2.1.

Experimental Setup

Figure 1 shows a schematic of our experimental setup. Light from a 100-W halogen tungsten lamp is coupled to the inverted prism through a multimode optical fiber for SPR excitation. A bandpass filter selects an input wavelength window with a center wavelength of 610 nm and full width at half maximum (FWHM) of 220 nm. The LCTF rapidly scans the output wavelength in a stepwise manner. For our LCTF unit (VariSpec, VIS-10-20-STD), the FWHM, spectral tuning resolution, and response time are, respectively, 10 nm, 0.5 nm, and 50 ms. In our experiments, we chose a wavelength step of 1 nm as this was found to have the lowest level of noise in the final detected signal. While the Kretschmann configuration has been adopted, the SPR cell consists of an equilateral prism made of BK7 glass, a microscope glass slide coated with a 48-nm-thick gold film, and a flow chamber for sample injection. The incident angle can be manually set to an optimized value calculated using Fresnel equations. In the present case, the angle for pure water with an excited wavelength of 660 nm is 72.5 deg. The output light is captured by a 12-bit CCD camera. The A-D card (model number: NI DAQ 6259) generates a triggering signal that synchronizes the operation of the LCTF and CCD camera. An exposure time of $50\mathrm{ms}/\text{wavelength}$ is used because of the compromise between camera capture speed and signal-to-noise ratio. In our experiments, it takes 50 ms for the LCTF to switch the wavelength before the CCD starts its exposure cycle. Consequently, the CCD camera records a long series of images taken at different wavelengths within a prescribed spectral range. Figure 2 shows a series of recorded CCD image frames that covers the spectral dip as selected by the LCTF. Each pixel within the CCD image frame provides one wavelength data point within the spectral SPR profile of a small region in the sensor surface. Theoretically, this one-to-one correspondence between the sensor surface and the 2-D CCD imaging pixel arrays may offer massively parallel-biosensing capability. Upon analyzing a series of images, one can build a set of spectral plots that covers all the pixels within the sensor surface. Here, we use a common second order polynomial-fitting algorithm to locate the spectral dip, which corresponds to the resonant wavelength of each pixel. The data analysis procedures are performed in an automatic mode using lab-designed software based on numerical comparison techniques.

Fig. 1

Schematic of our SPRi system in the Kreschmann configuration. Light from a halogen lamp is collected by a multimode fiber through a set of coupling optics. At the exit end of the fiber, the beam is collimated by a group of lenses and spatially filtered by an aperture before passing through the tunable spectral filter unit and a linear polarizer. The sensor surface is imaged by a CCD camera using two imaging lenses. ${\mathrm{L}}_1$ to ${\mathrm{L}}_7$: lens; BF: bandpass filter; MF: multimode fiber; DA: diaphragm aperture; LCTF: liquid crystal tunable filter; ${\mathrm{P}}_1$ and ${\mathrm{P}}_2$: polarizer.

Fig. 2

Schematic of the image recording process. Regions marked by A-J represent the sample areas on the chip. First to $N$’th frame represent a series of images recorded by the CCD camera as we sweep the center wavelength of the LCTF step-by-step. The intensity values of each pixel location taken from the frames constitute one SPR spectral plot, as indicated by the red curve. Therefore, the number of pixels in each image frame corresponds to the number of spectral plots from the system.

2.2.

Feedback Loop

The feedback loop shown in Fig. 1 is designed to automatically track the shift of the SPR dip caused by the binding of target biomolecules. Most importantly, incorporation of this feedback loop drastically reduces the spectral width one needs to sweep before hitting an SPR dip. For example, it takes 80 s to obtain the resonance dip with a traditional stepping monochromator based on a spectral range of 40 nm and a step-size of 1 nm, whereas with the help of a feedback loop, the time required by the present system to find the SPR dip is 4 s. Moreover, an LCTF can scan the incident wavelength across any spectral range according to an optimized strategy with minimal delay. Consequently, the amount of time required to arrive at an SPR resonance data point is significantly shorter.

Figure 3 schematically shows how the feedback loop operates. First, an initial SPR spectral profile is obtained by scanning the incident wavelength across a large spectral range to extract the baseline resonant wavelength ${\lambda }_0$. Then a spectral range ${\lambda }_0-x$ to ${\lambda }_0+x$ is chosen. With this, the desired SPR dip with sufficient resolution ($<0.01\mathrm{nm}$) (solid red straight line along horizontal axis in which $x$ is half the width of the wavelength-scanned range) can be obtained.²⁵ The corresponding partial SPR plot near the SPR dip (red dotted curve) is repeated if the RI $n_0$ of the baseline solution does not change. When a change of RI has taken place, e.g., from $n_0$ to $n_1$, a new partial SPR plot (solid red curve) is obtained by scanning the incident wavelength from ${\lambda }_0-x$ to ${\lambda }_0+x$. Thus, a new resonant wavelength ${\lambda }_1$ is extracted, then the new scanning spectral range is modified to ${\lambda }_1-x$ to ${\lambda }_1+x$ (solid green straight line along the horizontal axis). If the RI continues to change from $n_1$ to $n_2$, the corresponding partial SPR plot (green solid curve) is then updated by scanning the incident wavelength from ${\lambda }_1-x$ to ${\lambda }_1+x$. This process will continue until the system has identified the next new resonant wavelength ${\lambda }_2$, which will trigger further modification of the scanning range, i.e., from ${\lambda }_2-x$ to ${\lambda }_2+x$. In a more general representation, the $i+1$’th scanning spectral range, which covers ${\lambda }_{i+1}-x$ to ${\lambda }_{i+1}+x$ (solid yellow straight line along the horizontal axis), is generated from the resonant wavelength extracted from the partial SPR plot near the SPR dip obtained in the $i$’th scanning spectral range of ${\lambda }_{\mathrm{i}}-x$ to ${\lambda }_{\mathrm{i}}+x$ (solid black straight line along horizontal axis). For a 2-D sensor array, a number of different resonant wavelengths will be obtained from the sensor elements as each of them has been designated to perform a specific sensing task. In such a situation, our system will try to cover the entire range of all the elements by registering the lower and upper limits, i.e., ${\lambda }_{i\min }$ and ${\lambda }_{i\max }$, of the spectral dip, and the range will become ${\lambda }_{i\min }-x$ to ${\lambda }_{i\max }+x$.

Fig. 3

Schematic showing the tracking of SPR dip movement based on the feedback loop technique. Here, $n_0,n_1,\dots ,n_i$, and $n_{i+1}$ represent a series of RI values at the Au/sample interface when a molecular binding interaction is taking place on the sensor surface. When the RI changes continuously from $n_0$ to $n_i$, the spectral range to be scanned will also be adjusted automatically from the original spectral range [${\lambda }_0-x$, ${\lambda }_0+x$] to [${\lambda }_i-x$, ${\lambda }_i+x$] as guided by the feedback loop. This way, only a selected range around the resonance dip will be interrogated at all time, thereby drastically shortening the time to calculate the location of the SPR dip.

3.1.

Linearity and Resolution of Surface Plasmon Resonance Measurement

To estimate the performance of our SPRi system, we measured the RI values of different salt-water mixtures. Salt solutions with a concentration level ranging from 0% to 25% in increments of 5% by volume, which corresponds to a RI ranging from 1.3330 to 1.3793 RIU,²⁶ were prepared. In the experiment, an incident angle of 72.5 deg in the prism was chosen based on the theoretical simulation of the SPR angular dip of water. The scanning spectral range was about 40 nm, and the time for finding the minima of the spectral SPR dip was about 4 s. The $3\times 3$ array sensor sites, which were uniformly distributed on the chip, were spectrally monitored simultaneously to assess the performance of our system. The step changes of resonant wavelength obtained from different salt concentrations are shown in Fig. 4(a), and the corresponding shifts in spectral SPR dip, as caused by changes in RI, are shown in Fig. 4(b). Based on measurement data obtained in the wavelength range from 600 to 700 nm, the calculated resolution is $5.87{\times 10}^{-6}\mathrm{RIU}$, and the measured dynamic range of the system is $4.63{\times 10}^{-2}\mathrm{RIU}$. Here, the conversion from the minimum measurable wavelength shift to the corresponding change in refraction index value is performed with the equation ${\sigma }_{\mathrm{RI}}=\frac{\delta n}{\delta \lambda }·{\sigma }_{\mathrm{SD}}$,²⁶ where ${\sigma }_{\mathrm{RI}}$ is the RI resolution, $\delta n$ is the RI change of a bulk medium, $\delta \lambda$ is the calculated wavelength shift, and ${\sigma }_{\mathrm{SD}}$ is the estimated wavelength noise. Here, ${\sigma }_{\mathrm{SD}}=0.034\mathrm{nm}$ is obtained through averaging 10 measurements of the RMS noise of our system.

Fig. 4

(a) Shift of SPR dip versus salt concentration in water as detected by sensor sites within the $3\times 3$ array. (b) Relationship between resonant wavelength of the SPR dip and the shift in RI as detected by the $3\times 3$ array on the SPR chip. Each data point is obtained by averaging the intensity values of $30\times 20\text{pixels}$ to reduce the effect of noise with no compromise in spatial resolution.

3.2.

Application to IgG Interaction Monitoring

To assess the capability of our SPRi system for monitoring biomolecular interactions, we performed real-time monitoring of binding interactions between goat antirabbit IgG and rabbit IgG at different concentration levels. Rabbit IgG was dissolved in the array-spotting buffer at concentrations of 50, 100, 150, and $250\mu \mathrm{g}{\mathrm{ml}}^{-1}$. BSA was chosen as the positive control protein. A $3\times 5$ array of protein dots was spotted on the SPR chip by hand. Within each column, the three spots had the same solution. The first column on the left was $100\mu \mathrm{g}{\mathrm{ml}}^{-1}$ BSA solution; other columns were rabbit IgG solution with concentrations of 250, 150, 100, and $50\mu \mathrm{g}{\mathrm{ml}}^{-1}$, from left to right [see Fig. 5(a)]. The diameter of the spots was about 1 mm, and the distance between the centers of the two adjacent spots was about 2.5 mm.

Fig. 5

Measurement of goat antirabbit IgG and rabbit IgG antigen–antibody interaction. (a) White light image of the SPR sensor chip, (b–d) real-time wavelength response of antigen–antibody binding interactions, (e–g) SPR shift images of a $3\times 5$ array obtained after interaction with goat antirabbit IgG solutions of 5, 10, and $15\mu \mathrm{g}{\mathrm{ml}}^{-1}$, respectively. The first column sensor sites, respectively, marked by 11, 21, 31 corresponding to $100\mu \mathrm{g}{\mathrm{ml}}^{-1}$ BSA; the second column sites are, respectively, marked by 12, 22, 32, which correspond to $250\mu \mathrm{g}{\mathrm{ml}}^{-1}$ rabbit IgG; the third, fourth, and fifth column sites are marked by 13, 23, 33, and 14, 24, 34 and 15, 25, 35, which correspond to 150, 100, and $50\mu \mathrm{g}{\mathrm{ml}}^{-1}$ rabbit IgG, respectively. The color of individual SPR plot, i.e., pink, green, blue, red or black, denotes the probe area of $i_1$, $i_2$, $i_3$, $i_4$, or $i_5$, respectively.

The IgG interaction procedures were carried out as follows. First, a $3\times 5$ array chip was attached to the coupling prism using a small drop of RI matching liquid. A piece of molded PDMS was placed over the chip to form a sealed flow cell of $15\mathrm{mm}\times 15\mathrm{mm}\times 0.3\mathrm{mm}$ ($l\times w\times h$). We designed software to analyze the 15 sensing regions captured by the CCD. A set of 15 spectral SPR plots representing the binding interactions taking place on the spotted sensing regions were generated from the sensor chip. We manually tuned the incidence angle to achieve minimum intensity at the SPR dip for a phosphate buffer saline (PBS) sample. A glycine solution (100 mM, $\mathrm{pH}=2.0$) was injected for 5 min to regenerate the sensor surface. Next, a buffer solution, PBS $1\times$, was injected in the flow cell for 3 min with a micropump operating at a speed of $25\mu \mathrm{l}/\min$ to obtain a steady baseline. The initial small scanning range was set at ${\lambda }_{0\min }-20\mathrm{nm}$ to ${\lambda }_{0\max }+20\mathrm{nm}$, in which ${\lambda }_{0\min }$ and ${\lambda }_{0\max }$were, respectively, the minimum wavelength and maximum wavelength among the resonant wavelengths extracted from all the SPR spectral profiles obtained from the sensing regions. The half width of the scanning spectral range $x=20\mathrm{nm}$ was chosen to achieve the fastest wavelength interrogation and lowest noise. After obtaining a steady baseline for several minutes, the process of capturing the SPR image was automatically executed for one scan to obtain the baseline spectral SPR image, which was produced through extracting the spectral SPR plot of individual sensor sites. Then IgG solutions of 5, 10, and $15\mu \mathrm{g}{\mathrm{ml}}^{-1}$ were in turn pumped into the sensor chip. The subsequent antigen–antibody binding interaction taking place on the detection surface was monitored in real time, as shown in Fig. 5. The binding interaction measurement time was typically $\sim 3\min$ for each concentration with a constant flow of sample using a syringe pump. After completion of each binding interaction measurement, PBS $1\times$ solution was pumped into the flow chamber for 2 min to ensure adequate buffer washing. To see clearly the shift of the SPR wavelength dip as a function of protein concentration, we arranged the SPR plots into three groups corresponding to the upper, middle, and lower rows of the $3\times 5$ array. Each group included five members corresponding to four different concentration levels each of IgG and the BSA control sample. As shown in Figs. 5(b)–5(d), the first group showed that the shift of the SPR wavelength dip increases as protein concentration increases in the range from 5 to $15\mu \mathrm{g}{\mathrm{ml}}^{-1}$. Locations on the other two groups on the chip exhibited a similar trend. After completing buffer washing and regeneration steps, the system collects a steady baseline image using data gathered for 3 min, and a reference SPR dip image is produced. Consequently, a reference SPR dip images is available for each IgG concentration. All subsequent SPR dip images obtained while binding interactions were ongoing were analyzed and compared using the first set of reference images and the updated one as the shift of the SPR dip progressed. Figures 5(e)–5(g) show representative images of SPR shift obtained with different protein concentration levels. In the images, the left column of sensor sites shows no change because goat antirabbit IgG does not specifically bind to BSA. In images of other sensor sites, the presence of binding interactions as the SPR dips shift continuously is clearly revealed: the wavelength shift is approximately proportional to the change in concentration of the protein solution.

3.3.

Determination of Important Parameters

Within our SPRi technique, interrogation speed is determined by the choice of parameters including step size and total spectral width to be scanned. We analyzed the effects of step size and spectral width on system noise by measuring the variations in SPR dip obtained from pure water. As shown in Fig. 6, for a given step size, system noise decreased and reached a constant level as spectral width increased. Conversely, decreasing the step size decreased the system noise, and the overall interrogation time also increased. For our antigen–antibody interaction experiment, a spectral width of 40 nm and a step size of 1 nm were chosen because they provide optimal performance in terms of noise and interrogation speed under normal curve fitting situations. The salt solution experiment showed that our spectral SPR technique could monitor binding interactions at a rate of about 4 s per data point with an RI resolution and dynamic range similar to that of traditional spectral SPR technique.

Fig. 6

Root-mean-square (RMS) noise level versus scanning spectral width for different step sizes.

Spatial resolution is also an important parameter of the SPR imaging system. It primarily defines the density of the final sensor array. We measured the spatial resolution of our SPR imaging system using a standard resolution target. The $x$-axis and $y$-axis resolution limits are 27.8 and $35.8\mu \mathrm{m}$, which, respectively, correspond to 6 and 4 pixels of the CCD imaging chip. In our experiments, a $3\times 2\text{pixel}$ area was averaged to improve the signal-to-noise ratio in the process of producing the SPR image. But for all SPR plots, a larger area, i.e., $30\times 20\text{pixels}$ square, was used to obtain the SPR plots for the binding interaction to achieve an optimized signal-to-noise ratio.