Article Text
PDF
Abstract
Objective Sporadic early-onset colorectal cancer (EOCRC) has bad prognosis, yet is poorly represented by cell line models. We examine the key mutational and transcriptomic alterations in an organoid biobank enriched in EOCRCs.
Design We established paired cancer (n=32) and normal organoids (n=18) from 20 patients enriched in microsatellite-stable EOCRC. Exome and transcriptome analysis was performed.
Results We observed a striking diversity of molecular phenotypes, including PTPRK-RSPO3 fusions. Transcriptionally, RSPO fusion organoids resembled normal colon organoids and were distinct from APC mutant organoids, with high BMP2 and low PTK7 expression. Single cell transcriptome analysis confirmed the similarity between RSPO fusion organoids and normal organoids, with a propensity for maturation on Wnt withdrawal, whereas the APC mutant organoids were locked in progenitor stages. CRISPR/Cas9 engineered mutation of APC in normal human colon organoids led to upregulation of PTK7 protein and suppression of BMP2, but less so with an engineered RNF43 mutation. The frequent co-occurrence of RSPO fusions with SMAD4 or BMPR1A mutation was confirmed in TCGA database searches. RNF43 mutation was found in organoid from a leukaemia survivor with a novel mutational signature; and organoids with POLE proofreading mutation displayed ultramutation. The cancer organoid genomes were stable over long culture periods, while normal human colon organoids tended to be subject to clonal dominance over time.
Conclusions These organoid models enriched in EOCRCs with linked genomic data fill a gap in existing CRC models and reveal distinct genetic profiles and novel pathway cooperativity.
- colorectal cancer
- colorectal cancer genes
- colon carcinogenesis
- gene expression
- Early-onset colon cancer
- organoid models
- R-spondin fusion
- RNF43
- serrated neoplasia
Statistics from Altmetric.com
- colorectal cancer
- colorectal cancer genes
- colon carcinogenesis
- gene expression
- Early-onset colon cancer
- organoid models
- R-spondin fusion
- RNF43
- serrated neoplasia
Significance of this study
What is already known on this subject?
The incidence of early-onset colorectal cancer (EOCRC) is on the rise, but it remains poorly characterised with few studies, as the majority of published literature is focused on late-onset sporadic CRCs.
Organoids derived from EOCRCs remain limited, given its rarity and highly heterogeneous genetic background.
What are the new findings?
We established a living biobank enriched in early-onset CRC with diverse molecular repertoires not previously seen, including a novel mutational signature in a leukaemia cancer survivor.
We established and characterised the first human organoid models of PTPRK-RSPO3 fusion, demonstrated to be Wnt dependent.
Transcriptome analysis revealed similarity between RSPO fusion organoids and normal, but distinct differences from APC mutant organoids, with unique pathway dependency on BMP/SMAD signalling inactivation in the former, and high PTK7 stem cell marker expression in the latter.
As the largest long-term culture study examining genomic stability in normal and cancerous organoid models, we reveal subclonal heterogeneity in normal organoid cultures, with competition always ending in clonal dominance. The information has implications on the design and use of pooled normal organoids for CRISPR-Cas9 genetic engineering studies.
How might it impact on clinical practice in the foreseeable future?
Oral Wnt secretion inhibitors, anti-RSPO antibodies and a PTK7-targeting antibody-drug-conjugate have been developed, with some in early clinical trials, which may potentially have differential applications in RSPO fusion versus APC mutant CRCs.
Introduction
Colorectal cancer (CRC) is a heterogeneous disease, with a specific combination of tumour drivers in each molecular subtype, making standard cancer treatment ineffective. It remains the second most common cause of cancer death worldwide. Although there was a drop in the total global incidence of CRC, the incidence of early-onset colorectal cancer (EOCRC) (for patients aged 50 or younger) has been increasing in many countries over the past two decades.1 2 While a proportion of the EOCRCs are familial/hereditary, the majority are sporadic in origin with no known hereditary predisposing factors and/or have poorly understood pathogenic mechanisms. In general, sporadic EOCRC represents an aggressive subtype that has distinct characteristics from late-onset CRC, including presentation at more advanced stage, adverse features such as poorly differentiated, mucinous or signet ring morphology, venous invasion, and associated synchronous or metachronous tumours.2–4 Nevertheless, in terms of molecular characterisation, reports within the field remain scanty. Therefore, a comprehensive genome-wide study on sporadic EOCRC to delineate the underlying molecular alterations can shed light on the potential existence of unique tumour drivers, biomarkers and druggable targets for this subgroup of patients.
Since the ability to culture human intestinal organoids has been realised,5 several teams have established a library of patient-derived CRC organoid lines, encompassing different histological subtypes and clinical stages, as preclinical models for downstream genomic profiling.6–10 Given the fact that sporadic CRCs arising through the classical adenoma-carcinoma sequence, with mutational activation of the Wnt signalling pathway, constitutes ~80% of total CRC incidence, withdrawal of Wnt-3A and R-spondin1 in the culture medium can prevent normal organoid overgrowth, thereby increasing the success rate of tumour organoid establishment.6 As a result, the majority of the available CRC organoids described in the literature are derived from conventional adenocarcinoma with APC mutation. Organoids derived from EOCRC remain limited, given its rarity and highly heterogeneous genetic background. Our CRC biobank complements other published cohorts by representing an aggressive subtype of EOCRC, with distinct and diverse genetic profiles.6–10 We also evaluated the genomic stability of both normal and cancer organoids in terms of chromosomal patterns, somatic alterations as well as clonal architecture over long-term culture. This provides essential information for the community on the use of organoid culture as a cell model for biological studies.
Materials and methods
See online supplementary methods for details on the study cohort, organoid culture, Microsatellite instability (MSI) analysis, whole-exome sequencing (WES), whole-genome sequencing (WGS), RNA sequencing, single cell RNA sequencing, RSPO fusion detection, RNF43 splicing verification, bioinformatics analysis of The Cancer Genome Atlas (TCGA) cohort, cell viability assays, immunofluorescence staining and flow cytometry for PTK7, CRISPR/Cas9 genome engineering of APC or RN43 mutations in normal colon organoids and western blot analysis.
Supplemental material
Results
Establishment of an early-onset enriched living CRC biobank
There are 22 patients in our cohort, including 2 patients who underwent colonoscopy to resect 2 preneoplastic lesions and 20 patients with CRC. Seventeen out of 20 patients with CRC were below the age of 60 and 11 were under 50 (online supplementary figure 1 and online supplementary table 1). A total of 52 organoids were derived from this study cohort, including 18 normal colon organoids from different regions of the intestinal tract (NO/N1O, NAsCO, NileO, NCaeO, NTrCO), 1 hyperplastic polyp, 1 sessile serrated lesion (SSAO), 20 tumour organoids, with another 9 tumour organoids established from a second region of the primary tumour (TO/T1O/T2O), and 3 metastatic lymph node organoids (figure 1). The success rate of establishing colon cancer organoids (CCOs) and normal organoids was ~67% (20/30) and 97% (29/30), respectively. The tumour organoids were established as previously described,5 11 with minor modifications, including longer incubation periods with digestion enzymes for tumour specimens that were surrounded by tough fibrotic tissue. Moreover, normal organoid contamination was minimised by dissecting tumours from the serosal side, applying selection pressure on CCOs by withdrawing Wnt-3A and R-spondin1 to enrich for APC mutant cells or by adding nutlin3a to enrich for TP53 mutant cells, as well as manually selecting the tumour organoids under a microscope (online supplementary table 2). MSI analysis was performed on at least one CCO from each of the patients (online supplementary table 1), as described in the online supplementary methods, and all tested CCOs were confirmed to be microsatellite-stable (MSS) (data not shown). We excluded MSI early-onset patients in the current study, as they are mostly related to Lynch syndrome or cases with a hereditary or familial background.
Supplemental material
Supplemental material
Supplemental material
The genomic landscape of the CCOs collected from 22 patients. (A) Oncoplot of the germline and somatic cancer driver alterations in CCOs, including most of the known CRC driver events. Histogram at the top denotes the mutation spectrum. *Patient’s tumours showing associated features of serrated neoplasm. (B) Identification and validation of the PTPRK-RSPO3 fusion transcript in H011 and H019 frozen tumour and tumour organoids. RT-PCR results for H011 and H019 (top), and Sanger sequencing results for H011-TO (middle). An RNF43 splicing event identified in H015 TF and TO was confirmed by the presence of dominant isoform v2 with Exon8 skipping and v1 with a 3 bp deletion at Exon 8 by RT-PCR and confirmed by Sanger sequencing (bottom). (C) Distribution of mutation in four Wnt regulators across age of onset in the TCGA CRC cohort. (D) Bar chart showing the association between age of onset and driver mutations in terms of mutant frequency. Statistics were performed using the χ² test or the Fisher’s exact test, as indicated by b. AF, mutant allelic fraction; CCO, colon cancer organoid; CNV, copy number variation; CRC, colorectal cancer; fusion, PTPRK-RSPO3 fusion; HPO, hyperplastic polyp organoid; LNO, tumour organoid from lymph node metastasis; LOH, loss of heterozygosity; NF, normal frozen tissue; NO, normal organoid; SSAO, sessile serrated lesion organoid; T2O, tumour organoid from a second region; TF, tumour frozen tissue; TO/T1O,tumour organoid.
Genomic characterisation of EOCRC tumour organoids revealed several distinct features
We performed WES on all tumour and normal organoids derived in this cohort, as well as the 21 organoids that were cultured long-term (11 normal organoid late passage (NOL) and 10 TOL), to examine their genomic stability (online supplementary figure 1 and online supplementary table 3). We also sequenced 19 normal frozen colon tissues and 3 blood leucocyte DNA (from patient HX102, HX103 and H012) as reference germline genomes for alignment of sequencing reads for the identification of somatic mutations. Twenty-two frozen tumour tissues (TFs) with high tumour content were selected for WES and the sequencing results were compared with paired organoids. We doubled the coverage of DNA sequencing for all TFs to mitigate the effect of lower tumour purity. Overall, we achieved a mean coverage of >50× for the organoids and normal frozen tissues, and >90× for the TFs (online supplementary table 3). We uncovered several unique and novel molecular features from this EOCRC enriched cohort (figures 1 and 2 and online supplementary tables 4 and 5). First, while truncating APC mutations were found in 16 out of 20 patients (80%), we observed an enrichment of APC wild-type tumours among the patients with EOCRC, with all four patients being under the age of 50 (figure 1A). Three out of four APC wild-type EOCRCs demonstrated an association with the serrated neoplasia pathway, with the concurrent presence of sessile serrated lesions (SSLs) either at the tumour edge, or in the adjacent colon tissue. Their tumours carried either BRAF V600E or KRAS mutations, and we subsequently detected a previously described PTPRK(e7)-RSPO3(e2) fusion12 in H011 and H019 CCOs by RNA sequencing (RNAseq) (online supplementary table 6), which we validated by RT-PCR in CCOs and their paired TFs (figure 1B). To our knowledge, these organoids represent the first in vitro 3D tumour model of RSPO fusions. For the third patient, H012, both the TF and the CCOs showed no mutation or changes in copy number variation of known key-components of the Wnt signalling pathway, such as AXIN2 and CTNNB1, even though the CCO was subsequently found to be Wnt independent (figure 3B). The fourth patient with APC wild-type EOCRC, H015, carried a RNF43 splicing mutation with exon eight skipping as the dominant transcript, which was confirmed by RT-PCR and Sanger sequencing (figure 1B). To validate our observation, we examined the WES data from the TCGA cohort and confirmed a reduced incidence of truncating APC mutation in the EOCRCs (early 68.1% (47/69) vs late 81.4% (319/392) onset, p=0.012)(figure 1C,D), whereas alterations in other Wnt pathway genes, such as CTNNB1, RNF43 or RSPO fusions, contributed to varying degrees in these early-onset cases.
Supplemental material
Mutational signatures identified in colon cancer organoids. (A) A novel mutational signature identified in H015. Each colour represents one of the six potential base substitutions, with each substitution further subdivided into 16 categories according to the adjacent 5’ and 3’ nucleotides. Y-axis denotes the mutation frequency. (B) Comparison of trinucleotide mutation patterns between H015-TO and its subset of key mutated drivers of the Cosmic Cancer Gene Census genes. (C) Comparison of a mutational signature between H015-TO and a patient with lung cancer found in the TCGA database. (D) Comparison of the mutational signature in H020-TO (left) and H020-T2O (right) with the mutational signatures in COSMIC. (E) Heatmap of the inferred LRR data showing the somatic copy number alteration landscape of the colon cancer organoids separated by age of onset. Copy number changes are shown in shades of red for copy number gains and shades of blue for copy number loss. The intensity scale is indicated on the right side. The histogram on top summarises the frequency. LRR, log R ratio; T2O, tumour organoid from a second region; TO, tumour organoid.
RSPO fusion colon cancer organoids have a transcriptome that resembles normal organoids and are hypersensitive to Wnt withdrawal. (A) Hierarchical clustering shows the segregation of normal organoids from the CCOs. The five CCOs derived from two patients with RSPO fusion clustered with the normal organoids. The heatmap represents differential gene expression in terms of TPM log2 values after mean centring. The scattered plot shows the Wnt activity of individual organoids using AXIN2 expression as the readout (top right). (B) Representative H&E and brightfield images of selected CCOs cultured in complete culture medium shows the differential growth response of the CCOs on withdrawal of either Wnt-3A or R-spondin1. Statistics were calculated using student’s t-test. Brightfield images were taken at 5 weeks of culture or at 14 days for Wnt dependent organoids as indicated under the specified culture conditions. Cell viability was quantified at 14 days after growth factor withdrawal using a cell-titre Glo assay. The assay was repeated twice with similar results. Scale bar: 40 µm (H&E), 300 µm (brightfield). CCO,colon cancer organoid.
One case, H015, was a cancer survivor who received Cytarabine, Daunorubicin and Fludarabine chemotherapy treatment for acute myeloid leukaemia and cyclophosphamide and busulfan treatment for a bone marrow transplant at a young age, providing us with a unique opportunity to examine the mutational insult of cancer treatment on colon stem cells. Interestingly, we identified a substantial increase in number and density of single nucleotide variant (SNV) in both the normal (4.86 SNVs/Mb) and cancer organoids (124 SNVs/Mb) from H015, with the latter being 30 times above that of other CRCs (excluding H020-TOs) and approaching the number observed in the POLE proofreading hotspot mutant case H020 (online supplementary figure 2B and online supplementary table 4). We also noted a dominance of C to A transitions, similarly observed in the tumour tissue (figure 1A), while C to T transitions were the most common mutation in the other CRCs. We also observed a novel single base substitution mutational signature that is entirely different from those currently described in COSMIC13 or in a recent in vitro study involving 79 carcinogens14 (figure 2A). When we extracted the missense mutations affecting Cosmic Cancer Gene Census genes in H015 tumour organoids, the trinucleotide mutation distribution was very similar to this new mutational signature (r=0.987), suggesting that H015 acquired most of these driver mutations through the signature (figure 2B). Indeed, when we compared the trinucleotide mutational signature of H015 with those obtained from the TCGA across all tumour types (n=9104), we identified one lung cancer tissue with a highly concordant spectrum (r=0.986) (figure 2C; online supplementary table 7). However, this patient with lung cancer received no chemotherapy prior to the cancer resection and it would be interesting to further explore the actual cause of this novel mutational signature. We believe that this mutational process did not operate continuously, as the signature was not found in the new mutations gained by the subclones of H015-NO after long-term culture (see online supplementary figure 3).
Supplemental material
Next, we examined tumour organoids (H020-TO and T2O) with a POLE hotspot mutation (p.P286R) from a 35-year-old woman, and both displayed the ultramutation phenotype (figure 1A and online supplementary figure 2B), with a mutation density of 198 and 196 SNVs/Mb for H020-TO and T2O, respectively. A higher incidence of POLE mutation has been previously observed in EOCRC.15 The mutation spectrum of H020 highly resembled a combination of the mutational signatures 10a, 10b and 28 (figure 2D), which were previously described in cancer tissue with POLE hotspot mutations. Notably, in this case, we observed heterogeneity of cancer driver mutations, including TP53, between different tumour regions.
Apart from these unique and novel features identified from the EOCRC, our CCOs also carried tumour drivers which have been identified in previous genome-wide studies on CRC tissues, including TP53, KRAS, BRAF, SMAD4, TCF7L2, ERBB2, PIK3CA, ATM and ARID1A.12 16 We also performed germline mutation calling using normal tissues or blood DNA from the 22 patients and found that none of the patients with CRC carried any known cancer causing pathogenic germline mutations. HX103 was a patient with known serrated polyposis syndrome, for which we previously described a germline RNF43 mutation, c.953–1, G>A, with loss of heterozygosity in the SSAO as the second hit (figure 1A),17 and our current exome sequencing study confirmed that finding.
We further studied the copy-number alterations of the CCOs, based on the WES data, using the cnv_facets tool. Several previously well-defined arm-level changes, including gains of 7 p and q, 8 p and q, 20 p and q, loss of 4q, 5q, 8 p, 14q, 15q, 20 p and 22q were observed (figure 2E). Pairwise analysis of 22 frozen cancer tissues and the CCOs again showed that the organoid cultures closely recapitulate the in vivo tumour, as demonstrated in previous studies.6 7 18–21 In essence, the 32 CCOs carried most of the key drivers, including variants and copy number aberrations found in the original tumours, with a few exceptions mostly attributed by the low tumour content in the frozen tissues (online supplementary figures 2, 4 and 5 and online supplementary tables 8 and 9). Analysis of CCOs derived from different tumour regions or lymph node metastasis showed variable degrees of heterogeneity, with differences in DNA copy number or driver mutations (online supplementary figure 6 and online supplementary table 9). Overall, our CCOs carried most of the known key driver alterations that were previously reported in CRC. More importantly, this EOCRC-enriched cohort includes tumours evolved from the serrated neoplasia pathway, as well as tumours carrying newly identified drivers, such as RSPO fusions, tumours with a novel mutation spectrum (H015) and a hypermutated phenotype (H020), thereby representing a unique resource for future biological studies.
Transcriptome profiling of tumour organoids showed distinct gene expression patterns, revealing diverse Wnt signalling-target therapeutic opportunities
Next, we studied the transcriptome of the organoids in the biobank by RNAseq (online supplementary figure 1 and online supplementary table 3). Under complete organoid medium growth conditions, which is a stem cell enriching culture, hierarchical clustering still showed distinct transcriptome patterns between tumour and normal organoids (figure 3A). Tumour organoids derived from the same patients always clustered together. Interestingly, the five RSPO fusion tumour organoids from H011 and H019 formed a small subcluster, suggesting they shared a similar transcriptomic profile. Surprisingly, they clustered together with normal instead of with other tumour organoids. Here, we are the first to reveal the transcriptome of the human RSPO fusion tumours at a pure epithelial cell level, and we found that their gene expression patterns highly resembled normal stem cells, including the expression of key colonic lineages such as enterocyte and goblet cell markers. Moreover, similar to normal organoids, they retained the expression of genes in several key signalling pathways, such as the BMP and TP53 pathways (despite the fact that they all carried 2-hit TP53 inactivation), while other tumour organoids showed downregulation of those pathways. Wnt signalling activity (as read out by generic Wnt target genes such as ZNRF3, ASCL2 and AXIN2) in RSPO fusion tumours was lower than APC mutant organoids even though RSPO fusions are known to amplify Wnt signalling.12 This modest elevation in Wnt activity was also seen in RNF43 mutant organoids.
While APC mutant tumours are known to be Wnt and R-spondin1 independent, with constitutively active Wnt activity to self-support stem cell renewal, we sought to test the Wnt-3A and R-spondin1 dependency in the CCOs lacking APC mutation in vitro. We found that the growth of RSPO fusion CCOs (H011 and H019), similar to RNF43 mutant organoids (H015 and HX103), were highly dependent on Wnt-3A but not on R-spondin1 (figure 3B).17 The role of RSPO fusion in conferring R-spondin1 independence, as shown here, was further confirmed by a recent study on engineering RSPO fusion in human colon organoids via R-spondin1 withdrawal from the culture medium.22 We did not detect any APC mutation in H012-TOs. However, we found a complete loss of APC protein in H012-TOs (online supplementary figure 7). The loss of APC wild-type protein in the organoids led to in vitro growth that was completely independent from Wnt-3A and R-spondin1, as well as to the upregulation of Wnt signalling with activity comparable to other APC mutant organoids, as shown by the transcriptome profile. Thus, we believe that there is a yet-to-be identified mechanism of APC alteration present in these H012 tumour organoids (figure 3B). Since deregulation of Wnt signalling is the hallmark of CRC and accounts for almost 90% of total CRC incidence, many Wnt signalling-targeted therapeutic strategies have been put forward. The growth of Wnt dependent CCOs (both RNF43 mutant and RSPO fusion CCOs) can be inhibited by porcupine inhibitors, such as IWP2 and LGK974, while anti-RSPO3 mAb treatment specifically works for RSPO fusion tumours.23 However, patients with APC mutations will not benefit from inhibition at the receptor-ligand level because growth of these cancer cells is completely independent of secreted Wnt. Therefore, an alternative targeting strategy is needed for APC mutation tumours. Indeed, we found that protein tyrosine kinase-7 (PTK7) expression was upregulated in APC mutant CCOs (figure 3A) and it ranked first in the differentially expressed gene list between Wnt dependent and independent CCOs (figure 4A and online supplementary table 10). PTK7 is a transmembrane protein and is expressed in the crypt base marking the LGR5 positive stem population.24 A PTK7-targeted antibody-drug conjugate (PF-06647020) has been developed that can induce tumour regression in animal models of various epithelial cancers,25 and is now in Phase I clinical trial. We confirmed, by immunostaining and flow cytometry, the high expression of PTK7 in APC mutant organoids (H002 and H009) and low expression in APC wild-type tumours (either RNF43 mutant or RSPO fusion TO), as well as in normal organoids (figure 4B). More importantly, PTK7 expression was found to be upregulated in CRISPR-Cas9 engineered normal human colon organoids with APC mutation but not with RNF43 mutation, indicating that PTK7 expression is specific to cells lacking APC (figure 4C–E and online supplementary figure 8). Even a previous study on small intestinal organoids derived from Ptprk-Rspo3 fusion mice showed low Ptk7 expression, whereas Apc-deleted organoids had elevated expression compared with their wildtype counterparts.26 Thus, PTK7 targeting may be beneficial to patients with CRC with APC mutations.
PTK7 is upregulated in Wnt independent organoids and in normal organoids engineered with APC mutation. (A) Differential gene expression analysis between Wnt independent and dependent tumour organoids. The x-axis depicts the average log counts per million (logCPM) of the TMM normalised expression values across the samples for each gene. The y-axis depicts the logFC of Wnt independent/dependent organoids for each gene. Red dots denote genes identified as being differentially expressed (logFC >2 or <–2 and BH FDR<0.05). The top upregulated and downregulated genes (PTK7 and RSPO3), based on the p value between two groups, are highlighted. Scatter plots (right) show the expression values (log2 TPM) of AXIN2, PTK7, RSPO3 and BMP2 in the organoids based on Wnt dependency. Blue line represents the mean and error bar represents SD. (B) Representative IF images of organoids with differential expression of PTK7. Strong membrane staining of PTK7 was found in APC mutant organoids. PTK7 was weakly expressed in RSPO fusion or RNF43 mutant CCOs as well as in normal colon organoids. Scale bar: 20 µm. Expression of PTK7 in the organoids was quantified by flow cytometry and corresponded to the IF results. Grey peaks represent the unstained control and green peaks represent PTK7 positive cells. The y-axis denotes the cell count and the x-axis denotes the FITC intensity. (C) Upregulation of AXIN2 and PTK7 gene expression in normal colon organoids with engineered APC mutation by CRISPR/Cas9. Expression of AXIN2 and PTK7 was modestly increased in normal colon organoids with engineered RNF43 mutation. Expression of BMP2 had the opposite trend. Each dot represents an independent clonal line. (D) Representative IF images for PTK7 protein expression in normal colon organoids with engineered APC or RNF43 mutations (top), and the corresponding flow cytometry quantitation (bottom). Scale bar: 20 µm. (E) Heatmap showing the differential gene expression of normal organoids engineered with different driver mutations, in terms of TPMlog2 values after mean centring. CCO,colon cancer organoid; HP, hyperplastic polyp; IF, immunofluorescence; logFC, log2 fold-change; TMM, trimmed mean of M-values; SA, serrated lesion.
The similarities between RSPO fusion CCOs and normal organoids prompted us to examine their propensity for differentiation on modification of the growth conditions. Wnt has been shown to have a strong inhibitory effect on enterocyte differentiation. As a result, normal colonic organoid cultures grown in complete medium (that includes Wnt) mostly consist of stem and undifferentiated progenitors that differentiate into various lineages on Wnt withdrawal. The growth of APC mutant organoids does not rely on Wnt or R-spondin1.5 Our organoid models offer a unique opportunity to study this process, since pre-existing cell models, including the two known RSPO fusion CRC cell lines, VACO6 and SNU-1411, do not require Wnt as part of their culture conditions.27 28 Therefore, we performed single-cell RNAseq analysis on the two RSPO fusion tumour-normal paired organoids (H011 and H019), with and without differentiation via Wnt withdrawal (online supplementary figure 1 and online supplementary table 3). We compared the expression levels of several well-established markers, including stem (LGR5), transit amplifying (TA) (MYC) and differentiated cell lineage markers (enterocyte: KRT20, FABP1; goblet cell: SPINK4 and neuroendocrine cell: CHGA) (figure 5 and online supplementary figure 9). Both the normal organoids and the RSPO fusion CCOs were enriched in stem and TA cells when cultured in complete medium. When they were stimulated to differentiate on Wnt withdrawal for 4 days, they showed maturation with significant shrinkage of stem and TA cell populations, with a marked increase in cells expressing differentiation and enterocyte markers, KRT20 and FABP1, and to a lesser extent goblet and enteroendocrine markers. We also examined four APC mutant organoids that showed sustained growth without Wnt or R-spondin. We found that the majority of cells were in the stem or TA state with very low expression of maturation markers (figure 5 and online supplementary figure 9B). We further analysed the lineage differentiation trajectories, focusing on enterocyte lineage, as this constitutes the predominant cell type in organoid culture. We found orderly differentiation and state transition from stem to early enterocytes and then to enterocytes in both the normal and RSPO fusion organoids, but the APC mutant organoids did not demonstrate such orderly maturation (figure 5B). Here, our results further demonstrate, at single cell resolution, the resemblance between RSPO fusion tumours and normal organoids. In contrast to APC mutant tumours, the RSPO fusion tumours retained their differentiation potential. This is an important feature to understand, as it predicts the consequence of therapeutic Wnt secretion inhibition in vivo.
Single cell transcriptome analysis of RSPO fusion organoids compared with normal organoids and APC mutant organoids. (A) Circos plots showing the single cell transcriptome analysis of colonic normal and cancer organoids in full medium (denoted by +Wnt) or with Wnt withdrawal (denoted by -Wnt). The stem cells are denoted by the LGR5 +population (red bar), the transit amplifying cells are denoted by the MYC +population (orange bar), without the expression of maturation markers of enterocytes (KRT20 or a high level of FABP1), goblet cells (SPINK4) or enteroendocrine cells (CHGA). In contrast, each mature cell population was strongly positive for one of these specific maturation markers. Both the normal organoids and the RSPO fusion CRC organoids were enriched in stem and TA cells when cultured in full medium, which on Wnt-3A withdrawal for 4 days, showed maturation with shrinkage of stem and TA cell populations, and the presence of cells expressing enterocytes, goblet and enteroendocrine markers. The four APC mutant organoids showed sustained growth without Wnt and remained enriched in stem and TA cells without the expression of maturation markers. Each circle represents the expression of a specific gene (from inside to outside, LGR5, MYC, KRT20, SPINK4 and CHGA), with the colour of each dot and its y-axis position denoting its expression level in one cell. A cut-off of 0.2 was used to designate the positive or negative status for all genes except FABP1, where 4.0 was used as the cut-off due to a high basal level. (B) Diffusion maps showing the enterocyte cell lineage in each organoid sample at the single-cell level. For each organoid sample, only cells classified as stem cells, early enterocytes or enterocytes were used and analysed with the diffusion map algorithm. Each dot in the diffusion maps is coloured based on the annotated cell type (right). The lineage trajectory with maturation from stem to early enterocyte and then to enterocyte can be observed in RSPO fusion and normal organoids, but was absent in APC mutant organoids. CRC, colorectal cancer; DC, diffusion component; NO, normal organoid; NOd, normal organoid with differentiation; TA, transit amplifying; TO, tumour organoid; TOd, tumour organoid with differentiation.
Association between PTPRK-RSPO fusion and SMAD4 mutation in the serrated neoplasia pathway
Previously, we and others have found that both RSPO fusion transcripts and RNF43 mutation frequently occur in both SSLs and traditional serrated adenoma.17 29 We further showed that somatic RNF43 mutation is strongly associated with BRAF mutation in both MSI and MSS CRC, highlighting its distinct role in the serrated neoplasia pathway.17 In this study, we noted that those EOCRCs carried either an RSPO fusion or an RNF43 mutation, lacked APC mutations, and all carried SMAD4 mutations (figure 1A). We confirmed the functional significance of the SMAD4 mutations, as these organoids continued to grow in the absence of Noggin and addition of BMP4 (online supplementary figure 10). Furthermore, the resemblance in gene expression patterns between RSPO fusion tumours and normal organoids, with high expression of BMP2 and lower Wnt activity, may favour differentiation (figure 3A). Indeed, when we engineered RNF43 or APC truncating mutations by CRISPR/Cas9 into normal organoids and performed RNAseq, we found a marked upregulation of Wnt signalling, and suppression of BMP2 or enterocyte differentiation markers in APC mutant organoids, while only a modest effect was seen in RNF43 mutant organoids (figure 4C,E). Thus, interruption of the BMP pathway may confer an additional growth advantage for the serrated neoplasia pathway. To explore this, we studied the mutational incidence of several key CRC drivers and their associations using the TCGA WES data and fusion transcript data from 464 CRCs (online supplementary table 11). We found PTPRK-RSPO fusions to be significantly associated with SMAD4 (p=0.022), BMPR1A (p=0.001) and BRAF V600E (p=0.006) mutations (figure 6A). Strikingly, apart from the enrichment in SMAD4 mutation, 33% of the CRCs with RSPO fusion also carried BMPR1A mutation (p=0.001), even though BMPR1A was mutated in less than 3% of CRCs. When we combined the BMPR1A and SMAD4 mutations, major regulators of the BMP signalling pathway, 66.7% of the CRCs with RSPO fusion showed defective BMP signalling, compared with 14.3% in CRCs without RSPO fusion (p=0.001) or 14.1% in CRC with APC mutation (p=0.001). We further validated the association between RSPO fusion and SMAD4 mutation by analysing data from two independent CRC cohorts established from Genentech (p=0.006)12 and the OncoTrack consortium (p=0.03)9 (figure 6A). The key driver mutations in the 16 RSPO fusion CRCs from the TCGA and Genentech cohorts are shown in figure 6B, in which 11 (68.8%) carried SMAD4 or BMPR1A mutation. Independently, in two RSPO fusion PDX models, both tumours once again carried SMAD4 mutations.23
RSPO fusion is associated with SMAD4 mutation across various cohorts. (A) Bar charts showing the association between tumours with or without RSPO fusion and various driver mutations, in terms of mutant frequency, in the TCGA, Genentech and OncoTrack cohorts. The actual number of patients with the corresponding driver mutation is listed in the tables on the right. Statistics were performed using the Fisher’s exact test. (B) Oncoplot of the 15 RSPO fusion tumours found in both the TCGA and Genentech cohorts, showing enrichment in SMAD4, BMPR1A and BRAF mutations. CRC, colorectal cancer.
Stability of normal and tumour organoids in long-term culture
Previous organoid studies only examined the genomic stability of normal organoids at the chromosomal level by karyotyping,30 31 while more detailed genomic characterisation using WES was also performed on a small number of CRC, gastric, breast and liver tumour organoids.10 18–20 However, these organoids were either cultured with varying time frames or within rather short periods of time (<3 months) in some studies. Here, we aimed to achieve a deep evaluation of the genomic stability of the organoid culture platform by performing exome-sequencing analysis on 10 tumour and 11 normal organoids, before and after 6 months of continuous culture (online supplementary figure 1 and online supplementary tables 2 and 3, 5). We also studied the genomic heterogeneity and in vitro clonal stability after long-term culture in normal organoids (online supplementary figure 1 and online supplementary tables 3 and 12, 9).
We found a distinct difference in the genomic stability between normal and tumour organoids during long-term culture (figure 7 and online supplementary figures 3, 11 and 12). For tumour organoids, similar to our previous studies on long-term culture of gastric cancer organoids,20 we found that the majority of CCOs, with few exceptions, maintained high concordance of SNVs between short and long-term culture, with the discrepancies mostly occurring in non-COSMIC-CGC genes or non-protein altering variants (online supplementary figure 11). Furthermore, both the mutant allelic fraction of the somatic mutations, which infers information on subclonal architecture as well as the chromosomal aberration pattern were stably maintained after long-term culture (figure 7A and online supplementary figure 11). We feel that these data provide important validation of the cancer organoid approach, as it underscores that cancer gene mutation frequencies are stably maintained over a relatively long period of in vitro culture.
Colon cancer organoids’ subclonal architecture was stably maintained in long-term culture, but normal organoids showed clonal dominance. (A) Stack bar showing that the correlation coefficient of both the mutant allelic fraction and copy number (LRR) between early and late passage tumour organoids were high, implying that the subclonal architecture was stably maintained after long-term culture. Representative scatter plots are shown on the right for one sample. SNVs are denoted by black dots and indels by blue dots. (B) Presence of clonal dominance of normal colon organoids after long-term culture. Scatter plots illustrate the mutant allelic fraction, with SNVs denoted by black dots, indels by blue dots and COSMIC genes by red dots. Note that both the x-axis and y-axis are compressed after 0.6, as there were few mutations beyond this range. (C) Enlarged view of sample H015, the leukaemia survivor in panel B, highlighting the decline of two pathogenic TP53 mutations and enrichment of a RNF43 mutation on long-term culture. The matched cancer organoid or cancer tissue of the patient carries a distinctly different TP53 (p.G154V) and RNF43 (exon8 c.850–1G>T) mutation. (D) Frequency distribution of mutant allelic fraction of somatic SNVs present in NO or NOL showing a shift of the peak from 0.1 to 0.2 in NO to 0.5 in NOL. LRR,log R ratio; NO, normal organoid; NOL, normal organoid late passage; TO, tumour organoid; TOL, tumour organoid late passage.
As shown previously, an individual adult human colon crypt is a clonal population harbouring a variable number of unique somatic mutations that accumulate over the life time of a specific stem cell.32 Thus, our normal organoid culture derived from pooled normal crypts is expected to contain a heterogeneous mixture of cell populations. Indeed, the mutant allelic fraction of somatic mutations tended to be low in the majority of the normal organoids, suggesting the existence of heterogeneous subclonal populations. Interestingly, on long-term culture, we noted a trend towards clonal domination in most of the normal colon organoids (figure 7B–D and online supplementary table 9). We observed a consistent shift, in a subset of mutations, from a mutant allelic fraction in the range of 1%–20% during short-term culture, to around 50% after long-term culture, implying that 100% of the cells carrying the mutation were in a heterozygous state. In parallel, there were other mutations that decreased from a variable percentage to 0%, suggesting a decline of the clone carrying these mutations. In addition, we noted some mutations that were present only after long-term culture, which could represent new mutations acquired in vitro, or mutations that pre-existed in a minor population which was beyond our detection limit due to the sequencing depth that subsequently became the dominant clone. When we examined the somatic variants that were enriched in the dominant clones on long-term culture, most of them were non-COSMIC-CGC genes, with a few exceptions (such as ARID1A, in H008-NOL and H011-NOL, and RNF43 in H015-NOL) (figure 7B). In contrast, we also observed a decline in clones carrying key cancer drivers such as TP53 mutations (H015) (figure 7C). Notably, the majority of these mutations in the normal organoids were not shared with their corresponding TOs or TFs, suggesting they were not relevant to the carcinogenic process of the specific cancer in the patient nor due to tumour contamination (online supplementary table 9; online supplementary figure 13). The chromosomal copy number also remained mostly stable (online supplementary figure 11). We then established two clonal organoid lines (C6 and C7) from H015-NO and performed WGS. We found that each clone contained both specific and shared SNPs (online supplementary figure 3, online supplementary table 12), with many more unique SNPs found in C7. This confirmed the presence of different subclones in the pooled organoids. However, both clones were depleted in the corresponding NOL, suggesting that they were less competent as compared with others. We also performed long-term culture on the C6 and C7 clonal organoid lines for an additional 4.5 months to document in vitro stability, following the Bloczijl et al protocol.33 There were ~75–120 new mutations gained per month in vitro in each subclone of C6 and C7, which was similar to what was previously observed in normal small intestinal organoids.33 We did not observe the characteristic new mutational signature in the newly acquired mutations, thus suggesting the mutation signature was due to past insult. Overall, our data suggest that for normal colon cells grown under organoid in vitro conditions, there was clonal competition that resulted in a subpopulation taking over. This finding is interesting given that previous studies of the normal intestinal crypt in vivo also showed the initial presence of several stem cells that were generally overtaken by clonal dominance through neutral drift dynamics.34 35
Discussion
We established an EOCRC enriched organoid biobank representing 20 patients with CRC. Compared with other CRC organoid biobanks, ours possesses several unique and novel features with therapeutic applications. First, our biobank is enriched in sporadic early-onset CCOs, with 22 CCOs derived from 11 EOCRCs (under age 50), while the three other existing independent cohorts, with age information, together only report seven organoids derived from three young patients with CRC.6–8 Second, we obtained five organoid lines from two young patients carrying a PTPRK-RSPO3 fusion, a genetic lesion not included in previous organoid cohorts. Together with the CCOs from H012, a SSL (HX103) and a hyperplastic polyp (HX102), we have derived 10 organoid lines derived along the serrated neoplasia pathway, while this subtype is not clearly documented in previous studies, except for three serrated polyps from the biobank described by Fujii et al.8 Although there are “classical” 2D CRC cell lines with RSPO fusions, such as VACO6 and SNU1411,27 28 they are routinely propagated in Wnt independent culture conditions. Whether these models preserve the key biology of the Wnt signalling pathway remains to be elucidated, but is unlikely. Our 3D model derived from RSPO fusion human colon cancers was extremely reliant on Wnt-3A in vitro, while retaining the in vivo genetic and phenotypic features. This indicates that they can serve as a good in vitro cell models for functional characterisation of pathogenic roles and drug response assays. Our model complements those derived by other groups, including Rspo fusion mice and their corresponding organoids,26 and several patient derived CRC xenografts with RSPO fusions. Indeed, treatment of RSPO fusion mice with adenomas or RSPO fusion PDX using Wnt inhibitors or anti-RSPO mAb has been reported to lead to growth inhibition, downregulation of stem cell markers and upregulation of differentiation markers,23 26 36 which is concordant with our observations. In our model, the differentiation and induction of CDKN2A on Wnt withdrawal was observed despite being in a TP53 null background. This is in alignment with a previous study using a mouse model, in which APC restoration in APC mutant cells could re-establish functional crypt units, with both proliferative crypt bases and differentiated lineages, despite the presence of an underlying p53 mutation.37 Induction of differentiation represents an attractive approach for cancer therapy, with a prior successful example of inhibition of IDH2 hot spot mutation leading to the maturation of leukemic blast cells to mature blood leucocytes in patients with leukaemia, resulting in complete remission.38 39 While replacement of APC is more difficult to achieve, several strategies exist to target RSPO fusion CRCs that are already in phase I clinical trials. This offers an attractive strategy of converting neoplastic crypts back into normal crypt units. Indeed, our RSPO fusion organoid models with linked single cell transcriptome data enable a deeper understanding of the therapeutic effect of Wnt inhibition, potentially facilitating the design of effective drug combinations and predicting potential routes of therapy resistance. Finally, we also acquired several hypermutated MSS organoids from two patients with EOCRC, one patient with a POLE mutation, while the other carried a novel mutational signature, further enriching the diversity of our biobank.
Our exome and transcriptome study of the RSPO fusion organoids also revealed new and interesting biology, as well as novel oncogenic pathway cooperativity that is tightly linked to the signalling processes along the colonic crypt axis. We have previously shown a BMP gradient in the human colon, with high BMP antagonist levels at the crypt base that may contribute to the maintenance of stem cells, and a higher BMP level at the top of the crypt driving cell differentiation.40 The presence of higher BMP2 and a higher propensity for differentiation in Wnt deficient conditions for the RSPO organoids mimics the crypt top compartment in the colon. According to the WHO Classification of Tumours of the Digestive tract (5th edition), ectopic crypt formation is one of the key histological features in the diagnosis of traditional serrated adenoma. It has been shown that blocking BMP signalling leads to the formation of ectopic crypt and epithelial hyperplasia in transgenic mice, with histology resembling human Juvenile Polyposis commonly caused by BMPR1A or SMAD4 mutations.41 42 More recently, two transgenic mouse models demonstrated that inhibition of the BMP pathway by GREM1 hyperexpression (a BMP inhibitor)43 or SMAD4 inactivation44 can resensitise mature villus cells to form ectopic crypts, and concurrent activation of Wnt signalling has a synergistic effect in initiating neoplasia. A recent study has attempted to re-create the traditional serrated adenoma from normal human colon organoids through CRISPR-Cas9 chromosome engineering of R-spondin fusion, TP53 mutation and BRAF mutation.22 These organoids, when implanted into mice, only formed flat serrated lesions. However, when they also overexpressed GREM1, these organoids formed polypoid tumours resembling TSA, with ectopic LGR5 positive cells detectable at mid-crypt positions. This study has therefore provided functional confirmation of the importance of BMP inactivation in the conversion of a flat serrated lesion to TSA. Although we did not observe elevation of GREM1 in our RSPO fusion organoids with SMAD4 mutation, taken together, SMAD4 or BMPR1A mutation, or GREM1 overexpression may serve as alternative mechanisms to inactivate BMP, which is critical in the progression of the serrated neoplasia pathway for enabling stem cell conversion in mature epithelial cells, thereby contributing to ectopic crypt formation.
Our novel demonstration of the presence of competition between subclones in normal organoid cultures resulting in clonal dominance is intriguing. It remains to be seen how much the in vitro competition and growth advantage resembles the normal competition that exists between stem cells residing within a single colon crypt, although the Matrigel organoid culture is designed to mimic the in vivo crypt growth conditions. A previous study showed that mutation of ARID1A is frequently observed in liver cirrhosis, with functional screens confirming that ARID1A mutation promotes clonal expansion and regeneration of hepatocytes on liver injury.45 ARID1A mutation has also been detected in normal colon crypts.32 Our observation of clonal dominance of the ARID1A mutant clones on long-term culture in two independent normal organoid samples is suggestive of a role for conferring a non-neutral growth advantage in the normal crypt. Interestingly, on long-term culture of H015 (an organoid line derived from the leukaemia survivor with many cancer driver mutations in her normal colon organoids), we observed a decline of two TP53 mutations and an enrichment of an RNF43 mutation, whereas the matched cancer organoids carried a different TP53 and RNF43 mutation, both in homozygous states. While these TP53 missense mutations in normal organoids were likely functional, they may be present in a heterozygous state in a small subpopulation, as we could not recover Nutlin-3a resistant clones from H015-NO. However, this highlights the field effect that exists in normal colon organoids, with many harbouring the first-hit inactivation of a tumour suppressor, conferring an increased chance for the second-hit to occur. Indeed, Lee-Six et al sequenced normal colon crypts and estimated that probable driver mutations were present in around 1% of crypts,32 some involving tumour suppressor genes, such as TP53, RNF43, ARID1A and AXIN2, which overlapped with what we observed. Overall, the effect of these different cancer drivers on normal versus cancer cell growth may be different. However, knowing the existence of this clonal heterogeneity and eventual clonal dominance may have an immense impact on study designs using pooled normal organoids for biological studies, disease modelling and large-scale cell-based screening, especially when coupled with the CRISPR/Cas genome editing technology.
Data repository
Sharing of organoids in this study with the scientific research community will be evaluated under the ethics approval of the Institutional Review Board of the University of Hong Kong and the Hospital Authority Hong Kong West Cluster on receipt of a research protocol and subject to a material transfer agreement with The University of Hong Kong.
Acknowledgments
We thank clinicians in the Hong Kong West Cluster, Hospital Authority for clinical care and Ms Dorothy Cheng for sample coordination. We thank the Genomics and Bioinformatics Cores of the Centre for PanorOmic Sciences (CPOS), The University of Hong Kong for performing the next generation sequencing and single cell sequencing with baseline bioinformatics data extraction. We thank the Imaging and Flow Cytometry Core of the CPOS for providing imaging and flow cytometry services. This paper has made use of the data generated by The Cancer Genome Atlas managed by the NCI and NHGRI. Information about TCGA can be found at http://cancergenome.nih.gov. We thank the Contributing Investigators who submitted data from the original study to dbGaP, the primary funding organiser that supported the Contributing Investigators and the NIH designated data repository.
References
Footnotes
Contributors HHNY, SYL and HC conceived the study. HHNY and SYL directed the study. HHNY, SYL, HC and STY contributed to the project design. HCS, YG and JWHW performed the bioinformatics data analysis. HHNY, SLH, SSKY, WYT, DC, ASC, AHYM, BCHL, ASYC, AKWC, HSH, AKLC, STY and SYL collected data, performed experiments and/or analysed data. OSHL and WLL contributed clinical data, samples and critical comments on the manuscript. HC contributed protocols and/or reagents and critical comments on the manuscript. SYL and HHNY analysed and interpreted data, wrote the manuscript with assistance, comments and final approval from all authors.
Funding This work was majorly supported by a Health and Medical Research Fund from the Food and Health Bureau, The Government of Hong Kong Special Administrative Region (Project No. 02132886) (on organoid establishment and exome sequencing analysis) and partly by a theme-based research grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. T12-710/16 R); a donation from the Hong Kong Jockey Club Charities Trust (on single cell transcriptome analysis); a Croucher Foundation Endowed Professorship to SYL and the Kadoorie Charitable Foundation.
Competing interests SYL and STY have received research sponsorship from Pfizer, Merck, Servier and Curegenix. HC is an inventor on several patents relating to Wnt activity in cancers and a pending patent on growing organoids from patients with colorectal cancer. He is a cofounder and SAB member of Surrozen, a start-up in Silicon Valley; a SAB member of Kallyope (New York), Merus (Utrecht) and Decibel (Boston); a non-executive board member of Roche (Basel) and SAB member of the Roche subsidiary Genentech (San Francisco) since 2019; a scientific advisor for and investor in Life Sciences Partners, a biotech venture capital firm located in Amsterdam.
Patient consent for publication Not required.
Ethics approval Institutional Review Board of The University of Hong Kong/Hospital Authority Hong Kong West Cluster.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data are available in a public, open access repository. The whole genome, exome and RNA sequencing data have been deposited into the European Genome-Phenome Archive with accession number EGAS00001004063 (https://www.ebi.ac.uk/ega/studies/EGAS00001004063). Application to a data access committee as a standard procedure in the EGA for data access containing human genetic data is needed, for the use of data for medical research by bona fide researchers. Normalised single-cell RNA sequencing data were deposited to GEO with accession number GEO142116 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE142116).