2.1.

Spectroscopy System

Terahertz spectra were acquired using a commercial Z-3™ Time-Domain THz spectrometer (Zomega THz Corp., Troy, United States). In this system, optical excitation is achieved by a mode-locked Mai-Tai laser which emits fewer than 120 fs pulses centered at a wavelength of 800 nm, with an 80 MHz repetition rate and an average power of $\sim 1\mathrm{W}$. The measurements were conducted in the typical transmission geometry at room temperature ($\sim 298\mathrm{K}$), the optics were purged using nitrogen gas to remove the water vapor from the air to decrease the humidity down to less than 1%. The usable frequency range of the system is from about 0.1 to 2.5 THz, but the valid range tends to decrease when the sample is in an aqueous phase due to signal attenuation by the sample.

2.2.

Sample Holder

The sample holder was fabricated from TOPAS™ 5013L-10. This material (which is a cyclic olefin co-polymer) was chosen because it has very low attenuation at THz frequencies.²⁷ The hydrated protein formulation was pipetted into the sample holder with a liquid layer which is of the order of a few micrometers thick for the THz measurement. The clean homogeneous empty sample cell was also measured as a reference (refractive index $\sim 1.53$) enabling the spectroscopic properties of the sample to be accurately determined. Each ample was measured four times. For each measurement, the sample was extracted and repipetted in the sample holder. By averaging the spectra, systematic errors produced by wrong positioning, as well as present heterogeneities in the sample, were minimized.

2.3.

Protein and Monoclonal Antibody Preparation

The recombinant HA protein was purchased from Sino Biological (Cat no. 11229-V08H). Briefly, the DNA sequence encoding the extracellular domain of influenza A virus (A/HongKong/1073/1999; H9N2) HA was synthesized and expressed in human embryonic kidney 293 cells. The recombinant HA protein was not infectious and had no biohazard risk for any in-vitro study. Fully human monoclonal antibody F10 was expressed in 293T cells using TCAE5.3 vector²⁸ containing corresponded immunoglobulin heavy chain (VH) and light chain (VL) gene and purified using protein A beads (GE Healthcare). The concentration of the proteins was determined using the Protein A280 application module by NanoDrop spectrophotometer (Thermo Scientific). Fully human anti-CXCR4 monoclonal antibody (purchased from NBGen) was used as the irrelevant antibody control (irmAb) in this study. All of the H9 HA, F10 antibodys and irmAb were stored in sterilized phosphate-buffered saline (PBS) at pH 7.4. H9 HA was prepared at the following concentration using PBS as diluents: 7.5, 15, 29, 56, 113, 225 and $430\mu \mathrm{g}/\mathrm{ml}$.

2.4.

Enzyme Linked Immunosorbent Assay

The 100 μl H9 HA from each concentration was coated in the ELISA plate at 4°C overnight, followed by adding 100 μl of the F10 antibody in each well. After 1 h incubation at room temperature, the plate was intensively washed using phosphate buffered saline with Tween20 (PBST). Then the HRP conjugated goat anti-human IgG mAb was used as a secondary antibody for color development. For different concentrations of coated H9 HA, PBS and irmAb were used as the negative and irrelevant control. Optical density (OD) values of 450 nm were read and results were expressed as the OD reading at different H9 HA concentrations. All protein related experiments were performed by NBGen laboratory using the indicated methods.

3.1.

HA H9 Protein

The liquid phase of the H9 HA protein lacked any distinct peaks in the spectral region studied as shown in Fig. 2. Together with the average over multiple measurements, the 95% confidence intervals are plotted as error bars to determine whether differences are statistically significant. For each concentration, the absorption increases with increasing frequency. However, the absorption coefficient does not simply decrease with increasing protein concentration. Instead there is a more complex relationship.

Fig. 2

Absorption coefficients of H9 HA solution at concentrations from 7.5 to $430\mu \mathrm{g}/\mathrm{ml}$. Error bars represent 95% confidence intervals.

We see in Fig. 2 that the absorption coefficient for the $113\mu \mathrm{g}/\mathrm{ml}$ solution is higher than for all the other concentrations between 7.5 and $430\mu \mathrm{g}/\mathrm{ml}$. The absorption coefficient for the $29\mu \mathrm{g}/\mathrm{ml}$ solution is also seemingly out of sequence. To highlight these features and illustrate how the absorption coefficient depends on both frequency and concentration, we have plotted the absorption coefficients against concentration for H9 HA from 0.26 to 1.47 THz in Fig. 3.

Fig. 3

Dependence of the absorption coefficient of H9 HA on protein concentration between 0.1 and 1.5 THz.

If the protein solution followed the Beer–Lambert law, then the absorption coefficient a($\omega$) would be a linear function with increasing protein concentration. However, from Fig. 3, we see that the absorption coefficient decreases with increasing H9 HA concentration up to $29\mu \mathrm{g}/\mathrm{ml}$. Then, between concentrations of 29 and $113\mu \mathrm{g}/\mathrm{ml}$, the absorption coefficient increases, reaching its maximum at $113\mu \mathrm{g}/\mathrm{ml}$. As the concentration increases further, the absorption coefficient decreases. This remarkable “zig-zag” behavior of the absorption coefficient can be explained by considering the protein interactions present at different concentrations. At the lower concentrations (7.5 to $29\mu \mathrm{g}/\mathrm{ml}$), there is no dynamical hydration shell around the H9 HA molecules. The absorption coefficient of bulk water in the terahertz region is much higher than that of the protein, so as the H9 HA concentration increases, the absorption coefficient decreases (region A of Fig. 4). This continues until the hydration shell starts to form at around $29\mu \mathrm{g}/\mathrm{ml}$. The absorption of the hydration shell in the terahertz region is greater than that of the bulk water and as the protein concentration increases further, the absorption coefficient increases (start of region B of Fig. 4). This continues until the hydration shells start to overlap at about $113\mu \mathrm{g}/\mathrm{ml}$: the overlapping causes the absorption coefficient to decrease again, and it continues to decrease until the shells become saturated at about $225\mu \mathrm{g}/\mathrm{ml}$ and the absorption plateaus. Thus, the behavior for region A (black arrow in Fig. 4) could be modeled as a two component model (protein and bulk water)²⁹ and the behavior in region B (red arrow in Fig. 4) could be modeled by a ternary component model (protein, hydration water and bulk water). This assumes a distinct absorption coefficient of the water around the solute molecule due to the distinct properties of the solvation water.³⁰ By using the concentration at which the hydration shells become saturated, we calculate the hydration shell to be $8.84\AA$. This result corresponds to about two to three hydration water shells around a HA protein molecule ($\sim 3\AA$ average extension per water molecule) which is similar to previously reported results.^29–31

Fig. 4

The absorption coefficient of H9 HA at 1.21 THz as a function of concentration. The schematic molecules illustrate the formation and overlap of the hydration shells to interpret the nonlinear absorption behaviors. Region A follows a binary component model and a Region B follows a ternary component model.

3.2.

H9-F10 Binding Reaction

To investigate the protein binding reaction, F10 nAb ($230\mu \mathrm{g}/\mathrm{ml}$, $\sim 160\mathrm{KD}$) was mixed into H9 HA solutions with the concentration from 7.5 to $430\mu \mathrm{g}/\mathrm{ml}$ by 1:1 volume ratio. The irmAb was used as the negative control with the same reaction molar ratio. The mixtures were incubated with gentle mixing for $\sim 30\min$ at room temperature and then overnight at 4°C to allow binding reactions and complex formation. No precipitate was observed in the samples.

Table 1 presents the reaction ratio of the H9 HA binding F10 nAb in our study. As explained in Sec. 1, usually $\sim 3$ antigen-binding fragments (Fabs) per trimer H9 HA can saturate all the F10 binding sites. This gives an optimum ratio of binding sites for H9:F10 as 1:1 and 1:3 for ${\mathrm{H}9}_{\text{trimer}}:\mathrm{F}10$. To investigate the relationship between THz absorption and concentration, we chose the concentration of the F10 such that the optimum ratio would occur when the concentration of H9 HA is nearly $113\mu \mathrm{g}/\mathrm{ml}$.

Table 1

H9 hemagglutinin (HA) binding F10 antibody reaction ratio.

To obtain a full picture of how the absorption coefficient depends on both frequency and concentration, we present the 3-D plots of the concentration-dependent terahertz absorption for the H9 HA binding F10 nAb [Fig. 5(a)] and the irmAb [Fig. 5(b)] in the range of 0.1 to 1.5 THz. We use the RBS H9-F10 values instead of the protein concentration as it accounts for the concentrations of both F10 and H9 HA. Figure 5(c) uses the data at 1.21 THz to highlight the comparison of absorption properties against the RBS of H9-F10 and H9/irmAb solutions. In the spectral range from 0.1 to 1.5 THz, we have observed a remarkable absorption of H9 HA solutions at $113\mu \mathrm{g}/\mathrm{ml}$ because of the overlap of the hydration shell. Within our measurement for antibody/antigen mixtures, the absorption property has been similarly shown in H9/irmAb groups (negative control) but shifted to the smaller concentration of $56\mu \mathrm{g}/\mathrm{ml}$. However, the effect of the hydration shell was not present in the H9 HA binding F10 (red solid line) in Fig. 5(c). This is because the HA protein contains a large number of amino and carboxyl residues (e.g., ${\mathrm{COO}}^{-}$), which charged the surface of the protein in neutral solutions. Polarized water molecules and oppositely charged sites on antibodies (e.g., ${\mathrm{NH}}^{3+}$) shape the hydration shell around the protein by electrostatics. The binding interactions between antibodies and antigens cause the charged sites to reduce and even disappear. Consequently, the hydration shell becomes lose. As for the negative control group, there is no binding interaction between the H9 HA and irmAb molecules, and the hydration shells overlap at a lower concentration than for H9 HA because of the higher molecular density. The terahertz spectroscopy data for the H9-F10 indicate that the absorption coefficient decreases sharply with an increasing ratio of the binding sites of H9/F10 until about 1.35, which is the closest to the optimum ratio (see also Table 1). After that, the absorption coefficient increases gradually by reason of the excess of the H9 HA in solution.

Fig. 5

The concentration dependence of the absorption coefficient for (a) H9-F10 solution and (b) negative control (H9/irmAb) at frequencies 0.1 to 1.5 THz. (c) Plot of concentration against absorption coefficient of H9, H9/irmAb and H9-F10 solution at 1.21 THz. (RBS: ratio of binding sites).

3.3.

Specificity and Sensitivity Detection for the H9-F10 Binding

The dielectric loss tangent is defined by $\tan \delta ={\epsilon }^{\prime \prime }/{\epsilon }^{\prime }$, which quantifies a dielectric material’s inherent dissipation of electromagnetic energy to charge motion. Moreover, $\tan \delta$ measurements are extremely sensitive to protein molecular relaxation and local motions involving the reorientation of side-chains of residues as it is subjected to an oscillating electric field.³² This enables us to detect the specificity and sensitivity of antibody–antigen binding using THz spectroscopy. In Fig. 6, we show the Cole–Cole plots (${\epsilon }^{\prime }$ versus ${\epsilon }^{\prime \prime }$) for our data.

Fig. 6

Cole–Cole plots for (a) H9 HA solution, mixtures of H9 HA and irmAb and (b) H9 HA binding F10 nAb at different concentrations.

H9 HA has such a higher dielectric constant and dielectric loss compared to the H9/irmAb mixtures that the H9 HA group and the H9/irmAb group are clearly separated in the plot. The Cole–Cole plots for H9/irmAb mixtures are close to semicircular arcs. However, for the H9 HA, the data shape can be described as a skewed arc, indicating a distribution of relaxation times for these samples. Furthermore, it is interesting to find that for the distribution of H9-F10 binding complex in Cole–Cole plots, the values for the imaginary and real parts initially decreased from the range of H9/irmAb (concentrations 1 to 4) and then increased up to the range of H9 HA (concentrations 5 to 7) with the F10 nAb consumed.

Figure 7 shows the relationship between the dielectric loss tangent and the reaction concentrations of H9-F10 and H9/irmAb at 1.21 THz, respectively. The H9 solution is plotted by green marks for comparison. All H9 HA, H9-F10 complex solutions and H9/irmAb mixtures display polar polymer characteristics as they have a relatively large dielectric loss tangent (0.1 or more). PBS solution has the weakest polar characteristics and had the lowest tanδ of about 0.1 (dark gray star). The results indicate that the $\tan \delta$ values of H9 HA and H9/irmAb solutions are approximately independent of concentration and lie at 0.12 and 0.61, respectively, for a wide range of concentrations. However, the dielectric loss tangent of the H9-F10 solutions (red line) was found to have three significant stages. At stage (a), increasing the H9 HA content in a fixed concentration of F10 nAb caused the tanδ values to decrease sharply. This is because the F10 nAb solution has a higher dielectric loss tangent of about 1.18 (pink mark) and the reaction molar ratio of H9 HA and F10 in solutions is smaller than F10 itself at the beginning. After a relatively steady stage (b) during which the dielectric loss tangent remains close to 0.53, the magnitude of tanδ decreases again and approaches the value for the H9 HA solution of about 0.12, stage (c). From these results, we can deduce that the binding state of the antigen with its specific antibody is detected at stage (b) as the dielectric loss tangent reaches a stable stage which means an energy balance of antigen–antibody interaction. Stage (b) starts at $15\mu \mathrm{g}/\mathrm{ml}$, therefore, this is the minimum detectable limit of H9-F10 binding in our work. The nonlinear change in the dielectric loss tangent observed at $113\mu \mathrm{g}/\mathrm{ml}$ occurs at the same concentration at which the hydration shells of H9 HA started to overlap in Fig. 3 and is caused by the excess of the H9 HA as per the analysis in Table 1.

Fig. 7

Plot of $\tan \delta$ versus ratio of binding sites of H9 HA and F10 (red circle) and negative control (blue down arrow), H9 HA solutions presented by green marks for the comparison. F10 nAb solution has the higher dielectric loss tangent of about 1.18 (pink left arrow) and PBS solutions with the weakest polar characteristics indicate the lowest $\tan \delta$ of about 0.1 (dark gray star). The linear best fit is also shown (black solid line) for H9 and H9/irmAb.

3.4.

ELISA for H9-F10 Binding

An ELISA test was performed to determine the specificity and sensitivity of antigen binding events between H9 HA and F10 nAb. In Fig. 8, the result showed that while there was no observed binding of PBS and irmAb with H9 HA at each concentration, F10 was detected to bind to the H9 HA with increasing concentration. The ELISA test results were considered positive when the OD was $>0.1$ over the negative control. Here, the gradually escalated signal could be observed in the F10 binding curve (blue) starting at a concentration of $113\mu \mathrm{g}/\mathrm{ml}$: they surged after that, which indicated that the detection limit determined by ELISA for H9-F10 was $113\mu \mathrm{g}/\mathrm{ml}$ compared to $15\mu \mathrm{g}/\mathrm{ml}$ with the terahertz dielectric loss tangent. Thus, the standard ELISA test is not as sensitive to the binding as the terahertz dielectric loss tangent.

Fig. 8

Specificity and sensitivity measurements by ELISA. ${\mathrm{OD}}_{450}$ is the optical density at 450 nm.