The gene encoding wild-type Int8 (residues 341–419 of herstatin) (Shamieh et al., 2004) was constructed by overlap-extension PCR. To improve solubility, the N-terminus of Int8 was tagged with a 5 × Lys-tag using a Gly-Ser-Ser-Gly linker (MGKKKKKGSSG) (Kato et al., 2007). Additionally, a 6 × His-tag was linked to the C-terminus using a Ser-Ser-Gly linker (SSGHHHHHH) to facilitate protein purification. Codons were optimized for high-level expression in Escherichia coli using the DNAWorks server (Hoover and Lubkowski, 2002). The Int8 DNA fragment was ligated into the NcoI/BamHI sites of the pET-15b vector (MilliporeSigma, Burlington, MA, United States). The R371I mutation was introduced according to the method of the QuikChange site-directed mutagenesis kit (Agilent Technologies, Santa Clara, CA, United States). The expression levels and solubilities of the wild-type and mutant Int8 proteins were predicted using the ESPRESSO server (Hirose and Noguchi, 2013).

Unlabeled and ¹⁵N-labeled Int8, as well as the unlabeled R371I mutant of Int8 were expressed in E. coli BL21(DE3) cells (Nippon Gene, Tokyo, Japan) in 2×YT medium (for unlabeled protein) or M9 minimal medium (for ¹⁵N-labeled protein) containing ampicillin (50 μg/ml). The cells were incubated at 37°C and overexpression of Int8 was induced by addition of 1 mM isopropyl β-D-1-thiogalactopyranoside at an optical density of 0.8 at 600 nm. After incubation for an additional 5 h, cells were collected, resuspended in denaturation buffer containing 20 mM Tris-HCl (pH 8.0), 6 M guanidine hydrochloride (GdnHCl), and 20 mM imidazole, and sonicated on ice for 4 min using a Branson Sonifier 250D Advanced (Branson, Danbury, CT, United States). The lysate was subsequently centrifuged at 35,140 × g for 30 min at 4°C. The supernatant was filtered through a 0.45-µm pore size membrane filter and applied to a column containing nickel-nitrilotriacetic acid agarose gel (Qiagen, Hilden, Germany). The column was washed with the denaturation buffer and wash buffer containing 20 mM Tris-HCl (pH 8.0), 500 mM NaCl, and 20 mM imidazole. Int8 was eluted using elution buffer containing 20 mM Tris-HCl (pH 8.0), 500 mM NaCl, and 0.1–1 M imidazole. For CD and light scattering measurements, the eluate was further purified by size-exclusion chromatography using a Superdex 200 pg column (Cytiva, Marlborough, MA, United States) with the buffer containing 10 mM sodium phosphate (pH 6.0) and 50 mM NaCl. For NMR and SAXS measurements, the eluate was desalted using a PD-10 column (Cytiva) in buffer containing 10 mM sodium phosphate (pH 6.0) and 50 mM NaCl. The purity of the Int8 proteins was analyzed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis.

Protein concentrations were obtained using the following four methods: 1) absorption measurement at 280 nm in the presence of 6 M GdnHCl (Edelhoch, 1967; Gill and von Hippel, 1989), 2) absorption measurement at 280 nm under native conditions (Pace et al., 1995), 3) absorption measurement at 205 nm (Anthis and Clore, 2013), and 4) the Pierce BCA protein assay kit (Thermo Fisher Scientific, Waltham, MA, United States). Similar protein concentrations were obtained by all methods.

Far-UV CD spectra were obtained from a J-805 spectropolarimeter (Jasco, Tokyo, Japan) at 200–250 nm using a quartz cuvette of 1 mm path length at 25°C. The temperature was controlled using a thermostat-circulating water bath. The protein concentrations of the wild type and R371I mutant of Int8 were 30 and 31 μM, respectively. The samples additionally contained 10 mM sodium phosphate (pH 6.0) and 50 mM NaCl, in the presence or absence of 4 M GdnHCl. Mean residue ellipticity (MRE) was calculated as previously described (Arai and Iwakura, 2005). The spectra in the presence of GdnHCl were measured at 211–250 nm. The helix content (f _H) was estimated from the MRE at 222 nm according to the following equation (Chen et al., 1972): f H   =   - ( MR E 222nm + 2340 ) × 100 / 30300 (1)

One-dimensional (1D) pulsed-field gradient (PFG) NMR spectra and two-dimensional (2D) ¹H-¹⁵N heteronuclear single quantum coherence spectra were recorded using a Bruker Avance500 spectrometer in a buffer containing 10 mM sodium phosphate (pH 6.0), 50 mM NaCl, 10% D₂O, and 0.2 mM sodium 3-(trimethylsilyl)-1-propanesulfonate (DSS) at 25°C. All spectra were analyzed using NMRPipe (Delaglio et al., 1995) and NMRViewJ softwares (Johnson and Blevins, 1994). The protein concentrations of the wild-type and R371I mutant of Int8 were 350 and 150 μM, respectively.

PFG NMR measurements were performed using a bipolar longitudinal eddy-current-decay pulse sequence (Wu et al., 1995; Nakamura et al., 2013; Kunihara et al., 2019). The samples contained 0.05% dioxane as a standard for molecular size analysis. The diffusion delay was 100 ms, the gradient pulse width was 6 ms, and the pulse separation was 0.6 ms. The spectral center was 4.701 ppm, and the spectral width was 8012.820 Hz. Five spectra for the peak decay of dioxane and 15 spectra for Int8 were acquired while changing the strength of the diffusion gradient (g) from 5.35 G/cm (5%) to 53.5 G/cm (100%). The peak intensity (s) is related to g as follows: s s 0 = exp ( − d ⋅ g 2 ) (2) where s ₀ is the peak intensity at 0% field gradient and d is the observed decay rate, which is proportional to the diffusion coefficient of the solute. The signal decay was fitted using Eq. 2, and the hydrodynamic radius, R _h of Int8 was calculated using the d value, as follows: R h ( Int 8 ) = d ( dioxane ) d ( Int 8 ) R h ( dioxane ) (3) where the R _h of dioxane is 2.12 Å (Wilkins et al., 1999).

SAXS measurements were conducted at beamline (BL)-10C at the Photon Factory of the High Energy Accelerator Research Organization (KEK), Tsukuba, Japan. The camera length (1,996 mm) was calibrated by the diffraction of silver behenate using FIT2D software (Hammersley, 1997). Samples were loaded into a mica-windowed cuvette with a 1 mm path length and were irradiated by a monochromatic X-ray beam (1.488 Å). Int8 concentration was 4.1 mg/ml (390 μM), and sample was centrifuged for 30 min at 4°C before measurement. The temperature in the cuvette was maintained at 25°C using a circulating water bath. Scattering images were acquired by a PILATUS3 300 KW detector (DECTRIS Ltd., Baden, Switzerland) at 10 × 1 s exposures. The scattering intensity, I(Q), was collected from 0.005 to 0.27 Å⁻¹, where Q = 4π sin (θ/λ) (λ, wavelength; 2θ, scattering angle). Scattering of blanks (buffer alone) was measured before and/or after measurement of protein samples. Circular averaging of 2D scattering data was performed using FIT2D. Blanks were subtracted from protein solution scattering data to obtain the scattering profile of protein molecules. The subtracted scattering data were binned per four data points to increase the signal-to-noise ratio. A radius of gyration, R _g, was obtained based on the Guinier approximation within the Guinier region (R _g Q < 1.3) (Trewhella et al., 2017): ln ⁡ I ( Q ) = ln ⁡ I ( 0 ) − R g 2 3 Q 2 (4) where I(0) denotes the zero-angle scattering intensity (Glatter and Kratky, 1982; Arai et al., 2007). An R _g was also estimated using the approximation of the Debye function for a random coil within the range of (R _g Q)² < 3: I ( Q ) − 1 = a [ 1 + 0.359 ( Q R g ) 2.206 ] (5) where a is the y-intercept of the I(Q)⁻¹ versus Q ^2.206 plot (Calmettes et al., 1994). The pair-distance distribution function, P(r), was calculated using GNOM software (Svergun, 1992).

The SAXS data were further analyzed using the ensemble optimization method (EOM) 2.1 (Bernado et al., 2007; Tria et al., 2015; Vallet et al., 2018). In this analysis, a pool of 10,000 random conformations based on the Int8 amino acid sequence was first generated. Then, a genetic algorithm was run 100 times to select for an ensemble of conformations that best reproduces the experimental scattering curve.

Size-exclusion chromatography (SEC) measurements were performed using a high-performance liquid chromatography (HPLC) system (LP-20AP, Shimadzu, Kyoto, Japan) and a Superdex 200 increase 3.2/300 column (Cytiva). The column was pre-equilibrated with 10 mM sodium phosphate (pH 6.0) and 150 mM NaCl, and samples of the wild-type and R371I mutant of Int8 were injected onto the column. The protein concentrations were 129 and 515 μM for the wild type and 138 and 550 μM for the R371I mutant. Molecular weight of the sample was determined on the HPLC system coupled to a Viscotek TDA 305 light scattering detector (Malvern Panalytical, Malvern, United Kingdom) as described previously (Chang et al., 2020). Bovine serum albumin (66.7 kDa) was used as a standard for instrument calibration. Data analysis for estimating molecular weights was performed using OmniSEC software (Malvern Panalytical). The error in molecular weight was calculated from duplicate or triplicate measurements.

The isolated Int8 domain containing a C-terminal 6 × His-tag and an N-terminal 5 × Lys-tag was overexpressed and purified from E. coli, to characterize its structure (Section 2). The structure of Int8 under native conditions was measured by CD, NMR, and SAXS. The far-UV CD spectrum of Int8 under native conditions had small intensities at ∼222 nm but had a minimum at ∼200 nm, indicating a largely disordered structure (Figure 2A). The CD intensity at ∼222 nm was reduced to almost zero after protein unfolding by the addition of 4 M GdnHCl (Figure 2A). The difference spectrum obtained by subtracting the CD spectrum in the presence of 4 M GdnHCl from that measured under native conditions showed a minimum at 222 nm, indicating the presence of an α-helical structure (Figure 2B). These findings suggest that Int8 is largely disordered under native conditions, but retains a residual helical structure. The helix content as estimated from the CD intensity at 222 nm is ∼4% (Eq. 1 in Section 2).

1D and 2D NMR spectra of Int8 showed that chemical shifts of amide protons were confined in a narrow range of 7.7–8.7 ppm (Figures 3A,E). This indicates that the amide protons of Int8 are in similar environments, suggesting that Int8 is predominantly in a disordered state. PFG NMR measurements were performed to characterize the hydrodynamic radius, R _h. From the 1D NMR spectra of Int8, four peaks were selected at ∼1–2 ppm (Figure 3A) and peak intensities were plotted depending on the external magnetic field gradient, g (Figure 3B). By fitting the decay curves (Eq. 2 in Section 2), a parameter d was estimated, which is proportional to the diffusion coefficient. The d values were found to be almost identical for the four selected peaks. Using the Einstein-Stokes equation and the d value of the standard substance dioxane (R _h = 2.12 Å), which was included in the NMR sample, an R _h of 33 ± 2 Å for Int8 was obtained (Eq. 3 in Section 2).

SAXS measurements were performed to characterize the molecular size and shape of Int8 (Figure 4A). Scattering intensities, I(Q), at scattering vectors, Q (Å⁻¹), were analyzed by a Kratky plot [I(Q)Q ² versus Q plot] which provides information on the molecular shape of a protein (Glatter and Kratky, 1982; Arai et al., 2007). The presence of a peak in the plot indicates that the protein has a compact, globular shape, whereas the absence of a peak is indicative of a random coil-like disordered structure (Arai et al., 2007). The Kratky plot of Int8 lacked a peak and instead showed a plateau, indicating that Int8 has a random coil-like disordered structure under native conditions (Figure 4B). The pair-distance distribution function, P(r), calculated by Fourier transformation of I(Q), showed a peak at ∼20 Å, but the maximum distance, D _max, between two atoms in Int8 was ∼80 Å (Figure 4C). It is known that compact globular particles have a symmetric bell-shaped P(r), whereas unfolded particles have an extended tail in the P(r) function (Kikhney and Svergun, 2015). Thus, the P(r) of Int8 suggests an extended structure.

Next, the SAXS data were analyzed by generating a Guinier plot (i.e., ln I(Q) versus Q ² plot), which provides information on the molecular size of a protein, that is, the radius of gyration R _g, using the Guinier approximation (Eq. 4 in Section 2). The R _g of Int8 was estimated to be 26.2 ± 0.5 Å (Figure 4D). Furthermore, the R _g estimated using the Debye function for a random coil was 29.9 ± 0.5 Å (Figure 4E; Eq. 5 in Section 2). Since the Debye function is applicable to the scattering data at wider angles (R _g Q < 1.73) than the Guinier approximation (R _g Q < 1.3), the R _g value obtained by the Debye function is less affected by interparticle interference effects (Calmettes et al., 1994). Therefore, we used an R _g of 29.9 Å for Int8. The R _g values of typical native, intermediate (molten globule), and unfolded states of globular proteins are related to the number of residues, N, as R _g = R ₀ N ^ν , where R ₀ is the R _g of a single amino acid residue and ν is the scaling exponent (Wilkins et al., 1999; Kohn et al., 2004; Arai et al., 2007). The R ₀ and ν values are 3.68 ± 0.86 Å and 0.29 ± 0.02 for the native state (Wilkins et al., 1999), 3.68 ± 0.45 Å and 0.35 ± 0.09 for the molten globule-like folding intermediate (Arai et al., 2007), and 1.927 ± 0.271 Å and 0.598 ± 0.028 for the fully unfolded state (Kohn et al., 2004), respectively. A comparison of these values showed that the R _g of Int8 is almost the same as that of the fully unfolded state (Figure 4F).

The R _g/R _h ratio is known to depend on the molecular shape of proteins. For spherical molecules (either folded proteins or disordered proteins in compact conformations) the R _g/R _h ratio is 3 / 5 ∼ 0.78, whereas for proteins in the random-coil-like conformations it is 1.2–1.6 (Nygaard et al., 2017). The R _g/R _h ratio for Int8 was 29.9/33 = 0.91, suggesting that Int8 does not have a spherical shape, consistent with the P(r) function, but is not in a fully random-coil-like state.

Then, the ensemble of Int8 conformations that best fitted to the SAXS data was modeled by EOM (Bernado et al., 2007; Tria et al., 2015; Vallet et al., 2018). The ensemble best fitted to the scattering curve (Figure 4A; χ² = 1.799) contained five representative conformations of Int8 with an averaged R _g of 28.5 Å and D _max of 79.2 Å, which are consistent with the above analysis (Figure 5). The distributions of R _g and D _max for Int8 were smaller than those for the completely random pool (Figures 5A,B). Furthermore, R _flex, indicative of the flexibility of the conformational ensemble, was 71.7% for Int8, which was smaller than that for the random pool (86.2%). Thus, the SAXS data suggest that Int8 has a molecular size comparable to the fully unfolded state but is less flexible than random coils probably due to the presence of residual helical structure.

Secondary structure prediction of Int8 using the PSIPRED server (Buchan and Jones, 2019) showed that most regions of Int8 were in a random coil-like structure (Figure 6A). Only a small number of residues in the middle of Int8 were predicted to be organized into two α-helices [helix 1 (residues 366–370) and helix 2 (residues 374–381)] and one β-strand (residues 410–413). The AGADIR server (Muñoz and Serrano, 1994; Lacroix et al., 1998) predicted that helical propensity was low for both helices, although helix 2 (residues 373–382) had a higher helical propensity than helix 1 (residues 365–369) (Supplementary Figure S1). These predictions are consistent with the CD measurements that showed the presence of only a small amount of helical structure in Int8.

Disorder prediction was performed using nine prediction servers, namely SPOT-Disorder2 (Hanson et al., 2019), MFDp2 (Mizianty et al., 2013), MetaDisorderMD2 (Kozlowski and Bujnicki, 2012), DISOPRED (Jones and Cozzetto, 2015), PrDOS (Ishida and Kinoshita, 2007), IUPred2A (Meszaros et al., 2018), DisProt(VSL2B) (Obradovic et al., 2005), DisEMBL (Linding et al., 2003), and POODLE-I (Shimizu, 2014). Although the results obtained differed between individual prediction servers, many predicted that most regions of Int8, particularly in the N- and C-terminal regions were intrinsically disordered (Figure 6B), which was consistent with the secondary structure predictions. The disordered nature of Int8 is probably due to the presence of many proline and serine residues between residues 346–365 and 384–406. Thus, theoretical predictions suggest that Int8 is intrinsically disordered with a residual helical structure, consistent with our experimental results (Figures 2–5).

The structure of the R371I mutant of Int8, which does not bind HER2 (Shamieh et al., 2004), was predicted by PSIPRED and the disorder prediction servers to be similar to that of the wild-type Int8 (Supplementary Figure S2). The AGADIR server also predicted that the helical propensity of the R371I mutant is almost the same as that of the wild-type Int8 (Supplementary Figure S1).

The R371I mutant of Int8 was overexpressed in E. coli and purified. The CD spectrum of the R371I mutant of Int8 was found to be similar to that of wild-type Int8 (Figure 2A). The helix content of the mutant was estimated to be ∼5% (Eq. 1 in Section 2). The difference CD spectrum between those measured in the absence and presence of 4 M GdnHCl had a negative peak at ∼222 nm in the R371I mutant (Figure 2B). These results indicate that the mutation little affected the helix content of Int8, which is consistent with the theoretical predictions (Supplementary Figure S1).

The 1D NMR spectrum of the R371I mutant showed that peaks for amide protons are confined in a narrow range of 7.7–8.7 ppm (Figure 3C), indicating disordered structures. The molecular size of the R371I mutant of Int8 was estimated by PFG NMR measurement. Three peaks at ∼1–2 ppm in the 1D NMR spectra of the mutant were used for the analysis of peak intensity decays (Figures 3C,D). The R _h for the R371I mutant was 72 ± 2 Å. This value is more than two-fold larger than that of wild-type Int8, indicating the formation of soluble aggregates in the mutant. The aggregation was observed despite the use of a lower protein concentration for the R371I mutant (150 μM) than for the wild-type Int8 (350 μM). Therefore, these results indicate that the R371I mutant of Int8 is prone to aggregation.

To further investigate the aggregation propensity of the mutant in more detail, we performed SEC measurements for the wild type and R371I mutant of Int8 (Figure 7). The elution profile of the wild type showed a single peak without aggregates at both low (∼100 μM) and high protein concentrations (∼500 μM). The molecular weights estimated by static right-angle light scattering were 12.6 (±0.1) kDa and 13.2 (±0.1) kDa for the low and high concentration samples of the wild-type Int8, respectively (Supplementary Figure S3), both of which are close to the value expected for an Int8 monomer (10.5 kDa). In contrast, a large amount of aggregates were observed at a high concentration (∼500 μM) of the R371I mutant (Figure 7B) with a molecular weight of 20.3 (±0.5) kDa at the elution peak (Supplementary Figure S3). At a low protein concentration (∼100 μM), the formation of aggregates was suppressed (Figure 7A), and the molecular weight at the elution peak was of 11.3 (±0.6) kDa, which is close to the value for an Int8 monomer (Supplementary Figure S3). However, the mutant started to elute earlier than the wild type, indicating the presence of slight aggregates (Figure 7A). The large R _h of the mutant determined by the PFG NMR measurements at 150 μM may correspond to the molecular size of these aggregates. Taken together, these results clearly demonstrate that the R371I mutant of Int8 is prone to aggregation.

Three-dimensional structure prediction of the wild type and R371I mutant of Int8 was performed using the standalone version of AlphaFold2 (Jumper et al., 2021). The predicted structure of wild-type Int8 was largely disordered, but had two α-helices at residues 365–370 and 374–382 corresponding to helices 1 and 2, respectively (Figures 8A,C). The predicted structure of the R371I mutant was similar to that of the wild type (Figure 8B), except that helix 1 (residues 368–370) was shorter (Figure 8D), which may be consistent with the slightly reduced helical propensity predicted by AGADIR for the mutant (Supplementary Figure S1).