Cloning in Cattle: Nuclear architecture and epigenetic status of chromatin during reprogramming of donor cell nuclei

Chromosomes occupy non-random tissue-specific positions in the interphase nucleus thatcorrelate with the chromosome size and gene richness; they can also have probabilisticallypreferred neighbors. Earlier studies reached opposite conclusions about the degree of thetransmission through mitosis of chromosome arrangement present in a given cell. None ofthese studies, however, analyzed more than one cell cycle and used sufficiently rigorousmeasures for quantitatively assessing similarity of chromosome arrangement. We studied thearrangement of 6-7 chromosomes in clones of up to 32 cells (5 divisions). Nuclear positionsof chromosomes were visualized using FISH and the similarity in chromosome arrangementsbetween cells of a clone was quantified using 3 different measures employing landmarkbased registration. Both in HeLa and normal diploid cells the similarity in chromosomearrangement was lost after only two cell cycles, implying a low level of transmission throughmitosis. In a discussion of factors affecting the degree of the transmission we partiallyreconcile the contradicting results of earlier studies. INTRODUCTIONIn the interphase nucleus, chromosomes occupy distinct regions 1-3 referred to aschromosome territories (CTs). Their structure and spatial arrangement (relative positions in anucleus) are the structural basis of the functional landscape for nuclear processes such astranscription and transcriptional regulation . Recent studies demonstrated the importanceof spatial interactions between numerous chromatin regions from different CTs . Spatialinteractions raise the question of how chromatin regions from distant CTs establish transientcontacts; alternatively, they require a constant cell type specific proximity of such regions. Two types of ordered CT arrangement have been found. First, CTs can occupy more centralor more peripheral nuclear positions depending on gene density and chromosome size(referred to as the global radial order) . Second, cell type specific, probabilisticallypreferred proximities between individual CTs were observed (referred to as neighborhoodorder) . An important question is when these two types of order are established and howthey are maintained in cycling cells. Two different concepts were proposed based on ___________________________________________________________Appendix 167photobleaching experiments of fluorescent protein tagged chromatin in living cells . Bothgroups agree that major changes of chromosome neighborhood occur during prometaphase.According to Gerlich et al (2003) 23 a chromosome specific segregation timing mechanismrestores the CT arrangement present in the mother nucleus in the two daughter nuclei. Acomputational model for inheritance of chromosomal positions through mitosis predicted thatchromosome arrangement should be lost gradually over many cell cycles. By contrast,Walter et al (2003) 24 argued that prometaphase changes of chromosome neighborhoodscause major changes of CT arrangement from one cell cycle to the next. This controversy has not been settled . A strong change in chromosome arrangement aftermitosis was supported by a live cell study of labeled chromosomal loci in sister cells , whileanother group published live cell observations supporting faithful restoration of the CTarrangement after mitosis . Here, we present a novel combined experimental andcomputational approach that allowed a rigorous quantitative analysis of chromosomal patterntransmission over several cell generations. We show that the maternal CT arrangement isentirely lost after only two cell cycles, implying a low level of transmission through mitosis. RESULTSA new experimental and computational approach to study changes in chromosomearrangement.The contradictions between the conclusions of the earlier studies on the degree ofinheritance of chromosome arrangement seem to a great degree result from methodicalreasons. Therefore a step crucial for our work was to design a methodical approach assuringrobust conclusions. The two core elements of our approach are (1) the analysis of severalsuccessive mitoses and (2) direct quantitative estimation of differences between CTarrangements. To study several successive mitoses we analyzed the degree of change in CT arrangementsthat accumulated in a number of successive divisions. In in vivo studies, recovery offluorescence prevents tracking of chromosomal regions differentially visualized by partialhistone-GFP/YFP photobleaching over more than one mitosis. The degree of patterninheritance over several generations had therefore to be extrapolated from observations ofone cell cycle . In contrast to these studies, we traced inheritance of CT arrangementover up to 5 cell cycles by observing descendants of a single mother cell in cell clones.Seeding cells on gridded coverslips and monitoring selected clones by repeated imaging,first, assured that for further studies we used only true clones and, second, allowed us totrace the genealogy of cells within clones (for details see the Material and Methods section).In the cells of studied clones, CTs 4, 7, and 21 were visualized using 3D-FISH and confocalmicroscopy (Fig. 1, Fig. S1). The other core element of our approach was direct quantitative estimation of differences inCT arrangement between cells. While all earlier studies were based either on qualitativecomparisons or on indirect measures (distance and angle measurements, signal intensitydistributions) we developed a novel approach for direct quantification of difference betweenCT arrangements using landmark based registration . A landmark is a point that may beidentified in each cell. We used as landmarks the estimated geometrical centers of individualCTs and applied three variants of registration that allowed the CT arrangement differentdegrees of plasticity (Fig. 2): rigid registration is sensitive to any changes in size or shape,similarity registration corrects for size only, while elastic bending energy based registration 28corrects for all kinds of global and local continuous deformations. Each variant of registrationprovides its own measure of dissimilarity between CT arrangements in two nuclei (Fig. 2, fordetails see the Material and Methods section). Since it remains unknown which changes inchromosome arrangement are reversible and which are not, using three measures, verydifferent in this respect, assured robustness of our final conclusions.CT arrangement patterns quickly diversify in growing clones. ___________________________________________________________Appendix 168Using the approach described above, we assessed the dissimilarity of chromosomearrangement accumulated in HeLa cells after 1 to 5 divisions of a mother cell, i.e. in 2 toapprox. 32 cell clones. To exclude the influence of chromatin movements associated withmitosis, only clones in which all cells were in interphase were used for quantitative analysis.Such clones with 2, 4, 8, and even 16-cell clones could be found easily. We also analyzedseveral clones with 27-38 cells, again chosen because they satisfied the above condition.The dissimilarity of CT arrangement was calculated for each possible pair of cells within agiven clone. The overall dissimilarity of CT arrangement for this clone was defined as themedian of the dissimilarities for cell pairs (for details see Material and Methods). The differences in CT arrangement between cells of a clone accumulated with size of clonesand reached the level of unrelated cells in 4 or 8 cell clones, depending on the considereddissimilarity measure (Fig. 3a). The dissimilarity in CT arrangement for 8-cell clones andbeyond was very much comparable (Fig. S2). For clones with 2, 4, and more than 8 (8+)cells (49, 32, and 37 clones, respectively), dependence of dissimilarity on the clone size washighly significant (Kruskal-Wallis test, P<0.001 for all measures, followed by Dunn’s test forpairwise comparisons at the cutoff confidence level of 0.05). Importantly, sisters (i.e., cells in2-cell clones) were significantly more similar, than cells in the 4 cell clones. The differencebetween the 4-cell and 8+ clones was not statistically significant except for the bendingenergy based dissimilarity measure, which probably depended on the measure itself. Notethat in 4 cell clones some cells are sisters (share the mother) while some are cousins andshare only the grandmother (Fig. 1). The proportion of cousins to sister pairs is 2:1, andtherefore the similarity for cousin cells may be estimated based on the values for 2 and 4 cellclones. The estimated dissimilarity for cousin cells reached the level of unrelated cells (Fig.3b) for all measures. A further parameter characterizing clones of a given size is thedistribution of dissimilarity values for pairs of cells within clones of this age. Thesedistributions showed a clear difference between 2-, 4and 8+ cell clones, whereas the 8+clones was indistinguishable from unrelated cells (Fig. 3d). Hence, our data suggested strongly that the dissimilarity level of unrelated cells is reachedby cousin cells. To confirm this conclusion, we analyzed 28 4-cell clones for which thegenealogy of cells was traced. Indeed, cousin cells did not reveal any higher degree ofsimilarity in chromosome positioning than genealogically unrelated cells (Fig. 3c), whereassister cells were significantly more similar than cousin cells (Wilcoxon Signed Rank test,P<0.001). A computer simulation based on chromosome arrangements in unrelated cells (one fromeach clone: 196 in total) showed that dissimilarities between sister cells corresponded to shiftof all CTs in random direction at 2.4 3.5 μm (depending on the registration). Dissimilaritiesbetween cousins corresponded to a shift of 3.6 5.4 μm (data not shown). Though thedifference between shifts characteristic of sisters and cousins does not look dramatic, animportant border is crossed. In line with earlier observations , in unrelated cells thestudied chromosomes follow size-dependent radial distribution, but move freely andindependently along the respective orbits. They are situated at the average distance of ca.4.4 μm from the nuclear center (3.3, 4.5 and 5.1 μm for HSA 21, 4 and 7, respectively; Fig.S3). Being shifted by 4-5 μm, homologous CTs can partially “swap” their neighbors, and ourmodel shows that this should indeed happens very often. Taken together, our simulation datasuggest that sisters shared the same chromosome arrangement in 50-70% of sister pairs,while cousins nearly always had qualitatively different arrangements. ___________________________________________________________Appendix 169Pattern transmission in normal diploid cells is similar to HeLa cells.To exclude the possibility that the quick loss of similarity in CT arrangement in HeLa cellswas related to the malignant nature of these cells, we performed the same analysis fornormal diploid cells. Several normal cell types tested do not form compact clones whengrown in vitro. Human mammary epithelial cells (HMEC) grow as compact clones until 4-cellstage; thereafter the clones start to move as a whole and fuse with other clones (Fig. S4). Asin the case of HeLa cells, the similarity in 2 cell HMEC (sister cells) clones was significantlyhigher than in 4-cell clones (Fig. 3e; n=16 and n=24 clones, respectively; Mann-WhitneyRank Sum test, P< 0.001 for all measures). The estimated dissimilarity in CT arrangementfor cousin cells was again at the same level as that between unrelated cells (Fig. 3f). Inaddition, we analyzed human normal diploid fibroblasts which have been intensively used inchromosome arrangement studies . Fibroblasts move up to a few hundreds ofmicrometers a day hampering the identification of sister cells after cell division. To comparechromosome arrangement in sister nuclei we generated binucleated fibroblasts by treatmentof cells with Cytochalasin B, an inhibitor of cytoplasmic division (Fig. S5). In accordance withthe above findings for HeLa and HMEC, the dissimilarity in CT arrangement for sister nucleiwas clearly below the level of unrelated cells (Fig. 3g). Hence, normal diploid cells did notshow a higher degree of inheritance of interphase CT arrangement than HeLa. Interphase chromatin arrangement is modified by changes in nuclear shape, butremains stable otherwise.The above findings indicate clearly that maternal chromosomal order was entirely lost afteronly two cell cycles. We went on to study, at which stage of the cell cycle the observed lossof chromosomal order took place (c.f. Walter et al. (2003); Thomson et al (2004) ). Tostudy chromosomal rearrangements in mid interphase we labeled chromatin in HeLa cells byreplication labeling and studied in vivo 38 nuclei. In each nucleus, we traced 10-18 focidistributed throughout the whole volume of the nucleus for 5-12 hours during interphaseusing in vivo confocal microscopy followed by 3D reconstruction (Fig. S6). Dissimilaritybetween the initial and final arrangement of replication foci was then estimated for allpossible combinations of 7 foci in the same way as it was done for 7 chromosomes of HeLacells. Median of these values was used as the measure of dissimilarity between initial andfinal arrangements. We tested several parameters that might affect rearrangement of foci: accumulated lineardisplacement of the nucleus along the substrate surface, change of the area of the nuclearprojection to the substrate plane, change in the roundness of this projection, as well as therate of change of these parameters per hour. In accordance with our own preliminaryfindings, the degree of rearrangement correlated with the change in the nuclear shape.Other parameters studied showed no effect on chromosome arrangement. Interestingly,rearrangement correlated equally well with mean deformation over all observation period andwith maximal deformation observed within a single one-hour window over the same period(Table S1). This finding suggests that chromatin rearrangement was caused by quick strongsingle deformations, rather than by cumulative effect of small ones. Indeed, the mostpronounced rearrangements were observed when two cells collided causing a high degree ofnuclear deformation (Fig. 4). In 4 nuclei interphase rearrangement of replication foci resultedin dissimilarity comparable to the dissimilarity of CT positions in sister cells, and two of thesenuclei came in a temporary contact with another nucleus. In accordance with the reported low mobility of chromatin in interphase , only relativelysmall displacements of foci took place in the majority of cells. Median dissimilarityaccumulated over interphase was 3 to 6 times smaller (depending on the registration), thanthe respective values for sister cells. The median displacement for replication foci over midinterphase was 0.97 μm, that is, 3 and 4 times less than in sister cells and unrelated cells,respectively. As revealed by our computer simulations, chromosomal patterns are mostlyretained in interphase with this low dissimilarity level. We conclude that (1) overall changes innuclear shape were the major cause of chromatin rearrangement in interphase and (2) that ___________________________________________________________Appendix 170pronounced chromosomal rearrangements only rarely occurred in interphase in ourexperiment. DISCUSSIONTo analyze the degree of preservation of CT arrangements during clonal growth through upto 5 cell divisions, we quantified the accumulating differences using 3 different measures. Wefound for both HeLa and normal diploid cells that the similarity in CT arrangements betweencells of a clone was lost after two cell cycles. Major repositioning of chromatin regions duringinterphase occurred only in a small proportion of nuclei and was caused by strong changesin the shape of the nucleus. Below we discuss which phases of mitosis affect the degree ofpattern transmission and compare our results with those of earlier publications. Changes of chromosome neighborhood arrangements during prometaphasecongression of chromosomes.Gerlich et al (2003) formally described formation of the metaphase chromosomearrangement as projection of their interphase positions onto the plane defined by thecorresponding metaphase plate. Although this assumption simplifies chromosomemovements occurring in living cells – e.g., due to the asymmetry of spindle formation 37 andthe effect of delayed biorientation 38 – this model is a useful tool for the analysis of changesin chromosomal neighborhood arrangements induced by congression. It is howeverimportant to take in account that spindle formation during early prometaphase 39 can beinitiated in any direction . Later the spindle rotates and achieves its final orientation onlywith the onset of the metaphase . After completion of kinetochore attachment to thespindle, the spindle and the condensed chromosomes rotate together. Since the primaryorientation of the spindle determines the plane onto which chromosomes will congress, italso determines which chromosomes will become neighbors in the metaphase plate. For most of the possible primary orientations of the spindle, widely separated chromosomes(e.g., at the opposite poles of the forming spindle) congress to the same region of themetaphase plate (Fig. 5). An accurate transmission of the neighborhood pattern of prophasechromosomes to the ensuing metaphase plate is only possible if (1) nuclei are flat and (2) thespindle is initially oriented perpendicularly to the substratum (Fig. 5d). Such a primary spindleorientation is not common, e.g., it was only observed in 15% of HeLa prometaphase cells 37(and our own observations). Even if sister cells have an identical CT arrangement, spindlesof the next division are formed independently and will transform this arrangement differently(Fig. 5). For two HeLa cell cousin, the probability to have a similar CT neighborhoodarrangement is only 0.023 (0,15). This explains convincingly why the similarity inchromosome arrangement is lost already by cousin cells. The CT arrangement of the foundercell of a clone should therefore be lost after two mitoses. This mechanism also conforms wellto the observations by Walter and collaborators 24 (Fig. S7). It was suggested 23 thatchromosome-specific difference in timing of chromatid separation restores the maternalarrangement in daughter cell nuclei nearly completely. This restoration effect was notobserved in our experiments with both HeLa and normal diploid cells. Preservation of chromosome order during cell cycle progression from the metaphaseplate to the formation of sister nuclei.The average dissimilarity between sisters remained statistically significantly below the levelof unrelated cells and cousins. Yet, only in about 15% of sister nuclei CT neighborhoodarrangements looked clearly similar at face value, while in 12-23% of sister cell pairs(depending on the registration method) the dissimilarity of neighborhood CT arrangementsexceeded the respective median value for cousin cells. Previous studies suggested thatdissimilarity between sisters mainly depends on G1 . Metaphase plates and earlyanaphase plates are typically oriented vertically to the substratum, while at late anaphase-telophase plates re-align parallel to the surface. It was suggested 31,42 that anaphase platesrotate without major deformation, thus transmitting the chromosome arrangement and size-dependent radial distribution from the metaphase to the interphase nuclei. On the other ___________________________________________________________Appendix 171hand, in line with our live cell observations, the similarity of CT arrangements between sisternuclei was lost in a sarcoma cell line because these nuclei attained different shapes duringearly G1 . In contrast to sparse cultures we used for our experiments, in sub-confluentHeLa cultures we observed that after division cells were forced to move between theneighboring cells to establish contact with the substratum. These movements wereaccompanied by diverse major deformations of nuclei which likely strongly reduced thesimilarity in chromosome arrangement between sister cells. Concluding remark.In summary, our and other results imply that major changes of CT proximity patterns occurduring prometaphase and early G1. Remarkably, changes of chromosome arrangementsduring prometaphase were first described one hundred years ago by Theodor Boveri in astudy of the cleavage in the nematode Parascaris equorum (for review see ). In sometissues, the final orientation of the spindle is inherently related to differentiation 44,45 anddetermines the fate of daughter cells. The primary orientation of the spindle, however, hasnot been studied. Our data suggest that transmission of chromosome arrangement throughmitosis cannot serve as effective epigenetic mechanism. When chromosomes take specificinterphase positions depending on gene content 15,33 or have preferred neighbors ,special mechanisms should exist to establish these specific features in each cell individuallyand independently from the particular arrangement in the mother cell. ACKNOWLEDGEMENTSThe authors thank J.Fieres for C++ codes for the first version of Bending Energy calculationprogram used in this project; J.Walter (Till Photonics, Munich) for helpful discussions;C.Fauth (Technical University of Munich) for MFISH analysis of the used HeLa cells,R.Hessing (University of Munich) for help with illustrations. RE received support by the EU(3DHumanGenome) and the HFSP. RE also acknowledges support on multi-dimensionalimaging from Leica Microsystems CMS GmbH, Mannheim, Germany. TC was supported byDFG grant Cr59/20 and BMBF NGFN II-EP grant 0213377A. METHODSCell lines and growth of clones. HeLa cells 46 (kindly donated by K.Sullivan, The ScrippsResearch Institute), human mammary epithelial cells (HMEC, kindly donated by T. D. Tilsty,University of California), and normal human diploid fibroblasts were routinely cultured inRPMI (HeLa) or DMEM (HMEC, fibroblasts) supplemented with 10% fetal calf serum. Cellswere seeded sparsely on gridded coverslips (Bellco, USA) and after 5 h, phase contrastimages of appropriate areas with single cells were recorded; afterwards cells were observedand imaged every 24 h to monitor the growth of clones. HeLa cells, that have a very lowmobility, were allowed to grow for up to 6 days to pass up to 5 divisions (Fig. S1). We usedfor analysis 118 clones (2 cell: 49; 4 cell: 32; 8 cell: 19; 16 cell:11, one clone of 20 cells, 5clones of 27-32 cells, and one 38 cell clone) and 78 single cells. Clones of HMEC cells could be followed reliably only up to 4 cell clones: single epithelialcells have a low mobility, however, groups of 4 and more cells moves along the substrateand fuse with one another (Fig. S4). We used for analysis 40 HMEC clones (2 cell: 16; 4 cell:24). Fibroblasts do not form clones and move along substrate for hundreds microns per day.To observe sister nuclei, binucleated cells were induced by incubation with cytochalasin B (5μg/ml) for 9 h; cells were transferred to fresh medium for 5 h prior fixation (Fig. S5); 22binucleated cells were used for analysis. To grow 4-cell clones with completely traced genealogy, HeLa cells were seeded asdescribed above, allowed to attach, and images of a number of appropriate view fields wererecorded. Then cells were allowed to grow in live cell chamber as described . In oneexperiment cells were placed in the live cell chamber immediately after imaging theirpositions and clones were monitored starting from single cells. In another experiment cellswere grown for 16 h in Petri dishes to allow them to pass the 1st mitosis and then placed in ___________________________________________________________Appendix 172the live cell chamber. After placing coverslips to the microscope stage, the initially selectedand recorded view fields were found and imaged every 20 minutes for a period of 18 h thatallowed cells to pass the second division. 28 clones with traced genealogy were used foranalysis. In vivo observations on interphase rearrangement of replication foci. The HeLa cell lineused for clone study stably expresses H2B-GFP was also used for in vivo observations. Cellswere scratch-replication-labeled with Cy3-dUTP , cultured for about 2 cell cycles to allowlabeled chromatids to segregate, and then observed in live cell chamber under confocalmicroscope for a period of 5-12 hours. In each of the 8 experiments carried out, a field ofview was selected at low magnification. Then 3 smaller view fields containing several nucleiwith well-separated easily identifiable replication foci were chosen within this area and imagestacks were collected with high magnification every hour, so that the chamber itself did notmove during observations . These experimental conditions allowed cells to cycle normally:viability of cells was confirmed by the fact that in more than 60% of monitored regions ofinterest one or more cells underwent mitosis during observations time and cell divisionsremained common at the end of the observations. FISH and microscopy. Karyotype of the used HeLa cell line was examined by M-FISH 48and three chromosomes not involved in any re-arrangements were chosen: diploid HSA 4and 21 and triploid HSA 7 (Fig. S8). Tested cells were frozen and all experiments were thencarried out with subcultures grown from this stock to avoid any further changes in thekaryotype. Cocktail of chromosome paint probes labeled with different haptens (biotin,digoxigenin, and DNP for HSA 4, 7, and 21, respectively) was used for hybridization on allthree cell types. Cells were fixed and 3D-FISH was performed as previously described .Briefly, cells were fixed in 4% paraformaldehyde in 1x PBS for 10min, permeabilized with0.5% Triton-X100 for 20min, incubated in 20% glycerol for at least 1 h, repeatedly frozen inliquid nitrogen, and finally incubated in 0.1N HCl for 5 min. Prior hybridization, coverslips withfixed and pretreated cells were stored in 50% formamid/SSC at 4°C for about one week.Labeled paint probes were dissolved in a hybridization mix (50% formamide, 10%dextransulfate, 1x SSC) together with Cot 1 DNA, applied to cells; DNA-probes and cellularDNA were denatured simultaneously on a hot block at 75°C for 3 min. Hybridization wasperformed for 2-3 days at 37°C in humid boxes. Post-hybridization washes were performedwith 2×SSC at 37°C and 0.1×SSC at 60°C. Digoxygenin was detected with mouse-anti-dig(Sigma) and Cy3-conjugated sheep-anti-mouse antibodies (Jackson ImmunoResearchLaboratories); biotin was detected with of Streptavidin-Cy5 (Vector Laboratories); DNP wasdetected with rabbit-anti-DNP (Sigma) and goat-anti-rabbit conjugated to Alexa488(Invitrogen). Nuclei were counterstained with DAPI (Sigma) and mounted in Vectashieldantifade (Vector Laboratories). After FISH, clones recorded during growth were re-foundbased on their position on the gridded coverslips and series of light optical sections throughwhole nuclei with voxel size of 100x100x300 nm (X,Y,Z) were collected using a Leica TCSSP1 confocal microscope. Data evaluation. Chromosome territories (CTs) and replication foci were segmented usingAmira 2.3 TGS software. Tracing the positions of replication foci was carried out manually:3D reconstructions corresponding to 3-4 successive time points were observed together,which allowed step-wise reconstruction of the trajectories (Fig. S6) For further analysis,surfaces obtained from Amira reconstructions were exported as inventor (.iv) files, and thecoordinates of the centers of volume of CTs were calculated using a program (GAQ)developed in-house. Dissimilarity between CT configurations was calculated based ongeometrical centers of signals. Rigid registration was implemented according to Dryden andMardia (1998); similarity registration was performed using the same program after sizenormalization; in both cases reflection was allowed. Bending energy registration 28 was doneusing the program described earlier . For size independence, bending energy was usedafter normalization of the size of point configurations. The smaller of the two reciprocalbending energy values was used as dissimilarity measure to exclude the effect of rare very ___________________________________________________________Appendix 173high values. Since homologous chromosomes could not be distinguished in our data, for all 3registration variants all possible pairwise associations of homologous chromosomes from 2cells were tested and the minimal dissimilarity value found was used as respectivedissimilarity measure. Calculation of intraand cross-clone dissimilarities was performedusing a program (Nuclear Nightingale) developed in-house. Our data evaluation software isavailable on request. Statistical analysis. The same kind of statistical data evaluation was performed for all 3dissimilarity measures. Because of the strong asymmetry of the distribution of dissimilarity inthe case of bending energy based measure, each clone was characterized by clone median,i.e. the median dissimilarity for all possible pairs of cells in this clone. These medians wereused to compare clones of different size (stages). The distributions of medians remainedasymmetric for clones of smaller size. Therefore all comparisons were made using non-parametric tests (as implemented in SigmaStat software, Systat Software incorporated).Kruskal-Wallis one-way ANOVA on Ranks followed by Dunn’s test for pairwise comparisonsat the cutoff confidence level of 0.05 was used to compare stages; Wilcoxon Signed Ranktest was applied to compare sisters and cousins in clones with traced genealogy. Calculationof dissimilarity for unrelated cells was performed using cross-clone dissimilarity values andbootstrap on clones using the Nuclear Nightingale program (for details see Fig. S9). Thedistributions of individual dissimilarity values (Fig. 3d) were processed (“smoothed”) usingnaive estimator with bin size 2 (rigid and bending energy based registration) and 0.2(similarity registration), respectively. Since estimation of cousin dissimilarities from data on 2and 4 cells clones is only possible with means, all values shown in Figure 3b,f wererecalculated for this purpose based on clone and stage means, as described above for cloneand stage medians.