Browse the corpus

Single-cell analysis of 5-aminolevulinic acid intraoperative labeling specificity for glioblastoma. OBJECTIVE: Glioblastoma (GBM) is the most common and aggressive malignant primary brain tumor, and resection is a key part of the standard of care. In fluorescence-guided surgery (FGS), fluorophores differentiate tumor tissue from surrounding normal brain. The heme synthesis pathway converts 5-aminolevulinic acid (5-ALA), a fluorogenic substrate used for FGS, to fluorescent protoporphyrin IX (PpIX). The resulting fluorescence is believed to be specific to neoplastic glioma cells, but this specificity has not been examined at a single-cell level. The objective of this study was to determine the specificity with which 5-ALA labels the diversity of cell types in GBM. METHODS: The authors performed single-cell optical phenotyping and expression sequencing-version 2 (SCOPE-seq2), a paired single-cell imaging and RNA sequencing method, of individual cells on human GBM surgical specimens with macroscopically visible PpIX fluorescence from patients who received 5-ALA prior to surgery. SCOPE-seq2 allowed the authors to simultaneously image PpIX fluorescence and unambiguously identify neoplastic cells from single-cell RNA sequencing. Experiments were also conducted in cell culture and co-culture models of glioma and in acute slice cultures from a mouse glioma model to investigate cell- and tissue-specific uptake and secretion of 5-ALA and PpIX. RESULTS: SCOPE-seq2 analysis of human GBM surgical specimens revealed that 5-ALA treatment resulted in labeling that was not specific to neoplastic glioma cells. The cell culture further demonstrated that nonneoplastic cells could be labeled by 5-ALA directly or by PpIX secreted from surrounding neoplastic cells. Acute slice cultures from mouse glioma models showed that 5-ALA preferentially labeled GBM tumor tissue over nonneoplastic brain tissue with significant labeling in the tumor margins, and that this contrast was not due to blood-brain barrier disruption. CONCLUSIONS: Together, these findings support the use of 5-ALA as an indicator of GBM tissue but question the main advantage of 5-ALA for specific intracellular labeling of neoplastic glioma cells in FGS. Further studies are needed to systematically compare the performance of 5-ALA to that of potential alternatives for FGS.

Patients were given 5-ALA (Gleolan) at 20 mg/kg approximately 4 hours before surgery. Tissue specimens were stratified into 5-ALA–positive (red) and 5-ALA–negative (blue) tissues by the surgeon under violet-blue light and collected immediately after surgical removal. Collected tissue samples were dissociated using the adult brain dissociation kit (Miltenyi Biotec, catalog no. 130-107-677) on a gentleMACS Octo Dissociator with heaters (Miltenyi Biotec) according to the manufacturer’s instructions, with the following modifications. Calcein AM (Thermo Fisher Scientific, catalog no. C3100MP) was added to the dissociation enzyme mix to a final concentration of 4 µM to label live cells. Cells were finally resuspended in tris-buffered saline buffer at a concentration of 1000 cells per microliter. Dissociated cells were profiled using the SCOPE-seq version 2 (SCOPE-seq2) method as previously described.29 A detailed experimental protocol along with our imaging and sequencing analysis procedures is shown in the Supplementary Methods.

Patients were given 5-ALA (Gleolan) at 20 mg/kg approximately 4 hours before surgery. Tissue specimens were stratified into 5-ALA–positive (red) and 5-ALA–negative (blue) tissues by the surgeon under violet-blue light and collected immediately after surgical removal. Collected tissue samples were dissociated using the adult brain dissociation kit (Miltenyi Biotec, catalog no. 130-107-677) on a gentleMACS Octo Dissociator with heaters (Miltenyi Biotec) according to the manufacturer’s instructions, with the following modifications. Calcein AM (Thermo Fisher Scientific, catalog no. C3100MP) was added to the dissociation enzyme mix to a final concentration of 4 µM to label live cells. Cells were finally resuspended in tris-buffered saline buffer at a concentration of 1000 cells per microliter. Dissociated cells were profiled using the SCOPE-seq version 2 (SCOPE-seq2) method as previously described.29 A detailed experimental protocol along with our imaging and sequencing analysis procedures is shown in the Supplementary Methods. For clustering and visualization, we first subsampled the merged data set from all patients included in the study so that every sample had the same number of cells. We then factorized the subsampled gene count matrix using the single-cell hierarchical Poisson factorization algorithm30 with default parameters and k = 17 (where k is the number of factors in the model), and identified the top 100 genes in each factor as variable genes. We removed nuisance factors associated with heat shock protein genes and cell stress. To cluster the expression profiles, we computed the Spearman correlation distance from the variable-gene count matrix, constructed a k-nearest neighbors graph with k = 20, and clustered using the Phenograph implementation of Louvain community detection.31 To visualize the expression profiles, we generated a 2D embedding of the Spearman correlation distance matrix using uniform manifold approximation and projection (UMAP).32

unt matrix, constructed a k-nearest neighbors graph with k = 20, and clustered using the Phenograph implementation of Louvain community detection.31 To visualize the expression profiles, we generated a 2D embedding of the Spearman correlation distance matrix using uniform manifold approximation and projection (UMAP).32 For clustering and visualization of expression profiles from each sample, variable genes were selected as genes detected in fewer cells than expected given their apparent expression level using the dropout curve method, as previously described.26 Phenograph clustering and UMAP embedding were processed as described above. Genomic DNA was extracted from each bulk tumor tissue and peripheral blood mononuclear cell sample using the DNeasy Blood & Tissue Kits (Qiagen) according to the manufacturer’s instructions. DNA libraries were prepared using a Nextera DNA library preparation kit according to the manufacturer’s instructions. Raw sequencing data were aligned to the human genome (GRCh38) using a bwa-mem algorithm33 and analyzed as previously described.26 Briefly, we computed the number of de-duplicated reads that aligned to each chromosome for each patient and divided this by the average number of de-duplicated reads that aligned to each chromosome for two diploid germline samples, after normalizing both by total reads. We then normalized this ratio by the median ratio across all somatic chromosomes and multiplied by 2 to estimate the average copy number of each chromosome.

divided this by the average number of de-duplicated reads that aligned to each chromosome for two diploid germline samples, after normalizing both by total reads. We then normalized this ratio by the median ratio across all somatic chromosomes and multiplied by 2 to estimate the average copy number of each chromosome. Neoplastic cells in glioma were identified based on aneuploidies detected by scRNA-seq as described previously.34 Detailed methodological procedures are described in the Supplementary Methods. The TS543 human glioma cell line was cultured in human NeuroCult NS-A basal medium (Stem Cell Technologies, catalog no. 057a51), supplemented with 10% NeuroCult proliferation supplement, heparin (2 µg/ml; Stem Cell Technologies, catalog no. 7980), human basic fibroblast growth factor (10 ng/ml; Stem Cell Technologies, catalog no. 78003), and human epidermal growth factor (20 ng/ml; Stem Cell Technologies, catalog no. 78006). The HMC3 human microglia cell line was cultured in Eagle’s minimum essential medium (ATCC, catalog no. 30-2003) supplemented with 10% fetal bovine serum (FBS; Gibco, catalog no. A3160402). Cell lines were incubated at 37°C and 5% CO2.

human epidermal growth factor (20 ng/ml; Stem Cell Technologies, catalog no. 78006). The HMC3 human microglia cell line was cultured in Eagle’s minimum essential medium (ATCC, catalog no. 30-2003) supplemented with 10% fetal bovine serum (FBS; Gibco, catalog no. A3160402). Cell lines were incubated at 37°C and 5% CO2. For 5-ALA treatment, 5-ALA hydrochloride (Sigma, catalog no. A7793) was dissolved in phosphate-buffered saline (PBS) and pH was adjusted to 7.4 using NaOH. Cells were incubated in medium with 1 mM 5-ALA for 4 hours at 37°C and 5% CO2. TS543 or HMC3 cells were then dissociated with TrypLE Express enzyme (Gibco, catalog no. 12605028) and stained with Calcein AM live stain dye (Thermo Fisher Scientific, catalog no. C3100MP) at 4 µM for 5 minutes. For incubation with 5-ALA–treated cells, we first treated human glioma TS543 cells with 5-ALA as described above. Dissociated 5-ALA–treated TS543 cells were washed with PBS three times and resuspended in fresh TS543 culture medium without exogenous 5-ALA. We split the resuspension into two parts: one was referred to as 5-ALA–treated cells, and the other was further centrifuged and the supernatant was referred to as 5-ALA–treated supernatant. 5-ALA–untreated TS543 or HMC3 cells were dissociated with TrypLE Express enzyme and stained with Calcein AM as described above. We then mixed 5-ALA untreated cells with treated cells and incubated them at 37°C and 5% CO2 for 1 hour.

s further centrifuged and the supernatant was referred to as 5-ALA–treated supernatant. 5-ALA–untreated TS543 or HMC3 cells were dissociated with TrypLE Express enzyme and stained with Calcein AM as described above. We then mixed 5-ALA untreated cells with treated cells and incubated them at 37°C and 5% CO2 for 1 hour. GL261 cells were grown in Dulbecco’s modified Eagle medium (DMEM)–10% FBS at 37°C and 5% CO2. Adult C57Bl/6 mice were anesthetized with ketamine-xylazine (100 mg/kg and 10 mg/kg, respectively) and assessed for lack of reflexes by toe pinch. The hair was shaved and scalp skin incised. A burr hole was made with a 17-gauge needle 2 mm lateral and 2 mm anterior to the bregma. Intracranial injection (5 × 103 cells in 1 µl) was performed under stereotactic guidance, 2 mm deep, using a Hamilton syringe at a flow rate of 0.3 µl/min. Tumor growth was assessed through MRI. Mice were killed at 25 days postinjection by cervical dislocation. The brain was removed, placed into ice-cold PBS, and cut into 300- to 500-µm sections using a McIlwain Tissue Chopper. Sections were transferred onto Millicell cell culture inserts (0.4 µm, 30-mm diameter; Sigma-Aldrich, catalog no. PICMORG50) that were placed in 6-well plates containing 1.5 ml of medium consisting of DMEM/F12 with N-2 supplement and 1% antimycotic/antibiotic. Slices were first rested overnight with the medium in a humidified incubator at 37°C and 5% CO2. Then, the medium was replaced with prewarmed medium containing 1 mM 5-ALA or PBS as control. Slices were then cultured with the treatment medium in a humidified incubator at 37°C and 5% CO2 for 4 hours. After 5-ALA treatment, slices were fixed by 4% paraformaldehyde (Thermo Scientific, catalog no. 28908) for 24 hours at 4°C and incubated in 30% sucrose at 4°C until they were embedded in optimal cutting temperature compound for cryosectioning into 16-µm sections. Adjacent tissue slides of each slice culture were used for PpIX fluorescence measurement and H&E staining. For PpIX fluorescence measurement, slides were warmed up to room temperature, rehydrated with PBS for 10 minutes, stained with 10 nM SYTOX Green (Thermo Fisher Scientific) for 5 minutes, washed with PBS, and mounted with Fluoromount-G mounting medium (Invitrogen, catalog no. 00-4958-02).

e measurement and H&E staining. For PpIX fluorescence measurement, slides were warmed up to room temperature, rehydrated with PBS for 10 minutes, stained with 10 nM SYTOX Green (Thermo Fisher Scientific) for 5 minutes, washed with PBS, and mounted with Fluoromount-G mounting medium (Invitrogen, catalog no. 00-4958-02). For H&E staining, slides were placed in PBS for 5 minutes, stained with H&E (5–10 seconds each), washed with deionized water until clear, dehydrated (once in 70%, twice in 90%, twice in 100% ethanol, twice in xylene, 10 seconds each step), and mounted with Permount (Thermo Fisher Scientific).

e measurement and H&E staining. For PpIX fluorescence measurement, slides were warmed up to room temperature, rehydrated with PBS for 10 minutes, stained with 10 nM SYTOX Green (Thermo Fisher Scientific) for 5 minutes, washed with PBS, and mounted with Fluoromount-G mounting medium (Invitrogen, catalog no. 00-4958-02). For H&E staining, slides were placed in PBS for 5 minutes, stained with H&E (5–10 seconds each), washed with deionized water until clear, dehydrated (once in 70%, twice in 90%, twice in 100% ethanol, twice in xylene, 10 seconds each step), and mounted with Permount (Thermo Fisher Scientific). For cell cultures, images were obtained with an epifluorescence microscope (Nikon Eclipse Ti2) in two fluorescence channels and the bright-field (BF) channel with a wide-field 10 × 0.3 numerical aperture (NA) objective (Nikon, catalog no. MRH00101). BF images were obtained using an RGB light source (Lumencor, Lida). Fluorescence images were taken using an LED light source (Lumencor, SPECTRA X). Calcein AM fluorescence was excited at 470 nm and emission collected at 500–550 nm. PpIX fluorescence was excited at 395 nm and emission collected at 600–660 nm. We used ImageJ (NIH) for image analysis. Cell regions of interest were identified from the Calcein AM fluorescence using threshold and particle analyzer modules. Then, the cell mean PpIX fluorescence was measured for each individual cell. To compare PpIX fluorescence across treatment conditions in cell lines, we first downsampled cells in each treatment condition into the same number of cells and then performed an unpaired t-test implemented as ttest_ind from Python module scipy.

en, the cell mean PpIX fluorescence was measured for each individual cell. To compare PpIX fluorescence across treatment conditions in cell lines, we first downsampled cells in each treatment condition into the same number of cells and then performed an unpaired t-test implemented as ttest_ind from Python module scipy. For mice slice culture slides, fluorescence images were taken using a Nikon A1R MP confocal microscope with a 25 × 1.1 NA objective. SYTOX Green fluorescence was excited at 488 nm and emission collected at 500–550 nm. PpIX fluorescence was excited at 403 nm and emission collected at 663–738 nm. H&E images were taken using an upright microscope (Nikon Optiphot) with a 20 × 0.3 NA objective. We used ImageJ for image analysis. Tissue regions were identified from the SYTOX Green fluorescence using threshold; regions with SYTOX Green intensity < 10 grayscale units in the grayscale image were considered regions without tissue presence. Then, the mean PpIX fluorescence of the tissue was measured for each slice culture. To compare PpIX fluorescence across treatment conditions in slice cultures, we performed an unpaired t-test implemented as ttest_ind from Python module scipy.

units in the grayscale image were considered regions without tissue presence. Then, the mean PpIX fluorescence of the tissue was measured for each slice culture. To compare PpIX fluorescence across treatment conditions in slice cultures, we performed an unpaired t-test implemented as ttest_ind from Python module scipy. To compare PpIX fluorescence across cell types, we first downsampled cells in each cell type and each patient into the same number of cells. Then we performed a two-sided Mann-Whitney U-test as implemented by the mannwhitneyu command in the Python module scipy between two cell subpopulations within each patient sample. The resulting p values were corrected using the Benjamini-Hochberg method as implemented by the multipletests function in the Python package statsmodels. All animal protocols used in these studies were approved by the Columbia University Institutional Animal Care and Use Committee. For studies of human specimens, all tissue was procured from de-identified patients who provided written informed consent through a protocol approved by the Columbia University IRB.

Patients were given 5-ALA (Gleolan) at 20 mg/kg approximately 4 hours before surgery. Tissue specimens were stratified into 5-ALA–positive (red) and 5-ALA–negative (blue) tissues by the surgeon under violet-blue light and collected immediately after surgical removal. Collected tissue samples were dissociated using the adult brain dissociation kit (Miltenyi Biotec, catalog no. 130-107-677) on a gentleMACS Octo Dissociator with heaters (Miltenyi Biotec) according to the manufacturer’s instructions, with the following modifications. Calcein AM (Thermo Fisher Scientific, catalog no. C3100MP) was added to the dissociation enzyme mix to a final concentration of 4 µM to label live cells. Cells were finally resuspended in tris-buffered saline buffer at a concentration of 1000 cells per microliter.

Dissociated cells were profiled using the SCOPE-seq version 2 (SCOPE-seq2) method as previously described.29 A detailed experimental protocol along with our imaging and sequencing analysis procedures is shown in the Supplementary Methods.

For clustering and visualization, we first subsampled the merged data set from all patients included in the study so that every sample had the same number of cells. We then factorized the subsampled gene count matrix using the single-cell hierarchical Poisson factorization algorithm30 with default parameters and k = 17 (where k is the number of factors in the model), and identified the top 100 genes in each factor as variable genes. We removed nuisance factors associated with heat shock protein genes and cell stress. To cluster the expression profiles, we computed the Spearman correlation distance from the variable-gene count matrix, constructed a k-nearest neighbors graph with k = 20, and clustered using the Phenograph implementation of Louvain community detection.31 To visualize the expression profiles, we generated a 2D embedding of the Spearman correlation distance matrix using uniform manifold approximation and projection (UMAP).32 For clustering and visualization of expression profiles from each sample, variable genes were selected as genes detected in fewer cells than expected given their apparent expression level using the dropout curve method, as previously described.26 Phenograph clustering and UMAP embedding were processed as described above.

Genomic DNA was extracted from each bulk tumor tissue and peripheral blood mononuclear cell sample using the DNeasy Blood & Tissue Kits (Qiagen) according to the manufacturer’s instructions. DNA libraries were prepared using a Nextera DNA library preparation kit according to the manufacturer’s instructions. Raw sequencing data were aligned to the human genome (GRCh38) using a bwa-mem algorithm33 and analyzed as previously described.26 Briefly, we computed the number of de-duplicated reads that aligned to each chromosome for each patient and divided this by the average number of de-duplicated reads that aligned to each chromosome for two diploid germline samples, after normalizing both by total reads. We then normalized this ratio by the median ratio across all somatic chromosomes and multiplied by 2 to estimate the average copy number of each chromosome.

The TS543 human glioma cell line was cultured in human NeuroCult NS-A basal medium (Stem Cell Technologies, catalog no. 057a51), supplemented with 10% NeuroCult proliferation supplement, heparin (2 µg/ml; Stem Cell Technologies, catalog no. 7980), human basic fibroblast growth factor (10 ng/ml; Stem Cell Technologies, catalog no. 78003), and human epidermal growth factor (20 ng/ml; Stem Cell Technologies, catalog no. 78006). The HMC3 human microglia cell line was cultured in Eagle’s minimum essential medium (ATCC, catalog no. 30-2003) supplemented with 10% fetal bovine serum (FBS; Gibco, catalog no. A3160402). Cell lines were incubated at 37°C and 5% CO2. For 5-ALA treatment, 5-ALA hydrochloride (Sigma, catalog no. A7793) was dissolved in phosphate-buffered saline (PBS) and pH was adjusted to 7.4 using NaOH. Cells were incubated in medium with 1 mM 5-ALA for 4 hours at 37°C and 5% CO2. TS543 or HMC3 cells were then dissociated with TrypLE Express enzyme (Gibco, catalog no. 12605028) and stained with Calcein AM live stain dye (Thermo Fisher Scientific, catalog no. C3100MP) at 4 µM for 5 minutes.

adjusted to 7.4 using NaOH. Cells were incubated in medium with 1 mM 5-ALA for 4 hours at 37°C and 5% CO2. TS543 or HMC3 cells were then dissociated with TrypLE Express enzyme (Gibco, catalog no. 12605028) and stained with Calcein AM live stain dye (Thermo Fisher Scientific, catalog no. C3100MP) at 4 µM for 5 minutes. For incubation with 5-ALA–treated cells, we first treated human glioma TS543 cells with 5-ALA as described above. Dissociated 5-ALA–treated TS543 cells were washed with PBS three times and resuspended in fresh TS543 culture medium without exogenous 5-ALA. We split the resuspension into two parts: one was referred to as 5-ALA–treated cells, and the other was further centrifuged and the supernatant was referred to as 5-ALA–treated supernatant. 5-ALA–untreated TS543 or HMC3 cells were dissociated with TrypLE Express enzyme and stained with Calcein AM as described above. We then mixed 5-ALA untreated cells with treated cells and incubated them at 37°C and 5% CO2 for 1 hour.

GL261 cells were grown in Dulbecco’s modified Eagle medium (DMEM)–10% FBS at 37°C and 5% CO2. Adult C57Bl/6 mice were anesthetized with ketamine-xylazine (100 mg/kg and 10 mg/kg, respectively) and assessed for lack of reflexes by toe pinch. The hair was shaved and scalp skin incised. A burr hole was made with a 17-gauge needle 2 mm lateral and 2 mm anterior to the bregma. Intracranial injection (5 × 103 cells in 1 µl) was performed under stereotactic guidance, 2 mm deep, using a Hamilton syringe at a flow rate of 0.3 µl/min. Tumor growth was assessed through MRI. Mice were killed at 25 days postinjection by cervical dislocation. The brain was removed, placed into ice-cold PBS, and cut into 300- to 500-µm sections using a McIlwain Tissue Chopper. Sections were transferred onto Millicell cell culture inserts (0.4 µm, 30-mm diameter; Sigma-Aldrich, catalog no. PICMORG50) that were placed in 6-well plates containing 1.5 ml of medium consisting of DMEM/F12 with N-2 supplement and 1% antimycotic/antibiotic. Slices were first rested overnight with the medium in a humidified incubator at 37°C and 5% CO2. Then, the medium was replaced with prewarmed medium containing 1 mM 5-ALA or PBS as control. Slices were then cultured with the treatment medium in a humidified incubator at 37°C and 5% CO2 for 4 hours. After 5-ALA treatment, slices were fixed by 4% paraformaldehyde (Thermo Scientific, catalog no. 28908) for 24 hours at 4°C and incubated in 30% sucrose at 4°C until they were embedded in optimal cutting temperature compound for cryosectioning into 16-µm sections. Adjacent tissue slides of each slice culture were used for PpIX fluorescence measurement and H&E staining. For PpIX fluorescence measurement, slides were warmed up to room temperature, rehydrated with PBS for 10 minutes, stained with 10 nM SYTOX Green (Thermo Fisher Scientific) for 5 minutes, washed with PBS, and mounted with Fluoromount-G mounting medium (Invitrogen, catalog no. 00-4958-02).

For cell cultures, images were obtained with an epifluorescence microscope (Nikon Eclipse Ti2) in two fluorescence channels and the bright-field (BF) channel with a wide-field 10 × 0.3 numerical aperture (NA) objective (Nikon, catalog no. MRH00101). BF images were obtained using an RGB light source (Lumencor, Lida). Fluorescence images were taken using an LED light source (Lumencor, SPECTRA X). Calcein AM fluorescence was excited at 470 nm and emission collected at 500–550 nm. PpIX fluorescence was excited at 395 nm and emission collected at 600–660 nm. We used ImageJ (NIH) for image analysis. Cell regions of interest were identified from the Calcein AM fluorescence using threshold and particle analyzer modules. Then, the cell mean PpIX fluorescence was measured for each individual cell. To compare PpIX fluorescence across treatment conditions in cell lines, we first downsampled cells in each treatment condition into the same number of cells and then performed an unpaired t-test implemented as ttest_ind from Python module scipy.

To compare PpIX fluorescence across cell types, we first downsampled cells in each cell type and each patient into the same number of cells. Then we performed a two-sided Mann-Whitney U-test as implemented by the mannwhitneyu command in the Python module scipy between two cell subpopulations within each patient sample. The resulting p values were corrected using the Benjamini-Hochberg method as implemented by the multipletests function in the Python package statsmodels.

All animal protocols used in these studies were approved by the Columbia University Institutional Animal Care and Use Committee. For studies of human specimens, all tissue was procured from de-identified patients who provided written informed consent through a protocol approved by the Columbia University IRB.

To investigate whether 5-ALA specifically induces PpIX fluorescence in tumor cells, we used SCOPE-seq229 to perform integrated fluorescence imaging and RNA sequencing of individual cells collected from 4 resected GBM patient specimens (Fig. 1, Table S1) treated with 5-ALA (Fig. 2). We characterized morphological phenotypes and PpIX fluorescence of single cells captured in a microwell array device using a scanning epifluorescence microscope (Fig. S1). In SCOPE-seq2, we capture mRNA from individual cells using beads coated in barcoded DNA primers. We could identify the barcode sequence associated with each bead-cell pair in our device using sequential fluorescent probe hybridization. Because this same barcode sequence is associated with each single-cell complementary DNA library during Illumina sequencing, we could directly link PpIX fluorescence imaging data from each cell with its gene expression profile (see Methods). Representative MR and intraoperative images of 5-ALA in GBM. A: Preoperative MR images annotated with intraoperative neuronavigation. The position of resected tissue is indicated by the green crosshairs. B: 5-ALA–induced PpIX fluorescence of resected tissue located at the tumor margin under violet-blue light with both 5-ALA–positive (red fluorescence) and 5-ALA–negative (blue fluorescence) regions.

mages annotated with intraoperative neuronavigation. The position of resected tissue is indicated by the green crosshairs. B: 5-ALA–induced PpIX fluorescence of resected tissue located at the tumor margin under violet-blue light with both 5-ALA–positive (red fluorescence) and 5-ALA–negative (blue fluorescence) regions. Schematic illustration of experimental and analytical methods using SCOPE-seq2 for comparing 5-ALA–induced PpIX fluorescence among cell subpopulations in the tumor microenvironment. cDNA = complementary DNA; TSO = template-switching oligonucleotide; TTT = oligo(dT), a segment of repeating deoxythymidines; UMI = unique molecular identifier.

n of experimental and analytical methods using SCOPE-seq2 for comparing 5-ALA–induced PpIX fluorescence among cell subpopulations in the tumor microenvironment. cDNA = complementary DNA; TSO = template-switching oligonucleotide; TTT = oligo(dT), a segment of repeating deoxythymidines; UMI = unique molecular identifier. We obtained GBM surgical specimens from 4 patients who received 5-ALA preoperatively (Table S1). For each patient, we collected one tissue specimen with intraoperatively observed PpIX fluorescence from bulk tumor and performed SCOPE-seq2 including single-cell fluorescence imaging and RNA sequencing. For one of the cases (no. 0826), we received a second specimen from a nonfluorescent region of the tumor and performed single-cell fluorescence imaging in our device as a negative control. As expected, we observed that all 4 specimens for which PpIX fluorescence was observed intraoperatively exhibited higher PpIX fluorescence (p < 0.01, Mann-Whitney U-test) by single-cell imaging than the specimen in which PpIX fluorescence was not observed intraoperatively (Fig. 3A). We note that, as anticipated, there was some overlap between the single-cell fluorescence intensity distributions of the PpIX-positive and PpIX-negative samples because the PpIX-positive samples contain a mixture of strongly fluorescent and inefficiently labeled cells. Thus, the microscopic imaging that we performed on individual cells was consistent with what we observed intraoperatively.

cell fluorescence intensity distributions of the PpIX-positive and PpIX-negative samples because the PpIX-positive samples contain a mixture of strongly fluorescent and inefficiently labeled cells. Thus, the microscopic imaging that we performed on individual cells was consistent with what we observed intraoperatively. 5-ALA labeling in the tumor microenvironment. A: Ridge plot of cell average PpIX fluorescence. B: UMAP embedding of scRNA-seq expression profiles colored according to the 4 patients. C: UMAP embedding of scRNA-seq expression profiles colored according to ploidy score. D: UMAP embedding of scRNA-seq expression profiles colored by cell subpopulations. E: Heatmap of average expression of lineage marker genes from cell types in the tumor microenvironment in each cell subpopulation and patient. F: UMAP embedding of scRNA-seq expression profiles colored by average PpIX fluorescence. G: Representative fluorescence images of each cell subpopulation in patient 0409. Bar = 5 µm. H: Box plot of average PpIX fluorescence in each cell subpopulation and patient sample. PpIX fluorescence is compared between neoplastic glioma cells and other nonneoplastic cell subpopulations of the same patient. *q < 0.05, Mann-Whitney U-test.

uorescence images of each cell subpopulation in patient 0409. Bar = 5 µm. H: Box plot of average PpIX fluorescence in each cell subpopulation and patient sample. PpIX fluorescence is compared between neoplastic glioma cells and other nonneoplastic cell subpopulations of the same patient. *q < 0.05, Mann-Whitney U-test. To investigate whether 5-ALA specifically labels tumor cells in tissue with visible PpIX fluorescence, we first identified cellular subpopulations from single-cell expression profiles. We performed unsupervised clustering and embedded the expression profiles in two dimensions using UMAP (Fig. 3B–D).32 We used aneuploidies, which can be inferred from scRNA-seq data, to identify neoplastic cells as described previously.26 Chromosome 7 amplification and chromosome 10 deletion are two common aneuploidies in isocitrate dehydrogenase (IDH)–wild-type GBM, and chromosome 7 amplification may occur in IDH-mutant high-grade glioma.26,35 We verified the presence of these alterations using bulk, low-coverage, whole-genome sequencing of the samples processed with SCOPE-seq2 (Fig. S2). We distinguished neoplastic cells from nonneoplastic cells by identifying which cells harbored these copy number variants (Fig. 3C, Fig. S3) from scRNA-seq. For example, cells with amplification of chromosome 7 express genes on chromosome 7 at higher levels on average. By examining the expression of cell lineage marker genes in each cluster, the nonneoplastic cells were further classified into myeloid cells (C1QA, C1QB, CD14, CD163), oligodendrocytes (MBP, MOG, MAG), and T cells (CD3D, CD3E, TRAC; Fig. 3D and E). Overall, the cellular composition of the GBM specimens profiled in this study was highly concordant with previous analyses of GBM by scRNA-seq.26,27

nonneoplastic cells were further classified into myeloid cells (C1QA, C1QB, CD14, CD163), oligodendrocytes (MBP, MOG, MAG), and T cells (CD3D, CD3E, TRAC; Fig. 3D and E). Overall, the cellular composition of the GBM specimens profiled in this study was highly concordant with previous analyses of GBM by scRNA-seq.26,27 To compare 5-ALA labeling among cellular subpopulations, we analyzed the paired-cell PpIX fluorescence and expression profiling data (Fig. 3F–H). We observed that all cell types in the tissue with macroscopically visible PpIX fluorescence had higher PpIX fluorescence than cells from tissue without visible PpIX fluorescence (q < 0.001, Mann-Whitney U-test, where q is a false discovery rate–corrected p value; Fig. 3H, Table S2), suggesting that 5-ALA treatment results in PpIX fluorescence in both neoplastic and nonneoplastic cells within the tumor microenvironment. In some cases, the neoplastic cells were not even the most fluorescent cells in the tumor. While T cells often have lower PpIX fluorescence than neoplastic cells, oligodendrocytes were comparable to and myeloid cells were brighter than neoplastic cells (q < 0.05, Mann-Whitney U-test; Fig. 3H, Table S2).

croenvironment. In some cases, the neoplastic cells were not even the most fluorescent cells in the tumor. While T cells often have lower PpIX fluorescence than neoplastic cells, oligodendrocytes were comparable to and myeloid cells were brighter than neoplastic cells (q < 0.05, Mann-Whitney U-test; Fig. 3H, Table S2). To examine whether gene expression was related to cell PpIX fluorescence, we correlated the expression of protein coding genes with PpIX fluorescence in neoplastic cells for each patient. We ranked genes by their correlation and performed gene set enrichment analysis.36 This analysis did not identify any consistent, significant gene ontologies that were correlated with fluorescence intensity among the neoplastic glioma cells across patients (Fig. S4).

X fluorescence in neoplastic cells for each patient. We ranked genes by their correlation and performed gene set enrichment analysis.36 This analysis did not identify any consistent, significant gene ontologies that were correlated with fluorescence intensity among the neoplastic glioma cells across patients (Fig. S4). We found that nonneoplastic cells within the GBM tumor microenvironment were fluorescently labeled after 5-ALA treatment, but the labeling mechanism of nonneoplastic cells in the tumor microenvironment was unknown. One potential labeling mechanism is that nonneoplastic cells can take up exogeneous 5-ALA and convert it into fluorescent PpIX. Because neoplastic cells can generate and secrete PpIX into the extracellular space after 5-ALA treatment,37,38 another potential labeling mechanism is that nonneoplastic cells take up PpIX secreted by neoplastic cells. As shown above, myeloid cells such as brain-resident microglia are highly fluorescent in GBM after 5-ALA treatment.39 Thus, to determine whether nonneoplastic cells can take up 5-ALA and convert it into PpIX directly like neoplastic cells, we measured the PpIX fluorescence of the 5-ALA–treated human microglial cell line HMC3 and human GBM neurosphere cell line TS543. We observed that both neoplastic glioma cells and nonneoplastic myeloid cells had significantly increased PpIX fluorescence after 5-ALA treatment (p < 0.001, t-test; Fig. 4A and B, Table S3). To determine whether cells can take up PpIX secreted by neoplastic cells, we measured PpIX fluorescence of untreated TS543 or HMC3 cells after incubation with 5-ALA–treated TS543 cells. To differentiate untreated cells from 5-ALA–treated cells in images, we labeled untreated cells with live stain Calcein AM before incubation. To ensure that the observed PpIX fluorescence in 5-ALA–untreated cells was not induced by 5-ALA directly, we washed 5-ALA–treated cells thoroughly with PBS to remove exogeneous 5-ALA and resuspended the treated cells in fresh medium before incubation with 5-ALA–untreated cells, and we also measured the PpIX fluorescence of 5-ALA–untreated cells incubated with the resuspension supernatant. We found that both GBM cells and nonneoplastic microglia had significantly increased PpIX fluorescence after incubation with 5-ALA–labeled GBM cells (p < 0.001, t-test; Table S4) but not after incubation with the supernatant (Fig. 4C and D).

rescence of 5-ALA–untreated cells incubated with the resuspension supernatant. We found that both GBM cells and nonneoplastic microglia had significantly increased PpIX fluorescence after incubation with 5-ALA–labeled GBM cells (p < 0.001, t-test; Table S4) but not after incubation with the supernatant (Fig. 4C and D). These results suggest that PpIX fluorescence of nonneoplastic cells in a GBM tumor microenvironment can be induced directly by 5-ALA uptake and conversion to PpIX, or indirectly by uptake of PpIX secreted from surrounding neoplastic cells. Potential mechanisms of 5-ALA labeling of nonneoplastic cells. A: Example images of TS543 (glioma) and HMC3 (cultured microglia) cells treated with PBS control or 5-ALA. Calcein AM = live cells; PpIX = 5-ALA labeling. Bar = 100 µm. B: Box plot shows the quantification of PpIX fluorescence of TS543 and HMC3 cells treated with PBS control or 5-ALA. ***p < 0.001, unpaired t-test. C: Representative images of 5-ALA untreated TS543 and HMC3 cells before (None) or after treatment with 5-ALA–treated cell supernatant or 5-ALA–treated TS543 cells. Calcein AM = 5-ALA untreated cells; PpIX = 5-ALA labeling. Arrowheads indicate 5-ALA–untreated cells. Bar = 100 µm. D: Box plot shows the quantification of PpIX fluorescence of 5-ALA–untreated TS543 and HMC3 cells before (None) or after treatment with 5-ALA–treated cell supernatant or 5-ALA–treated TS543 cells. ***p < 0.001, unpaired t-test. NS = not significant.

ALA labeling. Arrowheads indicate 5-ALA–untreated cells. Bar = 100 µm. D: Box plot shows the quantification of PpIX fluorescence of 5-ALA–untreated TS543 and HMC3 cells before (None) or after treatment with 5-ALA–treated cell supernatant or 5-ALA–treated TS543 cells. ***p < 0.001, unpaired t-test. NS = not significant. Although 5-ALA labels both neoplastic and nonneoplastic cells in GBM, 5-ALA is reported to preferentially label tumor tissue.9 While this preference could arise from specific labeling of neoplastic glioma cells, our SCOPE-seq2 experiments eliminated this possibility. Another possibility is that tumor tissue-selective labeling is due to BBB disruption by the tumor, making 5-ALA more accessible to tumor tissue compared with nonneoplastic brain tissue. A third possibility is that metabolic alterations in the GBM microenvironment could result in PpIX accumulation. To remove the effect of the BBB during 5-ALA labeling, we generated acute brain slice cultures from mice injected with GL261 cells—which exhibit comparably increased PpIX fluorescence to human GBM neuroshperes (TS543) after 5-ALA treatment (Fig. S5)—and from control mice. GL261 injection resulted in the formation of syngeneic brain tumors, and we treated the slice cultures from the tumor-bearing and control mice with 5-ALA ex vivo (Fig. 5A). After 5-ALA treatment, we generated thin sections for each slice and measured the tissue average PpIX fluorescence using confocal microscopy. We then used H&E-stained adjacent sections to differentiate tumor and normal tissues. We observed that tumor tissues in GL261 mice showed significantly increased average PpIX fluorescence across the imaging field after 5-ALA treatment (p < 0.01, t-test; Fig. 5B and C, Table S5), while the average PpIX fluorescence of normal brain tissues in both GL261 and control mice did not differ significantly. We also observed that tumor tissues had higher average PpIX fluorescence than normal tissues after 5-ALA treatment in GL261 mice (t-test, p < 0.01), indicating that 5-ALA selectively labels GBM tumor tissue at the bulk level even given uniform access to the tumor and normal brain. At the tumor margin, we found that tissue regions inside and outside the tumor have increased average PpIX fluorescence after 5-ALA treatment, but the fluorescence intensity inside the tumor is higher than outside the tumor (Fig. S6, Table S6).

at the bulk level even given uniform access to the tumor and normal brain. At the tumor margin, we found that tissue regions inside and outside the tumor have increased average PpIX fluorescence after 5-ALA treatment, but the fluorescence intensity inside the tumor is higher than outside the tumor (Fig. S6, Table S6). These results suggest that although 5-ALA labeling is not specific to neoplastic glioma cells, selective labeling of tumor tissue over nonneoplastic brain tissue is independent of BBB disruption. In vitro treatment of 5-ALA on mouse slice cultures. A: Schematic illustration of in vitro 5-ALA treatment of mouse brain slice cultures and imaging. B: Representative images of normal and tumor tissues in control and GL261 mouse brain slice cultures treated with PBS control or 5-ALA. SYTOX Green labels all nuclei. PpIX = 5-ALA labeling. Bar = 100 µm. C: Box plot shows the quantification of the average PpIX fluorescence of normal and tumor tissues treated with PBS control or 5-ALA. Each dot represents a slice culture. **p < 0.01, unpaired t-test.

We obtained GBM surgical specimens from 4 patients who received 5-ALA preoperatively (Table S1). For each patient, we collected one tissue specimen with intraoperatively observed PpIX fluorescence from bulk tumor and performed SCOPE-seq2 including single-cell fluorescence imaging and RNA sequencing. For one of the cases (no. 0826), we received a second specimen from a nonfluorescent region of the tumor and performed single-cell fluorescence imaging in our device as a negative control. As expected, we observed that all 4 specimens for which PpIX fluorescence was observed intraoperatively exhibited higher PpIX fluorescence (p < 0.01, Mann-Whitney U-test) by single-cell imaging than the specimen in which PpIX fluorescence was not observed intraoperatively (Fig. 3A). We note that, as anticipated, there was some overlap between the single-cell fluorescence intensity distributions of the PpIX-positive and PpIX-negative samples because the PpIX-positive samples contain a mixture of strongly fluorescent and inefficiently labeled cells. Thus, the microscopic imaging that we performed on individual cells was consistent with what we observed intraoperatively.

We found that nonneoplastic cells within the GBM tumor microenvironment were fluorescently labeled after 5-ALA treatment, but the labeling mechanism of nonneoplastic cells in the tumor microenvironment was unknown. One potential labeling mechanism is that nonneoplastic cells can take up exogeneous 5-ALA and convert it into fluorescent PpIX. Because neoplastic cells can generate and secrete PpIX into the extracellular space after 5-ALA treatment,37,38 another potential labeling mechanism is that nonneoplastic cells take up PpIX secreted by neoplastic cells. As shown above, myeloid cells such as brain-resident microglia are highly fluorescent in GBM after 5-ALA treatment.39 Thus, to determine whether nonneoplastic cells can take up 5-ALA and convert it into PpIX directly like neoplastic cells, we measured the PpIX fluorescence of the 5-ALA–treated human microglial cell line HMC3 and human GBM neurosphere cell line TS543. We observed that both neoplastic glioma cells and nonneoplastic myeloid cells had significantly increased PpIX fluorescence after 5-ALA treatment (p < 0.001, t-test; Fig. 4A and B, Table S3). To determine whether cells can take up PpIX secreted by neoplastic cells, we measured PpIX fluorescence of untreated TS543 or HMC3 cells after incubation with 5-ALA–treated TS543 cells. To differentiate untreated cells from 5-ALA–treated cells in images, we labeled untreated cells with live stain Calcein AM before incubation. To ensure that the observed PpIX fluorescence in 5-ALA–untreated cells was not induced by 5-ALA directly, we washed 5-ALA–treated cells thoroughly with PBS to remove exogeneous 5-ALA and resuspended the treated cells in fresh medium before incubation with 5-ALA–untreated cells, and we also measured the PpIX fluorescence of 5-ALA–untreated cells incubated with the resuspension supernatant. We found that both GBM cells and nonneoplastic microglia had significantly increased PpIX fluorescence after incubation with 5-ALA–labeled GBM cells (p < 0.001, t-test; Table S4) but not after incubation with the supernatant (Fig. 4C and D).

Although 5-ALA labels both neoplastic and nonneoplastic cells in GBM, 5-ALA is reported to preferentially label tumor tissue.9 While this preference could arise from specific labeling of neoplastic glioma cells, our SCOPE-seq2 experiments eliminated this possibility. Another possibility is that tumor tissue-selective labeling is due to BBB disruption by the tumor, making 5-ALA more accessible to tumor tissue compared with nonneoplastic brain tissue. A third possibility is that metabolic alterations in the GBM microenvironment could result in PpIX accumulation. To remove the effect of the BBB during 5-ALA labeling, we generated acute brain slice cultures from mice injected with GL261 cells—which exhibit comparably increased PpIX fluorescence to human GBM neuroshperes (TS543) after 5-ALA treatment (Fig. S5)—and from control mice. GL261 injection resulted in the formation of syngeneic brain tumors, and we treated the slice cultures from the tumor-bearing and control mice with 5-ALA ex vivo (Fig. 5A). After 5-ALA treatment, we generated thin sections for each slice and measured the tissue average PpIX fluorescence using confocal microscopy. We then used H&E-stained adjacent sections to differentiate tumor and normal tissues. We observed that tumor tissues in GL261 mice showed significantly increased average PpIX fluorescence across the imaging field after 5-ALA treatment (p < 0.01, t-test; Fig. 5B and C, Table S5), while the average PpIX fluorescence of normal brain tissues in both GL261 and control mice did not differ significantly. We also observed that tumor tissues had higher average PpIX fluorescence than normal tissues after 5-ALA treatment in GL261 mice (t-test, p < 0.01), indicating that 5-ALA selectively labels GBM tumor tissue at the bulk level even given uniform access to the tumor and normal brain. At the tumor margin, we found that tissue regions inside and outside the tumor have increased average PpIX fluorescence after 5-ALA treatment, but the fluorescence intensity inside the tumor is higher than outside the tumor (Fig. S6, Table S6).

In this study, we sought to determine the specificity of fluorogenic labeling by 5-ALA in patients with high-grade glioma using single-cell resolution. While prior studies have reported the predictive value of 5-ALA labeling for identifying tumor tissue,7,9,18–20 the mechanism underlying this apparent specificity is unclear. One possibility is that neoplastic glioma cells themselves are specifically labeled by 5-ALA due to unique metabolic properties. Another possibility is that the tumor microenvironment as a whole shares metabolic alterations that are conducive to labeling or that certain cells specifically metabolize 5-ALA and others take up the resulting PpIX secreted by these cells. Less sophisticated FGS agents such as fluorescein preferably label the extracellular space in regions of BBB disruption and may be able to label the tumor margin where the BBB is less disrupted through diffusion, although the labeling efficiency at the tumor margin is unclear. Thus, although 5-ALA is clearly producing intracellular PpIX, BBB breakdown could also facilitate delivery of 5-ALA to glioma tissue. However, our results with the GL261 slice cultures show that selective labeling of tumor tissue is not dependent on BBB breakdown.

the labeling efficiency at the tumor margin is unclear. Thus, although 5-ALA is clearly producing intracellular PpIX, BBB breakdown could also facilitate delivery of 5-ALA to glioma tissue. However, our results with the GL261 slice cultures show that selective labeling of tumor tissue is not dependent on BBB breakdown. By combining scRNA-seq and live cell imaging with SCOPE-seq2, we learned that many cell types, including nonneoplastic cells in the glioma microenvironment, were brightly labeled by PpIX in glioma surgical specimens from patients who received 5-ALA. In fact, 5-ALA treatment does not even preferentially label neoplastic glioma cells in some tumors, and myeloid cells can be significantly brighter. As tumor-associated nonneoplastic cells such as myeloid cells are part of the glioma tumor microenvironment, labeling both neoplastic and nonneoplastic cells within the tumor microenvironment could be helpful in distinguishing tumor versus nontumor tissue in FGS. Our studies in acute slice cultures of a glioma mouse model and normal brain tissue showed that glioma tissue was specifically labeled by 5-ALA. In this experimental paradigm, differences in BBB status between tumor tissue and normal brain are irrelevant, because we administered 5-ALA directly to tissue slices. Further experiments in cultured human microglia and patient-derived glioma neurospheres showed that both cell types can convert 5-ALA into PpIX and that PpIX can be secreted and taken up by unlabeled cells. These results suggest that the apparent specificity of 5-ALA for glioma tissue is not due to BBB disruption and that both neoplastic and nonneoplastic cells in the tumor microenvironment can likely metabolize 5-ALA into PpIX. Furthermore, glioma cells can likely secrete PpIX, which can then be taken up by both unlabeled glioma cells and microglia.

e apparent specificity of 5-ALA for glioma tissue is not due to BBB disruption and that both neoplastic and nonneoplastic cells in the tumor microenvironment can likely metabolize 5-ALA into PpIX. Furthermore, glioma cells can likely secrete PpIX, which can then be taken up by both unlabeled glioma cells and microglia. Previous studies have reported that the metabolic status of a cell could affect intracellular PpIX accumulation;8,11–14 thus, we might have expected to see consistent enrichment of metabolic pathways associated with PpIX fluorescence intensity in neoplastic glioma cells. One possible reason that we do not find any consistently enriched metabolic pathway in our analysis is that there are multiple possible mechanisms by which a tumor cell could become labeled, including uptake of PpIX secreted by other cells in the microenvironment.

While previous studies have reported the utility of 5-ALA in FGS,7,9 our results demonstrate that the PpIX fluorescence arising from 5-ALA treatment does not originate exclusively from neoplastic glioma cells. Furthermore, although 5-ALA is typically considered to be an intracellular labeling method, there could be substantial extracellular PpIX secreted by multiple cell types in and around a tumor. This finding raises the possibility that 5-ALA could spuriously label nonneoplastic cells in the glioma margins, where labeling specificity is most critical. Thus, the major advantage of 5-ALA for FGS over inexpensive, simpler alternatives such as fluorescein is unclear, and future studies should systematically evaluate the performance of these two labeling approaches, preferably by simultaneous treatment of the same patients.

P.A.S. was funded by grant no. 75N910019C00029 from the NIH/National Cancer Institute (NCI). P.A.S., P.C., and J.N.B. were funded by grant no. R01NS103473 from the NIH/National Institute of Neurological Disorders and Stroke. This research was funded in part through the NIH/NCI Cancer Center support grant no. P30CA013696 and used the Genomics and High Throughput Screening Shared Resource, Molecular Pathology Shared Resource, and the Confocal and Specialized Microscopy Shared Resource of the Herbert Irving Comprehensive Cancer Center at Columbia University.

Ms. Liu reported a patent for WO/2020/264387 pending. Dr. Sims reported personal fees from Guardant Health outside the submitted work; in addition, Dr. Sims has a patent for WO/2020/264387 pending, a patent for WO/2016/191533A1 pending, and a patent for WO/2019/104337A1 pending.

Conception and design: Sims, Liu, Argenziano, Banu, Bruce, Canoll. Acquisition of data: Liu, Mela, Argenziano, Banu, Furnari, Kotidis, Sperring, Mahajan, Bruce. Analysis and interpretation of data: Sims, Liu, Mela, Banu, Bruce. Drafting the article: Sims, Liu, Banu, Canoll. Critically revising the article: Sims, Mela, Argenziano, Bruce. Reviewed submitted version of manuscript: Sims, Mela, Argenziano, Banu, Sperring, Canoll. Approved the final version of the manuscript on behalf of all authors: Sims. Statistical analysis: Sims, Liu. Administrative/technical/material support: Sperring, Humala, Bruce. Study supervision: Sims, Bruce.

Supplemental material is available with the online version of the article.Supplementary Materials. https://thejns.org/doi/suppl/10.3171/2023.7.JNS23122. Supplementary Materials. https://thejns.org/doi/suppl/10.3171/2023.7.JNS23122.

The sequencing data and count matrices reported in this paper are available in the Gene Expression Omnibus GSE218331 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE218331). Code is available at https://github.com/simslab/SCOPEseq2 and https://github.com/simslab/cluster_diffex2018.