Fluorescence lifetime imaging microscopy and its applications in skin cancer diagnosis

Fluorescence lifetime (FLT) of °uorophores is sensitive to the changes in their surrounding microenvironment, and hence it can quantitatively reveal the physiological characterization of the tissue under investigation. Fluorescence lifetime imaging microscopy (FLIM) provides not only morphological but also functional information of the tissue by producing spatially resolved image of °uorophore lifetime, which can be used as a signature of disorder and/or malignancy in diseased tissues. In this paper, we begin by introducing the basic principle and common detection methods of FLIM. Then the recent advances in the FLIM-based diagnosis of three di®erent skin cancers, including basal cell carcinoma (BCC), squamous cell carcinoma (SCC) and malignant melanoma (MM) are reviewed. Furthermore, the potential advantages of FLIM in skin cancer diagnosis and the challenges that may be faced in the future are prospected.


Introduction
In recent years, with the increase of people's outdoor activities and the aggravation of air pollution, the incidence of skin cancer has gradually increased. 1,2 Take malignant melanoma (MM) as an example, from 1975 to 2014, the incidence of MM increased year by year, as shown in Fig. 1.
According to the American Cancer Society, the number of new cases of cutaneous melanoma in the United States in 2018 is estimated to be 91,270, accounting for 5.3% of all new cancer cases. 3,4 In China, skin cancer accounts for about 1.5% of all malignant tumors. 5 According to di®erent types of skin cells that are primarily a®ected, skin cancer can be mainly divided into basal cell carcinoma (BCC), squamous cell carcinoma (SCC) and MM, etc. 6 BCC is the most common, accounting for 65-75% of all skin tumors. SCC is prone to regional lymph node metastasis. It is much more dangerous than basal cell carcinoma. MM is a highly malignant tumor derived from neuronal melanocytes, accounting for 1% of systemic malignant tumors. It is prone to lymph node metastasis and blood spread. MM has a poor prognosis and causes the most deaths. 7 As the largest organ of the integumentary system, human skin is the outer covering of the body, which provides accessibility and convenience for detection. If an appropriate method can be found for early diagnosis and timely treatment, about 90% of skin cancers can be controlled or even completely cured before they spread. 8 The traditional diagnosis of skin cancer is by biopsy and histopathological examination with hematoxylin-eosin (H&E) staining. However, it is traumatic and cannot be quantitatively detected. Recent technical advances o®er many noninvasive optical detection methods, such as confocal microscopy, two-photon microscopy, Raman spectroscopy,°uorescence spectroscopy, optical coherence tomography, super-resolution microscopy imaging and multispectral imaging technique, etc. [9][10][11][12][13] Among these newly developed imaging techniques, two-photon microscopy is one of the most famous landmark inventions, and°uorescence lifetime (FLT) imaging opens up new detection functions for two-photon microscopy imaging. 14 Unlike intensity-based measurements, the FLT is independent of spectral similarity and concentration of multiple°uorophore labels employed and moreover it is a sensitive means for evaluating microenvironment, which makes°uorescence lifetime imaging microscopy (FLIM) a critical research tool in biomedicine by producing spatially resolved images of°uorophore lifetime. 15 In recent years, the use of FLIM technology for skin cancer research has gradually increased and the statistical numbers of SCI articles since 1997 are shown in Fig. 2 (Based on the keywords of°uorescence lifetime and skin cancer, statistics in the Web of Science database).
FLIM provides a new way for cell biologists to detect, visualize, and investigate structure and function in biological systems and allows the direct diagnosis before tissue biopsy and the identi¯cation of tumor boundaries during surgery. 16 Many universities and scienti¯c research institutions in the world have performed FLIM-related researches, such as the construction on experimental platforms, FLIM data processing and analysis, biomedical diagnosis and applications. [17][18][19] The company Jenlab GmbH of Germany is a pioneer in exploring the application of two-photon FLIM to clinical research and its DermaInspect system is the world's¯rst two-photon FLIM device for clinical skin lesion detection. 12,20,21 Domestic research on FLIM started fairly late, and only a few research groups carried out this study. Among them, the group of Prof. Qu from Shenzhen University has been engaged in two-photon FLT imaging and its biomedical applications for many years and they have applied FLIM to the study on cancer identi¯cation and molecular diagnostics. [22][23][24] In this paper, the basic principle and common detection methods of FLIM will be introduced. And the latest research progresses based on FLIM in the diagnosis of three di®erent skin cancers, including BCC, SCC and MM will be reviewed. Furthermore, the potential advantages of FLIM in skin cancer diagnosis and the challenges that may be faced in the future will be prospected.

Fluorescence Lifetime Imaging
Microscopy 2.1. Principle FLT is a dynamic process of°uorescence intensity decay. When a substance is excited by a laser of a suitable wavelength, the molecule of the substance absorbs energy and transitions from the ground state to an excited state, and then emits°uorescence with longer wavelength in the form of a radiation transition to return to the ground state. 25 The FLT refers to the average amount of time that the molecule stays in its excited state before emitting a photon. Usually, it is de¯ned as the time required for the°uorescence intensity to decay from its peak value to 1/e of its peak value.
Speci¯cally, after the excitation is stopped, the°u orescence intensity that is proportional to the populations in the excited state will decay exponentially with time. For a¯xed°uorescent molecule, a single exponential function can be used to describe this process: Here, is the°uorescence lifetime, t is the time, and I 0 is the initial°uorescence intensity at t ¼ 0, and IðtÞ is the°uorescence intensity at time t, respectively.
In biological samples, there are more than one kind of substances that generate°uorescence when excited by a laser. The superposition process of the FLT of multiple°uorescent substances can be expressed by a multi-exponential function, as shown below: where i is the FLT of the ith°uorescent substance and i is the ith weighting factor.

Fluorescence lifetime measurement method
There are many methods for determining the FLT of a substance and they can be divided into two main categories: frequency domain method and time domain method. 26

Frequency domain method
The frequency domain method is also called phasemodulation method. It usually uses a sinusoidally modulated continuous-wave laser to excite a sample.
The°uorescence emitted from the sample is also sinusoidally modulated with the same modulation frequency as that of the excitation source. The FLT value is calculated by measuring the demodulation coe±cient and the phase shift between the°uorescence and the excitation light. 27 Di®erent modulation frequencies (generally the reciprocal of the°uorescence lifetime) can be chosen for di®erent samples, thereby expanding the measurement range of°uorescence lifetime. Figure 3 shows the schematic diagram of the frequency domain method. 28 Assuming that the phase shift of°uorescence relative to the excitation light is ' and the modulation factor is M, the formulas for calculating the FLT are Here, ! is the modulation frequency; for the one-component°uorescence decay, ' ¼ M (where ' is the phase lifetime and M is the modulation lifetime, respectively).

Time domain method
The time domain method is also called the pulse method. When a sample is excited by an ultra-short optical pulse, the time-resolved°uorescence will  Figure 4 shows the schematic diagram of the time domain method. 28 A simple case is one-component FLT measurement. By measuring the attenuation of the°uorescence intensity at di®erent time, the FLT of a molecule is calculated with the following formula: Here, I 1 and I 2 are the intensities at time t 1 and t 2 , respectively.

Fluorescence lifetime imaging microscopy system
FLT measurement and imaging can be obtained by di®erent implementations, either in time domain or in frequency domain as stated above. TCSPC-based FLIM has the best signal-to-noise ratio of any FLIM techniques and it is an accepted gold standard for FLT measurement. Here we show the schematic of a typical TCSPC-based FLIM system 31,32 in Fig. 5.
The system mainly consists of a picosecond/ femtosecond pulsed laser source, an inverted°uorescence microscope, a high speed confocal scanner, and a TCSPC module. A high repetition rate picosecond or femtosecond pulsed laser is employed to excite the sample through a laser scanning microscope. The°uorescence emitted from the sample is detected by the PMT after an emission¯lter and then sent to the time-correlated single photon counter for photon-number statistics. Besides the light delivered to the microscope for sample excitation, a small portion of the excitation light is used as a stop pulse for TCSPC. The time probability distribution of the arrival photons and the°uorescence intensity decay curve can be obtained. The FLT of the sample is measured point by point, and the FLT image of the region of interest is reconstructed.

Application of FLIM in Skin Cancer
Diagnosis 3.1. Basal cell carcinoma BCC is the most common skin cancer and occurs mainly in fair-skinned patients. It accounts for at least 32% of all cancers globally. In the United States, about 35% of white males and 25% of white females are a®ected by BCC at some stage in their lives. 33 BCC begins in the basal cell layer of the skin and can damage the tissue around it, but it has a low metastatic rate and is unlikely to cause death.
Existing research on BCC mainly focused on isolated tissues. In 2008, Galletly et al. 34 used FLIM technology to distinguish between BCC and surrounding normal skin tissue. The sample was excited by a 355 nm pulsed laser to obtain auto-FLT images of 25 unstained BCC tissues. The results showed that BCC and normal tissues could be clearly distinguished in the wide-¯eld pseudo-color FLT image, and the FLT of BCC was shorter than that of normal skin tissue.
In 2011, based on FLIM technology, Patalay et al. 35 used two spectral channels to detect°uorescence intensity and°uorescence lifetime, and then segmented the images to calculate the FLT of the regions of interest (ROI). A 760 nm laser was employed to excite freshly removed benign sputum (naevi) and malignant nodular basal cell carcinoma (nBCC). There was a statistically signi¯cant difference between the mean°uorescence lifetimes of nBCC and naevi, which may be derived from each spectral channel. The°uorescence lifetimes of NADH,°avin and melanin were di®erent. In the spectral channel of < 500 nm, the FLT of NADH contributed a lot. While in the channel of > 500 nm, the FLT of melanin and°avin contributed a lot.
In 2012, Patalay et al. 36 used multispectral multiphoton FLT imaging to di®erentiate BCC from normal skin tissue. The image data of the isolated BCC tissue was manually divided into 10,462 regions. The di®erence in spectrum and cell morphology between BCC and normal skin was quanti¯ed for the¯rst time. BCC and normal skin tissue were distinguished by image characteristics of FLT with sensitivity and speci¯city of 79% and 93%, respectively. In the same year, Seidenari et al. 37 compared the lifetime distribution of healthy skin tissue and BCC. Based on two-photon FLT imaging, several morphological and numerical descriptors that can distinguish BCC and other skin lesions were extracted to evaluate the sensitivity and speci¯city of Multiphoton laser tomography (MPT)/FLIM for improving BCC diagnosis and identi¯cation of tumor margins.
In 2014, Fan et al. 22 utilized a two-photon°uorescence imaging system with a 790 nm pulsed laser to excite the endogenous°uorophores of the isolated BCC samples and imaged the structure and morphology of the cells and extracellular matrix. By spectral scanning in the 450-700 nm band, they found that it was indistinguishable between normal epidermal tissue and BCC tissue because of their same auto°uorescence emission peak at 545 nm. However, by imaging the two-photon FLT of en-dogenous°uorescent substances, it was found that BCC tissue has a longer FLT relative to normal skin tissue.
In 2017, Luo et al. 38 distinguished between BCC, actinic keratosis (AK) and Bowen's disease (BD) by FLT and phasor analysis of H&E stained sections. The study found that BCC, AK and BD could be identi¯ed according to the distribution of°uorescence lifetimes of stratum corneum (SC), epidermal cells (ECs) and connective tissue (CT), as shown in Fig. 6. In addition, the analysis of the FLT of H&E stained tissue in the phasor space was performed by observing the phasor diagram. The coordinate values, diagonal slopes, and phasor regions of the skin layers in the vector diagram were di®erent. FLT imaging microscopy and phasor method (phasor-FLIM) provided a simple, noninvasive histopathological analysis for the diagnosis of skin cancer.

Squamous cell carcinoma
SCC, the second most common form of skin cancer is also known as epidermal carcinoma. More than 1 million cases of SCC are diagnosed and more than 15,000 people die from the disease each year in the United States. SCC is a malignant tumor that mostly occurs in the squamous epithelium and usually results from human papillomavirus heat damage, ultra violet radiation and chemical carcinogens, etc. 39 SCC is clinically and histopathologically diverse and the cancer cells have di®erent degrees of keratinization at di®erent stages of development. Early diagnosis and treatment can inhibit cancer cells to a certain extent.
In 2010, Martín-Villar et al. 40 used in vivo°uorescence resonance energy transfer/°uorescence lifetime imaging microscopy (FRET/FLIM) to measure the interactions between Podoplanin-eGFP (donor) and CD44s-mRFP (receptor). The experimental results showed that the expression of Podoplanin in tumor cells, especially SCCs, was linked to increased cell migration and invasiveness, and Podoplanin-CD44s interaction was associated on the plasma membrane of cells with a migratory phenotype. FRET/FLIM demonstrated that the Podoplanin-CD44s interaction was important for driving tumor cell migration during malignancy.
In 2011, Roberts et al. 19 used FLIM to study skin conditions. The FLIM images of the stratum granulosum, stratum spinosum and stratum basale from healthy, psoriasis lesional, atopic dermatitis lesional and SCC lesional skin were shown respectively. According to FLT decay of double-exponential function, the ratio of amplitudes (a 1 =a 2 ) of NAD(P)-H was used as an indication of metabolic rate. The results showed that the metabolic rates di®ered depending on the skin conditions. Although FLIM could distinguish between di®erent lesions and healthy skin, its combination with multiple complementary modalities such as spectral imaging, re°ectance and Raman imaging with confocal and multiphoton spectroscopy were suggested to achieve robust diagnostic discrimination between skin diseases and their boundaries.
In 2015, Miller et al. 41 used intensity, time and wavelength resolved multiphoton microscopy to measure changes in cellular metabolism. By analyzing the°uorescence emission of metabolic coenzymes such as the reduced form of NADH and the oxidized form of°avin cofactors, slight changes in the mitochondrial microenvironment and cellular metabolism can be quanti¯ed. Dual-channel FLIM was used to quantify metabolic changes both within and between high-and low-HER2 expressing SCC cell lines (SCC-74B and SCC-74A, respectively). Chemically induced changes in metabolic state resulted in measurable changes in lifetime distribution and°uorescence intensity within each cell line. In addition, SCC-74A and SCC-74B could be clearly distinguished based on the dynamic range of°u orescence intensity (the di®erence between uncoupled and inhibited metabolic states) and the lifetime distribution. In 2017, Miller et al. 42 characterized skin cancer in living mice with SCC by using multimodal°uorescence molecular imaging method. Firstly, the endogenous°uorophores, such as°avin and lipofuscin, were excited by 480 nm laser. According to the auto-FLT imaging, it was found that the shortlived component in SCC increased relative to normal skin, indicating that the FLT of the tumor was shorter than that of normal skin, as shown in Fig. 7. In addition, a near-infrared (NIR)°uorescent molecular probe was injected into a mouse su®ering from SCC and subjected to°uorescence molecular tomography (FMT), which provided a method to estimate the tumor volume and depth, as well as quantify the molecular probe concentration in SCC. All detection and imaging methods in this study were performed on live mice, providing a comprehensive model for skin tumor research.

Malignant melanoma
MM is a very aggressive and dangerous type of cancer that develops in the pigment-containing cells known as melanocytes. An estimated 178,560 cases of melanoma were diagnosed in the United States in 2018. It is especially important for MM's early diagnosis and treatment because of its high malignancy, easy to early metastasis and high mortality. 43 Early stage melanoma has an overall survival rate of nearly 100%, while metastatic melanoma can be rapidly fatal. Until now, MM has been studied using cell models, ex vivo tissues and animal models.
In 2008, Cicchi et al. 44 used the two-photon FLIM experimental system to perform lifetime imaging analysis of MM and melanocytic nevus (MN). They evaluated the lifetime distribution of a layer containing MM or MN cells using both single-and double-exponential decay function. The results showed that MM exhibited a broader mean lifetime distribution with respect to the corresponding MN mean lifetime distribution in both single-exponential and double-exponential decay analysis, and the di®erentiation between MM and MN occurred at the fast lifetime component level in double-decay analysis. Therefore, it could be discriminated between MM and MN by looking at the di®erent In 2009, by selective melanin imaging and spectral FLT imaging, Dimitrow et al. 45 found that the FLT distribution was in correlation with the intracellular amount of melanin. Compared to keratinocytes, MM exhibited a greater ratio of the intensity coe±cients a1=a2 (a1 was the shorter component and a2 was the longer component), so MM had a shorter°uorescence lifetime. Procedures of selective imaging as well as spectral FLT imaging supported diagnostic decisions and could improve the process of noninvasive early detection of melanoma.
In 2014, Pires et al. 46 performed experiments on 42 experimental melanoma lesions induced in balb/c nude mice using the cell line B16F10, with a 378 nm laser to excite NADH and a 445 nm laser to excite FAD. The experimental results showed that the two methods could distinguish normal skin and melanoma according to the change of°uorescence lifetime. In practice, the contribution of various endogenous°uorescent molecules in the tissue should be considered comprehensively. Using this time-domain°uorescence spectroscopy to detect melanoma in animal models, the sensitivity of 99.4%, speci¯city of 97.4% and accuracy of 98.4% were achieved.
In 2017, Pastore et al. 47 used a melanoma mouse model to compare the morphological and metabolic state changes of normal skin and melanoma tissue at three di®erent growth stages based on NADH signaling. It was found that with the growth of melanoma, the ratio of free NADH to protein-bound NADH increased signi¯cantly (Fig. 8), while the mean FLT of NADH decreased. This study demonstrated that the use of two-photon FLIM to detect the metabolic state characterized by NADH could quickly and sensitively re°ect the growth stage of melanoma. FLIM technology had the potential to be a powerful tool for the diagnosis and staging of melanoma.

Summary
Skin cancer is the most common form of cancer that is caused mostly by exposure to ultraviolet radiation from the sun. Research shows that skin cancer is associated with abnormal cellular metabolism and its early identi¯cation may introduce the possibility of intervention to prevent its progress to a deadly metastatic stage. FLIM enables noninvasive detection of complex molecular assemblies for studies of cell function and dynamics with subcellular resolution by imaging the lifetime of the°uorophore signal rather than its intensity. It can be used as an imaging technique in confocal microscopy, twophoton excitation microscopy and multiphoton tomography, etc. With these advantages, FLIM has been performed as a powerful imaging tool in cell biology and biomedicine. This paper brie°y summarizes the principle and detection methods of FLIM and reviews in detail its applications in the diagnosis of skin cancer, including BCC, SCC and MM. By measuring the FLT of endogenous (such as NADH, FAD, etc.) or exoge-nous°uorophores, FLIM provides information on both morphology and metabolism of the tissues under investigation at a subcellular level, which helps di®erentiate between normal skin tissue and cancerous tissue as well as presents a great potential in clinical application.
However, the application of FLIM in skin cancer is still in its infancy. No strict de¯nition to quantitatively describe normal skin tissue and cancerous tissue, limitations in hardware systems and imaging principles, and lack of proper°uorescent indicators, are the great challenges that FLIM faces in the study of skin cancer. The FLIM system used in clinical dermatology should be noninvasive, low cost, compact and portable, and with high speed, sensitivity and accuracy. Combination with multiple techniques, such as dermoscopy, will help to improve the system performance and e±ciency. We believe that with the development of imaging and detection technologies and the improvement of data analysis capabilities, FLIM, as a noninvasive pathological research tool, will play a role in the early diagnosis and treatment of skin cancer.

Con°ict of Interest
There are no con°icts to declare.