Label-free optical biomarker for prostate cancer diagnosis

Abstract

Background

Reprogramming of cellular metabolism in cancer leads to changes in the activity of metabolic coenzymes NAD(P)H and FAD which can be used as markers of tumorigenesis and metastasis. NAD(P)H and FAD form coupled redox pair and regulator of mitochondrial energy metabolism. Since NAD(P)H and FAD are endogenous and auto fluorescent, the coupled redox pair can hence be exploited as a reporter for metabolic activity in tumor microenvironment.

Methods

We have developed a 2P-FLIRR assay (2-photon Fluorescence Lifetime based Redox Ratio), a sensitive metabolic optical biomarker for tumor diagnosis in prostate cancer (PCa) patients’ tissue sections. We have imaged and used our novel 2P-FLIRR metric which is the ratio of the enzyme-bound fractions of NAD(P)H and FAD to evaluate PCa patients’ surgically resected tissue sections. These were PCa adenocarcinomas with different Gleason grades.

Results

In comparison with conventional H&E histopathology, our 2P-FLIRR assay showed the potential to distinguish tumor-enriched regions from normal prostate tissue in unstained frozen patient tissue sections. Lower FLIRR values were observed to correspond to tumor versus higher FLIRR values to normal prostate.

Conclusion

This pilot feasibility study demonstrates that 2P-FLIRR detects metabolic contrast between prostate cancer and normal prostate tissue in frozen human patient tissue sections. Our study identifies for the first time quantitative 2P-FLIRR as a potential sensitive diagnostic optical biomarker for PCa.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s10020-026-01605-2.

Background

Prostate cancer (PCa) is a leading cause of cancer and the second leading cause of death (36,320- cancer.org) in the USA. It is projected that 1 in 8 men will be diagnosed with prostate cancer in their lifetime. Worldwide it is projected to double to 2.9 million a year by 2040, with annual deaths predicted to rise by 85% (James et al. 2024). Urologists are challenged with determining which patients with clinically organ-confined prostate cancer require curative treatment (such as surgery or radiotherapy), and which can be appropriately managed with active surveillance. Among the primary drivers informing this decision is pathologic Gleason grading which is subject to high inter-observer variability (Humphrey 2017), leading to divergent management recommendations i.e. active surveillance vs definitive local therapy, with resultant impact on patients’ quality of life and healthcare costs.

Metabolic rewiring of cancer cells is linked with uncontrolled cell proliferation, tumor progression, and metastasis, affecting the cellular redox states of the tumor microenvironment (TME). Thus, changes in the redox state of the tumor microenvironment can be used as an indicator of tumorigenesis and tumor’s aggressiveness. Two-photon (2P)- Fluorescence Lifetime Imaging Microscopy (FLIM) of endogenous, auto-fluorescent metabolic coenzymes: Nicotinamide Adenine Dinucleotide (Phosphate) Hydrogen (NAD(P)H) and Flavin Adenine Dinucleotide (FAD) is a powerful, non-invasive, label-free approach to track metabolic alterations in cancer (Heikal 2010; Provenzano et al. 2009; Lee et al. 2018; Skala et al. 2007; Walsh et al. 2016). NAD(P)H, the reduced forms of NADH and NADPH and oxidized forms of FAD are auto fluorescent, their signal predominantly originating from mitochondrial matrix (Huang and Heikal 2002). Suitably normalized fluorescence of NAD(P)H and FAD approximates the oxidation and reduction equilibrium (redox state), providing a resolution of redox states in cells and tissues (Chance et al. 1979). 2P-FLIRR is a powerful approach to assess the metabolic state under different pathophysiological conditions (Alam et al. 2017; Wallrabe et al. 2018; Zhang et al. 2024) Since, in tumorigenesis the oxidation–reduction equilibrium is altered it allows us to exploit the NAD(P)H/FAD redox pair as an optical biomarker for tumor diagnosis and stratification. These co-enzymes exist in “free” or “enzyme-bound” states during cellular metabolic activity. Analysis of fluorescence lifetime decay curves by two-component model is used to assess the relative fractions of free and enzyme-bound states (Lakowicz et al. 1992; 2010; Cao et al. 2020). The shorter (0.4 ns) and longer (2.4 ns) lifetimes of NAD(P)H represent the “free (a₁%)” and “enzyme-bound (a₂%)” components, respectively. On the other hand, the shorter (0.12 ns) and longer (3.38 ns) lifetimes of FAD represent the “enzyme-bound (a₁%)” and “free (a₂%)” components, respectively (see Methods for details).

Intensity-based imaging was introduced almost 40 years ago (Chance et al. 1979), however it is prone to optical artifacts, including light-scattering and absorption barriers in biological specimens (Alam et al. 2017; Wallrabe et al. 2018). To overcome these limitations, we developed a Fluorescence Lifetime based Redox Ratio (FLIRR) metric. FLIRR is independent of absolute intensity measures; instead, it is calculated as the ratio of the enzyme-bound fractions reflected by the relative amplitudes of the lifetime decay curve for each molecule’s enzyme-bound fraction specifically NAD(P)H-a2%/FAD-a1%. We established that FLIRR serves as an indicator of mitochondrial OXPHOS (Oxidative Phosphorylation) activity. In cancer, since the interplay of glycolysis and mitochondrial OXPHOS is altered, FLIRR also serves as an indicator of mitochondrial dysfunction (Alam et al. 2017).

In this study we have reported the development of an innovative quantitative FLIRR approach, as an optical biomarker assay for prostate cancer diagnosis in patient tissue sections. As molecular changes precede morphological changes on which H&E (Hematoxylin & Eosin) staining and Gleason scoring are based, we have demonstrated that quantitative FLIRR approach can be used for PCa detection and potential risk stratification.

Methods

PCa patients and demographics

All patients unfixed tissue sections after surgical resection were obtained from University of Virginia (UVA), Biorepository and Tissue Research facility (BTRF) under the approved IRB-HSR #13,281. In this study, 12 core punch tissues with 6 normal and 6 tumors (n = 12) were included. All 6 tissue sections from PCa patients were confirmed to have adenocarcinoma. These patients were between 63–79 years old. Tumor tissue sections were stained by conventional staining methods using H&E (Fig. S1). Tumors were confirmed and annotated by the UVA pathologists. Demographics, histopathological data and stage of the prostate cancer according to the AJCC (American Joint Committee on Cancer) TNM (Tumor, lymph Nodes and Metastasis) system with PSA (Prostate Specific Antigen) and Gleason score are provided in Table 1. The patients were de-identified.

Table 1.

Demographics of patients

PCa patient code	Race	Age	Type	Gleason Grade	T	N	Type
MAD08-00550 X1-1	White	70	Prostate	3 + 4 = 7	2C	X	Adenocarcinoma
MAD12-01132 X1-6	White	70	Prostate	3 + 3 = 6	2A	X	Adenocarcinoma
MAD20-0512.X1-1	White	79	Prostate	4 + 3 = 7	3A	1	Adenocarcinoma
MAD20-0707.X1-4	White	65	Prostate	3 + 4 = 7	2	0	Adenocarcinoma
MAD20-00037 X1-4	White	70	Prostate	3 + 4 = 7	2	0	Adenocarcinoma
MAD20-00149 X1-1	White	63	Prostate	3 + 3 = 6	2	0	Adenocarcinoma

The surgically resected unfixed tissue blocks were preserved as frozen by embedding in Optimal Cutting Temperature (OCT) Medium for cryo-sectioning. Two- photon FLIM for NAD(P)H and FAD were performed on 10 μm thick tissue sections after cryo-sectioning.

Two-photon fluorescence lifetime redox ratio (2p-FLIRR) microscopy: lifetime imaging, lifetime fitting, analysis and FLIRR computation

Two-photon fluorescence lifetime imaging was performed on Zeiss LSM 780 confocal/multiphoton (MP) laser scanning system coupled to the Zeiss inverted epi-fluorescence microscope controlled with the ZEN software (Carl Zeiss, Inc). Two-photon excitation of NAD(P)H and FAD was achieved with 800 nm wavelength (Cao et al. 2020), using an ultrafast (150 fs) tunable Ti:sapphire laser (680–1060 nm), operating at 80 MHz repetition rate (Chameleon Vision II, Coherent, Inc.). The HPM-100 hybrid GaAsP detectors (Becker and Hickl) connected to the non-descanned (NDD) port of the microscope using two T-adapters (Zeiss) with proper dichroic and band pass filters collect fluorescence in the spectral ranges of 425–475 nm (NAD(P)H) and 520–560 nm (FAD). NADH and FAD channels also contain 690 nm short pass filter (Zeiss) in the beam path. The SPC-150 cards (Becker and Hickl) synchronized with the pulsed laser and the Zeiss LSM-780 scan head collects time-resolved fluorescence in TCSPC (Time Correlated Single Photon Counting) mode using SPCM (9.74) acquisition software. Details on the methodology of our FLIM set up can be found in the literature (Alam et al. 2017; Wallrabe et al. 2018; Cao et al. 2020; Cao et al. 2019).

After simultaneous acquisition of lifetime images for NAD(P)H and FAD, were fitted for 2-components using SPCImage software (v. 8.6, Becker & Hickl). Number of parameters was generated including photon images, Lifetime values: τ₁, τ₂, Amplitude weighted fractions: a₁%, a₂%, and Goodness of Fit or χ² for each pixel of each channel. Tissue sections were analyzed at 1 × 1 pixels/ROI (Regions of Interest) using the NAD(P)H photon image. The ROIs generated were used for further lifetime data analysis using our custom developed python codes for ‘flimanalyzer’ (Alam et al. 2017; Wallrabe et al. 2018; Cao et al. 2020; Cao et al. 2019).

Lifetime measurements were performed on the tissue sections with 10X N.A. 0.45 objective, with 2P-800 nm laser at 2.8% excitation or ~ 16 mW average laser power at the specimen plane, laser scan speed of 4 and tiling format of 12 × 12 at 128 × 128 pixel resolution. Lifetime imaging was performed on the normal and PCa tissue sections for NAD(P)H and FAD the redox pairs. Fluorescence lifetime decay curves of NAD(P)H and FAD were fitted and analyzed as previously described by us (Alam et al. 2017; Wallrabe et al. 2018; Cao et al. 2020; Cao et al. 2019). Color coded images of the FLIRR map were computed on pixel-by-pixel basis for the acquired lifetime images from enzyme bound fractions of NAD(P)H-a2% and FAD-a1%.

For deeper insight into the redox state specific to tumor vs normal regions, fluorescence lifetime imaging for NAD(P)H and FAD was performed on different areas/field of views (FOVs) using Zeiss 40X 1.3NA oil, (EC Plan-Neuofluar) immersion objective lens for better spatial resolution. The laser average power at the specimen plane (7 mW) and the acquisition time of 60 s was chosen to reduce any photodamage to the cells (Alam et al. 2022). The exported results from multiple tissue sections and FOVs were further analyzed as described in the literature (Alam et al. 2017; Wallrabe et al. 2018; Cao et al. 2019, 2020). FLIRR is the ratio of NAD(P)H-a₂%/FAD-a₁%.

Mitochondrial staining

Mitochondrial staining with MitoTracker CMXROS probe (Invitrogen) as a structural mitochondrial marker used in this study was performed at a concentration of 250 nM for 30 min followed by Hoechst nuclear probe (Invitrogen) for 10 min in DMEM FluoroBrite (Invitrogen). After staining, the slides were fixed with 3.7% paraformaldehyde and mounted in Fluoromount G (Southern BioTech). Confocal tiling of MitoTracker CMXROS for the normal and PCa tissue sections were performed on Zeiss 780 confocal microscope with Ex 561 nm at 0.8%, Em 566–671 nm, PMT 700v; for Hoechst 0.2% with 405 nm laser and Em 410–476 nm, PMT 650v with 10 × objective lens, 0.75NA.

FLIRR and gaussian mixture model

Gaussian Mixture Model (GMM) (scikit-learn user guide 2019) clustering suitable for homogenous data distribution as observed in normal tissue sections (Figs. 1A, 2A & 4A), was integrated into our python based flimanalyzer software (Zhang et al. 2024). The GMM approach was applied to the FLIRR data sets arising from all the normal tissue sections (n = 6). The GMM grouped the data into 2 distributions peaks (GMM full 1& 2).

Fig. 1.

PCa: Comparison of FLIRR map with H&E staining: Histology of the tissue sections with H&E staining in comparison with the FLIRR map is shown for normal prostate (A) and PCa (B) and (C) suspected tumor which was earlier reported as “normal” by H&E. Two-photon fluorescence lifetime measurements were performed at optimal average laser power (Alam et al. 2022) on the tissue sections. Lifetime decay curves of NAD(P)H and FAD are fitted and analyzed as previously described by us (Alam et al. 2017; Wallrabe et al. 2018; Cao et al. 2019, 2020). Color coded image of the FLIRR map is computed on pixel-by-pixel basis from the lifetime images of enzyme bound fractions of NAD(P)H-a2% and FAD-a1%. Lifetime fitted, computed and color coded FLIRR map shows higher FLIRR distribution for the normal prostate (A). Since tumor tissue section has a mix of normal and PCa regions two FLIRR histogram distribution peaks are seen (right side of the figure)- lower FLIRR distribution corresponding to PCa and higher to the normal areas (B). H&E staining for the suspected tumor (Fig. 1C) did not show the presence of cancerous lesions but FLIRR showed areas with lower FLIRR indicative of tumor lesion (squares with yellow color in Fig. 1C). Thus, highlighting the sensitivity of FLIRR over H&E and as a diagnostic optical biomarker. FLIM was performed with 10 × N.A. 0.45 objective, 800 nm at 2.8% pulsed laser excitation, laser scan speed 4; 12 × 12 tiling. FLIRR image: Scale bar 2 mm

Fig. 2.

Quantitative FLIRR for PCa diagnosis. Representative Fields of View (FOVs) for 2 patients’ normal prostate (A) and PCa (B) areas of tissue sections by FLRR. To match the representative FLIRR FOVs, lower magnification H&E images from Fig. 1 were further zoomed and rotated to identify similar histology patterns as can be seen by FLIRR and H&E staining. Color coded image of the FLIRR map showed distribution of higher FLIRR values for normal prostate (A) and lower FLIRR for PCa tissue sections (B). C Quantification of FLIRR data into bar graphs with ± SEM for both the patients respectively and combined showed higher FLIRR values in normal vs PCa. Lifetime measurements were performed with 40 × N.A. 1.3 objective, 800 nm at 2% pulsed laser excitation. “The results were statistically significant by Student t-Test assuming unequal variance at alpha value of 0.05 at the patient level for the different FOVs of the normal and PCa with p < 0.001 and p = 0.021 respectively, and between normal and PCa patients with p = 0.025. FLIRR was more sensitive in discriminating differences in the PCa tissue sections at the patient level diagnosed with the same Gleason grade 7 with p value = 0.045 at 95% confidence intervals and Effect Sizes η² = 0.83 having a large effect and thus identifying FLIRR as a highly impactful diagnostic factor in this study. Scale bar FLIRR images 50 µm.”

Fig. 4.

Comparison of tumor and normal with FLIRR. PCa patient tissue sections in Fig. 3 were further used to capture higher magnification NAD(P)H and FAD images using 40X N.A. 1.3 objective with 800 nm at 2% laser excitation. FLIRR map was computed for all the FOVs as described before. Representative color coded FLIRR images with differences in FLIRR distribution histograms as compared to the normal can be seen (A, C, E). The unsupervised Gaussian Mixture Model (GMM) applicable to dataset with homogenous distribution was applied to FLIRR data points arising from normal control tissue sections. B The GMM clustered the normal FLIRR datapoints into 2 clusters: GMM full 1 and 2. This approach allowed us to determine a threshold for FLIRR score (≥ 0.4) for B normal vs (≤ 0.4) D PCa. Furthermore, when we applied the GMM to a suspected tumor (MAD20-00512 X1-3, Fig. 1C), diagnosed by H&E as “normal” showed predominance in distribution of the lower FLIRR values as compared to all the normal tissue sections (Fig. 4F), underscoring the potential of FLIRR over H&E in detecting a missed cancerous lesion. However, further pathological correlation is required

FLIRR and K-means clustering

Several hundred thousand of single pixel data points are generated for each tissue section by the lifetime data analyses for every lifetime parameter τ₁, τ₂, a₁%, a₂%, and χ² for NAD(P)H and FAD from which FLIRR values are computed on a pixel-by-pixel basis. The ROIs generated were then used for further lifetime data analysis using our custom developed python codes for flimanalyzer (Zhang et al. 2024).

Large number of datapoints arising from the tissue sections were then processed using K-means clustering, an unsupervised clustering algorithm used for grouping datapoints into similar clusters by iteratively assigning data points to clusters based on their proximity to cluster centroids. The number of clusters is denoted by "k" (Liu et al. 2023; Timmerman et al. 2013). K-means clustering for FLIRR mean was performed on the tissue sections using our custom flimanalyzer software (Zhang et al. 2024).

Statistical analysis

Results represented in Bar graphs are expressed as mean ± SEM (Standard Error Mean). Student t-Test assuming unequal variance and single factor ANOVA analysis at alpha value of 0.05 were used for statistical significance between groups showing changes in mean FLIRR at the Field of View (FOV) and Patient levels (Figs. 2 & 5). Effect Sizes (η²) was further calculated from ANOVA statistical analysis to assess the impact of FLIRR as a diagnostic factor (Cohen 2013; Loth et al. 2021).

Fig. 5.

Heterogeneity in redox states and FLIRR in PCa patient tissue sections. For the quantitative FLIRR imaging and analysis we have used tissue sections from 6 PCa patients and compared them with 6 normal areas of the prostate after surgical resection. Representative color coded FLIRR map showed lower FLIRR distributions in PCa as compared to the normal prostate tissue sections (Figs. 1–2). Upon analysis all FLIRR datasets such as KDE (Fig. 5A) and bar plots (Fig. 5B) showed differences in FLIRR distribution in different PCa tissue sections at the patient level and compared to the combined normal FLIRR dataset. Heterogeneity in FLIRR distribution was observed. The results were statistically significant by ANOVA single factor analysis (p < 0.001) at 95% confidence intervals between the patients with Effect Size η² = 0.64 having a large effect, thus identifying FLIRR as a highly impactful diagnostic factor in this pilot study. Since the PCa tissue will have a mix of normal tissue along with tumor lesions, K-means clustering was performed on all datasets from patient tissue sections which yielded 2 FLIRR mean value clusters- segregating normal (higher FLIRR scores) and tumor lesions (lower FLIRR scores) respectively (Fig. 5C). Furthermore, smaller sub-clusters were also observed (red circle in Fig. 5B-C) based on disease severity either by tumor grading, infiltration to lymph nodes or higher PSA, showing similar FLIRR distributions in these PCa tissue sections at the patient level. A non-quantitative future framework for a hypothetical FLIRR grading table based on FLIRR distribution and unsupervised K-mean clustering with different Gleason grades is proposed (Fig. 5D-Table) which could enable easy interpretation for PCa status

Data availability

All data generated or analyzed in this study are included in this manuscript and in supplementary information.

Results

FLIRR mosaic: fluorescence lifetime imaging for PCa tissue

The current standard for histopathological diagnosis is by means of staining with exogenous dyes- Hematoxylin and Eosin (H&E). Gleason grading is based on H&E staining on biopsy tissue sections to identify morphological changes in the primary and secondary cellular arrangements patterns (Humphrey 2017). H&E staining is a time-consuming process that requires denaturation of the tissue using Formalin Fixation and Paraffin Embedding (FFPE). In addition, downstream processing in xylene, ethanol followed by staining prone to cross contamination and false positivity as the tissue sections are batch processed (Fig. S1). Unlike this conventional method, our FLIRR methodology is based on the fluorescence lifetime of auto-fluorescent (unstained) endogenous fluorophores, NAD(P)H and FAD, the metabolic coenzymes. Our FLIRR approach is applicable to live or frozen tissue sections. Freezing surgically resected tissues is a standard method of preservation and requires no chemical denaturation or any exogenous labeling of the tissue (Fig. S1). Fluorescence lifetime imaging of NAD(P)H and FAD can be achieved in their natural tumor microenvironment (TME) and can provide valuable molecular information on the redox state of the TME and normal tissues. The FLIRR images when compared to H&E image provide a very clear vivid histological overview of the normal and tumor (Fig. 1). Color coded images of the FLIRR map, computed on a pixel-by-pixel basis for the enzyme bound fractions of NAD(P)H-a2% and FAD-a1%, show a higher FLIRR distribution for normal prostate tissue sections (Fig. 1A). Since tumor tissue sections have a mix of normal and cancerous lesions, two FLIRR distribution peaks are seen: lower FLIRR distribution corresponding to tumor and higher to normal areas (Fig. 1B), thus serving as a potential diagnostic optical biomarker for PCa. Importantly, FLIRR revealed metabolically distinct regions (yellow color in Fig. 1C) in a tissue specimen interpreted as “normal” by H&E; further pathological correlation is required to establish the sensitivity of FLIRR in detecting a missed lesion.

Quantitative FLIRR for PCa diagnosis

To improve spatial resolution, we examined PCa tissue sections at higher magnification (40X oil immersion N.A. 1.3)—compared to Fig. 1—to assess and quantify ranges of FLIRR distribution more precisely for the normal and tumor regions. Representative fields of view (FOVs) for 2 patients’ tissue sections show similar histology patterns by FLIRR as seen by H&E staining. Pseudo-colored images of the FLIRR map showed distribution with higher FLIRR values for normal prostate and lower FLIRR for PCa surgically resected tissue sections (Fig. 2A-B). Quantification of FLIRR data grouped by patient or combined showed higher FLIRR (0.4–0.6) in normal vs lower FLIRR (0.1–0.3) in PCa tissue sections (Fig. 2C). ANOVA single factor analysis was performed to assess the discriminatory power of FLIRR for PCa detection between the 4 tissue sections (2 normal, 2 PCa) and the results were statistically significant (p < 0.001) between the groups. The 2 PCa tissue sections by H&E staining and Gleason grading were both categorized with Gleason score 7. However, FLIRR was more sensitive and able to discriminate differences in 2 tissue sections assigned with the same Gleason grade (Fig. 2C). The differences in FLIRR in the PCa tissue sections were statistically significant (p value = 0.045). Effect Sizes (η2) (Cohen 2013; Loth et al. 2021) was calculated from ANOVA single factor statistical analysis with η2 = 0.83 having a large effect; identifying FLIRR as a biomarker heavily driving the difference in patient tissue sections in this study and therefore making FLIRR a strong candidate for PCa diagnosis.

Potential risk-stratification relevance of FLIRR: comparison with mitochondrial staining

Higher mitochondrial content has been associated with prostate cancer disease progression (Grupp et al. 2013) where the prognostic value of various mitochondrial markers has been evaluated in advanced PCa (Sachdeva et al. 2022; Schöpf et al. 2020). Therefore, to assess the potential risk-stratification relevance of FLIRR, we first compared FLIRR images with mitochondrial staining of the tissue sections (Fig. 3, 10X, 0.75NA), as a confirmatory method since FLIRR signal predominantly comes from the mitochondria. All 3 methods: H&E staining, FLIRR and mitochondrial staining with mitochondrial structural marker MitoTracker CMXROS and Hoechst as conventional marker for nuclei, showed similar histological patterns in the surgically resected tissue sections (Fig. 3A-C). Next, we compared and analyzed normal and PCa tissue sections with different Gleason grades for mitochondrial staining by confocal laser scanning microscopy. Clear qualitative differences were observed between the normal and PCa (Fig. 3D-F). Furthermore, we evaluated different FOVs (n = 4) for those same tissue sections for normal and tumor region at higher magnification to monitor the spatial distribution (40X N.A. 1.3). We observed clear differences in histology in color-coded FLIRR images and their corresponding FLIRR histograms. Higher FLIRR values corresponded to normal and lower FLIRR to the tumor regions (Fig. 4A-C). Single pixel-by-pixel analysis of FLIRR from 5-6 FOVs yielded on average 100,000-160,000 non-overlapping data points from single PCa tissue section. To analyze the large volume of the data we used Gaussian Mixture Model (GMM) (scikit-learn user guide 2019) clustering suitable for homogenous data distribution like we saw in normal tissue sections (Figs. 1A, 2A & 4A). The GMM approach was applied to the FLIRR data sets arising from all the normal tissue sections (n = 6). The GMM grouped the data into 2 distributions peaks (GMM full 1& 2), allowing us to determine a FLIRR threshold for normal (≥ 0.4, Fig. 4B) and for PCa (≤ 0.4, Fig. 4D).

Fig. 3.

Comparison of FLIRR with H&E and mitochondrial staining. Normal prostate by H&E (A), FLIRR map (B) and confocal tiling of mitochondrial staining with mitochondrial structural marker MitoTracker CMXROS (C) shows similarity in tissue histology. However, H&E histology showed some dehydration due to Formalin-Fixation Paraffin Embedding (FFPE) with alcohol and xylene washing; unlike the frozen unfixed tissue sections used for FLIRR (without staining) and mitochondrial imaging. Similar, to FLIRR (B), MitoTracker CMXROS staining identified the histological differences in normal (C, D) and PCa tissue sections (E–F) from different patients with different blood PSA and Gleason grades. Scale bar-200 µm

Interestingly, the tissue sections for a suspected tumor (MAD20-00512 X1-3, Fig. 1C), diagnosed as “normal” by H&E showed a distribution shift from higher to lower FLIRR values when compared to all the normal tissue sections by GMM clustering (Fig. 4F), underscoring the potential sensitivity of FLIRR in detecting a missed cancerous lesion. However, further clinical investigation is required.

Heterogeneity in redox states and FLIRR in PCa patient tissue sections

We further analyzed different tissue sections of independent patients (n = 6) with different Gleason grades and blood PSA levels with their respective normal controls (n = 6), combined. The surgically resected tissue sections were subject to quantitative FLIRR imaging at higher magnification (40 × Oil, N.A. 1.3) to assess the heterogeneity in FLIRR distribution for PCa and normal prostate epithelia. Representative color-coded FLIRR maps showed distributions of lower for PCa vs higher FLIRR values in the normal prostate epithelia. Heterogeneity in FLIRR distribution was seen in the color-coded FLIRR images of different patient tissue sections and in histograms and FLIRR mean bar charts (Fig. 5A-B) for the entire datasets with p < 0.001 at 95% confidence intervals between the groups and Effect Size η² = 0.64, identifying FLIRR having a large effect as a diagnostic factor. K-means clustering on millions of FLIRR data points arising from different FOVs from different tissue sections with different Gleason grades was performed for data grouping and categorization, yielded 2 clusters. Tumor tissue sections have a mix of tumor regions and normal surrounding tissue and by K-means clustering we were able to separate the lower and higher FLIRR values corresponding to tumor vs normal regions, respectively. (Fig. 5C). As a non-quantitative future framework, a hypothetical FLIRR grading table for different Gleason grades is proposed (Fig. 5D-Table) which could enable easy interpretation for PCa status by the non-experts. Interestingly all G7 PCa tissue sections according to the AJCC and TNM system of grading were grouped together by K-mean clustering of FLIRR than the rest of the G6, less invasive tumors. Hence, underscoring the concordance of FLIRR with AJCC (American Joint Committee on Cancer) TNM (Tumor, lymph Nodes and Metastasis) system (Edge and Compton 2010), despite being two very different methodologies of PCa staging.

Discussion

Several imaging modalities have been previously applied to cancer detection in tissues. For example, hyperspectral imaging (HSI) which assesses difference in absorption peaks in tumor over a wide spectral range (Pertzborn et al. 2023) or chemical imaging approaches like the stimulated Raman scattering (SRS) microscopy that directly measures chemical composition of tissues arising from their Raman vibrational response (Kong et al. 2013; Becker et al. 2021). Similarly, multiphoton coherent anti-Stokes Raman spectroscopy (CARS) tomography for intratissue lipids/proteins for skin infection like dermatitis and psoriasis has been used (König et al. 2020). Also, the use of machine learning (ML) approaches in tumor diagnoses has been used in recent years because the results from hyperspectral or SRS imaging cannot be directly interpreted by pathologists and, thus, require the use of artificial intelligence (AI) techniques for interpretation (Pertzborn et al. 2023; Falahkheirkhah et al. 2023; Tsafas et al. 2022). Third Harmonic Generation (THG), another non-linear optical imaging technique has been applied to breast and ovarian cancer and relies on the inhomogeneities in tissue structures (Tsafas et al. 2022).

In this study, we have demonstrated the application of quantitative FLIRR as a sensitive optical biomarker for PCa diagnosis. Unlike the above-mentioned optical modalities which rely on chemical composition and structure inhomogeneities, FLIRR detects changes directly at the molecular level arising from metabolic reprogramming, a fundamental phenomenon in tumorigenesis. FLIRR could clearly detect and establish differences in its distribution based on the redox states of NAD(P)H and FAD enzyme bound fractions in normal and PCa tissue. We were able to corroborate our findings with H&E histopathology and mitochondrial staining with FLIRR. The sensitivity of FLIRR lies in the fact that it has a molecular correlation based on early on-set changes in metabolic reprogramming and tumorigenesis; Unlike the H&E staining which identifies late-onset morphological changes in primary and secondary cellular patterns. The advantage of FLIRR over H&E and mitochondrial staining or IHC also lies in the ease of usage of the approach without fixation with harmful chemicals and labeling with exogenous dyes. We have demonstrated that FLIRR has the advantage of identifying heterogeneity in tumor and subtle differences in redox state across tissue sections. K-means clustering method provides an unsupervised clustering approach (Liu et al. 2023; Timmerman et al. 2013) effectively processing millions of 1 × 1 pixel data points arising from patient tissue sections within few minutes. This approach allowed us to categorize the data points into K-means clusters of FLIRR, to identify differences in tumors and to establish FLIRR ranges that would assist us to classify and grade the tumors in patient tissue sections.

Thus, based on our results: FLIRR distribution and K-mean clustering of FLIRR dataset for the different patient tissue sections with different Gleason grades, we were able to define the FLIRR ranges for normal and malignant tumor area in the tissue sections. Overall, our results demonstrate FLIRR as a sensitive optical biomarker for PCa diagnosis with potential risk-stratification relevance. The innovative FLIRR approach would offer precision in cancer diagnosis not only for PCa biopsy tissues but for other cancers as well, where the interplay between glycolysis and OXPHOS is altered.

Conclusion

This study reports a quantitative label-free 2P-FLIRR assay to distinguish tumor versus normal in unfixed patients’ tissue specimens. Our study identifies for the first time, 2P-FLIRR as a sensitive diagnostic optical biomarker for PCa diagnosis in tissues and proposes a method of tumor grading for risk stratification based on FLIRR scores without inter-observer variability. Our 2P-FLIRR approach does not require any exogenous stains and tedious fixation protocols, cross contamination and inter-observer variability, inherent problems with H&E staining based histopathological diagnosis of the tumor. In conclusion, our study identifies 2P-FLIRR as a potential sensitive diagnostic optical biomarker for PCa diagnosis. Our innovative quantitative 2P-FLIRR assay can be extended to PCa patient biopsies, for the assessment of tumor margins and real-time diagnosis of the PCa tumor at the time of surgery.

Abbreviations

NAD(P)H

Reduced form of Nicotinamide Adenine Dinucleotide (Phosphate)

FAD

Flavin Adenine Dinucleotide

OXPHOS

Oxidative Phosphorylation

2P

Two-photon

FLIRR

Fluorescence Lifetime Redox Ratio

FLIM

Fluorescence Lifetime Imaging Microscopy

PCa

Prostate cancer

TME

Tumor Microenvironment

H&E

Hematoxylin & Eosin

UVA

University of Virginia

BTRF

Biorepository and Tissue Research facility

AJCC

American Joint Committee on Cancer

TNM

Tumor, lymph Nodes and Metastasis

PSA

Prostate Specific Antigen

OCT

Optimal Cutting Temperature

MP

Multiphoton

Fs

Femtosecond

Nm

Nanometer

MHz

Mega Hertz

NDD

Non-Descanned

TCSPC

Time Correlated Single Photon Counting

τ₁

Shorter Lifetime

τ₂

Longer Lifetime

a₁%

Amplitude weighted fractions associated with τ₁

a₂%

Amplitude weighted fractions associated with τ₂

χ²

Goodness of Fit

ROI

Regions of Interest

mW

Milli Watt

GMM

Gaussian Mixture Model

FOVs

Field of Views

FFPE

Formalin Fixation and Paraffin Embedding.

HSI

Hyperspectral Imaging

SRS

Stimulated Raman Scattering

CARS

Coherent anti-Stokes Raman Spectroscopy

ML

Machine Learning

AI

Artificial Intelligence

THG

Third Harmonic Generation

Authors’ contributions

Conceptualization: AP, SRA; Methodology: SRA, AP; Investigation: SRA; Visualization: SRA; Funding acquisition: AP; Project administration: AP, SRA; Supervision: AP; Writing – original draft: SRA; Software for data analysis-KS; Writing – review & editing: SRA, AP, KS & RD; final approval by AP.

Funding

National Institutes of Health to AP—OD016446 and OD025156.

Data availability

All data generated or analyzed in this study are included in this manuscript and in supplementary information.

Declarations

Ethics approval and consent to participate

Ethics approval and consent to participate was waived under the approved IRB-HSR (# 13281)

Human Ethics declaration: not applicable.

Competing interests

The authors declare no competing interests.