Macrophages (ΜΦs) are important immune effector cells in organs and tissues that act as border junctions to environments such as the gut, the airways, and the skin (Elhelu, 1983). Skin ΜΦs (Dong et al., 2016; Estandarte et al., 2016; Ryter, 1985) originate from circulating monocytes (Geissmann et al., 2010; Gordon and Taylor, 2005) via the same infiltration route into the dermis as monocyte-derived dendritic cells (Schmid and Harris, 2014; Figure 1a) and are mainly located in the papillary and reticular dermis in close proximity to blood vessels (Weber-Matthiesen and Sterry, 1990; Figure 1b–d). It has been known for more than 30 years that skin ΜΦs are abundant and heterogeneous, based on their morphology, localization, and staining properties (Weber-Matthiesen and Sterry, 1990). More recently, skin ΜΦs have been classified based on their function, and they fall into two phenotypes referred to as inflammation-promoting M1-polarised ΜΦs (classically activated) and anti-inflammatory M2-polarised ΜΦs (alternatively activated) (Arango Duque and Descoteaux, 2014; Figure 1a). M1 ΜΦs are activated by viral and bacterial infection (Benoit et al., 2008; Ferrer et al., 2019; Malmgaard et al., 2004), interferon-γ, lipopolysaccharide (LPS), and tumor necrosis factor (TNF), which is known as the classical activation pathway (Li and Liu, 2018). M2 ΜΦs are alternatively activated in response to IL-4, IL-13, and IL-33 (Furukawa et al., 2017; Sica and Mantovani, 2012). Recently, this paradigm was questioned, as a manifold of cytokines, biomarkers, and activators are involved in MФs functioning, resulting in a continuum of states between the M1 and M2 phenotypes (Mendoza-Coronel and Ortega, 2017; Murray et al., 2014). Furthermore, different markers like CXCL10 for M1 ΜΦs and CCL17 for M2 ΜΦs can have the function to attract t-cells. Macrophages in disease, cancer, or obesity can switch function from wound healing to inflammatory ΜΦs given the right signals and microenvironment (Mosser and Edwards, 2008). However, for simplicity, the terms M1 and M2 ΜΦs are used here with the activators in brackets, where applicable.

Figure 1

Download asset Open asset

Skin M1 ΜΦs are held to contribute to dermal innate immunity and homeostasis. This is supported by reports that M1 ΜΦs can phagocyte objects up to 20 µm in size (Morhenn et al., 2002), promote skin inflammatory and immune responses (Remmerie and Scott, 2018; Theret et al., 2019; Yanez et al., 2017), and produce nitric oxide and other reactive oxygen species (ROS) (Forman and Torres, 2001; Rendra et al., 2019). Skin M2 ΜΦs, on the other hand, are thought to promote dermal repair, healing, and regeneration, for example, by contributing to the formation of the extracellular matrix (ECM; Ploeger et al., 2013).

The precise role of skin ΜΦs and their M1 and M2 phenotypes in health and disease remain to be elucidated. In skin diseases, such as melanoma (Bardi et al., 2018), systemic sclerosis (Trombetta et al., 2018), Lupus (Chong et al., 2015), and LPS tolerance (O’Carroll et al., 2014), polarization of ΜΦs leading to mixed M1 and M2 phenotypes can be observed. It is not known whether and in what density mixed ΜΦ phenotypes are to be expected in healthy skin. The fluorescence properties of mixed phenotypes have not been studied.

Efforts to do so include their quantification in healthy human skin and in lesional and nonlesional skin of patients with skin diseases. Currently, the most common approach is to obtain skin biopsies and to visualize ΜΦs by immunohistochemistry. Skin biopsies, however, come with several important limitations, which include scarring, the risk of infection and bleeding, and artificial findings caused by the use of local anesthesia. In addition, histopathological analyses of skin biopsies are not well suited for characterizing ΜΦ functions such as phagocytosis and for long-term monitoring of ΜΦ distribution in the skin.

Fluorescence lifetime imaging (FLIM) employs NAD(P)H and fluorescence decay parameters of cellular compartments as specific indicators of cell types and phenotypes (Alfonso-García et al., 2016; Heaster et al., 2021). Combined with two-photon tomography, two-photon excited fluorescence lifetime imaging (TPE-FLIM) allows for label-free and noninvasive imaging of dermal cells. For instance, TPE-FLIM allows for in vitro imaging of mast cells, fibroblasts, neutrophils, and dendritic cells and in vivo imaging of mast cells in human skin (Kröger et al., 2020). Whether or not TPE-FLIM can be used to visualize human skin ΜΦs, their M1 and M2 phenotypes, and their functions, is currently unknown. There are, however, several independent lines of evidence that support this approach: First, previous studies have shown that TPE-FLIM can distinguish ΜΦs from other dermal cells and ECM, without prior labeling (Kröger et al., 2020). Second, the capillaries of the papillary dermis, which often are in close proximity to ΜΦs, show distinct TPE-FLIM signatures and are readily visualized (Shirshin et al., 2017). Third, M1 and M2 ΜΦs come with unique cytokine patterns, and the TPE-FLIM signatures of these cytokines and patterns could help to tell the two phenotypes apart. Finally, TPE-FLIM can distinguish between functional states of dermal cells, for example, resting and activated mast cells in vivo, ΜΦs ex vivo (Kröger et al., 2020), and T-cell activation in vitro (Walsh et al., 2021) may, therefore, potentially allow for monitoring ΜΦ functions in vivo (Szulczewski et al., 2016). Taken together, the morphological features of skin ΜΦs, their localization in the skin, and the expected differences in fluorescence decay parameters between ΜΦ phenotypes as well as between other dermal cells, make TPE-FLIM a promising strategy for their detection (Yakimov et al., 2019).

Here, we first investigated human skin ΜΦs, in vitro with clear M1 and M2 phenotypes, for their TPE-FLIM properties and how these differ from those of the main components of the ECM and other dermal cells such as fibroblasts, mast cells, and dendritic cells. We then applied the identified ΜΦ TPE-FLIM signatures to investigate M1 and M2 ΜΦs and their phenotypes in human skin biopsies, combined with traditional immunohistochemistry-based visualization. Finally, we used TPE-FLIM in vivo in humans to study skin ΜΦs, their phenotypes, and functions, and we developed, tested, and characterized TPE-FLIM signature-based machine learning algorithms for the detection of skin ΜΦs.

This is the first in vivo study to show that human skin ΜΦs can be distinguished from other dermal cells and quantified through visualization with label-free, completely noninvasive TPE-FLIM. This risk-free approach also allows for the identification of ΜΦ phenotypes, that is, M1 and M2 ΜΦs, and for the characterization of their functional stage, that is, resting versus phagocytosing M1 ΜΦs. Finally, TPE-FLIM can be used to implement sensitive and specific machine learning algorithms for ΜΦ detection in the skin.

Our initial work with CD-14 positive monocytes isolated from PBMC and then differentiated and polarised toward M1 (IFN-γ) and M2 (IL-4) ΜΦs was needed to establish their TPE-FLIM parameters. In fact, it showed that ΜΦs are fluorescence-active, and, more importantly, that their TPE-FLIM parameters are among the best differentiators of M1 (τ_m=479±106) and M2 (τ_m=1,185±170) ΜΦs. M1 ΜΦs are associated with a slightly higher TPE-AF intensity (Table 1), which is a prominent indicator for the metabolic stress of the cell on account of a shift in lifetimes by changing amounts of free and bound NAD(P)H (AlShabany et al., 2016) and generation of ROS in mitochondria, phagosomal vacuoles, and the cell membrane (Datta et al., 2015). Additionally to NAD(P)H, autofluorescence of lipids and other cell compartments was recorded. TPE-AF intensity is a parameter with limitation due to the nonlinear imaging technique. There is no linear correlation between excitation and emission intensity, also it is reduced due to scattering and absorption in the skin. The metabolism of LPS-induced M1 ΜΦs is characterized by higher glycolysis, indicating a shift toward shorter fluorescence lifetime (Li et al., 2020; Orihuela et al., 2016). The longer fluorescence lifetime τ₂ in M2 ΜΦs is best explained by oxidative phosphorylation and the emergence of fluorophores caused by fatty acid oxidation (Viola et al., 2019). NAD(P)H fluorescence is ubiquitously present in cells and exhibits the continuum of lifetimes in the 360–3400 ps range. Therefore, changes in TPE-FLIM parameters are likely a reason of the metabolic changes of the ΜΦs. Free NAD(P)H has a short lifetime of 360 ps. For bound NAD(P)H, longer lifetimes up to 2–4 ns have been reported (Alfonso-García et al., 2016). A higher ratio of bound to free NAD(P)H is associated with M2 MΦs resulting in longer TPE-FLIM parameters, while a lower ratio of bound to free NAD(P)H is associated with M1 MΦs resulting in faster TPE-AF decay (Blacker et al., 2014). Thus, a strong indicator for the ΜΦ polarization is the TPE-FLIM parameters of monocytes in between cohorts of ΜΦs.

It was observed that the quantity of fluorescence lifetimes in ΜΦs is vastly varying between M1 and M2 ΜΦs. Regarding the ΜΦ polarization, the paradigm shifts toward a less strict classification compared to M1 (IFN-γ/LPS-polarized) and M2 (IL-4-polarized). While this categorization is useful in clinical terms, the multitude of parameters leading to the differentiation process leaves ΜΦs with wide-ranging properties both in expression of markers and also in appearance and TPE-FLIM parameters (Murray, 2017).

M1 (IFN-γ/LPS-polarized) ΜΦs rely on the NADH oxidase and production of ROS, which is shown by fluorescent lifetimes of under 250 ps and mitochondrial fission, which can indicate the bright spots, whereas M2 (IL-4-polarized) rely on oxidative phosphorylation and fatty acid oxidation, together with mitochondrial fusion, it can explain the homogeneous appearance of M2 ΜΦs (Ramond et al., 2019; Swindle et al., 2002; Xu et al., 2016).

Translation of our in vitro findings to ΜΦs isolated from human skin confirmed that the latter share the TPE-FLIM signatures of the former, with shorter τ_m in dermal M1 ΜΦs and longer τ_m in dermal M2 ΜΦs. The classification into M1 and M2 ΜΦs in vitro based on their distinct TPE-FLIM parameters was supported by their differences in size, morphology and internal vacuole structure. That the τ₁ lifetime of dermal M2 ΜΦs is longer than that of monocyte-derived M2 ΜΦs is most likely due to the use of different polarization agents. ΜΦ colony-stimulating factor (M-CSF) and IFNγ for M1 ΜΦs and ΜΦ colony-stimulating factor (M-CSF) and IL-4 for M2 ΜΦs were used in PBMC-derived ΜΦs and microenvironment effects, like inflammatory signals, UV exposure (Kang et al., 1994), and immune responses (Theret et al., 2019) influencing ΜΦ functions, result in divergent fluorescence lifetimes (Zhang et al., 2014; Table 1, Figure 2d). The most important outcome of our work with dermal MΦs was the establishment of their phenotype-specific TPE-FLIM signatures, a prerequisite for our subsequent in vivo studies and for comparing skin MΦs and other dermal cells.

In fact, the use of the TPE-FLIM signatures of M1 and M2 ΜΦs clearly allowed to distinguish them from mast cells, dendritic cells, fibroblasts, and neutrophils (Table 1, Figure 4f; Kröger et al., 2020). We controlled this by independent markers. For example, ΜΦs exhibited a higher fluorescence intensity compared to other dermal cells. In addition, they were larger than dermal mast cells, and their morphology was clearly different from that of neutrophils and fibroblasts. Vacuoles, a defining feature of ΜΦs, were linked to ΜΦ TPE-FLIM parameters, whereas granules, which identify mast cells, were not. In short, dermal ΜΦs, in their TPE-FLIM profiles, do not superimpose with other dermal cells or structures. The only exception, a partial overlap between M2 ΜΦs and elastin, is not relevant for the visualization of the former, as they are readily distinguished from the latter based on their morphology.

To confirm that ex vivo TPE-FLIM overlap with TPE-FLIM signatures of M1 and M2 ΜΦs in vitro, we sequentially analyzed cells in skin biopsy cryo sections with TPE-FLIM and conventional immunohistochemistry. Indeed, both approaches identified and distinguished matching ΜΦ populations, that is, M1 and M2 ΜΦs, with strong fluorescence intensity and spindle shape appearance of M1 ΜΦs and lower fluorescence intensity and longer fluorescence decay in M2 ΜΦs (Figure 3). Interestingly, M2 ΜΦs are often found in an area of higher density of unknown dermal cells, presumably fibroblasts, compared to M1 ΜΦs. It is suspected that M2 ΜΦs in conjunction with collagen-synthesizing fibroblasts are acting toward and aiding in dermal repair and regeneration. However, this approach also revealed some challenges that come with ΜΦ visualization by TPE-FLIM. For example, it was more difficult to visualize M2 ΜΦs than M1 ΜΦs in biopsies due to the high fluorescence intensity of elastin in dried tissue and other ECM components and a decreased signal-to-noise ratio. In Figure 3d, more CD163 positive M2 ΜΦs are visible compared to the corresponding TPE-FLIM image in Figure 3c, which is due to previously mentioned challenges and the limited imaging plane of the two-photon tomograph (1.2–2.0 µm) compared to a significantly thicker biopsy section (10 µm), which was stained in an entire depth and visualized by bright field microscopy. Importantly, immunohistochemistry confirmed our ΜΦ phenotype-specific TPE-FLIM signatures, and, in addition, confirmed that they distinguish ΜΦs from other dermal cells. Skin mast cells, for example, stained for tryptase, showed a distinct TPE-FLIM signature, confirming our recently reported findings on dermal mast cells in vivo (Kröger et al., 2020), and distinguished them from M1 and M2 ΜΦs (Figure 3—figure supplement 2).

When we turned to the visualization of skin ΜΦs in vivo, we had first to develop a search strategy. Important considerations included the preferred localization of ΜΦs in the papillary and reticular dermis, the orientation of the focal plane parallel to the skin surface, and the need for maximal cellular cross-section visualization, which requires a high-resolution adjustment in depth to reconstruct an entire cell structure. The application of this search algorithm successfully visualized M1 and M2 ΜΦs in human skin in vivo. The in vivo TPE-FLIM parameters of M1 ΜΦs were in agreement with those observed in vitro and ex vivo. M2 ΜΦs in vivo were characterized by longer mean fluorescence lifetime τ_m compared to in vitro and ex vivo (Table 1), which can be explained by the influence of environment (Koo and Garg, 2019; Njoroge et al., 2001). ΜΦs measured in vivo differ from cells measured in vitro by their simplified microenvironment (Mosser and Edwards, 2008) with missing growth factors and cytokines (Melton et al., 2015) and an elevated level of nutrients, which leads to different polarization of ΜΦs and different contributions of fluorescence lifetimes. Membrane extensions were harder to detect in vivo due to the obscuring effect of the surrounding ECM. We also observed that TPE-FLIM parameters in the same ΜΦs can vary depending on their cellular substructures, for example, the nucleus, vacuoles, cytoplasm, or membrane (Figure 4a–d). The phasor plot shows the relative position of the categories of ΜΦs and other cells. Furthermore, it shows the contributions of long and short fluorescence lifetimes, their discrepancies (Table 1) being due to the computational method and the harmonics at the repetition frequency of 80 MHz. Further investigations are needed to clarify how the location, morphology, and function of M1 and M2 ΜΦs influence their TPE-FLIM parameters and TPE-AF intensity in vivo. Such studies should also address the reasons for the differences in TPE-FLIM parameters between M1 and M2 ΜΦs, which may include differences in their metabolic pathways. M1 ΜΦs, for example, rely on NADH oxidase and production of ROS, which is linked to short fluorescence lifetimes below 250 ps and mitochondrial fission. M2 ΜΦs, on the other hand, rely on oxidative phosphorylation and fatty acid oxidation, together with mitochondrial fusion (Ramond et al., 2019; Swindle et al., 2002; Xu et al., 2016).

The ability to visualize M1 and M2 ΜΦs by TPE-FLIM in vivo also makes it possible to explore how and why ΜΦs’ morphology, location, and functions are linked. When activated, the cytoskeletal structure and cellular appearance of ΜΦs change, and this may also affect their TPE-FLIM parameters. M1 ΜΦs are elongated, with a dense actin network along the cortex. M2 ΜΦs are more spherical with more randomly distributed actin (Porcheray et al., 2005; Vogel et al., 2014; Zhang et al., 2014; Figure 4b and d). Actin reorganization in M1 and M2 ΜΦ polarization and activation lead to bigger filament bundles of the actin cytoskeleton, which reduces cell plasticity (Colin-York et al., 2019; Pergola et al., 2017). As reported by Vogel et al., 2014, ΜΦ migration in the skin depends on their polarization. M1 ΜΦs, due to changes in actin cytoskeleton, migrate less far than M2 ΜΦs. Our TPE-FLIM findings confirm this, as we detected M1 ΜΦs via their high fluorescence and short autofluorescence lifetimes primarily in close proximity to blood capillaries. The irregular appearance of M1 ΜΦ detected by TPE-FLIM is likely a consequence of polarization-specific changes of the cellular cytoskeleton (McWhorter et al., 2013). The only morphological feature observed in both, M1 and M2 populations of ΜΦs, is that they are moderately dendritic, possibly because such ΜΦs are in the process of polarization, prior to cytoskeletal changes (Sica and Mantovani, 2012), or because polarization in ΜΦs is reversible polarizing (Sica and Mantovani, 2012; Yuan et al., 2017). Future studies should characterize the influence of cytoskeletal changes on TPE-FLIM parameters in detail and use TPE-FLIM to assess the impact of age, gender, and disease on the ratio, localization, and function of M1 and M2 ΜΦs in the skin (Fukui et al., 2018).

Time stability measurements were performed in vitro on M1 and M2 ΜΦs isolated from periocular skin within 1 hr (Figure 2—figure supplement 4) and in vivo on M1 ΜΦ within 30 min (Figure 4—figure supplements 2 and 3).

TPE-FLIM parameters were stable and varied within the standard deviation range shown in Table 1. It was possible to visualize single cells over this long-term time period in vitro (Figure 2—figure supplement 4) and in vivo (Figure 4—figure supplements 2 and 3), yet no unprompted change in fluorescence lifetime was observed, given the laser power was not high enough to induce photoproducts or photobleaching of the fluorophores. The NAD(P)H-related TPE-FLIM parameter showed no change over the course of the measurement nor did the fluorescence intensity. The limitation of long-term in vitro measurements is that the cells cannot be heated and will eventually cool down to room temperature. The limitation of long-term in vivo measurements is inherent to the individual. It is technically possible to investigate one subject over the course of multiple hours, but it is almost impossible to find the same cell again in another measurement.

Another limitation of this study is the simplified separation between M1 and M2 ΜΦs without taking into consideration the mixed ΜΦ phenotype. As mentioned in the introduction, mixed phenotype of ΜΦs is primarily observed in diseases like melanoma, systemic sclerosis, Lupus, and in recovery from LPS tolerance. Thus, we do not expect a significant amount of mixed phenotype ΜΦs in healthy and asymptomatic skin. However, the ΜΦs misidentified by the decision tree (Figure 2—figure supplement 3 and Figure 4—figure supplement 1) could potentially be ΜΦs of a mixed phenotype, which should be proved in additional experiment.

During phagocytosis, the generation of ROS by NAD(P)H oxidase leads to the highest degree of metabolic stress observed in M1 ΜΦs besides apoptosis (Dupré-Crochet et al., 2013; Shirshin et al., 2019), and ROS localization in vacuoles in phagocytosing M1 ΜΦs as a bactericidal mechanism (Dupré-Crochet et al., 2013; Myers et al., 2003). This is why phagocytosing M1 ΜΦs change their TPE-FLIM lifetimes toward shorter values and their vacuoles become visible as localized bright spots, which makes their in vivo detection possible (Figure 4—figure supplement 4, Table 1; Cannon and Swanson, 1992). The results shown here for possible phagocytosis are supported by the appearance (large size) and shortened fluorescence lifetime values for erythrophagocyting cells, presented by our group (Yakimov et al., 2019), resembling the cell shown in this study. The microenvironment in inflamed skin is known to be acidic with a pH<7.35 (Haka et al., 2009). Acidification is also known in phagocytosing ΜΦs (Teixeira et al., 2018). Together with the fact that fluorescence lifetime of fluorophores is shifted toward shorter values (Li et al., 2017) and cells especially ΜΦs produce ROS under phagocytosis and acidic conditions (Slauch, 2011), we have very compelling indications for the visualization of phagocytosing M1 ΜΦs. The influence of different phagocytosed materials in ΜΦs should be investigated in the future. TPE-FLIM potentially allows for the detection of possible phagocytosing M1 ΜΦs and is used as a confounder in the classification but cannot detect what is phagocytized. No internal structure was visible in TPE-FLIM images. The precision of measurements of cells and structures with short fluorescence lifetimes, such as phagocytosing ΜΦs, could be improved by reducing the value of the instrument response function (IRF), which is <100 ps in our measurements.

The construction of the feature vector and the resulting hyperparameter optimized decision tree model (Figure 4—figure supplement 5) yielded proficient results for the automatised classification of M1 and M2 ΜΦs, demonstrating that M1 and M2 ΜΦs can be separated from each other and other cells in the skin with high accuracy, that is, sensitivity and specificity, without need of additional staining using a supervised machine learning approach. The decision tree model uses only the independent TPE-FLIM parameters τ₁, τ₂, a₁, and a₂ and the dependent TPE-FLIM variables τ_m, τ₂/τ₁, a₁/a₂, (a₁−a₂)/(a₁+a₂). This indicates that macrophages could be distinguishable completely from other cells without the use of morphologic parameters, thus reducing the degree of freedom and saving calculation and annotation time by software and physicians. The high accuracy for M1 macrophages is owed to the fact that M1 ΜΦs have the shortest τ₁ and τ₂ fluorescence lifetimes, the highest ratio of a₁/a₂ and the highest fluorescence intensity of the cells in the model. M2 ΜΦs can be misclassified in rare cases with resting mast cells and in vitro dendritic cells. Concluding from these results that the classification is only dependent on TPE-FLIM parameters and not on the morphology of dermal cells.

Ideally the data sets consist of the same amount of entries for every three classes (M1 ΜΦs, M2 ΜΦs, and other cells). In the experimental reality, those three classes are not evenly distributed and could lead to overemphasize of certain classes in the classifier model. It is shown here by the methods above that the data set is useable for the decision tree classifier. Overfitting of the relatively small data set is avoided with the parameters for the decision tree model, first by randomly splitting the data set into training and test set 10,000 times resulting in a small standard deviation and with the use of k-fold cross-validation resulting in a mean of the cross-validation score of 0.90, which equates to 90% accuracy. A cross-validation score of 1 describes perfectly even distributed data in all folds. The robustness and accuracy of this approach can be improved further, by the introduction of a depth-adjusted cell size and refined cell shape parameters and by increasing the number of in vivo ΜΦs integrated into the algorithm and training data set.

The data have been deposited in Dryad.

1. 1. Kröger M
   2. Darvin M

   (2021) Dryad Digital Repository

   Macrophage FLIM raw data.

   https://doi.org/10.5061/dryad.8gtht76q2

1. 1. Alfonso-García A
   2. Smith TD
   3. Datta R
   4. Luu TU
   5. Gratton E
   6. Potma EO
   7. Liu WF

   (2016) Label-free identification of macrophage phenotype by fluorescence lifetime imaging microscopy

   Journal of Biomedical Optics 21:46005.

   https://doi.org/10.1117/1.JBO.21.4.046005

   • PubMed
   • Google Scholar
2. 1. AlShabany AJ
   2. Moody AJ
   3. Foey AD
   4. Billington RA

   (2016) Intracellular NAD+ levels are associated with LPS-induced TNF-α release in pro-inflammatory macrophages

   Bioscience Reports 36:e00301.

   https://doi.org/10.1042/BSR20150247

   • PubMed
   • Google Scholar
3. 1. Arango Duque G
   2. Descoteaux A

   (2014) Macrophage cytokines: involvement in immunity and infectious diseases

   Frontiers in Immunology 5:1–12.

   https://doi.org/10.3389/fimmu.2014.00491

   • PubMed
   • Google Scholar
4. 1. Bardi GT
   2. Smith MA
   3. Hood JL

   (2018) Melanoma exosomes promote mixed M1 and M2 macrophage polarization

   Cytokine 105:63–72.

   https://doi.org/10.1016/j.cyto.2018.02.002

   • PubMed
   • Google Scholar
5. 1. Benoit M
   2. Desnues B
   3. Mege JL

   (2008) Macrophage polarization in bacterial infections

   Journal of Immunology 181:3733–3739.

   https://doi.org/10.4049/jimmunol.181.6.3733

   • PubMed
   • Google Scholar
6. 1. Blacker TS
   2. Mann ZF
   3. Gale JE
   4. Ziegler M
   5. Bain AJ
   6. Szabadkai G
   7. Duchen MR

   (2014) Separating NADH and NADPH fluorescence in live cells and tissues using FLIM

   Nature Communications 5:3936.

   https://doi.org/10.1038/ncomms4936

   • PubMed
   • Google Scholar
7. 1. Botting RA
   2. Bertram KM
   3. Baharlou H
   4. Sandgren KJ
   5. Fletcher J
   6. Rhodes JW
   7. Rana H
   8. Plasto TM
   9. Wang XM
   10. Lim JJK
   11. Barnouti L
   12. Kohout MP
   13. Papadopoulos T
   14. Merten S
   15. Olbourne N
   16. Cunningham AL
   17. Haniffa M
   18. Harman AN

   (2017) Phenotypic and functional consequences of different isolation protocols on skin mononuclear phagocytes

   Journal of Leukocyte Biology 101:1393–1403.

   https://doi.org/10.1189/jlb.4A1116-496R

   • PubMed
   • Google Scholar
8. Book

   1. Breiman L
   2. Friedman JH
   3. Olshen RA
   4. Stone CJ

   (1984)

   Classification and Regression Trees, The Wadsworth Statistics/Probability Series

   Monterey, CA: Wadsworth & Brooks/Cole Advanced Books & Software.

   • Google Scholar
9. 1. Breunig HG
   2. Weinigel M
   3. Bückle R
   4. Kellner-Höfer M
   5. Lademann J
   6. Darvin ME
   7. Sterry W
   8. König K

   (2013) Clinical coherent anti-stokes raman scattering and multiphoton tomography of human skin with a femtosecond laser and photonic crystal fiber

   Laser Physics Letters 10:025604.

   https://doi.org/10.1088/1612-2011/10/2/025604

   • Google Scholar
10. 1. Cannon GJ
    2. Swanson JA

    (1992) The macrophage capacity for phagocytosis

    Journal of Cell Science 101 (Pt 4):907–913.

    https://doi.org/10.1242/jcs.101.4.907

    • PubMed
    • Google Scholar
11. 1. Cheng J
    2. Zhang Q
    3. Fan S
    4. Zhang A
    5. Liu B
    6. Hong Y
    7. Guo J
    8. Cui D
    9. Song J

    (2019) The vacuolization of macrophages induced by large amounts of inorganic nanoparticle uptake to enhance the immune response

    Nanoscale 11:22849–22859.

    https://doi.org/10.1039/c9nr08261a

    • PubMed
    • Google Scholar
12. 1. Chong BF
    2. Tseng L-C
    3. Hosler GA
    4. Teske NM
    5. Zhang S
    6. Karp DR
    7. Olsen NJ
    8. Mohan C

    (2015) A subset of CD163+ macrophages displays mixed polarizations in discoid lupus skin

    Arthritis Research & Therapy 17:324.

    https://doi.org/10.1186/s13075-015-0839-3

    • PubMed
    • Google Scholar
13. 1. Colin-York H
    2. Li D
    3. Korobchevskaya K
    4. Chang VT
    5. Betzig E
    6. Eggeling C
    7. Fritzsche M

    (2019) Cytoskeletal actin patterns shape mast cell activation

    Communications Biology 2:93.

    https://doi.org/10.1038/s42003-019-0322-9

    • PubMed
    • Google Scholar
14. 1. Darvin ME
    2. Richter H
    3. Ahlberg S
    4. Haag SF
    5. Meinke MC
    6. Le Quintrec D
    7. Doucet O
    8. Lademann J

    (2014) Influence of sun exposure on the cutaneous collagen/elastin fibers and carotenoids: negative effects can be reduced by application of sunscreen

    Journal of Biophotonics 7:735–743.

    https://doi.org/10.1002/jbio.201300171

    • PubMed
    • Google Scholar
15. 1. Datta R
    2. Alfonso-García A
    3. Cinco R
    4. Gratton E

    (2015) Fluorescence lifetime imaging of endogenous biomarker of oxidative stress

    Scientific Reports 5:1–10.

    https://doi.org/10.1038/srep09848

    • PubMed
    • Google Scholar
16. 1. Digman MA
    2. Caiolfa VR
    3. Zamai M
    4. Gratton E

    (2008) The phasor approach to fluorescence lifetime imaging analysis

    Biophysical Journal 94:L14–L16.

    https://doi.org/10.1529/biophysj.107.120154

    • PubMed
    • Google Scholar
17. 1. Dong B
    2. Almassalha LM
    3. Stypula-Cyrus Y
    4. Urban BE
    5. Chandler JE
    6. Nguyen TQ
    7. Sun C
    8. Zhang HF
    9. Backman V

    (2016) Superresolution intrinsic fluorescence imaging of chromatin utilizing native, unmodified nucleic acids for contrast

    PNAS 113:9716–9721.

    https://doi.org/10.1073/pnas.1602202113

    • PubMed
    • Google Scholar
18. 1. Dupré-Crochet S
    2. Erard M
    3. Nüβe O

    (2013) Ros production in phagocytes: why, when, and where?

    Journal of Leukocyte Biology 94:657–670.

    https://doi.org/10.1189/jlb.1012544

    • PubMed
    • Google Scholar
19. 1. Elhelu MA

    (1983)

    The role of macrophages in immunology

    Journal of the National Medical Association 75:314–317.

    • PubMed
    • Google Scholar
20. 1. Estandarte AK
    2. Botchway S
    3. Lynch C
    4. Yusuf M
    5. Robinson I

    (2016) The use of DAPI fluorescence lifetime imaging for investigating chromatin condensation in human chromosomes

    Scientific Reports 6:1–12.

    https://doi.org/10.1038/srep31417

    • PubMed
    • Google Scholar
21. 1. Ferrer MF
    2. Thomas P
    3. López Ortiz AO
    4. Errasti AE
    5. Charo N
    6. Romanowski V
    7. Gorgojo J
    8. Rodriguez ME
    9. Carrera Silva EA
    10. Gómez RM

    (2019) Junin virus triggers macrophage activation and modulates polarization according to viral strain pathogenicity

    Frontiers in Immunology 10:1–12.

    https://doi.org/10.3389/fimmu.2019.02499

    • PubMed
    • Google Scholar
22. 1. Fitzpatrick TB

    (1988) The validity and practicality of sun-reactive skin types I through VI

    Archives of Dermatology 124:869–871.

    https://doi.org/10.1001/archderm.124.6.869

    • PubMed
    • Google Scholar
23. 1. Forman HJ
    2. Torres M

    (2001) Redox signaling in macrophages

    Molecular Aspects of Medicine 22:189–216.

    https://doi.org/10.1016/s0098-2997(01)00010-3

    • PubMed
    • Google Scholar
24. 1. Fukui S
    2. Iwamoto N
    3. Takatani A
    4. Igawa T
    5. Shimizu T
    6. Umeda M
    7. Nishino A
    8. Horai Y
    9. Hirai Y
    10. Koga T
    11. Kawashiri SY
    12. Tamai M
    13. Ichinose K
    14. Nakamura H
    15. Origuchi T
    16. Masuyama R
    17. Kosai K
    18. Yanagihara K
    19. Kawakami A

    (2018) M1 and M2 monocytes in rheumatoid arthritis: a contribution of imbalance of M1/M2 monocytes to osteoclastogenesis

    Frontiers in Immunology 8:1–10.

    https://doi.org/10.3389/fimmu.2017.01958

    • PubMed
    • Google Scholar
25. 1. Furukawa S
    2. Moriyama M
    3. Miyake K
    4. Nakashima H
    5. Tanaka A
    6. Maehara T
    7. Iizuka-Koga M
    8. Tsuboi H
    9. Hayashida JN
    10. Ishiguro N
    11. Yamauchi M
    12. Sumida T
    13. Nakamura S

    (2017) Interleukin-33 produced by M2 macrophages and other immune cells contributes to Th2 immune reaction of IgG4-related disease

    Scientific Reports 7:1–10.

    https://doi.org/10.1038/srep42413

    • PubMed
    • Google Scholar
26. 1. Geissmann F
    2. Manz MG
    3. Jung S
    4. Sieweke MH
    5. Merad M
    6. Ley K

    (2010) Development of monocytes, macrophages, and dendritic cells

    Science 327:656–661.

    https://doi.org/10.1126/science.1178331

    • Google Scholar
27. 1. Gordon S
    2. Taylor PR

    (2005) Monocyte and macrophage heterogeneity

    Nature Reviews. Immunology 5:953–964.

    https://doi.org/10.1038/nri1733

    • PubMed
    • Google Scholar
28. 1. Haka AS
    2. Grosheva I
    3. Chiang E
    4. Buxbaum AR
    5. Baird BA
    6. Pierini LM
    7. Maxfield FR

    (2009) Macrophages create an acidic extracellular hydrolytic compartment to digest aggregated lipoproteins

    Molecular Biology of the Cell 20:4932–4940.

    https://doi.org/10.1091/mbc.E09-07-0559

    • PubMed
    • Google Scholar
29. 1. Heaster TM
    2. Heaton AR
    3. Sondel PM
    4. Skala MC

    (2021) Intravital metabolic autofluorescence imaging captures macrophage heterogeneity across normal and cancerous tissue

    Frontiers in Bioengineering and Biotechnology 9:644648.

    https://doi.org/10.3389/fbioe.2021.644648

    • PubMed
    • Google Scholar
30. 1. Kang K
    2. Hammerberg C
    3. Meunier L
    4. Cooper KD

    (1994)

    CD11b+ macrophages that infiltrate human epidermis after in vivo ultraviolet exposure potently produce IL-10 and represent the major secretory source of epidermal IL-10 protein

    Journal of Immunology 153:5256–5264.

    • PubMed
    • Google Scholar
31. 1. König K

    (2008) Clinical multiphoton tomography

    Journal of Biophotonics 1:13–23.

    https://doi.org/10.1002/jbio.200710022

    • PubMed
    • Google Scholar
32. 1. Koo SJ
    2. Garg NJ

    (2019) Metabolic programming of macrophage functions and pathogens control

    Redox Biology 24:101198.

    https://doi.org/10.1016/j.redox.2019.101198

    • PubMed
    • Google Scholar
33. 1. Kröger M
    2. Scheffel J
    3. Nikolaev VV
    4. Shirshin EA
    5. Siebenhaar F
    6. Schleusener J
    7. Lademann J
    8. Maurer M
    9. Darvin ME

    (2020) In vivo non-invasive staining-free visualization of dermal mast cells in healthy, allergy and mastocytosis humans using two-photon fluorescence lifetime imaging

    Scientific Reports 10:14930.

    https://doi.org/10.1038/s41598-020-71901-2

    • PubMed
    • Google Scholar
34. 1. Lakner PH
    2. Monaghan MG
    3. Möller Y
    4. Olayioye MA
    5. Schenke-Layland K

    (2017) Applying phasor approach analysis of multiphoton FLIM measurements to probe the metabolic activity of three-dimensional in vitro cell culture models

    Scientific Reports 7:42730.

    https://doi.org/10.1038/srep42730

    • PubMed
    • Google Scholar
35. 1. Li W
    2. Houston KD
    3. Houston JP

    (2017) Shifts in the fluorescence lifetime of EGFP during bacterial phagocytosis measured by phase-sensitive flow cytometry

    Scientific Reports 7:1–11.

    https://doi.org/10.1038/srep40341

    • PubMed
    • Google Scholar
36. 1. Li Y
    2. Liu TM

    (2018) Discovering macrophage functions using in vivo optical imaging techniques

    Frontiers in Immunology 9:1–20.

    https://doi.org/10.3389/fimmu.2018.00502

    • PubMed
    • Google Scholar
37. 1. Li Y
    2. He Y
    3. Miao K
    4. Zheng Y
    5. Deng C
    6. Liu TM

    (2020) Imaging of macrophage mitochondria dynamics in vivo reveals cellular activation phenotype for diagnosis

    Theranostics 10:2897–2917.

    https://doi.org/10.7150/thno.40495

    • PubMed
    • Google Scholar
38. 1. Malissen B
    2. Tamoutounour S
    3. Henri S

    (2014) The origins and functions of dendritic cells and macrophages in the skin

    Nature Reviews. Immunology 14:417–428.

    https://doi.org/10.1038/nri3683

    • PubMed
    • Google Scholar
39. 1. Malmgaard L
    2. Melchjorsen J
    3. Bowie AG
    4. Mogensen SC
    5. Paludan SR

    (2004) Viral activation of macrophages through TLR-dependent and -independent pathways

    Journal of Immunology 173:6890–6898.

    https://doi.org/10.4049/jimmunol.173.11.6890

    • PubMed
    • Google Scholar
40. 1. May RC
    2. Machesky LM

    (2001) Phagocytosis and the actin cytoskeleton

    Journal of Cell Science 114:1061–1077.

    https://doi.org/10.1242/jcs.114.6.1061

    • PubMed
    • Google Scholar
41. 1. McWhorter FY
    2. Wang T
    3. Nguyen P
    4. Chung T
    5. Liu WF

    (2013) Modulation of macrophage phenotype by cell shape

    PNAS 110:17253–17258.

    https://doi.org/10.1073/pnas.1308887110

    • PubMed
    • Google Scholar
42. 1. Melton DW
    2. McManus LM
    3. Gelfond JAL
    4. Shireman PK

    (2015) Temporal phenotypic features distinguish polarized macrophages in vitro

    Autoimmunity 48:161–176.

    https://doi.org/10.3109/08916934.2015.1027816

    • PubMed
    • Google Scholar
43. 1. Mendoza-Coronel E
    2. Ortega E

    (2017) Macrophage polarization modulates fcγR- and CD13-mediated phagocytosis and reactive oxygen species production, independently of receptor membrane expression

    Frontiers in Immunology 8:303.

    https://doi.org/10.3389/fimmu.2017.00303

    • PubMed
    • Google Scholar
44. 1. Morhenn VB
    2. Lemperle G
    3. Gallo RL

    (2002) Phagocytosis of different particulate dermal filler substances by human macrophages and skin cells

    Dermatologic Surgery 28:484–490.

    https://doi.org/10.1046/j.1524-4725.2002.01273.x

    • PubMed
    • Google Scholar
45. 1. Mosser DM
    2. Edwards JP

    (2008) Exploring the full spectrum of macrophage activation

    Nature Reviews. Immunology 8:958–969.

    https://doi.org/10.1038/nri2448

    • PubMed
    • Google Scholar
46. 1. Murray PJ
    2. Allen JE
    3. Biswas SK
    4. Fisher EA
    5. Gilroy DW
    6. Goerdt S
    7. Gordon S
    8. Hamilton JA
    9. Ivashkiv LB
    10. Lawrence T
    11. Locati M
    12. Mantovani A
    13. Martinez FO
    14. Mege J-L
    15. Mosser DM
    16. Natoli G
    17. Saeij JP
    18. Schultze JL
    19. Shirey KA
    20. Sica A
    21. Suttles J
    22. Udalova I
    23. van Ginderachter JA
    24. Vogel SN
    25. Wynn TA

    (2014) Macrophage activation and polarization: nomenclature and experimental guidelines

    Immunity 41:14–20.

    https://doi.org/10.1016/j.immuni.2014.06.008

    • PubMed
    • Google Scholar
47. 1. Murray PJ

    (2017) Macrophage polarization

    Annual Review of Physiology 79:541–566.

    https://doi.org/10.1146/annurev-physiol-022516-034339

    • PubMed
    • Google Scholar
48. 1. Myers JT
    2. Tsang AW
    3. Swanson JA

    (2003) Localized reactive oxygen and nitrogen intermediates inhibit escape of listeria monocytogenes from vacuoles in activated macrophages

    Journal of Immunology 171:5447–5453.

    https://doi.org/10.4049/jimmunol.171.10.5447

    • PubMed
    • Google Scholar
49. 1. Nielsen MC
    2. Andersen MN
    3. Møller HJ

    (2020) Monocyte isolation techniques significantly impact the phenotype of both isolated monocytes and derived macrophages in vitro

    Immunology 159:63–74.

    https://doi.org/10.1111/imm.13125

    • PubMed
    • Google Scholar
50. 1. Njoroge JM
    2. Mitchell LB
    3. Centola M
    4. Kastner D
    5. Raffeld M
    6. Miller JL

    (2001) Characterization of viable autofluorescent macrophages among cultured peripheral blood mononuclear cells

    Cytometry 44:38–44.

    https://doi.org/10.1002/1097-0320(20010501)44:1<38::aid-cyto1080>3.0.co;2-t

    • PubMed
    • Google Scholar
51. 1. O’Carroll C
    2. Fagan A
    3. Shanahan F
    4. Carmody RJ

    (2014) Identification of a unique hybrid macrophage-polarization state following recovery from lipopolysaccharide tolerance

    Journal of Immunology 192:427–436.

    https://doi.org/10.4049/jimmunol.1301722

    • PubMed
    • Google Scholar
52. 1. Orihuela R
    2. McPherson CA
    3. Harry GJ

    (2016) Microglial M1/M2 polarization and metabolic states

    British Journal of Pharmacology 173:649–665.

    https://doi.org/10.1111/bph.13139

    • PubMed
    • Google Scholar
53. 1. Pergola C
    2. Schubert K
    3. Pace S
    4. Ziereisen J
    5. Nikels F
    6. Scherer O
    7. Hüttel S
    8. Zahler S
    9. Vollmar AM
    10. Weinigel C
    11. Rummler S
    12. Müller R
    13. Raasch M
    14. Mosig A
    15. Koeberle A
    16. Werz O

    (2017) Modulation of actin dynamics as potential macrophage subtype-targeting anti-tumour strategy

    Scientific Reports 7:1–12.

    https://doi.org/10.1038/srep41434

    • PubMed
    • Google Scholar
54. 1. Ploeger DT
    2. Hosper NA
    3. Schipper M
    4. Koerts JA
    5. de Rond S
    6. Bank RA

    (2013) Cell plasticity in wound healing: paracrine factors of M1/ M2 polarized macrophages influence the phenotypical state of dermal fibroblasts

    Cell Communication and Signaling 11:29.

    https://doi.org/10.1186/1478-811X-11-29

    • PubMed
    • Google Scholar
55. 1. Porcheray F
    2. Viaud S
    3. Rimaniol AC
    4. Léone C
    5. Samah B
    6. Dereuddre-Bosquet N
    7. Dormont D
    8. Gras G

    (2005) Macrophage activation switching: an asset for the resolution of inflammation

    Clinical and Experimental Immunology 142:481–489.

    https://doi.org/10.1111/j.1365-2249.2005.02934.x

    • PubMed
    • Google Scholar
56. 1. Ramond E
    2. Jamet A
    3. Coureuil M
    4. Charbit A

    (2019) Pivotal role of mitochondria in macrophage response to bacterial pathogens

    Frontiers in Immunology 10:1–11.

    https://doi.org/10.3389/fimmu.2019.02461

    • PubMed
    • Google Scholar
57. 1. Remmerie A
    2. Scott CL

    (2018) Macrophages and lipid metabolism

    Cellular Immunology 330:27–42.

    https://doi.org/10.1016/j.cellimm.2018.01.020

    • PubMed
    • Google Scholar
58. 1. Rendra E
    2. Riabov V
    3. Mossel DM
    4. Sevastyanova T
    5. Harmsen MC
    6. Kzhyshkowska J

    (2019) Reactive oxygen species (ROS) in macrophage activation and function in diabetes

    Immunobiology 224:242–253.

    https://doi.org/10.1016/j.imbio.2018.11.010

    • PubMed
    • Google Scholar
59. 1. Ryter A

    (1985) Relationship between ultrastructure and specific functions of macrophages

    Comparative Immunology, Microbiology and Infectious Diseases 8:119–133.

    https://doi.org/10.1016/0147-9571(85)90039-6

    • PubMed
    • Google Scholar
60. 1. Schmid MA
    2. Harris E

    (2014) Monocyte recruitment to the dermis and differentiation to dendritic cells increases the targets for dengue virus replication

    PLOS Pathogens 10:e1004541.

    https://doi.org/10.1371/journal.ppat.1004541

    • PubMed
    • Google Scholar
61. 1. Shirshin EA
    2. Gurfinkel YI
    3. Priezzhev AV
    4. Fadeev VV
    5. Lademann J
    6. Darvin ME

    (2017) Two-photon autofluorescence lifetime imaging of human skin papillary dermis in vivo: assessment of blood capillaries and structural proteins localization

    Scientific Reports 7:1171.

    https://doi.org/10.1038/s41598-017-01238-w

    • PubMed
    • Google Scholar
62. 1. Shirshin EA
    2. Yakimov BP
    3. Darvin ME
    4. Omelyanenko NP
    5. Rodionov SA
    6. Gurfinkel YI
    7. Lademann J
    8. Fadeev VV
    9. Priezzhev AV

    (2019) Label-free multiphoton microscopy: the origin of fluorophores and capabilities for analyzing biochemical processes

    Biochemistry 84:69–88.

    https://doi.org/10.1134/S0006297919140050

    • Google Scholar
63. 1. Shirshin EA
    2. Shirmanova MV
    3. Gayer AV
    4. Lukina MM
    5. Nikonova EE
    6. Yakimov BP
    7. Budylin GS
    8. Dudenkova VV
    9. Ignatova NI
    10. Komarov DV
    11. Yakovlev VV
    12. Becker W
    13. Zagaynova EV
    14. Shcheslavskiy VI
    15. Scully MO

    (2022) Label-free sensing of cells with fluorescence lifetime imaging: the quest for metabolic heterogeneity

    PNAS 119:e2118241119.

    https://doi.org/10.1073/pnas.2118241119

    • PubMed
    • Google Scholar
64. 1. Sica A
    2. Mantovani A

    (2012) Macrophage plasticity and polarization: in vivo veritas

    The Journal of Clinical Investigation 122:787–795.

    https://doi.org/10.1172/JCI59643

    • PubMed
    • Google Scholar
65. 1. Slauch JM

    (2011) How does the oxidative burst of macrophages kill bacteria? still an open question

    Molecular Microbiology 80:580–583.

    https://doi.org/10.1111/j.1365-2958.2011.07612.x

    • PubMed
    • Google Scholar
66. 1. Swindle EJ
    2. Hunt JA
    3. Coleman JW

    (2002) A comparison of reactive oxygen species generation by rat peritoneal macrophages and mast cells using the highly sensitive real-time chemiluminescent probe pholasin: inhibition of antigen-induced mast cell degranulation by macrophage-derived hydrogen peroxide

    Journal of Immunology 169:5866–5873.

    https://doi.org/10.4049/jimmunol.169.10.5866

    • PubMed
    • Google Scholar
67. 1. Szulczewski JM
    2. Inman DR
    3. Entenberg D
    4. Ponik SM
    5. Aguirre-Ghiso J
    6. Castracane J
    7. Condeelis J
    8. Eliceiri KW
    9. Keely PJ

    (2016) In vivo visualization of stromal macrophages via label-free FLIM-based metabolite imaging

    Scientific Reports 6:1–9.

    https://doi.org/10.1038/srep25086

    • PubMed
    • Google Scholar
68. 1. Teixeira J
    2. Basit F
    3. Swarts HG
    4. Forkink M
    5. Oliveira PJ
    6. Willems P
    7. Koopman WJH

    (2018) Extracellular acidification induces ROS- and mptp-mediated death in HEK293 cells

    Redox Biology 15:394–404.

    https://doi.org/10.1016/j.redox.2017.12.018

    • PubMed
    • Google Scholar
69. 1. Theret M
    2. Mounier R
    3. Rossi F

    (2019) The origins and non-canonical functions of macrophages in development and regeneration

    Development 146:1–14.

    https://doi.org/10.1242/dev.156000

    • PubMed
    • Google Scholar
70. 1. Trombetta AC
    2. Soldano S
    3. Contini P
    4. Tomatis V
    5. Ruaro B
    6. Paolino S
    7. Brizzolara R
    8. Montagna P
    9. Sulli A
    10. Pizzorni C
    11. Smith V
    12. Cutolo M

    (2018) A circulating cell population showing both M1 and M2 monocyte/macrophage surface markers characterizes systemic sclerosis patients with lung involvement

    Respiratory Research 19:1–12.

    https://doi.org/10.1186/s12931-018-0891-z

    • PubMed
    • Google Scholar
71. 1. Viola A
    2. Munari F
    3. Sánchez-Rodríguez R
    4. Scolaro T
    5. Castegna A

    (2019) The metabolic signature of macrophage responses

    Frontiers in Immunology 10:1–16.

    https://doi.org/10.3389/fimmu.2019.01462

    • PubMed
    • Google Scholar
72. 1. Vogel DYS
    2. Heijnen PDAM
    3. Breur M
    4. de Vries HE
    5. Tool ATJ
    6. Amor S
    7. Dijkstra CD

    (2014) Macrophages migrate in an activation-dependent manner to chemokines involved in neuroinflammation

    Journal of Neuroinflammation 11:1–11.

    https://doi.org/10.1186/1742-2094-11-23

    • PubMed
    • Google Scholar
73. 1. Walsh AJ
    2. Mueller KP
    3. Tweed K
    4. Jones I
    5. Walsh CM
    6. Piscopo NJ
    7. Niemi NM
    8. Pagliarini DJ
    9. Saha K
    10. Skala MC

    (2021) Classification of T-cell activation via autofluorescence lifetime imaging

    Nature Biomedical Engineering 5:77–88.

    https://doi.org/10.1038/s41551-020-0592-z

    • PubMed
    • Google Scholar
74. 1. Weber-Matthiesen K
    2. Sterry W

    (1990) Organization of the monocyte/macrophage system of normal human skin

    The Journal of Investigative Dermatology 95:83–89.

    https://doi.org/10.1111/1523-1747.ep12874002

    • PubMed
    • Google Scholar
75. 1. Xu Q
    2. Choksi S
    3. Qu J
    4. Jang J
    5. Choe M
    6. Banfi B
    7. Engelhardt JF
    8. Liu ZG

    (2016) Nadph oxidases are essential for macrophage differentiation

    The Journal of Biological Chemistry 291:20030–20041.

    https://doi.org/10.1074/jbc.M116.731216

    • PubMed
    • Google Scholar
76. 1. Yakimov BP
    2. Gogoleva MA
    3. Semenov AN
    4. Rodionov SA
    5. Novoselova MV
    6. Gayer AV
    7. Kovalev AV
    8. Bernakevich AI
    9. Fadeev VV
    10. Armaganov AG
    11. Drachev VP
    12. Gorin DA
    13. Darvin ME
    14. Shcheslavskiy VI
    15. Budylin GS
    16. Priezzhev AV
    17. Shirshin EA

    (2019) Label-free characterization of white blood cells using fluorescence lifetime imaging and flow-cytometry: molecular heterogeneity and erythrophagocytosis [invited]

    Biomedical Optics Express 10:4220–4236.

    https://doi.org/10.1364/BOE.10.004220

    • PubMed
    • Google Scholar
77. 1. Yanez DA
    2. Lacher RK
    3. Vidyarthi A
    4. Colegio OR

    (2017) The role of macrophages in skin homeostasis

    Pflugers Archiv 469:455–463.

    https://doi.org/10.1007/s00424-017-1953-7

    • PubMed
    • Google Scholar
78. 1. Yang L
    2. Shami A

    (2020) On hyperparameter optimization of machine learning algorithms: theory and practice

    Neurocomputing 415:295–316.

    https://doi.org/10.1016/j.neucom.2020.07.061

    • Google Scholar
79. 1. Yuan R
    2. Li S
    3. Geng H
    4. Wang X
    5. Guan Q
    6. Li X
    7. Ren C
    8. Yuan X

    (2017) Reversing the polarization of tumor-associated macrophages inhibits tumor metastasis

    International Immunopharmacology 49:30–37.

    https://doi.org/10.1016/j.intimp.2017.05.014

    • PubMed
    • Google Scholar
80. 1. Zhang XM
    2. Lund H
    3. Mia S
    4. Parsa R
    5. Harris RA

    (2014) Adoptive transfer of cytokine-induced immunomodulatory adult microglia attenuates experimental autoimmune encephalomyelitis in DBA/1 mice

    Glia 62:804–817.

    https://doi.org/10.1002/glia.22643

    • PubMed
    • Google Scholar

Decision letter after peer review:

Thank you for submitting your article "Label-free imaging of macrophage phenotypes and phagocytic activity in the human dermis in vivo using two-photon excited FLIM" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, including Michael L Dustin as Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Aleksandra Walczak as the Senior Editor.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

There are a number of positive aspects of your study including:

Using optical methods to non-invasively detect cells has significant interest for clinical and basic studies, and the impact of this study is considered high.

Identification of different FLIM signatures for macrophages polarized towards different phenotypes in vitro, and were able to compare these signatures to those of other cell types and macrophages in the skin.

They identified a few cells in the skin that expressed markers associated with macrophage polarization, and also exhibited the FLIM signatures that were established from the in vitro polarization studies.

At the same time, there are limitations of the study that can be addressed through the following Essential Revisions:

1) The description of M1 and M2 macrophages in the skin (in the introduction) is overly simplistic and the use of single markers (in Figure 3) to establish correlations is also simplistic. This should be updated and mixed phenotypes acknowledged.

2) It would be very useful to see how the classification tree fails ~10% of the time in relation to the plots in Figure S5. It's striking that the FLIM parameters generate a near perfect classification of M1 and M2 so it almost seems that adding information like cell shape may make it less accurate. Can the authors indicate in S5 which cells were mis-identified by the decision tree and if the reason is clear? If the FLIM parameters alone are fully discriminative it's not clear why the other parameters are helpful. Is there an "F-test" that can be done to assess the statistical value of each parameter that is added to the tree and if the improvement is greater then just adding another degree of freedom. Eventually this seems destined for some kind of experimental medicine application and this information would help determine where we are in terms of feasibility toward these near term goals. It would be important to report the number of in vivo tomograms that were done for each subject, the volume analyzed and amount of time to collect the tomograms.

3) Additional phenotype markers and evaluation of many more cells seems required to test the classifier. It's possible that the current 90% reliability is not actually statistically significant due to the low numbers. There are some significant technical concerns given that macrophages are a highly heterogeneous population of cells, particularly in vivo during an activation event such as injury. The few cells analyzed in Figure 3 are not sufficient given the heterogeneity of macrophages in vivo. Mixed phenotypes are common in vivo, and it is unclear how the FLIM signatures would change in response to such mixed signatures.

4) Visualizing a single phagocytosing cell in Figure 5 is also not sufficient to conclude that the method is capable of detecting phagocytosis events. Experiments in vitro treating macrophages with phagocytic targets need to be performed. An experiment in vivo would also strengthen the study, where phagocytic targets are applied to a skin wound. Controls without targets should be examined, along with quantitation across many cells. You may wish to focus on the classification task and perhaps only bring up phagocytosis as a possible confounder in this class-action process, but demonstrating the ability to detect phagocytosis in vivo may be another paper.

5) Is the lifetime signature stable over time? Can a single cell be imaged over time. This seems possible as macrophages are generally sessile. The study would be strengthened with further analysis that correlates FLIM signatures with metabolic state (free vs. bound NADPH). Could the lifetime relate to metabolic cycles in the issues that could change with time? This is a fundamental question that should be addressed in a first paper on this topic.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Label-free imaging of M1 and M2 macrophage phenotypes in the human dermis in vivo using two-photon excited FLIM" for further consideration by eLife. Your revised article has been evaluated by Aleksandra Walczak (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

This is a very interesting study that uses a label free method to classify macrophages in human dermis using two photon imaging. Additional data analysis and reporting is requested to fully document the very important work.

1. Justify your fitting model and include the decays pixel by pixel and also the fit profiles together with the goodness of the fit for all cases.

2. Compare with the Phasor plot, all your data.

3. Present all images with the corresponding number of photons acquired per pixel and plot these versus your mean lifetimes.

4. Calculate the true mean lifetime according to the canonical formula.

5. Present all your data in a pixel by pixel format.

Reviewer #1 (Recommendations for the authors):

Thank you for your clear response to the reviewer's concerns. You have addressed the issue of mixed phenotypes in healthy skin and your in vitro analysis of both in vitro differentiated and freshly isolated M1 and M2 like cells from skin show consistent two photon film signatures. You have provided information about the time involved in the measurements for each subject, the volume scanned and the number of cells found. You have also addressed the temporal stability in vivo consistent with the observation that the isolate cells maintain the flim phenotype even after isolation. The reporting on phagocytosis is now qualified and the limitations of the study are now appropriately acknowledged. I have no further recommendations.

Reviewer #3 (Recommendations for the authors):

In general the approach and validation of the technique is appropriate. I would be interested though to check the relative number of photons collected per pixel versus the different lifetimes to make sure that all lifetimes collected and analyzed are independent on the intensity/number of photons.

Also, some of the lifetimes recovered were very short and close to the resolution of the instrument response function (IRF); which by the way seems not to be recovered experimentally but rather produced by the SPCI software. Please, clarify and also if it was not recovered experimentally produce these data.

This brings me to my next point. The authors employ a double exponential approach and present both the average lifetime (calculated based on tau1 and tau2 and not with the appropriate formula, see Padilla-Parra et al., 2008 Biophys J). The authors do not show the fluorescence decays for each cell, together with the double exponential fits nor the goodness of these fits or the Χ2 (chi square). This data is fundamental to understanding if all FLIM data recovered can be fitted to the double exponential model. Also, the authors assume a double exponential approach but they do not justify their choice. Which are the two populations you are assuming to co-exist in the cells/dermis? Are you always measuring NADPH/NADP? In this case, have you shown the value of these two lifetimes in vitro, and then you should fix these two lifetimes and recover different proportions, right? The model to fit your data should be discussed and justified. I assume that if you use a triple exponential most of your stats will be better. Please discuss this also in the context of the number of photons.

Also, all the parameters recovered from the fit should be also shown pixel by pixel so that we can understand how these data vary as a function of the number of photons, or the error (Chi Square).

In Figure 4 the authors decide to show the phasor plot (Digman et al., 2008 Biophys J). This is a nice approach that does not necessitate the assumption of a model. You should present all data comparing this approach and the fitting approach utilizing a double exponential and also discuss how your data varies depending on the number of photons. It is possible that some lifetime distributions arise from the fact that in some cells the high intensity values give longer lifetimes and in others where there is a lower signal to noise, lower lifetimes are observed. When plotting your lifetimes versus the number of photons pixel by pixel no dependence should be observed. This will help to validate your data.

At the same time, there are limitations of the study that can be addressed through the following Essential Revisions:

1) The description of M1 and M2 macrophages in the skin (in the introduction) is overly simplistic and the use of single markers (in Figure 3) to establish correlations is also simplistic. This should be updated and mixed phenotypes acknowledged.

We thank the reviewer for this important comment. We agree that the M1/M2 is simplified and it is clear that macrophage phenotype diversity goes beyond that. Not only differ macrophages with regards to their origin but they can adopt to various responsive phenotypes with shared and mixed signatures, depending on the stimulus they experience. This being said, macrophage polarization requires certain stimuli, which might not be present in healthy skin. In literature, spontaneously mixed phenotypes in healthy skin have not been found.

The precise role of skin ΜΦs and their M1 and M2 phenotypes in health and disease remains to be elucidated. In skin disease such as melanoma (Bardi GT, Smith MA, Hood JL. Melanoma exosomes promote mixed M1 and M2 macrophage polarization. Cytokine. 105: 63-72, 2018), systemic sclerosis (Trombetta AC, Soldano S, Contini P, et al. A circulating cell population showing both M1 and M2 monocyte/macrophage surface markers characterizes systemic sclerosis patients with lung involvement. Respir Res. 19(1): 186, 2018), Lupus (Chong BF, Tseng LC, Hosler GA, et al. A subset of CD163+ macrophages displays mixed polarizations in discoid lupus skin. Arthritis Res Ther. 17: 324, 2015) and Lipopolysaccharide Tolerance (O'Carroll C, Fagan A, Shanahan F, Carmody RJ. Identification of a unique hybrid macrophage-polarization state following recovery from lipopolysaccharide tolerance. J Immunol. 192(1): 427-436, 2014), ΜΦs polarizations leading to mixed M1 and M2 phenotypes can be observed.

Moreover, this study did not aim to fully characterize the FLIM signatures of all possible macrophage subpopulations but to describe a method to visualize macrophages in the healthy skin without the need of external labeling. Further studies are definitely required to investigate mixed phenotypes in health and skin disease and disorders.

The additional information is updated and acknowledged in the main text on page 1: line 46 and page 13: line 454.

2) It would be very useful to see how the classification tree fails ~10% of the time in relation to the plots in Figure S5. It's striking that the FLIM parameters generate a near perfect classification of M1 and M2 so it almost seems that adding information like cell shape may make it less accurate. Can the authors indicate in S5 which cells were mis-identified by the decision tree and if the reason is clear? If the FLIM parameters alone are fully discriminative it's not clear why the other parameters are helpful. Is there an "F-test" that can be done to assess the statistical value of each parameter that is added to the tree and if the improvement is greater then just adding another degree of freedom. Eventually this seems destined for some kind of experimental medicine application and this information would help determine where we are in terms of feasibility toward these near term goals. It would be important to report the number of in vivo tomograms that were done for each subject, the volume analyzed and amount of time to collect the tomograms.

We thank the reviewer for this important recommendation. The high accuracy for M1 macrophages is owed to the fact that M1 macrophages have the shortest fluorescence lifetimes τ₁ and τ₂ and the highest ratio of a₁/a₂ of the cells in the model. M2 macrophages can be confused in rare cases with resting mast cells or/and dendritic cells in vitro due to the overlap in τ₂ fluorescence lifetime. These TPE-FLIM parameters can be seen in Table 1 and the phasor plot in Figure 4. The failure is also visible in the 2D segmentation of isolated dermal macrophages where we find areas located closely together, which can be interpreted as superposition of parameters.

In the figures we show the cells (area marked red) which were misidentified by the decision tree in our in vitro and in vivo experiments, in Figure 2—figure supplement 3 and Figure 4—figure supplement 1 (new numbering).

Figure S3 and Figure S5 have been replaced by the new (Figure 2—figure supplement 3 and Figure 4—figure supplement 1 in a new numbering) in the supplementary materials showing additionally cells misidentified by the decision tree.

The misidentification was due to the different values of TPE-FLIM parameters and the proximity to the classification conditions.

As a reminder the independent TPE-FLIM parameters are a₁, a₂, τ₁, τ₂, the dependent TPE-FLIM variables τ_m, τ₂/τ₁, a₁/a₂, (a₁-a₂)/(a₁+a₂) and the fluorescence intensity.

With 10,000 repetitions of random training data and test data splits, we are sure the model is not overfitted to the data. We have to separate between true positive rate and true negative rate, meaning the difference between the ability of the model to predict macrophages as macrophages (true positive rate, TPR) and to predict other dermal cells as non-macrophages (true negative rate, TNR). The TPR for M2 macrophages is lower compared to M1 macrophages due to a more common FLIM signature, where less M2 macrophages get classified as M2 macrophages. Yet, the TNR is higher for M2 macrophages compared to M1 macrophages.

A “F-Test” is not needed and not applicable in the decision tree model, because the significance of the parameter is directly tied to the use in the decision tree model. As can be seen in the decision tree model in Figure 4—figure supplement 5 (new numbering), the best classification is done with TPE-FLIM parameters only, shape and fluorescence intensity wasn’t used in this model, thus the degree of freedom is limited by the relevant parameter in the decision tree model. The 8 TPE-FLIM parameters: τ₁, τ₂, τ_m, a₁, a₂, τ₂/τ₁, a₁/a₂, (a₁-a₂)/(a₁+a₂), obtained after bi-exponential approximation of the decay curve are enough for a meaningful classification. The fluorescence intensity will be a more important feature when only in vivo cells in the papillary and reticular dermis will are classified and the fluorescence intensity is comparable independent of absorption and scattering effects, due to the same depth in the skin. The cell shape and fluorescence intensity are promising indicators to separate functionally different macrophage phenotypes in a comprehensive classification model with more dermal cells in the future. The parameters used in the model are described again in the Results section on page 10 and discussed from line 286 and in the Discussion section on page 14: line 484.

“Six to twelve in vivo tomograms (different skin areas) were measured per subject, the investigated volume is from ≈70 µm depth in the papillary dermis to ≈130 µm depth in the reticular dermis with an image size of 150 µm. This adds up to (60×150×150) µm³ times six to twelve images, with a total volume of ≈0.0081 to 0.0162 mm³ of papillary and reticular dermis seen per subject, the average time spent was ≈30 minutes per subject and the acquisition time was 6.8 s per image.”

This information has been added to the methods section of main text on page 15: line 558.

3) Additional phenotype markers and evaluation of many more cells seems required to test the classifier. It's possible that the current 90% reliability is not actually statistically significant due to the low numbers. There are some significant technical concerns given that macrophages are a highly heterogeneous population of cells, particularly in vivo during an activation event such as injury. The few cells analyzed in Figure 3 are not sufficient given the heterogeneity of macrophages in vivo. Mixed phenotypes are common in vivo, and it is unclear how the FLIM signatures would change in response to such mixed signatures.

It is desirable that more cells have to be tested in the future to further strengthen the model and classification accuracy. In this pilot study all measures were met to avoid the overfitting of the data or the statistical insignificance of the classification method due to low numbers. First, the training data set and the test set was split randomly 10,000 times and the model was tested on it, this approach removes the possibility of random spikes in sensitivity and specificity and ensures a sufficient size of training and test sets. The standard deviation of both the specificity and sensitivity was reasonably low, indicating no irregularities. Additionally, a k-fold cross-validation with k=5 was executed to avoid unevenly distributed data, with the results: (0.87; 0.92; 0.87; 0.89; 0.94) and mean of k-fold scores using the cross_val_score method is 0.90. The macrophages misidentified by the decision tree (answer to the first reviewer’s question) and now shown in Figure 2—figure supplement 3 and Figure 4—figure supplement 1 (new numbering), could potentially be macrophages of a mixed phenotype, however this conclusion was not proved in this study and needs additional investigations. The main text is updated on page 10: line 296 in the Results section and on page 14: line 484 in the Discussion section.

Regarding the heterogeneity of macrophages please also refer to the response to the question 1, all subjects were investigated on healthy and asymptomatic skin areas without pre-existing skin conditions. It is unclear whether and to what amount the mixed phenotypes occurring in disease are existent in healthy and asymptomatic skin.

Hence, we focused on M1 and M2 phenotypes in this study, that is a simplified yet realistic model, and mixed phenotypes will be a subject of further investigations.

4) Visualizing a single phagocytosing cell in Figure 5 is also not sufficient to conclude that the method is capable of detecting phagocytosis events. Experiments in vitro treating macrophages with phagocytic targets need to be performed. An experiment in vivo would also strengthen the study, where phagocytic targets are applied to a skin wound. Controls without targets should be examined, along with quantitation across many cells. You may wish to focus on the classification task and perhaps only bring up phagocytosis as a possible confounder in this class-action process, but demonstrating the ability to detect phagocytosis in vivo may be another paper.

We thank the reviewer for the insightful comment and suggestion. In this paper we have focused on the classification task as it arises from the questions above with added k-fold cross validation and extended discussion. We have changed the title to “Label-free imaging of M1 and M2 macrophage phenotypes in the human dermis in vivo using two-photon excited FLIM”. Figure 5 has been moved to the supplementary information, see Figure 4—figure supplement 4 (new numbering).

The authors agree with the reviewer that in vitro and in vivo study of phagocytosis is required. However, the authors decided to adhere to these pilot results about visualization of phagocytosing macrophages in the skin in the manuscript due to the following reasons.

As a further evidence, the erythrophagocyting cells, presented by our group (Yakimov et al., Label-free characterization of white blood cells using fluorescence lifetime imaging and flow-cytometry: molecular heterogeneity and erythrophagocytosis, Biomedical Optics Express. 10(8): 4220-4236, 2019) exhibit strong shortening of fluorescence decay, resembling the cell shown in our study. As a defence mechanism phagocytosing ΜΦs reduce the acidify, their adjacent extracellular environment and simultaneously ROS are produced, which acts as bactericide and is known to decrease fluorescence lifetimes (Haka AS, Grosheva I, Chiang E, et al. Macrophages create an acidic extracellular hydrolytic compartment to digest aggregated lipoproteins. Mol Biol Cell. 20(23): 4932-4940, 2009; Teixeira J, Basit F, Swarts HG, et al. Extracellular acidification induces ROS- and mPTP-mediated death in HEK293 cells. Redox Biol. 15: 394-404, 2018; Slauch JM. How does the oxidative burst of macrophages kill bacteria? Still an open question. Mol Microbiol. 80(3): 580-583, 2011; Li, W., Houston, K. and Houston, J. Shifts in the fluorescence lifetime of EGFP during bacterial phagocytosis measured by phase-sensitive flow cytometry. Sci Rep 7: 40341, 2017). Thus, the observed cells in the skin and their TPE-FLIM parameters and morphology are in agreement with previous results and knowledge about phagocytosing macrophages.

We modified the paragraph “TPE-FLIM can distinguish resting from phagocytosing human skin M1 ΜΦs in vivo“, please see page 9: line 266 and the discussion on page 13: line 467.

5) Is the lifetime signature stable over time? Can a single cell be imaged over time. This seems possible as macrophages are generally sessile. The study would be strengthened with further analysis that correlates FLIM signatures with metabolic state (free vs. bound NADPH). Could the lifetime relate to metabolic cycles in the issues that could change with time? This is a fundamental question that should be addressed in a first paper on this topic.

We thank the reviewer for the interesting technical question. The fluorescence lifetime of in vitro cells is stable for at least one hour and lies in the range of the variation in lifetime presented in Table 1. The in vivo cells are stable over the course of the measurement and the stability of their TPE-FLIM parameters is also within the standard deviation. One indicator of the lifetime stability is that ex vivo experiments agree with the other measurements.

A single cell could be imaged over time, however the in vivo subjects cannot be measured for longer than half an hour due to human restraints.

We performed a time series images of macrophages isolated from periocular skin during 60 minutes acquired in vitro every 5 minutes at 6.8 s scanning time at 6 mW, this is a total of 81.6 s of scanning at 80 MHz repetition rate. The TPE-FLIM images for a mixture of M1 and M2 macrophages is presented for two time points: 0 and after 60 minutes. As can be seen, TPE-FLIM parameters of M1 and M2 macrophages are stable and vary within the corresponding standard deviation presented in Table 1. These TPE-FLIM images are added as Figure 2—figure supplement 4 in the paper.

We performed a time series of dermal M1 and M2 macrophage images during 30 minutes acquired in vivo every minute at 6.8 s scanning time and at 50 mW, this is a total of 210.8 s of scanning at 80 MHz repetition rate. The TPE-FLIM image for M1 and M2 macrophages in vivo is presented in the figure for seven time points: 0, 5, 10, 15, 20, 25 and 30 minutes. As can be seen, TPE-FLIM parameters for recorded M1 and M2 macrophages are stable and vary within the standard deviation presented in Table 1. These TPE-FLIM images are added as Figure 4—figure supplement 2,3 (new numbering) in the paper.

No significant changes can be observed in TPE-FLIM parameters in vitro and in vivo during the stability measurements over period of 30 to 60 min.

In the literature macrophage infiltration is measured in a matter of days by Baeten et al. (Baeten, Kurt, et al. Visualisation of the kinetics of macrophage infiltration during experimental autoimmune encephalomyelitis by magnetic resonance imaging. Journal of Neuroimmunology, 195:1-6, 2008) and Lauterbach et al. (Lauterbach, Mario A., et al. Toll-like receptor signaling rewires macrophage metabolism and promotes histone acetylation via ATP-citrate lyase. Immunity, 51: 997-1011, 2019) showed change in metabolism by toll-like receptors over the course of hours with a maximum at 4 hours. In the time frame of the measurements of our pilot study we don’t expect significant changes in fluorescence lifetime and only snapshots of the metabolic status of the in vivo macrophage can be taken.

The Discussion section has been completed with these results based on review’s suggestion, please see page 11: line 333.

Furthermore, changes in TPE-FLIM parameter can generally be induced by photo bleaching, changes in temperature, cellular metabolic changes and photoproduct formation.

The effect of photo bleaching is not present in this work. As it has been shown in the literature (L. Tiede and M. Nichols, Photobleaching of Reduced Nicotinamide Adenine Dinucleotide and the Development of Highly Fluorescent Lesions in Rat Basophilic Leukemia Cells During Multiphoton Microscopy, Photochemistry and Photobiology, 82: 656-664, 2006), no photo bleaching effects have been observed in cells up to 5 min of data collection using excitation power of 7 mW at 80 MHz repetition rate, 740 nm wavelength and an objective with NA of 1.3. Since in our case the acquisition time was 6.8 s per image and below 4 minutes in total at lower excitation power in vitro and due to absorption and scattering lower excitation power in vivo, defined by emission photons per mW, we assure that no photo bleaching occurred during the fluorescence measurements of NAD(P)H. The overall photon counting rate did not decrease during measurements, which is an indication of absence of any TPE-induced photo bleaching effects.

It cannot exclude the influence of excitation light on metabolism even in the absence of photo bleaching. To assess if any changes have been triggered upon light illumination, the changes in the level of DNA synthesis must be evaluated. This has been done, for example, in the work (N. Nichols et al., Reduction in DNA synthesis during two-photon microscopy of intrinsic reduced nicotinamide adenine dinucleotide fluorescence, Photochemistry and Photobiology, 81: 259-269, 2005). Evaluation of the data shows that we are working below the threshold doses that may trigger a reduction in DNA synthesis.

It is known that both fluorescence intensity and fluorescence lifetime of NAD(P)H depend on temperature. In general, an increase in temperature is accompanied by decreases in fluorescence lifetime and quantum yield due to increased efficiency of non-radiative processes related to thermal effects, such as collisions with solvent molecules, intramolecular vibrations, and rotations. We do not see a reduction in the fluorescence intensity during the experiments, therefore we believe that at the intensity levels we used, the thermal effects are negligible. In the skin in vivo the temperature is regulated by the transport of water, blood and lymph, and in vitro in the cell culture, the temperature is stable at room temperature for the whole measurement.

In recent research (Shirmanova, M.V., et al. "FUCCI-Red: a single-color cell cycle indicator for fluorescence lifetime imaging." Cellular and Molecular Life Sciences 78: 3467-3476, 2021), the correlation between the cell cycle and fluorescence lifetime of NAD(P)H has been examined and showed that proliferating cells have shorter mean lifetime of NAD(P)H, an indicator of glycolytic shift.

It is known that photo oxidation induces a broadband excitation spectrum and red edge excited fluorescence related to oxidation products of aromatic amino acids from an irradiation of 10 mW/cm intensity during 6 h at 254 nm (Semenov AN, Yakimov BP, Rubekina AA, et al. The Oxidation-Induced Autofluorescence Hypothesis: Red Edge Excitation and Implications for Metabolic Imaging. Molecules. 25(8): 1863, 2020), leading to shortened fluorescent lifetimes and an increase in fluorescence intensity. In our investigations given we used two-photon excitation at 760 nm and significantly shorter irradiation time, it can be excluded that photoproducts had an impact on the measured TPE-FLIM parameters in vitro and in vivo.

Regarding the ΜΦ polarization, the paradigm shifts towards a less strict classification compared to M1 (IFN-γ/LPS-polarized) and M2 (IL-4-polarized) (Murray PJ. Macrophage Polarization. Annu Rev Physiol. 79: 541-566, 2017). While this categorization is useful in clinical terms, the multitude of parameters leading to the differentiation process leaves ΜΦs with wide-ranging properties both in expression of markers, appearance and TPE-FLIM parameters.

M1 (IFN-γ/LPS-polarized) ΜΦs rely on the NADH oxidase and production of ROS, which is shown by fluorescent lifetimes of under 250 ps and mitochondrial fission, which can indicate the bright spots, whereas M2 (IL-4-polarized) rely on oxidative phosphorylation and fatty acid oxidation, together with mitochondrial fusion, it can explain the homogeneous appearance of M2 ΜΦs (Xu Q, Choksi S, Qu J, et al. NADPH Oxidases Are Essential for Macrophage Differentiation. J Biol Chem. 291(38): 20030-20041, 2016; Ramond E, Jamet A, Coureuil M and Charbit A. Pivotal Role of Mitochondria in Macrophage Response to Bacterial Pathogens. Front. Immunol. 10:2461, 2019; Swindle EJ, Hunt JA, Coleman JW. A comparison of reactive oxygen species generation by rat peritoneal macrophages and mast cells using the highly sensitive real-time chemiluminescent probe pholasin: inhibition of antigen-induced mast cell degranulation by macrophage-derived hydrogen peroxide. J Immunol. 169(10): 5866-5873, 2002).

Future investigations will address the change in TPE-FLIM parameter with metabolic stress and shifts in macrophage polarization.

Additional information was added in the Results section on page 5: line 169, page 8: line 245 and in the Discussion section on page 10: line 320, page 11: line 333 and page 13: line 440.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Reviewer #1 (Recommendations for the authors):

Thank you for your clear response to the reviewer's concerns. You have addressed the issue of mixed phenotypes in healthy skin and your in vitro analysis of both in vitro differentiated and freshly isolated M1 and M2 like cells from skin show consistent two photon film signatures. You have provided information about the time involved in the measurements for each subject, the volume scanned and the number of cells found. You have also addressed the temporal stability in vivo consistent with the observation that the isolate cells maintain the flim phenotype even after isolation. The reporting on phagocytosis is now qualified and the limitations of the study are now appropriately acknowledged. I have no further recommendations.

We are grateful to the Reviewer for the positive assessment of our work and responses.

Reviewer #3 (Recommendations for the authors):

In general the approach and validation of the technique is appropriate. I would be interested though to check the relative number of photons collected per pixel versus the different lifetimes to make sure that all lifetimes collected and analyzed are independent on the intensity/number of photons.

We agree with the Reviewer that in the FLIM data correlation between the fluorescence intensity (e.g. photon counts per pixel) and decay parameters could be observed. The question of the dependence of fluorescence decay parameters on the intensity is also connected with further questions of the Reviewer, namely, the dependence of fitting quality on the intensity. To address this issue, Appendix 1-figure 1 shows the dependence of fluorescence decay parameters for macrophages on the integral fluorescence intensity.

As can be seen, the fluorescence decay parameters were independent on the intensity (also note that here a1 is the normalized fraction of the first component, i.e., a₁ + a₂ = 1).

We also note that all the fluorescence decay parameters presented in the paper were calculated for individual cells, i.e., averaged over ~100 pixels with binning 3 (i.e., summation over 48 neighbouring pixels was performed), thus providing a reasonable number of photons for fitting. We would also like to add that the question of FLIM analysis over individual cells (averaging, the dependence of fitting errors on the number of photons etc) was discussed in our recent paper (Shirshin et al., PNAS, 2022, 10.1073/pnas.2118241119).

To further confirm the absence of artefacts connected with parameters dependence on the FLIM data quality and processing algorithms, in Appendix 1-figure 2 the dependence of fluorescence decay curves parameters on χ² is shown.

The dependence of χ² on fluorescence intensity (both integral intensity per pixel and amplitude of the fluorescence decay curve, “kinetic max”) is a textbook knowledge – lower signal to noise ratio results in a worse fitting quality and higher χ² (please see two panels to the right in Appendix 1-figure 2). Importantly, there was no correlation between the fluorescence decay parameters obtained from the decay curves and χ², hence, there were no artefacts like lower fitting quality resulting in lower (or higher) values of fluorescence decay parameters.

Summarizing, the fluorescence decay parameters obtained from biexponential fitting were independent on the intensity (number of photons per pixel) and fitting quality, thus making it possible to use them as the descriptors for classification of cells.

This discussion was added to the SI and briefly addressed in the Materials and methods section of the paper.

Also, some of the lifetimes recovered were very short and close to the resolution of the instrument response function (IRF); which by the way seems not to be recovered experimentally but rather produced by the SPCI software. Please, clarify and also if it was not recovered experimentally produce these data.

Indeed, for some cells the fluorescence decay was quite fast. The IRF of our setup was measured using the SHG signal and its FWHM was estimated as ~100 ps (please see Author response image 1).

Author response image 1

Download asset Open asset

In Author response image 1, fluorescence decay curves for individual cells are divided into two classes: those with τ₁ < 300 ps (left) and τ₁ > 300 ps (right). It can be seen that even in the case of short lifetimes, the IRF is more “narrow” than the fluorescence decay curve. We definitely agree with the Reviewer that some of the lifetimes are too short, however, (1) the absence of intercorrelation between the fluorescence decay parameters and (2) the fact that for the majority of curves τ₁ significantly exceeds the FWHM of the IRF prove that the decay parameters recovered by the procedure employed in our work can used can be used for classification of cells into subpopulations.

This brings me to my next point. The authors employ a double exponential approach and present both the average lifetime (calculated based on tau1 and tau2 and not with the appropriate formula, see Padilla-Parra et al., 2008 Biophys J). The authors do not show the fluorescence decays for each cell, together with the double exponential fits nor the goodness of these fits or the Χ2 (chi square). This data is fundamental to understanding if all FLIM data recovered can be fitted to the double exponential model. Also, the authors assume a double exponential approach but they do not justify their choice. Which are the two populations you are assuming to co-exist in the cells/dermis? Are you always measuring NADPH/NADP? In this case, have you shown the value of these two lifetimes in vitro, and then you should fix these two lifetimes and recover different proportions, right? The model to fit your data should be discussed and justified. I assume that if you use a triple exponential most of your stats will be better. Please discuss this also in the context of the number of photons.

This question of the Reviewer is connected with the experimental basis underlying the metabolic FLIM of NADH. Below, we’ll briefly justify our model based on the literature knowledge base and understanding of the NAD(P)H fluorescence.

It is commonly accepted that the fluorescence signal of a cell at 700–760 nm excitation can be ascribed to NAD(P)H (note that NAD+ and NADP+ do not fluoresce). The idea of using FLIM for metabolic imaging is based on the assumption that the first (fast) component in fluorescence decay of NAD(P)H is originated form free NAD(P)H and the second (slow) component is originated from NAD(P)H bound to enzymes. The reasons of the differences in lifetimes of free and bound NAD(P)H forms are extensively studied (Cao et al. Chem Phys Lett, 726: 18, 2019; Cao et al. J Phys Chem B 124: 6721, 2020; Gorbunova et al. J Phys Chem B 125: 9692, 2021). The basis of using the a1/a2 ratio in the fluorescence decay of NAD(P)H for assessing metabolic (redox) alterations in cells is reviewed in e.g. (Blacker et al. Free Radic Biol Med 100: 53, 2016; Kolenc et al. Antiox Redox Signal 30: 875, 2019).

The lifetimes of NAD(P)H can not be fixed during fitting due to the following reasons. Firstly, NAD(P)H in cells is bound to a manyfold of enzymes, and different enzymes cause different lifetimes of the bound NAD(P)H (in the 1.5–6 ns range Blacker et al. Free Radic Biol Med 100: 53, 2016; Kolenc et al. Antiox Redox Signal 30: 875, 2019; Shirshin et al. Biochem 84: 69, 2019). Hence, the second lifetime of NAD(P)H represents the average over different bound NAD(P)H molecules, and its variation is meaningful and corresponds to redistribution of NAD(P)H between distinct proteins. Secondly, free NAD(P)H in aqueous solution exhibits two lifetimes, ca. 200 and 800 ps (Cao et al. Chem Phys Lett, 726: 18, 2019; Cao et al. J Phys Chem B 124: 6721, 2020; Gorbunova et al. J Phys Chem B 125: 9692, 2021). The origin of these components is debatable and originates either from conformational states of the whole molecule or from isomerisation in the nicotinamide ring (Gorbunova et al. Phys Chem Chem Phys 22: 18155, 2020; Gorbunova et al. J Phys Chem B 125: 9692, 2021). The latter lifetime is close to that for some bound NAD(P)H forms (Vishwasrao et al. J Biol Chem 280: 25119, 2005). Hence, the mechanism of NAD(P)H decay in such a complex environment as a cell is complicated, and for metabolic FLIM a simple two-component model is used (see, e.g., the Becker and Hickl guideline https://www.becker-hickl.com/applications/metabolic-imaging/). Summarizing, the biexponential fitting procedure used in our work is a standard, and it is also used in the papers on macrophages (Lemire et al. Int J Mol Sci 23: 2338, 2022; Alfonso-García et al. J Biomed Opt 21: 46005, 2016).

We would also like to note that in the work of (Blacker et al. Nat Comm 5: 1, 2014) an approach to separate NADH and NADPH impacts using three exponentials was suggested, but it is not a standard and requires high intensity signals.

According to the theoretical estimations, if the number of photons is less than 10000, there is no need to use more than two exponentials (Kollner et al. Chem Phys Lett 200: 199204, 1992). As for the endogeneous fluorescence signal is rather weak (ca 1000 photons in the fluorescence decay curve maximum), biexponential approximation is appropriate, as stated above. To additionally illustrate this fact, we performed comparison of fitting of the fluorescence decay for macrophages with 2 and 3 exponents (Appendix 1-figure 3). As can be seen, an increase of the number of components does not result in a notable increase of the fitting quality.

Also, all the parameters recovered from the fit should be also shown pixel by pixel so that we can understand how these data vary as a function of the number of photons, or the error (Chi Square).

Please see the answer to the first question.

In Figure 4 the authors decide to show the phasor plot (Digman et al., 2008 Biophys J). This is a nice approach that does not necessitate the assumption of a model. You should present all data comparing this approach and the fitting approach utilizing a double exponential and also discuss how your data varies depending on the number of photons. It is possible that some lifetime distributions arise from the fact that in some cells the high intensity values give longer lifetimes and in others where there is a lower signal to noise, lower lifetimes are observed. When plotting your lifetimes versus the number of photons pixel by pixel no dependence should be observed. This will help to validate your data.

The absence of intensity dependence of fluorescence decay parameters was shown (please see Appendix 1-figure 1). Below we will address the correlations between the phasor approach and biexponential fitting.

The phasor approach is a fit-free method used to represent the FLIM data. The basics of the phasor plot is that the position of the point on the phasor plane represents the decay parameters of the fluorescence curve. We’ve performed a detailed analysis of the capabilities of phasor plot, biexponential fitting and K-means clustering in the analysis of the NAD(P)H FLIM data (similar to the presented paper) in our recent work (Shirshin et al. PNAS, 119: e2118241119, 2022). Briefly, the results of the biexponential fitting and phasor analysis are almost similar, with fitting being more beneficial in separating between cells subpopulations with close fluorescence lifetimes. Hence, these two approaches generally yield similar information when aiming at detecting cells subpopulations.

We also demonstrate the correlation between the phasor data (c and s) and the fluorescence decay parameters obtained from biexponential fitting for single macrophage cells in Author response image 2.

Author response image 2

Download asset Open asset

As can be seen, the c value (cosine, x-axis of the phasor plot) is correlated with all decay parameters, while s (sine, y-axis) mainly correlates with a1. We do not draw out any conclusions from these correlations, as the very idea of the phasor plot is the representation of similar data compared to the fitting procedure, hence, the parameters must be correlated (and it is so).

Author details

1. Marius Kröger

   Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt- Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Berlin, Germany

   Contribution

   Data curation, Software, Formal analysis, Investigation, Writing – original draft

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-0148-4225

2. Jörg Scheffel

   Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt- Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Berlin, Germany

   Contribution

   Resources, Data curation, Investigation, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

3. Evgeny A Shirshin

   Lomonosov Moscow State University, Faculty of Physics, Moscow, Russian Federation

   Contribution

   Formal analysis, Validation, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

4. Johannes Schleusener

   Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt- Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Berlin, Germany

   Contribution

   Formal analysis, Validation, Investigation, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

5. Martina C Meinke

   Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt- Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Berlin, Germany

   Contribution

   Resources, Supervision, Funding acquisition, Project administration, Writing - review and editing

   Competing interests

   No competing interests declared

6. Jürgen Lademann

   Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt- Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Berlin, Germany

   Contribution

   Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

7. Marcus Maurer

   Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt- Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Berlin, Germany

   Contribution

   Conceptualization, Data curation, Supervision, Funding acquisition, Methodology, Writing – original draft

   Competing interests

   No competing interests declared

8. Maxim E Darvin

   Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt- Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Berlin, Germany

   Contribution

   Conceptualization, Data curation, Formal analysis, Supervision, Validation, Methodology, Writing – original draft, Project administration

   For correspondence

   maxim.darvin@protonmail.com

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-1075-1994

• 755

  Page views

• 206

  Downloads

• 1

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.