Protein sequences were obtained from the following sources: Hemoglobin subunit alpha (Homo sapiens) (NCBI Reference Sequence: NP_000549.1), hemoglobin subunit beta (Homo sapiens) (NCBI Reference Sequence: NP_000509.1), hemoglobin subunit gamma (Homo sapiens) (Swiss-Prot: P68068.2), hemoglobin subunit delta (Homo sapiens) (GenBank: ABK79072.1) and Peroxiredoxin2 (Homo sapiens) (GenBank: AAH00452.1). The original PPI data of humans used in this work was downloaded from the Database of Interacting Proteins (DIP) (http://dip.doe-mbi.ucla.edu/) (32). The corresponding protein sequence information was extracted from the Universal Protein Resource (UniProt) (33). There were 1,962 protein pairs (involving 1,703 proteins) in both the positive and negative sets. The prediction of PPI was based on protein sequences. A descriptor using a conjoint quintuple (5-mer), which regarded any five continuous amino acids as a unit, was first proposed by us (34). The classification of amino acids and IDQD analysis was introduced in our previous work. The sensitivity (Sn), specificity (Sp), false positive rate (FPR), total accuracy (TA), correlation coefficient (CC), receiver operator characteristics (ROC) and the area under ROC curve (auROC) were used to access the accuracy of the PPI predictions. A three-fold cross-check is combined with the above four indicators, as well as the ROC curve and auROC, to evaluate the predictive power of the IDQD algorithm. The specific method is as follows: The data set is randomly divided into 3 parts, each of which contains an equal amount of positive and negative set data. Two of them were used as training sets, and the remaining one was used as a test set. The rotation was carried out three times in turn, and the average of the various indicators was used to evaluate the model performance.

All the experimental protocols of this study were approved by the Ethics Committee of Baotou Medical College. Fresh, normal anti-coagulated blood was collected from volunteers. Before the blood samples were collected, written consent was obtained from all of the volunteers. Then, the RBC suspensions and hemolysates were prepared in accordance with the approved guidelines (35).

A two percent starch-agarose mixed gel (starch: agarose=4:1) was prepared with 1× TEB buffer (pH 8.6) (36). Eight microliters of RBC suspension and hemolysate were loaded into the gel, and electrophoresis was performed at 5 V/cm for approximately 2 hours. After the electrophoresis, the red bands of RBC-HbA (RA), hemolysate-HbA (HA), RBC-HbA-A₂ (RA-A₂), hemolysate-HbA-A₂ (HA-A₂), RBC-HbA₂ (RA₂) and hemolysate-HbA₂ (HA₂) were cut out separately and frozen at -80°C for at least 30 min. Before use, the gel was taken out and thawed at room temperature. After centrifuging at 10,000 g for 10 min, the supernatants were pipetted into new Eppendorf tubes and concentrated with Sephadex G25.

Seven microliters of PMSF and 20 μL of agarose conjugate hemoglobin Beta (37-8) (Santa Cruz, sc-21757AC) were added into 700 μL of HA and RA samples, respectively. A negative control was prepared by adding 7 μL of PMSF and 20 μL of agarose conjugated mouse IgG into 700 μL HA and RA solutions, respectively. After incubating the samples overnight at 4°C on a flat platform, they were centrifuged at 10,000 g for 1 min at 4°C. Next, the beads were collected, and the supernatants were removed carefully. The beads were washed with 700 μL 1×PBS (pH 7.4) buffer six times and 0.1×PBS (pH 7.4) buffer one time. Forty microliters of 1× loading buffer was added into each sample, which were then heated at 100°C for 5 min. After centrifuging at 10,000 g for 5 min, the supernatant was ready for SDS-PAGE.

The samples of supernatant (20 μL) prepared by the co-IP complex procedure described above were separated by 5-12% SDS-PAGE (36). After electrophoresis, the gel was stained overnight with 0.08% Coomassie Brilliant Blue G250 and was destained in Milli-Q water until the background staining of the gel was low.

The desired gel bands were excised into 1 mm³ pieces and further destained and digested according to a routine protocol (36). The digested peptide samples were extracted with 50-100 mL of 5% trifluoroacetic acid (TFA) solution at 40°C for 1 hour and later extracted with the same volume of 50% CH₃CN/2.5% TFA solution at 30°C for 1 hour. Finally, the samples were extracted ultrasonically with 50 mL of CH₃CN solution. The extracted solutions were pooled and dried in a SpeedVac vacuum dryer and resuspended in 3-5 mL of 0.1% formic acid for MS analysis.

The peptide mixture was analyzed on a nanoliquid chromatography tandem mass spectrometry (LC-MS/MS) system, which consists of an Ultimate HPLC system (Dionex) and a quadrupole time of flight (Q-TOF) mass spectrometer (micrOTOF-Q II mass spectrometer, Bruker) equipped with a nano-ESI source. The MS/MS data were processed using Data Analysis 4.0 and then searched against the Swissprot protein sequence databases by MASCOT (http://www.matrixscience.com). The TOF mass analyzer of the instrument was calibrated by using 10 mM Na TFA (pos). For the MS/MS ion search, a peptide charge state of +2, +3 or +4 and peptide/fragment mass tolerance of ±0.1 Da was used. Probability-based MASCOT scores were estimated by comparing the search results against the estimated random match population and reported as −10 × log (P), where P is the absolute probability. The significance threshold was set at P < 0.05.

Twenty microliters of HA, RA, HA₂ and RA₂ were first separated by 5-12% SDS-PAGE, and then the protein was transferred to the PVDF membrane. The PVDF membrane was subsequently blocked in 5% skim milk for 1 hour at room temperature and then incubated with rabbit polyclonal anti-Prx2 antibody (Abcam, ab86295) at a 1:1000 dilution overnight at 4°C. Next, the membranes were hybridized with anti-rabbit IgG-conjugated horseradish peroxidase (Santa Cruz, sc-2004) at a 1:2000 dilution for 1 hour at room temperature. Finally, the protein signals were detected by the Pierce ECL Western Blotting Substrate and visualized by a Tanon 4200 (Tanon Science & Technology Co., Ltd.).

HA samples were collected from the starch-agarose gel by the freeze-thaw method after electrophoresis of the hemolysate (30). Peroxiredoxin 2 was purchased from Abcam (Abcam, ab8). XAFS experiments were carried out at the National Synchrotron Radiation Laboratory (USRL), University of Science and Technology of China. The XAFS data of the iron K-edge was collected at the U-7C end station in transmission mode. The electron energy in the storage ring was 0.8 GeV, and the ring current was approximately 160 mA. Higher harmonics were reduced sufficiently by detuning the double-crystal Si(III) monochromator at the beam line. The raw data were analyzed by using the NSRLXAFS 3.0 program. The basic EXAFS data, χ (k), was fitted to the following equation:

${\chi }_j\left(k\right)=\frac{S_0^2{N}_j}{k{R}_j^2}{F}_j\left(\pi ,k\right)\exp \left(-2k^2\Delta {\sigma }_j^2\right)\exp \left(-2R_j/{\lambda }_j\right)\sin \left(2k{R}_j+{\varphi }_j+{\Sigma }_j\right)$ (1)

and

${\Sigma }_j=-4\frac{\Delta {\sigma }_j^2}{R_j}K-\frac{4}{3}{\sigma }_j^3{k}^3$ (2)

where k is the photoelectron wavenumber, S₀ is the amplitude reduction factor, N_j is the coordination number of the j^th shell, R_j is the distance between the coordination atoms and central atoms in the shell, F_j (π,k), λ_j and ф_j are the _j^th atom backscattering factors, Δσ_j² is the Debye-Waller factor, and σ_j³ is the third cumulant (37).

In our approach, the PPI prediction model was constructed based on the training set. Three training sets and three testing sets were prepared by the sampling method described previously (34). Each training set consisted of 2,616 protein pairs, with half of the protein pairs randomly selected from the data of positive PPI pairs and the other half randomly selected from the negative protein pairs. Each testing set was constructed with another 1,308 protein pairs. Thus, three prediction models were generated for the three training sets of data. The best prediction results are listed in Tab 1, and the ROC curves are shown in Figure 1. The performance of the IDQD modules for the prediction of the PPIs in human samples was trained on different types of tests. The values showed that the performance was at the best threshold (ξ0). The ξ₀ of the self-consistency test and the 3-fold cross-validation are –0.45 and –0.08, respectively. The PPI prediction results between each two Hb subunits (Alpha, Beta, Gamma, Delta) and between each Hb subunit and Prx2 (P) were analyzed with an IDQD model. As shown in Figure 2, there were interactions between Beta and Beta, Gamma and Gamma, Beta and Alpha, Alpha and Alpha, P and Beta, Gamma and Alpha, P and Gamma, P and P, P and Alpha, Gamma and Beta, and Delta and Alpha subunits; however, there were no interactions between Beta and Delta, Delta and Delta, Delta and Gamma, or Delta and P subunits.

Figure 1.

Evaluation of the performance of IDQD models. ROC curves for IDQD model were plotted for the self-consistency test and the 3-fold cross-validation test. A mean auROC of 0.82 was obtained in the self-consistency test, while a mean auROC of 0.67 was obtained in the 3-fold cross-validation experiment.

Figure 2.

IDQD prediction results of PPI between Hb subunits (Alpha, Beta, Gamma, Delta) and Prx2 subunit. The scores were calculated by deducting -0.45, which is the threshold value (ξ₀), from the performance values between every two subunits. A positive score represents a positive interaction, while a negative score represents a negative interaction.

Hb subunits have been reported to interact with each other (38-47). To assess the predictive accuracy of the IDQD model, the predicted results of the Hb subunits were compared with the reported experimental results (Tab 2). The total predictive accuracy is 80% (the interactions of BetaBeta, GammaGamma, AlphaBeta, AlphaAlpha, AlphaGamma, AlphaDelta, BetaDelta and DeltaGamma are correct, but those of BetaGamma and DeltaDelta are incorrect), and this result demonstrates that the accuracy of the algorithm for the PPI prediction is good. The interactions between the Hb subunits (Alpha, Beta, Gamma, Delta) and Prx2 were then analyzed by this algorithm. The results showed that Prx2 had interactions with the Alpha, Beta and Gamma subunits of Hb but had no interaction with the Delta subunit (Figure 2). These results indicate that Prx2 should interact primarily with HbA, which consists of Alpha and Beta subunits.

The interaction between HbA and Prx2 predicted by IDQD was confirmed by co-IP. Co-IP was performed with anti- beta globin antibody to pull down the interacting complex of HbA from HA and RA samples. HA and RA represent the HbA band of the hemolysate and red blood cell samples during starch-agarose mixed gel electrophoresis, respectively (Figure 3). The immunoprecipitating complexes were separated by 5%-12% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) (Figure 4). Five protein bands were pulled down by the beta-globin antibody from both the HA and RA samples. These five bands were subsequently cut out, digested, and detected by ESI-Q-TOF. The MS detection results showed that the ~52 kD, ~22 kD and ~16 kD bands were the immunoglobulin heavy chain, Prx2 and beta subunits of hemoglobin, respectively (Tab 3). Thus, Prx2 was confirmed to be a composite component of the beta-globin immunoprecipitating complex. However, the detection results of the ~20 kD and ~30 kD bands were not satisfactory, so they could not be identified in this study.

Figure 3.

Starch-agarose mixed gel electrophoresis of human Hb. H represents a hemolysate sample, and R represents a RBC sample. HA represents hemolysate-HbA, RA represents RBC-HbA, HA₂ represents hemolysate-HbA_2, and RA₂ represents RBC-HbA₂.

Figure 4.

Separating Beta-globin immunoprecipitating complexes by SDS-PAGE. Negative control was prepared by immunoprecipitating RA and HA samples with mouse IgG; HA and RA represent hemolysate-HbA and RBC-HbA; M represents a protein marker.

To determine whether Prx2 can interact with HbA₂, the freeze-thawed protein extracted from the starch-agarose gel of hemolysate-HbA, RBC-HbA, hemolysate-HbA₂, RBC-HbA₂, hemolysate-HbA-HbA₂ and RBC-HbA-HbA₂ were immunoblotted with anti-Prx2 and anti-Alpha-globin (internal control) antibodies, respectively. The results showed that Prx2 was present in hemolysate-HbA, RBC-HbA, RBC-HbA₂, hemolysate-HbA-HbA₂ and RBC-HbA-HbA₂ but not in hemolysate-HbA₂ (Figure 5).

Figure 5.

Detecting Prx2 in different Hb components by Western blot. HbA, HbA₂ and HbA-A₂ represent the protein samples extracted from the starch-agarose gels of HbA, HbA₂ and Hb located between HbA and HbA₂; H represents hemolysate; and R represents RBCs.

Figure 6A shows the iron K-edge absorption curve of HbA, while Figure 6B shows the iron K-edge absorption curve of the HbA-Prx2 complex. Both curves show the leading-edge absorption peaks in the X-ray absorption near-edge structure (XANES) spectrum of each sample. The formation of a prepeak feature is due to a 1s→3d transition, which is forbidden in the octahedral coordination but is allowed in coordination geometries (distorted octahedral and tetrahedral), which is lacking a center of inversion (48). Behind the main peak, there is a flattened peak that represents the simple sinusoidal behavior due to the octahedral symmetry of the first coordination sphere (49). Our results showed that heme-Fe had the same local geometry, regardless of whether HbA binds with Prx2. Figures 6C and 6D are the Fe K-edge Fourier transform spectra of HbA and the HbA-Prx2 complex. The results showed that the distance of the first shell (the first shell containing a planar ligand formed by four Fe-N ligands and one axial ligand composed of oxygen and nitrogen (48, 50)) had not been changed. However, after the addition of Prx2, the distance of the second and other extended shells increased significantly.

Figure 6.

Detecting the change in the iron K-edge of HbA by XAFS. (A) The iron K-edge absorption curve of the HbA group. (B) The iron K-edge absorption curve of the HbA-Prx2 group. (C) The iron K-edge Fourier transform spectra of the HbA group. (D) The iron K-edge Fourier transform spectra of the HbA-Prx2 group.