Genomic profiling and pre-clinical modelling of breast cancer leptomeningeal metastasis reveals acquisition of a lobular-like phenotype

Abstract

Breast cancer leptomeningeal metastasis (BCLM), where tumour cells grow along the lining of the brain and spinal cord, is a devastating development for patients. Investigating this metastatic site is hampered by difficulty in accessing tumour material. Here, we utilise cerebrospinal fluid (CSF) cell-free DNA (cfDNA) and CSF disseminated tumour cells (DTCs) to explore the clonal evolution of BCLM and heterogeneity between leptomeningeal and extracranial metastatic sites. Somatic alterations with potential therapeutic actionability were detected in 81% (17/21) of BCLM cases, with 19% detectable in CSF cfDNA only. BCLM was enriched in genomic aberrations in adherens junction and cytoskeletal genes, revealing a lobular-like breast cancer phenotype. CSF DTCs were cultured in 3D to establish BCLM patient-derived organoids, and used for the successful generation of BCLM in vivo models. These data reveal that BCLM possess a unique genomic aberration profile and highlight potential cellular dependencies in this hard-to-treat form of metastatic disease.

Fig. 1. Molecular profiling of BCLM by whole exome sequencing.

a Schematic showing the sites of material collected, created with BioRender.com. b Flowchart of sample processing pipeline. cfDNA, cell-free DNA; ctDNA, circulating tumour DNA, DTC, disseminated tumour cell; PDO, patient-derived organoid. c–e Samples underwent WES to identify mutation (exonic variants excluding synonymous SNVs, plus splice variants) and copy number aberration (CNA) events. c Chemotherapy-related COSMIC Single Base Substitution (SBS) mutational signatures compared between sites. Signature contributions of the six chemotherapy-related SBS signatures (SBS11, SBS17b, SBS28, SBS31, SBS35, SBS86) were compared between CSF cfDNA, plasma cfDNA, primary and metastasis tumour samples. Stacked bar plots display the mean signature contribution for each SBS signature across all samples in each category. Significantly different total chemotherapy-related SBS signature contribution between CSF and plasma cfDNA is shown (Mann-Whitney two-tailed test). All other comparisons were ns. d WES of paired CSF cfDNA samples collected at BCLM diagnosis and primary tumour samples (n = 17 pairs). Bars show proportion of total variants unique to CSF (median 47.8%; IQR 43.0–56.4%), unique to primary tumour (median 20.4%; IQR 16.6–31.5%) and shared (median 23.0%; IQR 13.8–33.7%). e WES of paired CSF cfDNA and plasma cfDNA samples collected at BCLM diagnosis (n = 11 pairs). Bars show proportion of total variants unique to CSF (median 24.6%; IQR 16.0–38.6%), unique to plasma (median 16.8%; IQR 11.3–26.6%) and shared (median 43.4%; IQR 24.2–65.1%).

Fig. 2. BCLM clonal evolution modelling.

ClonEvol was used to infer consensus clonal evolution models, using the variant clusters generated by PyClone. (a) KCL566, (b) RMH010, (c) KCL553. Remaining models are presented in Supplementary Figs. 7–13. Box plots show the variant clusters identified in each sample, and displays the variant allele frequency (VAF) of each mutation, with the median VAF depicted by black lines. Bell plots show the ClonEvol estimated cancer cell fraction (CCF) and evolution order of clones within each sample. Phylogenetic trees for each case depicts the evolution pattern. Each branch represents the evolution path of a clone from its parental clone, and each node marks the clone number of the preceding branch. Clone colour and number match those in box and bell plots. Samples are annotated at clone nodes when no further subclone evolution occurs. Branch lengths are scaled by the square root of number of mutations per clone, and branch width is inversely proportional to branch length. Mutations per clone are shown to top left of the trees. Gene names and variant type are annotated on branches as follows; black font if appear in the list of BCLM frequently altered cancer driver genes (shown in Fig. 3); green font if variant is OncoKB annotated as potentially therapeutically actionable (Supplementary Table 5). Variant type abbreviations; fs frameshift, nfs non-frameshift, ns nonsynonymous, sg stop-gain, spl splice-site variant.

Fig. 3. Cancer driver gene aberration in BCLM CSF cfDNA samples.

Uppermost bar plot show tumour mutational burden (TMB) by sample. Stacked bar plot below shows mutational subtype for each sample as fraction of all variants (excluding synonymous). Banners indicate primary tumour histological subtype and ER and HER2 status. Main plot shows the 35 most frequently altered cancer driver genes (alterations include non-silent somatic mutation or high-level copy number change) in CSF cfDNA as follows; Cancer Gene Census (CGC) gene altered in at least 4 cases; breast cancer driver gene (defined in Methods) altered in at least 3 cases. Right side bar plot shows percentage of CSF cfDNA samples (n = 21) with gene alterations that are unique to the CSF sample (green) or shared with any other site per individual (grey). Left side bar plot shows genomic alteration rate in the MBC cohort. See Supplementary Fig. 14 for extended data on matched plasma cfDNA, primary tumour and metastasis samples, and Supplementary Table 4 for frequently altered cancer driver genes in plasma cfDNA. Gene names are highlighted in red if genomic alteration rate was significantly different (p ≤ 0.05 by Chi-square test two-sided) in BCLM cohort compared to a publicly available dataset of 216 metastatic breast cancer (MBC cohort) samples (see Methods). Extended data in Supplementary Data 4 displays the significance values for this comparison.

Fig. 4. BCLM CNA landscape and gene set enrichment.

a Genome wide purity-adjusted CNA plots of all samples (CSF cfDNA (n = 21), plasma cfDNA (n = 11), primary tumour (n = 18) and metastasis (n = 8) showing frequency of copy number gain (red bars, positive y-axis) and copy number loss (blue bars, negative y-axis) at a gene level. b Volcano plots displaying comparison of CNA frequency per gene, between CSF cfDNA and primary tumours (upper panels) and between CSF cfDNA and plasma cfDNA (lower panels). Delta proportion (y-axis) is the proportional difference in CNA frequency between sample types. Significance (x-axis, -log₁₀ p-value) obtained by Chi-squared proportion test (two-sided) compares CNA event proportion per gene between sample types (source data are provided as a Source Data file). Gene names are annotated if delta proportion is ≥ ± 0.20 and -log₁₀ p-value ≥ 1.5 (as shown in Supplementary Fig. 15). c Gene set enrichment analysis. Genes enriched for alterations (including non-silent somatic mutation or high-level copy number change) in CSF cfDNA (n = 257) and/or plasma cfDNA (n = 160) vs. matched primary tumour, underwent statistical overrepresentation test using GO biological process (PANTHER). Displayed terms were retained if -log₁₀ p-value was ≥ 3.0 in either sample type (Fisher’s exact test, adjusted for multiple testing using the Benjamini-Hochberg procedure with an overall false discovery rate of < 0.05). d Focused alteration plot of genes from the indicated GO biological processes overrepresented in CSF cfDNA vs. matched primary tumour, but not in plasma cfDNA vs. primary tumour. Banners above indicate; histological subtype (primary tumour) and CDH1 mutation status of CSF cfDNA. Left panel shows gene names and chromosomal location, grouped by GO biological processes. Bars to right show the frequency of gene alteration across CSF cfDNA samples (n = 21) with ‘cfDNA only’ indicating alterations unique to CSF cfDNA or CSF and plasma cfDNA at the time of BCLM diagnosis, and ‘shared’ occurring both in CSF cfDNA and primary tumour or metastasis sample. See Supplementary Fig. 16 for extended data on matched plasma cfDNA, primary tumour and metastasis samples.

Fig. 5. Generation of BCLM patient-derived organoids (PDOs) established from CSF DTCs.

Histology and immunohistochemistry of BCLM PDOs and matched primary tumours. A minimum of 2 biological replicates were performed for each PDO immunohistochemical staining (range 2–4 replicates) with representative images shown. Scale bars, 50 µm.

Fig. 6. Characterisation of BCLM PDOs.

a Genomic comparison of CSF cfDNA and matched BCLM PDO (early passage) by WES. Top panel, correlation plots of copy number Z-score (one data point per gene) with regression line plotted (purple) and Pearson’s r-value plus two-sided significance values are shown. Lower panel, Venn diagrams of somatic mutations (excluding synonymous) present in CSF cfDNA and matched BCLM PDO, values shown are mutation count. b Comparison of Pearson r-values from Fig. 6a and Supplementary Fig. 17 by Mann-Whitney, two-tailed test showing a higher level of correlation in CNA status between PDO (n = 5) and CSF cfDNA (n = 5) than between PDO and primary tumour DNA (n = 5) (left panel) and higher fraction of shared variants between CSF cfDNA and PDO than primary tumour and PDO (right panel), source data are provided as a Source Data file. c RNAseq of PDOs (fresh early passage) and matched primary tumours (FFPE). Expression levels of adherens junction components are shown as Z-scores of log2-TPM values (transcripts per million). d RTqPCR analysis of CDH1 expression in PDOs as compared with cell line controls. n = 1 biological replicate, n = 3 technical replicates, source data are provided as a Source Data file. e BCLM PDOs and cell line control (DU4475) were cultured in the presence of methotrexate in 0.2% DMSO or 0.2% DMSO alone (vehicle alone). Cell viability was measured by CellTiter-Glo after 14 days. Data represents mean of n = 4 wells per datapoint ± SD, normalised to vehicle alone treatment, source data are provided as a Source Data file.

Fig. 7. Variable engraftment rate of BCLM PDOs in the orthotopic site.

Female NSG mice were inoculated with mChLuc2-labelled PDOs as follows. a Non-dissociated BCLM PDOs (~ 1 × 10⁶ cells) were injected transcutaneously into the 4th mammary fat pad (unilateral), n = 4–6 mice per PDO. Upper panel, tumour size was measured bi-weekly. Middle panel, mammary fat pad tumours stained for H&E, lamin A/C, ER and E-cadherin. Scale bar, 100 µm. Lower panel, bar charts show percentage of mice with metastatic dissemination in individual organs. Pale bars indicate % tissues with single or small cell clusters (1–10 cells), dark bars indicate tissues with metastatic deposits ( > 10 cells), source data are provided as a Source Data file. b 4 × 10⁴ dissociated KCL320 PDO cells per gland were inoculated via the mammary gland intraductal (MIND) route into the 3rd, 4th and 5th mammary glands, bilaterally (6 glands/mouse; n = 6 mice). Mice were IVIS imaged on day 0 and day 165 (endpoint). c 2 × 10⁶ dissociated KCL320 PDO cells injected intraperitoneally (n = 4 mice) with IVIS imaging at day 0 and day 81 (endpoint). Created with BioRender.com.

Fig. 8. Modelling BCLM by intracardiac or intracerebroventricular injection of PDOs.

BCLM PDOs were dissociated into single cells. a 0.5–1.0 × 10⁶ cells inoculated into the left cardiac ventricle under ultrasound guidance. Upper panel, IVIS imaging immediately after cell injection (day 0) and at endpoint for each mouse (development of significant morbidity), day 40–210. Middle panel, metastatic site involvement assessed by lamin A/C staining of FFPE sections. Bar charts show percentage of mice with disseminated tumour cells in individual organs, with pale bars indicating with single or small clusters of cells (1–10 cells) and dark bars indicating tissues with metastatic deposits ( > 10 cells), source data are provided as a Source Data file. Bottom panel. IHC staining for human lamin A/C of whole cranium and spine FFPE sections showing tumour cells growing along the meninges (orange arrowheads) or in the brain parenchyma (blue arrowheads), in the dura (red arrowheads) and pituitary (green arrowhead). Scale bars; Left, 100 µm; other panels, 1 mm. b 1 × 10⁵ cells inoculated intracerebroventricularly (ICV) (n = 4 mice). Left panel, IVIS imaging on day 0 and day 81 (endpoint) showing development of metastasis in two mice. Right panels, cranium and spine FFPE sections stained for human lamin A/C. Orange arrowheads indicate leptomeningeal involvement. Scale bars; 250 µm. Created with BioRender.com.