Samples and sequencing

After radical nephrectomy, four tumour and one normal kidney samples were biopsied and divided into fresh tissue for dissociation and fresh frozen for subsequent WGS and bulk RNA-seq. An additional perioperative blood sample was taken. DNA, RNA, and single cell suspension were extracted from the resulting fresh frozen samples.

DNA was extracted from DNA libraries of 150 bp length were prepared for Illumina NovaSeq 6000 paired end sequencing with a 500 bp insert size. The four tumour samples were sequenced to an average coverage of 76.73x, and the normal kidney and blood samples were initially sequenced to similar depth, then topped up to 496.69x and 531.69x, respectively, to identify shared variants.

Bulk RNA libraries for the five renal samples were prepared with 450–500 bp fragment size for 150 bp paired end sequencing with Illumina 2500 HiSeq chemistry to a mean coverage of 68.66x.

Single cells were dissociated in PBS, loaded, barcoded, and processed for 10X library preparation using the same protocol as reported in Li, et al. Cancer Cell 2022 [27]. Samples were sequenced using Illumina NovaSeq 6000 paired end machines.

Histology preparation and processing

Paxgene fixed tissue blocks were stained with H&E and imaged with a Hamamatsu NanoZoomer S60 slide scanner.

Genomic data processing, variant calling, and phylogeny analysis

Genomic data were aligned to the GRCh 37d5 reference genome with the Burrows-Wheeler transform (BWA-MEM). Chromosomal copy number changes were identified using both ASCAT NGS (https://github.com/cancerit/ascatNgs/tree/dev) version 4.3.3 and Battenberg (https://github.com/cancerit/cgpBattenberg) version 3.5.3 (Supplementary Table 3 and Supplementary Fig. 1a). SNVs were called using CaVEMan (Cancer Variants through Expectation Maximisation, https://github.com/cancerit/CaVEMan) version 1.15.1 and indels were called using an in-house Pindel build (https://github.com/cancerit/cgpPindel) version 3.3.0, both run in unmatched mode calling variants against a simulated reference genome dataset. We selected SNVs with fewer than half of supporting reads as clipped (CLPM = 0) and a median alignment score greater than or equal to 140 (ASMD ≥ 140). Mapping quality and base quality cut-offs were set to minimums of 30 and 25, respectively.

We identified germline and somatic variants using previously described exact binomial testing [26]. A Shearwater-like filter was deployed to find probable sequencing artefacts by comparing called SNVs to an internal panel of 21 in-house, unrelated renal WGS samples. To select variants shared by all four tumours, we performed an additional Fisher’s exact test comparing renal variants (tumour and normal) to blood, using a cut-off of p < 0.01 to select for somatic, non-artefactual SNVs enriched in the tumours. To check for pathogenic germline predisposition syndromes, we looked in the list of germline mutations for coding mutations in the following genes: FH, TSC1, TSC2, FLCN, SDHA, SDHB, SDHC, SDHD, BAP1, and PTEN. There was one nonsynonymous mutation, a C to T missense variant in FLCN at chr17:17118598 resulting in a A445T amino acid change. This uncommon single nucleotide polymorphism is not directly associated with Birt-Hogg-Dubé syndrome and its pathogenicity is labelled as “Benign-Likely Benign” in ClinVar (build 156). Given that the patient had no other somatic mutations in FLCN, we concluded that she had no definitive germline cancer predisposition. This was repeated for indels that passed Pindel filtering, revealing no potentially predisposing germline indels.

The phylogeny was drawn using both a manual and a Dirichlet process with a 10,000 burn-in on the filtered somatic SNVs.

In generating Fig. 1d, we assumed shared variants to be clonal and accordingly adjusted variant allele frequencies (VAF) to a theoretical mean of 0.5 while accounting for tumour-specific chromosome copy number changes. Raw and adjusted VAFs are included in Supplementary Tables 1 and 2, respectively.

Bulk transcriptomic data processing

Bulk transcriptomic data were mapped to the hg37d5 reference genome using the Ensembl 75 transcriptome with aligner STAR (https://github.com/alexdobin/STAR) version 2.5.0.

Differential expression analysis reported in Supplementary Table 4 was computed using DESeq2 version 1.30.1. We compared all four tumours against a panel of the one matched normal kidney and 10 GTEx normal kidney cortex samples from female donors aged 20–59. The test also included a batch correction between our in-house and GTEx sequencing. Patterns of gene programme and pathway changes were calculated from genes ranked by log-fold change in the GTEx panel DE analysis using fgsea (https://github.com/ctlab/fgsea) version 1.16.0 with Hallmark pathways, with all results in Supplementary Table 5 and enriched gene programmes with padj < 0.05 and Net Effect Size (NES) > 1.5 plotted in Fig. 2c.

Expression of the mutated MTOR allele was determined using a samtools search of RNA-seq BAM files spanning the MTOR gene coordinates.

Single cell transcriptomic data processing

scRNA-seq datasets were mapped with CellRanger and processed using Seurat version 4.0.4. Because oncocytoma and chromophobe tumours are known to have abnormally high numbers of mitochondria, we did not filter out cells with high mitochondrial reads and instead removed cells with unexpectedly too few (<200) or too many (>2000) expressed genes.

These filtered data were then processed using NormalizeData(), FindVariableFeatures() with VST selection method, and ScaleData() Seurat functions. To retain as many tumour cells, which may divide at a higher rate than normal cells, cycling cells were identified using the CellCycleSorting() function but not removed. RunPCA() was performed before inspecting the data. Lastly, neighbourhoods were defined using the FindNeighbors() function with 30 dimensions and cells were clustered using the FindClusters() function with a resolution of 2.1. Clusters were visualised in UMAP space (Supplementary Fig. 1b).

We then used AlleleIntegrator (https://github.com/constantAmateur/alleleIntegrator) to impute tumour cell identity by calling heterozygous SNPs from tumour DNA samples and searching for patterns of chromosome losses in single cells [27]. Cells with allelic imbalances that matched the chromosomal copy number profiles of tumours A and D called from WGS data were labelled as tumour, which enabled identification of tumour cell clusters (Supplementary Fig. 1c).

To identify expression of the mutated MTOR allele, we searched scRNA-seq BAM files downsampled to MTOR gene coordinates for the 12 bp duplication using samtools.

Inferring cell type identity and contribution from transcriptomic data

We used two methods of integrating bulk and single cell RNA-seq data. First, we started with bulk deconvolution using Cell Signal Analysis (https://github.com/constantAmateur/cellSignalAnalysis), comparing the four tumour bulk transcriptomes to a normal renal scRNA-seq reference [27, 28]. The top nine most similar cell type identities are shown in Fig. 1f.

Next, we trained a logistic regression model on the same renal scRNA-seq reference as previously described [29]. The most likely cell type was manually assigned to each Seurat-defined cluster. Tumour cell clusters determined by allelic pattern imbalances are displayed in Fig. 1g along with several clusters determined by this method to be renal cell types.

Cloning of recombinant human mTOR wild-type and EWEDdup constructs

For the production of human mTORC1 complex, the three subunits were cloned individually into a pCAG mammalian expression vector as described before [30, 31]. Cloning of the mTOR ‘EWED’ duplication mutant was carried out using a fragment-assembly based approach: two fragments were PCR amplified with overlaps containing the desired 12 bp 1455EWED1458 duplication region at one end and overlaps with the digested vector at the other, and then they were assembled with the digested vector using NEBuilder HiFi DNA Assembly. mTOR subunit has an N-terminal tandem Strep-tag II followed by a TEV cleavage site, while RAPTOR and mLST8 subunits are without tags. Human RICTOR was PCR-amplified from IMAGE:9021161 clone, and then cloned with an N-terminal 3X Flag-tag in the pCAG vector for expression in mammalian cells. Human Sin1.1 was PCR-amplified from Addgene 73,388 plasmid (gift from Taekjip Ha [32]) and then cloned into pcDNA4TO. Subsequently, the promoter-gene (SIN1.1)-terminator cassette was PCR amplified from this plasmid and cloned into a pCAG vector already containing mLST8 gene, to allow co-expression of the two proteins from a single plasmid.

Recombinant protein expression and purification

mTORC1 and mTORC2 complexes, containing either mTOR_WT or mTOR_EWEDdup mutant, were expressed by transient transfection of Expi293F cells grown in Expi293 media (Thermo Fisher A1435102) in a Multitron Pro shaker set at 37 °C, 8% CO2 and 125 rpm. A total of 1.1 mg DNA/L cells was co-transfected into cells at a density of 2.5 × 106 cells mL−1 using PEI (Polyethyleneimine "MAX", MW 40,000, Polysciences, 24,765, total 3 mg PEI/L cells). After 52 h (mTORC1) or 68 h (mTORC2), cells were harvested by centrifugation and cell pellets were frozen in liquid N2.

mTORC1 was purified from cell pellets from 2 L Expi293F culture as described before [31], by affinity purification on a Strep-Trap HP resin, followed by Strep-tag cleavage by TEV protease overnight on the column. The cleaved protein was further purified by anion-exchange chromatography (AEX) on a 5 mL HiTrap Q column (Cytiva), concentrated with Amicon Ultra-4 100 kDa concentrators, flash frozen in liquid N2 and stored at −80 °C.

mTORC2 was purified from 2L Expi 293F cell pellets. Cells were lysed in lysis buffer consisting of 50 mM BICINE, pH 8.5, 300 mM NaCl, 2 mM MgCl2, 0.5 mM TCEP and the clarified cell lysate loaded onto 2 tandem Strep-Trap HP 5 ml columns. The loaded column was washed with 20 CV lysis buffer followed by 20 CV wash buffer (50 mM BICINE, pH 8.5, 150 mM NaCl, 2 mM EDTA, 0.5 mM TCEP) and protein eluted with wash buffer supplemented with 5 mM desthiobiotin. mTORC2 containing fractions were pooled, diluted to <100 mM NaCl and further purified on a 5 ml HiTrap Q column equilibrated in Q-buffer containing 50 mM BICINE, pH 8.5, 50 mM NaCl, 0.5 mM TCEP. Protein was eluted by a linear gradient of Q-buffer containing 1 M NaCl. mTORC2 containing fractions were pooled, concentrated and stored as described for mTORC1.

Human 4EBP1 was expressed as a GST-4EBP1 fusion in E. coli strain C41(DE3) and purified by affinity chromatography on Glutathione-Sepharose 4B beads, followed by GST-tag removal by incubating with TEV protease overnight. The cleaved 4EBP1 was passed through a Q column and the flow-through fractions containing 4EBP1 were concentrated and run on a Superdex 75 16/60 column equilibrated in 50 mM HEPES pH 8.0, 100 mM NaCl, and 1 mM TCEP.

Human RHEB was expressed as a GST-RHEB fusion in E. coli strain C41(DE3) and purified as described previously [31]. Briefly, GST-RHEB was purified on Glutathione-Sepharose 4B beads, followed by GST-tag cleavage with TEV protease overnight. To separate His6-tagged TEV protease from RHEB protein, sample was passed through a HisTrap FF column, followed by gel filtration on a Superdex 75 16/60 column. After removal of bound nucleotide by incubating the protein for 1 h on ice with EDTA buffer containing 20 mM HEPES pH 7, 100 mM NaCl, 20 mM EDTA and 1 mM TCEP, the buffer was exchanged to 50 mM HEPES pH 7, 100 mM NaCl, 5 mM MgCl2, 1 mM TCEP before concentrating the protein to 25 mg mL−1. The concentrated RHEB was then incubated with 1 mM GMPPNP (Jena Bioscience NU-401-50) for 60 min at 4 °C and the protein was flash frozen in liquid nitrogen and stored at −80 °C.

Human AKT1_D274A was generated by overlapping PCR mutagenesis using wild-type AKT1 (a gift from Thomas Leonard, Addgene plasmid 86561 [33]) and cloned into a pAceBac1 vector with N-terminal His10-StrepII-(tev) tag. The recombinant protein was expressed in Sf9 insect cells and purified by affinity purification on Strep-Trap HP resin, followed by simultaneous overnight tag cleavage by TEV and dephosphorylation by λ-protein phosphatase (NEB, P0735). The cleaved, dephosphorylated protein was further purified by anion exchange chromatography on a 5 mL HiTrap Q column and AKT1 containing fractions were concentrated with an Amicon Ultra-4 30 kDa concentrator, flash frozen in liquid N2 and stored at −80 °C.

mTORC1 activity assays

All reactions were performed in kinase buffer (KB) consisting of 25 mM HEPES, pH 7.4, 75 mM NaCl, 0.9 mM TCEP, 5% glycerol, 0.5 mg mL−1 BSA, at 30 °C for a duration of 30 to 45 min for the apo mTORC1_WT or for 2 to 4 min for the hyperactivated mTORC1_EWEDdup mutant or mTORC1_WT activated by RHEB. Reactions were set up by preincubating mTORC1 with 4EBP1 for 10 min on ice. After that, the reactions were equilibrated at 30 °C for 15 s, and kinase assays were started by the addition of 250 µM ATP and 10 mM MgCl2 (final concentrations). The reactions were stopped by the addition of 8 µL of 2.5X LDS sample buffer containing 4 mM ZnCl2 to 12 µL of sample. The samples were analysed by SuperSep Phos-tag (50 µM), 7.5% precast gels (FUJIFILM Wako Pure Chemical Corporation, 192-17381), with MOPS (no EDTA) running buffer supplemented with 5 mM sodium bisulphate. Western blots were performed using a 0.2 µm pore size nitrocellulose membrane (Invitrogen IB301002) and the iBlot dry blotting transfer system (Invitrogen). After the transfer, membranes were blocked with 5% Marvel in TBST buffer (100 mM Tris-HCl, 150 mM NaCl, 0.1% Tween 20). Incubation with the primary antibody, anti-4EBP1 (Cell Signalling, Cat. No. 9452S), was done in 5% BSA in TBST at 4 °C overnight, using 1:1000 dilution of the antibody. Incubation with the secondary antibody (anti-Rabbit IgG, HRP-linked Antibody, Cell Signalling Cat. No.7074) was at room temperature for 1 h, using 1:5000 dilution of the secondary antibody. Detection was performed using a ChemiDoc Touch Imaging System (Bio-Rad). Kinetic parameters kcat and KM were calculated using Prism 9 and nonlinear regression fitting of the data assuming Michaelis-Menten kinetics.

mTORC2 activity assays

mTORC2 activity assays were performed in kinase buffer (KB) consisting of 25 mM HEPES, pH 7.5, 100 mM NaCl, 0.5 mM TCEP and 5% glycerol. Reactions were set up by serial dilution of AKT1_D274A (10–120 μM) in KB, followed by pre-incubation with 0.2 μM mTORC2_WT or 0.01 μM mTORC2_EWEDdup for 5 min on ice. Samples were equilibrated to 30 °C and reactions started by addition of 1 mM ATP and 10 mM MgCl2. Reactions were terminated after 10 min by addition of 4X LDS sample buffer containing 8 mM ZnCl2. For each reaction, a sample volume equivalent to 1 μg of AKT1 was analyzed by SuperSep Phos-tag gel electrophoresis as described for mTORC1. Phosphorylated AKT was visualised by Coomassie staining (InstantBlue, Abcam) and imaged using a ChemiDoc Touch Imaging System. The fraction of phosphorylated AKT was determined by densitometry and kinetic parameters calculated as for mTORC1.

Cryo-EM sample preparation

Purified mTORC1 (1 µM), 4EBP1 (20 µM) and 1 mM AMPPNP (Jena Bioscience NU-407-10) were mixed in ~300 µL and incubated for 1 h on ice. The sample was crosslinked with 0.2 mM BS3 for 15 min at 4 °C, followed by further crosslinking with 0.03% glutaraldehyde (Sigma G5882) (added from a 1% glutaraldehyde stock in buffer A containing 50 mM HEPES pH 7.5, 0.1 M NaCl, 1 mM TCEP, 10% glycerol), for 15 min at 4 °C. The reaction was quenched by the addition of Tris pH 8.0 (final concentration 100 mM). The sample was then immediately subjected to a gradient centrifugation on a 12 mL gradient of 10–30% glycerol in 50 mM HEPESpH 7.5, 0.1 M NaCl, 1 mM TCEP, preformed in a SW40 rotor tube (Ultra-Clear, Beckman 344060) using a gradient maker (Biocomp Instruments). The sample was centrifuged in an SW40 rotor (Beckman) at 33,000 rpm for 16 h. After centrifugation, 0.40 mL fractions were collected, analyzed by SDS-PAGE, and the fractions containing crosslinked material were pooled, and concentrated to 500 µL using an Amicon Ultra-15 100 kDa concentrator. Cross-linked mTORC1_EWEDdup variant was further run on a Superose 6i 10/300 column equilibrated in 50 mM HEPES pH7.5, 250 mM NaCl, 5 mM MgCl2 and 1 mM TCEP and the peak fractions were concentrated to 0.8 OD280 and incubated further with 1 mM AMPPNP and 20 µM of 4EBP1 for 10 min and used immediately for cryo-EM grid preparation.

Cryo-EM data collection and processing

UltraAuFoil R 1.2/1.3 (Au 300 mesh) grids were glow-discharged using an Edwards Sputter Coater S150B for 1 min at 40 mA. A 3 μL aliquot of freshly prepared, crosslinked mTORC1_EWEDdup at a concentration of 0.5 mg mL−1 was added to the grids and blotted immediately for 3.5 s at 14 °C (95% humidity) and then plunge-frozen in liquid ethane using a Vitrobot (Thermofisher). A total of 11,197 micrographs were acquired on a FEI Titan Krios electron microscope operated at 300 keV. Zero-energy loss images were recorded on a Gatan K3 Summit direct electron detector operated in super-resolution mode with a Gatan GIF Quantum energy filter (20 eV slit width), using EPU for automated collection. Images were recorded at a magnification of 105,000 (calibrated pixel size of 0.86 Å), with a dose rate of ~16 electrons/Å2/s. An exposure time of 2.3 s was fractionated into 50 movie frames, giving a total dose of 50 electrons/Å2. For data collection, the defocus-range was set to −2.6 to −1.2 µm.

All image-processing steps were done using the RELION 4 software package [34], which includes Gctf [35], MotionCor2 [36], and ResMap [37]. A total of 11,197 micrographs were processed using GPU-accelerated MotionCor2 to correct for electron beam-induced sample motion, while the contrast transfer function (CTF) parameters were determined using Gctf. Particles were picked using RELION autopicking. In total, 1,069,382 particles were extracted with a particle box size of 512 by 512 pixels. Two rounds of reference-free 2D classification (using a mask with a diameter of 352 Å) resulted in a selection of 285,221 particles. This set of particles was subjected to a 3D classification over 25 iterations in point group C1, using a low-pass filtered (40 Å) ab-initio reference, which was created from the de-novo 3D model generated by the SGD algorithm in Relion4. Selection of reasonably looking classes by visualisation in Chimera and by paying attention to the rotational and translational accuracies for six classes reduced the number of particles to 264,890 sorted into four 3D classes. Without providing a mask around the mTORC1 complex, 3D auto-refinement of these particles, with C1 symmetry, led to a reconstruction of 4.2 Å resolution, based on the gold-standard FSC = 0.143 criterion [38, 39]. To correct for beam-induced particle movements, to increase the signal-to-noise ratio for all particles, and to apply radiation-damage weighting, the refined particles were further 'polished' using the Bayesian approach implemented in Relion 4.0. Following this step, another 3D auto-refinement using a mask around the mTORC1 complex as well as applying solvent-flattened FSCs yielded a 4.0 Å resolution reconstruction (FSC = 0.143 criterion). After a CTF- and beamtilt-refinement for the estimation of per-particle defocus and beam-tilt values for the complete set of selected particles, a 3D auto-refinement resulted in a 4.0 Å resolution reconstruction (FSC = 0.143 criterion). After correction for the detector modulation transfer function (MTF) and B-factor sharpening (sharpened with a negative B-factor as listed in Supplementary Table 6), the post-processed map was used for inspection in Chimera [40] and model building in Coot [41]. Local resolutions were estimated using ResMap. The 3D FSC server was used to determine the directional FSC and sphericity of the maps (Supplementary Fig. 4b) [42]. One of the mTORC1 protomer possess better EM densities compared to the other protomer. The mTORC1_EWEDdup dimer lacks significant densities in one of the RAPTOR molecules at the HEAT and WD40 domains. The region of FAT domain around 1455EWED1458 duplication shows extreme flexibility and lacks EM densities in both mTOR_EWED protomers. Focused classification with signal subtraction of the better protomer yielded 178,379 particles. A 3D refinement of this set of particles led to a reconstruction at 3.4 Å resolution. A focused classification with signal subtraction of the better protomer particles with the masks covering the mTOR residues 769–2549 and RAPTOR residues 57–366 (mTOR∆N-RAPTOR∆C), and subsequent refinement of mTOR∆N-RAPTOR∆C yielded a reconstruction at 3.1 Å (Supplementary Table 6).

Cryo-EM structure refinement and validation

The model for one protomer of the apo mTORC1_WT structure (6BCX) was rigid body fit into the locally refined mTORC1_EWEDdup EM protomer density and into the focused refined mTOR∆N-RAPTOR∆C density. The models were then manually adjusted using Coot. The StarMap plugin for ChimeraX [43] was used to generate Rosetta scripts for rebuilding and refining the model [44]. One of the half-maps was lowpass filtered and used for Rosetta refinement. The other half-map was used for FSC validation. The refined model was manually checked in COOT and B-factor refinement was performed using Rosetta. Further real space refinement was carried out in Phenix (version 1a, 4620) [45] with default settings and the following additions: (1) a nonbonded weight of 1000 was used; (2) rotamer outliers were fit with the target ‘fix_outliers’. MolProbity [46] was carried out for validation and manual adjustments were made in COOT followed by re-refinement. Additional validation of the model was performed in Phenix Validation. The above refined protomer model was then rigid-body fit into both protomers of the mTORC1_EWEDdup dimer map, followed by the same refinement and validation procedure (Supplementary Table 6).

Thermal stability assay

Before performing thermal stability assays, defrosted samples were gel filtered on a Superose 6 Increase (10/300) column equilibrated in 50 mM HEPES, pH 7.5, 200 mM NaCl, 1 mM TCEP. Thermal unfolding was followed by differential scanning fluorimetry (DSF) by measuring intrinsic protein fluorescence using a Prometheus NT.48 (NanoTemper Technologies). Aliquots (10 μl) of purified mTORC1_WT or mTORC1_EWEDdup variant were loaded into the Prometheus capillaries (Cat. PR-C002, NanoTemper Technologies), and the fluorescence intensity at 330 nm and 350 nm was recorded as a function of temperature from 10 to 90 °C. The experiment was repeated two times with three replicates per sample. Melting points were calculated using PR.ThermControl.