Xinqidi,Your biological research partner
Xinqidi Biotech Co.,Ltd,Wuhan,China 2008-2021
R&D 13th year

Somatic mutations and single-cell transcriptomes reveal the root of malignant rhabdoid tumours

Issuing time:2021-03-08 11:37


Malignant rhabdoid tumour (MRT) is an often lethal childhood cancer that, like many paediatric tumours, is thought to arise from aberrant fetal development. The embryonic root and differentiation pathways underpinning MRT are not firmly established. Here, we study the origin of MRT by combining phylogenetic analyses and single-cell mRNA studies in patient-derived organoids. Comparison of somatic mutations shared between cancer and surrounding normal tissues places MRT in a lineage with neural crest-derived Schwann cells. Single-cell mRNA readouts of MRT differentiation, which we examine by reverting the genetic driver mutation underpinning MRT, SMARCB1 loss, suggest that cells are blocked en route to differentiating into mesenchyme. Quantitative transcriptional predictions indicate that combined HDAC and mTOR inhibition mimic MRT differentiation, which we confirm experimentally. Our study defines the developmental block of MRT and reveals potential differentiation therapies.


Malignant rhabdoid tumours (MRT) are soft tissue cancers that predominantly affect infants. Although they may arise in any body part, MRT usually form in isolation or synchronously in the kidney and the brain (where they are referred to as atypical teratoid/rhabdoid tumours (AT/RT)). MRT, especially metastatic MRT, remain one of the most lethal childhood cancers, even following intense multimodal treatment. The sole driver event of MRT is the occurrence of biallelic mutations in the genes encoding SMARCB1(INI1, 95% of cases) or SMARCA4 (BRG1, 5% of cases), the core subunits of the SWItch/Sucrose Non-Fermentable (SWI/SNF) chromatin-remodelling complex1,2,3. In about one-third of cases, one of the variants is present in the germline, thus predisposing children to the development of MRT4.

Like most childhood cancers5, MRT are thought to arise during embryogenesis, a notion that has recently been substantiated in studies of mouse models of Smarcb1 loss. Rhabdoid tumours, albeit the majority being AT/RT, only developed in these mice when Smarcb1 was inactivated during very early embryogenesis, but not at later fetal stages or in adult animals. Renal MRT were never observed6. Analyses of bulk, and more recently of single-cell transcriptomes, suggest that MRT retain an overall fetal transcriptome with neural as well as mesenchymal signals7,8,9,10. These findings suggest as a plausible source of rhabdoid tumours the ectoderm-derived neural crest, which is uniquely capable of generating cell types across the boundaries of the germ layers, mesoderm and ectoderm.

The fetal origin of MRT may be exploitable therapeutically by promoting differentiation of MRT along developmental pathways. The possibility of devising differentiation treatments for childhood cancer has recently gained traction with the advent of high-throughput single-cell assays5,10,11,12. Single-cell transcriptomic readouts enable precise, comprehensive and quantitative comparisons of cancer cells to the transcriptional changes underpinning normal cellular development, thus potentially revealing therapeutic avenues for promoting cellular maturation.

Here, we define the developmental root of MRT and reveal opportunities for differentiation therapy, combining phylogenetic analyses of tumours and surrounding normal tissues, single-cell mRNA readouts, and perturbation experiments in patient-derived MRT organoids.


Malignant rhabdoid tumours are phylogenetically related to neural crest-derived tissues

The starting point of our investigation were phylogenetic analyses of MRT, to establish whether the origin of MRT lies in the neural crest in humans. We have previously shown that it is feasible to reconstruct the developmental relationship between childhood tumours and normal tissues from the distribution of somatic mutations across tissues13. Applying these principles to MRT, we used DNA whole-genome sequencing (WGS) to study two cases of MRT along with corresponding normal tissues.

First, we examined tissue obtained from a child presenting with the most common type of extracranial rhabdoid tumour, renal MRT. The child carried a pathogenic germline SMARCB1mutation. We performed WGS of tumour (n = 2), blood cells, kidney parenchyma, and renal hilar tissue (n = 2) (Fig. 1a, Supplementary Table 1). Using an established variant calling pipeline13,14,15,16, we determined somatic variants in each tissue, from which we derived phylogenetic relations between tumour and normal tissues. The possibility of observing shared mutations due to tumour contamination was addressed by histological examination and by quantitative assessment through a mixture model (“Methods” section). The key finding in this first case was that some, but not all somatic mutations of the tumour, were present in hilar tissues, occupied by ganglion cells and Schwann cells (Fig. 1a, b, Supplementary Fig. 1). Both these cell types are derived from the neural crest. However, there were no shared somatic mutations between tumour and blood or kidney parenchyma, bar ubiquitous early embryonic mutations (Fig. 1a, Supplementary Fig. 1). These findings place MRT on an ectodermal, neural crest-derived lineage with Schwann cells, distant from mesodermal blood and kidney parenchymal lineages.

Fig. 1: MRT are phylogenetically closely related to neural crest-derived Schwann cells.

A Phylogenetic tree representing the somatic genetic relation of a renal MRT and normal tissues. Percentages: clone size in tissues. Numbers inside circles: mutation burden within cluster. Red or white coloured rectangles: SMARCB1 mutations status (red = mutant; white = wild type). LOH: loss of heterozygosity. H&E (B) staining of biopsies and INI1 (C) immunostaining, showing INI1 negative Schwann cells in hilum biopsy 2. Scale bars = 100 µm. D Pattern of positive INI1 staining in Schwann cells of normal nerve sheath from control hilar regions of two independent donors. Scale bars = 100 µm. E Phylogenetic tree representing the somatic genetic relation of an extradural (spinal) MRT and normal tissues. Embryonic clusters of mutations are denoted (a–d). The annotation otherwise follows A. H&E staining (F), clone size of the different mutational clusters (a–d, G), and INI1 immunostaining (H) of dorsal nerve root, ventral nerve root, tumour and (I) bone marrow of the same donor. The latter showing positive INI1 staining. Scale bars = 100 µm.

Examining shared mutations between tumour and hilar tissues more closely, we found that one hilar biopsy, occupied mainly by ganglion cells, shared only a small number (n = 6) of variants with the tumour, whereas the second, composed of Schwann cells, shared 175 mutations with the tumour (Fig. 1a, b, Supplementary Data 1) including copy number-neutral loss of heterozygosity of SMARCB1 (Supplementary Fig. 2). To verify this finding, we performed immunohistochemistry of the SMARCB1 protein, INI1 (Fig. 1c). As predicted from the distribution of mutations, the first hilar biopsy showed only occasional INI1 negative cells, consistent with a heterozygous germline mutation of SMARCB1. By contrast, the Schwann cells of the second hilar biopsy, which should have stained ubiquitously and intensely positive for INI1 (Fig. 1d), did not exhibit INI1 staining (Fig. 1c), consistent with biallelic loss of SMARCB1 predicted from the somatic genome of this tissue.

Next, we examined the tissues obtained post mortem from a child, who succumbed to an MRT of the cervical spine. The tumour bulk was situated ventrally in the extradural space. The child did not carry germline SMARCB1 mutations. No early mosaic (i.e. present in blood) variant affecting SMARCB1 was discovered in this patient. We studied tumour tissue along with nine normal tissues: skin (n = 2), fat (n = 2), muscle (n = 2), blood, dorsal, and ventral nerve roots (Fig. 1e, Supplementary Table 1). Pursuing the same analyses as before, we found that the tumour was somatically related to neural crest-derived Schwann cells sampled in nerve roots (Fig. 1e, f, Supplementary Fig. 1, Supplementary Data 1), but not to any other normal tissue. The clonal composition underlying this phylogenetic relation was complex. Based on variant allele frequencies and distribution of mutations, we were able to discern four clones (Fig. 1g), two of which were shared between Schwann cells and tumour. In addition, the tumour and Schwann cell lineages possessed a private clone each, alluding to a sustained potential of tumour and Schwann cells for subclonal diversification. Analysis of copy number variants (CNVs) revealed a biallelic loss of SMARCB1 in tumour and both nerve roots (Fig. 1e, Supplementary Fig. 2, Supplementary Data 1), which again we were able to validate through INI1 staining. INI1 negative Schwann cells were more readily found in the ventral root, consistent with the larger clone sizes in this tissue (~40% vs. ~20%, Fig. 1h, i). Together, these observations provide the most direct evidence yet that human MRT is phylogenetically related to the neural crest lineage and firmly places its origin in fetal life.

SMARCB1 reconstitution drives MRT differentiation

To establish the differentiation stage of MRT within neural crest development, we studied the consequences of reversing the loss of SMARCB1, the principal genetic driver of MRT. As a model of MRT, we utilised patient-derived MRT organoids, which have been shown to faithfully recapitulate the genetic, transcriptional, and epigenetic features of primary MRT tissue17. We reconstituted SMARCB1 expression in three MRT organoid cultures17 (60T, 78T and 103T; Supplementary Table 2) by lentiviral transduction with either a control or SMARCB1 expression plasmid (Fig. 2a, Supplementary Fig. 3a). DNA methylation profiles of MRT organoids resembled those of primary MRT tissue, irrespective of SMARCB1 status (Supplementary Fig 3b). Reconstitution of SMARCB1 expression induced a proliferation arrest in all MRT cultures (Supplementary Fig. 3c) with a morphological transformation of cells (Fig. 2b, Supplementary Fig. 4a, c). While both 60T and 103T transformed from a grape-like to a neural- or fibroblast-like morphology with long extensions protruding from the cell body, 78T stopped proliferating without an apparent morphological change. To assess the transcriptional profiles underpinning these phenotypic changes, we subjected organoid cultures, without and with SMARCB1 re-expression, to single-cell mRNA sequencing (10× Genomics Chromium platform, n = 16,133 cells post filtering). Cell cycle profiles generated from single-cell transcriptomes confirmed the growth arrest induced by SMARCB1, with 78T showing the least penetrant effect (Supplementary Fig. 3d). UMAP clustering revealed that, as expected, most transcriptomic variance of MRT single cells can be explained by donor, as cells first separate by patient line (Supplementary Fig. 3e). After transduction, the majority of cells expressing SMARCB1 segregated into independent cell clusters for each patient line (Fig. 2c, Supplementary Fig. 5a–c). This segregation was not explained by batch effects of individual cultures, as unsuccessfully transduced cells co-clustered with cells from cultures transduced with the control plasmid (Supplementary Fig. 3f). As our phylogenetic analyses revealed that MRT are neural crest-derived (Fig. 1), we subsequently assessed the similarity of MRT cells with and without SMARCB1 re-expression relative to single-cell signals of (murine) neural crest development18 using logistic regression12 (Fig. 2a). At baseline (i.e. no SMARCB1 re-expression), MRT organoid transcriptomes primarily resembled mesenchymal and neural cells (Fig. 2d, Supplementary Fig. 3g), as previously shown7,8,10. In addition, each individual patient line exhibited different signals of neural crest differentiation stages (Fig. 2d, Supplementary Fig. 3g). Patient lines 60T and 103T primarily resembled mesenchymal cells, whereas 78T exhibited a more neural signal. Examining cellular mRNA profiles upon SMARCB1 reconstitution, MRT cells appeared consistently more differentiated. That is, they resembled their normal counterpart more strongly, as similarity to most neural crest cell types increased (Fig. 2d, Supplementary Fig. 3g). Further assessment of cell type similarity showed that SMARCB1 reconstitution promotes a neural to mesenchyme conversion that is consistent among all three MRT organoid cultures (Fig. 2e). These results were validated using a second independent mouse single-cell mRNA reference of early neural and mesenchymal cell types19 (Supplementary Fig. 3h). In agreement, analysis of neural crest differentiation genes showed a significant upregulation of mesenchymal markers (Fig. 2b, f, g, Supplementary Fig. 4b–d, Supplementary Data 2). Additional cell typing was performed for each single-cell cluster separately to evaluate intra-organoid heterogeneity (Supplementary Fig. 5a–f, Supplementary Data 2), showing that single-cell clusters exhibited variable neural crest differentiation signals. However, SMARCB1+ clusters consistently induced a relative gain of mesenchymal differentiation signal, with the exception of minor cluster 60T_S2, which retained a more neural identity. Altogether, our findings place MRT on a developmental trajectory of neural crest to primarily mesenchyme conversion, which is promoted by SMARCB1 reconstitution.

Fig. 2: SMARCB1 reconstitution drives MRT differentiation.

A Schematic representation of SMARCB1 reconstitution in patient-derived MRT organoids and subsequent single-cell transcriptome comparison to fetal mouse neural tube and neural crest cell types. Branching tree represents differentiation trajectories of mouse neural crest. Abbreviations are indicated. B Representative immunofluorescence images of MRT control (C) and SMARCB1+ (S) organoids. White: DAPI (nuclei), red: phalloidin (membranes), green: MMP2 (mesenchymal marker). Scale bars equal 50 µm. C UMAP representation of single cells from MRT control (grey) or SMARCB1+ (green) organoid lines (60T control/SMARCB1+: 8059/425 cells, 78T control/SMARCB1+: 3195/806 cells, 103T control/SMARCB1+: 2694/953 cells). D Dot plots represent similarity of MRT control (circles) or SMARCB1+ (squares) cells to neural crest differentiation trajectories. Colours represent the average probability (prob) that the MRT cells are similar to the indicated neural crest cell type (predicted similarity score estimated by logistic regression12). Changes in similarity score between control and SMARCB1+ cells were assessed for cell types with average similarity score >0.5. P values were calculated using an unpaired Student’s t test (two-tailed): *<1e−3, **<1e−9, ***<1e−15 (−log10 (p value): 60T D = 45, S = 27, M2 = 66, ME = 3.7; 78T NT = 9, D = 54, M1 = 14, S = 22, M2 = 4.4; 103T D = 198, EM = 40, S = 7.8, M2 = 314, ME = 3.2). E Stacked bar plot represents relative frequencies of single-cell annotations for MRT control (−) and SMARCB1+ (+) organoids, showing a consistent conversion of neural to mesenchymal signals. Cell type annotation was assigned for each single-cell based on the highest similarity score. Colours represent neural crest cell types depicted in Fig. 2a. Cell type migratory2 (M2) was assigned as either migratory mesenchyme (ME(M2)) or migratory autonomic (A(M2)) based on the highest similarity score. The relative frequency of the mesenchymal/autonomic (ME/A) branch was compared between control and SMARCB1+ organoids for each patient line. P values were calculated using a chi-square test: *<0.01, ***<1e−15 (p value: 60T = 0.0048; 78T = 4.9e−48; 103T = 1.0e−32). F Dot plot shows expression levels (exp) of mesenchymal marker MMP2 for MRT control (−) and SMARCB1+ (+) organoids for each patient line. Colour-code from grey to red refers to average MMP2 transcript levels (unique molecular identifier (UMI)). Dot size refers to the percentage of cells (pct) showing MMP2 expression. G Box plot representation of gene module scores for MRT control (grey) and SMARCB1+ (green) single cells (n = 60T control/SMARCB1+: 8059/425 cells; 78T control/SMARCB1+: 3195/806 cells; 103T control/SMARCB1+: 2694/953 cells), showing consistent upregulation of mesenchymal/autonomic differentiation genes for SMARCB1+ cells. Box plots indicate median (middle line), 25th and 75th percentile (box). Whiskers represent the range excluding outliers (dot). Module scores were generated by averaging gene expression levels per set of genes. Gene sets include marker genes for either sensory (S) or mesenchymal/autonomic (ME/A) differentiation branches, distinguishing early and late differentiation genes. Module scores were assessed by comparing control and SMARCB1+ cells. P values were calculated using an unpaired Student’s t test (two-tailed): *<1e−3, **<1e−9, ***<1e−15 (−log10 (p value) ME/A late 60T = 28, 78T = 64, 103T = Inf; ME/A early 60T = 77, 78T = 134, 103T = Inf; S early 60T = 5.6, 78T = 72, 103T = 54; S late 60T = 28, 78T = 16, 103T = 11).

Mimicking SMARCB1 reconstitution pharmacologically

Reconstitution of SMARCB1 to drive differentiation of MRT would appear to be an attractive, non-cytotoxic treatment strategy. However, reinstating SMARCB1 expression genetically in children is not feasible at present. An alternative strategy is to find agents that mimic the changes induced by SMARCB1 re-expression. Using bulk mRNA-seq, we defined a SMARCB1+ transcriptional programme based on genes upregulated upon SMARCB1 re-expression in our three MRT organoid cultures (Supplementary Fig. 6a, Supplementary Data 3). We could validate the SMARCB1+ programme in MRT tissue and found a positive correlation with SMARCB1 expression levels in normal tissues (Supplementary Fig. 6b). To explore therapeutic avenues, we searched a publicly available perturbation data base20 for drugs that induce expression changes of SMARCB1 reconstitution (Fig. 3a). This analysis identified a variety of HDAC and mTOR inhibitors (Supplementary Fig. 6c) as the top hits. Interestingly, HDAC inhibitors have previously been identified for treatment of rhabdoid tumours by orthogonal approaches21. We tested the phenotypic and transcriptional effects of these agents, alone or in combination, across the three MRT organoid cultures. HDAC inhibition alone induced a morphological transformation akin to SMARCB1 reconstitution (Fig. 3b, Supplementary Fig. 7a). Furthermore, there was a significant correlation between gene expression changes of bulk culture transcriptomes of SMARCB1 re-expression and HDAC inhibition (Fig. 3c, Supplementary Fig. 6d). Inhibition of mTOR signalling primarily constrained organoid growth, which, however, was readily reversible by drug washout (Fig. 3f, g). Combination of HDAC and mTOR inhibition induced the phenotypic and transcriptional changes of SMARCB1 reconstitution/HDAC inhibition as well as a marked proliferation arrest (Fig. 3b, c, Supplementary Fig. 7a). The action of HDAC and mTOR inhibition was synergistic, as corroborated by assessment of the two drugs in dose-response matrices (Fig. 3d, e, Supplementary Fig. 8a). Furthermore, the combined effects of the drugs on viability were more durable than single-agent treatment. On its own, anti-proliferation effects of each drug were readily reversible upon washout (bar HDAC inhibition in MRT organoid 103T). By contrast, combination treatment had more lasting effects on proliferation in all tested MRT organoid cultures (Fig. 3f, g, Supplementary Fig. 7b). While regrowth of 60T and 103T was completely diminished after drug washout, 78T showed minor regrowth, which could relate to the remnant proliferating cells that were also observed upon SMARCB1 reconstitution (Supplementary Fig. 3d). To determine whether MRT are in particular sensitive to HDACi and mTOR inhibition, we tested the sensitivity of normal kidney organoids22 to both drugs. Normal kidney organoids were significantly more resistant to single agents as well as combination treatment, and in contrast to MRT, showed significant regrowth upon washout of drug combination (Fig. 3f, g, Supplementary Figs 7b, 8b–d). Mechanistically, the longevity of the effects of combined HDAC and mTOR inhibition may be mediated through interference with MYC, as MYC-driven cancer cell lines seem to be particularly susceptible to this drug combination23. In our experiments, we interrogated hallmark pathways and perturbation gene sets24, which validated our MRT SMARCB1+ model, as we found enrichment for SWI/SNF-related perturbation gene sets upon SMARCB1reconstitution (e.g. SNF5 (SMARCB1) and subunits of polycomb repressive complexes25) (Supplementary Fig. 6e, f). Further, we found that MYC target genes were strongly downregulated upon SMARCB1 re-expression (Supplementary Fig. 6e, h). This was mimicked by combined HDAC/mTOR inhibition, significantly more strongly than by single-agent treatment (Supplementary Fig. 6i). Further examination of differentially expressed genes showed that identified pathways are largely shared between SMARCB1 reconstitution and combination treatment (Supplementary Fig. 6g, Supplementary Data 3). Together, these analyses identify combined HDAC/mTOR inhibition as pharmacological mimics of SMARCB1 reconstitution that prohibit proliferation and induce differentiation in MRT.

Fig. 3: Combined HDAC/mTOR inhibition mirrors SMARCB1 reconstitution.

A Overview of methodology used for discovery of potential differentiation therapeutics. BRepresentative immunofluorescence images of MRT organoids treated with DMSO control or a combination of vorinostat (HDACi, 1 µM) and sirolimus (mTORi, 2 nM). White: DAPI (nuclei), red: phalloidin (membranes), green: MMP2 (mesenchymal marker). Scale bars equal 50 µm. C Heatmaps represent gene expression values (n = 2 independent experiments) of MRT control or SMARCB1+ organoids, or MRT organoids treated with vorinostat (HDACi, 1 µM) or both vorinostat and sirolimus (combination,1 µM/2 nM). Heatmaps are subset for genes differentially expressed upon SMARCB1 re-expression (Supplementary Data 3). Genes are ordered by the average mRNA changes induced by SMARCB1 re-expression and treatment. Colour-code represents gene expression values scaled by gene. Pearson correlation coefficients (corr.) were generated by comparing mRNA changes induced by either SMARCB1 re-expression or HDACi/combination treatment. P values are indicated for Pearson’s correlation tests (two-tailed): ***<1e−15 (−log10 (p value): Combi 60T = 217, 78T = Inf, 103T = 221; HDACi 60T = 268, 78T = 306, 103T = 192). D Schematic overview of the dose-response matrix setup to find synergy between HDAC (vorinostat) and mTOR (sirolimus) inhibitors in MRTs. E Graphs show zero interaction potency (ZIP) scores that indicate either synergistic (red) or antagonistic (blue) effects of combination treatment. ZIP scores are generated by calculating the observed deviation from a reference model that assumes drugs are non-interacting (synergy when ZIP > 10%51). The dashed rectangles highlight the drug concentration ranges where synergy between the two drugs is the strongest. Source data are provided as a Source Data file. F Schematic overview of the regrowth assay. G Bar graphs represent cell viability values normalised to timepoint 1 (T1) DMSO controls for each MRT or normal kidney organoid line. Mean and SD (error bars) of independent experiments (dot) are indicated (n = 60T/103T: 3, 78T mTOR/HDAC 1 µM/Combi 2 nM 1 µM: 6, 78T HDAC 3 µM/Combi 2 nM 3 µM: 4. normal kidney: 7). Each independent experiment is an average of four technical replicates. Source data are provided as a Source Data file. Additional effect of combination treatment on cell viability was determined by comparing combination (T2) with HDACi (T2) treatment. Regrowth capability was assessed by comparing T2 to T1. P values were calculated using a paired ratio Student’s t test (two-tailed): *<0.05, **<0.01, ***<0.001 (p value: Combi 1 µM T1 vs. HDACi 1 µM T1 60T = 0.020, 78T = 0.012; Combi 3 µM T2 vs. Combi 3 µM T1 78T = 0.013, normal kidney donor 1 = 2.5e−5, donor 2 = 1.8e−5).

Article classification: Biological abstract
Share to:
Add:Room A11-329, 1st Floor, No.1, SBI Venture Street, Optics Valley, East Lake
New Technology Development Zone, Wuhan, China.
Certificate NO.:U18Q28010569R0S