Epigenetics ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/kepi20 The DNA methylation landscape of human cancer organoids available at the American type culture collection Ricky Joshi , Manuel Castro De Moura , David Piñeyro , Damiana Alvarez- Errico , Carles Arribas & Manel Esteller To cite this article: Ricky Joshi , Manuel Castro De Moura , David Piñeyro , Damiana Alvarez- Errico , Carles Arribas & Manel Esteller (2020) The DNA methylation landscape of human cancer organoids available at the American type culture collection, Epigenetics, 15:11, 1167-1177, DOI: 10.1080/15592294.2020.1762398 To link to this article: https://doi.org/10.1080/15592294.2020.1762398 © 2020 The Author(s). Published by Informa Published online: 12 May 2020. UK Limited, trading as Taylor & Francis Group. Submit your article to this journal Article views: 2477 View related articles View Crossmark data Citing articles: 1 View citing articles Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=kepi20 EPIGENETICS 2020, VOL. 15, NO. 11, 1167–1177 https://doi.org/10.1080/15592294.2020.1762398 BRIEF REPORT The DNA methylation landscape of human cancer organoids available at the American type culture collection Ricky Joshia, Manuel Castro De Moura a , David Piñeyro a , Damiana Alvarez-Erricoa, Carles Arribasa, and Manel Esteller a,b,c,d a Josep Carreras Leukaemia Research Institute (IJC), Badalona, Barcelona, Spain; bInstitucio Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain; cCentro de Investigacion Biomedica en Red Cancer (CIBERONC), Madrid, Spain; dPhysiological Sciences Department, School of Medicine and Health Sciences, University of Barcelona (UB), Barcelona, Spain ABSTRACT ARTICLE HISTORY One caveat in cancer research is the dependence of certain experimental systems that might not Received 30 March 2020 really reflect the properties of the primary tumours. The recent irruption of 3D cultured cells Revised 17 April 2020 termed organoids could render a better representation of the original tumour sample. However, Accepted 21 April 2020 every laboratory has its own protocol and tissue-provider to establish these cancer models, KEYWORDS preventing further dissemination and validation of the obtained data. To address this problem, Organoids; cancer; DNA the Human Cancer Models Initiative (HCMI) has selected the American Type Culture Collection methylation; epigenetics; (ATCC) to make available organoid models to the scientific community. In this regard, no microarray; primary epigenetic information is available for these samples and, overall, the DNA methylation profiles tumours; cell lines; of human cancer organoids are largely unknown. Herein, we provide the DNA methylation validation landscape of 25 human cancer organoids available at the ATCC using a microarray that inter- rogates more than 850,000 CpG sites. We observed that the studied organoids retain the epigenetic setting of their original primary cancer type; that exhibit a DNA methylation landscape characteristic of transformed tissues excluding an overgrowth of normal-matched cells; and that are closer to the DNA methylation profiles of the corresponding primary tumours than to established 2D cell lines. Most importantly, the obtained DNA methylation results are freely available to everyone for further data mining. Thus, our findings support from the epigenetic standpoint that the ATCC human cancer organoids recapitulate many of the features of the disorder in the patient and are excellent tools to be shared among investigators for further tumour biology research. Background models cancer cell lines that grow in two- dimensions (2D). Since the first immortalized cell Cancer is still a devastating disorder with an line (HeLa) was established [1] there are now increasing incidence worldwide. Thus, a major hundreds of human cancer cell lines that reflect effort to understand the cellular and molecular all tumour types and have been characterized at pathways involved in its origin and evolution is the genomic and drug-sensitivity level [2–4]. Many a necessity. In this regard, advances in new tar- important discoveries have been produced from geted therapies and rescued pharmacological hits, these cancer models, but several issues that limit such as immunotherapy agents, are providing their applications are also known. These include more hope that the disease might eventually be that 2D cancer cell lines do not reconstitute the controlled. However, a major problem encoun- architecture of the original tumour tissue, the tered in the cancer research arena is the lack of acquisition of immortalized cell lines from pri- good laboratory models that recapitulate the fea- mary tumours is very inefficient, and that the tures of the disorder in the patient, a caveat that cancer cells could have changed to survive in could explain the lack of success of several clinical a monolayer on a plastic surface [5]. One comple- trials that at the pre-clinical stage were very pro- mentary approach would be the use of patient- mising. Many of these early studies in both aca- derived tumour xenografts (PDXs) that can demic and pharma laboratories use as initial CONTACT Manel Esteller, mesteller@carrerasresearch.org Josep Carreras Leukaemia Research Institute (IJC), Carretera de can Ruti, Camí de Les Escoles S/n, 08916 Badalona, Barcelona, Catalonia, Spain © 2020 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License (http://creativecommons.org/licenses/by-nc- nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way. 1168 R. JOSHI ET AL. maintain to a certain degree the heterogeneity of Cancer Research UK (CRUK), Hubrecht Organoid the primary tumour, including the cross-talk Technology (HUB), and Wellcome Sanger Institute between the cancer cells and the surrounding cell (WSI) have committed its efforts to make organoid microenvironment constituted by fibroblasts, models available to the scientific community [11,12]. endothelial and immune cells [6]. However, In this regard, many of the advances in tumour biol- major drawbacks for this technology are the lim- ogy in the last decades have been possible by the use of ited implantation rate for some tumour types and 2D established cancer cell lines with high-quality the very long time required to grow these PDXs standards provided to researchers around the world until they are useful for further experimentation. by the American Type Culture Collection (ATCC) One alternative scenario that arose in these pre- and, for this reason, the ATCC has been selected as clinical models of cancer was initiated by the use of the distributor of the HCMI models that are now three-dimensional (3D) culture methods, originally available to everyone (https://www.lgcstandards-atcc. established for neural stem cells that grew as neuro- org/hcmi). Importantly, DNA sequencing data are spheres [7]. From these systems, tumour spheroid available for these human cancer organoids, but they cultures were used for cancer stem cells and the field are missing a key layer of biological information: the exploded by the possibility to culture normal epithe- epigenetic setting constituted by DNA methylation lial stem cells as organoids. Organoids can be defined [13–16]. Thus, we have herein obtained and studied as a 3D structure derived from stem cells that mimic the DNA methylation landscapes of human cancer the cell types and self-organization present in the organoids available at the ATCC, in a similar manner original in vivo sample. The exceptional advance- that we provided the DNA methylation profiles for ment for epithelial-derived organoids, originally dis- the 1,001 Sanger [17] and NCI-60 [18,19] cancer cell covered in intestinal cells, was possible by achieving lines panels. Most importantly, we have deposited all WNT pathway activation by R-spondin1 and BMP the available DNA methylation data in the public pathway inhibition [8]. This procedure has now been Gene Expression Omnibus (GEO) repository to extended beyond colon cells to other tissues such as serve the scientific community in any further studies pancreas, liver, lung, breast, and others, and not only that use these very promising human cancer models. allowing the growth of the normal tissues but also of the corresponding tumour types [9,10]. The advan- tages of human cancer organoids are many and Results and discussion among these, we can cite the current efficiency of Human cancer organoids show tumour-type derivation from the most common epithelial specific DNA methylation profiles tumours types, the possibility to grow the neighbour- ing normal tissue part, the relative conservation of The genomic DNA from 25 embedded 3D cultures intratumoral heterogeneity, the great versatility to (organoids) available at the American Type Culture perform genetic manipulation and drug testing Collection (ATCC) (https://www.lgcstandards-atcc. and, finally, the capacity to test them for a specific org/HCMI#Organoids) were interrogated for their patient in a timely manner [9,10]. DNA methylation landscape using the Epigenetic The extraordinary expectations derived from the Infinium MethylationEPIC BeadChip (EPIC) emergence of human organoids for cancer research microarray from Illumina, as previously described can be limited by the required expertise and sophisti- [20]. These organoids represented five tumour cation to initially obtain, grow, and characterize these types: Colorectal (n = 11), Pancreatic (n = 7), cells, in addition, to the problem of exchanging data Oesophageal (n = 4), Stomach (n = 2), and Lung from highly diverse samples established around the (n = 1) cancer. The characteristics of these organoids, world. One possible solution is the creation of orga- including ATCC ID, model ID, primary site, acqui- noid biobanks that can expand long term, cryopre- sition site of the sample, clinical tumour diagnosis, serve, and provide these living materials to the histological type, and gender and race of the donor research laboratories. To achieve this aim, the are described in Table 1. The complete DNA methy- Human Cancer Models Initiative (HCMI), consti- lation data are freely available at the GEO repository tuted by the National Cancer Institute (NCI), under accession number GSE144213: EPIGENETICS 1169 Table 1. General information regarding the human cancer organoids used in the DNA methylation study. Clinical tumour ATCC ID Model ID Primary site Acquisition site diagnosis Histological type Gender Race PDM-2 HCM-CSHL-0057-C18 Colon Sigmoid colon Colorectal cancer Adenocarcinoma Male White PDM-4 HCM-CSHL-0060-C18 Colon Sigmoid colon Colorectal cancer Adenocarcinoma Male White PDM-44 HCM-SANG-0267-D12 Colon Transverse colon Colorectal cancer Tubulovillous adenoma Male White PDM-46 HCM-SANG-0269-C18 Colon Sigmoid colon Colorectal cancer Adenocarcinoma Male White PDM-5 HCM-CSHL-0061-C18 Colon Sigmoid colon Colorectal cancer Adenocarcinoma Female White PDM-7 HCM-CSHL-0063-C18 Colon Sigmoid colon Colorectal cancer Adenocarcinoma Male African PDM-8 HCM-CSHL-0064-C18 Colon Sigmoid colon Colorectal cancer Adenocarcinoma Male Asian PDM-9 HCM-CSHL-0065-C20 Colon Liver Colorectal cancer Adenocarcinoma Male Asian PDM-43 HCM-SANG-0266-C20 Rectum Rectum Colorectal cancer Adenocarcinoma Female White PDM-47 HCM-SANG-0270-C20 Rectum Rectum Colorectal cancer Adenocarcinoma Female White PDM-6 HCM-CSHL-0062-C18 Rectum Rectosigmoid Colorectal cancer Adenocarcinoma Female White junction PDM-36 HCM-CSHL-0089-C25 Pancreas Pancreatic head Pancreatic cancer Adenocarcinoma ductal Male Unknown type PDM-37 HCM-CSHL-0090-C25 Pancreas Pancreatic head Pancreatic cancer Other Female Unknown PDM-38 HCM-CSHL-0091-C25 Pancreas Pancreatic tail Pancreatic cancer Adenocarcinoma ductal Female Unknown type PDM-39 HCM-CSHL-0092-C25 Pancreas Pancreatic head Pancreatic cancer Adenocarcinoma ductal Male Unknown type PDM-40 HCM-CSHL-0093-C25 Pancreas Pancreatic head Pancreatic cancer Adenocarcinoma ductal Female Unknown type PDM-41 HCM-CSHL-0094-C25 Pancreas Pancreatic head Pancreatic cancer Adenocarcinoma ductal Female Unknown type PDM-90 HCM-SANG-0315-C25 Pancreas Other Pancreatic cancer Other Female White PDM-120 HCM-BROD-0100-C15 Oesophagus Oesophagus – distal Oesophageal cancer Oesophagus Male White third adenocarcinoma PDM-66 HCM-SANG-0290-C15 Oesophagus Oesophagus Oesophageal cancer Oesophagus Male White adenocarcinoma PDM-67 HCM-SANG-0291-C15 Oesophagus Oesophagus Oesophageal cancer Oesophagus Male White adenocarcinoma PDM-71 HCM-SANG-0295-C15 Oesophagus Oesophagus Oesophageal cancer Oesophagus Female White adenocarcinoma PDM-135 HCM-BROD-0115-C16 Stomach Lymph node Stomach cancer Other Male White PDM-146 HCM-BROD-0208-C16 Stomach Stomach (NOS) Stomach cancer Other Male White PDM-3 HCM-CSHL-0058-C34 Lung Right upper lobe lung Lung cancer Acinar adenocarcinoma Male White https://www.ncbi.nlm.nih.gov/geo/query/acc. its previous version, the 450 K microarray [23], the cgi?acc=GSE144213 EPIC microarray interrogates additional regions We observed discrete DNA methylation profiles with poor CpG content, intergenic sequences, between different organoid tumour types, which and distant regulatory regions such as enhancers enabled their distinction on the basis of the origin of regions provided by the ENCODE [24,25] and the primary using unsupervised hierarchical cluster- FANTOM5 [26] projects. However, the majority ing with multi-step bootstrap resampling (Figure 1 of DNA methylation information available from (a)). The final hierarchical clustering represented as databases, i.e., Gene Expression Omnibus (GEO) a tree-based dendrogram also distributed at two arms or The Cancer Genome Atlas (TCGA) is still the studied organoids: one arm corresponding to all derived from the 450 K platform. Because around the colorectal cancer organoids and the other 91% of the 450 K CpGs are also included in the arm that included the rest of organoid cancer types EPIC 850 K array [20], we speculated if limiting (Figure 1(a)). Dimensionality reduction analysis using our study to these shared methylation sites we t-Distributed Stochastic Neighbour Embedding could obtain similar results for organoid tumour (t-SNE) [21,22] provided very similar results type classification as described above. Using the (Figure 1(b)). same unsupervised hierarchical clustering with The herein used MethylationEPIC BeadChip multi-step bootstrap resampling for these loci, we Infinium microarray analyzes the methylation sta- again found specific DNA methylation profiles tus of 853,307 CpG sites (850 K). In comparison to that enabled the classification of the organoids 1170 R. JOSHI ET AL. Colorectal Esophageal Lung Pancreatic Stomach a c b d Figure 1. Cancer organoid DNA methylation landscapes retain the original tumour type epigenetic profile. (a and c) Heatmaps demonstrating unsupervised hierarchical clustering with bootstrap resampling of DNA methylation profiles for all organoid samples using (a) Infinium MethylationEPIC BeadChip (EPIC) (>850,000 CpGs) and (c) Infinium HumanMethylation450 BeadChip (450 K) (>450,000 CpGs). Red represents methylated CpGs, green unmethylated CpGs. (b and d) t-SNE plots further highlight organoid clustering based on their DNA methylation characteristics from EPIC (b) and 450 K (d). Each organoid sample is labelled according to its respective ATCC ID (b and d). Organoid DNA methylation profiles are displayed as brown (colorectal cancer), violet (oesophageal cancer), light blue (lung cancer), dark blue (pancreatic cancer), and yellow (stomach cancer). according to their cancer type (Figure 1(c)). The that for both EPIC (850 K) and 450 K sites in tree-based dendrogram also showed the two the hierarchical clustering (Figure 1(a,c)) and described arms; the colorectal cancer branch and t-SNE (Figure 1(b,d)) studies appeared among the other arm that included all the other remain- the colorectal cancer organoids. We applied, in ing tumour types (Figure 1(c)). Dimensionality this case, a further DNA methylation analysis, the reduction analysis by t-SNE yielded again identical EPICUP test, that provides the tissue of origin in results with all colorectal tumours in one group Cancer of Unknown Primary (CUP) [27] and col- whereas the other cancer types grouped away orectal cancer was the second obtained (Figure 1(d)). These results prompted us to use diagnosis. Thus, the epigenetic features of this the CpG sites common in both DNA methylation oesophageal cancer organoid closely resemble microarray platforms for further comparisons with a colorectal tumour. In this regard, it is worth other samples. mentioning that oesophageal adenocarcinomas It is worth mentioning one exception to the harbouring certain molecular defects, such as clear classification of organoids by tumour type ARID1A mutations and microsatellite instability, according to DNA methylation profiles. It was exhibit clinicopathological features fairly similar to the case of the oesophageal organoid PDM-66 colorectal adenocarcinomas [28,29]. EPIGENETICS 1171 Human cancer organoids maintain Human cancer organoids exhibit DNA cancer-specific DNA methylation characteristics methylation profiles that are closer to primary tumour samples than to established cancer cell Having shown that human cancer organoids lines available at the ATCC exhibit distinct DNA methylation profiles according to their tissue of As mentioned in the introduction, many studies origin, we wondered if they also keep the epige- dealing with the cellular and molecular features of netic dysregulation characteristic of the trans- tumours have been performed using human can- formed cells derived from each organ. This is cer cell lines that have proven to be a useful tool in a relevant point because tumour organoids do many cases, but these models also show important not grow quicker than their corresponding nor- caveats that question how they really mirror the mal organoid counterpart [9]. Importantly, in biology of the primary sample. In this regard, several cases, the cancer organoid grows slower organoids have been proposed to replicate better than the matching normal tissue, maybe related the features of the real cancer case within the to a higher percentage of mitotic failures and patient. Thus, to validate or not this concept in ulterior cell death in the tumour organoid the studied set of ATCC cancer organoids, we [30,31]. Strategies to avoid the growth of the compared the DNA methylation landscapes of normal tissue instead of the real tumour involve these tumour organoids with those of primary initiating the culture either with pure tumour tumours and cancer cell lines according to each material or the use of selective culture condi- cell type. The 450 K DNA methylation profiles for tions, such as the deprivation of growth factors the primary tumours were obtained from the that are essential for normal-tissue-derived orga- TCGA database (TCGA Research Network: noids [9]. Thus, to check if in any of the studied https://www.cancer.gov/tcga) (n = 500, 100 cases cancer organoids the healthy cells from residual for each one of the five studied tumour types) and normal tissue have outgrown the cancer cells the cancer cell lines from the Sanger collection and are the majority cell type in the available [17] (n = 204, corresponding to 49 colon, 31 organoid, we compared the DNA methylation pancreas, 35 oesophagus, 28 stomach and 61 lung profiles of the cancer organoids with those of cancer cell lines). the matching normal tissues. The DNA methyla- Applying the unsupervised hierarchical cluster- tion profiles for 50 normal tissues were obtained ing with multi-step bootstrap resampling metho- from the TCGA database (https://portal.gdc.can dology, we observed that overall the cancer cer.gov/), corresponding to 10 cases of each tis- organoids clustered within the primary tumours sue type: colon, pancreas, oesophagus, stomach, and rarely among the established cancer cell lines and lung. (Figure 4(a,b) and Figure 5(a-c)). In this regard, Using the unsupervised hierarchical clustering the tree-based dendrogram yielded two evident with multi-step bootstrap resampling approach, arms: one for the primary tumours that included we found specific DNA methylation profiles in also most of the organoids and another arm that the cancer organoids that enabled their distinction corresponded to the established cancer cell lines. from their corresponding normal tissues obtained Dimensionality reduction studies by t-SNE from the TCGA (Figure 2(a-e)). In this regard, the mimicked the hierarchical clustering data with tree-based dendrogram provided two clear arms the cancer cell lines grouped apart, whereas the for each tissue of origin: one for the cancer orga- majority of the cancer organoids clustered with the noids and another for the corresponding normal primary tumours (Figure 4(c-d) and Figure 6 tissues. Dimensionality reduction analyses by (a-c)). Thus, derived from the unsupervised hier- t-SNE mirrored the hierarchical clustering data archical clustering and t-SNE approaches, our data (Figure 3(a-e)). Thus, these results strongly sup- favour the idea that human cancer organoids avail- port that the studied organoids are really derived able at the ATCC resemble more the epigenetic from tumour cells without evident normal tissue setting of primary tumours than long-term 2D contamination. cultured cancer cell lines and can be very useful 1172 R. JOSHI ET AL. Colorectal Esophageal Lung Pancreatic Stomach Normal a b c d e Figure 2. Cancer organoid DNA methylation profiles are distinct to their healthy tissue counterparts according to unsupervised hierarchical clustering. (a–e) Heatmaps showing unsupervised hierarchical clustering with bootstrap resampling DNA methylation clustering of each cancer organoid against its healthy tissue equivalent. All healthy samples are represented in light grey and are pertinent to the organoid cell type. For example, the heatmap represented in Figure 2a, light grey represents normal colorectal tissue methylation, in Figure 2b, light grey represents methylation clustering in healthy oesophageal cells, etc. In all heatmaps, red represents methylated CpGs, green unmethylated CpGs, and black. Organoid DNA methylation profiles are displayed by cancer types: brown (colorectal cancer), violet (oesophageal cancer), light blue (lung cancer), dark blue (pancreatic cancer), yellow (stomach cancer), and light grey (corresponding normal tissue). All interrogated CpGs were analysed using only EPIC and 450 K shared probes. models to recapitulate the disease in the experi- organoids represent an excellent tool for cellular mental laboratory setting. and molecular studies of human cancer. Most importantly, their availability in ATCC constitutes an advantage to develop more objective, coordi- Conclusions nated, and shareable investigations in this area by We know very little about how well the molecular scientists worldwide. setting of human cancer organoids represents the cellular defects that occur in primary tumour sam- ples. Our ignorance in this area is larger if we just Methods focus in the epigenetic landscape where only in Sample preparation a very few cases the DNA methylation patterns have been recently explored in these tumoural Genomic DNA from 25 embedded 3D cultures experimental models [32,33]. Our analyses of the (organoid) representing 5 cancer subtypes were DNA methylation profiles in human cancer orga- purchased from the Human Cancer Models noids available from ATCC demonstrate that these Initiative, American Type Culture Collection biological systems maintain their epigenetic signa- (ATCC). The clinical tumour diagnoses for the ture characteristic of the original primary cancer five organoid cancer subtypes were Colorectal type, are truly transformed cells distinct from their (n = 11), Pancreatic (n = 7), Oesophageal corresponding normal counterparts and are more (n = 4), Stomach (n = 2) and Lung (n = 1) cancer. proximal to primary tumours than to long- Organoid IDs are listed in Table 1. Genomic DNA established cancer cell lines. Thus, these tumour was quantified using Pico Green fluorescence kit/ EPIGENETICS 1173 Colorectal Esophageal Lung Pancreatic Stomach Normal a b c d e Figure 3. Cancer organoid DNA methylation profiles are distinct to their healthy tissue counterparts according to t-SNE. (a–e) t-SNE of cancer organoid DNA methylation vs its normal tissue counterpart and each organoid sample is labelled according to its respective ATCC ID. Organoid DNA methylation profiles are displayed by cancer types: brown (colorectal cancer), violet (oesophageal cancer), light blue (lung cancer), dark blue (pancreatic cancer), yellow (stomach cancer), and green (corresponding normal tissue). All interrogated CpGs were analysed using only EPIC and 450 K shared probes. Qubit® 2.0 Fluorometer (life technologies). DNA house for quality control. idat files created by the integrity of all samples was checked by electro- Infinium HumanMethylation450 (450 K) BeadChip phoresis in a 1.3% agarose gel. Bisulphite conver- (Illumina) from primary tumours and healthy tissues sion of genomic DNA was carried out using the were obtained from The Cancer Genome Atlas EZ DNA Methylation Kit (Zymo Research) (TCGA) database using their recommended GDC following the manufacturer’s protocol and hybri- client data transfer tool. TCGA is a publicly available dized to the Infinium MethylationEPIC BeadChip database providing essential genomics resources to (Illumina). the cancer community. idat files corresponded to TCGA projects that reflected the clinical tumour diagnosis of the organoids samples; TCGA-COAD and READ (colorectal cancer), TCGA-ESCA DNA methylation microarrays (Oesophageal cancer), TCAG-LUAD (Lung cancer), Six hundred ng of bisulphite converted genomic TCGA-PAAD (Pancreatic cancer), and TCGA- DNA from all organoids were randomly distributed STAD (Stomach cancer). A random selection of in a 96 well plate and subjected to genome-wide 500 primary tumours (100/tumour type) and 50 DNA methylation examination of >850,000 CpG healthy samples (10/tissue type) from each project sites by the Infinium MethylationEPIC (EPIC) were analysed in this study. BeadChip (Illumina). DNA was hybridized to the 450 K data from the Sanger panel of cancer cell EPIC BeadChip following Illumina Infinium HD lines were obtained from our previous publication Methylation protocol and scanned using HiScan [17]. Where both EPIC and 450 K arrays were SQ system (Illumina). The resulting raw intensity examined in conjunction, idat files from EPIC files (idat) were subsequently pre-processed in- arrays were back-converted to 450 K arrays and 1174 R. JOSHI ET AL. Colorectal Pancreatic Primary tumor Cancer Cell line a b C D Second dimension Second dimension First dimension First dimension Figure 4. DNA methylation profiles from colorectal and pancreatic cancer organoids in the context of primary tumours and cancer cell lines. Heatmaps representing unsupervised hierarchical clustering with bootstrap resampling of DNA methylation in colorectal (a) and pancreatic (b) cancer organoid samples and their primary tumour [The Cancer Genome Atlas or (TCGA)] and cancer cell line [Wellcome Sanger Institute (Sanger)] counterparts. Red represents methylated CpGs, green unmethylated CpGs. (c and d) t-SNE plots show organoid, primary tumours and cancer cell lines clustering based on their DNA methylation profiles representing (c) colorectal cancer and (d), pancreatic cancer. Each organoid sample is labelled according to its respective ATCC ID. (a – d) DNA methylation profiles are displayed as cancer types and tissues; brown (organoid colorectal cancer), dark blue (organoid pancreatic cancer), primary tumour (black) and cancer cell lines (dark grey). The primary tumours and cancer cell lines are respective of the organoid tissue in each figure. All CpGs were analysed using only EPIC and 450 K shared probes. combined at the first step of data processing to Problematic probes such as failed probes (detection minimize batch effects. >90% of 450 K CpG sites p value >0.01), cross-reacting probes, and probes are present on the EPIC array [20]; therefore, only that overlapped single nucleotide variants within ± shared CpGs were retained for evaluation. 1 bp of CpG sites were removed. Background correc- tion and dye-based normalization were performed using ssNoob algorithm (single-sample normal- DNA methylation quality control, normalization, exponential out-of-band). Lastly, all sex chromo- and filtering some probes were discarded. Final DNA methyla- Raw signal intensity values were initially QC’d and tion values for each CpG represented as β-values pre-processed from subsequent idat files in were analysed. Downstream analyses were per- R statistical environment (v3.6.1) [34] using the formed under the R statistical environment (v3.6.1) minfi Bioconductor package (v1.32.0) [35,36]. [34]. The complete DNA methylation data are freely Quality control steps were applied to minimize available on the GEO repository under accession errors and remove poor probe signals. Potential number GSE144213: labelling errors were discerned by examining methy- https://www.ncbi.nlm.nih.gov/geo/query/acc. lation status at sex chromosomes of each individual. cgi?acc=GSE144213 EPIGENETICS 1175 Esophageal Lung Stomach Primary tumor Cancer Cell line a b c Figure 5. DNA methylation profiles of oesophageal, lung, and stomach cancer cluster predominately with primary tumours according to unsupervised hierarchical clustering. (a – c) Heatmaps demonstrating unsupervised hierarchical clustering with bootstrap resampling of oesophageal cancer organoids, oesophageal primary tumours (TCGA), and oesophageal cancer cell lines (Sanger) (a), lung (b) and stomach (c) DNA methylation. Red represents methylated CpGs, green unmethylated CpGs. Organoids are represented in violet (oesophageal cancer), light blue (lung cancer), and yellow (stomach cancer). In black, all pertinent primary tumours and dark grey, the respective cancer cell lines. All CpGs were examined using only EPIC and 450 K shared probes. Esophageal Lung Stomach Primary tumor Cancer Cell line a b c Second dimension Second dimension Second dimension First dimension First dimension First dimension Figure 6. DNA methylation profiles of oesophageal, lung, and stomach cancer cluster predominately with primary tumours according to t-SNE. (a–c) t-SNE plots of oesophageal organoid DNA methylation visualization (a), lung (b) and stomach (c) with primary tumour (TCGA) and cancer cell line (Sanger) equivalents. Organoids are represented in violet (oesophageal cancer), light blue (lung cancer), and yellow (stomach cancer). In black, all pertinent primary tumours and dark grey, the respective cancer cell lines. All CpGs were examined using only EPIC and 450 K shared probes. Hierarchical clustering with multi-step bootstrap variance agglomerative method for hierarchical clus- resampling tering formation was applied using the R package Comparison of intrinsic epigenetic characteristics hclust function [38]. The final hierarchical clustering between organoids and primary tumours, cancer cell was represented as a tree-based dendrogram and lines and healthy tissues was conducted by examining plotted in a heatmap generated by 5000 random β empirical β values for >400,000 CpG sites genome values and constructed using R package heatmap.2. wide. Robust correlations with multistep bootstrap Unmethylated CpGs where β values are 0 are repre- resampling and Euclidian distance measures for unsu- sented in green, heterozygous methylation where β pervised hierarchical clustering were performed using values are 0.5 are represented in black and methylated the R package pvclust [37]. Ward.d2 minimum CpGs where β values are 1, are represented in red. 1176 R. JOSHI ET AL. Dimensionality reduction analysis using References t-Distributed Stochastic Neighbour Embedding [1] Scherer WF, Syverton JT, Gey GO. Studies on the propa- (t-SNE) gation in vitro of poliomyelitis viruses. IV. Viral multi- To further assess clusters found by hierarchical clus- plication in a stable strain of human malignant epithelial cells (strain HeLa) derived from an epidermoid carci- tering with multi-step bootstrap resampling and to noma of the cervix. J Exp Med. 1953;97:695–710. improve methylation data visualization, t-Distributed [2] Shankavaram UT, Varma S, Kane D, et al. CellMiner: Stochastic Neighbour Embedding technique was a relational database and query tool for the NCI-60 applied using R package Rtsne [21,22]. Fifty- cancer cell lines. BMC Genomics. 2009;10:277. thousand β values from across the genome in all tissue [3] Garnett MJ, Edelman EJ, Heidorn SJ, et al. Systematic types were randomly selected in 5000 iterations to identification of genomic markers of drug sensitivity in cancer cells. Nature. 2012;483:570–575. assemble and visualize sample features similarity. [4] Barretina J, Caponigro G, Stransky N, et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012;483:603–607. Author’s Contributions [5] Sasaki N, Clevers H. Studying cellular heterogeneity RJ and ME conceived and designed the approach, interpreted and drug sensitivity in colorectal cancer using organoid the results, wrote and revised the manuscript. M.C.dM. and technology. Curr Opin Genet Dev. 2018;52:117–122. D.P. provided bioinfoinformatic support. D.A.E. and C. [6] Bleijs M, van de Wetering M, Clevers H, et al. A. coordinated and performed the epigenomic experiments. Xenograft and organoid model systems in cancer All authors have read and approved the final manuscript. research. Embo J. 2019;38:e101654. [7] Eiraku M, Watanabe K, Matsuo-Takasaki M, et al. Self- organized formation of polarized cortical tissues from ESCs and its active manipulation by extrinsic signals. Disclosure statement Cell Stem Cell. 2008;3:519–532. ME is a consultant for Ferrer International and Quimatryx. The [8] Sato T, Vries RG, Snippert HJ, et al. Single Lgr5 stem remaining authors declare that they have no conflict of interest. cells build crypt-villus structures in vitro without a mesenchymal niche. Nature. 2009;459:262–265. [9] Drost J, Clevers H. Organoids in cancer research. Nat Funding Rev Cancer. 2018;18:407–418. [10] Tuveson D, Clevers H. Cancer modeling meets human We thank CERCA Programme/Generalitat de Catalunya for organoid technology. Science. 2019;364:952–955. institutional support. This work was supported by the Health [11] van de Wetering M, Francies HE, Francis JM, et al. Department PERIS-project no. SLT/002/16/00374 and Prospective derivation of a living organoid biobank of AGAUR-projects no. 2017SGR1080 of the Catalan colorectal cancer patients. Cell. 2015;161:933–945. Government (Generalitat de Catalunya); Ministerio de [12] Palechor-Ceron N, Krawczyk E, Dakic A, et al. Ciencia e Innovación (MCI), Agencia Estatal de Conditional reprogramming for patient-derived cancer Investigación (AEI), and European Regional Development models and next-generation living biobanks. Cells. Fund (ERDF) project no. RTI2018-094049-B-I00; the Cellex 2019;8:1327. Foundation; and ‘la Caixa’ Banking Foundation (LCF/PR/ [13] Esteller M. Epigenetics in cancer. N Engl J Med. GN18/51140001). 2008;358:1148–1159. [14] Baylin SB, Jones PA. A decade of exploring the cancer epigenome - biological and translational implications. Data availability Nat Rev Cancer. 2011;11:726–734. The complete DNA methylation data are freely available on [15] Feinberg AP. The key role of epigenetics in human disease the GEO repository under accession number GSE144213: prevention and mitigation. N Engl J Med. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= 2018;378:1323–1334. GSE144213 [16] Berdasco M, Esteller M. Clinical epigenetics: seizing oppor- tunities for translation. Nat Rev Genet. 2019;20:109–127. [17] Iorio F, Knijnenburg TA, Vis DJ, et al. A landscape of pharmacogenomic interactions in cancer. Cell. ORCID 2016;166:740–754. Manuel Castro De Moura http://orcid.org/0000-0002- [18] Nogales V, Reinhold WC, Varma S, et al. Epigenetic 8488-5306 inactivation of the putative DNA/RNA helicase David Piñeyro http://orcid.org/0000-0001-5633-3339 SLFN11 in human cancer confers resistance to plati- Manel Esteller http://orcid.org/0000-0003-4490-6093 num drugs. Oncotarget. 2016;7:3084–3097. EPIGENETICS 1177 [19] Reinhold WC, Varma S, Sunshine M, et al. The NCI-60 unstable Barrett Esophagus-associated adenocarcinoma. methylome and its integration into cellminer. Cancer Am J Surg Pathol. 2011;35:647–655. Res. 2017;77:601–612. [30] Drost J, van Jaarsveld RH, Ponsioen B, et al. Sequential [20] Moran S, Arribas C, Esteller M. Validation of a DNA cancer mutations in cultured human intestinal stem methylation microarray for 850,000 CpG sites of the cells. Nature. 2015;521:43–47. human genome enriched in enhancer sequences. [31] Verissimo CS, Overmeer RM, Ponsioen B, et al. Epigenomics. 2016;8:389–399. Targeting mutant RAS in patient-derived colorectal [21] van der Maaten LJP, Hinton GE. Visualizing cancer organoids by combinatorial drug screening. high-dimensional data using t-SNE. J Mach Learn Elife. 2016;15:5. Res. 2008;9:2579–2605. [32] Roerink SF, Sasaki N, Lee-Six H, et al,. Intra-tumour [22] Maaten LVD. Accelerating t-SNE using tree-based diversification in colorectal cancer at the single-cell algorithms. J Mach Learn Res. 2014;15:3221–3245. level. Nature. 2018;556:457–462. [23] Sandoval J, Heyn H, Moran S, et al. Validation of [33] Tao Y, Kang B, Petkovich DA, et al. Aging-like spon- a DNA methylation microarray for 450,000 CpG sites taneous epigenetic silencing facilitates Wnt activation, in the human genome. Epigenetics. 2011;6:692–702. stemness, and BrafV600E-induced tumorigenesis. [24] ENCODE Project Consortium. An integrated encyclo- Cancer Cell. 2019;35(315–328.e6). DOI:10.1016/j. pedia of DNA elements in the human genome. Nature. ccell.2019.01.005. 2012;489:57–74. [34] Core Team R. R: A language and environment for statistical [25] Siggens L, Ekwall K. Epigenetics, chromatin and gen- computing. Vienna, Austria: R Foundation for Statistical ome organization: recent advances from the ENCODE Computing. 2006. https://www.R-project.org/ project. J Intern Med. 2014;276:201–214. [35] Aryee MJ, Jaffe AE, Corrada-Bravo H, et al. Minfi: [26] Lizio M, Harshbarger J, Shimoji H, et al. Gateways to A flexible and comprehensive Bioconductor package the FANTOM5 promoter level mammalian expression for the analysis of Infinium DNA Methylation atlas. Genome Biol. 2015;16:22. microarrays. Bioinformatics. 2014;30:1363–1369. [27] Moran S, Martínez-Cardús A, Sayols S, et al. Epigenetic [36] Fortin JP, Triche TJ, Hansen KD. Preprocessing, profiling to classify cancer of unknown primary: normalization and integration of the Illumina a multicentre, retrospective analysis. Lancet Oncol. HumanMethylationEPIC array with minfi. 2016;17:1386–1395. Bioinformatics. 2017;33:558–560. [28] Drage MG, Tippayawong M, Agoston AT, et al. [37] Suzuki R, Shimodaira H. Pvclust: an R package for Morphological features and prognostic significance of assessing the uncertainty in hierarchical clustering. ARID1A-deficient esophageal adenocarcinomas. Arch Bioinformatics. 2006;22:1540–1542. Pathol Lab Med. 2017;141:970–977. [38] Müllner D. fastcluster: fast Hierarchical, Agglomerative [29] Farris AB 3rd, Demicco EG, Le LP, et al. Clustering Routines for R and Python. J Stat Softw. Clinicopathologic and molecular profiles of microsatellite 2013;53:1–18.