Degree and site of chromosomal instability define its oncogenic potential

Most human cancers are aneuploid, due to a chromosomal instability (CIN) phenotype. Despite being hallmarks of cancer, however, the roles of CIN and aneuploidy in tumor formation have not unequivocally emerged from animal studies and are thus still unclear. Using a conditional mouse model for diverse degrees of CIN, we find that a particular range is sufficient to drive very early onset spontaneous adenoma formation in the intestine. In mice predisposed to intestinal cancer (ApcMin/+), moderate CIN causes a remarkable increase in adenoma burden in the entire intestinal tract and especially in the distal colon, which resembles human disease. Strikingly, a higher level of CIN promotes adenoma formation in the distal colon even more than moderate CIN does, but has no effect in the small intestine. Our results thus show that CIN can be potently oncogenic, but that certain levels of CIN can have contrasting effects in distinct tissues.

Colorectal cancer (CRC) is a very common and deadly cancer type that has a high prevalence of CIN and aneuploidy. CRC is divided in heritable and sporadic types, both mainly consisting of microsatellite unstable (MIN, 15%) and CIN (85%) tumors 20 . CIN correlates with poor patient prognosis: a meta-analysis of 63 studies with a total of 10.126 CRC patients (60% aneuploid tumors, as a proxy for CIN) showed that CIN tumors responded worse than non-CIN tumors to 5-Fluoruracil treatment, and (progression free) survival was lower in CIN patients 14 .
Despite the correlations described above, the roles of CIN and aneuploidy in tumor formation are still unclear. Mouse models of CIN have occasionally shown sporadic, spontaneous tumors with very long latency (>12-18 months), and predominantly in spleen and lung [21][22][23][24][25][26][27][28][29] , suggesting CIN is not a potent cancer driver. In mice predisposed to cancer, CIN is either neutral [30][31][32] , promotes tumor formation 23,26,27,30,[33][34][35][36][37][38][39] , or, in some conditions, suppresses it 29,30,32,37 . Comparisons between these studies is however exceedingly difficult due to the use of different oncogenic backgrounds, to differences in tissues that were examined 40 , and to the manner and time by which the tissues were exposed to CIN. Moreover, technical limitations often precluded direct measurements of CIN in the relevant tissues, and oncogenic effects in the different models cannot be attributed to distinct CIN levels.
We therefore established a genetic mouse model that allows controlled induction of various degrees of CIN in a tissue-specific manner. With this model, we make direct comparisons and find striking differences in the consequences for tumor formation between the various degrees of CIN: moderate to high CIN levels are sufficient to drive very early intestinal tumor formation, with moderate CIN being the most effective. Moreover, the capacity of similar CIN levels to drive or promote tumor formation is not the same in distinct tissues.

Results
An allelic series for graded increases of CIN in vivo. To enable tissue-specific induction of a range of CIN levels, we created mouse strains carrying a conditional T649A (TA) or D637A (KD; kinase-dead) mutation in the spindle assembly checkpoint kinase Mps1 (Fig. 1a, b, Supplementary Fig. 1A-C). Similar mutations in human cell lines caused mild or severe CIN, respectively 41 . We reasoned that combining these Cre-inducible Mps1 knock-in (CiMKi) alleles (together and with wild-type (WT) Mps1) would result in an allelic series of CIN, ranging from very low (few missegregations with mostly mild errors) to very high (many missegregations with mostly severe errors).
CiMKi mice were born healthy and at Mendelian ratios. Activation of Cre recombinase by addition of 4-hydroxytamoxifen (4-OHT) to mouse embryonic fibroblasts (MEFs) from CiMKi; Rosa26-CreER T2 mice resulted in efficient recombination and expression of mutant Mps1 mRNA ( Supplementary Fig. 1D), from which mutant MPS1 protein was translated to comparable levels as wild-type protein ( Supplementary Fig. 1E). As expected, the allelic series caused graded reductions in MPS1 activity, as evidenced by acceleration of mitosis after mutant induction 42 (Fig. 1c) and reduced MAD1 levels at kinetochores 43 (Supplementary Fig. 1F). Time-lapse microscopy of 4-OHT-treated immortalized MEFs showed a striking increase in mitotic errors with diminishing MPS1 kinase activities (Fig. 1d, Supplementary Movies 1, 2), verifying the predicted phenotypes of the allelic series. As expected, induced CIN resulted in increased aneuploidy in primary MEFs ( Supplementary Fig. 1G). Mutant induction also occurred efficiently in vivo in four-week old CiMKi;Rosa26-CreER T2 mice ( Supplementary Fig. 1H), and analysis of anaphase figures in intestinal tissue sections showed that the expected range of CIN was induced (Fig. 1e). Moreover, single cell whole genome karyotype sequencing (scKaryo-seq 44 ) showed that both aneuploidy and karyotype heterogeneity were enhanced in vivo in the small intestine one week after induction of moderate, high, and very high CIN (Fig. 1f, Supplementary Fig. 1I). We thus conclude that the CiMKi mouse model enables spatio-temporal control of a range of CIN in vivo.

Moderate CIN leads to early intestinal tumor initiation.
Whole-body mutant inductions in CiMKi;Rosa26-CreER T2 mice disrupted small intestinal tissue organization (Fig. 2a). The extent of disorganization correlated with the degree of CIN and likely explained the severe weight loss seen in mice with high and very high CIN (Supplementary Fig. 2A). To study early CIN induction in the intestine without adverse effects on other organs, we generated CiMKi;Villin-Cre mice to enable mutant induction specifically in the intestinal tract from 12.5 days post coitum (dpc) 45 . All CiMKi;Villin-Cre mice were healthy and normally fertile, and showed no signs of intestinal dysfunction like diarrhea, weight loss ( Supplementary Fig. 2B) or any other signs of health problems like abnormal posture, immobilization, or unresponsiveness. Also, we did not observe any abnormalities in general tissue characteristics at the age of 12 weeks or 8 months, except for the moderate CIN groups ( Supplementary Fig. 2C, D). Moreover, moderate CIN had caused one or more lesions in the small intestine of these mice by as early as 12 weeks of age, as judged by methylene blue staining ( Fig. 2b-d). Using histological analyses, we confirmed that these mice had indeed developed multiple low-grade adenomas (Fig. 2e, Supplementary Fig. 2C, D) of variable sizes. These adenomas were positive for nuclear βcatenin, showing that CIN was sufficient to induce constitutive Wnt pathway activation (Fig. 2f). Also, in this moderate CIN group, we detected a large low-grade adenoma in the colon of one mouse (Fig. 2g).
To get more insight in the differences in responsiveness between the various degrees of CIN on tumor initiation, we further analyzed colon and small intestinal tissues from the 12weeks old mice for differences in survival and proliferative activity. We found no significant differences in the number of viable crypts (as determined by pHH3 positivity in each crypt; Supplementary Fig. 2E, F), nor in proliferative activities (ki67; Supplementary Fig. 2G, H). In both colon and small intestine, we observed an increasing trend with rising degrees of CIN for mitotic (pHH3 positive) cells per crypt ( Supplementary Fig. 2I, J), and for apoptotic cells (based on morphology) per crypt ( Supplementary Fig. 2K, L). However, we did not observe significant differences between the moderate to very high degrees of CIN.
CIN can thus lead to very early onset, spontaneous tumor initiation. This effect of CIN differed between the various degrees of CIN, with the strongest effect in mice with moderate CIN (Fig. 2c, d, g). Together, these data suggest that the various CIN levels differentially affect the chance of spontaneous intestinal tumor initiation. Examination of general tissue characteristics after CIN induction did not, however, provide an obvious explanation for this.
Degree and site define oncogenic potential of CIN. Human colorectal cancers (CRCs) are often aneuploid 46 , and the vast majority of these cancers are caused by loss-of-function mutations in genes of Wnt pathway components such as APC [47][48][49] . Moreover, loss of heterozygosity (LOH) of APC causes extensive polyp growths in patients with familial adenomatous polyposis coli (FAP) syndrome [50][51][52] . To examine the impact of CIN on a tissue predisposed to cancer, we next investigated CiMKi mice carrying a mutant Apc allele (Apc Min/+ ). Apc Min/+ mice normally develop around 30 adenomas in the small intestine and no or very few in the colon 53 . Note that the expected degrees of CIN in this tissue in CiMKi mice were directly verified by in situ analyses (see Fig. 1e). In contrast to CiMKi;Villin-Cre mice, very high CIN in the Apc Min/+ background (CiMKi KD/KD ;Apc Min/+ ;Villin-Cre) was embryonic lethal, precluding further analysis of this level of CIN. While mice with low CIN were sacrificed by the expected 12 weeks of age 30,36,54 , mice with moderate or high CIN had to be sacrificed at 6-8 weeks due to severe weight loss. Apc Min/+ mice with moderate CIN presented with a striking increase in the amount of small intestinal adenomas (Fig. 3a-c, Supplementary  Fig. 3A). Neither high nor low CIN, however, had the same effect, suggesting that adenoma formation in the small intestine is sensitive to a narrow range of CIN.
Macroscopic examination of the colons of the same CiMKi; Apc Min/+ ;Villin-Cre mice revealed that in contrast to control and low CIN mice, colons from moderate and high CIN mice were widely covered with large adenomas (Fig. 3d- 3B). This was most striking in the distal region of the colon, where aneuploid APC-mutant tumors also most frequently occur in humans 55,56 . Incidence was 100% ( Supplementary Fig. 3C), and the number of adenomas was substantially higher than reported for other CIN models in the Apc Min/+ background 30,36,37 . Importantly, CIN in organoids established from these colon adenomas still corresponded to the expected levels ( Fig. 3g, Supplementary Movies 3, 4), suggesting that high CIN levels were not selected against after adenoma formation and that there was no drift towards an 'optimal' CIN level. Of note: while individual adenoma sizes were comparable between all induced CIN levels (Supplementary Fig. 3D-G), adenomas with moderate or high CIN level had reached this size substantially earlier (6-8 vs. 12 weeks). We thus hypothesize that CIN advanced initiation, accelerated growth, or both. In humans, tumors in the distal part of the colon are often considered CIN as they are typically aneuploid and karyotypically heterogeneous 20,57 . Since the CiMKi;Apc Min/+ ;Villin-Cre mice with moderate and high CIN mimicked such distal colon tumors, we next assessed aneuploidy and heterogeneity of copy number alterations (CNAs) of colon adenomas. scKaryo-seq showed that both aneuploidy and karyotype heterogeneity were increased with moderate and high CIN (Fig. 3g, h, Supplementary Fig. 3H). Chromosome 18, which harbors the Apc allele (that is often subject to LOH in human FAP tumors 47,53 ), was diploid in the vast majority of cells. Since adenoma formation in Apc Min/+ mice requires LOH of wild-type Apc 58 and since CiMKi colon adenoma organoids grew independently of Wnt ligands, this indicated that LOH of Apc by CIN occurred in a manner other than whole chromosome 18 loss, as previously suggested 30 . Targeted PCR detected only Apc Min alleles ( Supplementary  Fig. 3I), strongly suggesting that LOH was accomplished either by double non-disjunction events of both chromosomes 18 or by somatic recombination 30,59 , the latter of which is likely the cause of APC LOH in FAP patients 51,60 . Both these processes could be accelerated by CIN.
Enhanced proliferation in colon but not in small intestine. Our data thus far show that the effect of CIN on karyotype heterogeneity and tumor formation in identical genetic backgrounds depends on the degree of CIN and the tissue in which CIN occurs. As high CIN caused massive colonic adenomas but did not increase adenoma formation in the small intestine, the effects of a similar range of CIN can be profoundly different between tissues. To better understand the tissue-dependent sensitivities to CIN, we first assessed the possibility that the same CiMKi mutations had resulted in different CIN levels in small intestine vs. colon. Although time-lapse imaging of colon and small intestinal organoids showed that CIN levels were not identical between the two tissues (in Apc Min background), genotypes with drastically different impacts on tumor formation (Fig. 3c, f) had comparable CIN levels (e.g., CiMKi TA/TA in colon vs. CiMKi TA/KD in small intestine) (Fig. 4a, b). Strikingly however, moderate and high CIN caused significant expansion of the proliferative compartments in the colons of CiMKi; Apc Min/+ ;Villin-Cre mice at four weeks of age (roughly the time of adenoma initiation), but not in the small intestine ( Fig. 4c-f, Supplementary Fig. 4A, B). The percentage of proliferating cells within the compartment (proliferative index) was similar across genotypes ( Supplementary Fig. 4C, D), thus the total amount of proliferating cells in the crypts was enhanced (Fig. 4c, d, Supplementary Fig. 4E, F). Cells might therefore be more readily retained in a proliferate state in the colons of moderate and high CIN mice, increasing the chance that transformed cells propagate in colonic crypts. The fact that this increased proliferative state was not observed in the small intestine again underscores the difference in CIN response between these tissues. Although previous studies have reported spontaneous tumor formation in CIN mice, it occurred sporadically and with late onsets of more than 12 months of age [21][22][23][24][25][26][27][28][29] . By contrast, our CiMKi mice with moderate CIN (CiMKi TA/TA ) developed a substantial number of lesions in the small intestine, and in one case in the colon, as early as 12 weeks of age. This shows that CIN is a more potent driver of tumor initiation than previously thought. As our model was able to probe the effects of a wide range of CIN, it is possible that the optimal CIN level for tumor induction was not reached in prior studies. The inducible nature of the CiMKi model is most probably a prerequisite for reaching the higher CIN levels. Many other CIN models were not inducible or tissue-specific, resulting in embryonic lethality in homozygous knock-outs, possibly due to severe missegregations during the developmental stages 22,23,[61][62][63] . Also, we used the CiMKi model here to induce CIN locally in the gastro-intestinal tract, thereby preventing adverse effects in other tissues.
Despite the outstanding tumorigenic potential of specifically the moderate degree of CIN, we did not find significant differences in survival or proliferative activities of crypts, nor in apoptotic responses between the moderate and the higher degrees of CIN. This indicates that additional factors may play a role, and/or that the moderate degree of CIN is the most effective in inducing aneuploidies that are beneficial for adenoma initiation or growth. Importantly, we confirmed with scKaryo-seq that the higher degrees of CIN lead to aneuploid populations in healthy intestinal tissue. Aneuploidies were also observed in adenomas that arose after induction of moderate or high CIN, suggesting that at least a portion of the aneuploid populations induced by CIN are being propagated as proliferative cells in the adenomas.
Our model enabled us to directly compare the effects of the various CIN levels between two different tissues -small intestine and colon-within the same mice. Whereas in the Apc Min/+ background moderate CIN levels markedly increased adenoma burdens in both tissues, a higher CIN level did not affect the small intestine, but increased the adenoma burden in the colon even more. Our data therefore do not support the previously proposed model in which low CIN levels promote tumorigenesis while high CIN leads to cell death and tumor suppression 32,64 . Instead, we argue that the role of different CIN levels is much more complicated, as similar levels of CIN can have contrasting effects in distinct tissues.
Several factors might account for the different effects between small intestine and colon: our finding that moderate and high CIN leads to an enhanced proliferative state in colon but not in small intestine shows a remarkable difference in response between the two tissues. Hyperproliferation of crypts could be an adaptive response to cell death or arrest due to high CIN. In that case, differences in tolerance for high CIN between the two tissues could play a role. Enhanced proliferation can increase the chance that transformed cells propagate in colonic crypts 65 , however it does not account for the tumorigenic effect of moderate CIN in the small intestine. Another explanation might be differential impacts of Apc loss on CIN between colon and small intestine: it was previously reported that loss of Apc impacts on fidelity of chromosome segregation 66 . However, in a more recent study it was reported that inactivating APC mutations in human organoids do not significantly induce CIN 67 . Importantly, in our own experiments we found that the frequency of errors in Apc Min/+ organoids (CiMKi wild-type) is very low, and that the Apc Min/Min tumors are diploid, suggesting that the Apc mutation alone does not induce substantial CIN. Therefore, other factors besides CIN itself might play a role as well, such as the normal variations throughout the gut in stem cell number and physiological Wnt activity 68 or in the adaptive immune landscape 69 . It will be exciting to further investigate the underlying mechanisms, as it can impact on future treatment strategies for different cancer types. The tissue-specific inducible nature of the CiMKi model enables studying the impact of various CIN levels in many other organs as well.
Apc Min/+ mice 54,70 have been widely used to model human FAP, the hereditable form of CRC 52 . Tumorigenesis in these mice requires LOH of Apc, which directly mimics the human disorder, and is genetically comparable to tumorigenesis in FAP patients. However, whereas in human FAP patients mostly colon tumors occur, Apc Min/+ mice develop many adenomas in the small intestine but very few in the colon. It is therefore intriguing that addition of moderate to high CIN to this model causes early occurrence of colon adenomas, thus making it a better model for human disease. Furthermore, as in human FAP, colonic adenomas of CiMKi;Apc Min/+ mice are predominantly located in the distal colon and are aneuploid and karyotypically heterogeneous 71 . Also, in contrast to adenomas from Apc Min/+ mice 58 , LOH in CiMKi;Apc Min/+ mice did not occur by loss of (part of) the wild-type allele. Instead, we found disomy of the mutant chromosome 18 in the adenomas, similar to a CIN model driven  Supplementary Fig. 3H). Graphs show individual cells (horizontal lines) of one example per genotype. Colors indicate copy number state for a given chromosome. Statistics for panels c, f, and g: one-tailed Welch's t-test, comparing each group to CiMKi WT/WT ;Apc Min/+ ;Villin-Cre, exact p-values are indicated when p < 0.05, and p < 0.0001 is indicated by asterisks (****). Source data for panels c, f, and g are provided as a Source Data file.
by Bub1 insufficiency 30 . Other processes by which LOH can be achieved are somatic recombination events 59 , as was described for APC LOH in human FAP tumors 51,60 , or double non-disjunction events during mitosis 30 . CIN can be involved in both processes: DNA damage and double strand breaks as a result of CIN may be repaired through recombination with the mutant allele, and doubling of the mutant chromosome accompanied or followed by loss of the wild-type allele by non-disjunctions during anaphase can lead to disomy of the mutant. So even though the exact mechanism remains to be uncovered, LOH of Apc in the CiMKi; Apc Min/+ mice is most probably accelerated by CIN. Importantly, in the colon but not in the small intestine, moderate CIN had a very high tumorigenic potential in Apc Min/+ mice, but only very modest in the wild-type background. This could indicate that specifically in the colon, CIN promotes (by LOH) rather than initiates tumor formation, again underlining the differences in responses to CIN between the two tissues. Taken together, CiMKi;Apc Min/+ mice are a useful model to study sporadic and hereditary human CRC.
In conclusion, it is now possible with the CiMKi model to accurately study the interaction between CIN and tumor development in a host of tissues and genetic backgrounds. Because of  Supplementary Fig. 4A, B). Alternating gray and white scale bars 50 μm. Crypts were selected from similar regions (~2/3 from proximal site). Statistics for panels a-d: one-tailed Welch's t-test, comparing each group to CiMKi WT/WT ;Apc Min/+ ;Villin-Cre(ER T2 ), exact p-values are indicated when p < 0.05, and p < 0.0001 is indicated by asterisks (****). Source data for panels a-d are provided as a Source Data file.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-15279-9 ARTICLE tight spatio-temporal control of CIN, CiMKi also enables investigations into the effect of various levels of CIN on cancer cell dissemination, as well as on possible tumor regression. The latter may greatly aid ongoing efforts that examine if exacerbating CIN, for example by MPS1 small molecule inhibitors, has potential as cancer therapy.

Methods
Mice: strains, experiments, and analysis. All animal experiments were approved by Animal Experimental Committee and the Dutch Central Authority for Scientific Procedures on Animals (CCD). All animals were bred and housed under standard conditions (humidity 50-60%, 22-23°C, inverted 12/12 h light/dark cycle, water and standard chow ad libitum) at the animal facility of the Gemeenschappelijk Dieren Laboratorium (GDL), Utrecht, the Netherlands, and the Hubrecht animal facility.
CiMKi Individual clones were analyzed for presence of the CiMKi alleles using standard PCR (Primers used: Forward TCTATGGCTTCTGAGGCGG and Reverse AAGGGACATCAGGGAAGCAA). DNA from targeted ES cells yielded a band of 2.8 kb.
Southern blot was performed according to standard protocols to confirm correct integration of the CiMKi alleles. 5' probe (500 bp) was obtained from genomic DNA from 129/Ola-derived IB10 ES cells, and labeled using a standard Rediprime II Random Prime labeling system (GE healthcare) and radioactive [α −32P] dCTP. Digestion of genomic DNA from ES cells with EcoRV and hybridization with the 5' probe resulted in a 9.5 kb band (when wildtype) or 4.7 kb band (when targeted with CiMKi allele).
Confirmed targeted ES cell clones for both CiMKi mutations were injected into C57BL/6 blastocyst, which were then transplanted into pseudo-pregnant females (standard techniques, performed under the license of the GDL Utrecht). Chimeric mice were bred with C57BL/6 mice to obtain germline transmission. Agouti mice were then backcrossed six times into a C57BL/6 background. Genotypic analysis of offspring was performed using standard PCR and targeted sequencing (Supplementary Table 1). To remove the puro cassette from the original pAC16 construct, CiMKi mice were bred with ACT-Flp mice (C57BL/6 background). Only lines that showed loss of the puro cassette (as confirmed by standard PCR) were used to maintain CiMKi lines.
To induce loxP recombination, MEFs were treated with 4-hydroxy-tamoxifen (4-OHT; 1 µM, Sigma H6278). Mice were injected intraperitoneally with Tamoxifen (1 mg dissolved in corn oil; Sigma, C8267). In CiMKi;Apc Min ;Villin-Cre mice, CiMKi alleles were induced at 12.5 dpc. when the Villin promotor is activated. To confirm recombination, RNA was isolated with a quick RNA kit (Zymo Research). cDNA was prepared using standard procedures, subjected to PCR and subsequently sequenced to determine the presence of T649A or D637A. For primers see Supplementary Table 1.
Mice were sacrificed at four weeks, twelve weeks or eight months of age, and immediately dissected. Small intestine was separated from colon, both were flushed with PBS and pieces of tissue were snap-frozen for later RNA/protein analysis. The organs were stored in formalin until further processing.
Histology and immunohistochemistry. Formalin fixed intestines were cut open longitudinally and stained with 0.25% methylene blue in dH 2 0, and rinsed with PBS. Pictures were taken with ×6.3 magnification using an Olympus SZX stereo microscope to count the number of lesions in the small intestine and colon. After washing with PBS to remove the methylene blue, intestines were rolled into "Swiss rolls" for paraffin embedding.
For identification and assessment of lesions 4-μm sections of paraffinembedded tissue were cut and stained with hematoxylin/eosin (H&E). These slides were scanned (Nanozoomer XR, Hamamatsu) for digital image analysis using NDP.view2 Software from Hamamatsu. Grading of dysplasias was done following the existing guidelines for human intestinal adenomas.
Apoptotic bodies were recognized on H&E stained sections, according to strict morphological criteria such as cell shrinkage with retracted pink to orange cytoplasm, chromatin condensation and nuclear fragmentation and separation of cells by a halo from adjacent enterocytes.
For proliferation measurements slides were incubated with ki67 antibody ( Isolation of MEFs. CiMKi mice were bred with Rosa26-CreER T2 mice and maintained in a stable homozygous CreER T2 background. Pregnant females were sacrificed at 13-17 dpc. by cervical dislocation. Uterine horns were dissected out and placed in tubes containing PBS. Embryos were separated from their placenta and surrounding membranes. Red organs, brains and tail (for genotyping) were removed. Embryos were finely minced using razor blades and the remaining cells/ tissues were suspended in a tube containing 2 ml Trypsin and kept at 37°C for 15 min. Two volumes of media (DMEM supplemented with 10% FBS, nonessential amino acids, glutamin and Pen/Strep) were added and remaining tissues were removed by allowing them to settle down at the bottom of the tube. Supernatant was subjected to centrifugation for 5 min at 1000 rpm, cell pellet was resuspended in medium and plated in 10 cm dishes.
MEFs (mitotic spreads and Western blot). For mitotic spreads, MEFs were treated with STLC (1 μM, Sigma 164739) for 4 h. Mitotic cells (isolated by shakeoff) were treated for 10 min in hypotonic buffer (75 mM KCl), fixed with acetic acid/methanol, dropped onto glass cover slides and stained with DAPI (1 mg ml −1 , Sigma 32670). Images were acquired on a DeltaVision RT system (Applied Precision) with a ×100/1.40NA UPlanSApo objective (Olympus) using SoftWorx software. Chromosomes were counted manually using Image J software.
To establish stable organoid lines expressing H2B-mNeon, organoids were transduced with an H2B-Neon expressing lentivirus (pLV-H2B-Neon-ires-Blasticidin) 67,74 , and selected with blasticidin (InvivoGen; 20 μg ml −1 ). For induction of CiMKi alleles organoids were treated with 1 μM 4-OHT for 56 h. Organoids were seeded and imaged in 8-chamber IBIDI slides using a confocal spinning disk (Nikon/Andor CSU-W1 with Borealis illumination), equipped with atmospheric and temperature control. Organoids were imaged in XYZT-mode (12 to 20 z-sections at 2.5 μm intervals, for 8 to 12 h) at 37°C at 3-min intervals, using a ×30 silicon objective and an additional ×1.5 lens in front of the CCD-camera. 3% 448 nm laser and 50 nm disk pinhole were used. Raw data were converted to videos using an ImageJ macro 67,76 . Fidelity of all observed chromosome segregations was scored manually, guided by the custom-made ImageJ macro for ordered data output.
Single cell karyotype sequencing (scKaryo-seq). Snap-frozen adenoma tissue was stained with 10 μg ml −1 Hoechst 34580 (Sigma-Aldrich) and minced in a petri dish, on ice, using a cross-hatching motion with two scalpels. The minced tissue was kept on ice for 1 h after which it was filtered through 70 μm and 35 μm strainer. Nuclei were sorted in a 384-well plate containing 5 μl of mineral oil (Sigma) in each well and stored at −20°C until further processing for library preparation and sequencing. For library preparation primers consisted of a 24 bp polyT stretch, a 4 bp random molecular barcode (UMI), a cell-specific 8 bp barcode, the 5′ Illumina TruSeq small RNA kit adapter and a T7 promoter. mRNA of each cell was then reverse transcribed, converted to double-stranded cDNA, pooled and in vitro transcribed. Illumina sequencing libraries were prepared with the TruSeq small RNA primers (Illumina). Libraries were sequenced on an Illumina Nextseq 500 with 1 × 75 bp single-end sequencing 44,77 . The fastq files were mapped to GRCH38 using the Burrows-Wheeler Aligner. The mapped data was further analyzed using custom scripts in Python, which parsed for library barcodes, removed reads without a NlaIII sequence and removed PCR-duplicated reads. Copy number analysis was performed performed with the AneuFinder1.6.0 pipeline 78 .
Statistics and data reproducibility. Power analysis predicted the number of animals that had to be used in each group to detect differences with 80% power and 95% confidence. Animals were not randomized, but assigned to the experimental groups based on their genotype. Statistical analyses were done with GraphPad Prism software. Comparisons between CiMKi wildtype and CiMKi mutants were made with a one-tailed Welch's t-test or with an ordinary one-way ANOVA, uncorrected Fisher's LSD test (stated in legends). Data is presented as mean ± SD unless otherwise stated in legends. All images and micrographs shown are representative for experimental groups of at least three individual mice in each experiment.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The scKaryo-seq data have been deposited in the European Nucleotide Archive database under the accession code PRJEB31573 [https://www.ebi.ac.uk/ena/data/view/ PRJEB31573]. All the other data supporting the findings of this study are available within the article and its supplementary information files and from the corresponding author upon reasonable request. Source data for Fig. 1c-