TROP2 targeting reveals therapy-driven cell state dynamics in colorectal cancer

TROP2 targeting reveals therapy-driven cell state dynamics in colorectal cancer Vaquero-Siguero · Georgakopoulos · Puschhof et al. — Nature, 1 July 2026 · Open Access 10.1038/s41586-026-10705-2

TL;DR — A ≈152,000-cell single-cell RNA-seq atlas (7 studies · 55 samples · human + mouse) mapping how colorectal-cancer cells switch between WNT-high LGR5⁺ intestinal-stem and WNT-low TROP2⁺ fetal-like states — and how conventional chemotherapy induces the TROP2⁺ state while a TROP2 antibody–drug conjugate (Sacituzumab Govitecan) depletes it. Ships baseline maps, drug time courses, pseudotime trajectories, and figure-reproducing code.


0. Summary

Biological premise. Metastasis remains the leading cause of cancer-related mortality and is driven by pronounced tumour-cell plasticity. Colorectal cancer is the second leading cause of cancer-associated death worldwide; roughly 90% of cases are microsatellite-stable and are treated with combination chemotherapy — 5-fluorouracil together with irinotecan (FOLFIRI) or with oxaliplatin (FOLFOX) — yet the five-year survival for stage IV disease remains around 12%, underscoring a persistent unmet need in advanced disease.

This study identifies the transmembrane glycoprotein trophoblast cell-surface antigen 2 (TROP2) as a marker of poor-prognosis colorectal cancer. TROP2 marks WNT-low, fetal-like tumour-cell states that are linked to metastasis and therapy resistance; functionally, TROP2-positive cells show context-dependent stem-like capacity and the ability to initiate metastatic outgrowth. Because these detrimental states converge on a single cell-surface antigen, the authors test therapeutic targeting with clinically relevant TROP2-directed antibody–drug conjugates (Sacituzumab Govitecan).

Time-resolved single-cell profiling shows that tumours shift dynamically between WNT-high, LGR5-positive intestinal-stem states and WNT-low, TROP2-positive fetal-like states. Conventional chemotherapy promotes the induction of TROP2-expressing cells, whereas the TROP2 antibody–drug conjugate selectively depletes them and remodels the cell-state landscape. Exploiting this plasticity, combined chemotherapy and TROP2 targeting enhances anti-tumour efficacy in patient-derived models — nominating TROP2 as a therapeutic vulnerability and highlighting cell-state–directed therapy as a route to overcome resistance in advanced colorectal cancer.

Every dataset in this collection is a single-cell RNA-sequencing readout of that state landscape — sampled across patient tumours, patient-derived organoids and organoid xenografts, and genetically engineered mouse models, captured at baseline or along drug time courses.

Assay and processing. Single-cell RNA-seq (10x Chromium 3' v3.1, TotalSeq-B cell hashing for multiplexing). Processing: Cell Ranger (7.1.0 multi / 6.1.0 count) → scanpy 1.9.3 / Seurat (QC, scrublet doublet removal, HVG, Harmony integration, Leiden clustering, UMAP), MAGIC imputation, Palantir 1.4.0 trajectory inference, Decoupler + PyDESeq2 pseudobulk DE, CellTypist annotation. All 7 studies ship raw integer count matrices (Tier-1) plus log1p-normalized matrices, per-cell metadata with cluster/state annotations, feature lists, and UMAP embeddings; two studies additionally ship Palantir .h5ad objects and one ships a MAGIC-imputed matrix.

Sources.

ArrayExpress / BioStudies studies (processed count matrices, metadata, embeddings; SDRF/IDF):

  • E-MTAB-16583 — Primary CRC tumour tissue + liver metastasis (patient biobank)

  • E-MTAB-16585 — PDOX colorectal tumours (biobank subcutaneous)

  • E-MTAB-16433 — PDOX tumours treated with vehicle or SG (Sacituzumab Govitecan)

  • E-MTAB-16849 — SG treatment time course of liver-metastasis models

  • E-MTAB-16843 — SG vs untargeted IgG1-SN-38 time course of PDOs in vitro

  • E-MTAB-16836 — FOLFIRI chemotherapy time course of PDOs

  • E-MTAB-16835 — Murine MDO-derived tumours (VAKPS and VKPN)

  • Code: github.com/ORCA-HD/TROP2_in_CRC (GPL-3.0; Jupyter notebooks + utils reproducing the figures)

  • Nature supplements & source data: Supplementary Information (MOESM1), Reporting Summary (MOESM2), Supplementary Tables 1–11 (MOESM3), Source Data Fig. 5 (MOESM4), Source Data Extended Data Fig. 12 (MOESM5)

  • PMIDs: none assigned yet (Nature, published 01 July 2026)

DiseaseColorectal cancer, including liver metastasis
Assay / modalitySingle-cell RNA-sequencing (scRNA-seq) — 10x Genomics Chromium Next GEM Single Cell 3′ Reagent Kit v3.1 (dual index) + TotalSeq-B cell hashing
OrganismsHomo sapiens; Mus musculus
Model systemsPatient tissue; patient-derived organoid (PDO); patient-derived organoid xenograft (PDOX); mouse-derived organoid (MDO) from a genetically engineered mouse model (GEMM)
RepositoriesArrayExpress / BioStudies (EMBL-EBI); GitHub; Springer Nature Electronic Supplementary Material
AccessionsE-MTAB-16583, E-MTAB-16585, E-MTAB-16433, E-MTAB-16849, E-MTAB-16843, E-MTAB-16836, E-MTAB-16835
AccessPublic. Processed matrices are mirrored in this dataset's input/ folder; raw FASTQ reads are brokered to the European Nucleotide Archive (ENA) and are not mirrored.
Number of studies7
Number of files121
Total size72,783,904,535 bytes (≈ 72.78 GB)
Biological samples55 — throughout this document, "sample" = one ArrayExpress SDRF Source Name (one profiled specimen/library). Counts per study: 9 + 26 + 4 + 6 + 4 + 4 + 2.
SequencerIllumina NovaSeq 6000
Reference build (human)refdata-gex-GRCh38-2020-A
Reference build (mouse)refdata-gex-mm10-2020-A
Code licenseGPL-3.0 — ORCA-HD/TROP2_in_CRC

1. Datasets — sample sources, protocols, experiment design and metadata

Landed sources at a glance.

SourceAccessFiles
ArrayExpress / BioStudies (7 studies)public / OPEN-processed88
GitHub ORCA-HD/TROP2_in_CRCpublic (GPL-3.0)28
Nature supplements & source datapublic5

The seven studies at a glance.

AccessionDescriptionOrganismSamplesScale
E-MTAB-16583Primary CRC tumour tissue + liver metastasis (patient biobank)human9 samples785 epithelial / 6,429 total cells
E-MTAB-16585PDOX colorectal tumours (biobank subcutaneous)human26 samples10,135 epithelial / 12,020 total cells
E-MTAB-16433PDOX tumours treated with vehicle or SG (Sacituzumab Govitecan)human4 samples7,330 cells
E-MTAB-16849SG treatment time course of liver-metastasis modelshuman6 samples0–48 h 11,385 / 48–120 h 13,984 / mets 44,779 cells
E-MTAB-16843SG vs untargeted IgG1-SN-38 time course of PDOs in vitrohuman4 samples46,906 cells
E-MTAB-16836FOLFIRI chemotherapy time course of PDOshuman4 samplesLM2 2,727 / LM3 1,173 cells
E-MTAB-16835Murine MDO-derived tumours (VAKPS and VKPN)Mus musculus2 samples (2 genotypes; 6 mice pooled)1,573 epithelial / 4,951 total cells

1.1 Sample origins and model-system lineage

All material derives from two roots — a human patient biobank and a set of genetically engineered mouse models — which branch into four model systems:

Human root (Heidelberg patient biobank; informed consent, ethics S-136/2021 and S-871/2021):

  • Patient tissue — primary tumour or liver-metastasis biopsy, profiled directly → E-MTAB-16583
  • Patient-derived organoid (PDO) — organoid grown in vitro from a patient tumour → E-MTAB-16843, E-MTAB-16836
  • Patient-derived organoid xenograft (PDOX) — a PDO engrafted into a mouse → E-MTAB-16585, E-MTAB-16433, E-MTAB-16849

Mouse root (genetically engineered mouse models; DKFZ):

  • Mouse-derived organoid (MDO) — organoids from genetically defined mouse tumours, grown as subcutaneous allografts → E-MTAB-16835

Human donors are de-identified with HD#### codes; clinicopathology (age, sex, site, stage, driver mutations) lives in the SDRF Characteristics[…] fields, in each sample_metadata.txt, and in Supplementary Tables 1–4.

1.2 Shared wet-lab protocol (bench → sequencer)

Applies to all seven datasets: E-MTAB-16583, E-MTAB-16585, E-MTAB-16433, E-MTAB-16849, E-MTAB-16843, E-MTAB-16836 and E-MTAB-16835.

Each ArrayExpress Investigation Description Format (IDF) file registers six or seven protocol steps (P-MTAB-…): sample collection → nucleic-acid extraction → library construction → sequencing → (treatment, where applicable) → normalization → alignment. In bench terms, the pipeline runs from tumour tissue / organoid / allograft through to paired FASTQ:

  1. Dissociation — enzymatic dissociation (Miltenyi Tumour Dissociation Kit, gentleMACS) to a single-cell suspension.
  2. Sorting — fluorescence-activated cell sorting selects live cells (or EPCAM-positive epithelium for xenografts) and excludes dead cells (DAPI / Zombie NIR).
  3. Cell hashing — stain with TotalSeq-B hashtag-oligo (HTO) antibodies so several samples can be pooled into one 10x lane.
  4. Library construction — 10x Genomics Chromium Next GEM Single Cell 3′ v3.1 (dual index): droplet capture, reverse transcription, cDNA. Library geometry from the SDRF: cell barcode 16 bp, unique molecular identifier (UMI) 12 bp, cDNA read 91 bp; PAIRED layout, PolyA selection, TRANSCRIPTOMIC SINGLE CELL source, 3′ tag.
  5. Sequencing — Illumina NovaSeq 6000 → paired FASTQ (deposited in ENA; per-run URIs listed in each SDRF).

Multiplexing rationale. Hashtags let several biological samples share one lane and be separated computationally afterwards, which is why most datasets ship an HTO_library.txt / feature_ref*.csv reference and carry HTO_classification columns in their per-cell metadata.

1.3 Shared data-processing protocol (reads → cell-by-gene tables)

Applies to all seven datasets: E-MTAB-16583, E-MTAB-16585, E-MTAB-16433, E-MTAB-16849, E-MTAB-16843, E-MTAB-16836 and E-MTAB-16835. The specialised trajectory and imputation steps apply to a subset: Palantir pseudotime to E-MTAB-16849 and E-MTAB-16836; MAGIC imputation to E-MTAB-16849.

The pipeline runs left to right — FASTQ → align & count → de-multiplex → quality control → normalise & integrate → cluster, embed & annotate → specialised — and each step leaves a recognisable fingerprint in the deposited files:

StepTool (version)Fingerprint in the files
Align + countCell Ranger v7.1.0 (multi) / v6.1.0 (count)*_counts.txt (raw), *_features.txt
De-multiplex hashtagsCell Ranger multi (confidence 0.9) / Seurat HTODemuxHTO_classification* columns in *_metadata
Quality controlscanpy 1.9.3 / Seurat; scrublet 0.2.3retained cells listed in *_metadata
Normalise + integratelog1p; harmonypy 0.0.9*_counts_log1p.txt; *_woCC.txt (cell-cycle regressed)
Cluster + embedleidenalg 0.10.1 → UMAPleiden column; *_umap.txt
Annotatemarkers + CellTypist 1.6.3cell-state labels in *_metadata
Trajectory / imputePalantir 1.4.0; MAGIC 3.0.0*_palantir.h5ad; *_magic.csv
Pseudobulk differential expressionDecoupler 1.8.0 + PyDESeq2 0.4.10reproduced by the GitHub notebooks

E-MTAB-16583 — Primary colorectal tumour tissue and liver metastasis

ArrayExpress record · 12 files · 1.71 GB

  • Sample source. Fresh primary colorectal-cancer biopsies and liver-metastasis biopsies taken directly from patients (native tumour genotype, characterised by whole-exome sequencing in Supplementary Table 4). Model system: patient tissue.

  • Samples and retention. 9 patient biopsies (individuals HD4181, HD4236, HD4241, HD4246, HD4249, HD4254, HD4272, HD4309, HD4362; a mix of primary rectum/colon and liver metastasis), pooled and multiplexed across libraries (4 sequencing runs). After Cell Ranger + quality control → 6,429 cells retained in the full atlas, of which 785 are tumour epithelium (≈ 12% epithelial gating for the state analysis).

  • Experiment design. Observational baseline; design axis = tissue site (primary tumour vs liver metastasis). Purpose: define the TROP2-positive population in real patient tumours.

  • Metadata available. Sample level → sample_metadata.txt (organism, developmental stage, disease, organism part, individual, sex, age, cell type, sampling site). Cell level → barcode, sample_id, leiden, plus HTO_classification.

  • Files (annotated).

    FileWhat it is
    E-MTAB-16583.idf.txtStudy design, protocol descriptions (P-MTAB-…), contacts, publication
    E-MTAB-16583.sdrf.txtPer-sample↔run table: characteristics, 10x geometry, ENA / FASTQ links
    HTO_library.txtHashtag reference used to de-multiplex the pooled samples
    sample_metadata.txtPer-patient clinical annotation (9 biopsies)
    biobank_tissue_full_counts.txtRaw integer count matrix — all cell types (6,429 cells)
    biobank_tissue_full_features.txtGene identifiers (matrix rows) for the full atlas
    biobank_tissue_full_metadata.txtPer-cell labels for the full atlas
    biobank_tissue_full_counts_log1p.txtLog-normalised full matrix
    biobank_tissue_epi_counts.txtRaw counts — tumour-epithelial subset (785 cells)
    biobank_tissue_epi_features.txtGene identifiers for the epithelial subset
    biobank_tissue_epi_metadata.txtPer-cell labels for the epithelial subset
    biobank_tissue_epi_counts_log1p.txtLog-normalised epithelial matrix
  • How this experiment became a dataset. 9 patient biopsies → dissociated, hashed, pooled → 10x + NovaSeq → Cell Ranger count → scanpy quality control → two-level atlas (all cells + epithelial subset) with raw counts, log-normalised counts, gene lists and per-cell labels.

  • Completeness (scRNA-seq). Tier 1 (raw counts, gene features, per-cell metadata): complete. Tier 2 (log-normalised matrices, sample metadata): complete; no UMAP embedding deposited for this dataset. Tier 3 (raw reads, reference genome): reads brokered to ENA (not mirrored); human reference build linked above.


E-MTAB-16585 — Patient-derived organoid xenograft (PDOX) tumours (biobank baseline)

ArrayExpress record · 14 files · 5.31 GB

  • Sample source. Patient-derived organoids injected subcutaneously into mice, grown into tumours, then profiled. Model system: patient-derived organoid xenograft (PDOX).

  • Samples and retention. 26 xenograft tumours from 7 donors (HD42363, HD42466, HD42541, HD42722, HD42775, HD43094, HD43627), pooled into three hashed libraries (12 sequencing runs). After processing → 12,020 cells retained in the full atlas, of which 10,135 are tumour epithelium (≈ 84% epithelial — xenografts are epithelium-rich). The 10,135-cell epithelial map corresponds to Figure 1f.

  • Experiment design. Observational baseline; design axis = sample pooling across donors. Purpose: characterise the LGR5 ⇄ TROP2 state structure in a controlled xenograft setting.

  • Metadata available. Sample level → sample_metadata.txt (adds Xenograft identifier per tumour). Cell level → barcode, sample_id, donor_id, library_id, batch, leiden.

  • Files (annotated).

    FileWhat it is
    E-MTAB-16585.idf.txt / .sdrf.txtStudy design; per-sample↔run metadata + ENA/FASTQ links
    HTO_library.txtHashtag reference (3 pooled libraries)
    sample_metadata.txtPer-tumour annotation (26 xenografts, 7 donors)
    biobank_subcut_full_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, all cell types (12,020)
    biobank_subcut_full_counts_log1p.txtLog-normalised full matrix
    biobank_subcut_full_umap.txtUMAP coordinates, full atlas
    biobank_subcut_epi_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, epithelial subset (10,135)
    biobank_subcut_epi_counts_log1p.txtLog-normalised epithelial matrix
    biobank_subcut_epi_umap.txtUMAP coordinates, epithelial subset (Figure 1f)
  • How this experiment became a dataset. 26 PDOX tumours from 7 donors → hashed, pooled → 10x + NovaSeq → Cell Ranger multi de-multiplex → scanpy quality control → integrated full + epithelial atlases with raw/normalised matrices, gene lists, labels and UMAP embeddings.

  • Completeness (scRNA-seq). Tier 1: complete. Tier 2 (log-normalised matrices, UMAP embeddings, sample metadata): complete for both full and epithelial atlases. Tier 3: reads brokered to ENA (not mirrored); human reference build linked above.


E-MTAB-16433 — PDOX tumours treated with vehicle or Sacituzumab Govitecan

ArrayExpress record · 8 files · 1.68 GB

  • Sample source. Patient-derived organoids (line HD4246) grown as subcutaneous xenografts, then dosed in vivo. Model system: patient-derived organoid xenograft (PDOX).

  • Samples and retention. 4 samples (vehicle vs Sacituzumab Govitecan, with replicates; 8 sequencing runs). Tumours reached 100 mm³, were treated twice weekly for 28 days, then harvested. After processing → 7,330 tumour cells retained (corresponds to Figure 3h).

  • Experiment design. Interventional; design axes = compound · dose · replicate (the richest ArrayExpress Factor Value set among these seven studies). Purpose: measure how a TROP2-directed antibody–drug conjugate reshapes the state landscape in vivo.

  • Metadata available. Cell level → barcode, sample_id, library_id, treatment, mouse_ID, batch, leiden. Design factors (Factor Value[compound | dose | replicate], with concentration unit) are in the SDRF.

  • Files (annotated).

    FileWhat it is
    E-MTAB-16433.idf.txt / .sdrf.txtStudy design (incl. treatment protocol); per-sample↔run metadata + ENA/FASTQ links + compound/dose factors
    HTO_library.txtHashtag reference
    treatment_HD4246_SG_counts.txtRaw integer count matrix (7,330 cells)
    treatment_HD4246_SG_features.tsvGene identifiers (matrix rows)
    treatment_HD4246_SG_metadata.tsvPer-cell labels incl. treatment (vehicle vs SG)
    treatment_HD4246_SG_counts_log1p.txtLog-normalised matrix
    treatment_HD4246_SG_umap.txtUMAP coordinates
  • How this experiment became a dataset. PDOX tumours dosed with vehicle or Sacituzumab Govitecan → hashed, pooled → 10x + NovaSeq → Cell Ranger multi → scanpy quality control → one treatment-annotated epithelial matrix with raw/normalised counts, genes, labels and UMAP.

  • Completeness (scRNA-seq). Tier 1: complete. Tier 2 (log-normalised matrix, UMAP embedding): complete; sample-level design captured in the SDRF Factor Value fields. Tier 3: reads brokered to ENA (not mirrored); human reference build linked above.


E-MTAB-16849 — Liver-metastasis time course under Sacituzumab Govitecan

ArrayExpress record · 19 files · 49.60 GB

  • Sample source. Human colorectal-cancer organoids (line HD4246) injected intrasplenically to seed liver metastases in mice, then treated. Model system: patient-derived organoid xenograft (PDOX), liver-metastasis model.

  • Samples and retention. 6 samples across a time course (0 h, 6 h, 24 h, 48 h, 72 h, 120 h; Sacituzumab Govitecan vs non-targeting IgG1-SN-38 control; 18 sequencing runs). Delivered as three cell-by-gene datasets: 0h–48h → 11,385 cells (Figure 3n), 48h–120h → 13,984 cells, metastasis map → 44,779 cells.

  • Experiment design. Interventional time course; design axis = single-cell library / timepoint. Purpose: trace, hour by hour, how tumour cells shift state under a TROP2 antibody–drug conjugate. Includes trajectory objects for pseudotime.

  • Metadata available. Cell level → barcode, sample_id, line, treatment, timepoint, replicate, library, leiden, batch. Trajectory metadata (pseudotime, terminal states) lives inside the *_palantir.h5ad objects.

  • Files (annotated).

    FileWhat it is
    E-MTAB-16849.idf.txt / .sdrf.txtStudy design; per-sample↔run metadata + ENA/FASTQ links
    feature_reference.csvHashtag / feature-barcode reference
    HD4246_SG_0h-48h_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, 0–48 h window (11,385)
    HD4246_SG_0h-48h_counts_log1p_woCC.txtLog-normalised, cell-cycle-regressed matrix, 0–48 h
    HD4246_SG_0h-48h_umap.txtUMAP coordinates, 0–48 h
    HD4246_SG_0h-48h_palantir.h5adAnnData with Palantir pseudotime + MAGIC layer, 0–48 h
    HD4246_SG_48h-120h_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, 48–120 h window (13,984)
    HD4246_SG_48h-120h_counts_log1p_woCC.txtLog-normalised, cell-cycle-regressed matrix, 48–120 h
    HD4246_SG_48h-120h_umap.txtUMAP coordinates, 48–120 h
    HD4246_SG_48-120h_palantir.h5adAnnData with Palantir pseudotime, 48–120 h
    HD4246_SG_mets_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, metastasis map (44,779)
    HD4246_SG_mets_magic.csvMAGIC-imputed expression matrix for the metastasis map (largest file, ≈ 23.38 GB)
  • How this experiment became a dataset. Liver-met–bearing mice dosed with Sacituzumab Govitecan vs control across six timepoints → hashed, pooled → 10x + NovaSeq → Cell Ranger multi → scanpy quality control → three time-resolved matrices, plus Palantir trajectory objects and a MAGIC-imputed expression table for smooth pseudotime gene trends.

  • Completeness (scRNA-seq). Tier 1: complete for all three windows. Tier 2 (log-normalised matrices, UMAP embeddings): complete for the 0–48 h and 48–120 h windows; the metastasis map ships raw counts plus a MAGIC-imputed matrix rather than a UMAP. Specialised: Palantir trajectory objects provided. Tier 3: reads brokered to ENA (not mirrored); human reference build linked above.


E-MTAB-16843 — Organoid time course: Sacituzumab Govitecan vs untargeted control

ArrayExpress record · 8 files · 10.57 GB

  • Sample source. Human colorectal-cancer organoids (line HD4246) treated in vitro. Model system: patient-derived organoid (PDO).

  • Samples and retention. 4 samples (8 sequencing runs) spanning on-treatment timepoints {0 h, 3 h, 6 h, 9 h, 12 h} and post-washout timepoints {12 h + 12 h, 12 h + 36 h, 12 h + 60 h}, comparing Sacituzumab Govitecan against non-targeting IgG1-SN-38. After processing → 46,906 cells retained (the largest single epithelial map in the atlas).

  • Experiment design. Interventional time course with drug washout; design axis = library / timepoint. Purpose: isolate the drug-specific (versus payload-specific) effect and test reversibility of the state shift.

  • Metadata available. Cell level → barcode, sample_id, treatment, timepoint, leiden (with HTO_classification). Control comparator IgG1-SN-38 encoded in treatment.

  • Files (annotated).

    FileWhat it is
    E-MTAB-16843.idf.txt / .sdrf.txtStudy design; per-sample↔run metadata + ENA/FASTQ links
    feature_ref.csvHashtag / feature-barcode reference
    HD4246_SG_org_counts.txtRaw integer count matrix (46,906 cells)
    HD4246_SG_org_features.txtGene identifiers (matrix rows)
    HD4246_SG_org_metadata.txtPer-cell labels incl. treatment + timepoint
    HD4246_SG_org_counts_log1p.txtLog-normalised matrix
    HD4246_SG_org_umap.txtUMAP coordinates
  • How this experiment became a dataset. Organoids treated with Sacituzumab Govitecan vs IgG1-SN-38 across on-treatment and washout timepoints → hashed, pooled → 10x + NovaSeq → Cell Ranger multi → scanpy quality control → one large time-annotated matrix with raw/normalised counts, genes, labels and UMAP.

  • Completeness (scRNA-seq). Tier 1: complete. Tier 2 (log-normalised matrix, UMAP embedding): complete. Tier 3: reads brokered to ENA (not mirrored); human reference build linked above.


E-MTAB-16836 — Organoid time course under FOLFIRI chemotherapy

ArrayExpress record · 16 files · 3.17 GB

  • Sample source. Two human colorectal-cancer organoid lines (LM2, LM3) derived from liver metastases of the same patient, treated in vitro. Model system: patient-derived organoid (PDO).

  • Samples and retention. 4 samples (4 sequencing runs) across on-treatment timepoints {0, 12, 24, 48, 72, 120 h} and washout timepoints {12, 24, 48, 72 h} after a 72 h treatment. After processing → LM2: 2,727 cells; LM3: 1,173 cells retained (Figure 4).

  • Experiment design. Interventional chemotherapy time course; design axis = library / timepoint. Purpose: show that chemotherapy induces the TROP2-positive state (the mechanistic counterpoint to Sacituzumab Govitecan depletion).

  • Metadata available. Cell level → barcode, sample_id, Timepoint, leiden_0.5, HTO_classification, HTO_classification.global.

  • Files (annotated).

    FileWhat it is
    E-MTAB-16836.idf.txt / .sdrf.txtStudy design; per-sample↔run metadata + ENA/FASTQ links
    Feature_ref_multiplexing.csvHashtag / feature-barcode reference
    Org_chemo_LM2_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, line LM2 (2,727)
    Org_chemo_LM2_counts_log1p.txtLog-normalised LM2 matrix
    Org_chemo_LM2_counts_log1p_woCC.txtLog-normalised, cell-cycle-regressed LM2 matrix
    Org_chemo_LM2_umap.txtUMAP coordinates, LM2
    Org_chemo_LM3_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, line LM3 (1,173)
    Org_chemo_LM3_counts_log1p.txtLog-normalised LM3 matrix
    Org_chemo_LM3_counts_log1p_woCC.txtLog-normalised, cell-cycle-regressed LM3 matrix
    Org_chemo_LM3_umap.txtUMAP coordinates, LM3
    Org_chemo_LM3_palantir.h5adAnnData with Palantir pseudotime (LM3 trajectory)
  • How this experiment became a dataset. Two matched organoid lines treated with FOLFIRI across on-treatment and washout timepoints → hashed → 10x + NovaSeq → Cell Ranger → scanpy quality control → per-line matrices (raw, log-normalised, cell-cycle-regressed) with labels, UMAP, and a Palantir trajectory for LM3.

  • Completeness (scRNA-seq). Tier 1: complete for both lines. Tier 2 (log-normalised and cell-cycle-regressed matrices, UMAP embeddings): complete for both lines. Specialised: Palantir trajectory object for LM3. Tier 3: reads brokered to ENA (not mirrored); human reference build linked above.


E-MTAB-16835 — Murine colorectal-cancer tumours (genetically engineered models)

ArrayExpress record · 11 files · 726.9 MB

  • Sample source. Mouse-derived organoids from two genetically engineered adenocarcinoma models, grown as subcutaneous allografts. Model system: mouse-derived organoid (MDO) / genetically engineered mouse model (GEMM).

    • VAKPS = Villin1-CreER; Apc^fl/+; Kras^G12D/+; Trp53^fl/fl; Smad4^fl/fl
    • VKPN = Villin1-CreER; Kras^G12D/+; Trp53^fl/fl; Rosa26^N1icd/+
  • Samples and retention. 2 samples (one per genotype: VAKPS, VKPN; 2 sequencing runs), pooling 6 mice (3 per genotype: AKPS_NV16_4/5/6, KPN_NV14_1/2/3). After processing → 4,951 cells retained in the full atlas, of which 1,573 are tumour epithelium (≈ 32%; the 1,573-cell epithelial map corresponds to Figure 1k).

  • Experiment design. Observational cross-genotype comparison; design axis = single-cell library construction. Purpose: test whether the human LGR5 ⇄ TROP2 program is conserved in genetically defined mouse tumours.

  • Metadata available. Cell level → barcode, sample_id, library_id, line, replicate, leiden, HTO_classification, HTO_classification.global, HTO_maxID.

  • Files (annotated).

    FileWhat it is
    E-MTAB-16835.idf.txt / .sdrf.txtStudy design; per-sample↔run metadata + ENA/FASTQ links
    HTO_library.txtHashtag reference
    AKPS_KPN_full_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, all cell types (4,951)
    AKPS_KPN_epi_counts.txt / _features.txt / _metadata.txtRaw counts + genes + labels, epithelial subset (1,573)
    AKPS_KPN_epi_counts_log1p.txtLog-normalised epithelial matrix
    AKPS_KPN_epi_umap.txtUMAP coordinates, epithelial subset (Figure 1k)
  • How this experiment became a dataset. Two engineered mouse tumour genotypes → hashed → 10x + NovaSeq → Cell Ranger multi (mapped to mm10-2020-A) → scanpy quality control → full + epithelial atlases with raw/normalised counts, genes, labels and UMAP.

  • Completeness (scRNA-seq). Tier 1: complete for both full and epithelial atlases. Tier 2 (log-normalised matrix, UMAP embedding): complete for the epithelial subset; the full-atlas object ships raw counts and metadata only. Tier 3: reads brokered to ENA (not mirrored); mouse reference build linked above.


1.4 Publication-level metadata (Nature Electronic Supplementary Material)

Shared across all datasets and stored in this dataset's input/supplement/ folder:

  • Supplementary Tables 1–11 (MOESM3): M0/M1 case–control cohort, UICC stage-II cohort, patient-derived-organoid driver mutations, and the gene-set definitions (fetal, intestinal-stem/LGR5, high-relapse, TROP2, YAP …).
  • Source Data for Figure 5 (MOESM4) and Extended Data Figure 12 (MOESM5).
  • Supplementary Information (MOESM1) and Reporting Summary (MOESM2).

2. Analyses this atlas supports

Concrete analyses this atlas directly enables, with the files to start from:

  • Quantify the LGR5 ⇄ TROP2 state balance. Score the intestinal-stem/LGR5 and fetal-like/TROP2 signatures (gene sets in Supplementary Table 11) on any *_epi_counts.txt + *_metadata.txt, then compare state fractions across conditions. Baselines: biobank_subcut_epi_* (E-MTAB-16585) and AKPS_KPN_epi_* (E-MTAB-16835).
  • Map therapy-induced plasticity in time. Use the Sacituzumab Govitecan time courses (HD4246_SG_0h-48h_*, HD4246_SG_48h-120h_* in E-MTAB-16849; HD4246_SG_org_* in E-MTAB-16843) and the FOLFIRI course (Org_chemo_LM2/LM3_* in E-MTAB-16836) to track how the TROP2-positive fraction rises or falls per timepoint — chemotherapy induces it, the antibody–drug conjugate depletes it.
  • Model state transitions with pseudotime. The *_palantir.h5ad objects (E-MTAB-16849, E-MTAB-16836) carry trajectories and terminal states; pair with HD4246_SG_mets_magic.csv (MAGIC-imputed) for smooth gene-trend curves of TROP2 (TACSTD2), LGR5, and fetal markers along pseudotime.
  • Dissect drug-target heterogeneity and escape. Examine TACSTD2 (TROP2) expression distribution in treated vs control cells (E-MTAB-16433 treatment axis; E-MTAB-16843 SG vs IgG1-SN-38) to locate TROP2-low escape populations — candidates for combination targeting.
  • Generate combination-therapy hypotheses. Contrast post-treatment survivor states (SG-resistant in E-MTAB-16433 / E-MTAB-16849) against chemotherapy-induced TROP2-high states (E-MTAB-16836) to nominate co-targets that close the resistance loop.
  • Study tumour-microenvironment crosstalk. The *_full_* matrices (E-MTAB-16583, E-MTAB-16585, E-MTAB-16835) retain immune, endothelial and fibroblast compartments for ligand–receptor / niche analyses around each tumour state.
  • Test cross-species conservation. Align the mouse VAKPS/VKPN epithelial states (E-MTAB-16835) to the human maps to check whether the TROP2 program is conserved — useful for justifying mouse models in follow-up work.
  • Connect states to clinical outcome. Bridge single-cell signatures to the bulk M0/M1 and UICC stage-II cohorts (Supplementary Tables) to ask whether a high TROP2 / high-relapse signature stratifies patient prognosis.
  • Reproduce or extend the pseudobulk differential expression. Re-run the authors' Decoupler 1.8.0 + PyDESeq2 0.4.10 workflow (Figure_01_pbulk.ipynb) on any raw-count dataset to derive condition-level differential-expression tables.
  • Re-align from raw reads. For allele-level or intron-aware work, pull FASTQ from the ENA via the SDRF FASTQ_URI fields and re-quantify with Cell Ranger against GRCh38-2020-A (human) or mm10-2020-A (mouse).

Reproduce the figures. git clone https://github.com/ORCA-HD/TROP2_in_CRC; run figures/get_data_from_arrayExpress.ipynb (reassembles AnnData objects from the deposited files), then Figure_01/03/04.ipynb and the Ext_Figure_*.ipynb notebooks. Conda environments are in env_specs/ (env_general, env_palantir, env_decoupler_v1.8); per-dataset helpers are in utils/.


Citation

Vaquero-Siguero, N., Georgakopoulos, N., Puschhof, M. C. et al. TROP2 targeting reveals therapy-driven cell state dynamics in colorectal cancer. Nature (2026). https://doi.org/10.1038/s41586-026-10705-2

About this dataset

This scRNA-seq (10x Chromium Next GEM Single Cell 3' v3.1, dual index; cell hashing / TotalSeq-B multiplexing) dataset provides single-cell gene expression from human and mus musculus colon, rectum, liver, subcutaneous xenograft and intestine samples, with disease state recorded as Colorectal cancer. 10x Genomics published it on July 1, 2026. The data were processed with Cell Ranger; scanpy; Seurat; Palantir; MAGIC; harmonypy; scrublet; Decoupler; PyDESeq2; CellTypist; Cell Hashing (HTODemux) Cell Ranger 7.1.0 (multi) / 6.1.0 (count); scanpy 1.9.3; Seurat; Palantir 1.4.0; MAGIC 3.0.0; harmonypy 0.0.9; Decoupler 1.8.0; PyDESeq2 0.4.10, and the 0 output files total 67.8 GB.

Analyze it in Drylab

Open this dataset in Drylab and ask a question in plain language. The AI agent loads the files, runs the analysis in a ready sandbox and explains each result. A typical first pass:

  1. Load the filtered feature-barcode matrix (HDF5) and run quality control on counts, genes and mitochondrial reads per cell.
  2. Normalize, select variable genes, then cluster the cells and draw a UMAP.
  3. Annotate clusters with marker genes and compare the cell-type mix between conditions.
  4. Run differential expression between cell types or samples to characterize colorectal cancer biology.