2.1.

Multimodal Multiphoton Nonlinear Optical Microspectroscopy Imaging System

Figure 1 shows a schematic of our multimodal multiphoton nonlinear optical microspectroscopy imaging system based on a small GRIN lens probe. This multimodal multiphoton microendoscope imaging modality was modified and developed from a previously described system.^23–25 The laser system used a home-made Ti:sapphire femtosecond oscillator pumped by a green laser source (Verdi V5, Coherent). The output beam from the oscillator generates a 120 fs, 56.1 MHz pulse train, and a 460 mW average output power with a center wavelength of 800 nm. Our laser system is described in greater detail in our previous paper.²⁵ The laser beam passed through an optical isolator (IO-5-NIR-LP, Thorlabs) and was divided into two beams (3:1) by a beam splitter (BP133, Thorlabs). The reflected beam was transmitted via a series of optical components for spatial filtering, and a “pump beam” in a narrow band for the CARS process, with a pulse of 0.8 nm and center wavelength of 800 nm, was generated by combining two filters (LL01-808-12.5, Semrock and custom-made filter from Andover Corporation). The transmitted pulse, in contrast, was coupled into the photonic crystal fiber (FemtoWHITE 800, NKT Photonics) via an objective lens to make ultrabroadband “Stokes pulses.” Both the narrowband pump and broadband Stokes pulses were spatially and temporally overlapped by a long-wavelength pass filter (BLP01-785R-25, Semrock), and the two beams were passed through galvanometric mirrors and microendoscope. We demonstrated multiphoton microendoscopy by using a GRIN lens [GT-MO-080-018-810, 0.8 numerical aperture (NA), $4.8\times$, WD: $200\mu \mathrm{m}$, GRINTECH], which was fixed under the microscope by a clamp (VK250, Micro V-Clamps, Thorlabs) attached to a three-axis stage. The objective lens (MPlanFLN, 0.3 NA, $10\times$, Olympus) was used to inject light into the GRIN lens, and the total magnification of the microendoscope was $48\times$. We used a comparable microscope objective lens (CFI Plan Fluor, 0.8 NA, $40\times$, W, Nikon) to compare optical images obtained by an objective lens and endoscopy. The backwards-scattered nonlinear signals were collected by an epi-detection configuration. They were separated from the excitation beams by a dichroic mirror (custom-made mirror from CVI) and passed through the short-pass filter (FF01-775/SP-25, Semrock). The CARS, TPEF, and SHG signals were spectrally separated by dichroic mirrors and bandpass filters (CARS: HQ650/20 m, Chroma, TPEF: FF01-550/88, SHG: FF01-390/18, Semrock) to obtain multimodal nonlinear images. All of their signals were simultaneously recorded by two different photomultiplier tubes, one for CARS and the other for TPEF and SHG (H-8249-102, H-7827-012, Hamamatsu). A monochromator (DM500i, DongWoo) was used to obtain the CARS spectrum, and a CCD camera (iDus DV401A-BV, Andor) having an effective spectral resolution of $5{\mathrm{cm}}^{-1}$ was used to record it. To convert the signals to nonlinear optical images, a data acquisition program written in LabVIEW 8.6 was used. Typically, powers of 20 mW for the pump and 3 mW for the Stokes pulses were used at the sample for imaging.

Fig. 1

Schematic of a multimodal multiphoton microspectroscopy imaging system based on gradient-index (GRIN) lenses for simultaneous coherent anti-Stokes Raman scattering (CARS), two-photon excited fluorescence (TPEF), and second-harmonic generation (SHG) microendoscopic imaging observations. ISO, isolator; BS, beam splitter; OL, objective lens; PCF, photonic crystal fiber; LPF, long-wavelength pass filter; NBPF, narrow bandpass filter; GS, galvanometer scanner; DM, dichroic mirror; SPF, short-wavelength pass filter; PMT, photomultiplier tube; and FM, flip mirror.

2.2.

Brain Tissue Preparation and Immunohistochemistry Method

All the experiments were performed according to the provisions of the Gwangju Institute of Science and Technology (GIST) Animal Care and Use Committee. Transgenic mice ($5\times$ FAD) were obtained originally from the Jackson Laboratory, and as a kind gift from professor Woo Jin Park (GIST). For brain sample preparation, 29-weeks old wildtype and $5\times$ FAD mice were transcardially perfused with 4% paraformaldehyde in $1\times$ phosphate-buffered saline (PBS) solution under anesthesia with isoflurane. The whole brain was obtained, fixed in 10% neutral formalin solution (Biochemical, BBC), and embedded in paraffin via a Leica EG1140C. Coronal hippocampal sections $5\mu \mathrm{m}$ in thickness were obtained with a Leica RM2135 microtome. After sectioning, brain tissue was dewaxed and rehydrated. It is reported that the formalin and paraffin fixed methods do not substantially affect the CARS spectra on lipids in brain tissue samples.²⁶

For the detection of amyloid beta (Aβ) oligomers, sections were first rehydrated then endogenous peroxidase activity was quenched using 3% ${\mathrm{H}}_2{\mathrm{O}}_2$ (Dako), and subjected to block with 1% bovine serum albumin in PBS for 1 h RT. Sections were incubated in amyloid beta 6E10 primary antibody (1:500, Covance) at 4°C for 24 h followed by biotin-labeled secondary antibodies, horseradish peroxidase-labeled avidin–biotin complex, and 3,3′-diaminobenzidine (Dako). Sections were counterstained with hematoxylin (Harris), mounted, and covered with Deckgläser coverslips. The hippocampus was visualized with a Zeiss upright light microscope in $5\times$, $10\times$, and $20\times$ (Zeiss).

3.1.

Comparison of Multiphoton (CARS/TPEF/SHG) Images Taken with Objective Lens and Microendoscope

We focused on the hippocampal CA-1 and DG regions in our measurements because these two areas are damaged during AD and are well-known to be involved in learning and memory.

Figure 2 compares the multimodal multiphoton images from the same position in the CA-1 region in AD samples taken with the objective lens (a)–(d) and microendoscope (e)–(h). CARS, TPEF, and SHG images were obtained simultaneously with the objective lens in Figs. 2(a)–2(c) and with the microendoscope in Figs. 2(e)–2(g), respectively, and the corresponding merged CARS, TPEF, and SHG images are shown in Figs. 2(d) and 2(h), respectively. Each image was $90\times 90\mu {\mathrm{m}}^2$ ($300\times 300\text{pixels}$); the acquisition time per image was within 5 s, and the size of the scale bar was $20\mu \mathrm{m}$. The CARS images were obtained from the resonant CARS signal of the aliphatic $\mathrm{C}─\mathrm{H}$ stretching vibration mode region from lipids. The TPEF and SHG images were observed using the autofluorescence signal from amyloid fibers and collagen, respectively.^15,27–29 TPEF Figs. 2(i) and 2(j) compare the CARS intensity profiles of the $X$- and $Y$-axes along the white dotted line in the objective lens image in Fig. 2(a) (red line in profiles) and the endoscope image in Fig. 2(e) (blue line in profile). The CARS signal strength for the objective lens is 23% higher, but the optical resolutions are similar owing to their identical NA. Figure 2 confirms that our microendoscope system maintains the image quality and spatial resolution of images obtained using an objective lens.

Fig. 2

Comparison of multimodal multiphoton nonlinear images from the same Alzheimer’s disease (AD) brain tissue region taken with (a)–(d) objective lens and (e)–(h) microendoscope. (a, e) CARS images, (b, f) TPEF images, (c, g) SHG images, and (d, h) merged images. (i), (j) Intensity profiles of $X$- and $Y$-axes along the white dotted line for (a) objective lens and (e) endoscope images. All scale bars are $20\mu \mathrm{m}$.

3.2.

Comparison of Microendoscopic Multiphoton Images of Normal and AD Samples in CA-1 Region

In Fig. 3, we compare simultaneously observed multimodal multiphoton (CARS/TPEF/SHG) microendoscopic images of normal (a)–(d) and AD (e)–(h) samples from the CA-1 region of each sample. Each image was $80\times 80\mu {\mathrm{m}}^2$ ($300\times 300\text{pixels}$); the acquisition time per image was within 5 s, and the size of the scale bar was $20\mu {\mathrm{m}}^2$. In the normal sample, we observed a low signal from lipids in Fig. 3(a) and low concentrations of amyloid beta fiber in Fig. 3(b) and collagen in Fig. 3(c) in the CA-1 region. In contrast, in the AD sample, we confirmed an increased CARS signal from lipids and very high amounts of amyloid beta fiber (white arrows) and collagen (yellow arrows), with large differences from the normal sample.

Fig. 3

Comparison of typical multimodal multiphoton microendoscope images of (a)–(d) normal and (e)–(h) AD samples from hippocampal brain CA-1 region. (a, e) CARS images, (b, f) TPEF images, (c, g) SHG images, and (d, h) merged images. Red: CARS (lipid-rich structure). Green: TPEF (amyloid fiber or other fluorescent particles, yellow arrows). Blue: SHG (collagen, yellow arrows). All scale bars are $20\mu \mathrm{m}$.

In order to prove our results, we observed amyloid beta fiber, which is known as a representative hallmark of AD, in normal and AD samples by using immunostaining method. Figure 4 indicates the representative histology images of normal and AD samples. Dark brown color represents amyloid beta fiber (black arrows). As a result, we confirmed that amyloid beta fiber is observed in AD samples much more than in normal samples as shown in Fig. 4. Amyloid beta plaques are regularly developed and observed in 29-week old AD mice.

Fig. 4

Histology image of (a, c) normal brain and (b, d) AD brain ($5\times$ FAD) with immunostaining method. Dark brown color represents an amyloid beta fiber.

3.3.

Comparison of Microendoscopic Multiphoton Images of Normal and AD Samples in DG Region

Figure 5 shows our observations of the DG region, a different region in the hippocampus, and compares multiphoton microendoscopic images of normal (a)–(d) and AD (e)–(h) samples. The image size, acquisition time, and size of the scale bar in Fig. 5 are the same as in Fig. 3. Figures 5(a) and 5(e) are CARS images of the normal and AD samples obtained using the resonant CARS signals of the aliphatic $\mathrm{C}─\mathrm{H}$ stretching vibration mode region, as described previously. In the normal tissue, we observed a well-distributed CARS signal, but in the AD sample, there were high signals in some areas [yellow in Fig. 5(e)] due to GABA, a neurotransmitter reported as the most prominent feature in AD in many research papers.^{5–7,30–32} This means that GABA is more highly accumulated in the DG in the AD sample than in the normal sample. We also observed higher autofluorescence signals, as shown in Fig. 5(f) (white arrows) than in Fig. 5(b), and these come from amyloid beta fiber (red dotted circle) and fluorescent particles such as NAD(P)H. Amyloid fiber is also an important hallmark of the AD brain.³³ In addition, we obtained very strong SHG signals (yellow arrows) from collagen as shown in Fig. 5(g).

Fig. 5

Comparison of typical multimodal multiphoton microendoscope images of (a)–(d) normal and (e)–(h) AD samples from hippocampal DG region. (a, e) CARS images, (b, f) TPEF images, (c, g) SHG images, and (d, h) merged images. Red: CARS (lipid-rich structure). Green: TPEF (amyloid beta fiber or other fluorescent particles, yellow arrows). Blue: SHG (collagen, yellow arrows). All scale bars are $20\mu \mathrm{m}$.

We confirmed the increased amounts of lipids, amyloid fiber, and collagen in the AD samples compared to normal brain samples. In particular, higher multiphoton signals (CARS/TPEF/SHG) are generated in the DG region than in CA-1. Several research papers showed that DG is the primary gateway for cortical input to the hippocampus and is crucial during memory formation and recall.^6,7 Therefore, in AD, the DG region has more prominent markers than CA-1, and our microendoscope system will be a powerful technique for observing these markers.

3.4.

Quantitative Analysis of Spectral Difference Between Normal and AD Samples with CARS Microendoscope Spectroscopy

We measured the spectral difference between the normal and AD samples using CARS microendoscope spectroscopy. We performed the experiment with 20 normal and 20 AD samples each day and repeated it four times, so a total of 160 spectra were measured. Figure 6 compares typical CARS spectra of representative normal (blue squares) and AD (red triangles) samples obtained from the region in the blue dotted circles in Figs. 5(a) and 5(e), which were overlapped as shown in Fig. 6(a). The CARS spectra were measured near the $\mathrm{C}─\mathrm{H}$ stretching vibration mode and were obtained at an arbitrary point of lipids inside the tissue, not by integrating the entire tissue area. Each point in the normal and AD groups had the same spectral pattern group by group except for the height of the intensity. The exposure time for taking a CARS spectrum was 1 s. In the normal brain sample, two main peaks were observed, at 2845 and $2930{\mathrm{cm}}^{-1}$, which correspond to the ${\mathrm{CH}}_2$ symmetric stretching vibration and ${\mathrm{CH}}_3$ symmetric stretching vibration, respectively. On the other hand, in the AD brain sample, only one peak was strongly observed, at $2850{\mathrm{cm}}^{-1}$, which corresponds to the ${\mathrm{CH}}_2$ symmetric stretching vibration mode, whereas ${\mathrm{CH}}_3$ symmetric stretching vibration mode was clearly weakened. Typically, the CARS signal intensity of AD samples is almost six times stronger than the intensity of the control sample at the ${\mathrm{CH}}_2$ symmetric stretching vibration as shown in Fig. 6(a). This is because considerable GABA, the chemical characteristics of which are responsible for the ${\mathrm{CH}}_2$ symmetric stretching mode, accumulates as AD progresses.^5,6,30–32

Fig. 6

(a) Comparison of typical CARS spectra of control (blue squares) and AD (red triangles) in $\mathrm{C}─\mathrm{H}$ stretching vibration region. (b) Normalized version of CARS spectra in Fig. 5(a).

To analyze quantitatively the difference in the CARS spectra of the control and AD brain samples, all the data were normalized by setting the $2845{\mathrm{cm}}^{-1}$ peak equal to 1.0. Figure 6(b) shows the normalized versions of the spectra in Fig. 6(a). For the normalized CARS spectra from 80 normal and 80 AD brain samples, the $2845{\mathrm{cm}}^{-1}$ peak intensity $I_{2845}$ was divided by the $2930{\mathrm{cm}}^{-1}$ peak intensity, $I_{2930}$. Table 1 shows the average and standard deviation of the values of $I_{2845}/I_{2930}$ for each group. The AD:normal ratio of the values of $I_{2845}/I_{2930}$ is $\sim 4:1$, indicating that ${\mathrm{CH}}_3$ symmetric stretching is suppressed in the AD brain sample. We speculate that this is due to changes in the chemical structure of lipids during the process of oxidative stress, which is implicated as a cause of AD.^30,31

Table 1

Average and standard deviation of I2845/I2930 in control (normal brain) and Alzheimer’s disease (AD) brain.