Functional impact of cancer-associated cohesin variants on gene expression and cellular identity

Abstract

Cohesin is a ring-shaped protein complex that controls dynamic chromosome structure. Cohesin activity is important for a variety of biological processes, including formation of DNA loops that regulate gene expression. The precise mechanisms by which cohesin shapes local chromosome structure and gene expression are not fully understood. Recurrent mutations in cohesin complex members have been reported in various cancers, though it is not clear whether many cohesin sequence variants have phenotypes and contribute to disease. Here, we utilized CRISPR/Cas9 genome editing to introduce a variety of cohesin sequence variants into murine embryonic stem cells and investigate their molecular and cellular consequences. Some of the cohesin variants tested caused changes to transcription, including altered expression of gene encoding lineage-specifying developmental regulators. Altered gene expression was also observed at insulated neighborhoods, where cohesin-mediated DNA loops constrain potential interactions between genes and enhancers. Furthermore, some cohesin variants altered the proliferation rate and differentiation potential of murine embryonic stem cells. This study provides a functional comparison of cohesin variants found in cancer within an isogenic system, revealing the relative roles of various cohesin perturbations on gene expression and maintenance of cellular identity.

Introduction

Proper gene expression is essential for maintaining cellular identity and dysregulation of gene control is associated with many human diseases, including cancer (Roy and Hebrok 2015). During eukaryotic gene regulation, long-range DNA interactions often form between gene promoters and distal cis-regulatory elements (Ong and Corces 2011; Bonev and Cavalli 2016). In order for a promoter to physically contact a distant cis-regulatory element, the intervening DNA must be looped out as the two elements are brought into close spatial proximity (Eagen 2018; Rowley and Corces 2018). Enhancer–promoter contacts are developmentally dynamic, and their dysregulation is associated with developmental disorders and tumorigenesis (Hill et al. 2016; Hnisz et al. 2016a; Long et al. 2016). There is limited understanding of the mechanisms through which these contacts form and subsequently function to control gene expression.

Formation of DNA loop structures that regulate transcription is dependent on the cohesin complex. Cohesin is a ring-shaped protein complex composed of three core subunits (SMC1A, SMC3, and RAD21), along with various accessory and regulatory factors including STAG1, STAG2, PDS5A, PDS5B, NIPBL, WAPL, and Sororin (Remeseiro and Losada 2013). The accessory factors associate with cohesin in combinatorial fashion and are thought to modulate and regulate its activity on the genome, although many open questions remain (Rudra and Skibbens 2013). Upon loading onto the genome at sites of transcriptional activity, cohesin appears to translocate or extrude chromatin thus generating a loop of DNA (Dowen and Young 2014; Eagen 2018; Rowley and Corces 2018). Extrusion is arrested when cohesin encounters specific chromatin-bound roadblocks, in particular the transcriptional apparatus or the 11-zinc-finger transcription factor CTCF (Hansen 2020). These sites with arrested cohesin serve as anchors of DNA loop structures that regulate local transcription in at least two ways. First, the formation of a cohesin-mediated DNA loop between an enhancer and promoter can increase expression of the gene, while loss of the interaction can decrease gene expression. Secondly, cohesin-mediated DNA loops anchored by pairs of distal CTCF sites can restrict enhancer activity to the region inside of the loop. DNA loops of this type often surround genes encoding master lineage regulators and have been termed insulated neighborhoods for their ability to constrain an enhancer’s activity to a promoter within the DNA loop defined by CTCF sites (Dowen et al. 2014; Ji et al. 2016). Disruption of an insulated neighborhood boundary can cause aberrant enhancer targeting to genes normally located outside the neighborhood and cause inappropriate gene expression (Dowen et al. 2014; Ji et al. 2016; Hnisz et al. 2016a). In addition to its role in organizing interphase chromatin, cohesin is also required for sister chromatid cohesion, DNA replication, DNA repair and dynamic restructuring of chromosomes during cell division. Because of its participation in these varied biological processes, the cohesin complex is essential for organismal and cellular function. While complete loss of cohesin function is not viable, less severe defects such as mutations or loss of non-essential subunits are often tolerated and cause diverse cellular consequences (Peters et al. 2008). Molecular insight into how cohesin organizes genome structure and impacts genome function is still lacking.

Cancer genome and exome sequencing has revealed that cohesin subunits undergo a wide spectrum of mutations in cancer (Losada 2014; Hill et al. 2016; Waldman 2020). Somatic mutations, insertions, and deletions in cohesin complex components are observed in a wide range of tumor types, whereas germline mutations in cohesin and its regulators are found in a group of developmental disorders known as cohesinopathies (Horsfield et al. 2012; Kandoth et al. 2013; Ley et al. 2013; Thol et al. 2014). Several types of mutations have been associated with disease, including missense mutations, haploinsufficiency from monoallelic gene inactivation, and complete gene inactivation. Recent studies have begun to investigate the wide variety of cellular consequences that result from distinct cohesin defects. Cohesin subunits STAG2 and SMC1A acquire somatic mutations in diverse tumor types, including bladder cancer and myeloid neoplasia (Kon et al. 2013; Ley et al. 2013; Kim et al. 2016). The STAG2 gene is X-encoded and is a frequent target of inactivating mutations, which are only partially compensated for by its paralogue, STAG1 (Hill et al. 2016; Arruda et al. 2020). Some cancers with inactivated STAG2 exhibit decreased cell viability and disrupted cell cycle, but conflicting data exist on whether these defects are attributable to cohesin’s role in mediating sister chromatid cohesion (Solomon et al. 2011, 2013; Balbás-Martínez et al. 2013; Lawrence et al. 2014; Taylor et al. 2014; Mullenders et al. 2015; Kim et al. 2016; Viny and Levine 2018; Benedict et al. 2020). An alternative hypothesis suggests that, rather than interfering with the cell cycle, cohesin mutants may contribute to disease pathology by altering genome structure and gene expression. Aberrant DNA looping could cause misregulation of oncogenes or tumor suppressor genes during tumorigenesis or alter expression of developmental regulators during cell state maintenance or differentiation (Horsfield et al. 2012; Hnisz et al. 2016a; Norton and Phillips-Cremins 2017; Bompadre and Andrey 2019). Further studies are needed to determine the diverse molecular and cellular phenotypes that arise from distinct cohesin perturbations, in order to understand both normal cohesin function in the cell and defects observed in human disease. In particular, comprehensive studies investigating multiple distinct alleles, from missense to loss of function mutations, in an isogenic system are essential for determining the importance of individual sequence variants to a pathogenic state and for further understanding of normal cohesin biology.

To investigate the model that oncogenic cohesin mutations may disrupt cohesin’s role as a spatial organizer of gene control, we utilized the CRISPR-Cas9 system to introduce disease-associated deletions and amino acid variants into cohesin subunits in murine embryonic stem cells (mESCs). Introduction of cohesin sequence variants did not alter overall levels of cohesin complex members in cells, indicating that the resulting phenotypes were not simply due to reduction or absence of the cohesin complex. Instead, most of the mESCs harboring cohesin sequence variants exhibited altered transcriptional regulation, including misexpression of genes at insulated neighborhoods and cell identity factors. Cellular differentiation and proliferation were impaired by some cohesin sequence variants but not others. The results indicate that individual cancer-associated cohesin variants are sufficient to cause misregulation of gene expression and suggest that aberrant cohesin function may contribute to human disease by altering expression of genes important for proper cellular identity and function.

Materials and methods

Cell culture

Murine embryonic stem cells (V6.5, male, derived from a C57BL/6(F) × 129/sv(M) cross) were grown under standard ESC culture conditions (Arruda et al. 2020; Justice et al. 2020). mESC media contained DMEM KO (Thermo Fisher Scientific, 10829-018), 15% fetal bovine serum (VWR, 97068-085), homemade leukemia inducing factor (LIF), 100 U/ml penicillin, 100 µg/ml streptomycin (Thermo Fisher Scientific, 15140-122), 100 µM beta-mercaptoethanol (Thermo Fisher Scientific, 21-985-023), 1× Non-essential amino acids (Thermo Fisher Scientific, 11140-050), and 1× Glutamax (Thermo Fisher Scientific, 35050-061). mESCs were grown on gelatinized tissue culture dishes and were passaged using TrypLE (Thermo Fisher Scientific, 12-604-039).

Genome editing

Genome-edited mESC lines were generated as previously described, with modifications (Arruda et al. 2020; Justice et al. 2020). mESCs were transfected with a single-stranded donor oligonucleotide (ssODN) repair template, and plasmids encoding a synthetic guide RNA (sgRNA), Cas9 and a fluorescent gene (eGFP or mCherry) using Lipofectamine 2000 (Thermo Fisher, 11-668-027). After 1–4 days, single cells that were GFP positive and/or mCherry positive were sorted by either UNC Flow Cytometry Core Facility staff using a FACSAria II (BD Biosciences, San Jose, CA, USA) or on a CytoSort Array using a CellRaft AIR System (Cell Microsystems). Fluorescent cells were either sparsely plated on irradiated murine embryonic fibroblast monolayers to form colonies or sorted into individual wells of a tissue culture dish. Individual colonies were expanded into clonal cell lines, screened for genome edits by PCR and Sanger sequencing, and cryogenically stored. Individual allele sequences were determined by PCR of the region surrounding the edit site, followed by TOPO-TA cloning (Thermo Fisher, K4575J10) and Sanger sequencing. sgRNA and ssODN sequences are provided in Supplementary Table S1 and were designed using the CRISPR tool (crispr.mit.edu) (Cong and Zhang 2015). Two independent clones were obtained for each genotype, except for Stag2^V181M which only had a single clone obtained.

RT-qPCR

All mESC lines were handled side-by-side during culturing, RNA extraction, and quantitative PCR measurements. Briefly, cells were resuspended in 1 ml Trizol (Thermo Fisher, 15596018), incubated for 5 min at room temperature and either stored at −80 °C or further processed. 200 µl chloroform (Sigma Aldrich, C2432) was added and mixed before centrifugation to separate organic and aqueous phases. The aqueous phase was recovered, mixed with 200 µl additional chloroform, and centrifuged. The aqueous phase was collected and RNA was precipitated by addition of isopropanol. Total RNA was quantified using a NanoDrop instrument (Thermo Fisher). cDNA was prepared with Superscript IV and oligo-d(T) primers (Thermo Fisher, 18091050) according to the manufacturer’s instructions. Quantitative PCR was performed using SYBRgreen Master Mix on an Applied Biosystems QuantStudio 6 qPCR instrument using primers found in Supplementary Table S2. For each clone, measurements were performed in triplicate for each of four biological replicates and normalized to Tbp.

Western blotting

Adherent cells were collected by washing with PBS, scraping and transferring to a tube for centrifugation. Cell pellets were either frozen on dry ice for storage or nuclear extracts were prepared. Cell pellets were resuspended in 10 ml Lysis Buffer A (10 mM HEPES pH 7.9, 10 mM KCl, 0.1 mM EDTA, and 0.1 mM EGTA) containing 1× protease inhibitor cocktail (PIC) (Sigma Aldrich, 11697498001) and incubated at 4 °C while rocking for 15 min. 1 ml 10% NP-40 was added, samples were immediately vortexed, and pelleted at 1350×g for 5 min at 4 °C. The pellet was resuspended in 1 ml of cold Buffer TEN250/0.1 (50 mM Tris-HCl pH 7.5, 250 mM NaCl, 5 mM EDTA, and 0.1 mM NP-40) containing 1× PIC and incubated for a minimum of 30 min while rotating at 4 °C. After spinning at max speed at 4 °C for 10 min, the nuclear fraction (supernatant) was collected. Precision Plus Protein Dual Color Standard (Bio-Rad, 1610374) and samples were run in 4–20% Tris-Glycine gels (BioRad, 4568094) and transferred to PVDF membranes (VWR 29301-856). Membranes were blocked for 1 h with 5% blocking grade buffer (BioRad, 170-6404) and incubated overnight with primary antibodies while rocking at 4 °C. Membranes were washed 3 × 10 min with TBS-T at room temperature and incubated with secondary antibodies for 1 h while rocking at 4 °C. Antibodies used: STAG1 (Bethyl Laboratories, A300-0157A), STAG2 (Bethyl Laboratories, A300-0158A), SMC1A (Bethyl Laboratories, A300-055A), H3 (Abcam, ab1791), anti-Rabbit secondary (GE Healthcare, NA934), anti-Goat secondary (Abcam, ab97100). Membranes were imaged using either Thermo SuperSignal West Pico PLUS or Thermo SuperSignal West Femto chemiluminescent substrates with an Amersham Imager 600 (GE Healthcare). ImageJ was used to quantify protein abundance.

Embryoid body differentiation

mESCs were differentiated into EBs using hanging droplet cultures as described (Behringer et al. 2016), with modifications. Briefly, mESCs were dissociated into a single-cell suspension and 1,000 cells were placed into 30 µl droplets hanging from the lid of a 10 cm dish, in media lacking LIF, and PBS was placed in the dish. After three days, images were acquired using an EVOS FL light microscope (Thermo Fisher). ImageJ was used to determine the area of the EB. If multiple EBs were found in a single droplet, the areas of each EB were summed and reported. For further differentiation, individual droplets were transferred to gelatinized wells of a six-well plate and maintained in media lacking LIF. Cultures were monitored daily for rhythmic contractions using an EVOS FL light microscope (Thermo Fisher). Significance was examined using Dunnet’s multiple comparisons test.

Proliferation assay

5 × 10⁴ cells were plated into wells of a gelatinized six-well tissue culture plate. At the indicated timepoints (24, 48, 60, and 72 h) cells were resuspended with TrypLE (Thermo Fisher Scientific, 12-604-039) and resuspended in PBS. Cells were then mixed with trypan blue and counted on a Countess II FL instrument (Life Technologies). Measurements were performed in triplicate, with three identical wells for each sample at each timepoint. The experiment was completed four times. All calculations are represented as a fraction of initial plating density and plotted using GraphPad Prism (GraphPad). Significance was determined using Dunnet’s multiple comparisons test.

Data availability

Cell lines are available upon request. The authors affirm that all data necessary for confirming the conclusions of the article are present within the article, figures, and tables. Supplemental Material available at figshare: https://doi.org/10.25386/genetics.13962011.

Results

Sequence variants in cohesin complex members

Large sequencing projects that aim to catalog disease-associated human variation have identified mutations in genes that encode members of the cohesin complex. We examined publicly available databases of disease-associated sequence variation (COSMIC, TCGA, and ENSEMBLE) to identify mutations that may impact cohesin activity (cancer.sanger.ac.uk) (Weinstein et al. 2013; Forbes et al. 2017). We observed a similar distribution of sequence variant types across these databases and report here the proportions from the COSMIC database in Figure 1. The Stag1 gene frequently acquired missense mutations, with some additional nonsense and silent mutations observed (Figure 1A). The location of missense mutations along the length of the STAG1 protein sequence reveals a broad distribution with only a few hotspots detected near the N-terminal end (Figure 1B). The Stag2 gene is encoded on the X chromosome and displays a distinct mutational spectrum from Stag1. While Stag2 frequently acquires missense mutations, it also shows an increased proportion of nonsense and frameshift insertions and deletions (indels) relative to Stag1, that are predicted to disrupt stable STAG2 protein levels (Figure 1C). The missense mutations in Stag1 and Stag2 may cause subtle defects by disrupting sites of post-translational modifications or a protein–protein interaction interface, preventing a conformational change, or may cause no defect at all. The location of missense mutations along the length of the STAG2 protein sequence is similar to STAG1, with a broad distribution punctuated by a few hotspots (Figure 1D). STAG1 and STAG2 appear to act redundantly in cohesin localization on the genome, yet loss of individual STAG proteins causes both distinct and overlapping gene expression changes (Arruda et al. 2020; Casa et al. 2020). It is interesting to note that the single copy of the Stag2 gene on the X chromosome is frequently inactivated in cancers by nonsense and frameshift insertions and deletions, whereas the two copies of the Stag1 gene tend to acquire missense mutations. While the STAG proteins are mutually exclusive members of the cohesin complex, SMC1A is a core member. We found that the X-encoded Smc1a gene rarely undergoes inactivating mutations, consistent with its essential functin in cohesin-mediated dynamic chromosome restructuring during the cell cycle (Forbes et al. 2017) (Figure 1E). Rather, Smc1a tends to acquire missense mutations predicted to be hypomorphic and not cause complete loss of function (Figure 1F). These results indicate that the Stag1, Stag2, and Smc1a genes acquire somewhat distinct mutation types at various frequencies in human cancers, which could in turn influence their contribution to disease pathology.

Figure 1.

Cancer-associated sequence variation in cohesin components. (A) Proportion of Stag1 mutations that are missense, in-frame insertion/deletion, nonsense, frameshift insertion/deletion or silent. Mutations are derived from the COSMIC database of somatic mutations in cancer (cancer.sanger.ac.uk) (Tate et al. 2019). (B) Lollipop plot showing frequency of missense and nonsense mutations in STAG1. STAG, stromal antigen domain; SCD, stromal in conserved domain; HEAT repeats, which mediate the interaction with RAD21 are indicated (Zhang et al. 2013; Hara et al. 2014; Kim et al. 2016). (C) Proportion of Stag2 mutations that are missense, in-frame insertion/deletion, nonsense, frameshift insertion/deletion or silent. Mutations are derived from the COSMIC database of somatic mutations in cancer (cancer.sanger.ac.uk) (Tate et al. 2019). (D) Lollipop plot showing frequency of missense and nonsense mutations in STAG2. STAG, stromal antigen domain; SCD, stromal in conserved domain; HEAT repeats, which mediate interaction with RAD21 are indicated (Hara et al. 2014; Kim et al. 2016). Two specific mutations are indicated and annotated with their one letter amino acid abbreviations (for example, V181M refers to p. Val118 changed to Met). (E) Proportion of Smc1a mutations that are missense, in-frame insertion/deletion, nonsense, frameshift insertion/deletion or silent. Mutations are derived from the COSMIC database of somatic mutations in cancer (cancer.sanger.ac.uk) (Tate et al. 2019). (F) Lollipop plot showing frequency of missense and nonsense mutations in SMC1A. ATPase domains and the hinge domain are indicated. Four specific mutations are indicated and annotated with their one letter amino acid abbreviations (for example, R96H refers to p. Arg96 changed to His).

Engineering isogenic cohesin variant mESCs

To investigate whether specific cohesin sequence variants cause defects in cohesin function, we created isogenic mESC lines harboring various individual cohesin sequence variants for direct comparison of the cellular and transcriptional consequences of impaired cohesin activity. The use of stable embryonic stem cell lines allows for interrogation of individual cohesin perturbations independent of other coexisting sequence variation in cancer cells and separate from the natural human variation of individual patients. mESCs also provide a well-characterized model system for investigating mechanisms of transcriptional control of cell state that are broadly applicable to other cell types. Therefore, CRISPR/Cas9 genome editing was used to engineer individual cohesin sequence variants found in human disease into their endogenous loci in the murine embryonic stem cell genome. Of particular interest are recurrent missense mutations of cohesin subunits, that are observed in multiple patients with distinct cancer types, as these variants could represent alleles of cohesin with partial loss of function that could provide insight into cohesin biology. In addition to frequent disease-associated variants, we were also interested in sequence variation within the N-terminal region of STAG2, a reported protein–protein interaction interface with CTCF (Xiao et al. 2011). Two particular disease-associated sequence variants, STAG2-V181M and STAG2-S202L, are located within the N-terminus of STAG2 and were, therefore, selected for further study. The cohesin variant mESCs generated include: frameshift indels in Stag1 and Stag2 (Stag1^−/− and Stag2^−/−), in-frame deletion in Stag2 (Stag2^Δ164-196), and amino acid variations in Stag2 (Stag2^V181M and Stag2^S202L) and Smc1a (Smc1a^R96H, Smc1a^R586W, Smc1a^R711W, and Smc1a^R816G) (Table 1). The vast majority of these mutations did not co-occur with other mutations to cohesin encoding genes (Table 1). In one instance, the SMC1A-R711W and STAG2-V181M mutations co-occurred in a highly mutated colorectal adenocarcinoma. Importantly, the genes that encode cohesin subunits show strong conservation between human and mouse, with the six amino acid variants engineered in this study being 100% conserved between the two species. The sgRNA sequences and repair templates used to generate these cell lines are indicated in Supplementary Table S1. Also, included in the table are the number of potential off-target locations when allowing for mismatches between the sgRNA sequence and the genomic sequence, as deteremined using the Cas-OFFinder tool from RGEN Tools (Bae et al. 2014). The genomic edits were identified by Sanger sequencing.

Table 1.

Cohesin sequence variants found in human disease

mESC line genotype	Sequence variant type	Disease relevance	Reported cases	Co-occurring cohesin mutationsa
Stag1−/−	Nonsense or frameshift indel	diverse cancer types	42	NDb
Stag2−/−	Nonsense or frameshift indel	diverse cancer types	391	NDb
Stag2V181M	V181L	lung adenocarcinoma	1	0
	V181M	hematopoietic neoplasm	2	0
	V181M	large intestine adenocarcinoma	1	4 (inc. SMC1A-R711W)
Stag2S202L	S202L	bladder carcinoma	1	0
	S202 nonsense	bladder carcinoma	1	0
Stag2Δ164-196	In-frame indel	none reported	0	ND
Smc1aR96H	R96H	acute myeloid leukemia	6	0
	R96H	endometroid carcinoma	1	1
Smc1aR586W	R586W	endometroid carcinoma	1	0
	R586W	acute myeloid leukemia	6	1c
	R586Q	acute myeloid leukemia	6	0
	R586Q	stomach carcinoma	1	0
	R586Q	colon adenocarcinoma	1	0
	R586Q	kidney Wilm’s tumor	1	0
Smc1aR711W	R711W	large intestine adenocarcinoma	2	6d (inc. STAG2-V181M)
	R711W	endometroid carcinoma	1	0
	R711W	acute myeloid leukemia	1	0
	R711W	adult T cell lymphoma/leukemia	1	0
	R711W	Congenital muscular hypertrophy- cerebral syndrome	1	0
	R711Q	chronic myelomonocytic leukemia	2	0
	R711Q	endometroid carcinoma	2	0
	R711L	large intestine adenocarcinoma	1	0
Smc1aR816G	R816H	acute myeloid leukemia	1	0
	R816G	adenoid cystic carcinoma	1	0
	R816G	Cornelia de Lange syndrome	1	N/A

^aCo-occurrence of indicated mutation with a mutation in Smc1a, Smc3, Rad21, Stag1, Stag2, Pds5a or Pds5b.

^bNot determined due to the large number of cases.

^cOne instance of SMC1A-R586W co-occurred with a STAG2-R110* nonsense mutation.

^dFive of these occurred in one case, the sixth in a second case.

One simple mechanism by which cohesin sequence variation may impair function is by reducing the transcript and/or protein levels of cohesin complex components. We investigated how cohesin sequence variants impact the expression or stability of the variant proteins, as well as other subunits of the cohesin complex. For each sequence variant, two independent CRISPR clones were analyzed, except for Stag2^V181M for which only a single clone was successfully generated. We examined Stag1, Stag2, and Smc1a transcript levels by reverse transcription-quantitative PCR (RT-qPCR), and found that Stag1^−/− cl2, but not Stag1^−/− cl1, showed greatly reduced levels of Stag1 transcripts, and normal levels of Stag2 and Smc1a transcripts (Figure 2A). Both Stag2^−/− and Stag2^Δ164-196 mESCs showed greatly reduced levels of Stag2 transcripts, but had relatively normal levels of Stag1 and Smc1a transcripts. Stag2^V181M and Stag2^S202L mESCS showed normal levels of Stag2, Stag1, and Smc1a transcripts. The mESCs with amino acid variants in Smc1a showed nearly normal levels of Smc1a, Stag1 and Stag2 transcripts, with less than a 2-fold change detected in any cell line (Figure 2B). These results indicate that the amino acid variants in Stag2 and Smc1a do not dramatically alter expression of the gene in which they are encoded, or expression of other cohesin complex members. Nearly all of the frameshift indels in Stag1 and Stag2, as well as the in-frame deletion in Stag2, lead to decreased transcript levels of the edited gene but did not alter levels of the other cohesin transcripts tested. We next examined the levels of SMC1A, STAG1, and STAG2 proteins in cohesin variant mESCs. Whereas the proteins containing amino acid variants were expressed at similar levels as in wild-type mESCs, the proteins that contained frameshift indels or a large in-frame deletion were not stably expressed (Figure 2C–F, Supplementary Figure 1). We note that Stag1^−/− cl1 contains an in-frame deletion within an exon which lead to normal levels of Stag1 transcripts but STAG1 protein was not detected. Both Stag2^−/− and Stag2^Δ164-196 mESCs showed loss of Stag2 transcripts and STAG2 protein, yet Stag2 amino acid variants, Stag2^V181M and Stag2^S202L, showed near wild-type transcript and protein levels. Some Smc1a variant mESCs displayed an up to 2-fold increase in cohesin transcripts, though this increase was not observed at the protein level. In contrast, the Smc1a^R586W mESCs showed slightly reduced STAG1 levels yet normal STAG2 protein levels. Overall, mESCs harboring cancer-derived amino acid variants exhibited near normal levels of cohesin subunits, while large deletions resulted in a dramatic reduction in protein levels. Importantly, the introduction of an amino acid variant or a large sequence deletion in one cohesin subunit did not disrupt levels of other cohesin complex members.

Figure 2.

Expression of cohesin components in isogenic mESCs harboring various cohesin sequence variants. (A) RT-qPCR analysis of Stag1, Stag2, and Smc1a transcript levels in Stag1 and Stag2 variant mESCs. n = 4 biological replicates. Cl1 refers to clone 1 and cl2 refers to clone 2. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05. (B) RT-qPCR analysis of Stag1, Stag2, and Smc1a transcript levels in Smc1a variant mESCs. Two independent clones were analyzed for each genotype. n = 4 biological replicates. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05. (C) Levels of STAG1, STAG2, and SMC1A proteins in Stag1 and Stag2 variant mESCs. Nuclear extracts were immunoblotted for the indicated proteins and GAPDH as a control. Blots are representative of two experiments. (D) Levels of STAG1, STAG2, and SMC1A proteins in Smc1a variant mESCs. Nuclear extracts were immunoblotted for the indicated proteins and GAPDH as a control. Blots are representatitive of two experiments. (E) Quantification of 2 biological replicates of the blots presented in C. (F) Quantification of 2 biological replicates of the blots presented in D.

Aberrant transcriptional insulation in cohesin variant mESCs

To understand if cohesin perturbations impact gene expression, we first investigated the activity of genes sensitive to transcriptional insulation. Insulated neighborhoods are DNA loop structures mediated by cohesin and CTCF, that constrain the activity of enhancers to specific target genes and prevent their activation of other nearby genes (Figure 3A). Insulated neighborhoods that contain super-enhancers and their highly expressed target genes are known as Super-enhancer Domains (SDs). SDs focus the activity of an enhancer on the highly expressed gene inside and prevent inappropriate activation of genes outside, which can occur if integrity of the insulating CTCF- and cohesin-mediated loop is compromised (Figure 3A). To test whether cohesin variants display altered gene expression at SDs, we performed RT-qPCR to measure transcript levels of gene pairs, where one gene is located inside and the other gene is outside of the SD: Gpa33/Ildr2, Laptm/Sdc3, Tdh/Xkr6, Dmtn/Bmp1, and Jam2/App. These insulated neighborhoods were previously shown to be sensitive to removal of CTCF binding sites or altered recruitment of cohesin or CTCF (Dowen et al. 2014; Arruda et al. 2020; Justice et al. 2020). Overall, mESCs lacking STAG1 showed moderate changes in gene expression at SDs with the outside gene often showing a significant increase in expression (Figure 3B). The Stag2^−/−, Stag2^Δ164-196 and Stag2^S202 mESCs showed mild changes to expression of genes inside and outside of SDs (Figure 3B). Stag2^V181M did not alter expression of the genes examined. Smc1a^R586W mESCs showed the strongest misregulation of gene expression at the SDs tested, with up to 16-fold changes in gene expression detected (Figure 3C). Smc1a^R711W, and Smc1a^R816G mESCs showed mild gene expression changes at SDs, while Smc1a^R96H did not alter expression of the genes examined (Figure 3C). Occasionally, two clones of the same mutation displayed some differences from one another. To address this, we first repeated the analysis of two such mutant cell lines, Stag2^−/− and Smc1a^R816G mESCs, and found that clonal differences were generally recapitulated (Supplementary Figure 2A). We next investigated whether off-target edits may underlie differences between clones. Our in silico analysis of potential off-target activity of these sgRNAs suggests that such events would be exceedingly rare, with Stag2^−/−, Smc1a^R816G, and Smc1a^R586W having no potential off-target locations when allowing for one mismatch with the sgRNA, and only 3, 0, and 1 potential off-target locations respectively, when allowing for 2 mismatches with the sgRNA (Supplementary Table S1). Importantly, the presence of two mismatches between a sgRNA and genomic sequence significantly reduces the chance of Cas9 cutting a site and, therefore, reduces the likelihood of generating an off-target edit (Anderson et al. 2015). We performed an experimental analysis of two clones of wild-type mESCs, termed WT^{Smc1a-R586W exp}, that experienced the sgRNA and clonal isolation process alongside two clones of Smc1a^R586W mESCs. The results show no significant gene expression changes in WT^{Smc1a-R586W exp} versus wildtype mESCs that were not exposed to the genome editing procedure (Supplementary Figure 2B). Overall, these results suggest that loss of STAG1 or STAG2, or introduction of recurring amino acid variants in cohesin observed in cancer, are sufficient to alter gene expression at Super-enhancer Domains, consistent with aberrant transcriptional insulation at cohesin-mediated insulated neighborhoods.

Figure 3.

Cohesin variant mESCs display altered gene expression at insulated neighborhoods. (A) Model of an insulated neighborhood where a cohesin and CTCF-mediated DNA loop focuses the activity of a Super-enhancer (pink) on a target gene inside the DNA loop (teal). This prevents the Super-enhancer from acting on the gene outside the DNA loop (orange). When transcriptional insulation is lost or impaired, the Super-enhancer can act on the orange gene outside the DNA loop and decrease activity on the teal gene inside the DNA loop. (B) RT-qPCR analysis of genes at five insulated neighborhoods in Stag1 and Stag2 variant mESCs. The inside gene (teal) and outside gene (orange) for each SD are adjacent to each other, and the log2 fold change value is indicated for the five insulated neighborhoods in each cell line. n = 4 biological replicates. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05. (C) RT-qPCR analysis of genes at five insulated neighborhoods in Smc1a variant mESCs. The inside gene (teal) and outside gene (orange) for each SD are adjacent to each other, and the log2 fold change value is indicated for the five insulated neighborhoods in each cell line. n = 4 biological replicates. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05.

Loss of pluripotency in cohesin variant mESCs

To further investigate the impact of cohesin variants in gene regulation, we examined expression of genes involved in cell identity. Embryonic stem cells maintain the pluripotent state through expression of the master transcription factors OCT4, SOX2, and NANOG. Decreased expression of these regulators in ESCs results in altered cell identity and differentiation (Nichols et al. 1998; Niwa et al. 2000; Chambers et al. 2003; Boyer et al. 2005; Loh et al. 2006). ESC differentiation, either upon receiving developmental cues or experimental disruption of pluripotency transcription factor expression, can result in entry into the endodermal, mesodermal, or ectodermal cell lineages (Keller 2005). In order to assess the cellular identity of mESCs harboring cohesin sequence variants, we performed RT-qPCR to measure transcript levels of pluripotency master transcription factors OCT4 (Pou5f1), SOX2 (Sox2), and NANOG (Nanog). Levels of pluripotency factors were generally decreased in all cohesin variant mESCs compared to wild-type, with Nanog transcripts being the most consistently reduced (Figure 4A). Expression of the ectodermal regulator PAX6 (Pax6) was significantly increased in all cohesin variant mESCs, whereas NESTIN (Nes) was mostly unchanged (Figure 4B). The endodermal regulators GATA6 (Gata6) and SOX17 (Sox17) were generally increased in expression in most cohesin variant mESCs compared to wild-type (Figure 4C). The mesodermal lineage-specifying factors FOXA2 (Foxa2) and Brachyury (T) showed inconsistent changes across the cohesin variant mESCs (Figure 4D). Many of the cohesin variants examined displayed some degree of altered stem cell identity via decreased expression of pluripotency factors and/or increased expression of ectodermal or endodermal lineage-specifying factors. To investigate whether the genome editing procedure contributes to changes in cellular identity, we performed additional analyses on two clones of wild-type mESCs that experienced the sgRNA and clonal isolation process alongside two clones of Smc1a^R586W mESCs. The results show some increased expression of the ectodermal marker PAX6 and no significant increases in pluripotency, endodermal, or mesodermal markers in WT^{Smc1a-R586W exp} mESCs versus wildtype mESCs that were not exposed to the genome editing procedure (Supplementary Figure 3). The uniform increase in Pax6 expression in all cell lines that experienced the genome editing process and clonal isolation process suggests that it may be a consequence of the procedure. While the mesodermal marker T and ectodermal marker Nestin were decreased in WT^{Smc1a-R586W exp} mESCs relative to wildtype, the decreased expression of a gene poised for upregulation upon lineage commitment is not a clear indicator of differentiation.

Figure 4.

Altered cellular identity of cohesin variant mESCs. (A) RT-qPCR analysis of pluripotency factors. Levels of Pou5f1 (OCT4), Sox2, and Nanog transcripts were measured in Stag1, Stag2, and Smc1a variant mESCs. n = 4 biological replicates. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05. (B) RT-qPCR analysis of ectodermal regulators. Levels of Pax6 and Nestin transcripts were measured in Stag1, Stag2, and Smc1a variant mESCs. n = 4 biological replicates. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05. (C) RT-qPCR analysis of mesodermal regulators. Levels of Foxa2 and T (Brachyury) transcripts were measured in Stag1, Stag2, and Smc1a variant mESCs. n = 4 biological replicates. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05. (D) RT-qPCR analysis of endodermal regulators. Levels of Gata6 and Sox17 transcripts were measured in Stag1, Stag2, and Smc1a variant mESC lines. n = 4 biological replicates. Significance was determined by Dunnet’s multiple comparisons test; *P < 0.05.

Differentiation potential is compromised in cohesin variant cells

To investigate the ability of cohesin variant mESCs to differentiate, we performed the embryoid body (EB) differentiation assay (Behringer et al. 2016). In this assay, 1,000 cells were cultured in a droplet of media lacking leukemia inhibitory factor (LIF), to allow for differentiation. The droplet was suspended from the lid of a sterile petri plate, which prevented cells from adhering to the surface of the dish, and cells were allowed to grow for three days (Figure 5A). During this process, cells differentiate into various lineages and form spherical EB structures in a process that mimics in vivo gastrulation (Sene et al. 2007). Cell lines that showed the strongest gene expression changes, were investigated for their ability to differentiate: Stag1^−/−, Stag2^−/−, and Smc1a^R586W mESCs. The overall size of EBs was significantly decreased in all Stag1^−/−, Stag2^−/−, and Smc1a^R586W mESC clones, as measured by their total area per droplet (Figure 5B). The two Smc1a^R586W variant clones displayed some heterogeneity, with one of the two clones showing a greater reduction in size than the other (Figure 5B). To further evaluate the differentiation potential of these EBs, we transferred each droplet to an individual well of a tissue culture plate and monitored the cells as they continued to grow in media lacking LIF. During subsequent days, the cells in the EB spread out, adhered to the surface of the plate and some patches of cells began rhythmically contracting, consistent with development of ES-derived beating cardiomyocytes (Boheler et al. 2002). In this un-directed differentiation assay, Stag1^−/−, Stag2^−/−, and Smc1^R586W cells have slightly reduced capacity to differentiate into cardiomyocytes compared to wild-type cells (Figure 5C). To test whether the decrease in EB size and potential defect in differentiation into cardiomyocytes were due to reduced cellular proliferation, we monitored cell growth rate over three days. Indeed, Smc1a^R586W mESCs show a reduced proliferation rate relative to wild-type, whereas Stag1^−/− and Stag2^−/− mESC clones showed modest or no defects (Figure 5D). These results demonstrate that Smc1a^R586W cells have a strong reduction in proliferation and possibly differentiation potential, whereas Stag1^−/− and Stag2^−/− mESCs have mild to no defects in cellular growth and differentiation.

Figure 5.

Impact of cohesin variants on cellular differentiation. (A) Schematic of the hanging droplet assay. 1,000 mESCs were plated in a 30 µl droplet of media hanging from the lid of a tissue culture dish. After 3 days of growth in media lacking LIF, embryoid bodies were imaged and their size was measured. Droplets were collected, dissociated and moved to individual wells of a 48-well tissue culture dish. Wells were monitored for 10 days for the presence of rhythmically contracting cells. (B) Total size of embryoid bodies after 3 days of differentiation in hanging droplet cultures lacking LIF. Within each droplet the total area of EBs was measured. If more than one EB was present in a droplet, the areas were summed and presented. WT n = 70, Stag1^−/− cl1 n = 70, Stag1^−/− cl2 n = 71, Stag2^−/− cl1 n = 69, Stag2^−/− cl2 n = 71, Smc1a^R586W cl1 n = 72, Smc1a^R586W cl2 n = 71. Asterisks indicate significant differences from wild-type mESCs. ****P < 0.0001, *P = 0.0124 as measured by Kruskal–Wallis test. Data merged from two biological replicates for each clone. (C) Propensity of Stag1 (top), Stag2 (middle), and Smc1a (bottom) variant mESCs to differentiate into cardiomyocytes. The proportion of wells with EB differentiation into at least one patch of rhythmically contracting cells. Error bars represent one standard deviation. No significant differences were detected as measured by two-way ANOVA. n = 2 biological replicates for each clone. (D) Proliferation rate of various Stag1 (top), Stag2 (middle), and Smc1a (bottom) variant mESCs, relative to the 50,000 cells plated at time 0. *P = 0.0479 at 72 h between wild-type and Stag1^−/− cl1; *P = 0.0489 at 72 h between wild-type and Smc1A^R586W cl1; *P = 0.0109 at 72 h between wild-type and Smc1a^R586W cl2 mESCs as measured by two-way ANOVA. n = 4 biological replicates.

Discussion

Here, we report the functional impacts of various oncogenic cohesin sequence variants on the regulation of gene expression and cellular identity. By engineering cohesin variant cell lines, we were able to directly compare the phenotypes of individual variants in an isogenic system, free of confounding effects from use of different cancer cell lines from distinct lineages and with varied mutational burdens. Additionally, our generation and independent analysis of multiple clonal cell lines limits the possibility of off-target effects of CRISPR/Cas9 editing contributing to the phenotypes observed in the variant mESCs. Several cohesin variants were found to be sufficient to alter gene expression and the maintenance of embryonic stem cell identity, implicating aberrant cohesin activity as a potential contributor to disease via misregulation of gene expression. Generally, the cohesin variants examined did not disrupt expression or steady state protein levels of the other cohesin complex members, indicating that the phenotypes observed resulted from altered cohesin function instead of loss of the complex. The Smc1a^R586W variant exhibited the strongest phenotypes with regard to altered expression of genes in insulated neighborhoods, cell identity genes and defects in proliferation and possibly differentiation. The results also indicate that the cohesin accessory subunits STAG1 and STAG2 are important regulators of gene expression and cellular identity. This work reveals that a subset of cohesin sequence variants found in cancer are sufficient to cause misregulation of gene expression and cellular identity, thus providing evidence that altered cohesin activity may contribute to disease through processes other than genome instability, aneuploidy, and altered DNA replication.

Many cohesin sequence variants tested in this study caused altered gene expression (Figure 6). Changes at insulated neighborhoods in Stag1^−/−, Stag2^−/−, and Smc1a^R586W mESCs were consistent with potential re-wiring of enhancer-gene pairs, with up-regulation of genes located outside of the DNA loop structures observed. These changes are consistent with those observed when individual insulated neighborhood boundary elements are deleted (Dowen et al. 2014; Ji et al. 2016; Hnisz et al. 2016a). Interestingly, loss of STAG1 or STAG2 showed gene expression changes in the same direction as the presence of SMC1A^R586W variant complexes, indicating that they might disrupt the function of cohesin at the anchors of DNA loop structures through a shared mechanism. SMC1A^R586W causes more pronounced changes in gene expression than STAG1 or STAG2 loss, which is consistent with previous findings indicating that disruption of core cohesin subunits causes stronger changes in genome architecture than disruption of accessory subunits. Whereas inactivation of RAD21 or NIPBL was shown to cause a strong breakdown of TADs, depletion of STAG2 caused modest reductions in the number and strength of TADs (Haarhuis et al. 2017; Rao et al. 2017; Schwarzer et al. 2017; Wutz et al. 2017; Kojic et al. 2018; Cuadrado et al. 2019). Other cohesin amino acid variants analyzed in our study, including the R96H, R711W, and R816G variants in SMC1A and the STAG2 amino acid variants V181M and S202L, caused moderate to minimal impacts on SD function and pluripotency. Therefore, these variants may be only weakly hypomorphic, but we do not rule out impacts at specific loci or subsets of cohesin sites and DNA loop structures in the genome. Occasionally, two independent clones of the same mutation displayed some differences from one another. We found that these differences were repeatable and were unlikely to arise from distinct off-target mutations since relatively few off-target locations were detected when allowing for 1, 2, or 3 mismatches with the sgRNA (Supplementary Table S1). To experimentally investigate potential off-target events, we analyzed wildtype cells that experienced the genome editing process and could potentially have off-target edits, but do not have on-target edits. The results show that wild-type cells that were exposed to the sgRNA and clonal isolation process did not display gene expression changes that resolve the differences between mutant clones. Rather, in the case where two mutant clones display some differences, this may be due to on-target effects from the genome editing process followed by downstream stochastic changes in gene expression and cellular differentiation that caused independent clones to take distinct paths of aberrant cellular identity.

Figure 6.

Summary of the impact of Cohesin variants on transcriptional insulation. (A) Proper transcriptional insulation is retained in wildtype, Stag2^V181M,and Smc1a^R96H mESCs, with proper gene expression largely retained. (B) Mild gene expression changes are observed in Stag2^−/−,Stag2^S202L and Smc1a^R711W, and Smc1a^R816G mESCs, consistent with altered transcriptional insulation. (C) Moderate gene expression changes are observed in Stag1^−/− and Smc1a^R586W mESCs, consistent with altered transcriptional insulation.

While cohesin and CTCF are recognized as important structural regulators of the genome, the molecular mechanisms by which the spatial organization of DNA impacts gene expression are poorly understood. Several studies have shown that deletions or alterations of individual DNA loop boundaries disrupt transcriptional insulation causing changes in gene expression (Lupiáñez et al. 2015; Guo et al. 2015, 2018; Flavahan et al. 2016; Hnisz et al. 2016b). In this study, we show that cancer-associated cohesin mutations are also capable of disrupting transcriptional insulation causing altered gene expression. These results are consistent with other work in the field showing that altered enhancer specificity and activity can lead to pathogenic transcriptional programs in disease contexts (Sur and Taipale 2016; He et al. 2019). How the various biological functions of cohesin are spatially and temporally directed on the genome is not well understood. In particular, the recruitment of cohesin to specific cis-regulatory elements and their subsequent spatial organization into DNA loops, hubs, or domains that influence transcription requires additional investigation. Future studies addressing extrusion speed and directionality, residence time of cohesin molecules at specific genomic sites, and the multi-way interactions that bring together combinations of enhancers and promoters will provide important molecular insights.

Introduction of cancer-associated cohesin sequence variants into mESCs caused misexpression of lineage-specifying factors. Ectoderm- and endoderm-promoting factors were upregulated, while pluripotency-defining factors were downregulated in nearly all of the cohesin variant mESCs. In addition, the differentiation potential may be reduced or delayed in Smc1a^R586W variant mESCs compared to wild-type. While a decreased proliferation rate may contribute to reduced differentiation, it does not fully explain the defects in Smc1a^R586W mESCs. Since cardiomyocytes develop from mesodermal precursors, it is possible that the cohesin variant mESCs have impaired differentiation due to the simultaneous expression of ectodermal and endodermal lineage determining factors. Co-expression of potentially opposing transcriptional programs could disrupt a coordinated differentiation process by allowing for cell fate plasticity rather than cell fate restriction into cardiomyocytes. Together, these data suggest that oncogenic cohesin variants alter gene expression and disrupt maintenance of the pluripotent state.

Our results are consistent with and extend previous work investigating specific STAG2 alleles. Previously, STAG2-V181M was identified in a myeloid neoplasm (Kon et al. 2013), and STAG2-S202L was identified in a bladder cancer (Solomon et al. 2013). Prior studies showed that both the expression of STAG2-V181M and STAG2-S202L and their incorporation into cohesin complexes occurs at wild-type levels (Kim et al. 2016). Furthermore, STAG2-V181M and STAG2-S202L did not impact chromosome stability, cellular proliferation rate, or sister chromatid cohesion when introduced into HCT-116 cells (Kim et al. 2016). Our results indicate that STAG2-V181M and S202L cause modest changes in expression of cell identity genes, which may indicate altered cohesin activity in the context of regulating gene expression. In combination with other mutations, these point mutations may therefore contribute to tumorigenesis by impairing maintenance of cell identity-controlling transcriptional programs.

Our findings of transcriptional dysregulation in SMC1A variant or STAG-null mESCs are consistent with results in other model systems. In previous work, SMC1A-R711 was mutated to glycine in murine hematopoietic stem and progenitor cells (HSPCs) and shown to alter expression of hematopoiesis-controlling transcription factors, accessibility at transcriptional regulatory elements, and impair differentiation (Mazumdar et al. 2015). Knockdown or knockout of Stag2 had similar effects on gene expression and differentiation in HSPCs (Mullenders et al. 2015). A direct comparison of the impacts of Stag1 and Stag2 ablation on HSPC function revealed that Stag2, but not Stag1, loss resulted in altered transcription and differentiation (Viny et al. 2019). In those cells, Stag2 loss was further associated with reduced transcriptional insulation and a loss of intra-TAD contacts around master regulator genes (Viny et al. 2019). STAG2 amino acid variants have also been implicated in cancer, as STAG2-D193N reduces STAG2 incorporation into cohesin and may participate in drug resistance in melanoma (Shen et al. 2016).

This study provides insight into how cohesin sequence variants may contribute to cancer. We find that specific cohesin variants disrupt transcriptional regulation and may act in disease contexts to destabilize cell identity. Indeed, in a murine model of AML, Smc3 haploinsufficiency caused transcriptional changes to the HSC cell identity program, but tumorigenesis only occurred once the proliferative driver FLT3-internal tandem duplication (ITD) was introduced (Viny et al. 2015). Similarly, Stag2 inactivation is a frequent second hit in Ewing sarcoma, and is associated with disease recurrence and poor clinical outcomes (Brohl et al. 2014; Crompton et al. 2014; Tirode et al. 2014). Our data further support a role for cohesin variants in destabilizing cellular identity as part of multiple genetic hits that ultimately result in tumorigenesis. Indeed, a current hypothesis of tumorigenesis involves epigenetic priming of cancer-initiating cells before additional oncogenic hits establish the cancer cell state (Vicente-Dueñas et al. 2018); we propose that our data are consistent with a similar role for cohesin sequence variants in cancer. Finally, the study of cancer-derived cohesin variants provides an avenue for identifying potential hypomorphic and separation-of-function alleles for study, rather than the total loss-of function conditions mostly studied to date. Further studies are needed to elucidate the roles of individual cohesin cancer variants in cohesin-mediated gene control on chromatin.