1. Introduction

The last decade saw the emergence of nonlinear microscopy as a powerful tool for imaging optically thick biological specimens based on intrinsic emission. Two- and three-photon excited autofluorescence of endogenous fluorophores such as NAD(P)H, FAD, retinol, tryptophan, elastin and collagen enabled in-depth imaging of live tissues [1–3]. Furthermore, second-harmonic generation of collagen and other endogenous structural proteins increased the potential of nonlinear microscopy in applications such as tissue diagnostics [4–6].

Tissue autofluorescence has long been investigated for potential applications in medical diagnostics. It has been shown that there are significant differences in the autofluorescence spectra between normal and diseased human tissues [7–11]. Although autofluorescence spectroscopy showed apparent statistical successes in tissue diagnostics, the origin of the spectral differences observed is still not completely clear [12].

In the last decade, studies have been made to link autofluorescence spectra with the morphology of the tissue. One technique involves characterization of the autofluorescence spectra at different depths within the tissue [13–15]. Another technique allows simultaneous recording of spectra and image. This is known as spectral imaging and it has been applied to fluorescence microscopy [16, 17] and FRET measurements [18–20]. Also, there have been a limited number of studies on spectral imaging based on tissue autofluorescence [21–23].

In this study, we integrated autofluorescence spectroscopy with nonlinear microscopy. We designed and developed a system that is capable of imaging tissue autofluorescence and second harmonic with each pixel represented by a spectrum. A combination of 1) a high sensitivity detection system consisting of two prisms and a CCD array and 2) special UV-optics allowed us to record autofluorescence and second-harmonic spectral images of living tissues over a broad wavelength range. The 100-channel spectral images, visualized as “real-color” images, enabled us to identify morphological features. Furthermore, depth-resolved spectral analysis and linear spectral unmixing provided important clues to the origins of the tissue autofluorescence.

2. Experiment

2.1. Spectral imaging system

A diagram of the experimental setup is shown in Fig. 1. The excitation light source was a mode-locked Titanium: sapphire laser (Tsunami, Spectra-Physics, Sunnyvale, CA) generating 70 to 100 fs pulses of 1 W average output power at a repetition rate of 82 MHz. The laser light was attenuated by a dual ND filter wheel (Model 5254, New Focus, CA, USA) before passing an UV-VIS-IR achromatic lens (Bernhard Halle Nachfl., Berlin, Germany). The computer-controlled laser-scanning head consisted of an XY scanning mirror, an XYZ piezo translation (sample) stage (Physik Instrumente, Karlsruhe/Palmbach, Germany), and a microscope objective (Fluor 40X/1.30 oil immersion, 160 mm tube length, Nikon).

Light collected by the objective was filtered by a dichroic mirror and a short-wavelength pass colored glass filter set (total thickness 7mm, BG40, Schott). The spectrograph consisted of two dispersion prisms and a CCD camera (Princeton Instruments, Spec-10:2KBUV, 2048×512 pixels, 16-bit, ST-133 controller, typical read noise 8 e^- rms at 1-MHz digitization rate). Fluorescence spectra can be recorded at a maximum rate of 500 emission spectra per second at a nominal spectral resolution of 2.7 nm from 330 nm to 600 nm (100 channels). The spectra were corrected for the overall wavelength-dependent sensitivity of the system. In this work, the fluorescence spectral images (224×224 pixels, 100 channels/spectrum) were acquired at 2.1 ms per pixel with an average excitation power of 5 mW. No photobleaching was observed at this low power level.

 

Fig.1. he experimental setup for autofluorescence spectral imaging of live mouse tissues.

Download Full Size | PDF

2.2. Animal model

Two female inbred albino hairless mice (SKH1 HR, Charles River, Someren, Netherlands) were examined in this study. The experimental protocol was approved by the Committee on Animal Research of the Erasmus University Rotterdam. Prior to the experiments animals were fed on a diet free of chlorophyll (Hope Farms BV, Woerden, Netherlands) for a minimum of two weeks in order to remove the autofluorescence emission from mouse skin centered at 675 nm attributed to pheophorbide-a [24]. Before imaging, the mouse was anaesthetized using intra peritoneal injection of Hypnorm, 0.5 ml kg-1 (Janssen Pharmaceutica, Tilburg, Netherlands) and diazepam, 2.5 ml kg-1. These injections were repeated every hour until the end of the experiment. The duration of the experiment was not more than 5 hours. To prevent dehydration the mouse was also injected intra peritoneal with 0.3 ml of 0.9% sterile NaCl solution. The mouse was placed on a temperature-controlled microscope stage and a cover slip.

2.3. Spectral image data analysis

To visualize the three-dimensional (i.e. xy dimensions and wavelength channel) images, they were transformed into real-color RGB images. Formally, a 32-bit, 100-wavelength channel spectral image can be represented as image planes A _ij(m) where i and j denote the x- and y-positions in the image respectively (i,j=1 to 224) and m denotes the wavelength channel number (m=1 to 100). The transformation into an RGB image involved two main steps:

1. Data reduction to 8-bit, 11-wavelength channel spectral image:

where n is the new wavelength channel dimension (n=1 to 11) and k _n is the wavelength channel-dependent correction factor to compensate for the nonlinearity of the wavelength interval and N is the normalizing factor for the data bit-size reduction. Here, the first 25 wavelength channels (330 nm to 360 nm) were not used since they contain no information. The next 25 wavelength channels (m=26 to 50 corresponding to 380±20 nm) were summed up to form channel 1 of the reduced spectral image and each subsequent 5 channels were summed up to form wavelength channels 2 to 11 (400 nm to 600 nm) of the reduced spectral image;

2. Conversion of the reduced spectral image into a 24-bit real-color RGB image:

where r, g, and b are the red, green and blue respective values in RGB color space, T _rgb(m) is the wavelength channel-dependent real-color RGB value derived from approximations of the RGB value for visible wavelengths [25].

3. Results and discussion

Figure 2 shows real-color RGB representation of the nonlinear spectral images from living mouse tissue at different relative depths from 5- to 40-µm below the skin surface. These images represent sections that contain morphological structures most relevant to the study of skin. The spectral imaging system can typically image as deep as 100 µm. The system’s ability to image deep structures is predominantly limited by the index of refraction mismatch between the sample and the oil-immersion objective.

Individual cells are distinctly observed in the stratum spinosum at about 5 µm below the surface of the stratum corneum [Fig. 2(a)]. The blue fluorescence of these cells mainly originates from the cytoplasm. In previous studies, it has been suggested that the main source of intracellular autofluorescence is NAD(P)H [13]. It is fluorescent only when reduced and has a characteristic blue fluorescence peak at around 460 nm. Another source of redox related autofluorescence comes from cellular flavins [13]. In contrast to NAD(P)H, these molecules are fluorescent in their oxidized state and has a characteristic yellow fluorescence peak at around 535 nm. In between the cells, green-fluorescing structures [white arrow heads in Fig 2(a)] are observed which we assume to be tonofilaments or keratin filaments. The cells of the stratum spinosum are connected to one another by desmosomes and reinforced by tonofilaments.

A deeper section (10 µm) in the skin shows higher number of cells in the same image area [Fig. 2(b)]. The cells appear smaller in cross-section than the cells in Fig. 2(a). This is because these cells are more columnar in shape while the cells near the stratum corneum are flattened out. Hair follicles appear as round green-fluorescing structures (white arrows). These hair follicles, mostly inactive in these hairless mice, become more apparent at a depth of about 15 µm [Fig. 2(c)]. Here, the purple color in the image increased in intensity. This purple color refers to the second-harmonic signal generated by collagen. This is a clear indication that the cells in Fig. 2(c) are mostly basal cells of the epidermis near the dermal-epidermal junction.

At about 20 µm deep, the fiber structure of the collagen becomes more distinct [Fig. 2(d)]. Some basal cells are still present in this section. Rings of cells surround the hair follicles and are still visible at 30 µm deep within the tissue [Fig. 2(e)]. At this depth, the collagen fibers are thicker and bright blue-fluorescing fiber-like structures are observed along the fibers (yellow arrows). These blue fiber-like structures are likely to be elastin. At a depth of about 40 µm [Fig. 2(f)], the collagen and elastin fibers are more distinct. Also, a featureless green fluorescence structure is observed (red arrow).

 

Fig. 2. Real-color spectral images at: (a) 5 µm; (b) 10 µm; (c) 15 µm; (d) 20 µm; (e) 30 µm; and (f) 40 µm below the surface of the live mouse skin. The excitation wavelength is 764 nm and the objective is a 40X/1.30 Oil immersion objective.

Download Full Size | PDF

The average spectra for the 100-channel spectral images (same imaging region as shown in Fig. 2) at different measurement depths were plotted using a color map [Fig. 3(left)]. The average spectra at 5 µm, 10 µm, 15 µm and 40 µm are shown in Figs. 3(a)–3(d), respectively. Three main spectral components were observed: 1) narrow-band emission at 382 nm (black arrow); 2) narrow-band emission at 409 nm (white arrow) and; 3) broad-band emission from 450 nm to 550 nm (arrow head). Near the surface of the skin, the 382 nm and 409 nm components were relatively low in peak intensity compared to the broad-band component. However, measurements deeper than 15 µm show an increase in intensity of the 382 nm and 409 nm components and a drop of the broadband component. At about 30 µm from the skin surface, the whole spectrum gradually decreased in intensity.

The strong narrow-band emission at 382 nm is positively identified as the second harmonic generated by collagen. The intensity of emission increased with depth until about 30 µm, and it slowly decreased for larger depths. The increase in intensity of the second harmonic emission was interpreted as an increase in number of collagen fibers. The collagen is exclusively located in the thin dermis of the mouse skin. We believe the decrease in second harmonic intensity to be due to the decrease in collagen fiber density and the increase in scattering.

The weak narrow-band emission at 409 nm was observed starting at depths of 15 µm. This peak is depicted in the Fig. 3(c). We could not possibly attribute this peak to any known chromophore in the skin mainly because of its narrow band character. Moreover, since all the data presented here were corrected for the spectral sensitivity of the system, this peak can not be considered an artifact. As of this moment, the origin of this narrow peak remains to be identified in future studies.

The real-color representation of the spectral images provided visual differentiation between the different tissue structures attributed to the differences in their biochemical makeup. Furthermore, spectral analysis at different tissue depths presented a coarse indication of the origins of the tissue emission. To further narrow down the fluorophores responsible for the tissue emission, particularly on the broadband emission from 400 nm and 600 nm, we applied linear spectral unmixing to the measured spectra [20, 21].

 

Fig. 3. Left: A color map of the skin emission spectra vs measurement depth. The white dotted lines correspond to the plots on the right. Right: Tissue emission spectrum at: (a) 5 µm; (b) 10 µm; (c) 15 µm; and (d) 40 µm. The excitation wavelength used was 764 nm.

Download Full Size | PDF

We fitted different combinations of spectra for each measured depth-resolved emission spectrum. These spectral components were mathematically modelled Gaussian curves with peaks and spectral widths that mimic the spectra of known endogenous fluorophores found in skin. A combination of five components was found to fit well with the spectra: collagen/elastin (430 nm) [26], NAD(P)H (460 nm) [26–28], keratin (475 nm) [29], and FAD (535 nm) [23, 26, 30] and a component of unknown origin (409 nm).

Shown in Figs. 4(a) and 4(b) are results of the linear spectral unmixing of depth-resolved emission spectra at 5 µm and 15 µm, respectively. At about 5 µm, three components, 460 nm, 475 nm and 535 nm components were found to fit well. The same components fitted the emission spectrum at 10 µm (not shown). At depths of about 15 µm, two additional components, 409 nm and 430 nm were required to fit the spectral data. A plot of the amplitudes of the fitted components as a function of depth is shown in Fig. 4(c). The increase of the collagen and elastin spectral components at depths of about 15 µm is in good agreement with the earlier analysis. The origin of dermal autofluorescence is mainly from collagen and elastin components. Collagen types I and III which represent the major fibrillar collagen types in skin, have been shown to be located throughout the dermis [31], generally together with small amounts of collagen V. On the other hand, the epidermal autofluorescence is mostly contributed by NAD(P)H, keratin, and FAD components.

It must be emphasized that the characterizations by spectral unmixing described here are, at best, approximations; they are not direct measurements. Many factors are at play, including tissue physiology and the interactions of multiple fluorophores with similar spectra. Further work must be done to characterize the fluorophores responsible for tissue fluorescence.

A movie of three-dimensional in vivo mouse skin tissue was produced from a series of real-color RGB images [see Fig. 4(d)]. It begins with an image of the outermost layer of the epidermis, the stratum corneum and ends in the dermal layer at 2 µm intervals.

 

Figs. 4. (a), (b) Linear spectral unmixing of the spectra obtained from 5 µm and 15 µm image sections, respectively. (c) Bar graph of the amplitudes of the fitted spectral components as a function of depth from the surface of the skin tissue. (d) (523 kB) Movie of three-dimensional in vivo mouse skin tissue. The movie is a series of spectral images in real color from the stratum spinosum of the epidermis to the dermis at 2 µm intervals. The field of view is 45 µm x 45 µm.

Download Full Size | PDF

4. Conclusion

In conclusion, we combined nonlinear microscopy with autofluorescence spectroscopy to simultaneously record images and emission spectra of the autofluorescence and second harmonic of live mouse skin tissues at different depths. The experimental results on tissue spectral imaging demonstrate that: 1) visualization of the spectral images as RGB images provides a direct impression on tissue morphology as well as identification of cellular and extra-cellular structures; 2) depth-resolved spectral analysis offers a simple method of discriminating tissue layers and; 3) linear spectral unmixing presents the possibility of identifying the main fluorophores that contribute to the tissue spectral emission. We find that linking the morphological information with the spectral information is a very powerful approach for tissue biochemical analysis. In the future, we plan to apply spectral imaging to in vivo human skin. However, the present depth limitation of the spectral imaging system of about 100 µm will allow us to image only until the basal layer of the in vivo human skin. Further improving the system, for instance by using a water-immersion microscope objective may permit deeper imaging into the dermis.