Multidimensional quantitative characterization of the tumor microenvironment by multicontrast nonlinear microscopy

Abstract

Characterization of the microenvironment features of tumors, such as its microstructures, biomolecular metabolism, and functional dynamics, may provide essential pathologic information about the tumor, tumor margin, and adjacent normal tissue for early and intraoperative diagnosis. However, it can be particularly challenging to obtain faithful and comprehensive pathological information simultaneously from unperturbed tissues due to the complexity of the microenvironment in organisms. Super-multiplex nonlinear optical imaging system emerged and matured as an attractive tool for acquisition and elucidation of the nonlinear properties correlated with tumor microenvironment. Here, we introduced a nonlinear effects-based multidimensional optical imaging platform and methodology to simultaneously and efficiently capture contrasting and complementary nonlinear optical signatures of freshly excised human skin tissues. The qualitative and quantitative analysis of autofluorescence (FAD), collagen fiber, and intracellular components (lipids and proteins) illustrated the differences about morphological changes and biomolecular metabolic processes of the epidermis and dermis in different skin carcinogenic types. Interpretation of multi-parameter stain-free histological findings complements conventional H&E-stained slides for investigating basal cell carcinoma and pigmented nevus, validates the platform’s versatility and efficiency for classifying subtypes of skin carcinoma, and provides the potential to translate endogenous molecule into biomarker for assisting in rapid cancer screening and diagnosis.

1. Introduction

Basal cell carcinoma (BCC) [1,2] and pigmented nevus [2] are common skin conditions worldwide; BCC is the most prevalent type of nonmelanoma skin cancer, and its incidence increases every year [1]. However, the false-positive clinical diagnosis of skin pathology caused by the complexity of the skin structure and limitations of preoperative biopsy could possibly leads to unnecessary tumor resection and exorbitant treatment costs [3,4]. Histological examination, an essential technique for preoperative screening, intraoperative diagnosis and prognosis assessment, is performed by visualizing the microstructures and morphological changes in standard hematoxylin and eosin (H&E)-stained tissue sections to determine the degree and differentiation of the disease [5,6]. However, the preparation of H&E-stained pathological sections, including fixation, dehydration, embedding, sectioning and staining, is a complicated and time-consuming process that slows the clinical intraoperative diagnosis [6]. Moreover, histological processing is a destructive process, possibly resulting in distortion and loss of natural biomolecular information and even abnormal tissue necrosis or damage due to alien reagents. In contrast to the tissue stains, certain immunohistochemical staining techniques can characterize specific marker molecules, but require more complicated preparations and alien reagents. Overall, conventional evaluation methods require specific dyes/probes/plasmids to intrusively gain specific functions of metabolic and molecular dynamics in the tumor microenvironment, thereby have a limitation in comprehensive and objective interpretation of tumor development and invasion.

In recent decades, nonlinear optical imaging (NLOI) technologies have been proposed and optimized to address unstained pathological tissues and provide a wealth of biochemical information about morphology and endogenous molecular metabolism [7,8]. Specifically, second-harmonic generation (SHG) [9] occurs in noncentrosymmetric materials and hyperpolarizable biological molecules, and third harmonic generation (THG) occurs in a nonhomogeneous medium requiring a condition of negative phase mismatch [10]. Two-photon excited fluorescence (TPEF) [11,12] and three-photon excited fluorescence (3PEF) [13,14]are well-known methods to visualize various endogenous fluorophores, such as nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD) [15,16]. The above approaches mainly concentrate on morphological features to match HE-stained histology. Yet, the fluorescence lifetime of autofluorescence sheds light on metabolic changes associated with carcinogenesis by indicating redox metabolic mechanism in the tumor microenvironment [17–19]. Furthermore, coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS) are capable of biochemical visualization and quantification of lipid and protein contents to highlight the molecular metabolism [20]. The integration of these approaches devotes to provide more precise and comprehensive cancer-associated features about microstructure and molecular metabolic according to the unique contrast mechanisms and imagery capabilities of various nonlinear modalities [21].

Here, we introduced a multidimensional optical characterization method (Supplementary Fig. 1 and Supplementary Note 1 in Supplement 1 ^{(17.1MB, pdf)} ) proposed in previous work, which integrates four optical imaging modalities [22], including TPEF, SHG, SRS and two-photon fluorescence lifetime microscope (TP-FLIM), to image and identify the abundant nonlinear optical properties of specific molecules and structures in skin tissues simultaneously, as shown in Fig. 1. This multiplex label-free nonlinear optical imaging method enables the direct observation of macroscopic tissues features and microstructure histomorphological information of skin tissues with sub-micron resolution. The extracted information matches the H&E-stained digital images with complex preparation and emphasizes the tumor microenvironment related to molecular components and metabolic activity. For different pathological subtypes of skin tissues, TPEF, rather than 3PEF with a low signal-to-noise ratio (SNR) due to higher order nonlinear process, can obtain the distribution of elastin fibers and FAD in the cytoplasm. SHG, which is superior THG that visualizes vesicles [13,15] in identifying cancer invasion, is applied to visualized tumor-associated collagen fibers in biological specimens [15,23]. The merged images of TPEF and SHG is widely used for confirming the microstructure of skin tissue according to the optical properties of different endogenous biomolecules and revealing the morphological changes of different skin pathological tissues [24–27]. Moreover, SRS without a nonresonant background can recognize and quantify lipids and proteins in the CH stretching region, and the Raman spectral signatures supplement tumorigenic mechanism about biochemical metabolism of lipids and proteins in skin conditions. TP-FLIM empowered by the fit-free phasor approach provides more precise relative-variation of fluorescence lifetime, and separates free and protein-bound FAD to imply the metabolic specificity in the tumor microenvironment. Mesoscopic structures, lipid-protein transitions, and metabolic events of cellular molecule of freshly excised unstained skin tissues can be exhibited graphically through our versatile multidimensional optical characterization platform, which can conduce to better interpretation of tumor-associated changes in carcinogenesis. Together with the analysis approaches, this system has great clinical potential for the diagnosis and prognosis assessment of tumor invasiveness and differentiation.

Fig. 1.

Multidimensional optical characterization of clinicopathological unperturbed skin tissues. (a) Multidimensional optical imaging method reveals the tumor microenvironment of skin tissues, including the microstructure characteristics of FAD and collagen fiber, and molecular metabolism related to the lipid-protein transition and redox state associated with FADs. (b) Multimodal nonlinear images and H&E-stained image of skin specimens, in which SRS at 2930 cm^-1 indicates protein (b3), TPEF reveals the distribution of FAD (b5), SHG highlights collagen fibers (b6), and TP-FLIM implies the lifetime features of free- and bound-FAD (b4). (b2) These combined nonlinear image can provide label-free molecular profiling of skin specimens, which matches the H&E-stained digital images (b1). Scale bars, 30 µm.

2. Methods

2.1. Sample preparation

All tissue samples, including 5 normal skin, 15 pigmented nevi and 15 BCC, were collected from 7 patients in The Sixth People’s Hospital of Shenzhen with approval by the Ethics Committee with informed written consent. All suspected and confirmed skin cancer patients were recruited, and the tissue samples were surgically removed by surgeons, snap frozen in liquid nitrogen and preserved in a –80 °C refrigerator. We cut these samples into several micrometer sections for multimodal nonlinear imaging and histological imaging applications using a freezing microtome (CM1850, Leica, Germany). Unstained frozen tissue sections were covered with a coverslip only. A few of the frozen section were further stained to obtain H&E-stained slides histological sections, which were imaged for cancer identification and classification by an experienced dermatological oncologist.

2.2. Optical setups

We used an ultrafast laser with a repetition rate of 80 MHz (Chameleon discovery, Coherent Inc.) to provide two synchronized excitation beams: a pump beam at 800 nm and a Stokes beam at 1040 nm. The pulse widths of the pump beam and Stokes beam were chirped from 126 fs and 107 fs to 2.0 ps and 1.8 ps using SF75 glass rods respectively, as measured by an autocorrelator (pulseCheck, APE). An additional group delay dispersion (GDD) of −7000 fs² was applied to the pump beam within the laser to achieve the same linear chirp parameter $\beta$ of the pump and Stokes beams, because the total GDDs were approximately 80488 fs² for the pump beam and 72742 fs² for the Stokes beam [22]. This matched linear chirp parameter resulted in a spectral resolution of 16 cm^-1 for SRS. A linear motion stage (M-ILS250CC, Newport) is placed in the pump path for time overlap of the pump and Stokes pulses and positional scanning to obtain the Raman spectra. For the pump-probe scheme, the intensity of the Stokes beam is modulated at a high frequency of 20 MHz using a resonance EOM (EO-AM-R-20-C2, Thorlabs). The samples were imaged with a 60× water immersion objective lens (UPLSAPO, 60×, 1.2 NA, Olympus). This objective collected the autofluorescence and harmonic signals, which were separated by a dichroic mirror (500 nm) and bandpass filters (550/40 nm and 400/10 nm) and detected by two PMTs (PMT2101/M, Thorlabs). An oil condenser (U-ACC, 1.4 NA, Olympus) collected the transmitted lights, where the modulated Stokes beam was removed by a hard-coated BP filter (825/150 nm, Chroma) with a high optical density (OD >6). We built a transimpedance amplifier with a bandwidth of 100 MHz to convert the current signal of the pump light translated by a PD (FDS1010, Thorlabs) to a voltage signal and send to a lock-in amplifier (HF2PLL, Zurich Instruments) for demodulation of stimulated Raman loss (SRL). Each image was captured at 512 × 512 pixels and with a dwell time of 2 µs. The SRS spectral range was approximately 2750-3000 cm^-1 with a depth of 80 spectral frames.

We routed a femtosecond pulse of the 800 nm beam before chirping for excitation of TPEF, SHG and TP-FLIM to avoid low excitation efficiency by picosecond pulses [22]. We adjusted the delay of this femtosecond pulse to be approximately 100 ps away from the picosecond pulses. This temporal interval is large enough to separate femtosecond pulse from the picosecond pump and Stokes pulses to reduce the thermal effect and small enough to avoid lifetime convolution in FLIM with a resolution of approximately 150 ps. Then, the unbiased lifetime of FAD was detected by a high-speed time-resolved detector (HPM-100-40, Becker & Hickl GmbH) connected to the TCSPC module (SPC-150 and DCC-100, Becker & Hickl GmbH). In addition, we also generated the XY triangular waves for scanning, three digital signals for frame, line, and pixel synchronization of FLIM measurement using an additional DAQ (PXIe 6341, National Instruments). These signals are sent to the SPC module (SPC-150 Becker & Hickl GmbH) for photon counting and image reconstruction.

2.3. Data processing

SHG imaging was obtained repeatedly during SRS spectral scanning and FLIM photon accumulation. Thus, the SNR of SHG images was $\sqrt{\text{N}}$ -fold enhanced by binning N frames into one (N = 80 sequential images for one SRS and FLIM acquisition).

The SNR of the SRS image was greatly enhanced using our previous global spectral denoising (GSD) method [28]. Briefly, we cross-correlated each spectrum in the SRS image stack, $I_{SRS}(\omega )$ with a Lorentzian impulse response, $I_{Lorentz}(\omega )$ :

$$I_{GSD}({\omega }_0)=\text{corr}[I_{\text{SRS}}(\omega ),I_{Lorentz}(\omega -{\omega }_0)],$$ (1)

$$I_{Lorentz}(\omega )=1/\pi \gamma \left[1+\frac{{\omega }^2}{{\gamma }^2}\right],$$ (2)

where 2 $\gamma$ is the FWHM of $L(\omega )$ , which can be set as half to a quarter of the FWHM of the Raman spectrum. ${\omega }_0$ was assigned to every value of spectral point $\omega$ . This correlation extracts the Lorentzian Raman profile from the whole spectral image stack and suppresses the background noise without causing distortion of the spectrum.

The dermis, stratum corneum and granulosum, melanocytes and stratum basale were distinguished using the phasor-mapped multicomponent analysis method [21]. Briefly, the decay traces of the FLIM images are converted to Fourier space using the following relations:

$${\text{s}}_{i,j}(\omega )=\frac{{\int }_0^{\mathrm{\infty }}I(t)\sin (n\omega t)dt}{{\int }_0^{\mathrm{\infty }}I(t)dt},$$ (3)

$${\text{g}}_{i,j}(\omega )=\frac{{\int }_0^{\mathrm{\infty }}I(t)\cos (n\omega t)dt}{{\int }_0^{\mathrm{\infty }}I(t)dt},$$ (4)

where i and j represent the pixel of the image, sine (s) and cosine (g) are phasor coordinates, $\omega =2\pi f$ ( $f$ is the laser repetition rate) and n is the harmonic frequency. The same components of the skin feature approximate decay signatures and hence form a cluster in the phasor plot [21].

We used raw images for data analyses and presented the images with proper dynamic adjustment (brightness/contrast) in ImageJ to show better discernible morphological features. We programmed MATLAB scripts to extract Raman spectra from SRS image stacks and multivariate curve resolution algorithms [29] for reconstruction of tissue components of lipids and proteins and extraction of optimized decomposed spectra. The photon distribution and decay curve in FLIM images were analyzed using SPCImage (Becker & Hickl GmbH). All statistical analyses of the mean, standard deviations (SD), standard error of measurement (SEM), and significant difference were performed using GraphPad Prism.

3. Results

3.1. Noninvasive multidimensional nonlinear imaging unscrambles the structural complexity of skin tissues

To correctly interpret skin pathologies that host to tumor-associated changes and events, we first study the microstructure of human skin by simultaneously capturing multidimensional nonlinear optical images. The pathological abnormalities of skin subtype (pigmented nevus and BCC) mainly occur in the dermis and epidermis (Fig. 2), and the epidermis consists of four layers, including stratum corneum, stratum granulosum, stratum spinosum, and stratum basale [30]. These different skin structures can be explicitly divided on the basis of multiparameter nonlinear properties in this study. The multiple images display the certain molecular and structural components, and the multinonlinear pentagonal radar plot (Fig. 2(d)) offers a representative method to demonstrate the molecular profiles in each skin structure: dermis with SHG characteristics and long fluorescence lifetime, stratum corneum and granulosum featured high lipid and protein signals from SRS images, and melanocytes and stratum basale with high fluorescence intensity and short fluorescence lifetime. The immersive visualization results of high-multiplex images were verified by digital images of colocated H&E-stained slides. In detail, multimodal nonlinear optical contrasts (Fig. 2(a)) from TPEF-visible FAD (Fig. 2(b), 2), SHG-visible extracellular matrix (ECM) composed of collagen fibers and elastic fibers (Fig. 1(b), 1), as well as SRS-visible lipids and proteins (Fig. 2(c)) effectively distinguish the dermis and subdivided epidermis in skin tissues, and the intensity curves (Fig. 2(b), 3) of TPEF and SHG image can quantify the thickness of epidermis in response to the epithelial cell proliferation. Protein signals are generated from keratin and collagen proteins in the epidermal and dermal tissues, and lipid signals (indicated by orange arrows) are produced from the intercellular lipid mixtures composed of phospholipids and free fatty acids in the stratum corneum and spinous layer. The SRS spectra analysis (Fig. 2(c), 1), including position and intensity of Raman characteristic peaks, is particularly suitable for quantifying conformation and content of lipids at 2885 cm^-1 and proteins at 2930 cm^-1. These molecular-level image-based results exhibit the considerable advantages of qualitative and quantitative characterization over the H&E-stained images with macroscopic morphological description. Moreover, the above mesostructured features are coregistered with the TP-FLIM images (Fig. 2(e)), which relate to intracellular redox reaction and supplement the dynamic metabolic signatures of endogenous FAD in tumor microenvironment. The segmentation and analysis of skin microstructure is completed by phasor plot (Fig. 2(e), 4), which matches previous report [31] . Specifically, keratinocytes in the stratum corneum and granulosum exhibit an intermediate fluorescence lifetime (Fig. 2(e), 2), and the fluorescence lifetime of FAD in the dermis (Fig. 2(e), 1) is longer than that in the epidermal, whereas cells or structures with a shorter fluorescence lifetime (Fig. 2(e), 3) and enhanced fluorescence intensity correspond to melanocytes or pigmented keratinocytes and melanosomes in stratum basale. Through the optical features of the various nonlinear imaging modalities and complementary imaging contrasts, we can accurately distinguish complicated skin structures in light of morphology and molecular function information, which is conducive to further interpret and evaluate the different pathological types of skin sections.

Fig. 2.

Macroscopic morphology and microstructures revealed in multiplex nonlinear optical images of normal skin specimens. (a) Pseudocolor presentation merging SHG of collagen (blue), TPEF of FAD (green) and SRS of protein (magenta). 1: dermis; 2: stratum corneum and granulosum; 3: melanocytes and stratum basale. (b) The merged image of SHG (b1) and TPEF (b2). (b3) Normalized intensity profile across the dotted white line in image b. (c) SRS image of lipids at 2885 cm^-1 (cyan) and proteins at 2930 cm^-1 (magenta). (c1) Overall SRS spectra for the entire FOV of image c. (d) Pentagonal radar multinonlinear profile describes the microstructure of skin tissue. (e) TP-FLIM image of FAD. (e4) The cluster areas of phasor-mapped images (1, 2 and 3) correspond to dermis (e1), stratum corneum and granulosum (e2), and melanocytes and stratum basale (e3). The color bar in (a) represents the pseudocolor of the three modalities, the color bar in (e) represents the lifetime range, and those in the other images represent the normalized intensity. Scale bars, 30 µm.

Fig. 3.

Morphological interpretations related to pathological diagnosis of unperturbed freshly clinicopathological skin tissues. Also see Supplementary Figs. 3-5 in Supplement 1 ^{(17.1MB, pdf)} . (a) The optical properties of normal skin(a), pigmented nevus (b) and BCC (b) demonstrated by TPEF (green) and SHG (blue) images. And the normalized intensity profile across the white dotted line in normal skin(a1), pigmented nevus (b1) and BCC (c1) images quantify morphological signatures of epidermis. (d) Statistical analysis of epidermal thickness with the Mann–Whitney test in different cancer types (n = 30). ****: p < 0.0001, ***: p < 0.001, **: p < 0. 01, *: p < 0.05. (e, f) The digital images of H&E-stained pathological section of pigmented nevus and BCC. (e1, f1) The combined images of TPEF (e2, f2) and SHG (e3, f3) in the dermis tissue affected by BCC and pigmented nevi. Scale bars, 30 µm.

Fig. 4.

Quantification of molecular metabolic features related to tumor microenvironment in epidermal tissue. Also see Supplementary Figs. 3–5 in Supplement 1 ^{(17.1MB, pdf)} . Two-color SRS images of clinicopathological normal skin (a), pigmented nevus (b) and BCC (c). Lipids are false-colored cyan, with proteins in magenta. (e, f, g) TP-FLIM pseudocolor images of FAD in normal skin, pigmented nevus and BCC and corresponding phasor plots (h1, i1, j1) based on multicomponent analysis method of FADs. (h) Significant difference analysis of lipids and proteins metabolism in different pathological subtypes of skin tissues from the molecular content and content ratio, R_{lipid/protein} (n = 10). (i) Statistical results of mean fluorescence lifetimes, component lifetimes and ratios (R_bound) about FAD in the stratum corneum and granulosum of epidermis (n = 8). (j) Pentagonal radar multinonlinear profile summarizes the optical properties of the skin epidermis. ****: p < 0.0001, ***: p < 0.001, **: p < 0. 01, *: p < 0.05. Scale bars = 30 µm.

3.2. Multicontrast nonlinear images reveal the morphological features correlated with pathological diagnosis

The heterogeneity of cancer cells, microstructural characteristics of ECM and macro-distribution of endogenous molecules in different clinicopathological skin tissues are significant morphological features related to tumor subtype differentiation. The unperturbed freshly skin tissues, including normal skin, pigmented nevus and BCC, were imaged using label-free multimodal nonlinear optical imaging platform, and we revealed the morphological changes of epidermis and dermis separately in clinicopathological skin tissues due to the complexity and diversity of molecular composition in different skin tissue structures. All clinicopathological tissues were also identified histologically with standard pathological examination and preliminary diagnostic interpretation, by board-certified experienced pathologists, containing morphological description of nucleus, cytoplasm and matrix, as shown in Supplementary Fig. 2. Macroscopic morphological characteristics related to proliferation of the melanocyte and epithelial tumor cells in the epidermal tissue of human skin have been identified and analyzed from TPEF and SHG images [32], which are highlighted by the pseudocolor image of green (TPEF) and blue (SHG), and separated from dermis with white dotted line, as shown in Fig. 3. There are several differences and similarities among the morphological information of tumor microenvironment in pigmented nevus and BCC compared to normal epidermal tissues [27,33] (Fig. 3(a)). Pigmented nevus specimens (Fig. 3(b)) exhibit a higher FAD-generated TPEF signal caused by the increased melanocytes in the epidermis, especially in the stratum basale [24]. Nevertheless, the epidermis of BCC (Fig. 3(c)) features cellular pleomorphism, including irregular cellular contours and larger intercellular spaces, as well as epidermal cells with lost cohesion are disposed of in random order [25]. Moreover, the epidermis of two skin pathological tissues are ulcerated and thickened [27], and we measure the epidermal thickness of normal skin (Fig. 3(a), 1), pigmented nevus (Fig. 3(b), 1), and BCC (Fig. 3(c), 1) to quantify the morphological differences for pathologic researches. The pathological skin tissues (nevus: 49 ± 3 µm, BCC: 80 ± 2 µm) had a thicker epidermis than normal skin (17 ± 1 µm). This morphological difference is significant (p < 0.0001 for nevus and BCC with normal tissue) by quantitative statistical analysis of the merged images of SHG and TPEF (Fig. 3(d)) in the skin epidermis. These similarities and differences of the epidermis are utilized to distinguish tumor regions and classify cancer subtype.

We further analyzed and discussed the morphological characteristics of dermal tissue in pigmented nevus and BCC. The ECM composed of collagen fibers, microfibrils and elastic fibers is an essential component of the dermis in normal skin and features unique optical properties of SHG, nevertheless, various cancerous skin tissues exhibit structural diversity due to tumor invasion and metastasis. Morphologic manifestations of nevus and BCC can be directly demonstrated by multiple nonlinear imaging, especially TPEF and SHG. Concretely, “dermal cell clusters” or isolated cells (melanocytes) can be observable in the dermis of pigmented nevi (Fig. 3(e), 1), which represents a significantly enhanced TPEF signal and disappearance of the SHG signal [34–36]. Nevertheless, round or oval tumor nodules filled with mucin and cystic space are surrounded by fibers in the dermis of BCC (Fig. 3(f), 1), and palisading phenomena are visible at the edge of the nodule [36,37]. Therefore, plentiful morphologic information of tumor microenvironment related to cancer aggressiveness and progression, including ultrastructural morphology of epidermal cells, macroscopic distribution of melanocytes, and degradation of cancer-associated fibroblasts, can be identified by multi-parameter nonlinear optical signals of endogenous FAD and collagen fibers. These multimodal label-free optical imaging modalities perform substantial advantages to reveal abundant pathology, which are verified by the images of colocated H&E-stained pathological slides of pigmented nevi (Fig. 3(e)) and BCC (Fig. 3(f)), further over traditional H&E-stained section with time-consuming preparation.

3.3. Molecular metabolism of tumor microenvironment related to pathological findings in epidermis

Cancerous specimens are featured with extensive heterogeneity, yet the comprehending of which were generally only limited within tumor cells [38] and determined by the morphological findings. The tumor microenvironment, as a remarkably complex and diverse cellular ecosystem [39], can be understood by the metabolic and functional information, which may be obliterated during the preparation of H&E-stained histology. As expected, the SRS technology described the spatial distribution of tumor-associated lipids and proteins in the epidermis and dermis, which provide a specificity to analyze the various skin conditions. TP-FLIM images of FAD combined with the bicomponent cluster analytical method in phasor plot offer abundant pathological functional information initiated by oxidation-reduction in tumor microenvironment of skin samples. Our findings show that lipids diffuse to the stratum granulosum and dermis among the cancerous skin cases (Fig. 4(b) and Fig. 4(c)) contrast to lipids distributed in the stratum basale of normal skin tissues (Fig. 4(a)), the macroscopic molecular distributions of diffusion-driven lipids imply the dynamic metabolism in tumor aggressiveness. The quantified lipids and proteins are demonstrated by the position and intensity of characteristic peaks in normalized Raman spectra of different pathological types of skin epidermis (Fig. 4(d)), and we observe the Raman shift of proteins at 2932 cm^-1 in normal epidermis (indicated by magenta arrows) to the 2924 cm^-1 in pathological epidermis (indicated by red arrows), and there is an evident distinction in the intensity of Raman peaks of lipids at 2885 cm^-1. The differences of position and intensity of Raman peaks at lipids and proteins are recognized as being associated with keratin and lipid mixtures in the epidermis, and have the potential as the special biomarker to indicate cancer-related tissues changes for cancer screening and diagnosis. TP-FLIM technology empowered with the phasor approach is an essential approach to segment the skin layer [40] and intuitively reveal the fluorophore fraction of free and protein-bound FADs [41,42]. There is an increasing trend of the average lifetime from the stratum corneum and granulosum of cancer-free tissues (Fig. 4(e)) to that of the pigmented nevi (Fig. 4(f)) and BCC (Fig. 4(g)), which can directly observe the color changes of TP-FLIM pseudocolor images. Abundant functional signatures associated with the double-component FADs, especially the decreasing component ratio of pigmented nevi and BCC, are identified with the phasor plots in different skin pathologies, as shown in Fig. 2(h), 1 - Fig. 2(j), 1. A high density of granulosum featured short lifetime are found in the stratum basale of pigmented nevi and BCC, which validate that TP-FLIM as a power tools for indicating the pathological changes related to melanocytes [31].

We investigated lesions in seven patients (three males, four females), including 1 normal skin, 3 pigmented nevi and 3 BCC, and statistically analyzed the molecular metabolic differences in epidermis of normal and cancerous skin tissues with the Mann–Whitney test. Figure 4(h) reveals that the intensities of Raman peaks at 2885 cm^-1 and 2930 cm^-1 vary greatly in the epidermis of normal skin, nevus and BCC, this reaction are initiated by different metabolic performances of the corresponding lipids (CH2) and proteins. We further extracted results of the content ratio of lipids and proteins (right axis in Fig. 3(d)) to quantify the difference in proteins and lipids, in view of absolute strength affected by sample preparation processes and subjectively selected imaging areas. A large keratinization areas, malignant cells invasion in pigmented nevus [43] and a large amount of mucin in BCC [44,45] cause the increased protein content in carcinoma than normal tissues. Moreover, the statistical data demonstrated that the quantified lipids intensities are highly related to the histologic skin subtype, and the content ratio of lipids and proteins represents a great difference between the two types of skin cancer (p = 0.0003). Figure 4(i) provides the statistical information of the average lifetime, component lifetime, and component ratio of double-component FADs in the stratum corneum and granulosum. The average lifetime shifts from 1.05 ns for normal skin to approximately 2.2 ns for the two pathological types (nevus: 2.24 ns; BCC: 2.19 ns), which mainly initiated by their component ratios with significant differences (p = 0.0002 between normal and pathological tissues). Furthermore, the two pathological types of skin tissue with similar average lifetimes can be distinguished via the dissimilar lifetime of bound-FAD (p = 0.0006) caused by different pathological mechanisms. The molecular metabolism of lipids, proteins, and FAD acquired by SRS and TP-FLIM modalities supplement the macroscopic morphological features, and eventually fill the cancer-associated events in tumor microenvironment. A representative pentagonal radar plot (Fig. 4(j)) based on the multiple nonlinear optical imaging combined with quantitative method gathers the comprehensive and complementary pathological signatures, and offers an intuitive method for interpreting specific pathological changes of different skin pathological tissues, including thickening of pathological epidermal tissue from TPEF and SHG intensity curves (TP/SHG), increasing of lipid and protein content in epidermis from SRS spectra, and lengthening of fluorescence lifetime from TP-FLIM images.

3.4. Multidimensional quantification of ECM related to cancer invasion in dermis

The morphology and function characteristics of clinicopathological tissues vary with the histological structure of skin, thus, illustrating the metabolic characterization of different skin structure is essential for orienting the clinicians to cope with pathological variations in any skin sites. The morphology changes of dermis composed with ECM in pigmented nevus and BCC have been verified by TPEF and SHG images, here, we further discuss the molecular metabolism caused by lesions in the dermis to establish the heterogeneity of different skin sits for random needle-biopsy screening. Due to the degradation of collagen fibers in ECM caused by secretions during carcinogenesis, dermal tissues perform significant differences in molecular metabolism. Specifically, SRS images of the dermis layer in pigmented nevi and BCC visualize the distribution of lipids and proteins in cancerous regions. The lipids at 2885 cm^-1 corresponding to unsaturated fatty acids (CH3) in the nevus (Fig. 5(a)) are mainly distributed in the location of melanocytes, which can be distinguished easily from the collagen-rich dermal stroma. There are plentiful saturated fats with the Raman characteristic peak at 2850 cm^-1 in the cancer nest of BCC (Fig. 5(b)), which can be considered as a specific marker for distinguishing the cancer nest from the surrounding ECM. In addition, we found a clear dissimilarity in the Raman spectra (Fig. 5(c)) between two pathological skin cases, including the position shift of proteins in pigmented nevi (indicated by red arrows) and BCC (indicated by magenta arrows), as well as the intensity of lipids. The absolute intensity histogram and the content ratio of lipid to proteins (Fig. 5(d)) are applied for statistical quantification of the content variations of lipids and proteins with the Mann–Whitney test. There were remarkable significant differences among normal skin, pigmented nevus and BCC, which is initiated by the pigmentation and hyperkeratosis of keratin in pigmented nevi, the excessive mucus protein around the cancer nest, and decomposition of collagen in BCC [46] during cancer invasion.

Fig. 5.

Determination of molecular metabolism in dermis. Also see Supplementary Fig. 6. (a, b) Two-color SRS images of lipids (cyan) and proteins (magenta) in the dermis of pigmented nevi and BCCs. (b1) Overall normalized SRS spectra for the ROIs of cancer nests (2) and extracellular matrix (1) in BCCs. (c) Overall normalized SRS spectra for the entire FOV of images a and b. (d) Statistical intensities of lipids and proteins and their intensity ratio, Rlipid/protein, for pathological tissues (n = 10). (e) Pentagonal radar multinonlinear profile summarizes the molecular metabolic properties of skin dermis. TP-FLIM pseudocolor images of FAD in the dermis of pigmented nevus (f) and BCC (g) with corresponding phasor plots (f1, g1) based on multicomponent analysis method. Statistical average fluorescence lifetime, component lifetime and ratio (R_bound) of extracellular matrix (h) and cancerous regions (i) in the dermis (n = 8). ****: p < 0.0001, ***: p < 0.001, **: p < 0. 01, *: p < 0.05. Scale bars, 30 µm.

In the false-color lifetime maps of pigmented nevi (Fig. 5(f)) and BCC (Fig. 5(g)), the regions of the anomalous events and the surrounding ECM in pathological tissues are segmented by the phasor method to provide significant visual contrast, as shown in Fig. 5(f), 1 and Fig. 5(g), 1. These functional actions about the fluorescence lifetimes of FADs are recognized as being related to the changes of microenvironments in the neoplastic tissue and the disturbance by nests of tumor cells invading the dermis [47,48], which echo the findings of the metabolic function associated with lipids and proteins. In addition, Fig. 5(h) provides the statistical information of average lifetimes, component lifetimes, and component ratios in ECM close to tumor areas. The average lifetime of normal tissue was significantly different from that of BCC (p < 0.0001), and its component ratio also significantly differed from that of pigmented nevus (p < 0.0001). The results indicate that the near invasive regions in the cancer tissues can also be distinguished, which provides a possibility to detect early cancers based on metabolic features. Furthermore, pigmented nevus and BCC (Fig. 5(i)) also exhibited significant differences (p < 0.01) in the average lifetime, component lifetime, and component ratio in the cancerous regions, which demonstrates that the metabolic features revealed by the phasor-empowered FLIM can be recognized as an effective marker for cancer diagnosis [37]. The concurrence of macroscopic distributions and molecular biochemical features of lipids, proteins, and FADs extracted from SRS and TP-FLIM images direct reveal the discrepant results between the different pathogenic skin tissues by pentagonal radar profile (Fig. 5(e)). These findings suggest the feasibility for investigating the complexity of the tumor microenvironment and molecular metabolism by implementing the multidimensional nonlinear optical characterization method, which have the future potential for fundamental comprehension of carcinogenesis in clinicopathological analysis.

4. Discussion and conclusion

The tumor microenvironment features of skin conditions, such as microstructure, cancer-associated collagen fibers, endogenous fluorescence, and molecular metabolism, are important for tumor margin evaluation, tumor subtype discrimination, and cancer diagnosis. We introduced the multiplexed nonlinear optical approach for the acquisition and interpretation of tissues signatures, as well as several powerful analysis methods for information-rich datasets to provide unbiased, quantitative, and comprehensive analysis of high-throughput measurements. These multidimensional optical imaging of TPEF, TP-FLIM, SHG, and SRS offer abundant complementary information and molecular contrast from unperturbed skin tissues. We demonstrated the applicability of TPEF and SHG to separate the microstructures and morphological features of skin. Phasor-FLIM analysis, on the other hand, has been demonstrated to be capable of high temporal resolution and fit-free, high-precision quantification of FAD associated-metabolism differences in normal and pathological skin tissues. We further demonstrated the capability of this approach for unbiased SRS analysis with high spectral resolution and background-free, high-specificity determination of protein and lipid metabolism in skin pathogenetic processes. The comprehensive and complementary tumor-associated pathological information about morphological and molecular metabolism obtained from four-modality nonlinear imaging combined with qualitative and quantitative analysis, instead of single mode characteristics, is the key to accurately reveal the pathological characteristics of different skin pathological types. The information obtained from these experiments allowed for simultaneous multi-modality nonlinear imaging with subcellular visualization and objective quantification of the tumor microenvironment with diverse biophysical signatures, which provided rich biophysical properties in clinicopathologic diagnosis.

Nevertheless, the frame rate and optical resolution of the system are still limited by the galvanometer mirrors and the objective. TPEF and SHG data can be obtained in real time with large photon fluxes, while FLIM and SRS require single-photon counting and spectral scanning, respectively, and are still slow in comparison. These can be overcome using frequency-domain FLIM and fast spectral scanning using a polygon scanner [49]. Moreover, although extension of SRS measurement from CH regions to fingerprint regions and NADH FLIM can be realized by tuning the pump wavelength and replacing the collect filters, the imaging quality will suffer from weak signal intensity and high noise level and background. These challenges, as well as the strong scattering problem in intravital imaging, can be alleviated by the introduction of AI algorithms [50,51], which have been demonstrated to greatly improve the Raman image quality in the fingerprint region [49]. The multimodal system can be further upgraded, and the analysis method can be substituted by AI models for further foreseeable improvements.

To conclude, in contrast to previous multiplexed imaging techniques that rely primarily on colorful fluorescent dyes and probes to distinguish different components [52–55], we provide the skin conditions pathological examination by the system featured label-free noninvasive detection. In the era of biomedicine, we believe the inaccessible, profound pathological characteristics can be acquired by the multidimensional optical imaging system with high precision, multiple contrast mechanism, and high information content to assist in multiple pathological types of cancer diagnosis.

A minor correction was made to the text.

Funding

National Key Research and Development Program of China10.13039/501100012166 (2021YFF0502900, 2017YFA0700402); National Natural Science Foundation of China10.13039/501100001809 (62225505, 61935012, 62175163); Shenzhen Talent Innovation Project (RCJC20210706091949022); Shenzhen Key Projects (JCYJ20200109105404067); Shenzhen International Cooperation Project (GJHZ20190822095420249).

Disclosures

The authors declare no conflicts of interest.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Supplemental document

See Supplement 1 ^{(17.1MB, pdf)} for supporting content.