Transcriptomic intratumor heterogeneity of breast cancer patient-derived organoids may reflect the unique biological features of the tumor of origin

Surgical specimens were obtained by the core needle biopsy of surgically removed tumors. The specimens were dissociated into single cells using a MACS Tumor Dissociation Kit and a gentle MACS dissociator (Miltenyi Biotec, North Rhine-Westphalia, Germany) in accordance with the manufacturer’s instruction. For pleural effusion specimens, drainage fluid was collected after trocar placement, centrifuged, and purified with Red Blood Cell Lysis Solution lysis buffer (Miltenyl Biotec). The organoids were established from the obtained cells as described previously [8]. Briefly, cell pellets were resuspended in BME gel (R&D Systems, Minnesota, USA) at 4 °C and seeded to form domes in 24-well or 48-well plates (IWAKI, Tokyo, Japan), polymerized at 37 °C for 10–20 min, and the prepared medium was added at a volume of 500 mL per 50 µL BME dome. The medium was changed every 3–4 days and the PDO was passaged depending on growth. Cultrex Organoid Harvesting Solution (Thermo Fisher Scientific, Massachusetts, USA) and 0.25 w/v% Trypsin-1 mM EDTA 4Na Solution with Phenol Red (FUJIFILM, Tokyo, Japan) were used for passaging of organoids. Expression of ER and PgR in surgical specimens was assessed using the Allred score [12].

Organoids were dissociated into a single-cell suspension using harvesting solution and trypsin, and the cells were cryopreserved for the scRNA-seq experiment. Moreover, BC specimens were dissociated into single cells, as described above. The scRNA-seq libraries were prepared using the BD Rhapsody Single-Cell Analysis system (BD, New Jersey, USA) in accordance with the manufacturer's instruction. Briefly, after thawing, cells were labeled with BD Single-Cell Multiplexing Kit. The labeled cells were washed, pooled, and loaded onto the BD Rhapsody microwell cartridge; then cDNA was synthesized using the BD Rhapsody Whole Transcriptome Analysis Amplification Kit. The resulting gene expression and cellular label libraries were sequenced on Illumina NextSeq 550 platform (Illumina, California, USA) with paired-end reads (read1, 75 bp; index1, 8 bp; read2, 75 bp). Sequencing data were processed using the BD Rhapsody Analysis pipeline on the Seven Bridges Genomics platform and converted to a gene expression count matrix.

We utilized the web-based analysis pipeline for BD Rhapsody and BD Precise ASSAYS (https://​www.​sevenbridges.​com/​bdgenomics/​) to generate the count matrixes and Seurat [13] for downstream analysis. RSEC counts were used as the input count matrixes. Low-quality cells with over 40% of mitochondrial RNAs and < 400 or > 9000 features were filtered out. To avoid misassignment between cells and samples, we performed pre-clustering of each PDO using Seurat’s standard flow and manually removed clusters that exhibited distant embedding from major clusters and different distribution of each sample tag that were possible contaminants from other samples.

For high-quality scRNA-seq data of each PDO, transcripts count matrixes were normalized to the total number of counts for the cell and multiplied by a scaling factor of 10,000. The normalized values were subsequently natural-log transformed using Seurat’s “NormalizeData()” function and a linear transformation was applied by “ScaleData()”. The principal component analysis was performed by “RunPCA()” with top 2,000 highest variable features identified by “FindVariableFeatures()” with vst selection method. We excluded PCs that contained ribosomal protein-encoding genes in more than half of the top 20 features. Then we performed Seurat’s standard clustering procedures using “FindNeighbors()” and “FindClusters()” with the top 20 PCs and a resolution of 0.4. To visualize data, “RunUMAP()” was used with the same PCs to identify the clusters.

After clustering of each PDO, we identified differentially expressed features as cluster-specific gene expression signatures (ClustGS) by using the “FindAllMarkers()” function with “only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0”. We removed mitochondrial RNAs or ribosomal protein-encoding genes and selected the genes with an adjusted P-value of less than 0.05. The top 25 most significant genes in each cluster of each PDO were selected as ClustGS. Any clusters with fewer than 25 significant genes (adjusted P-value < 0.05) were not considered for further analysis. Meta-clusters were manually defined by the Jaccard similarity index. All genes included in at least two ClustGS were defined as meta-cluster specific genes (meta-ClustGS).

Gene enrichment analysis for each ClustGS was performed using the “enrichr()” function of ClusterProfiler v3.18.1 [14] with MsigDB Hallmark genes obtained through msigdbr’s msigdbr(species = “Homo sapiens,” category = “H”) function. The false discovery rate (FDR) was used as the adjusted P-value.

Gene overlap analysis between ClustGS and cluster-specific expressed genes of corresponding primary tumors or different passage organoids was performed using “GeneOverlap” package’s newGeneOverlap() and testGeneOverlap() functions. To obtain cluster-specific expressed genes in each dataset, we selected the genes with < 0.05 adjusted P-values calculated by Seurat’s FindAllMarkers() with options of “only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0”.

The extracellular cytokines in the culture medium were comprehensively semi-quantified using the Proteome Profiler Array Human XL Cytokine Array Kit (R&D Systems) in accordance with the manufacturer’s instruction. Briefly, culture medium from four organoids (PDOs 155, 165, 166, 180) was collected 72 h after the change of fresh medium and frozen at − 20 °C. Between 59.5 and 846.2 µL of the culture medium, normalized by RNA quantity from the PDOs in the same culture, was loaded into the cytokine array. The Odyssey XF Imaging System (LI-COR, Nebraska, USA) was used to detect fluorochromes; the absorbance was set to 685 nm and the exposure time was set to 10 min. The signal intensity corresponding to each factor in the array was quantified by ImageJ software, and the expression level was calculated relative to the positive control.

Tumor sections were stained with hematoxylin–eosin (HE) and immunohistochemically examined for CHI3L1 and CST3 using a Nichirei Histofine system (Nichirei Biosciences Inc.). Tissue sections were incubated with primary anti-CHI3L1 antibody (Cell Signaling, #47,066) and anti-CST3 antibody (Abcam, ab109508) at a 1:800 and 1:4000 dilution, respectively, and detected with HRP-labeled polymer-conjugated secondary antibody (Histofine Simple Stain MAX PO, multi, Nichirei). Color development was achieved with 3,3′-diaminobenzidine tetrahydrochloride. Organoids were fixed and paraffin-embedded using Epredia HistoGel Specimen Processing Gel (Thermo Scientific) and Tissue-Tec VIP (SAKURA, Japan). Immunohistochemical staining on the organoid sections was performed by incubating with anti-CHI3L1 antibody (Cell Signaling, #47,066) and anti-CST3 antibody (Abcam, ab109508) at a 1:400 and 1:500 dilution, respectively. Signals were detected using Dako EnVision + Dual Link System-HRP (Agilent).

Ten PDO lines established during a certain period of time were subjected to the analysis in this study (Fig. 1A). The clinicopathological features of the 10 cases are presented in Table 1. Briefly, eight were from patients with primary breast cancer, one case (PDO210) was local recurrence in the chest wall after total mastectomy, and one case (PDO207P) was lung metastasis with pleural effusion. In addition, one of the eight cases of primary breast cancer (PDO155) developed metastasis soon after surgery. Tumors with high proliferative potential may have been more frequently established as organoids; there were five cases with nuclear grade 3 and Ki67 > 20%.

The speed of organoid growth differed for each PDO. PDO155, 166, and 207P had the fastest growth rate, with a doubling time of approximately 7 days, which continued even after ten passages. PDO 165 and 180 grew more slowly than the three PDO lines listed above, with a doubling time of approximately 14 days. The other lines grew more slowly.

Organoid morphology was also heterogeneous among PDOs (Fig. 1B, C). Spheroids were common in all PDOs, but some PDOs had mixed populations of organoids with different shapes. For example, PDO155 consisted of two different cell populations: spheroids and densely expanding structures. PDO154, 165, and 180 appeared to form sheet-like structures around the spheroids that dragged each spheroid along, altering their 3D position every few days. PDO207P contained grapevine cluster-like formations. PDO166, 195, 202, 203, and 210 formed homogenous spheroids.

Collectively, our results present the inter- and intratumor heterogeneity for organoid growth and morphology, suggesting that the PDOs reflect the biological properties of the original tumors.

We performed single-cell transcriptome analysis of 10 PDOs by multiplexing scRNA-seq using the BD Rhapsody Single-Cell Analysis System. After assignment and quality filtering, we obtained the high-quality expression profiles of 9548 cells: an average of 6635 counts and 2203 genes per cell (Additional file 1: Figure S1; “Method” Section). For simple quality assessment, we used the Seurat package to cluster the cells and examine the expression of marker genes (Additional file 1: Figure S1). Almost all cells expressed epithelial markers such as EPCAM and keratins, and we considered that the PDO lines used in this study contained nearly no stromal cells. Therefore, we hereby note that the subsequent analysis assesses the ITH of cancer cells.

The cancer cells were clustered as per patient samples in scRNA-seq analysis (Additional file 1: Figure S1), which is consistent with previous reports [15‐18]. To investigate the ITH in each PDO, we developed the analytical approach as follows: (1) cluster the cells in each PDO, (2) identify cluster-specific expressed genes, and (3) calculate their similarity and classify clusters (Fig. 2A). Through a standard analytical flow in Seurat, the cells were classified into 3–6 clusters per each PDO (3 clusters in PDO180, 195, and 202; 6 clusters in PDO210) (Fig. 2B, C). Thus, we identified a set of genes expressed specifically in each cluster within each PDO and defined the most significant 25 differentially expressed genes as the cluster-specific gene signature (ClustGS). Any clusters with fewer than 25 significant genes (adjusted P-value < 0.05) were not considered for further analysis. Each PDO had at least one cluster with ClustGS, and in total, 38 clusters with a ClustGS were identified (Fig. 2C).

For example, PDO165 had four clusters with a ClustGS (Fig. 2D). ClustGSs in PDO165#C1 contained luminal-lineage transcription factors (TFs) (FOXA1, GATA3); PDO165#C2 contained TGFB2 and DKK1; PDO165#C3 contained cell cycle-related genes (CDK1, MKI67); and PDO165#C4 contained basal keratins (KRT5, KRT14) (Fig. 2D; Additional file 2: Table S1). Consistently, the molecular signatures database (MSigDB) hallmark gene enrichment analysis suggested that PDO165#C1 was related to estrogen response; PDO165#C2 and PDO165#C4 were related to the epithelial–mesenchymal transition (EMT); and PDO165#C3 was related to the cell cycle (Additional file 1: Figure S2). These results suggest that each PDO is composed of multiple subpopulations with distinct cell states and lineages with the cell cycle, estrogen response, and EMT-like gene expression programs.

We next assessed whether the ITH observed in the PDOs reflected the ITH in the original tumor tissues and whether it was maintained after repeated PDO passages. We performed scRNA-seq experiments on two additional original breast cancer tissues (165 and 210) and two different passages of PDO lines (180 and 210). Subsequently, we performed the same analysis as above, and confirmed the existence of ITH in each sample (Additional file 1: Figure S3; Additional file 2: Table S2). Comparison of the characteristics of the identified cell populations by looking at the overlap of cluster-specific expressed genes revealed that they were composed of somewhat similar cell populations between original tissues and PDOs as well as between different passages of the same PDO lines (Additional file 1: Figure S4; Additional file 2: Table S3), although not completely one-to-one correspondence. Thus, we decided to analyze these PDOs further, considering that they retain some extent of ITH of the original tissues even after successive passages.

To compare the degree of ITH between PDOs, we calculated the Jaccard similarity index across 38 ClustGS from each cluster. Based on the similarity of ClustGS, seven meta-clusters were identified with the hierarchical clustering guide, representing a set of cell clusters characterized by the same expression programs across PDOs (Fig. 3A). To annotate each of the meta-clusters, we selected the overlapping genes across ClustGS in at least two PDOs as a meta-cluster (“meta-ClustGS”; Additional file 2: Table S4). Gene enrichment analysis for meta-ClustGSs using the MsigDB Hallmark gene set suggested that meta-ClustGSs 1–2 were associated with cell cycle, notably the G2/M phase for meta-ClustGS 1 and G1/S phase for meta-ClustGS 2; meta-ClustGSs 3, 4, and 6 were associated with the EMT; and meta-ClustGSs 5 and 7 were associated with estrogen response (Fig. 3B; Additional file 1: Figure S5). Although meta-ClustGSs 3, 4, and 6 all contained EMT-associated genes, each gene signature was different: basal cytokeratin (such as KRT5, KRT14) in meta-ClustGS 3, mesenchymal marker ACTA2 in meta-ClustGS 4, and the metastasis-associated gene ANXA1 in meta-ClustGS 6. There were also two meta-ClustGSs associated with estrogen response: luminal-lineage specific TFs FOXA1 and GATA3 in meta-ClustGS 5, CCND1 in meta-ClustGS 7. These results suggested that each PDO contained several cell populations that were regulated by the defined gene expression programs that were common across PDOs.

We subsequently characterized each PDO using meta-ClustGSs. The meta-ClustGS 1 (Cell cycle—G2/M) was present in 7 of 10 PDO lines (Figs. 3A, 4A, B; Additional file 1: Figure S6), indicating that the proliferating cell populations were common in PDOs. Interestingly, both meta-ClustGS 3 (EMT1) containing basal markers (KRT5 and KRT14) and meta-ClustGS 5 (estrogen response 1) containing luminal-lineage TFs (FOXA1 and GATA3) were observed in four PDO lines (PDO155, 165, 166, and 180) derived from patients with ER+ /HER2– or ER–/HER2+ tumors, suggesting that these PDOs had distinct ITH of co-existing of basal-like and luminal cells (Fig. 4C–F; Additional file 1: Figure S6). Clusters of cells with meta-ClustGS 5 expression were observed in PDO155 derived from ER–/HER2+ tumors, suggesting that a small number of cells expressed ER target genes that were not detected by the bulk assessment, i.e., immunohistochemistry, but were observable at a single-cell resolution (Fig. 4E; Additional file 1: Figure S6). Most cells in three PDOs (PDO180, 195, and 202) derived from ER+ tumors expressed the meta-ClustGS 3 (EMT1, basal-like), implying that the property of ER+ cancer cells was altered in organoid culture or that part of organoids was derived from juxta-tumoral basal epithelial cells in these PDOs (Fig. 4C, D).

Another estrogen response-related meta-cluster, meta-ClustGS 7, was observed in PDO154, 207P, and 210. Interestingly, they had distinct clinical features such as ER + /HER2 + (PDO154), pleural effusion sample (PDO207P), and early recurrence (PDO210). These suggested that meta-ClustGS 7 may reflect a different cellular state to normal ER-positive breast cancer cells with a good prognosis, even though it is annotated with the same estrogen response by MsigDB analysis. For example, meta-ClustGS 7 contained CCND1, which has been implicated in endocrine resistance and poor outcome [19, 20], suggesting that this gene signature might reflect the peculiarity of a cell population (Additional file 1: Figure S6).

These results indicate that meta-ClustGSs can describe each PDO with a qualitative and quantitative assessment of ITH.

In the process of identifying meta-clusters, we also identified nine PDO-specific clusters (Fig. 5A; Additional file 2: Table S5). These ClustGSs corresponded to four PDO lines, each derived from cases with characteristic clinical courses or histopathology: PDO155 from inflammatory breast cancer (IBC), PDO203 from invasive micropapillary carcinoma, PDO207P from metastatic breast cancer, and PDO210 from recurrent breast cancer. Thus, ClustGSs not shared with any other PDOs possibly reflected the clinical or molecular features for each PDO.

For example, PDO207P#C4 ClustGS included the mesenchymal marker VIM and CYP4Z1, which are associated with poor prognosis and high-grade tumors in breast cancer [21]. This ClustGS was highly expressed in a certain population of PDO207P (Fig. 5B, C). As PDO207P was derived from the metastatic pleural effusion, this cluster probably reflected the distinct features of mesenchymal and aggressive disease in PDO207P.

We further investigated PDO155 which was derived from a patient with IBC. The patient developed liver metastases early after surgery and had a poor clinical outcome [22, 23]. To our knowledge, no reports are available on PDO lines being established from patients with IBC and then precisely analyzed. We therefore considered that it might be possible to find the biological features of IBC in the PDO155. We focused on a specific cell population, PDO155P#C1 (Fig. 2A), with a ClustGS that was not similar to other ClustGSs. In the gene signature, genes associated with RNA transport and translation initiation functions were highly enriched (EIF3H, PABPC1, EIF3E, and EIF2S3; Additional file 2: Table S5). In this cell population, translation was expected to occur preferentially to cell cycle progression (Fig. 4C), suggesting that cell size may increase [24, 25]. This may be related to the morphological heterogeneity observed specifically in PDO155, in which some of the organoids formed gigantic spheroids (Fig. 1B, C).

The gene signature also included several genes encoding cytokines, such as CST3 and CHI3L1. We thought that the specific cytokines released from cancer cells might be related to the pathophysiological characteristics of IBC. Thus, we examined cytokines in the culture medium in PDO155 and other PDOs using a cytokine array (Fig. 6A). CST3 and CHI3L1 were highly detected in PDO155; in particular, the secretion of CST3 was specifically observed in PDO155. CST3 is known to have various functions on cancer progression [26, 27], which may be related to the feature that part of PDO155 forms huge clumps. Interestingly, these genes were not uniformly expressed in all cells; they were heterogeneously and non-randomly expressed (Fig. 6B). By performing immunohistochemistry of PDO155 and pathology specimens of this patient, we confirmed that CST3 and CHI3L1 proteins were heterogeneously expressed in PDO155 and even in the original tumor tissue (Additional file 1: Figure S7). This indicates that only a subset of the cell population may only exhibit the specific characteristics observed in the bulk assay; conversely, a unique feature visible in the ITH may display the representative feature of that PDO.