PRMT5 Maintains Tumor Stem Cells to Promote Pediatric High-Grade Glioma Tumorigenesis

Abstract

Pediatric high-grade gliomas (PHGG) are aggressive, undifferentiated central nervous system tumors with poor outcomes, for which no standard-of-care drug therapy currently exists. Through a knockdown (KD) screen for epigenetic regulators, we identified PRMT5 as essential for PHGG cell growth. We hypothesized that, similar to its effect in normal cells, PRMT5 promotes self-renewal of stem-like PHGG tumor-initiating cells essential for tumor growth. We conducted in vitro analyses, including limiting dilution studies of self-renewal, to determine the phenotypic effects of PRMT5 KD. We performed chromatin immunoprecipitation sequencing (ChIP-Seq) to identify PRMT5-mediated epigenetic changes and performed gene set enrichment analysis to identify pathways that PRMT5 regulates. Using an orthotopic xenograft model of PHGG, we tracked survival and histologic characteristics resulting from PRMT5 KD or administration of a PRMT5 inhibitor ± radiation therapy. In vitro, PRMT5 KD slowed cell-cycle progression, tumor growth and self-renewal, and altered chromatin occupancy at genes associated with differentiation, tumor formation, and growth. In vivo, PRMT5 KD increased survival and reduced tumor aggressiveness; however, pharmacologic inhibition of PRMT5 with or without radiation therapy did not improve survival. PRMT5 KD epigenetically reduced tumor-initiating cells’ self-renewal, leading to increased survival in preclinical models. Pharmacologic inhibition of PRMT5 enzymatic activity may have failed in vivo due to insufficient reduction of PRMT5 activity by chemical inhibition, or this failure may suggest that nonenzymatic activities of PRMT5 are more relevant.

Implications: PRMT5 maintains and promotes the growth of stem-like cells that initiate and drive tumorigenesis in PHGG.

Introduction

As a group, pediatric high-grade glioma (PHGG) is the deadliest childhood tumor (1). The 5-year survival rates of PHGG range from 2% to 20% depending on the subtype. PHGG is highly invasive and grows diffusely among normal cells, which limits surgery as a therapeutic option. Radiation therapy (RT) is transiently effective, but tumors almost always recur. Despite hundreds of clinical trials of cytotoxic and targeted chemotherapy, RT remains the only accepted post-surgery standard-of-care therapy.

Epigenetic regulation plays an important role in PHGG tumorigenesis. The two diffuse PHGG subtypes are defined by epigenetic alterations. In diffuse midline glioma (DMG), histone 3 lysine 27 is altered to methionine (H3K27M); diffuse hemispheric glioma has a defining mutation of H3 glycine 34 to arginine or valine (H3G34R/V; refs. 2, 3). These histone mutations prevent post-translational modifications, such as methylation, at the affected site. In DMG, for example, the H3K27M alteration prevents trimethylation of H3K27 (H3K27me3), resulting in the failure to silence expression of many genes and producing a broad group of oncogenic changes in transcriptional regulation (4). PHGG also includes a nonmutant histone 3 subtype that, despite being wild-type for histone variants (H3-wt), may nonetheless rely on epigenetic regulation for oncogenesis and tumor growth (5, 6).

We conducted a short-hairpin RNA (shRNA)-based screen of known epigenetic regulators in H3K27M (7) and H3-wt PHGG cell lines to identify genes that are important for tumor cell survival. The screening identified PRMT5 as essential to PHGG cell growth. The protein product of PRMT5 methylates the two NH₂ groups of arginine’s guanidino functional group (known as symmetric dimethylation). Methylation by PRMT5 occurs throughout the proteome. Its targets include both histone and nonhistone proteins (8). During normal mammalian development, PRMT5 maintains pluripotency of stem cell populations by upregulating expression of stemness genes and repressing genes associated with differentiation (9). Stem cell differentiation is accompanied by downregulation of PRMT5 expression (9). PRMT5 regulates transcription by methylating histone 3 arginine residues H3R2 and H3R8 (10, 11). This inhibits formation of H3K27me3, which produces a permissive environment for transcription of nearby genes (10). PRMT5’s arginine methyltransferase activity thus promotes expression of self-renewal genes otherwise silenced by H3K27me3 occupancy (10). PRMT5 also indirectly regulates the transcription-activating mark H3K4me3. In this study, PRMT5’s effect is context-dependent and can increase or decrease H3K4me3 levels (12, 13).

In cancers that depend on stem-like tumor cells for growth, PRMT5 has emerged as an important oncogene. PRMT5 expression correlates with increased tumor aggressiveness (grade) in adult HGG and promotes its cell growth in vitro (14). PRMT5 also maintains self-renewal of tumor-initiating cells and promotes tumor growth in preclinical models of breast cancer (11). In leukemia cell lines, PRMT5 promotes cell-cycle progression through crosstalk-mediated inhibition of H3K27me3, resulting in expression of genes that would otherwise be transcriptionally silent (10).

PHGG has a stem-like phenotype, suggesting that it originates from stem-like tumor-initiating cells (TIC) (5, 15–21). TICs are self-renewing and form progeny that divide rapidly and uncontrollably, forming highly invasive tumors (15, 20, 22). They are essential for tumor growth, resist current therapeutics, and may be a primary cause of treatment failure (16, 20, 23–25). We hypothesized that PRMT5 promotes PHGG oncogenesis by promoting TIC formation and growth.

We investigated PRMT5’s role in PHGG, both in vitro and in vivo. PRMT5 regulates cell self-renewal in vitro and its genetic depletion suppresses PHGG xenograft growth in vivo. Chromatin immunoprecipitation (ChIP-seq) analysis showed that PRMT5 mediates epigenetic alterations to H3K4me3 and H3K27me3 occupancy at genes important to maintain stem-like and oncogenic characteristics.

Materials and Methods

Cell culture

PHGG cell lines BT-245, SU-DIPG-IV (DIPG4), HSJD-DIPG-7 (DIPG7), SU-DIPG-XIII* (DIPG13), and HSJD-GBM-001 (GBM1) were cultured in suspension in low-attachment 250 mL/75 cm² flasks (Greiner #658; Table 1). BT-245 was a gift from Keith Ligon (Harvard University, Boston, MA); DIPG4 and DIPG13 were gifted by Michelle Monje (Stanford University, Stanford, CA); DIPG7 and HSJD-GBM-001 were gifted by Angel Montero Carcaboso (Hospital Sant Joan de Deu, Barcelona, Spain). Cell lines were authenticated using short tandem repeat testing conducted at least one time per year throughout the course of the study; Mycoplasma testing was also performed at least once per year throughout the course of the study using e-Myco PLUS PCR Mycoplasma Detection Kit. The medium for DIPG4, DIPG7, DIPG13, and GBM1 consisted of equal parts of DMEM/F12 (Gibco) and neurobasal A (Gibco), 2% of B-27 growth supplement without vitamin A (Gibco), 1% each of L-glutamine (Gibco), 2-[4-(2-hydroxyethyl)piperazin-1-yl]ethanesulfonic acid (HEPES; Gibco), nonessential amino acids (Gibco), sodium pyruvate (Gibco), and antibiotic/antimycotic (Gibco), 0.1% heparin (StemCell Technologies), and EGF, FGF, and platelet-derived growth factor (PDGF) A/B at 20 ng/μL (Shenandoah Biotechnology). The medium for BT-245 consisted of NeuroCult with proliferation supplement (StemCell Technologies), 1% penicillin-streptomycin antibiotic (Gibco), 0.1% heparin (StemCell Technologies), and EGF, FGF, and PDGF A/B at 20 ng/μL (Shenandoah Biotechnology). The cells were passaged as required by trituration to break up neurospheres and for replacement of the medium.

Table 1.

Cell line characteristics.

Name	Type	Origin	Genetics	PDX
BT-245	Thalamic DMG	Biopsy	H3.3K27M TP53 mutant	Y
SU-DIPG-IV (DIPG4)	Pontine DMG	Autopsy	H3.1K27M ACVR1 G328V	N
HSJD-DIPG-7 (DIPG7)	Pontine DMG	Autopsy	H3.3K27M ACVR1 R206H	N
VUMC-DIPG10 (DIPG10)	Pontine PHGG	Autopsy	H3K27 wt NF1 Q209*	N
SU-DIPG-XIII (DIPG13)	Pontine DMG	Autopsy	H3.3K27M	N
HSJD-GBM-001 (GBM1)	Cortical PHGG	Resection		Y

PRMT5 KD

We obtained three shRNAs targeting PRMT5 in a lentiviral format (TRCN0000379612, TRCN0000303446, and TRCN0000381130) and an empty vector control (SHC202) from the Functional Genomics Facility at the University of Colorado Anschutz Medical Campus. Cells to be transduced were plated in 24-well ultra-low–attachment plates (Corning, #3473) at 100,000 cells in 2 mL of medium. Polybrene (1 μg/mL) was added to the cell suspension and, 20 to 30 minutes later, a medium was added containing lentiviral shRNA particles at a multiplicity of infection of 0.25 to 0.33. Following 4 to 6 hours of incubation, the medium was changed and the cells were incubated for 48 to 72 hours. After 72 hours, puromycin (1 μg/mL) was added. Cells were selected for one or more passages prior to beginning the experiments.

shRNA screen

We conducted an shRNA screen in two cell lines, SU-DIPG-IV, derived post-autopsy from a pontine tumor with H3.1K27M and ACVR R328V mutations (DIPG4; ref. 7), and VUMC-DIPG-10, derived post-autopsy from an H3K27-wt pontine tumor with an NF1 Q209* variant and MYCN amplification (DIPG10). The shRNA screening library comprised 4,139 shRNAs targeting 405 genes involved in the epigenetic regulation of gene expression (Supplementary Materials S1 and S2). Cells were transduced with the barcoded shRNA library at a multiplicity of infection calculated to obtain one shRNA per cell and placed into selection. Samples were collected after 72 hours of selection and 21 days later. The barcodes were amplified and sequenced, and the resulting data were analyzed to identify changes in barcode frequency in the samples.

Cell growth assay

Cells previously transduced with an empty vector control or shRNA targeting PRMT5 were triturated to obtain a single-cell suspension and placed into 24-well plates (100,000 cells per well). In long-term longitudinal assays, the cells were triturated into single cells and counted once or twice per week for up to 5 weeks. Following each counting, 100,000 cells were replated. In short-term longitudinal assays, 10,000 cells transduced with an empty vector control or one of three PRMT5 KD constructs were plated into 96-well plates in triplicate. The cells were then incubated in Incucyte S3 Live Cell Analysis Instrument (Sartorius) for 6 days, during which imaging was conducted every 4 hours. Image data were analyzed to ascertain cell growth based on a neurosphere proliferation model (Sartorius).

Extreme limiting dilution assay of self-renewal

Cells were triturated to form a single-cell suspension and plated into 100 μL of medium in a 96-well plate at 1 cell per well (20 wells), 2 cells per well (10 wells), 4 cells per well (10 wells), 8 cells per well (10 wells), 16 cells per well (5 wells), and 32 cells per well (5 wells). The cells were incubated for 2 to 3 weeks. Wells with neurospheres were identified using a brightfield microscope that accepts a 96-well plate (Keyence). The results were analyzed as previously described to yield point estimates of mean stem cell frequency and a P value comparing PRMT5 KD to empty vector control cells (26).

Apoptosis assay

Cells were placed into 96-well white wall plates at 2,000 cells per well with three technical replicates per sample. Caspase-Glo 3/7 3D Assay (Promega, #G8981) was added, and the mixture was incubated as per the manufacturer’s instructions. After incubation, luminescence was measured using a Synergy plate reader (BioTek).

Cell-cycle analysis

Cells were fixed in 70% ethanol at 4°C for at least 12 hours. Cells to be analyzed were placed into a 96-well plate at 100,000 cells per well in 200 μL of medium, spun down, rinsed with PBS, and resuspended in Guava Cell Cycle Staining Reagent (Millipore). Data acquisition was performed on a Guava PCA-96 system (Millipore). Data were analyzed using FlowJo (RRID: SCR_008520) to identify cells in G1, S, and G2 cell-cycle phases.

Western blot

Cells were lysed using RIPA buffer containing 1× Halt protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, 1861281) and gentle vortexing for 5 seconds. Cell lysates were then placed on ice for 5 minutes, followed by centrifugation at 13,000 rpm for 10 minutes at 4°C. Protein concentration was determined using BCA Protein Assay Kit (Pierce, 23227). Protein samples were reduced and denatured using Laemmli 4× SDS sample buffer (Boston BioProducts, BP-110R), heated at 95°C for 10 minutes, loaded onto precast Mini-Protean TGX Gels (Bio-Rad, 456-1093), and electrophoresed at 125 V for approximately 1 hour. Protein was wet-transferred from the gels onto polyvinylidene difluoride membranes (Bio-Rad, 162-0115) by applying a voltage of 75 V to the membranes for 90 minutes at 4°C. A solution of 4% to 5% BSA (Fisher BioReagents, 1605-100) was prepared in 1× TBS (Bio-Rad, 170-6435) and 0.1% Tween-20 (Fisher BioReagents, BP337-500) and used to block the membranes for 20 minutes. Following blocking, the membranes were incubated in primary antibodies overnight at 4°C using the following antibodies and dilutions: PRMT5 [Cell Signaling Technology, #79998 rabbit mAb (1:1,000)]; β-actin [Millipore, #MAB1501 mouse mAb (1:1,000)]; β-tubulin [Cell Signaling Technology, #79998 rabbit mAb (1:1,000)]; and symmetric dimethyl arginine (SDMA) [Cell Signaling Technology, #13222 rabbit mAb mix (1:1,000)]. The membranes were then incubated in horseradish peroxidase–conjugated secondary antibodies (Cell Signaling Technology, 7074S or 7076p2) for 1 hour at room temperature. The membranes were developed using Luminata Forte Western horseradish peroxidase substrate (Millipore, WBLUF 0500) and imaged using a G:BOX digital imaging system with GeneSys (version 1.8.8.0) software (Syngene).

Bulk RNA sequencing

We conducted bulk RNA sequencing (RNA-seq) of PRMT5 KD and empty vector control cells. Samples were obtained from BT245, DIPG4, DIPG7, DIPG13, and GBM1 cells and comprised one empty vector control and three KD samples with different shRNA vectors. RNA extracted from 1 × 10⁶ cells was sent to the Anschutz Genomics Shared Resource for sequencing. Sequencing libraries of poly-A RNA were prepared using TruSeq Library Preparation Kit v2 (Agilent). Paired-end sequencing, with 50 million reads per sample, was performed using NovaSeq 6000 (Illumina). Data were mapped to the human genome (GRCh38) using gSNAP. Expression (FPKM) was derived using Cufflinks (RRID: SCR_014597), and differential expression was analyzed with ANOVA in R. FPKM read data were analyzed using gene set enrichment analysis (GSEA) and Metascape (RRID: SCR_016620).

ChIP-seq

ChIP-seq was performed using antibodies against H3 lysine 4 trimethyl [H3K4me3, Cell Signaling Technology #9751S, 1:50], H3 lysine 27 trimethyl (H3K27me3, Cell Signaling Technology, #9733S, 1:50), and H3 lysine 27 acetyl (H3K27ac, Cell Signaling Technology, #8173S, 1:50). We also obtained and sequenced an input sample and a sample immunoprecipitated with rabbit IgG only (Cell Signaling Technology, #3900S, 2.5 μg total antibody per reaction). Approximately 5 to 10 million cells were fixed in 1% formaldehyde, after which cells were lysed using FL lysis buffer (27). Nuclei were sonicated at low energy to fully isolate them from the cytosol. Nuclei were then sonicated at high energy in a Bioruptor Plus (Diagenode). Sonication parameters to produce DNA fragments in the 200 to 600 bp range were optimized beforehand. Samples were incubated with the antibodies overnight at 4°C with rocking. Magnetic beads (CST, #9006) were used for immunoprecipitation. DNA was eluted from the beads using 2× ChIP Elution Buffer (Cell Signaling Technology, #7009) and incubating at 65°C for 30 minutes with shaking. Following elution, DNA was purified using spin columns (Cell Signaling Technology, #14209). DNA was sequenced on a NovaSeq 6000 (Illumina) using 150 paired-end cycle reads with 50 million paired reads per sample. Following filtering, reads were aligned to GRCh38 using BowTie software. BAM files were filtered to remove secondary alignments, improperly paired reads, and alignments with mapping quality <30. Peak calling was performed using MACS3 (28). Peaks were analyzed using DiffBind (3.8.4; ref. 29) and ChIPpeakAnno (3.32.0; ref. 30) in R (4.2.2), followed by differential gene expression, GSEA, and Metascape analyses.

Tumor implantation in mice

All animal experiments were conducted in accordance with an existing Institutional Animal Care andUse Committee protocol administered by the University of Colorado Anschutz Medical Campus. BT-245 (a DMG cell model) cells genetically modified to express a luciferase gene were implanted into the right pons of athymic nude mice (Charles River Laboratories) by intracranial injection using a stereotactic frame (31). Following injection, the mice were monitored daily for evidence of illness, including decreased activity, hunched posture, poor grooming, and ataxia. In addition, the mice were weighed weekly. Animals that were moribund and/or had lost more than 15% of their initial body weight were euthanized for necropsy. Tumor growth was monitored by MRI, as well as bioluminescence imaging for cell lines recombinantly modified by the introduction of a luciferase gene. In experiments involving the injection of PRMT5 KD cells, differences in survival were assessed. In experiments of RT and drug treatment with PRMT5 inhibitor LLY-283 (SelleckChem), survival was evaluated in four experimental arms: vehicle control, RT only, LLY-283 only, and LLY-283 plus RT. The LLY-283 dose was 50 mg/kg of body mass. LLY-283 was given on three consecutive days each week. LLY-283 was administered by oral gavage in vehicle consisting of 1% hydroxyethyl cellulose by weight, 0.25% Tween 80 (Sigma) by volume and 0.05% Pluronic F-68 (Gibco) by volume. RT total dose was 8 Gy given in equal fractions of 2 Gy over four days. We used eight mice per arm in each experiment, which enabled the detection of a 50% survival difference at 80% power with an α of 0.05.

In vitro cell survival experiments

Cells were plated into 96-well plates at 20,000 cells per well in 90 to 100 μL medium with three replicates per condition. In experiments utilizing RT, cells were irradiated using a cesium-137 source. In experiments using the PRMT5 inhibitor LLY-283, drug dissolved in DMSO and diluted in PBS was added to each well at concentrations ranging from 0.316 nmol/L to 10 μmol/L by half-log₁₀ concentration increments. Cells were then incubated for 120 hours. Following incubation, CellTiter 96 Aqueous One Solution Assay (Promega) was added (20 μL per well), and cells were incubated 1 to 4 hours. Plates were then analyzed for absorbance at 490 nmol/L in a 96-well Synergy plate reader (BioTek).

Data availability

Sequencing data were deposited in the NCBI Gene Expression Omnibus (RRID: SCR_005012) database (GSE261512).

Results

PRMT5 is essential for PHGG cell growth

A shRNA KD screen identified epigenetic regulators that are important for tumor cell growth in DMG (7) and H3-wt pontine PHGG cell lines (Fig. 1A and B). Analysis of the DMG screen in conjunction with the previously unpublished H3-wt PHGG screen revealed two epigenetic regulators, PRMT5 and HDAC2, with significant decreased fold changes (FC) in both cell lines (FC < 1; P < 0.05; Fig. 1A and B; Supplementary Table S1). RNA interference (RNAi) data from the Broad Institute’s DEMETER2 screen showed that PRMT5 promotes growth in adult glioma and other central nervous system (CNS) cancers (Fig. 1C; Supplementary Fig. S1A), whereas HDAC2 promotes growth in peripheral nervous system tumors but not CNS tumors. The magnitude of PRMT5’s growth effect did not depend on its initial expression level, suggesting that oncogene addiction is not required for potential therapeutic interventions (Fig. 1D). To exclude the possibility that PRMT5-related growth effects occurred because it is essential for the survival of all cells, we compared DEPMAP CRISPR knockout (KO) data (Broad Institute) with DEMETER2 RNAi KD screening results (Fig. 1E; Supplementary Fig. S1B) in cancer and noncancer cell types (32). In KO experiments, both tumor and nontumor cells having no PRMT5 expression died, suggesting the absence of a therapeutic window from PRMT5 KO. In KD experiments, normal cells survived PRMT5 KD, but cancer cells reliant on PRMT5 expression did not. We verified that targeting PRMT5 with shRNA reduced, but did not eliminate, its gene and protein expressions in PHGG cells and that shRNA-mediated KD acted on target (Fig. 1F–H). A functional assay showed decreased symmetric dimethylation of arginine residues (PRMT5’s sole enzymatic activity and target) as a result of PRMT5 KD (Fig. 1I). Symmetric dimethylation also decreased when PHGG cells were treated with the clinical-grade PRMT5 inhibitor LLY-283 (Fig. 1I).

Figure 1.

PRMT5 is essential for growth of PHGG and other CNS cancer cells. A, shRNA KD screen of epigenetic regulators identified those that are important for cell growth. The Venn diagram shows numbers of genes important for tumor cell survival in DIPG4 (H3K27M DMG cell line), DIPG10 (H3-wt pontine PHGG), and overlap between the two cell types. B, shRNA screen results showing genes important for cell growth in DIPG4 (left) and DIPG10 (right); significance thresholds in A and B are FC < 0.6, P < 0.025 for DIPG4 cells and FC < 0.75, P < 0.025 for DIPG10 cells. C, In RNAi KD screens, CNS cancer and glioma cell lines depend on PRMT5 to proliferate, validating PRMT5 as a potential therapeutic target in these cancer types. D, Scatter plot of PRMT5 expression vs. CRISPR and RNAi gene effects for individual cell lines assayed by the Broad Institute, both CRISPR and RNAi gene effects are independent of endogenous PRMT5 expression level (trend line slope is not significantly different from 0). E, Summary of Broad Institute CRISPR and RNAi screening data for PRMT5; shaded areas represent the proportion of cell lines for each measured gene effect; DepMap CRISPR KO screening shows that PRMT5 is essential for cell survival at a level similar to housekeeping genes in 97.5% (1,001/1,027) of cell lines tested, suggesting that PRMT5 KO is lethal to cancerous and noncancerous cell types; in the RNAi KD screen, PRMT5 KD is “strongly selective” among different cell types, suggesting that it may slow tumor growth without adversely affecting normal tissue. F,PRMT5 KD effect on PRMT5 expression in cell models of PHGG. Table 1 describes the cell lines. G, Rescue experiment in BT245 cells shows that PRMT5 KD targeting 3′-untranslated region is rescued by transfection of PRMT5 cDNA. H, Western blots in DMG (DIPG7) and PHGG (GBM1) cell lines illustrate effect on PRMT5 protein levels of PRMT5 KD. Densitometry results representing PRMT5/H3 (top) and PRMT5/β-tubulin (bottom) are annotated onto blots to facilitate comparison of expression levels. I, Left: Western blot of symmetric dimethylation of arginine residues throughout the proteome (SDMA, multiple bands) illustrates in BT245 cells the effect of PRMT5 KD on PRMT5 enzymatic activity; right: Western blot of SDMA after LLY283 treatment illustrates the effect of PRMT5 inhibition on SDMA; 15 kD bands of SDMA are histone proteins, densitometry-annotated as in H. NS, not significant; SDMA, symmetric dimethyl arginine.

PRMT5 KD reduces self-renewal of PHGG stem-like cells, reduces proliferation and cell-cycle progression, and increases apoptosis

We evaluated the effects of PRMT5 KD and PRMT5 inhibition using LLY283 on PHGG cells. In in vitro extreme limiting dilution assays, PRMT5 KD and LLY283 treatment both decreased the tendency of PHGG cells to form neurospheres, from which we calculated that PRMT5 KD or inhibition decreased the frequency of self-renewing cells by a factor of 6.6 or more (P < 7e−9; Fig. 2A and B). We interrogated a single-cell RNA-seq dataset of PHGG patient samples and identified a likely TIC population that expressed NES, SOX2, and OLIG2, which are also expressed in neural stem cells and oligodendrocyte progenitors (Supplementary Fig. S2A and S2B). PRMT5 KD decreased PHGG cell growth in both long-term and short-term experiments (Fig. 2C and D). The effect of PRMT5 inhibition using LLY283 on PHGG cells was dose-dependent, but at least 50% of cells remained viable at the highest dosage levels (Fig. 2E; Supplementary Fig. S2C). LLY283 showed no effect on survival when applied to a normal human astrocyte cell line, suggesting that a therapeutic window may exist in PHGG (Fig. 2E; Supplementary Fig. S2C). PRMT5 KD inhibited cell-cycle progression, increasing the number of cells in the G1 phase and decreasing the number of cells in the S or G2 phases (Fig. 2F). PRMT5 KD also increased apoptosis by 4% to 5% in in vitro experiments (Fig. 2G).

Figure 2.

PRMT5 KD decreases self-renewal, inhibits cell growth and cell-cycle progression, and increases apoptosis vs. the empty vector control. A, In extreme limiting dilution assay, PRMT5 KD decreases stem cell frequency (measure of self-renewal) in DMG cell lines. B, PRMT5 inhibition using LLY283 decreases stem cell frequency in DMG cell lines. LLY283 concentration was selected as the greatest concentration shown in Fig. 1E and Supplemeantary Fig. S2C, at which a substantial percentage of cells survived PRMT5 inhibition; (C) PRMT5 KD results in decreased cell growth in DMG cell lines. D,PRMT5 KD results in decreased cell growth in the PHGG cell line, GBM1; (E) PRMT5 inhibition using LLY283 has a dose-dependent effect on survival of BT245 DMG and GBM1 PHGG cells but leaves a large resistant population. LLY283 had no observable effect on normal human astrocytes (N = 3 for all data points, error bars show the SEM, and error bars not shown were too close to the central estimate to display). F,PRMT5 KD slows cell-cycle progression by increasing cells in the G1 phase and decreasing cells in the G2 phase. G,PRMT5 KD increases cell susceptibility to apoptosis. NHA, normal human astrocyte.

The overall effects of PRMT5 KD/inhibition in vitro thus included decreased PHGG TIC frequency, stem cell gene expression, and cell growth. PRMT5 KD also inhibited PHGG cell-cycle progression and increased apoptosis. These effects are consistent with the hypothesis that PRMT5 maintains cell self-renewal in PHGG stem-like cells.

PRMT5 KD increased expression of genes associated with stem cell maintenance and oncogenesis

We conducted bulk RNA-seq on PRMT5 KD versus control cells. Data were analyzed by taking the two KD samples with the lowest expression levels of PRMT5 and comparing them with the empty vector control for each cell line. Differentially expressed genes as a result of PRMT5 KD included several genes involved in stem cell maintenance (PAX3, CDH4, KIF1A, and UCN2) as well as oncogenesis (CDH4, MMP14, and ARPC1B; Fig. 3A; Supplementary Table S2). We conducted pathway analyses using GSEA and Metascape (33–35). PRMT5 KD had both disparate and common pathway effects in H3K27M DMG versus H3-wt PHGG samples. DMG was depleted in cell growth (mitotic spindle and E2F) and inflammatory pathways (TNFα/NF-κB, IFNα and IFNγ), whereas H3-wt PHGG was depleted in gene sets of core stem cell genes (Fig. 3B). Both DMG and H3-wt PHGG cells with PRMT5 KD were enriched in DNA repair pathway expression and depleted in genes expressed during hypoxia and epithelial–mesenchymal transition (EMT; Fig. 3B and C). Metascape analysis was consistent with the GSEA results (Supplementary Fig. S3A and S3B).

Figure 3.

PRMT5 KD decreases expression of genes important to stem cell growth and decreases expression of important cancer pathways. A, Differential expression analysis following PRMT5 KD in four DMG and one PHGG cell lines; the genes shown are 6 of the 11 most highly differentially regulated genes in the assay (N = 3 for all data points, and error bars not shown were too close to the central estimate to display). B, Dot plot of GSEA NES and −log(FDR) following PRMT5 KD. The DMG results represent the aggregate of four cell lines (BT245, DIPG4, DIPG7, and DIPG13), and NES reflects the magnitude of depletion. C, GSEA NES values for gene sets important for PHGG inception and growth following PRMT5 KD (FDR <0.05 for all NES values). NES, normalized enrichment score.

Similar to its phenotypic effects, PRMT5 KD downregulated pathways involved in tumor growth and EMT in both DMG and H3-wt PHGG. PRMT5 KD also downregulated stem cell maintenance pathways that are epigenetically regulated in H3-wt PHGG only, suggesting that the H3K27M alteration in DMG may interfere with this regulation at the pathway level. Notably, however, individual genes involved in stem cell regulation, such as PAX3, CDH4, and KIF1A, were downregulated in both DMG and H3-wt PHGG.

In vivo studies of PRMT5 KD in PHGG showed increased survival and decreased tumor aggressiveness

We orthotopically implanted mice with PRMT5 KD cells and empty vector control cells from a DMG cell line (BT245) and conducted a survival study. PRMT5 KD mice survived significantly longer than control mice; however, all mice eventually died of tumor-related effects (Fig. 4A). MRI and bioluminescence imaging confirmed that the PRMT5 KD tumors were less aggressive than the control tumors (Fig. 4B and C). Hematoxylin and eosin staining showed that the tumor extent and cell density were lower in PRMT5 KD cells (Fig. 4D). Staining for Ki-67 verified that the frequency of proliferating cells was lower in PRMT5 KD cells than in control tumor cells (Fig. 4E).

Figure 4.

PRMT5 KD produces less aggressive tumors with greater survival in a mouse PDX model of PHGG. A, Kaplan–Meier survival curve showing that mice orthotopically injected with BT245 patient-derived tumor cells with PRMT5 KD survive 2.5–3 times longer than mice injected with empty vector control tumor cells, n = 8 mice per arm. B, MRI images of empty vector (control) and PRMT5 KD tumors; arrowheads show enhancement of tumor areas. C, Bioluminescence images showing tumor progression. D, Histologic sections with H&E staining showing representative tumors from PRMT5 KD and empty vector control mice. E, Ki-67 staining (top) and quantification showing a greater number of proliferating cells in PRMT5 KD vs. empty vector control tumors (P < 0.05). F, In vitro, PHGG cells with PRMT5 KD are more susceptible to RT than empty vector control cells (N = 3 for all data points, and error bars not shown were too close to the central estimate to display). G, Kaplan–Meier survival curve showing no significant differences in survival of mice treated with the PRMT5 inhibitor vs. vehicle control with and without RT (BT245 PDX model). EV, empty vector; H&E, hematoxylin and eosin; Veh, vehicle.

We also conducted a mouse survival study with BT245 cells using PRMT5 inhibitor LLY283 and the current standard-of-care therapy, RT. In vitro experiments combining RT with PRMT5 KD showed 17.1% to 30.3% lower survival rates when PRMT5 KD versus control cells were exposed to RT (Fig. 4F). These results suggested that PRMT5 depletion sensitizes cells to the effects of RT. Because a large fraction of tumor cells survived in vitro treatment with LLY-283 alone (Fig. 2E), we hypothesized that combining PRMT5 inhibition with RT might lead to increased cell death but that LLY283 alone would be ineffective. We further hypothesized that applying RT as soon as tumor engraftment could be verified would produce a population of primarily TICs with few differentiated tumor cells, maximizing the possibility of successfully treating with LLY283. We found, however, that LLY283, with or without RT, did not significantly increase survival in the BT245 PDX model (Fig. 4G; Supplementary Fig. S4A). To assess whether LLY283 successfully reached the tumor tissue and inhibited PRMT5 activity, we took viable cells from the cortex and tumors of mice treated with vehicle or LLY283 in the LLY283/RT experiment. A Western blot for SDMA levels in the treated and untreated samples verified that the normal cortex and tumor cells both absorbed LLY283 (Supplementary Fig. S4B). The observed inhibitory effect was greater in the normal cortex than in the tumor tissue, possibly suggesting resistance to treatment in the tumor tissue, though we note that the cortical and tumor tissues were mouse and human samples, respectively.

PRMT5 KD in our mouse PDX experiment had effects similar to those observed in vitro. However, the combination of RT and PRMT5 inhibition did not increase the survival rate. The reasons and potential solutions for the treatment failure are further addressed in the “Discussion section”.

PRMT5 KD alters chromatin occupancy at important epigenetic regulatory sites

We performed ChIP-seq in the DMG cell line used in the mouse patient derived xenograft (PDX) experiments (BT245) and the H3-wt PHGG cell line, GBM1. We immunoprecipitated H3K4me3, H3K27ac, and H3K27me3, performed high-throughput sequencing of the extracted DNA, and analyzed the changes in chromatin occupancy (Fig. 5; Supplementary Figs. S5–S9). Overall, in the DMG cell line, PRMT5 KD increased occupancy at the H3K4me3 and H3K27me3 marks (Fig. 5A; Supplementary Fig. S5A–S5C). In H3-wt PHGG cells, PRMT5 KD produced overall decreases in occupancy at both H3K4me3 and H3K27me3 (Fig. 5B; Supplementary Fig. S5D–S5F). PRMT5 KD produced no discernible overall change at the H3K27ac mark in either cell line, though we note that overall occupancy at this mark is greater in H3-wt PHGG than in DMG (Fig. 5A and B). Principal component analysis comparing occupancy changes showed that tumor type [H3K27-mutant (BT245) versus H3K27-wt (GBM1)] explained the greatest amount of occupancy differences at all three histone marks (PC1, 60%–91%), whereas PRMT5 KD was responsible for the next greatest amount of occupancy variation (PC2, 4%–15%; Supplementary Fig. S5G).

Figure 5.

ChIP-seq in PRMT5 KD vs. empty vector control cells shows consistent differences in H3K27me3 and H3K4me3 occupancy. A and B, occupancy levels of 1,000 randomly selected genes from 1,500 bp downstream (−1,500) to 1,500 bp upstream of the transcription start site in PRMT5 KD and empty vector control cells; (A) shows BT245 cells; (B) shows GBM1 cells; (C and D) top: Venn diagram of GSEA results showing the number of TFs whose target genes had significant occupancy changes as the result of PRMT5 KD at H3K4me3 (left) or H3K27me3 (right) histone marks, overlap shows genes with significant occupancy changes at both marks; bottom: list of TFs for which occupancy at target genes changed by histone marks, blue type = depletion, red type = enrichment. Dn, downregulation; TSS, transcription start site; Up, upregulation.

Next, we identified binary changes in occupancy (occupied/unoccupied) by genomic locus for the three ChIP-seq samples (two PRMT5 KD constructs versus one empty vector control; Supplementary Fig. S6A–S6F). To understand the transcriptional effects of PRMT5 KD, we performed GSEA on the ChIP-seq results using the Hallmark, curated (C2), and ontology (C5) gene sets provided in the molecular signature database (MSigdb, Broad Institute), ranking them by change in peak height attributable to PRMT5 KD (33, 34, 36). We identified gene sets with “target” in their names because they typically represent a set of target genes acted upon by a single transcription factor (TF; Fig. 5C and D; Supplementary Tables S3–S6). In both cell lines, occupancy at these “target” sites overwhelmingly decreased with PRMT5 KD at both the H3K4me3 (BT245: 106/106 sites with FDR <0.05; GBM1: 30/31 gene sets with FDR <0.05) and H3K27me3 marks (BT245: 62/70 gene sets with FDR <0.05, GBM1: 90/90 gene sets with FDR <0.05; Fig. 5C and D). The genes whose targets had decreased occupancy at H3K4me3 included TFs involved in early embryonic development or stemness maintenance in highly undifferentiated cells, whereas decreases in H3K27me3 occupancy tended to occur at targets of TFs associated with more differentiated cells (Fig. 5C and D). Notably, H3K4me3 was the primary locus of occupancy changes in DMG whereas H3K27me3 was the primary locus in H3-wt DMG. A few genes important for stem cell maintenance, for example, SOX2 in GBM1 cells, decreased in H3K27me3 occupancy as the result of PRMT5 KD (Fig. 5D). This decrease does not necessarily equate to increased transcription but instead may mark a change in the SOX2 locus being poised for transcriptional silencing as differentiation occurs (37). A total of 17.3% of the targets in BT245 and 10% of targets in GBM1 decreased at both H3K4me3 and H3K27me3 (Fig. 5C and D). Bivalent shifts at these loci may likewise suggest a shift from higher transcriptional levels to a locus poised for silencing upon cellular differentiation (37, 38).

Differential gene occupancy analysis identified several consistent enrichment or depletion differences (Supplementary Fig. S7A). These included genes of the SEMA4 family that are involved in cellular differentiation and have been identified as important to the development of cancer (39); CACUL1, which regulates the G1/S cell-cycle transition (40); PWWP2A, which recruits histone deacetylases to gene promoters and is essential for mitosis (41); SMURF1, a ubiquitin ligase specific to bone morphogenetic proteins (42, 43); and TOP3B, a topoisomerase involved in recombination that has oncogenic characteristics (Supplementary Fig. S7A; ref. 44). Pathways suggested by these occupancy differences included regulation of cellular proteins through destruction (ubiquitination and the clathrin pathway), bone morphogenetic protein regulation, cell division, and transcriptional regulation (DNA repair and RNA polymerase II regulation), cellular processes important for oncogenesis such as differentiation and angiogenesis, and the cellular stress response (heat shock proteins; Supplementary Fig. S7B). We performed an ontologic analysis using Metascape for each of the two cell lines (Supplementary Figs. S8 and S9). The gene ontology terms from the Metascape analysis, which included cellular signaling, ligand/receptor interaction, differentiation, and cell fate commitment, angiogenic signaling, and brain development, were consistent with the GSEA and differential expression analyses (Fig. 5C and D; Supplementary Figs. S8 and S9).

Analysis of the ChIP-seq data suggests that transcriptional regulation through the H3K27me3 and H3K4me3 histone marks is an important aspect of PRMT5’s regulation of stem-like TIC cells important for the development of PHGG. Genes whose occupancy decreased at H3K4me3 with PRMT5 KD were primarily those involved in stem cell maintenance. Occupancy changes at H3K27me3 were consistent with the inception of differentiation. The differential gene expression and Metascape analyses were consistent with the GSEA results in showing that PRMT5 regulates stem cell characteristics in DMG and PHGG through modulation of the H3K4me3 and H3K27me3 chromatin marks. The patterns of occupancy changes between DMG (alterations primarily at H3K4me3) and H3-wt PHGG (alterations primarily at H3K27me3) are consistent with the limitations the H3K27M alteration places upon regulation occurring through H3K27me3 in DMG.

Discussion

Our results show that PRMT5 plays an important role in the development and growth of PHGG through epigenetic control of gene expression in stem-like TICs. PRMT5’s regulation of TIC self-renewal and growth underlies the phenotypic effects of PRMT5 KD in PHGG. Those phenotypic effects include decreased proliferation and cell-cycle progression, increased susceptibility to apoptosis, and decreased self-renewal. Genes downregulated by PRMT5 KD include PAX3, which plays a key role in neural stem cell maintenance, neurogenesis, and astrogenesis (45, 46). GSEA showed that PRMT5 KD downregulated pathways that maintain self-renewal and other stem cell characteristics and increased or established conditions poised for increased expression of genes involved in cellular differentiation. The pathway alterations, however, differed between DMG and H3-wt PHGG. PRMT5 KD also decreased hypoxia and mesenchymal gene expression and increased DNA repair. These changes are associated with decreased tumorigenesis in DMG and H3K27-wt PHGG.

Our ChIP-seq studies offer some clues about the mechanism by which PRMT5 KD inhibits PHGG growth in H3K27-wt and H3K27M cells. The H3K4 and H3K27 methylation sites were relevant because PRMT5, though it encodes an arginine methyltransferase, regulates methylation through a cross-talk–mediated mechanism (10). H3K4me3 occupancy decreased with PRMT5 KD in genes that maintain self-renewal and other stem cell characteristics important to oncogenesis. At the H3K27me3 mark, PRMT5 KD produced greater depletion in H3-wt PHGG than in DMG. The differing chromatin regulation patterns in DMG (H3K27M) and H3-wt PHGG suggest that PRMT5’s regulation is less effective at the (mutant) K27 mark of histone 3 in DMG, as would be expected.

An important question for DMG (H3K27M) cells is the extent to which changes in H3K27me3 affect gene expression. Although the H3K27M mutation is estimated to be less than 20% penetrant among DMG cells, it sequesters the PRC2 complex that is responsible for adding the trimethyl mark at the H3K27 locus. This decreases the number of H3K27me3 sites to a greater extent than the frequency of the H3K27M mark alone would suggest (4). The alteration of occupancy at some H3K27me3 sites that occurs with PRMT5 KD in DMG implies, however, that at least some of PRMT5’s regulation of DMG occurs through the H3K27me3 mark. The nature of the downregulated pathways differed between H3K4me3 and H3K27me3, with the pathways at H3K4me3 comprising primarily stem cell maintenance and those at H3K27me3 comprising primarily cell division and differentiation. Our results suggest that in DMG, PRMT5 regulation at the H3K27me3 mark is subsidiary to the regulation occurring at H3K4me3. In contrast, PRMT5 regulation in H3-wt PHGG is more prevalent at the H3K27me3 site. These differences suggest that although similarities exist in the overall phenotypic changes that PRMT5 KD produces, mechanistic differences in PRMT5 regulation may be important in designing therapies targeting TICs in the respective tumor types.

Our ChIP-seq results included H3K27me3 and H3K4me3 occupancy changes with PRMT5 KD in several multigene pathways, including ubiquitination, the bone morphogenetic protein (BMP), and the clathrin pathways. To date, these pathways have not received much attention in PHGG. Ubiquitination controls the degradation of intracellular proteins, which play key roles in protein regulation. Clathrin-mediated vesicular transport enables intercellular communication as well as intracellular protein transport. BMP signaling regulates early CNS development and patterning and, later, regulates fate specification of CNS cells, including the oligodendroglial-to-neuronal shift (47). BMP signaling is also involved in the shift from neuronal to astrocyte fate toward the end of neurogenesis (47). PDGF signaling, which is an important driver of PHGG growth, is active in these same processes. The epigenetic changes introduced by PRMT5 KD thus highlight previously underappreciated pathways that might be important and deserve further study in PHGG.

PRMT5 KD in our mouse PDX model of DMG showed a survival benefit and decreased aggressiveness compared with empty vector control cells. This result is consistent with our in vitro results as well as the Broad Institute’s screening data. The Broad Institute data imply that a therapeutic window exists for PRMT5 KD, in which a baseline level of expression persists following shRNA treatment, but not for PRMT5 KO, in which ablation of PRMT5 is lethal in all cell types. Our mouse PDX survival curve for PRMT5 KD (Fig. 4A) suggests a long period during which the PRMT5 KD mice were disease-free followed by a rapid survival decline. We attribute this to the initial effect of the KD in interfering with TIC maturation to form tumor cells. We expect the KD effect to dissipate over time as tumor cells adapt or cells without KD become dominant based upon their faster growth. At the endpoint, the Ki-67 results showed that the KD tumor had a 40% to 60% proliferation rate observed as compared with the control proliferative rate of about 80%. This is consistent with the difference in the slopes of the Kaplan–Meier survival curves, which are steeper for the controls than for the KD population.

PRMT5 inhibition with a clinical-grade drug failed to show survival or phenotypic differences in mice. This was true whether the inhibitor was administered alone or in conjunction with RT. Another group also recently reported that PRMT5 inhibition, given without RT, failed to confer a survival benefit (48). We verified that the inhibitor reached the tumor tissue and had an inhibitory effect. This suggests that PRMT5 inhibition, or the level of PRMT5 inhibition that we were able to induce, was ineffective in altering tumor trajectory. Several possible reasons may exist for this. First, as noted, the degree to which PRMT5 activity was inhibited may have been too small to slow tumor growth. Second, the timing of treatment may have been incorrect. Finally, in vivo tumor dynamics may diverge from the in vitro models in a manner that renders PRMT5 inhibition ineffective in vivo. These possibilities are further explored below.

Our rationale for applying RT prior to PRMT5 inhibition was to destroy all except tumor-initiating cells, enabling PRMT5 inhibition to target cells from which the tumor can grow or recur. This strategy may have failed because RT either did not completely destroy the proliferating tumor cells or because they regrew to a sufficient extent to power tumor growth before inhibitor treatment was started. It is also possible that RT altered the TICs in a way that rendered them less dependent on PRMT5. Our data suggest this possibility. Specifically, our GSEA results found that PRMT KD downregulated hypoxia and EMT pathways. RT’s effects include upregulation of EMT and hypoxia. It is thus possible that RT and PRMT5 KD acted antagonistically to one another. The inhibitor treatment may also have failed if PRMT5 acts nonenzymatically in PHGG, for example, as a scaffold protein, to drive tumor growth.

Based upon our results, it is unclear whether there is a path forward to pursue PRMT5 as a therapeutic target in PHGG. We feel, however, that several ideas may be worth exploring. RNAi screening showed that initial levels of PRMT5 expression do not affect the efficacy of PRMT5 KD. The phenotypic effects of PRMT5 KD may thus depend on reducing PRMT5 levels to a fixed level. Because of the success of PRMT5 KD, we suggest that a more robust method of inhibition that approaches the expression levels achieved with KD could be effective. One approach might be to combine PRMT5 inhibition with another means of regulating its activity. One possibility is to reduce the availability of the PRMT5 substrate S-adenosyl methionine, which is required for PRMT5 activity. This approach has been attempted successfully with other epigenetic regulatory drugs in preliminary preclinical experiments (49). An alternative approach suggested by the success of shRNA-mediated KD would be to degrade PRMT5 in vivo. Degradation strategies have been used successfully to treat other tumors (50). Another possibility is to reverse the order of RT and PRMT5 inhibition. The rationale would be to allow PRMT5 inhibition to reduce the stem-like phenotype in the TIC population without having to work against the potentially antagonistic background of RT. We are planning to explore some of these strategies and encourage other investigators to do so as well.

The strengths of this study include the use of multiple PHGG cell models, the availability of a clinical-grade inhibitor for PRMT5, and the consistent levels of PRMT5 KD expression achieved across cellular models and shRNA constructs. The limitations include the inability to directly model the effect of PRMT5 KD on isolated PHGG TIC cells, the limitation that only one H3-wt PHGG cellular model was available, and the possibility in the mouse experiments that TICs were altered to a PRMT5-resistant phenotype by the initial application of RT.

Supplementary Material

Supplementary Figure 1

Figure S1. a. tumor phylogeny for CNS/Brain category used in the Broad Institute DepMap project, OncoTree (oncotree.info) Kundra et al., JCO Clinical Cancer Informatics 2021; b. individual sample distribution for the shaded curves shown in Figure 1e.

Supplementary Figure 2

Figure S2. a. single cell RNA-Seq analysis shows PHGG includes a population of stemlike cells that express stemness (NES, SOX2) and oligodendrocyte lineage (OLIG2) genes; –log p value of expression in stemlike population versus other tumor cell populations from a set of 19 PHGG patient samples; b. left panel - dot plot showing average differential expression and percentage of cells expressing each gene in stemlike population 17; right panel - UMAP projection of cell types, arrowhead points to the stemlike population (17) depicted in left panel; c. survival curve showing effect of LLY283 in DMG cell lines not included in Figure 2e (N=3 for all data points, error bars show SEM, error bars not shown were too close to central estimate to display)

Supplementary Figure 3

Figure S3. Metascape pathway analysis showed enrichment of differentiation pathways with PRMT5 KD.

Supplementary Figure 4

Figure S4. a. BLI imaging results for mice in RT and RT + LLY283 arms of experiment shown in Figure 4g, RT was given on Day 0 and LLY283 treatment began day 27. Mice without Vehicle or Inhibitor listed died following RT and prior to the start of LLY283 treatment; b. PRMT5 inhibitor LLY283 reaches brain and tumor tissue in treated mice; left panel - Western blot showing SDMA levels in cortex and tumor of mice treated with LLY 283 versus vehicle; right panel - quantification of Western blots shows differing effects of LLY283 treatment on SDMA in cortical and tumor tissue (N=2 biological)

Supplementary Figure 5

Figure S5. Overall effects of PRMT5 KD on H3K27me3, H3K4me3 and H3K27ac occupancy in BT245 and GBM1 cells.

Supplementary Figure 6

Figure S6. Differential effects of PRMT5 KD on H3K27me3, H3K4me3 and H3K27ac occupancy in BT245 and GBM1 cells.

Supplementary Figure 7

Figure S7. Summary of genes and pathways at which H3K4me3 or H3K27me3 were differentially expressed in PRMT5 KD versus control cells.

Supplementary Figure 8

Figure S8. Metascape analysis of differential gene expression effects in BT245 cells.

Supplementary Figure 9

Figure S9. Metascape analysis of differential gene expression effects in GBM1 cells.

Supplementary Table 1

Genes identified in shRNA screen of DIPG10 with FC<0.75 and p<0.025.

Supplementary Table 2

Genes with p < 0.05 when comparing expression FC between PRMT5 KD vs shNULL

Supplementary Table 3

List of Hyper-fucosylated glycans for prostate cancer analysis

Supplementary Table 4

BT245 H3K27me3 GSEA gene sets containing target genes

Supplementary Table 5

GBM1 H3K4me3 GSEA gene sets containing target genes

Supplementary Table 6

GBM1 H3K27me3 GSEA gene sets containing target genes

Authors’ Disclosures

No disclosures were reported.

Authors’ Contributions

J. DeSisto: Conceptualization, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. I. Balakrishnan: Investigation, methodology. A.J. Knox: Investigation, methodology. G. Link: Investigation. S. Venkataraman: Conceptualization, supervision, methodology. R. Vibhakar: Conceptualization, resources, supervision, methodology, project administration, writing–review and editing. A.L. Green: Conceptualization, resources, supervision, project administration, writing–review and editing.