Simultaneous label-free autofluorescence and multi-harmonic imaging reveals in vivo structural and metabolic changes in murine skin

: Simultaneous quantiﬁcation of multifarious cellular metabolites and the extracellular matrix in vivo has been long sought. Simultaneous label-free autoﬂuorescence and multi-harmonic (SLAM) microscopy has achieved simultaneous four-channel nonlinear imaging to study tissue structure and metabolism. In this study, we implemented two laser systems and directly compared SLAM microscopy with conventional two-photon microscopy for in vivo imaging. We found that three-photon imaging of adenine dinucleotide (phosphate) (NAD(P)H) in SLAM microscopy using our tailored laser source provided better resolution, contrast, and background suppression than conventional two-photon imaging of NAD(P)H. We also integrated ﬂuorescence lifetime imaging with SLAM microscopy, and enabled diﬀerentiation of free and bound NAD(P)H. We imaged murine skin in vivo and showed that changes in tissue structure, cell dynamics, and metabolism can be monitored simultaneously in real-time. We also discovered an increase in metabolism and protein-bound NAD(P)H in skin cells during the early stages of wound healing.


Introduction
In vivo imaging to visualize dynamic biological processes in their natural environment is a primary goal for not only biological applications but also for clinical diagnosis and disease or treatment monitoring. The transition from in vitro to in vivo, and from pre-clinical animal to clinical human studies, can be quite challenging since many of the markers or contrast agents such as fluorescent dyes or nanoparticles that are routinely used for animal studies are not readily applicable or approved for human use. Fortuitously, intrinsic molecules within cells and tissues can provide meaningful contrasts for many purposes, including diagnostic and functional studies, when excited by nonlinear optical processes and visualized by labelfree optical imaging. Over the past three decades, nonlinear optical imaging technologies based on multiphoton-excitation fluorescence (MPEF) or multiphoton harmonic frequency conversions have been extensively developed into powerful platforms to image living cells or intact tissue [1][2][3][4][5][6]. Endogenous fluorescent biomolecules such as flavins, nicotinamide adenine dinucleotide (phosphate) (NAD(P)H) coenzymes, and melanin allow one to monitor the metabolic changes of tissue through MPEF microscopy [7][8][9][10][11][12]. Endogenous structural molecules such as elastin, collagen, and keratin, on the other hand, generate structural tissue contrast [13][14][15][16][17]. This information, usually difficult to obtain by conventional optical imaging means, allows scientists to understand living tissue without the perturbative effects of exogenous dyes or particles, contributing to new diagnostic strategies and ways to study metabolism [18,19]. Multiphoton imaging tools have transitioned from bench-top to intravital platforms, especially for in vivo imaging of skin [5,10,[20][21][22][23][24][25][26][27].
Similarities of excitation requirements between different multiphoton imaging modalities allow for the integration of multiple capabilities into a single microscopy platform, and as such, have been used by many groups for nonlinear optical imaging [17,[28][29][30][31][32]. However, when transitioned to in vivo imaging and measurements, scientists have encountered new challenges. The dynamic molecular and structural nature of living tissue requires multimodal images to be acquired simultaneously, in order to prevent motion artifacts, ambiguity, or information loss caused by the rapid changes in the sample. Sequential image collection, as used in most multimodal imaging systems, is unsuitable to fulfill the task. Simultaneous multimodal nonlinear microscopy requires careful selection of optical windows for all channels, as well as a laser source for simultaneous excitation of all modalities. Stringari et al. demonstrated simultaneous multicolor NAD(P)H and flavin adenine dinucleotide (FAD) imaging by utilizing wavelength mixing, such as sum-frequency generation, by overlapping two laser pulses synchronized both temporally and spatially [33]. You et al. recently demonstrated a simultaneous label-free autofluorescence multi-harmonic microscopy (SLAM) microscopy platform using a single widely-coherent supercontinuum light source combined with pulse shaping. This approach enabled high-speed and simultaneous acquisition of autofluorescence from NAD(P)H and FAD, and harmonic signals from structural molecules (e.g. collagen) and biological interfaces in living tissue following surgical exposure of the tissue site [34].
Finding the best imaging parameters for multiphoton microscopy is critical for expanding its applications in biology and medicine. Although SLAM microscopy best emphasized the simultaneous image acquisition of autofluorescence and multi-harmonic modalities, there has been no direct comparison between SLAM microscopy and conventional MPEF microscopy for NAD(P)H and FAD imaging. We implemented two laser sources in a single imaging platform and directly compared three-photon excitation fluorescence (3PEF) of NAD(P)H in SLAM microscopy using an 1100 nm laser and two-photon excitation fluorescence (2PEF) of NAD(P)H using a 750 nm laser. Although a higher-order nonlinear optical process and a longer wavelength for excitation is used, 3PEF shows better quality for imaging NAD(P)H by offering better spatial resolution, contrast, and background suppression compared to 2PEF. In addition, SLAM microscopy separates different autofluorescent molecules based on their distinct emission frequencies, but cannot distinguish the same fluorescent molecules in different conformations, which is critical to understand metabolic activity related to mitochondria function in skin cells. Therefore, we incorporated fluorescence lifetime microscopy (FLIM) into our SLAM microscope, which allowed for label-free detection of both free and bound NAD(P)H molecules in vivo. We demonstrate the capacity of this imaging system by simultaneously monitoring structural and metabolic changes in skin in real-time. We also used our microscope to explore metabolic changes during wound healing, and found a decrease of fluorescence lifetime during the early stages of wound healing, possibly due to the increase in protein-bound NAD(P)H in skin cells.

Methods
In order to excite NAD(P)H and FAD coenzymes simultaneously, a custom-designed widely coherent light source was developed and integrated into the optical setup shown in Fig. 1(A). We pumped a nonlinear photonic crystal fiber (PCF, LMA-PM-15, NKT Photonics) using a high-power femtosecond laser centered at 1040 nm (Femtotrain, Spectra-Physics, 10 MHz, 370 fs, USA) ( Fig. 1(A)). A Faraday isolator (Pavos Series, EOT, USA) was installed after the laser output to prevent laser back-reflection from optical components in the system. The PCF generates broadband pulses spanning from 870 nm to 1300 nm, which were subsequently tailored to a bandwidth from 1025 nm to 1175 nm (150 nm bandwidth) by a pulse shaper to program the excitation windows for MPEF, second harmonic generation (SHG), third-harmonic generation (THG), and FLIM simultaneously ( Fig. 1(B)). All of the nonlinear signals from the samples were acquired using a lab-built multimodal multiphoton imaging system integrated with an epi-detection microscope, and all nonlinear signal emissions were spectrally split into 4 different detection channels. The detection channels were designed to collect short-wavelength signals first. THG images, which appear in Channel 1, are generated by using a 405 nm dichroic beam splitter (Di02-R405-25 × 36, Semrock) and a 370 nm (bandwidth: 36 nm) bandpass filter (FF01-370/36-25, Semrock) to collect the THG signal excited by the 1100 nm (30 nm bandwidth) portion of the broadband pulse. To image NAD(P)H autofluorescence, a spectral window from 405-504 nm (FF01-451/106-25, Semrock) was selected for Channel 2 after a 520 nm dichroic beam splitter (FF520-Di02-25 × 36, Semrock). In addition, FLIM microscopy was integrated with this channel, using the same low-noise hybrid photon counting PMT (H7421-40, Hamamatsu), allowing both free and bound NAD(P)H to be distinguished according to differences in their fluorescence lifetimes. A time-correlated single-photon counting unit (HydraHarp 400, PicoQuant GmbH) was used for FLIM signal acquisition. Channel 3 was designed for SHG, by integrating signal from 542 nm to 566 nm using a bandpass filter (FF-01-554/23-25, Semrock) and a dichroic beam splitter edged at 593 nm (FF593-Di03-25 × 36, Semrock). The last channel, Channel 4, primarily acquired 2PEF signal from FAD, within the range from 593 nm to 700 nm, excited by the 1040 nm portion of the broadband pulse. A dichroic beam splitter and a bandpass filter were used to select the signal.
All samples were mounted on a motorized XY-axis piezo sample stage (SLC-24150-LC, SmarAct, Germany) for imaging. The objective lens (XLPLN25XWMP2, Olympus) was mounted on a piezoelectric positioning system (SLC-24120-LC, SmarAct) to control the imaging depth. A single image containing 256×256 pixels was collected at a speed of 2.5 frames per second. Typically, an average power of 6 mW was used at the sample for imaging.
To evaluate the performance of our tailored light source for simultaneous multimodal imaging, we compared imaging results with those acquired using a standard tunable femtosecond laser (Discovery Chameleon, 80 MHz, 100 fs, Coherent). Using this laser, 750 nm was used to excite NAD(P)H and 950 nm was used to excite FAD for 2PEF signal generation ( Fig. 1(B)). A flip mirror was used to quickly switch between the two laser sources ( Fig. 1(A)). We collected fluorescence signal from both NAD(P)H and FAD solutions, using both excitation laser sources, and compared the results in Fig. 1(C). The two excitation lasers have different repetition rates and pulse widths, but using the same input average power, we found that our tailored light source provides higher signal intensity, likely due to the higher peak power originating from the lower repetition rate. We used Channel 2 and Channel 4 to normalize the signal intensities from NAD(P)H and FAD solutions, respectively, and compared signal leaking into other channels. It was found that the tailored broadband source contributed to less leaking of NAD(P)H signal to longer wavelength channels, and was similar to the tunable source for FAD signal detection.
The FLIM images were analyzed by the FLIMfit software. Fluorescent decay curves were fitted using a 2-component exponential decay model to determine the lifetimes and relative contributions of free and bound-NAD(P)H at each pixel. To obtain sufficient photons for fitting the FLIM signal decay curve, each FLIM image was integrated 20 times (∼51 seconds).
Male BKS.Cg-Dock7 m +/+ Leprdb/J (db/db) diabetic mice (Jackson Laboratory, Harbor, Maine) were used in this study. All mice used in the experiment were anesthetized and imaged alive on the microscope stage. A monitoring system was used to monitor the respiratory rate for optimal isoflurane supply. The total exposure time under anesthesia was less than 1 hour for each mouse, and all mice survived the imaging experiments. An electric heating pad was used to keep the animals warm during imaging. For the wound healing study, hair from the back of the mouse was removed on Day 0. A 1 mm wound area was made in the shaved area using a punch biopsy tool on Day 1. Images near the wounded area were acquired on Day 0, 1, 3, 7, 10, and 14. All studies were conducted in accordance with the GSK Policy on the Care, Welfare, and Treatment of Laboratory Animals, and were reviewed and approved by the Institutional Animal Care and Use Committee at University of Illinois at Urbana-Champaign.

Better resolution and contrast from 3PEF imaging of NAD(P)H
Three-photon excitation is a higher-order nonlinear process that offers better spatial resolution [35,36]. 3PEF can dramatically reduce out-of-focus background noise, thus improving the signal-to-noise ratio (SNR) by orders of magnitude compared to 2PEF [2,[36][37][38][39][40]. Our tailored light source at wavelengths above 1100 nm allows for three-photon excitation of NAD(P)H. Figure 2(A) compares 2PEF and 3PEF images from the same location in the epidermal layer of in vivo mouse skin, using the tunable and tailored laser sources, respectively. The 3PEF image using our tailored light source clearly shows better image contrast and resolution, which was also compared through the intensity profile from the same region of interest as shown in Fig. 2(C).
Since nuclei are not generating NAD(P)H signal, we used the nuclei area as the background to compare the rejection of background autofluorescence. We selected 10 cells (as indicated in Fig. 2(B)) and using the cytosol region as signal and nuclei as background, we found that the SNR for 2PEF and 3PEF were 4.55 ± 1.38 and 1.23 ± 0.84, respectively, with statistical significance (p < 0.0005), clearly showing the effective rejection of background fluorescence using 3PEF. We also compared 2PEF and 3PEF of NAD(P)H autofluorescence from deeper tissue regions in the dermis, ∼40 µm from the skin surface. Figures 2(D) and 2(E) show 2PEF imaging of a skin area containing blood vessels using, respectively, 6 mW and 50 mW excitation laser energy at 750 nm. The blood vessels have low contrast due to the strong fluorescence background from the surrounding tissue. However, using 6 mW excitation at 1100 nm, the 3PEF image from the same field of view shows much better contrast (Fig. 2(F)). It was possible to clearly visualize the blood vessel walls (yellow arrows) in 3PEF, but not in 2PEF, even with a much higher incident laser power (50 mW). In addition, 2PEF imaging using 750 nm from the tunable laser generates strong autofluorescence signal from hair follicles (Figs. 2(G) and 2(H). Using long-wavelength excitation from our tailored light source, such signal from hair follicles can be significantly reduced, allowing for higher contrast FAD images showing the FAD distribution in skin and vascular structures.
The above results indicate our tailored laser source for 3PEF provides better imaging quality and background fluorescence suppression compared to conventional 2PEF imaging of NAD(P)H and FAD.

Simultaneous in vivo metabolic and structural imaging of murine skin
Nonlinear optical imaging provides intrinsic sectioning capability, which allows for rapid 3D imaging. To evaluate the performance of our multimodal microscope, we imaged mouse skin in vivo in 3D. To extend the original 4 imaging modalities in SLAM microscopy, we integrated an additional FLIM channel with the 3PEF imaging of NAD(P)H. This enabled the enhanced capability to differentiate NAD(P)H in the free or protein-bound form in samples. These 5 simultaneous imaging modalities provide structural tissue information (from THG and SHG) as well as metabolic information (from NAD(P)H, FAD, and FLIM) imaging. Figure 3(A) shows selected sections from mouse skin from the surface to ∼50 µm in-depth, which includes the surface stratum corneum down to the dermis. The images from THG reveal overall membrane and extracellular structures, while SHG highlights collagen fibers found only in the dermis. NAD(P)H signal was found mainly in the cytoplasm in the stratum basale layer and in elastin fibers in the dermis, while the FAD signal was found largely in the stratum basale. Based on the NAD(P)H and FAD intensity signals, the redox ratio, which is closely related to the cellular states and conditions from the sample, can be calculated by FAD/(FAD + NAD(P)H), as in our previous publication [34]. Visualization 1 displays depth-resolved 3D imaging of skin using all modalities. We found that the redox ratio is lower in the epidermis, and higher in the dermis, indicating different metabolic rates in different layers, which agrees with other observations of mitochondrial activities [41]. FLIM images, on the other hand, show distributions of free and bound NAD(P)H in cells and in the extracellular matrix. With these images, we found evidence that free and bound NAD(P)H are not homogeneously distributed in the epidermis and dermis. Figs. 3(B) and 3(C) show magnified FLIM images from the stratum corneum and the stratum basale layers, highlighting the inhomogeneity of fluorescence lifetime in skin cells, possibly due to the inhomogeneous distribution of free and protein-bound NAD(P)H in mitochondria [41].
To demonstrate the capability of our multimodal optical imaging platform for monitoring cellular dynamics, we performed time-lapse mouse skin imaging, both in vivo and noninvasively. We imaged the dermal layer, which has an abundant network of blood and lymphatic vessels. Time-lapse images were acquired over a period of 180 seconds where the time interval between each image was 2 seconds (Visualization 2). Figure 4(A) shows THG, 3PEF, SHG, and 2PEF images from this dermal layer, where blood vessels are present. Simultaneous multimodal imaging concurrently captured particle movement and displacement/disruption of collagen structures. Figure 4(B) shows time-lapse images from four imaging channels at a 2 s time interval. Particle movement was clearly seen in all THG, 3PEF, and 2PEF channels (indicated by arrows in these channels; dashed lines are used to show the relative position of the particle). These image features suggest the particle is likely a cell moving outside the blood vessel. In addition, the SHG channel reveals displacement/disruption of collagen structures during the particle movement (indicated by the arrows in the SHG channel; dashed lines are used as position reference). These results indicate that simultaneous multimodal imaging allows for monitoring metabolic and structural signatures of tissue at the same time. Such highly dynamic information cannot be obtained with sequential serial collection and then computational assembly or merging of multimodal images. . Image interval is 2 seconds. Arrows point out the moving particle in each image. For THG, TPEF, and 2PEF images, the arrows point to a microparticle, likely a cell. For SHG, the arrows point to collagen fibers. Dashed lines are used as location references.

Simultaneous measurement of metabolic changes during wound healing
To further validate our multimodal imaging platform for the study of in vivo metabolism, we tracked the wound healing of mouse skin over the course of 14 days. This longitudinal animal study enabled comparisons between wounded and non-wounded skin using both the calculated redox ratio from the autofluorescence intensity, as well as from the FLIM measurements. The animals were placed into two groups: non-wounded diabetic mice and wounded diabetic mice, with 2 mice in each group. Diabetic mice were used here because their wound healing process is slower and less characterized than for wild type mice. Although the wound healing process has been widely studied using wild-type mice [42,43], the metabolic activities of the healing process in diabetic mice are less understood. Several previous studies used multimodal imaging microscopy to evaluate the ear skin wound healing process in diabetic mice [44,45]. However, while this location is often preferred because of accessibility, low density of hair, and thin skin architecture, different skin regions likely have different healing characteristics and rates. Here, we applied our imaging system to investigate the longitudinal changes of skin metabolism in diabetic mice during the healing of a wound area on the dorsal (back) skin region. Figure 5(A) is a photo that illustrates a typical wound area and an imaging area on the dorsal skin. Figure 5(B) is a photograph showing an anesthetized mouse on our imaging stage, positioned for in vivo imaging of the wounded skin area on the back. We could not find cells at the center of the wound. Therefore, we performed large-area mapping of the wounded area and determined the best imaging sites. We found that skin tissue surrounding the wound center shows similar metabolic features, reflected in both FLIM and multiphoton images, indicating a relative homogeneous metabolic signature of the wound-surrounding cells. We selected one field of view containing many skin cells for each time point for quantitative analysis. Figure 5(C) shows representative redox ratio images from the non-wounded (top) and wounded (bottom) groups collected at different days. The imaging location for the wounded group was approximately 300 µm away from the wound site. Mean values of redox ratio intensities were calculated for each group and are plotted in Fig. 5(E). The results show that the redox ratio did not change significantly over the measurement period for the non-wounded group, while for the wounded group, a sharp increase in redox ratio was observed on Day 1, which subsequently rapidly fell back to the original baseline level after Day 3. This increase in the redox ratio suggests an overall increase in the level of FAD compared to NAD(P)H. Such a change is primarily attributed to the increase in oxidative phosphorylation through the electron transport chain in mitochondria, which converts NAD + to NADH, and FADH 2 to FAD. Therefore, the results indicate a surge in mitochondrial activity during the early days of wound healing, which is expected to be an essential process for healing [46].
From the FLIM channel ( Fig. 5(D)), we found that Day 1 corresponds to a sharp increase in the mean lifetime of NAD(P)H for the wounded group, while the non-wounded group reveals a short and stable lifetime over all 14 days (Fig. 5(F)). Longer fluorescence lifetime is associated with protein-bound NAD(P)H. The increased amount of protein-bound NAD(P)H suggests more protein activities at the beginning of wound healing. Such a change is likely related to NAD(P)H:ubiquinone oxidoreductase (mitochondrial complex I), a mitochondria membrane protein in the electron transport chain which is responsible for the oxidative phosphorylation process. This agrees with the results obtained from our redox ratio measurement. In addition, complex I is a major source of reactive oxygen species in mitochondria that contributes to superoxide production and cellular oxidative stress. It is also possible that our observation is related to superoxide molecule production during wound healing [47,48]. These molecules are signaling molecules that can upregulate vascular endothelial growth factor (VEGF) in keratinocytes to enhance wound healing [49,50].

Discussion
From this study, we show that our 3PEF imaging of NAD(P)H offers better resolution and imaging contrast compared with conventional 2PEF imaging. SLAM microscopy allows simultaneous visualization of skin metabolic and structural changes when applied to in vivo skin imaging. After the integration of FLIM microscopy with SLAM microscopy, we were able to separate free and bound NAD(P)H in living samples, which offers additional contrasts to understand the metabolic activities of the skin. We also demonstrated increases in skin cell metabolism and mitochondria activities during the early days of wound healing. These results highlight that SLAM microscopy, in combination with FLIM microscopy, can be a powerful label-free method to study skin metabolism and structures in vivo.
Our imaging technology can benefit from several future improvements. First, a fiber laser would help improve environmental stability for microscope operation and reduce the size of the system. In addition, fiber lasers might facilitate integration with PCFs, and largely reduce the amount of free-space optics. An all-fiber-based excitation light source is a future target for advancement. Second, the maximum imaging speed of our current galvanometer-based microscope is 1 frame per second to image an area of ∼200 × 200 µm with 256 × 256 pixels, which is not fast enough to capture all the dynamics occurring in vivo, such as blood flow or all cell movements in 3D. The imaging speed can be further improved using a resonant-mirror-based scanning scheme. Third, our FLIM imaging speed is much lower than the other four imaging modalities due to the requirement of collecting enough photons for the generation of the lifetime decay curve. The FLIM images acquired in this study were from the relatively static tissue areas and we did not observe strong motion artifacts. However, such a slow imaging speed limits our FLIM modality from imaging highly dynamic features in living samples. In the future, we will integrate a fast FLIM method developed by our lab [51] into this imaging platform. Fourth, the imaging platform is still constructed using an inverted microscope frame. While this is suitable for use in a pathology lab, for other clinical applications, it will be necessary to develop a handheld probe for maximum flexibility. Such a transition requires reducing the footprint of the system and using fiber delivery and collection of the excitation beams and signal photons, respectively. Several groups have been working on building flexible probes or even endoscopes for in vivo multiphoton microscopy [52][53][54][55][56]. Simultaneous imaging of 5 modalities using fiber delivery and signal collection remains a challenging task but will be explored in our future research.
Current image interpretation is primarily based on the optical properties of intrinsic biomolecules. Future work will also focus on validating conclusions obtained from this multimodal imaging platform using other approaches. For example, a large and more systematic biologically-based study will be carried out to explore protein-bound NAD(P)H levels during wound healing to understand the potential role of mitochondrial complex I and the release of superoxide molecules. Immunofluorescence imaging targeting complex I, silencing the related gene (e.g. NDUFS6) though shRNA, or using specific inhibitors (e.g. rotenone), would validate the role of this respiratory complex I during wound healing.

Conclusion
By implementing two laser sources in a single multimodal imaging platform, we have directly compared metabolic imaging of NAD(P)H and FAD using SLAM microscopy (excitation at ∼1100 nm) and more conventional multiphoton microscopy (excitation at ∼750 nm). We show excitation at longer wavelengths can improve resolution and contrast through higherorder nonlinear absorption, and reduce background fluorescence for skin imaging. The new incorporation of FLIM with SLAM microscopy allows for not only 5 imaging modalities with 4 channels of label-free contrast, but also the ability to detect both free and bound NAD(P)H distributions in vivo, offering a more comprehensive set of informative on metabolism. These capabilities, together with THG and SHG imaging which provide in vivo structural information, allow investigators to better understand both the physical and chemical changes in highly dynamic living tissues or organisms. This was demonstrated by imaging mouse skin in vivo, and by monitoring simultaneous metabolic and structural changes using SLAM microscopy. We also performed a longitudinal study to track changes in skin metabolism during wound healing and found both an increase in metabolic activity and a protein-bound NAD(P)H surge during the first days of the wound healing process. While these murine skin studies serve as representative examples for dynamic biological processes, the simultaneous analyses of in vivo structural and metabolic information will likely yield new diagnostic biomarkers for many types of diseases, as well as provide new metrics for monitoring responses to treatment interventions.

Funding
GlaxoSmithKline