Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , The C-G2P framework. Mice are HDTV injected with a pool of barcoded perturbation plasmids leading to sleeping beauty (SB)-transposon-mediated stable integration into the genome of hepatocytes. Higher-order combinatorial perturbations drive mosaic liver tumour development in a conceptual 2 n combination space for clonal selection (RUBIX). Direct barcode identification is achieved by linking perturbations to 50-nt barcode sequences that are captured and identified by RTL probes as embedded in the 10X Visium for FFPE platform (PERTURB-CAST). Endogenous transcripts are captured alongside barcodes, hence enabling simultaneous mapping of genotypes (as defined by the presence of perturbations) and phenotypes (as defined by transcriptional signatures) on the same tissue section. b , PERTURB-CAST barcode selection. Transcripts not expressed in murine liver are identified using public databases. Their respective 50-nt RTL-probe capture sequences are used as barcodes detected by redeployed commercially available RTL probes provided with the 10X Visium for FFPE mouse kit . Barcodes derived from chemosensory receptor transcripts are embedded in perturbation plasmids as triplet arrays.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Injection, Selection, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , Frequency of liver tumour samples with at least k potential driver mutations per sample in The Cancer Genome Atlas Program (TCGA) HCC dataset. Potential drivers were defined as either amplification or fusion of known COSMIC oncogenes, or homozygous deletion, nonsense mutation, splice site mutation or frameshift deletion/insertion in tumour-suppressor genes. b , Frequent alterations observed in human liver cancer (The Cancer Genome Atlas Program HCC dataset) are ‘geno-copied’ in a C-G2P mouse model (oncoprint based on https://www.cbioportal.org/study/summary?id=lihc_tcga ). ORF, open reading frame; RNAi, RNA interference. c , RUBIX mouse model generated in this study. Schematic overview of sleeping beauty transposon perturbation plasmids to ectopically overexpress genes of interest (oncogenic-driver perturbations) or shRNA to enable gene knockdown (tumour-suppressor perturbations). Functional elements are highlighted. BC, barcode in which three redeployed RTL-probe capture sequences (as indicated) are embedded; EF1, polymerase II promoter; IR, inverted/direct repeats of sleeping beauty transposon; pA: polyadenylation signal; sh, shRNA embedded in miRE context. Note that we used Visium mouse transcriptome probe set v1 to derive barcodes. Each 50-nt barcode is separated and flanked by spacer sequences of approximately 20 nt to avoid potential steric hindrance during hybridization. Further information in .

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Amplification, Mutagenesis, Generated, shRNA, Knockdown, Functional Assay, Hybridization

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: Schematic overview of Sleeping Beauty transposon perturbation plasmids to ectopically overexpress genes-of-interest (oncogenic-driver perturbations) or shRNA to enable gene knockdown (tumor-suppressor perturbations). Functional elements are highlighted. IR: Inverted/direct repeats of sleeping beauty transposon; GFP: green fluorescent protein, mK2: mKate2 red fluorescent protein, EF1: Polymerase II promoter; U6: Polymerase III promoter; pA: polyadenylation signal; sh: short hairpin RNA embedded in mir-E context; BC: a barcode in which 3 redeployed RTL-probe capture sequences (as indicated) are embedded. Note that we used Visium Mouse Transcriptome Probe Set v1 to derive barcodes. Each 50 nt barcode is separated and flanked by ca. 20 nt spacer sequences to avoid potential steric hindrance during hybridization. Spacer sequences used were derived from T7 and T3 promoters and/or AsCas12a-DR sequences and/or 10X Capture sequences cs1 and cs2 (not shown). Functionality of spacer sequences was not tested in this study. Note that plasmids were equipped with multiple orthogonal barcodes at varying positions. See Methods for further information.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: shRNA, Knockdown, Functional Assay, Hybridization, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , RUBIX with a pool-of-8 plasmid mix results in rapid liver tumour development. Injection of a shRNA targeting Renilla (shRen; matching total plasmid concentration for pool-of-8 mix) served as control, n = 2 each group. Representative H&E-stained samples revealing multiple tumour nodules from the two individual animals are shown. Absence of tumours in the control group (shRen only) indicates that random integration of transposon plasmids itself is unlikely to contribute to tumorigenesis. b , Tissue preprocessing. Following liver tumour development, livers were extracted, divided and processed to FFPE as well as fresh frozen specimens. FFPE samples were initially sectioned to enable sample selection. c , C-G2P liver samples. Overview of three representative FFPE samples used in this study. A total of 513 tumour nodules (red outline) were identified based on histopathological examination (based on H&E). d , Overview of ROIs selected for 10X Visium. 6 segregated regions were selected across 3 FFPE samples. Squares indicate approximate position of ROIs selected for ST. Orange: first 10X Visium run, light orange: first 10X Visium run, replicate ROI; blue: second 10X Visium run; black: 10X Visium CytAssist run. Note overlap between ROIs, where serial sections are used for 10X Visium. e , 10X Visium workflow. Samples for ST are derived directly from FFPE blocks and mounted on 10X Visium slides. f , 10X Visium CytAssist workflow. Samples for ST are derived from sections already mounted on glass slides and transferred to 10X Visium slides using the 10X CytAssist instrument. g , Overview of all samples used for ST. 12 samples from a single RUBIX experiment with two animals were used for 10X Visium in this study. Respective H&E stainings are depicted. Note that the utility of sample ML-II_B_2Cyt is constrained by tissue detachment of the sample during the processing for 10X Visium CytAssist and was not included for further analyses (asterisk). QC summary stats for each sample related to Visium runs performed are provided.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Plasmid Preparation, Injection, shRNA, Concentration Assay, Control, Staining, Selection, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , RUBIX establishes hundreds of coexisting tumours in the context of native tissue. Respective H&E-stained tissue samples for six topographically separated regions (approximately 6 × 6 mm) that were used for 10X Visium for FFPE-spatial transcriptomics analysis. A total of 324 nodules (colour-coded and numbered) were annotated. Colours were chosen arbitrarily. b , PERTUB-CAST allows perturbation-specific barcode identification. Average log(1 p )-transformed expression of all 38 barcode-associated transcripts used (left). Combined data for the reference control liver datasets from ref. and the six main spatial transcriptomics samples (C-G2P). Spatially resolved expression of triplet barcodes (as indicated in Fig. ) for each of the eight perturbations (top right). Aggregated log(1 p )-transformed and quantile-rescaled expression per 10X Visium spot. A representative sample is shown. Average log(1 p )-transformed expression of individual barcodes in each triplet array for each perturbation averaged across the six spatial transcriptomics samples (bottom right). c , Conversion of PERTURB-CAST barcode signals to perturbation maps. Spatially resolved visualization of the inferred probabilities indicating the presence or absence of each of the eight perturbations associated with annotated tumour nodules . A representative sample is displayed. b , c , Both the quantitative barcode expression ( b ) and probabilities ( c ) of all samples can be explored through the interactive web browser https://chocolat-g2p.dkfz.de/ . d , Validation of inferred perturbation integration. A GLM model predicts the phenotype expression signals based on the estimated probabilities of perturbation presence using Bayesian modelling . Phenotypes are defined as direct target transcripts associated with perturbations such as shPten– Pten and NICD– Notch1 . Expression data were log(1 p )-transformed. GS, a well-established marker for active WNT signalling in murine livers, was used to infer mtCtnnb1-GS-positive phenotype via IHC on a corresponding serial section. Baseline depicts background phenotype marker expression. Data are presented as feature coefficients shown as mean and error bars depict 3σ confidence intervals (CIs). Data are derived from 324 nodules across six topographically separated regions used for 10X Visium from a single RUBIX experiment with two animals. Mapping GS IHC data are derived from three corresponding sections from a single RUBIX experiment with two animals. A representative sample section is displayed.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Staining, Transformation Assay, Expressing, Control, Biomarker Discovery, Marker, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , Quantitative reproducibility. Scatterplots of inferred perturbation probabilities for nodules on the primary section to those on the corresponding replica sections, with Pearson’s correlation values displayed. In total, 136 nodule pairs were analysed. b , Matching nodules across samples. Spatial maps of the inferred probabilities (as in Fig. ) indicating the presence or absence of each of the 8 perturbations associated with annotated tumour nodules for all samples that have matching ROIs from a single RUBIX experiment with two animals. Matching nodules were manually annotated (Methods). Addition of “_Cyt” in sample name indicates use of 10X CytAssist. Number of matching nodules is indicated.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques:

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a – c , Perturbation–phenotype association. Spatially resolved visualization of the inferred probabilities indicating the presence or absence of each of the 8 perturbations associated with annotated tumour nodules (Methods). a , A representative sample is displayed. A generalized linear model (GLM) predicts phenotype expression signals based on the estimated probabilities of perturbation presence (Methods). b , Phenotypes are defined as direct target transcripts associated with perturbations such as shKmt2c-Kmt2c and shTrp53-Trp53. Expression data are log1p-transformed. Note that shTrp53 is linked to a GFP peptide barcode and shKmt2c is linked to a RFP barcode (Extended Data Fig. ). c , Hence we infer shTrp53-GFP-positive phenotype and shKmt2c-RFP-positive phenotype. Representative IHCs for a corresponding ROI on a serial section. Baseline depicts background phenotype marker expression. Data are presented as feature coefficients shown as mean and error-bars depict 3σ confidence intervals. As in Fig. , and d . d , H&E and IHC for GFP, RFP, GS. Three representative FFPE samples from a single RUBIX experiment with two animals were sectioned and stained for H&E (see Extended Data Fig. ). GFP and RFP IHC staining was performed on two individual serial sections. GS IHC staining was performed on serial sections. GFP and RFP were embedded in perturbation plasmids as orthogonal barcodes (see Methods and Extended Data Fig. ). GS is a well-known marker for liver WNT/mtCtnnb1-signalling activity (see Fig. ). e , GFP and RFP detection for the same hepatocellular tumour nodules to spatially map peptide barcode combinations associated with introduced perturbations. Left: H&E. Zoom-in: GFP and RFP respectively. Note that one tumour nodule is positive for RFP alone whereas the other tumour nodule is positive for both GFP as well as RFP peptide barcodes. A representative example is shown. Data is derived from 324 nodules across 6 topographically separated regions used for 10X Visium from a single RUBIX experiment with two animals. Mapping GFP and RFP IHC data is derived from 3 corresponding sections from a single RUBIX experiment with two animals.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Expressing, Transformation Assay, Marker, Staining, Immunohistochemistry, Activity Assay, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , Genotype maps (right); 2 8 powerset embedding of spatially mapped perturbations encompassing 324 nodules across six topographically separated regions. Each of the 256 combinations is colour-coded. Perturbation probabilities for a representative nodule are depicted (left; black text highlights present perturbations, while grey text highlights absent perturbations). Nodules sharing representative similar genotypes are encircled and indicated in b (solid lines, Myc + mtCtnnb1 + NICD; dashed lines, Myc + mtCtnnb1 + shKmt2c + NICD + shRen + shTrp53 + shPten). b , Clonal selection. Observed occurrences of genotypically defined tumour clones (median and 95% CI) across 2 8 powerset embedding (top). Grey text indicates combinatorial complexity. The highlighted genotypes are encircled in a . Probability p (O > E) that observed occurrences (O) deviate from the expected baseline distribution (E) ( ; bottom). Deviations of >0.5 indicate increased tumorigenic potential (orange), whereas values of <0.5 suggest potentially disadvantageous combinations (blue). c , Combinatorial order distribution. Observed distribution of the perturbation integration order (mean and 95% CI). A binomial distribution with p = 0.5 is included as a reference of a random unbiased integration rate (red line). d , Ranking of cancer-driving perturbations. Marginal frequencies of individual perturbations in descending order (mean and 95% CI). e , Pairwise co-occurrence and mutual exclusivity patterns. An OR > 1 suggests co-occurrence, whereas OR < 1 indicates mutual exclusivity . Perturbations are ordered according to d . f , Identification of pairwise genetic interactions. Comparison of observed versus expected frequencies (median and 95% CI) for selected gene pairs, calculated using multiplicative models of gene interaction. We simulated the expected probabilities for the pairwise groups under the assumption of no interaction OR, which indicates the direction of the gene interaction effect (arrows), are reported along with the corresponding P values. OR values were estimated from 5,000 posterior samples. A softmax GLM with interaction fixed at one defined the null. P values reflect two-tailed deviations of observed double-positive proportions from the null based on 5,000 draws . Data are derived from 324 nodules across six topographically separated regions used for 10X Visium from a single RUBIX experiment with two animals. Bayesian modelling of perturbation probabilities was used to infer the occurrence of individual perturbation combinations per nodule. From the inferred Bayesian posterior, we sampled 5,000 points and computed the median and CI for the frequencies of individual perturbations as well as individual genotypes and calculated the OR . H0, null hypothesis.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Selection, Clone Assay, Comparison, Two Tailed Test, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , The tumour ecosystem. Spatial maps of tumour-intrinsic phenotypes (top) and TME phenotypes (bottom) across six topographically separated regions. Colour shade depicts aggregated log(1 p )-transformed expression of phenotype-associated transcripts (colour-code as in b and d ). Nodule borders are highlighted (grey). The aggregated values for all samples and underlying quantitative data of individual transcript expression can be explored through the interactive web browser interface ( https://chocolat-g2p.dkfz.de/ ). b , Co-clustering of tumour-intrinsic phenotypes by associated transcripts. Tumour phenotypes are colour-coded. c , Associations between tumour-intrinsic and TME phenotypes. Pearson’s correlation coefficient for each pair of tumour-intrinsic and TME phenotype-associated transcripts across all nodules. d , Co-clustering of TME phenotypes by associated transcripts. TME phenotypes are colour-coded. b , d , Clustering based on Spearman correlations. Phenotypes are subdivided using hierarchical clustering. Scaled ( p 10 ) estimated plasmid probabilities per nodule are indicated ( b (bottom) and d (left) . Data are derived from 324 nodules across six topographically separated regions used for 10X Visium from a single RUBIX experiment with two animals.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Transformation Assay, Expressing, Plasmid Preparation, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , Identification of genotype–phenotype relationships. Comparison of the prevalence of perturbations between phenotypic groups and the remainder of the nodules (total n = 324) for tumour-intrinsic phenotypes (top) and TME (bottom) using ORs. OR > 1 indicates enrichment of perturbations within the phenotypic group; OR < 1 indicates depletion . The number of nodules with a given phenotype ( n ) is indicated. Note that groups are not mutually exclusive. The median and 90% CI are reported. Significant relationships are indicated (exact P values are provided); two-tailed deviations from one, computed with 20,000 samples from the posterior ; *** P < 0.001, ** P < 0.01, * P < 0.05. b , Identification of genotype–phenotype relationships for genes associated with cholangiocytes. A GLM model predicts gene expression signals at each 10X Visium spot using estimated probabilities of perturbation presence . Feature coefficients, shown as the mean and 3σ CIs, indicate associations between gene expression and perturbations for representative transcripts. Bayesian modelling of perturbation probabilities was used to infer the occurrence of individual perturbations per nodule . Data are derived from 324 nodules across six topographically separated regions used for 10X Visium from a single RUBIX experiment with two animals.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Comparison, Two Tailed Test, Gene Expression, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , Tumour-intrinsic genotype–phenotype relations. A generalized linear model (GLM) predicts gene expression signals at each 10X Visium spot, using estimated probabilities of perturbation presence (Methods). Data are presented as feature coefficients shown as mean and error bars depict 3σ confidence intervals. Feature coefficients indicate associations between gene expression and perturbations for representative transcripts of four tumour-intrinsic phenotypes. b , TME-related genotype–phenotype relations. As in a for representative transcripts of two exemplary TME phenotypes. c , GLM-inferred genotype–phenotype associations. Top: heatmap of 1,283 genes with at least one significant (3σ) feature weight, ordered by 1D UMAP embedding. Bottom: Detailed views of four representative clusters linked to marker genes of known phenotypic groups. Data is derived from 324 nodules across 6 topographically separated regions used for 10X Visium from a single RUBIX experiment with two animals.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Gene Expression, Marker, Derivative Assay

Journal: Nature Biomedical Engineering

Article Title: Integrated in vivo combinatorial functional genomics and spatial transcriptomics of tumours to decode genotype-to-phenotype relationships

doi: 10.1038/s41551-025-01437-1

Figure Lengend Snippet: a , Spatially resolved co-occurence of VEGFA and mutual exclusivity of mtCtnnb1 for the CCA tumour subtype as revealed by C-G2P. Magnified views of three representative nodules ((i)–(iii)) identified as CCA. Nodules identified as CCA (left; the area covered by the tumour nodule is indicated as 10X Visium spots in yellow) as well as the mtCtnnb1 (middle; as in Fig. ) and VEGFA (as in Fig. ) perturbation probabilities are shown. Bayesian modelling of perturbation probabilities is used to infer the occurrence of individual perturbations per nodule . Data are derived from 324 nodules across six topographically separated regions used for 10X Visium from a single RUBIX experiment with two animals. Perturbation probabilities for all samples can be explored through the interactive web browser https://chocolat-g2p.dkfz.de/ . b , Experimental design. Parallel RUBIX mouse models were performed using the leave-one-out experimental design. c , Time to tumour occurrence. Animals were palpated twice weekly to monitor tumour development. d , Histological quantification of liver tumour subtypes. H&E images were analysed, and tumour nodules were counted and classified as either HCC (top) or CCA (bottom); two independent liver-tissue sections per animal. The median ± s.d. alongside individual tumour counts are indicated. Group comparisons used a two-sided Kruskal–Wallis test with Dunn’s post-hoc test (Holm–Bonferroni correction). Exact adjusted P values are shown. e , Abundance of CCA. CK19 IHC was used as a cholangiocyte marker. Representative samples from a total of two separate sections per animal are depicted. b – e , n = 4 animals per group.

Article Snippet: The endogenous transcripts associated with the REDPRO-BCs used in this study are illustrated in Extended Data Fig. and the respective nucleotide sequences for 10X Visium RTL-probe capture barcodes (reverse complement to RTL-probe sequence provided by 10X Genomics) are listed under the section ‘Molecular cloning’.

Techniques: Derivative Assay, Marker