Engineering a multimodal optical system for SRH and TPEF
We modified an existing, FDA-registered clinical system for stimulated Raman histology (Fig. 1a-b) by adding a two-channel epi-fluorescence module to allow simultaneous detection of SRH and PpIX fluorescence from the same optical section in fresh human brain tumor specimens. Stimulated Raman scattering (SRS) is excited with a custom dual-wavelength picosecond (2-3ps) fibre laser source with a fixed wavelength pump beam at 790 nm and a Stokes beam tunable from 1010 nm to 1040 nm27. SRS is detected in transmission mode with a large area photo-diode (PD). As described previously25, we combine high-frequency lock-in detection at 10MHz with high-speed auto-balanced detection in order to achieve high detection sensitivity. We use a high-NA water-immersion objective lens (Olympus 25x 1.05NA) to achieve high-resolution in tissue specimens. For SRH, we image the sample sequentially at 2845cm-1 to detect primarily lipids and at 2940cm-1 to detect primary protein and DNA and apply a pseudo-color scheme that mimics traditional H&E staining.25
Two photon-fluorescence is generated by the same excitation beams used to achieve stimulated Raman scattering (Fig. 1a). We use a 700 nm long-pass dichroic (Edmund Optics 69-903) to direct the epi-fluorescence signal toward the dual-channel detector. The emission is filtered with a high optical-density 700 nm short-pass filter (Thorlabs FESH0700), split with a 50/50 beam splitter (Thorlabs BSW10R), filtered with a 640/75 nm (Edmund Optics 67-022) and 590/43 nm band-pass filter (Edmund Optics 67-020), and detected with two high-sensitivity photo-multiplier tube (PMT) detectors (Hamamatsu H10723-20), respectively. The two emission bands were chosen to match the 640 nm emission peak of PpIX and to be distinct but spectrally close in order to detect auto-fluorescence from species such as lipofuscin, NADH, and elastin (Fig 1C).
We detected the output from the PMT detector with two low-frequency analog-to-digital (ADC) detectors and display the 640 nm signal from PpIX and auto-fluorescent specimens in red and the 590 nm signal from primarily auto-fluorescent specimens in green (Fig. 1c). As a result, auto-fluorescence appears in hues of yellow and green depending on the specific emission ratio while signal in the red channel is primarily driven by the presence of PpIX. We display both the multiwavelength fluorescence and SRH images and display images from the same optical section side-by-side. The images are linked to enable zoom/pan functionality, allowing for a quick correlation of the images during surgery (Fig. 1b, Supplemental Videos 1-5).
Experimentally, we determined that the PpIX fluorescence was maximal when both the 790 nm and 1030 nm beams were present at the sample. However, this condition also creates coherent anti-Stokes Raman scattering (CARS) at 641 nm, which coincidentally falls into the 640 nm emission band and overwhelms the weaker PpIX signal. For this reason, we de-tune the time-delay between the 790nm and 1030nm beams, such that CARS is no-longer generated but two-photon excited fluorescence (TPEF) from 790 nm and 1030 nm is generated independently and measured with the 640 nm detector. We therefore modified the excitation sequence, in which we first image the sample with SRS at 2845cm-1, then with SRS at 2930 cm-1, and finally with TPEF, essentially scanning the same image strip three times. The order of this sequence is reversed during consecutive strips to minimize laser tuning and optimize imaging speed (i.e., the TPEF is detected first for every other strip). Striping is not visible in the stitched images, indicating that photo-bleaching does not present a major issue in this acquisition scheme.
Performance of paired SRH/TPEF microscope in standards and tissue
We validated the multimodal imaging system by measuring the fluorescence intensity of serial dilutions of soluble PpIX at known concentrations and found that 1.0 𝜇g PpIX/ml was the threshold for detection (Supplementary Fig. 1). As expected, fluorescence intensity increased with increasing PpIX concentration between 1.0-62.5 𝜇g/ml (Fig. 1d), a range that mirrors the concentration of PpIX clinically encountered in tumor-infiltrated tissue from glioma patients receiving 5-ALA.28
We then evaluated the ability of the SRH/TPEF microscope to visualize PpIX in brain tumor specimens from patients treated with 5-ALA prior to resection or biopsy of suspected high-grade glioma. We noted that the SRH/TPEF microscope produced SRH images acquired at the 2845 cm-1 and 2930 cm-1 Raman shifts of comparable character to previous SRH microscopes without the TPEF module, and that the TPEF images could be acquired in corresponding fields of view (Fig. 2). Based on morphologic evaluation, it was possible to visualize the same tissue architecture and cytology in the SRH and TPEF images.
Clinical detection of PpIX in brain tumor tissue typically occurs via microscopes that excite PpIX at approximately 405nm and detect emission at 600-710nm via a single photon excitation fluorescence (SPEF). To ensure the comparable image generation between the conventional SPEF microscopes commonly employed in the operating room and the laboratory and the TPEF-based system we engineered, we used both techniques to image comparable optical sections from the same specimens (Fig. 2J, Supplementary Figs. 2 and 3) and verified that the pattern of PpIX tissue distribution was comparable between SPEF and TPEF across a series of 30 specimens from 20 patients. We also measured the emission from 569-667nm with 405nm excitation of sectioned tumor specimens with SPEF and observed a spectral signature consistent with PpIX fluorescence with peak emission at 630nm (Supplementary Fig. 2). We also performed a photobleaching experiment with SPEF confocal microscopy to confirm that the areas of signal intensity and patterns attributed to PpIX fluorescence decayed with exposure to 405nm excitation while the areas with high signal intensity and patterns attributed to autofluorescence do not (Supplemental Fig. 4).
Measuring the correlation of tumor cellularity and fluorescence
Both neoplastic and non-neoplastic cells contain endogenous fluorophores29. To explore the possibility that the observed fluorescence in PMT1 was specific to 5-ALA administration, we imaged tissue from 5 high-grade glioma patients who had not received 5-ALA prior to resection. Importantly, in these cases the observed pattern of fluorescence revealed only structures that fluoresced in both channels (rather than solely in PMT1) suggesting that there was absence of PpIX present (Supplementary Fig. 5).
Having validated the performance of the SRH/TPEF imaging system in PpIX solutions and a small number of human specimens, we sought to systematically evaluate PpIX distribution in a cohort of high-grade glioma patients (n=70, Supplementary Table 1). The multimodal optical microscope we developed offered a unique opportunity to quantitatively study the relationship between tumor cellularity and fluorescence in high-grade gliomas. To make the best use of the complete microscopic imaging dataset we collected, we utilized QuPath’s validated cell-detection module30, to count all of the cells (n=2,504,733 cells, Fig. 3a) in all imaged specimens (n=163 and a total area of 3,175.5 mm2) and developed a method for fluorescence quantification in the imaged specimens (Fig 3b). To quantify the fluorescence of PpIX, the signal from the PMT2 channel is regularized and subtracted from the PMT1 channel.
We noted a wide range of cellularity (average: 744 cells/mm2, range: 285-1506 cells/mm2) and fluorescence intensity amongst the specimens identified by operating surgeons as lesional and/or fluorescent during surgery (Supplementary Fig. 6). Using regression analysis of our entire dataset we also noted that there was no correlation between fluorescence and cellularity (R=-0.21, Fig. 3c, 3d).
Defining patterns of PpIX accumulation in high grade glioma tissue
Given the unexpected lack of correlation between cellularity and tissue fluorescence, we sought to better understand the nature of fluorescence in high-grade glioma tissue by characterizing the patterns of PpIX accumulation within tissue specimens. PpIX appeared to accumulate in both extracellular and intracellular spaces. Accumulation occurred to a variable degree both within individual specimens (Supplementary Fig. 7) and amongst patients (Fig. 4, Supplementary Fig. 6). Generally, there were three patterns of PpIX accumulation: (1) primarily autofluorescence (minimal PpIX accumulation, <5 a.u.), (2) dim PpIX fluorescence (5 - 40 a.u.) or (3) bright PpIX fluorescence (>40 a.u.). Intracellular concentration of PpIX occurred in two patterns, including: (1) axonal accumulation (935/166,743 or 0.6% fields of view, Supplementary Fig. 8) and (2) cytoplasmic accumulation (19,057/166,743 or 11.4% of fields of views, Supplementary Figs. 9 and 10). To better understand the variability in observed fluorescence, we examined a number of clinical variables with potential influence on the degree of observed fluorescence in study patients: the interval between 5-ALA dosing and specimen imaging (Supplementary Fig. 11), proportion of enhancing tumor, pattern of enhancement, and Ki-67 index.31 Using a linear fixed effects model, none of these variables were found to correlate with the measured concentration of PpIX (Supplementary Table 2).
Identifying the cells concentrating PpIX in human brain tumors
Abundant evidence demonstrates intracellular accumulation of PpIX within glioma cell lines after treatment with 5-ALA in culture.17–20 Notably, a definitive correlative study revealing PpIX within glioma cells in human tissues has not been reported. Intriguingly, an approach for PpIX visualization relying on light-sheet microscopy revealed accumulation of PpIX in a small minority of cells, many of which were in the perivascular space. The SRH/TPEF microscope we developed is particularly well suited for localization of PpIX in brain tumor tissue because it enables comparison of histomorphologic and fluorescent images of the same specimen in the same physical location with submicron resolution (Supplementary Videos 1-5).
Though rare in our dataset, we noted that 130,044/2,504,733 cells accumulated PpIX in the cytoplasm and that they generally exhibited morphology more consistent with histiocytes than tumor cells including reniform-shaped nuclei, a granular cytoplasmic appearance and variable cellular shape including both spherical and elongated forms (Supplementary Fig. 9). There was a high degree of variability in the abundance of cells with high intracellular PpIX concentration amongst and within subjects. However, cells with high intracellular PpIX concentrations were observed either along blood vessels (32/70 patients, Supplementary Fig. 10) or infiltrating amongst tumor cells in 62/70 patients in a pattern consistent with the known distribution of histiocytes within high-grade gliomas32.
Cytologic and histoarchitectural features led us to suspect cells concentrating PpIX intracellularly were myeloid cells, a subset of tumor cells or pericytes. We therefore employed immunostaining for markers of histiocytes (CD163+), tumor cells (GFAP+), and pericytes (SMA+) and in 5 cases (Supplementary Fig. 12).33 In these cases, we noted that the abundance and morphology of cells with intracellular PpIX accumulation mirrored that of CD163+ cells in the imaged specimens. We then evaluated the abundance of CD163 in the 45 study subjects, in which imaged tissue was available for CD163+ immunohistochemistry by comparing the degree of CD163 positivity to the quantity of cells concentrating PpIX within their cytoplasm. The abundance of cells concentrating PpIX within their cytoplasm was closely associated with quantified CD163 positivity (Supplementary Fig. 12). The difference in the number of cells concentrating PpIX in the cytoplasm was greatest between tissues with the greatest CD163 positivity and those with the least CD163 positivity (p=0.002) but remained significant even when tissues with moderate CD163 positivity were compared to those with the least CD163 positivity (p=0.02 (Supplementary Fig. 12).
5-ALA-induced fluorescence in non-glial brain tumors has been previously reported.34–38 Consequently, we hypothesized that PpIX might accumulate within CD163+ cells in non-glial tumors. In addition, there have been sporadic reports suggesting CD163 expression by glioma cells39. We examined tissue from three patients in whom 5-ALA was administered given a suspicion for high-grade glioma but were ultimately diagnosed with diffuse large B-cell primary central nervous system lymphoma. As observed in the glioma patients in the study, in tissue from CNS lymphoma biopsies, there was abundant intracellular accumulation of PpIX in cells that mirrored the population of CD163+ histiocytes, rather than in the neoplastic B-lymphocytes, in quantity and morphology (Supplementary Fig. 13).
Protoporphyrin IX is enriched in the myeloid cell population
To confirm the observation of elevated PpIX fluorescence intensity in macrophages, we leveraged recent integrated spatially resolved metabolomics (MALDI-FTICR-MSI) and transcriptomics (Visium 10X) data acquired from six GBM patients treated with 5-ALA40 (Fig. 5a-b). PpIX is metabolized from 5-ALA via an intermediate compound, coproporphyrinogen III (CpPIII). We hypothesized that active PPIX metabolism is marked by the simultaneous expression of the heme biosynthesis enzymes CPOX, HMBS and FECH and the accumulation of intermediate metabolites (CpPIII and PpIX). Spatial correlation of the recently described regional transcriptional expression patterns (Reactive Immune, Reactive Hypoxia, Spatial OPC, Radial Glia and Neuronal Development40) with both 5-ALA intermediate metabolites (CpPIII, PpIX) and gene expression of CPOX, HMBS and FECH revealed high co-localization of 5-ALA metabolism with immune active regions (Reactive Immune) (Fig. 5c). The Reactive Immune regions are marked by a high content of mesenchymal-like malignant cells with CD163/HMOX1 immunosuppressive macrophages41,42,43. Heme oxygenase 1 (HMOX1) has been shown to be an important marker of increased Heme metabolism44, which also suggests a key role of this subpopulation of macrophages in PPIX metabolism. We noted that this observation was also consistent with the finding that myeloid cells, unlike CD45 negative cells, have a significantly enhanced PpIX fluorescence signal45.
To verify the accumulation of PPIX in macrophages on a cellular level, we performed single cell deconvolution using the state-of-the-art Robust Cell Type Decomposition (RCTD)46 and the GBMap reference dataset containing approximately ~1 million cells47. The inherent heterogeneity of PpIX accumulation in our specimens enabled us to compare cellular composition across areas of low and high PpIX accumulation (Fig. 5d). RCTD provided cell type likelihood scores for each spot which were spatially correlated to the expression of enzymes (in red) and metabolite intensities (in blue) (Fig. 5d and Supplementary Fig. 14a). Immunosuppressive tumor associated macrophages (TAM’s) and inflammatory microglia demonstrated the strongest correlation with PpIX intensities and HMBS expression (Fig. 5d) with astrocyte-like malignant cells also showing a lesser degree of PpIX accumulation and HMBS and FECH expression. By extracting the mass spectra (regions of interest (ROI)) from both high CD163 expressing and non-expressing control regions we confirmed that higher PpIX intensity was present in macrophage-enriched regions (padj=0.0032) (Supplementary Fig. 14c-d).
To further validate our findings on an external, publicly available dataset, we utilized bulk RNA-seq obtained from PpIX positive cells separated with fluorescence-activated cell sorting (FACS) (Fig. 5e)48. Cell type deconvolution from bulk data by Multi-subject Single-cell Deconvolution (MuSiC)49 and the GBMap dataset, demonstrated enrichment of myeloid and oligodendroglial lineages amongst PpIX positive cells. Focusing on the analysis of RNA-seq data from the tumor infiltration zone (Fig. 5f and Supplementary Fig. 14.d-e) is particularly valuable since it contains both malignant and non-malignant cells (Fig. 5f and Supplementary Fig. 14.d-e). Notably, malignant cells represented only 5% of the PpIX-positive cells isolated from infiltration zone tissue which was predominantly composed of tumor associated macrophages and oligodendroglial cells (Fig. 5f). The cellular composition of the PpIX-positive fraction was further investigated by inferred copy number alterations demonstrating the lack of the characteristic gain of chromosome 7 and loss of chromosome 10 in PpIX-positive cells (Supplementary Fig. 14f-g and Supplementary Fig. 15). In conclusion, IHC, spatially resolved metabolomics and transcriptomics, and PpIX enriched RNA-seq suggest that PPIX is synthesized and accumulates predominantly within myeloid cells.