In vivo non-invasive staining-free visualization of dermal mast cells in healthy, allergy and mastocytosis humans using two-photon fluorescence lifetime imaging

Mast cells (MCs) are multifunctional cells of the immune system and are found in skin and all major tissues of the body. They contribute to the pathology of several diseases including urticaria, psoriasis, atopic dermatitis and mastocytosis where they are increased at lesional sites. Histomorphometric analysis of skin biopsies serves as a routine method for the assessment of MC numbers and their activation status, which comes with major limitations. As of now, non-invasive techniques to study MCs in vivo are not available. Here, we describe a label-free imaging technique to visualize MCs and their activation status in the human papillary dermis in vivo. This technique uses two-photon excited fluorescence lifetime imaging (TPE-FLIM) signatures, which are different for MCs and other dermal components. TPE-FLIM allows for the visualization and quantification of dermal MCs in healthy subjects and patients with skin diseases. Moreover, TPE-FLIM can differentiate between two MC populations in the papillary dermis in vivo—resting and activated MCs with a sensitivity of 0.81 and 0.87 and a specificity of 0.85 and 0.84, respectively. Results obtained on healthy volunteers and allergy and mastocytosis patients indicate the existence of other MC subpopulations within known resting and activated MC populations. The developed method may become an important tool for non-invasive in vivo diagnostics and therapy control in dermatology and immunology, which will help to better understand pathomechanisms involving MC accumulation, activation and degranulation and to characterize the effects of therapies that target MCs.

www.nature.com/scientificreports/ vessels and hair follicles 18 , which allows the transmition of local inflammatory signals to distant target cells and tissues 1 . Under physiological conditions, MCs appear to be in a resting state 19 . MCs are metabolically heterogeneous and cytoplasmic granules' maturation occurs throughout their entire life cycle 20 . Immature MCs contain a few large cytoplasmic granules with a few vesicles and lower histamine content 21 . In the process of maturation, MCs accumulate biologically active substances including heparin, histamine and various proteases including tryptase and chymase in their granules, which decrease in size and increase in number during differentiation 22 . The size of these granules ranges from 0.3 to 2 µm, while the size of dermal MCs is ≈10 µm 23 . MC activation via IgE-dependent and -independent stimuli induce a process of rapid release of their granular content into the extracellular space, a process called degranulation, leading to increasing vascular permeability and orchestration of the inflammatory responses 24 . In addition, lipid mediators such as prostaglandins and leukotrienes as well as cytokines and chemokines are generated by de novo synthesis and then secreted 23,[25][26][27] . Resting and activated/degranulated MCs are illustrated schematically in Fig. 1.
MCs are non-homogeneously distributed in the skin, and their concentration is increased at sites of inflammation in various pathological conditions including urticaria, psoriasis, atopic dermatitis and mastocytosis, where they are held to impact on the pathogenesis 12,28 . However, it is not yet clear how the accumulation of MCs contributes to local immunological reactions since MCs have been shown to exhibit both, inflammatory, as well as anti-inflammatory functions 29,30 . Also, MC accumulation may occur without any immediate pathophysiological consequences 10 .
To date, the routine method for visualization of MCs in the skin is immunohistochemistry of punch biopsies 31 , which requires surgical intervention and cannot be performed in patients with wound healing disorders. Other methods like electron microscopy and atomic force microscopy 1,32 are also not applicable in vivo. The imaging of fluorescent MCs has been described in the skin of transgenic mice via in vivo two-photon microscopy 18,33 . This technique has conceivable advantages over conventional in vivo microscopic procedures 34 . It allows, for example, for monitoring of fluorescence-labeled cell migration 35,36 . Using this technique, Tong et al. 37 developed the "immune atlas" of the skin, which describes the distribution of immune cells across different locations in the skin. However, in this study, the in vivo experiments were performed on anesthetized transgenic mice using fluorescent labelling.
The intrinsic fluorescence signal of the dermis originates from all dermal components and, therefore, the component-specific fluorescence is superimposed, making the fluorescence-based separation of certain molecules challenging 38,39 . Recently presented multi-channel two-photon/SHG (second harmonic generation)/threephoton/THG (third harmonic generation) in vivo rat tissue imaging 40 does not separate between cell types. In contrast, in vivo imaging of dermal MCs without labeling or the use of fluorescent tags in murine skin has been reported by Li et al. 41 based on the tryptophan-originated autofluorescence, which is not MC-specific.
There are clinical situations where a non-invasive technique for visualizing skin MCs may help to better manage patients, e.g. improve diagnosis (urticaria, mastocytosis, mastocytoma) without the need for a biopsy, to monitor the disease course and treatment responses in single cells. Current methods have major limitations such as the requirement for serial skin biopsies which would additionally come from different locations making a direct comparison difficult. Label-free in vivo imaging can overcome these challenges and allow for the investigation of MCs and their responses at a single cell level, at the same location over a long time period.
Thus, in order to solve the challenges associated with in vivo visualization of MCs in human skin, an imaging technique which is able to separate between component-specific fluorescence contributions and to measure at depths exceeding 70 µm, is required. Two-photon tomography in combination with the fluorescence lifetime imaging (TPT/FLIM) perfectly fulfills these requirements. The TPT/FLIM imaging technique allows to observe and screen skin in vivo deeper than 200 µm 42 , and provides the potential to separate between dermal components www.nature.com/scientificreports/ based on differences in two-photon excited fluorescence lifetime imaging (TPE-FLIM) signatures. This was recently shown for the main extracellular matrix proteins and capillaries of the papillary dermis in vivo 43 . The aim of this study was to visualize human skin MCs in vivo with a non-invasive and staining-free approach.

Results
characteristic tpe-fLiM signature of primary human skin Mcs in vitro. Cultured primary human skin MCs (hsMCs) were analyzed using TPT/FLIM. MCs were characterized by homogenous and densely packed cytoplasmic granules, had a close-to-spherical shape, and were ≈10 µm in size (Fig. 2a). Ionomycininduced MC activation resulted in the loss of cell membrane adjacent granules and a decrease in MC size by ≈1.5 µm (Fig. 2b) from 10.2 ± 0.8 to 8.7 ± 0.4 µm. Exteriorized granules and released content were occasionally seen in the culture medium. In addition, β-hexosaminidase activity was detected in the culture supernatant indicating degranulation (data not shown). The appearance of activated/degranulated MCs was less homogenous as seen in the τ m distribution and less round in shape compared to non-activated resting MCs. Analyses of the TPE-FLIM parameters (τ 1 , τ 2 , τ m , a 1 /a 2 and average two-photon excited autofluorescence (TPE-AF) intensity) revealed significant differences (p < 0.0001 for all parameters) between both groups with higher TPE-AF intensity and longer τ m for resting MCs (n = 43) compared to lower TPE-AF intensity and shorter τ m for ionomycin-treated, activated/degranulated (n = 13) MCs ( Fig. 2a-d; Table 1). 30% of the in vitro observed MCs are degranulated, while the remaining 70% are in a resting state.
In the tissue, MCs are sensitized by IgE bound to the high affinity receptor FcεR1 occupying most receptors. However, over time in culture the amount of MC-bound IgE decreases. To investigate the effect of IgE sensitization on the TPE-FLIM signature, cultured hsMCs were sensitized with human IgE by incubation with 1 µg/ml of purified plasma IgE for 1 h. As shown in the IgE-sensitized MC culture in vitro (Table 1), TPE-FLIM parameters τ 1 and τ m of MCs slightly decreased after sensitization of MCs with IgE (p < 0.05) for both resting and activated MC populations.
The distribution of TPE-FLIM parameters τ 1 , τ 2 and τ m of the major constituents of MC granules heparin and histamine, and corresponding phasor plot are presented in Supplementary Fig. 1. Tryptase has very weak A detailed analysis of the TPE-FLIM parameters obtained from bi-exponential regression of the fluorescence decay data is presented in the Supplementary Fig. 2. A clear separation is visible for τ 1 (τ 2 ), τ 1 (a 1 /a 2 ), τ 2 (a 1 /a 2 ) and τ 2 ((a 1 − a 2 )/(a 1 + a 2 )) plots. Hence, resting and activated/degranulated MCs can be readily distinguished by their distinct TPE-FLIM parameters.
Ex vivo dermal cell specific immunofluorescence confirms in vitro identification of MCs. Comparison of TPE-FLIM parameters of main papillary dermis cells, such as macrophages, dendritic cells and fibroblasts measured in vitro in the cell culture showed that resting and activated/degranulated MCs can be distinguished from other cell types (Table 1, Supplementary Fig. 3), indicating their potential successful visualization in the skin ex vivo and in vivo.
To further confirm the specificity of the MC signature, side by side comparison of human skin cryo-sections was performed ex vivo, which were first analyzed by TPT/FLIM and then followed by conventional immunohistochemistry staining with tryptase. As marked in orange ellipses in Fig. 3, the suspected MCs (Fig. 3a) are superimposed with the MCs, verified by staining (Fig. 3b). The contrast of shorter lifetimes of MCs compared to the longer lifetimes of the collagen and elastin dense dermis makes MCs distinguishable. In an effort to increase the sensitivity and specificity for MC visualization additional criteria such as MC size and shape were used. Moreover, in the TPE-FLIM signatures of these skin cryo-sections the MCs match the population of the MCs with a longer τ m (1 in Fig. 3a) and MCs with a shorter τ m (2 in Fig. 3a), measured in vitro in the culture as resting and activated/degranulated MCs ( Table 1). The proportion of observed activated/degranulated MCs is ≈44%. The TPE-FLIM parameters of the staining-proved MCs determined ex vivo are summarized in Table 1. The enlarged TPE-AF signal intensity is potentially related to the dehydration of the skin biopsy samples.
The 2D correlations of the TPE-FLIM parameters for the staining-proved MCs with a longer τ m (n = 9) and MCs with a shorter τ m (n = 4) measured ex vivo are presented in Supplementary Fig. 4 and show good separation in τ 1 (τ 2 ), τ 1 (a 1 /a 2 ), τ 2 (a 1 /a 2 ) and τ 2 ((a 1 -a 2 )/(a 1 + a 2 )) plots, which is similar to the separation performed for in vitro measurements ( Supplementary Fig. 2). To exclude false positive signals from other immune cells in the papillary dermis, sections were further stained for macrophages and dendritic cells and overlaid with the corresponding Table 1. TPE-FLIM parameters τ 1 , τ 2 , τ m , a 1 /a 2 and TPE-AF intensity of the pure and IgE-sensitized resting and activated/degranulated MCs, of the main dermal cells measured in vitro in the cell culture and of the shorter and longer τ m MCs measured in vivo on the volar forearm of healthy volunteers, allergic and mastocytosis patients and ex vivo in human skin cryo-sections.  www.nature.com/scientificreports/ TPE-FLIM images (Fig. 4, white/gray circles). Using the developed algorithm, neither macrophages nor dendritic cells were erroneously assigned as MCs (Fig. 4, orange circles). Fibroblasts, the most concentrated cells of the dermis, have normally an elongated shape and are not visible in the skin in vivo due to their low TPE-AF intensity in comparison to the surrounding extracellular matrix (mainly elastin). However, fibroblasts can be visualized in in vitro collagen matrix ( Supplementary Fig. 5a-d).
In the cell culture, fibroblasts are round in shape and their TPE-AF intensity is comparable to other dermal cells. However, fibroblasts are spreading in the collagen matrix and adopt their elongated shape. The TPE-FLIM parameters for six fibroblasts, measured in the collagen matrix, are shown in the Supplementary Fig. 5e and are summarized in Table 1.
In summary, all cells that were recorded and marked by TPE-FLIM stained positive for tryptase in immunohistochemistry demonstrating TPE-FLIM signatures to be a reliable tool to identify MCs in frozen skin biopsies.

TPT/FLIM identifies human skin MCs of healthy volunteers in vivo. Based on our in vitro and
ex vivo data we hypothesized that MCs' TPE-FLIM signatures can be clearly identified in the skin in vivo using TPT/FLIM. To identify the MCs, we developed a search algorithm, which is presented as flowchart in Fig. 5a. This algorithm assumes that skin MCs are present in the dermis and thus are expected at depths exceeding 70 µm, directly inside the dermal papillae, in close proximity to the blood capillaries. Thus, the focal spot of TPT www.nature.com/scientificreports/ should be moved into the papillary dermis and the capillary loop should be found by fast scanning (for instance, 3.2 s per image, typical image size ≈150 µm × 150 µm, power 25-50 mW) regime. The dermal cells have higher TPE-AF intensity than surrounding tissue in the dermis (the epidermal content is usually highly fluorescent). Therefore, "bright fluorescent spots" should be found and zoomed in to find cell-like structures. MCs should have an average size of ≈ 10 µm with close-to-round shape in loose connective tissue and an elongated to ovoid shape when in close proximity to blood vessels 44,45 . When such a cell is found, a TPT/FLIM image should be recorded and the TPE-FLIM parameters should be compared with known TPE-FLIM parameters of resting and activated/degranulated MCs (Fig. 2, Table 1). Matching of these TPE-FLIM parameters indicates if the measured cell is resting or activated/degranulated MC. A representative example of TPE-FLIM visualized MCs in the papillae is presented in Fig. 5b. The image was recorded from the skin of a healthy volunteer at the inner forearm at the depth of 80 µm. Capillaries (1 in Fig. 5b) surrounded with elastin (2 in Fig. 5b) and collagen III (3 in Fig. 5b) and two MC populations characterized by low TPE-AF intensity (≈ 300 photons/mW) and longer τ m (> 1,000 ps) (4 in Fig. 5b) as well as by high TPE-AF intensity (≈ 800 photons/mW) and a shorter τ m (< 800 ps) (5 in Fig. 5b) are clearly visible in the TPE-FLIM image of the papilla. An exemplary application of this algorithm is demonstrated in Fig. 6. Taking the MC density into consideration, determined as ≈174 ± 120 MCs/mm 2 for 10 µm biopsies in the papillary dermis using MC-specific staining of cryo-sections ( Supplementary Fig. 6), we assume that an image of ≈150 µm × 150 µm will contain ≈1-4 MCs over the whole depth of 120 µm, which was observed in the experiments. In the first stage, during the TPT fast scanning (3.2 s per image, typical image size ≈150 µm × 150 µm), "bright fluorescent spots" of 10 µm and certain shape, established in vitro and ex vivo were searched in the papillary dermis (typical depths ≈80-120 µm) using TPE-AF/SHG imaging (Fig. 6a).
The TPE-AF intensity of the "bright fluorescent spots" was considerably higher in comparison to the fluorescence background of the surrounding extracellular matrix, whose fluorescence intensity is due to elastin and collagen 43,46 (Fig. 5b). Subsequently, the image magnification was increased in order to focus on the "bright fluorescent spots" (Fig. 6c, e) and the TPE-FLIM images (Fig. 6d, f) were recorded for further analysis. A celllike structure with the size of inner granules between 0.5 and 1.5 µm, that is typical for MCs 47 , was often visible  Table 1.
The recorded dermal cells could be grouped according to their TPE-FLIM parameters into cells with longer τ m (> 1,000 ps) and shorter τ m (< 800 ps). The obtained TPE-FLIM parameters (Table 1) match well with the ones obtained from cultured hsMCs before and after activation with ionomycin ( Fig. 2c; Table 1) suggesting that the recorded dermal cells are MCs of two populations. However, TPE-AF intensities were inversed. MCs with a longer τ m presented with a significantly lower intensity compared to cells with a shorter τ m ( Table 1). The reason for this difference remains to be investigated. MCs with a longer τ m (Fig. 6c, d) had a circular cell body and substructures barely distinguishable, due to tightly packed granules. In contrast, MCs with a shorter τ m (Fig. 6e, f) had an elongated shape and clearly distinguishable granularly substructure. In total, 119 MCs were recorded. Analysis of the TPE-FLIM signatures revealed 71 MCs with a longer τ m and 48 MCs with a shorter τ m (Fig. 6g, h, Table 1).
Importantly, TPE-FLIM signatures of the two MC populations identified in vivo are significantly different (p < 0.0001 for τ 1 , τ 2 , τ m , a 1 /a 2 and average TPE-AF intensity, Table 1) and resemble these of cultured MCs in vitro (Figs. 1, 6; Table 1). Figure 6h shows the phasor plot of one exemplary MC with a longer τ m (blue points) and one MC with a shorter τ m (orange points) measured in vivo and presented in Fig. 6d, f. Both MC populations are partly superimposed in the phasor plot, but can be well separated.
Approximately 67% of the dermal MCs corresponds to the MC population with a longer τ m , thus matches the resting MC population from in vitro experiments. However, the MCs with a shorter τ m , matching the activated MC population measured in in vitro experiments, could be also detected in vivo. Their visualization is much easier due to the higher TPE-AF intensity ( Table 1).
The presented results for the decision tree classifier were calculated for 407 cells, the cell distribution is described in the materials and methods section. The parameters for the classifier are criteria of quality of a split is Gini impurity, minimal samples for split is 2, maximum depth of the tree is 6, and the samples had equal weight. The test set was 40% of the whole set and the training set included the remaining 60% of the whole set and split was tested 1,000 times. The first model classified MCs against other cells. The sensitivity was 0.86 ± 0.03 and the specificity was 0.82 ± 0.07, which suggests a high accuracy of 0.84 for the TPE-FLIM method for MCs. When activated/degranulated MCs were classified the sensitivity was 0.87 ± 0.04 and the specificity 0.84 ± 0.04. For resting MCs, the sensitivity was 0.81 ± 0.06 and the specificity was 0.85 ± 0.03 ( Supplementary Fig. 8). These results meet our expectations for a robust classification. Thus, we see that resting and activated/degranulated MCs can be separated from other cells in the skin with high accuracy without additional staining.  (2) and collagen III (3) and located in close proximity to blood capillary MCs characterized with a longer τ m (> 1,000 ps) (4) and a shorter τ m (< 800 ps) (5), measured in vivo without staining in the volar forearm at 80 µm depth (b). The mean fluorescence lifetime τ m is shown by the color gradient ranging from 300 to 1,300 ps. Scale bar: 10 µm. www.nature.com/scientificreports/

TPT/FLIM identifies human skin MCs of mastocytosis and allergy patients in vivo. TPE-FLIM
images from four mastocytosis patients (total of 12 cells) and three allergy patients (total of 17 cells) were further recorded. For one allergy patient the IgE in serum is increased at 110.0 kU/l compared to the reference value of < 85.0 kU/l and increased IL-8 at 314 pg/ml compared to < 15 pg/ml, which can be produced by MCs 48 . In the lesional skin of mastocytosis patients, the number of MCs was visually higher as compared to non-lesional skin areas and the skin of healthy volunteers. As in volunteers, the majority of MCs were of the longer τ m population (resting MCs). Importantly, the appearance of MCs in the lesional and non-lesional dermis of mastocytosis patients was similar to that of MCs of healthy volunteers (see Figs. 6c, 5d). Dermal mastocytosis MCs were round or slightly elongated in shape and ≈10 µm in size with very densely packed granules (Fig. 7a-c). Also, TPE-FLIM signatures of MCs recorded from mastocytosis patients with longer τ m (n = 9) as well as shorter τ m (n = 3) were not significantly different from MCs of healthy volunteers (Table 1; Fig. 7a-c), showing ≈ 75% resting MCs in mastocytosis and ≈ 60% resting MCs in healthy subjects. In contrast, a majority of MCs recorded from the allergy patient were MCs with a shorter τ m (Table 1; Fig. 7d-f). The cytoplasmic granules of 0.5-1.5 µm size showed a high TPE-AF intensity, which was not previously observed. Moreover, this patient had visually more MCs of both populations compared to healthy volunteers.

Discussion
In this study, we demonstrate for the first time, that the non-invasive TPT/FLIM technique can identify MCs in the papillary dermis of healthy and diseased human skin in vivo and distinguish them from other cell types and dermal structures by specific TPE-FLIM signatures. In addition, in vivo TPE-FLIM imaging can differentiate between two skin MC populations, characterized by either longer or shorter τ m , which match the TPE-FLIM signatures of resting and activated cultured MCs, respectively. Ionomycin-induced MC activation in vitro results in significant (p < 0.0001) changes in all TPE-FLIM parameters (τ 1 , τ 2 , τ m ) towards shorter lifetimes (Fig. 2d). Importantly, ionomycin itself did not show any detectable fluorescence signal. Therefore, changes in TPE-FLIM parameters are the result of cellular changes. The phasor plot, showing the fluorescence decay parameters' distribution for two representative MCs, indicated in Fig. 2a, b, that blue points correspond to the longer τ m (1 in Fig. 2a) and orange points to the shorter τ m (2 in Fig. 2b) MC phenotypes, which can be distinguished in vitro and in vivo (Figs. 2c, Fig. 6). The TPE-FLIM parameters are characteristic for MCs and distinguishes them from other cells in the papillary dermis, such as macrophages, dendritic cells, neutrophils and fibroblasts (Table 1, Fig. 4, Supplementary Fig. 3). Therefore, the TPE-FLIM signature can be used to visualize MCs in human skin in vivo.
Considering the presence of different cell types in the dermis, the identification of the origin of the fluorescence signal is crucial to distinguish between different cell types. Generally, at 760 nm excitation, the fluorescence signal from cells is due to the presence of free and bound NAD(P)H 49 with fluorescence lifetime components centered at τ 1 ≈3 60 ps and τ 2 ≈ 2,200 ps 50,51 . For different cell types, as well as for changes in cell metabolism or activation redistribution between NAD(P)H forms and its binding to different enzymes may cause pronounced changes in TPE-FLIM signatures, which enables the separation between cells at different stages of differentiation, normal and cancer cells, cells with oxidative stress, etc. 39 Fig. 1), probably due to the formation of oxidation products or aggregation-induced fluorescence [56][57][58] . Moreover, the comparison of the phasor plots for MCs in vitro (Fig. 2c) and heparin and histamine solutions ( Supplementary Fig. 1b) reveals a correlation between heparin, histamine and the MC activation state. The TPE-FLIM distributions presented in Supplementary Fig. 1 are significantly different from the parameters of NAD(P)H 50,51 . Moreover, the data indicate an impact of IgE molecules bound to the high affinity receptor FcεRI on the TPE-FLIM signature, in particular τ 1 and τ m ( Table 1). These data lets suggest that TPE-FLIM signatures might be MC-specific, yet further investigations are required to reveal the exact nature of fluorescence in the MCs and the FLIM signal.
For in vivo applications focusing on MCs, it is essential to selectively and specifically differentiate dermal MCs from other TPE-FLIM signals in the dermis, originating from the extracellular matrix, whose fluorescence is mainly provided by elastin and collagen III. The data show that dermal MCs have a higher TPE-AF intensity and are visible as "bright fluorescent spots" of specific shape, size and τ m distribution in the skin (Fig. 6a). Importantly, their TPE-FLIM parameters are significantly different to those of the surrounding extracellular matrix components, such as elastin, collagen III and blood capillaries (green-blue and dark red in the FLIM images, Fig. 6b, Supplementary Fig. 3) and match the TPE-FLIM parameters of cultured MCs measured in vitro (Fig. 2d, Table 1). Moreover, a precursor for the presence of cells in the dermis is the collagen I-free niches, which are visible by measuring the collagen I-specific SHG signal (red in Fig. 6a, c, e). In addition, MCs can be clearly distinguished from other dermal cells such as fibroblasts, dendritic cells and macrophages. They are localized in the proximity of blood capillaries, which are visible in dark red in the TPE-FLIM images (Fig. 6b, d, f) due to ultrafast fluorescence decay of Hb photoproducts 59 . The presence of cytoplasmic granules is a specific hallmark feature of MCs. Additionally, the size of measured cells was always ≈ 10 µm, which is specific for MCs. As can be seen in the phasor plot ( Supplementary Fig. 3), TPE-FLIM parameters of resting MCs and fibroblasts are partly superimposed. However, fibroblasts are characterized with substantially lower TPE-AF intensity in comparison to MCs. Moreover, the size of fibroblasts in the dermis (> 20 µm) is much larger than of other dermal cells (MCs, macrophages, dendritic cells, etc.) with a typical size of ≈10 µm, and taking into account that fibroblasts do not contain distinct granules, it can be concluded that fibroblasts cannot be visualized in the dermis in vivo at the detection parameters used in this work. Therefore, fibroblasts could not be mistaken as MCs in the skin during in vivo or ex vivo experiments. It should be taken into consideration that due to lower TPE-AF intensity of resting MCs, their concentration in the skin could be underrepresented and, thus their percentage is most likely higher than the reported 67%.
The ex vivo human skin biopsies stained for the presence of MCs were measured using TPT/FLIM (Fig. 3). The ex vivo MCs with a longer τ m shows an elongated shape that had never been observed in the in vivo experiments. In total, 17 histology-matched MCs (nine MCs with a longer τ m and eight MCs with a shorter τ m ) were recorded, which all showed an elongated shape and a size below 10 µm, which is probably a result of stress induced during freezing, excision, water loss and staining of the skin biopsies in the ex vivo experiment, which can initiate spontaneous MC degranulation and changes of MC size, shape and elasticity in comparison to the in vivo experiment. As can be seen in Fig. 3a, the orange-marked areas have different τ m values than the main extracellular matrix component elastin (τ m = 1,600 ± 110 ps). The nucleus is not visible, probably due to the masking of the TPE-AF by the granules' signals, which are observed in this image as clouded or dotted points, www.nature.com/scientificreports/ as confirmed in the staining image (Fig. 3b). The images do not show a perfect superposition. Yet, the burned square and the epidermal features ensure the same skin position. The distortion is most likely caused by the staining of the biopsy section. Under physiological in vivo conditions, human dermal MCs should be in a resting state and their degranulation is induced by IgE-independent triggers such as neurotransmitters or complement or IgE-mediated, e.g. by cross-linking of surface-bound IgE with corresponding antigen/allergen 2,60,61 . MC activation induces changes in the TPE-FLIM signal and a shift towards a shorter τ m . However, degranulation could not be confirmed in ex vivo and in vivo experiments. Therefore, it cannot be definitely concluded, that the measured MCs with a shorter τ m correspond to degranulated MCs. They could also correspond to MCs in an early differentiation stage, or even to a distinct MC population with a distinct composition of MC granules. For this reason, MCs with a shorter τ m are called activated/degranulated MCs. These aspects should be a subject of future studies.
In the allergy patient with increased IgE value the fluorescence lifetimes are shorter, similar to the in vitro results, which is an additional indicator that the observed dermal cells are indeed MCs. The TPE-FLIM parameters of MCs in allergy and healthy skin were only slightly different (p > 0.05) with obvious differences presented in the phasor plot for the resting MCs (Fig. 7i). MCs of allergy patients with a longer τ m determined as resting MCs show a broader distribution with a minor superposition with the activated/degranulated MC population of healthy volunteers. The broader distribution of the resting MCs in allergy compared to healthy indicate that resting MCs are presented in at least two subpopulations in allergic skin. The reason for that can be an influence of allergy on the MC maturation or degranulation processes, which require additional investigations on a larger cohort of patients. In the mastocytosis patients, a higher density of MCs than in healthy subjects was observed. This leads to the conclusion that the observed MC distribution is consistent with the criteria for mastocytosis with a higher percentage of resting MCs compared to allergy and normal skin. The TPE-FLIM parameters of MCs in mastocytosis and healthy skin were only slightly different (p > 0.05). However, some differences are obvious in phasor plot for both resting and activated/degranulated MC populations (Fig. 7i). It is obvious that each MC population consists of at least two subpopulations. This assumption as well as the origin of the observed changes should be a topic of future investigations. It is unknown that there are significant differences in IgE in mastocytosis patients compared to healthy volunteers. It is assumed that the resting and activated/degranulated MC populations are identical between healthy and diseased skin, however the results of this pilot study performed on allergy and mastocytosis patients suggest an existence of other MC populations, which should be proved in future investigations.
Data for different types of cells can be partially divided based on the TPE-FLIM parameters and the phasor approach, however, there are superimposed regions that do not allow 100% separation. Since various parameters were determined, such as shape of the cells, TPE-AF intensity and bi-exponential approximation of lifetime decay, and representation of data in the phase space, it was decided to combine some of these data and construct a classification rule to search for hidden dependencies. The basic decision tree classifier, which is available in the public domain and was proven in practice, was used 62 . The feature vector was created by combining the characteristics described in the materials and methods section and had the final dimension: 1 (circular form) + 1 (normalized TPE-AF intensity of cell normalized by area and power) + 256 (decay curve) + 7 (TPE-FLIM parameters obtained after bi-exponential approximation of decay curve) = 265. The exponential model is described in the statistical analysis and classification algorithm section. The MC size in vivo could be slightly different from MCs measured in vitro in suspension and ex vivo in biopsies, therefore the MC size was not included in the feature vector for the classification model. As a classification accuracy, two parameters, sensitivity and specificity were chosen. These parameters reflect the quality of the classification and indicate Type I and Type II errors. The classification algorithm is robust with the decision tree classifier showing a sensitivity of 0.87 ± 0.04 and a specificity 0.84 ± 0.04 for activated/degranulated MCs. For resting MCs, the sensitivity was 0.81 ± 0.06 and the specificity was 0.85 ± 0.03. For the classification of MCs of both populations versus other cells, the sensitivity is 0.86 ± 0.03 and the specificity is 0.82 ± 0.07. The diagnostic ability is shown as a ROC-curve in Supplementary  Fig. 8. Despite the relatively low number of entries, for a machine learning approach, future challenges are to integrate the algorithm into a pipeline and to optimize the training data set in order to further increase the sensitivity and specificity. Considering these results, it is concluded that the observed "bright fluorescent spots" (Fig. 6) matching presented specific criteria are dermal MCs, which are characterized either by the longer τ m (Fig. 6d) or shorter τ m (Fig. 6f). Both MC populations can be well distinguished from each other (Fig. 6), other dermal cells and extracellular matrix components in the papillary dermis ( Supplementary Fig. 3). In order to show time course results of the same MC before and after activation treatment (such as ionomycin or IgE) the MC should be stabilized during the in vitro measurement, or activation should be performed during the ex vivo measurement, which is currently not realized and serves a promising future task.
The presented results demonstrate the possibility to identify, track and characterize MCs in human skin in vivo in a non-invasive and staining-free mode using the TPT/FLIM technique. To the best of our knowledge, selective in vivo staining-free imaging of dermal MCs and potentially also other cell types in human skin has, as of yet, not been described. Hence, our data let suggest a possible extension of the TPT/FLIM technique for the assessment of MC number and function within the dermal composition in vivo. www.nature.com/scientificreports/ power of 3-5 mW was applied in order to exclude potential photochemical processes or laser influence on cells.

Two-photon tomography in combination with fluorescence lifetime imaging (TPT/FLIM).
For 760 nm excitation, a 410-680 nm band pass filter was used to detect two-photon excited autofluorescence (TPE-AF) and a 375-385 nm band pass filter was used to detect the second harmonic generation (SHG) signal, which is specific to collagen I. The acquisition time to record one image was 6.8 s. The lateral and axial resolutions were approx. 0.5 µm and 1.2-2.0 µm, respectively 63 . Two-photon excited fluorescence lifetime imaging (TPE-FLIM) data were processed using the SPCImage software version 8.0 (Becker&Hickl, Berlin, Germany). Fluorescence lifetime decay in each pixel was determined by bi-exponential non-linear regression using the weighted least squares method, i.e. with fast (τ 1 , a 1 ) and slow (τ 2 , a 2 ) lifetime and amplitude decay components. The temporal shift of the signal in relation to the instrument response function was fixed. The intensity threshold was chosen depending on the image quality. The fluorescence lifetime decay was binned over the 48 neighboring pixels (binning value 3), yielding photon count values larger than 200 photons/mW for in vivo measurements and larger than 800 photons/mW for in vitro and ex vivo measurements, resulting in reduced residual errors. Introducing the average fluorescence intensity of the cell as the photon count at a binning value of 3, the intensity of the pixel in focus and the 48 neighboring pixels were calculated. The obtained lifetimes (τ 1 and τ 2 ) and amplitudes (a 1 and a 2 ) were exported and used for the evaluation of the TPE-FLIM distributions and 2D segmentation 43 . The mean lifetime was defined as τ m = (a 1 τ 1 + a 2 τ 2 )/(a 1 + a 2 ) and used as a characteristic signature of the fluorescence lifetime decay. The dependences of TPE-FLIM parameters such as τ 1 (τ 2 ), τ 1 (a 1 /a 2 ), τ 2 (a 1 /a 2 ) and τ 2 ((a 1 − a 2 )/(a 1 + a 2 )) were used for 2D segmentation and separation between different cell types. The τ 1 /τ 2 ratio was calculated by the SPCImage software for each pixel. The TPE-FLIM data were also analyzed and represented in phasor plots, which are based on the transformation of the fluorescence decay data in the frequency domain, whereas the decay is described as amplitude and phase values of the first Fourier component 64 . The phasor plots' x-axis is described by the cosine of the phase value multiplied by the amplitude, the y-axis represents the sine of the phase value multiplied with the amplitude 39,65 . The position of the mean lifetime is on the secant from τ 1 and τ 2 , the distance to the circle is given by the proportion of a 1 and a 2 . The TPE-FLIM data were normalized to maximum intensity and the threshold of 70% was set to represent in the phasor plot. The utilized TPT/FLIM system has been previously described in detail elsewhere 43 Cryo-sections were scanned using TPT/FLIM for cells with MC-specific TPE-FLIM signatures and the corresponding TPE-FLIM signature was recorded. To retrieve the TPE-FLIM measured position for histomorphometric analysis, skin biopsies were labeled by burning a squared area of 28 µm × 28 µm located near the suspected MCs. The burning was done with the laser at 760 nm using a power of 50 mW for 3 s at maximal zoom. The TPE-FLIM images containing suspected MCs with the quadratic label were recorded and the same sections were then used for histomorphometric analysis. All incubations were performed at room temperature unless otherwise stated. In brief, sections were fixed for 10 min in cold acetone (− 20 °C) and rinsed in TBS (Agilent, Santa Clara, CA, USA). Sections were blocked with serum-free protein followed by incubation for 1 h with anti-tryptase antibody (clone AA1) diluted 1:1,000 in antibody diluent (all Agilent, Santa Clara, CA, USA). Slides were rinsed three times with TBS and endogenous peroxidase was blocked with 3% H 2 O 2 in TBS for 5 min followed by incubation with anti-mouse EnVision + labelled polymer (Agilent Technologies, Santa Clara, USA) for 30 min. Slides were rinsed in TBS as before and incubated with AEC substrate-chromogen (Agilent Technologies, Santa Clara, USA) for 10 min. Nuclei were counterstained with Mayer's hemalum solution (Merck, Darmstadt, Germany). Stained MCs have a brown-red color, which enables to visually distinguish them from other cells and the extracellular matrix. After the staining procedure, target MCs and squared labels of the skin sections were identified by light microscopy and overlaid with TPE-FLIM images matching an appropriate magnification and image orientations.
To prove that the TPE-FLIM parameters of other dermal cells, namely, macrophages and dendritic cells do not match or superimpose with parameters of MCs, negative control measurements were performed. The procedure was similar as described for the verification of MCs in the skin biopsies using specific immunofluorescence, but six human skin cryo-sections were stained for the presence of macrophages and two for dendritic cells, respectively.
For staining of macrophages, the macrophage-specific anti-CD68 (clone ab955) (Abcam, Cambridge, UK) and anti-CD163 antibodies (clone GHI/61.1) (Miltenyi, Bergisch Gladbach, Germany) were used to account for M1 and M2 macrophage phenotypes, respectively. For staining of dendritic cells anti-CD11c antibody (clone B-Ly6) (BD Biosciences, Franklin Lakes, NJ, USA) was used after fixing the cryo-section for 10 min in cold acetone (− 20 °C) and rinsed in TBS. www.nature.com/scientificreports/ Label free in vivo imaging. The volar forearm skin of 27 healthy volunteers  year old, skin type II-III according to Fitzpatrick classification 71 ) without declared immune or skin diseases, disease-affected and non-affected skin areas of four untreated mastocytosis patients  year old, skin type II) and an unsymtomatic forearm of female atopic dermatitis patient (32 year old, skin type II) at thighs and forearms with reported allergies to grass pollen, dogs and cats with immune suppression therapy (oral application of daily 100 mg of Azathioprine, 20 mg cetirizine and 150 mg Ranitidine) were included for non-invasive in vivo screening of the papillary dermis (60-150 µm depths) using the TPT/FLIM technique. The volunteers were screened between April 2018 and November 2019. The oil objective of the TPT/FLIM was covered by a 150 µm thin cover glass slide embedded into a metal ring and fixed to the volunteer's forearm. A 10 µl droplet of water was placed between the skin surface and cover glass for maintaining the optical contact. The hair was occasionally removed using scissors without influence on the skin integrity. The TPT/FLIM technique does not have any known side effects, considering a correct power-depth profile for the laser.
Statistical analysis and classification algorithm. TPE-FLIM data for all MCs were recorded, and descriptive statistics was applied using Matlab R2016a (MathWorks, Natick, MA, USA). All values are given as mean ± standard deviation. Differences between distributions were compared using the non-parametric Kolmogorov-Smirnov test and p < 0.05 was considered significant. All machine learning classifications were carried out in Python 3.7 utilizing Scikit-learn 0.22 using a Decision tree classifier. The model was trained using training set sizes of 60% of the complete data set, involving all relevant cells and repeated for 1,000 times with randomized training and test sets in every cycle. Results are exported in form of a confusion matrix and subsequent parameters true positive rate (meaning of sensitivity) and true negative rate (meaning of specificity) are presented as mean ± standard deviation. These parameters reflect the quality of the classification and indicate Type I and Type II errors. For the decision tree 72 the TPE-FLIM parameter variables τ 1 , τ 2 , τ m, a 1 , a 2 , a 1 /a 2 , (a 1 − a 2 )/(a 1 + a 2 ), TPE-AF intensity, shape (where 1 is circular and 0 is non-circular) and decay curve, were used for each cell measured in vitro, ex vivo and in vivo. The exponential approximation of lifetimes was averaged over the whole cell and the fluorescent intensity normalized per mW power was averaged over 49 pixel. A total of 407 cells, where 85 were MCs in vitro, 17 MCs ex vivo, 148 MCs in vivo, 54 macrophages in vitro, 4 macrophages ex vivo, 58 macrophages in vivo, 14 dendritic cells in vitro, 6 fibroblasts in vitro and 21 neutrophils in vitro were used as input for the model. Given data vectors from x i ∈ R n , i = 1,…,l and a label vector y i ∈ R l , where a decision tree recursively separates the data into classes. With the mode m represented as Q . For each node a split θ = (j, t m ) has to decide with the feature j and threshold t m . The node splits the data into subsets Q left (θ) and Q right (θ) At the mode m the impurity is calculated by the impurity function H() With the parameter for minimised impurities the subsets are recourse until N m = 1.
The classification criterion has values of 0 for activated MCS, 1 for resting MCs and 2 for other cells, 0 for MCs and 1 for other cells for node m in the region R m and N m observation, the proportion of class k observations in node m is p mk = 1/N m x i ∈R m I(y i = k).
The diagnostic ability is presented in form of a receiver operating characteristic (ROC)-curve where the true positive rate is plotted against the false positive rate of the respective outcomes ( Supplementary Fig. 8).
ethics. Volunteers for intravital microscopy provided their written informed consent before participation.
Skin samples taken from plastic skin reduction surgery for MC preparation and all human skin investigated in this study were used after written informed consent was obtained, as approved by the ethics committee of the Charité-Universitätsmedizin Berlin (EA1-078-18, EA4-193-18, EA1-141-12) according to the Declaration of Helsinki (59th WMA General Assembly, Seoul, October 2008).