the absence of Ca2+ than the native proteins. Significant deleterious effects of the mutations became apparent when the proteins were studied with 10–100 mM CaCl2. Under these conditions, the D432A mutant of dOtolC1q was unable to form stable trimers, contrary to native dOtolC1q. Interestingly, this mutant seemed to be accidentally optimized for the binding of Tb3+ relative to Ca2+, as the mutation increased the apparent affinity of the protein towards Tb3+. This mutant also formed stable trimers only in the presence of Tb3+ and not in the presence of Ca2+. The most important observation, however, is that although the mutation of the Ca2+ binding sites of hOtolC1q or dOtolC1q seemed to have a stabilizing effect, native hOtolC1q and dOtolC1q were the most thermally stable forms in the conditions simulating the interiors of otoliths and otoconia. Additionally, hOtolC1q E417A, hOtolC1q E470A and all mutants of dOtolC1q were difficult to handle in the laboratory: the expression yields were smaller than for the native forms, and protein loss during purification was significantly higher. The mutants were also difficult to concentrate. The highest achievable concentration of hOtolC1q was 4 mg/mL, and of dOtolC1q was 1.5 mg/mL. In contrast, hOtolC1q E417A and all mutants of dOtolC1q began to precipitate at approximately 0.5 mg/mL.

Crystallographic data show that Ca2+ is bound at the axis of a gC1q domain trimer29,48. Mutation of the Ca2+-binding residues of the gC1q domain of C1q-like protein 3 (C1QL3) significantly decreased its thermal stability48. It was plausible to conclude that mutation of a Ca2+ binding site in the gC1q domain of otolin-1 could disrupt the formation of trimers due to the lack of a stabilizing effect of Ca2+. However, mutations of the acidic residues forming the Ca2+ binding site to alanine actually facilitated the trimerization of hOtolC1q and dOtolC1q, explained by the neutralization of the repulsive negative charge at the trimerization interface by replacing acidic residues with uncharged residues. Interestingly, a gC1q domain of collagen VIII also forms trimers in the absence of Ca2+27. In this case, the Ca2+ binding sites are occupied by side chains of lysine residues from the same subunits, also leading to neutralization of the negative charge. We conclude that Ca2+-dependent neutralization of a negative charge at the Ca2+-binding site is crucial for the formation of the gC1q domain trimer.

Although our data do not explain why an increased preference of dOtolC1q D432A for Tb3+ occurs, the binding of Tb3+ is known not to be fully equivalent to the binding of Ca2+, although these ions have similar ionic radii due to the higher charge and higher observed coordination number of Tb3+ complexes compared to Ca2+ complexes with proteins40,60,61. Moreover, changes in hydrogen bonding around the metal binding site were observed when lanthanide ions were bound instead of Ca2+ to compensate for the charge difference40. In our case, this inequivalence apparently leads to the formation of stable dOtolC1q D432A trimers in the presence of Tb3+ but not in the presence of Ca2+, possibly also the reason for severe destabilization of dOtolC1q in the presence of Tb3+.

In light of the results of Tb3+-binding experiments, which seem to bind strongly to and thermally stabilize hOtolC1q, dOtolC1q and their mutants, it is surprising that the proteins were actually destabilized if we consider their solubility. For dOtolC1q, strong immediate precipitation in the presence of Tb3+ was observed at 0.5 mg/mL protein. During prolonged incubation, some precipitation was also observed at 0.2 mg/mL. Human isoforms seemed to better tolerate the presence of Tb3+. In the LRET experiment, Tb3+ seemed to bind to mutants that were not responsive to Ca2+. In the case of dOtolC1q, this did not lead to thermal stabilization of its nonbinding mutants. Tb3+ ions may have been bound away from the Ca2+ binding sites. Such “adventitious” binding of lanthanide ions at sites that do not bind Ca2+ was observed in the crystal structures of proteins and was characterized by a lower coordination number. The remaining coordination sites of lanthanide ions were speculated to be occupied by water40. In the case of dOtolC1q, the available coordination sites of Tb3+ could possibly become occupied by functional groups from other trimers and lead to the formation of heavy aggregates of dOtolC1q. Weak binding of Tb3+ at the adventitious sites could result in aggregation occurring at a relatively high protein concentration, above 0.2 mg/mL.

In summary, we discovered that the basis of the engagement of Ca2+ in the formation of gC1q domain trimers is the neutralization of the negative charge at the Ca2+ binding site. Despite the apparently stabilizing effect of substitutions of acidic Ca2+-binding residues with neutral alanine residues, hOtolC1q and dOtolC1q were ultimately the most stable in their native forms saturated with Ca2+. We also determined that hOtolC1q and its mutants were more stable than dOtolC1q and its mutants under all tested conditions, including a Ca2+-free environment. Apparently, the human protein became optimized to function at lower concentrations of Ca2+. We conclude that this predisposes otolin-1 to years-long function in biomineral organic matrices.

## Methods

### Accession numbers

Human OTOL1 gene Ensembl accession ID: ENSG00000182447.

Zebrafish OTOL1A gene Ensembl accession ID: ENSDARG00000001771.

Human otolin-1 UniProt accession ID: A6NHN0.

Zebrafish otolin-1 UniProt accession ID: A5PN28.

Human otolin-1 E470A SNP variant ID: rs3921595.

### Key resources

Synthetic cDNAs encoding full-length otolin-1 from human and zebrafish were codon-optimized for E. coli and provided by GeneArt (currently Thermo Fisher Scientific, Warsaw, Poland). Nucleotide primers were provided by Genomed (Warsaw, Poland). The pQE-80 L plasmid expression vector was purchased from Qiagen (Hilden, Germany). E. coli Top10 cells, Phusion DNA polymerase, restriction enzymes, T4 DNA ligase and LB broth were obtained from Thermo Fisher Scientific. One-fusion DNA polymerase was obtained from GeneOn (Ludwigshafen am Rhein, Germany; distributed by ABO, Gdańsk, Poland). Agar, agarose, tris(hydroxymethyl)aminomethane (Tris), ethylenediaminetetraacetic acid (EDTA), carbenicillin, isopropyl β-d-1-thiogalactopyranoside (IPTG), NaCl, glycerol, 2-mercaptoethanol, imidazole, glycine, sodium dodecyl sulfate (SDS) and CaCl2 were obtained from Carl Roth (Karlsruhe, Germany; distributed by Linegal Chemicals, Warsaw, Poland). E. coli BL21(DE3) cells, TB broth, 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), lysozyme, phenylmethylsulfonyl fluoride (PMSF), DNase I, RNase A, terbium(III) chloride hexahydrate, xylenol orange disodium salt and SYPRO Orange were from Sigma (currently Merck, Warsaw, Poland). The NP40 substitute was from Amresco (currently VWR, Gliwice, Poland). Empty Tricorn, Superdex 200 Increase 10/300 GL and Superdex 75 16/60 Prep Grade columns were obtained from GE Healthcare Life Sciences (currently Cytiva, Warsaw, Poland). TALON® Metal Affinity resin and 6xHis monoclonal antibody (albumin-free) were purchased from Takara Bio (Mountain View, CA, USA; distributed by Biokom, Janki, Poland). HRP horse anti-mouse IgG antibody was purchased from Vector Laboratories (Burlingame, CA, USA; distributed by Biokom, Janki, Poland).

### Identification of single nucleotide polymorphisms

The Ensembl genome browser (www.ensembl.org/) was queried for known single nucleotide polymorphisms (SNPs) in otolin-1 genes. The gene identifiers were ENSG00000182447 (OTOL1) for human otolin-1 and ENSDARG00000001771 (OTOL1A) for zebrafish otolin-1. Boundaries of gC1q domains were retrieved from the UniProt database (A6NHN0 for human otolin-1 and A5PN28 for zebrafish otolin-1) as 338–477 for human and 351–489 for zebrafish otolin-1.

### Prediction of mutation effects

Natural and designed mutants were assessed using online tools to predict their effects on protein structure and function. Data for the natural E470A SNP accompanied a respective entry in the Ensembl database, from which the results for SIFT (https://sift.bii.a-star.edu.sg/)62 and PolyPhen2 (http://genetics.bwh.harvard.edu/pph2/)63 tools were chosen for comparisons, as these tools were also useful to predict the effects of artificial mutations for both hOtolC1q and dOtolC1q. Additionally, for all investigated mutations, the SNP MuSiC (https://soft.dezyme.com/) tool was used to predict the effects on protein stability64. Models of gC1q trimers, which were used as a template, were based on already published EOM analysis conducted on the basis of SAXS data33. Default parameters were used in all predictions. The structure model was visualized using VMD software (University of Illinois, https://www.ks.uiuc.edu/Research/vmd/)65. Sequences of otolin-1 from selected fish, birds and mammals were aligned using Clustal X66 and visualized using Jalview67.

### Preparation of mutated gC1q genes

Genes coding the mutants of hOtolC1q: E417A, D425A and E417A D425A and of dOtolC1q: D432A, D440A and D432A D440A were prepared by amplifying the synthetic cDNAs of hOtolC1q and dOtolC1q in two parts using Phusion DNA polymerase. PCR products were digested using the Esp3I (BsmBI) restriction enzyme, which cleaves DNA 5 bp away from the recognition site. Cohesive ends of the cleavage products contained codons, which introduced the mutations. The gene fragments were subcloned into the pQE-80 L vector using BamHI and HindIII restriction sites. In the case of hOtolC1q E470A, synthetic cDNA of hOtolC1q subcloned into the pQE-80 L plasmid expression vector33 was used as a template in modified QuickChange®, which was conducted as described68. One-fusion DNA polymerase was used in the mutagenic PCR. For the calculation of the annealing temperatures of the primers, the concentration of KCl in the reaction mixture was assumed to be 0.1 M, as in the assay buffer of the polymerase. Plasmids were propagated in E. coli TOP10 cells. The progress of the cloning was followed by agarose gel electrophoresis. All mutated genes were analysed by DNA sequencing (Genomed).

### Protein expression and purification

Escherichia coli BL21(DE3) cells were chemically transformed by heat shock and grown overnight on plates containing LB broth with 1.5% agar and 100 µg/mL carbenicillin at 37 °C. Single colonies were picked and used to inoculate starter cultures containing 100 ml of TB broth with carbenicillin, which were incubated overnight at 37 °C and 200 rpm. Portions of 500 ml TB with carbenicillin were inoculated with a 2% volume of starter culture and incubated at 29 °C and 200 rpm. After reaching the optical density at 600 nm of 0.5, cultures were cooled to 15 °C, and the expression of the protein of interest was induced by 0.5 mM IPTG. The culture was continued overnight (16–18 h) at 15 °C and 200 rpm. Cells were collected by centrifugation at 5000×g at 4 °C for 15 min and resuspended in H10Na500G5 buffer [HEPES 10 mM, pH 7.0 (20 °C), NaCl 500 mM, glycerol 5% (v/v)] with 1 mM 2-mercaptoethanol and 1 mg/ml lysozyme. Suspensions of cells expressing hOtolC1q E470A were additionally supplemented with 0.1% NP40 substitute (Amresco, currently VWR, Gliwice, Poland). The cells were kept frozen at − 80 °C.

Cell lysis was initiated by thawing in a room-temperature water bath. After thawing, 0.2 mg/mL PMSF, 20 µg/mL DNase I and 20 µg/mL RNase A were added. Lysis was facilitated by 10 cycles of ultrasonic treatment for 30 s with 1-min breaks in a Cole-Parmer CPX 500 ultrasonic processor with a microtip and amplitude set at 35% (Cole-Parmer, Vernon Hills, IL, USA). The cell suspension was kept on ice to maintain the temperature below 10 °C. Lysates were clarified by centrifugation at 18,500×g for 30 min at 4 °C and incubated with 1 mL bed volume of TALON® Metal Affinity resin for 1 h in a cold room (4–6 °C) in an orbital mixer set at 5 rpm. The resin was separated by centrifugation at 700×g for 5 min at 4 °C, washed with 20 bed volumes of H10Na500G5, centrifuged again and packed in a glass Tricorn column. The column was connected to ÄKTA Avant chromatography system (GE Healthcare Life Sciences) with a flow rate of 1 mL/min. Contaminants were washed away with 20 bed volumes of H10Na500G5 and subsequently with 20 bed volumes of the buffer with 30 mM imidazole. The protein of interest was eluted with buffer containing 200 mM imidazole. The eluate was concentrated in Amicon Ultra centrifuge filters with a 10 kDa cut-off (Merck) and subjected to gel filtration using a Superdex 200 Increase 10/300 GL column operated at 0.75 mL/min (hOtolC1q and its mutants) or Superdex 75 16/60 Prep Grade operated at 1.5 mL/min (dOtolC1q and its mutants) with H10Na500G5 as a mobile phase. Eluate before void volume of the column was collected for use as a reference in further studies. Fractions containing proteins were collected and analysed using SDS-PAGE with an acrylamide percentage of 4% in a stacking gel and 12% in a resolving gel in a Laemmli buffer system (Tris–glycine-SDS)69. The identity of the protein was confirmed by Western blotting with murine 6xHis monoclonal IgG as the primary antibody and HRP horse anti-mouse IgG antibody as the secondary antibody. Pure protein samples were stored at − 80 °C.

### Tb3+ binding fluorescence

The binding of Tb3+ ions to hOtolC1q, dOtolC1q and their mutants was assessed using steady-state fluorescence. Terbium(III) chloride was dissolved in Milli-Q water to a final concentration of approximately 0.5 M. The exact concentration of TbCl3 was determined by titration of diluted stock solution with EDTA in the presence of xylenol orange. Aliquots of diluted TbCl3 were added to 2 ml 3.7 µM protein solution in a 10 × 10 mm quartz cuvette and incubated for 15 min at room temperature. Subsequently, fluorescence emission at 520–580 nm was recorded using an excitation wavelength of 280 nm using a Fluorolog-SPEX fluorimeter (HORIBA Scientific, Jobin–Yvon, Kyoto, Japan) equipped with a Peltier temperature control accessory set at 20 °C. The bandwidth was set at 5 nm for both excitation and emission monochromators. The fluorescence intensities obtained were processed and fitted to a model based on work by Gonzalez et al.70,71. Fluorescence intensity was corrected for sample dilution and background fluorescence using the following equation:

$${I}_{corr}=\frac{{I}_{max}}{{I}_{max-15}} \cdot \frac{{v}_{0}+{v}_{i}}{{v}_{0}}$$
(1)

where, Icorr—corrected fluorescence, Imax—fluorescence at emission maximum (here: 545 nm), Imax–15—fluorescence at local emission minimum (here: 530 nm), v0—initial sample volume, vi—added volume of working solution of TbCl3.

Corrected fluorescence was fitted to a single binding site model to determine the apparent dissociation constant Kd:

$${I}_{corr}={I}_{0}+\frac{{K}_{d}+{c}_{p}+{c}_{t}-\sqrt{{\left({K}_{d}+{c}_{p}+{c}_{t}\right)}^{2}-4 \cdot {c}_{p} \cdot {c}_{t}}}{a}$$
(2)

where, I0—corrected fluorescence intensity at 0 µM Tb3+ (fitted), cp—concentration of the protein, ct—concentration of Tb3+, a—proportionality constant (fitted).

Fitting of the data to the model above was conducted using Origin Pro 9.0 software (OriginLab, Northampton, MA, USA).

### Circular dichroism

Circular dichroism of 0.2 mg/mL proteins in H10Na500G5 with 1 mM EDTA, 0.1 mM CaCl2, 1 mM CaCl2, 10 mM CaCl2 or 100 mM CaCl2 was measured in 1 mm quartz SUPRASIL® cuvettes (Hellma Analytics, Müllheim, Germany) using a Jasco J-815 spectropolarimeter (Jasco, Easton, MD, USA) with a Peltier temperature control accessory set at 20 °C. The proteins were incubated with EDTA/CaCl2 at room temperature for at least 1 h before the measurements. The spectra were collected between 200 and 260 nm every 1 nm at a scanning speed of 50 nm/min with five accumulations. Data for which the photomultiplier voltage was below 600 V were analysed. CD spectra of the proteins were corrected for buffer background signal and normalized for protein composition and concentration using the equation below72:

$${\theta }_{mrw}=\frac{\theta \cdot MRW}{10 \cdot c \cdot l} \left[\frac{deg \cdot {{cm}^{2}}}{dmol}\right]$$
(3)

where $${\theta }_{mrw}$$ is the mean residue ellipticity, $$\theta$$—ellipticity [degrees], MRW—mean residual weight of a protein [g/mol], c—protein concentration [g/L] and l—optical pathlength of a cuvette [cm].

### Analytical ultracentrifugation

Sedimentation velocity analytical ultracentrifugation (SV AUC) was conducted in a Beckman Coulter ProteomeLab XLI analytical ultracentrifuge (Beckman Coulter, Brea, CA, USA) with an An60Ti rotor and assembled cells with two-channel 12 mm charcoal filled Epon® centerpieces and quartz windows. The proteins were analysed at approximate concentrations of 0.1, 0.25 and 0.5 mg/ml in H10Na500G5 cells with 1 mM EDTA or 10 mM CaCl2. Assembled cells with the samples were preincubated in an ultracentrifuge for 3 h at 20 °C and then centrifuged at 50,000 rpm (approximately 200,000×g at the bottom of the cell) overnight. Scans of absorbance at 280 nm were collected continuously with 0.003 cm resolution. The scans representing the whole sedimentation process were time-corrected73 and analysed in SEDFIT (available at https://sedfitsedphat.nibib.nih.gov/) using a continuous c(s) distribution model74 with at least 20 points per 1 S. Partial specific volumes of the proteins, densities and dynamic viscosities of the solvents were calculated using SEDNTERP (available at http://www.jphilo.mailway.com/download.htm). Maximum entropy regularization with p = 0.95 was used. Simplex and Marquardt–Levenberg algorithms were used until the RMSD value converged. Among the results of the calculations were sedimentation coefficients (s), sedimentation coefficients corrected for water at 20 °C (s20,w), weight-averaged sedimentation coefficients ($$\overline{{s_{{20,w}} }}$$), apparent molecular weights (MWapp) and frictional ratios (f/f0). c(s) distributions were visualized using GUSSI (available at https://www.utsouthwestern.edu/labs/mbr/software/)75 and Origin Pro 9.0 software.

### Differential scanning fluorimetry

Differential scanning fluorimetry assays were conducted as described76. Five micromolar solutions of the proteins in H10Na500G5 supplemented with 10 × SYPRO Orange (500-fold dilution of delivered stock solution) were measured in the presence of 1 mM EDTA, 0.1 mM CaCl2, 1 mM CaCl2, 10 mM CaCl2 or 100 mM CaCl2. Additionally, hOtolC1q and its mutants were measured in the presence of a sevenfold molar excess of TbCl3, dOtolC1q and its mutants in the presence of a 14-fold excess of TbCl3. Final sample volume was 20 µl. The samples were aliquoted into a 96-well plate in triplicate, covered with optically clear foil and incubated at room temperature for at least 1 h before the measurements. Fluorescence of SYPRO Orange was measured using an Applied Biosystems ImageQuant5 qPCR thermal cycler (Thermo Fisher Scientific) with optical filters set as x1 - m3 (excitation at 470 ± 15 nm, emission at 587 ± 10 nm) between 20 and 99 °C during heating at 0.033 °C/s. The data were analysed using Protein Thermal Shift software (Thermo Fisher Scientific). Tm were determined from the derivative of fluorescence with increasing temperature (dF/dT).