Heritable transcriptional defects from aberrations of nuclear architecture

Cell culture and cell line construction

Cells were cultured at 37 °C in 5% CO2 atmosphere with 100% humidity. Telomerase-immortalized RPE-1 retinal pigment epithelium cells (CRL-4000, American Type Culture Collection), U2OS osteosarcoma cells (HTB-96, American Type Culture Collection) and derivative cell lines were grown in DMEM/F12 (1:1) medium without phenol red (Gibco) supplemented with 10% FBS, 100 IU ml−1 penicillin and 100 μg ml−1 streptomycin. For cell lines with doxycycline-inducible constructs, tetracycline-free FBS (X&Y Cell Culture) was used.

Stable cells lines H2B–eGFP and TDRFP–NLS RPE-1, mRFP–H2B and eGFP–BAF RPE-1, mRFP–H2B RPE-1 and TDRFP–NLS U2OS were generated by transduction of RPE-1 or U2OS cells using lentivirus or retrovirus vectors carrying the genes of interest as previously described26. RPE-1 cells with transient expression of a dominant-negative variant of telomeric repeat-binding factor 2 (TRF2-DN)43 were treated as previously described11. RPE-1 clones derived from single cells with CRISPR–Cas9-mediated telomere loss on chromosome 4 (chromosome 4 bridge) and their derived clones were generated in a previous study11. Control parental RPE-1 subclones were generated by FACS and expansion in 96-well plates. The HDAC inhibitor vorinostat (SAHA, Sigma-Aldrich, SML0061) was used at 0.5 μM concentration, as described in the Extended Data Fig. 5d.

Generation of cells expressing SNAP-MDC1

The RPE-1 GFP-H2B RFP–NLS SNAP–MDC1 cell line (Fig. 3 and Extended Data Fig. 9) was generated by lentiviral transduction of the SNAP–MDC1-bearing lentiviral vector. This vector was generated by cloning a synthesized SNAPf fragment (sequence from pBS-TRE-SNAPf-WPRE; plasmid 104106, Addgene) with AgeI and BstBI restriction sites into the pLenti CMV/TO GFP-MDC1 (779-2) (plasmid 26285, Addgene, gift from E. Campeau; Genewiz) backbone, substituting SNAPf with eGFP at the N terminus of MDC1. Stably transduced cells were selected by FACS around 10 days after transduction for SNAP–MDC1 expression.

Generation of the modified U2OS 2-6-3 transcription system

Our modified U2OS 2-6-3 cells contain GFP–H2B, Cuo-LacI–SNAP and MS2–Halo (Figs. 2 and 4 and Extended Data Fig. 8). These cells were generated from the original U2OS 2-6-3 cells31 (gift from D. Spector). In brief, the 2-6-3 transgene consists of 256 tandem copies of the lac operator, which enables visualization of the transgene genomic locus, 96 tetracycline response elements (TREs) to control the reporter transgene and 24 MS2 translational operators (MS2 repeats) for the visualization of the reporter nascent transcript31. The 2-6-3 transgene was introduced into a single euchromatic locus on chromosome 1p36 (ref. 31). We modified the system as follows. We introduced a lentivirus with the coding sequence of LacI fused to SNAP, under the control of a cumate-inducible promoter. Independent control of LacI–SNAP and the MS2 reporter enabled the identification of the reporter in micronuclei in generation 1, followed by assessment of MS2-marked transcription in generation 2. We also stably introduced genes expressing LacI–SNAP, rtTA and MS2 coat protein (MCP) used for visualizing the MS2 aptamers.

Specifically, U2OS 2-6-3 cells were transduced with pLenti CMV rtTA3 Blast (w756-1, plasmid 26429, Addgene; gift from E. Campeau) for the expression of rtTA, a lentiviral vector, phage ubc nls ha 2×mcp HALO, for the expression of MCP–Halo (plasmid 64540 Addgene; gift from J. Chao) and lenti Cuo-LacI-SNAP, for the expression of LacI–SNAP. Our LacI–SNAP expression vector, CuO-LacI-NLS–SNAPf, contains the coding sequence for the SNAPf-Tag (sequence from pBS-TRE-SNAPf-WPRE; plasmid 104106, Addgene) followed in-frame with the coding sequence for LacI-NLS (sequence taken from Cherry-LacRep; plasmid 18985, Addgene). This sequence was subcloned into pCDH-EF1-CymR-T2A-Puro (QM200VA-1, System Biosciences SBI) using NheI and BstBI restriction sites. The final modified U2OS 2-6-3 cell line was obtained by selection for hygromycin resistance conferred by the 2-6-3 transgene and for blasticidin resistance conferred by the rtTA expression construct, followed by FACS to identify MCP–Halo and LacI–SNAP expression. Note that binding of LacI–SNAP to LacO was transiently inhibited by adding 1 mM isopropyl β-d-1-thiogalactopyranoside before FACS. Transient inhibition was done to avoid genetic instability from LacI binding to the Lac operators, which are a barrier to replication fork progression. Full maps of the constructs used are available upon request. All cell lines used in this study were monitored for mycoplasma contamination.

Generation of RPE-1 megaDam cells

The RPE-1 3×Cherry Dam AID Smash cell line (RPE-1 megaDam) (Fig. 4 and Extended Data Fig. 10) was generated by lentiviral transduction of the megaDam vector into a RPE-1 cell line that has a doxycycline-inducible transgene expressing the E3 ligase, OsTIR1, integrated at the ROSA26 locus44. The megaDam vector (Fig. 4a) was generated by synthesizing (Genewiz) a sequence containing three copies of mCherry (based on the sequence from pHAGE-EFS-N22p-3XRFPnls; plasmid 75387, Addgene) and the sequences encoding the mAID and SMASh degrons (from ref. 44). The Dam coding sequence was taken from TS52_pT_damonly (van Steensel lab). The sequence encoding the Dam–mCherry double-degron fusion was cloned and introduced into the lentiviral vector pCW57.1 (plasmid 41393, Addgene; gift from D. Root, Genewiz). A stably expressed RPE-1 megaDam cell line was obtained by puromycin selection.

Cell cycle synchronization and methods to generate micronuclei or bridges

To synchronize cells and to generate micronuclei, most experiments in this study used a previously described nocodazole block and release protocol8,11,26 unless otherwise stated. In brief, approximately 15 h after TP53 siRNA (Horizon Discovery) treatment, cells were treated with 100 ng ml–1 nocodazole for 6 h followed by a mitotic shake-off procedure. Alternatively (for Figs. 1f and  4a–c and Extended Data Figs. 4f and 10), cells were synchronized at the G2/M border with a treatment of 9 μM RO-3306 (MilliporeSigma), a CDK1 inhibitor, for 18 h. G2/M-arrested cells were next released into mitosis by washing five to seven times with medium, followed by addition of 1 μM NMS-P715 (MilliporeSigma) to impair chromosome segregation through inhibition of the MPS1 kinase45.

For the analysis of cells after bridge formation (Extended Data Fig. 12a), RPE-1 TRF2-DN cells were treated as previously described11. In brief, the cells were incubated in 0.1 μg ml–1 doxycycline (Millipore Sigma) for about 20 h to generate chromosome bridges. Bridges begin to form during cell divisions that occur at least 8 h after the washout of doxycycline. At 20 h after doxycycline washout, cells were synchronized in G2 using 9 μM RO-3306 (Millipore Sigma) for another 18 h. Finally, the cells were fixed and analysed 6 h after release from RO-3306, at the next interphase after bridge resolution and cell division. Data in Fig. 5 and Extended Data Figs. 11 and 12b–d were generated using RPE-1 clones derived in a previous study11.

Detection of nascent transcripts marked with 5-ethynyl uridine

To detect nascent transcripts, cells were incubated for 30 min with 1 mM 5-ethynyl uridine, which was added approximately 23 h after mitotic shake-off from nocodazole release. Incorporation of 5-ethynyl uridine was detected using a Click-iT RNA Alexa Fluor 488 imaging kit according to the manufacturer’s instructions (ThermoFisher Scientific).

Cas9 RNP transfection

The method for targeting a specific chromosome arm to a micronucleus in RPE-1 cells, complete characterization of the editing efficiency of the sgRNA used in this study and the frequency of generation of micronuclei harbouring the targeted chromosome have been previously described in detail25. In brief, a Trueguide Synthetic gRNA system (ThermoFisher Scientific) was used to generate the sgRNA for chromosome 5q with the sequence 5′-G*U*U*GGCCUCCCAAACCACUA-3′ (asterisks indicate modified 2′-O-methyl bases with phosphorothioate linkages). RPE-1 GFP–H2B RFP–NLS cells were synchronized in G0 by serum starvation for 23 h and were then transfected with the Cas9–gRNA RNP complexes 22 h after release. Cell synchronization and transfection were performed on cells seeded onto MembraneRing 35 rings (415190-9142-000, Carl Zeiss), which enabled cell isolation by laser capture (see below). Live-cell imaging started 3–5 h after transfection, and cells with micronuclei and their siblings were followed until late G2 phase before cell capture for single-cell RNA-seq (scRNA-seq).

Live-cell imaging

For the majority of the live-cell imaging experiments, images were collected on Nikon (Ti-E) or (Ti2) wide-field inverted microscopes equipped with Perfect Focus, an environmental enclosure to maintain cell culture conditions (37 °C and humidified 5% CO2), a ×20/0.75 NA Plan Apochromat Lambda objective (Nikon) or a ×40/0.95 NA Plan Apochromat Lambda objective, and a Zyla 4.2 sCMOS camera (Andor). At each time point, three 2-μm-spaced Z-focal plane image stacks were acquired. Live imaging by confocal microscopy (for the experiments with the adapted U2OS 2-6-3 and MDC1-expressing cells, see below) was performed at 15 min time intervals on a Ti2 inverted microscope fitted with a CSU-W1 spinning disk confocal head (Nikon). At each time point, three 2-μm-spaced Z-focal plane image stacks were acquired using a ×40/0.95 NA Plan Apochromat Lambda objective. The microscopes were controlled using Metamorph (v.7.10.2.240; Molecular Devices) or NIS Elements (v.4.30 AR or newer versions; Nikon Instruments).

Live-cell imaging for Look-Seq2 experiments

Imaging for Look-Seq2 experiments (Figs. 1 and 2 and Extended Data Figs. 13, 6 and 7) was performed using a wide-field inverted microscope and a ×20 objective (see above). Note that at this imaging resolution, we can confidently detect the presence of micronuclei and micronuclear rupture. RFP–NLS or GFP–BAF were used for the assessment of NE integrity of the generation 1 samples as previously described26.

We acknowledge, however, that with ×20 wide-field imaging, some events such as micronuclei of small size and borderline cases of NE rupture, fine bridges and rare cases of micronuclei-like structures connected to the PN may not be resolved.

Live-cell imaging of cells containing the U2OS 2-6-3 transcription reporter

For live-cell imaging of our modified U2OS 2-6-3 nascent transcript reporter cells (Figs. 2 and 4 and Extended Data Fig. 8), cells were seeded on 35-mm ibiTreat Grid-500 dishes (Ibidi) with a gridded imaging surface after mitotic shake-off. SNAP-tagged and Halo-tagged proteins were labelled using 250 nM JF549-cpSNAP-tag and 50 nM JF646-HaloTag ligands (Janelia Materials) for 15 min before the start of imaging. To induce LacI–SNAP expression, cumate (30 µg ml–1; System Biosciences) was added in the medium immediately after the mitotic shake-off step and washed out before imaging. Doxycycline (1 µg ml–1; MilliporeSigma) was used to induce expression of the MS2 transcription reporter approximately 2 h before the start of imaging and was maintained in the medium for the remainder of the experiment (Fig. 4d). Confocal imaging started 16–19 h after mitotic shake-off and was performed as described above for about 24 h or until most of the cells of interest had divided and could be imaged in the generation 2 cycle.

Live-cell imaging of MDC1-expressing cells

For the live-cell imaging experiments tracking damaged MN chromosomes marked by MDC1 (RPE-1 GFP–H2B RFP–NLS SNAP–MDC1 cells; Fig. 3 and Extended Data Fig. 9), cells were seeded in 35-mm ibiTreat dishes (ibidi) after mitotic shake-off and SNAP-tagged proteins were labelled using 250 nM JF646-SNAP-tag ligand (Janelia Materials) for 15 min before the start of imaging. Imaging was started 16–19 h after mitotic shake-off and images were collected at 15 min intervals with a ×40 objective and 2 × 2 image stitching. An exception was the experiments with high temporal resolution, for which images were collected at 6 min intervals (Extended Data Fig. 9c and Supplementary Video 4) and at 3 min intervals (Supplementary Videos 5 and 6) without image stitching.

Capture of single-cells for Look-Seq2

The isolation of live cells for scRNA-seq was performed in two ways. Our initial experiments were performed in an analogous manner to our early Look-Seq procedure8 for single-cell whole-genome sequencing (which was subsequently refined in ref. 11). For these experiments, cell isolation was accomplished by trypsinization followed by FACS of single cells into 384-well μClear imaging plates (Greiner). Micronucleated cells were identified by imaging, and then after the division of these cells, daughter cells were isolated for scRNA-seq by trypsinization and replating into 384-well μClear plates after serial dilution. This procedure was used for a subset of the generation 2 experiments to assess the effect of MN chromosome re-incorporation into normal daughter nuclei (10 out of 127 of the total generation 2 samples; Supplementary Table 1). We subsequently developed the laser capture microdissection (LCM) procedure (Extended Data Fig. 1b, described below) and used this method to collect MN cells and MN sisters (generation 1) and MN daughters and MN nieces (the majority of generation 2 samples) for data presented in Figs. 1 and 2, Extended Data Figs. 13, 6 and 7 and Supplementary Tables 1 and 2. The main advantages of our modified system over previous LCM methods23 are as follows: (1) minimized cell stress (because the cells are kept in medium in the microchamber setup that prevents the culture to dry out) throughout capturing; (2) higher throughput; and (3) an enhanced ability to capture cell families (again because the cells are maintained in medium throughout capture of multiple cells).

Live-cell imaging for Look-Seq2 experiments

Cells were treated as described in ‘Cell cycle synchronization and methods to generate micronuclei or bridges’ for the induction of micronuclei. After mitotic shake-off, cells were handled as described below.

For experiments using the older cell capture method8,11, cells were plated into 384-well μClear imaging plates and imaged using wide-field fluorescence microscopy at time intervals of 15 min for up to 48 h or until the majority of cells had progressed through mitosis. For LCM capture experiments, cells were instead plated on MembraneRing 35 rings (hereafter membrane rings; 415190-9142-000, Carl Zeiss) (see details in ‘Development of the modified LCM capture method’ below) and imaging was performed at 15 min intervals as described above, with the difference that image stitching (2 × 2) was used to track mobile cells across different fields of view.

Development of the modified LCM capture method

We adapted a previously developed LCM system (Palm Microbeam, Carl Zeiss) and re-designed the capturing and imaging setup as described below and in Extended Data Fig. 1b. A custom-designed aluminium adapter was constructed (SeqTech) to enable imaging of cells plated on the membrane rings. We also custom-designed and 3D printed an adapter using VeroWhite material (opaque white Polyjet resin, SeqTech) to allow placement of the membrane rings in a flipped orientation on the Palm Microbeam LCM microscope with a DishHolder 50 CC (415101-2000-841, Carl Zeiss; Extended Data Fig. 1b). This enabled capturing of cells in a multi-well capture plate that could be placed close to the cells, which helped increase the capturing speed, efficiency and therefore the throughput of the method. The designs of the adapters are available upon request. The Look-Seq2 method is described in detail in a provisional patent46. A hydrophobic barrier was applied at the periphery of the surface of the membrane rings using an ImmEdge PaP Pen (Vector Laboratories) to prevent evaporation of the medium. Next, the cells were plated on the membrane rings.

At the time of cell capturing, the cells were supplemented with medium containing HEPES buffer. Next, the membrane rings were flipped upside down and positioned on the custom-made adapter after the application of a glass 20 mm glass coverslip (Neuvitro), which we refer to as the microchamber (Extended Data Fig. 1b). The cells in the microchamber were transferred to a Palm Zeiss LCM microscope on the custom-made adapter. Cells of interest were identified by extrapolation of the coordinates on the imaging microscope to the LCM microscope using a custom MatLab script and reference marks that were applied to the membrane rings. Snapshots of all imaging channels of the cells of interest were taken immediately before LCM to ensure accurate assessment of the micronuclear NE integrity and cell viability (from the maintenance of nuclear RFP–NLS). Cells of interest on small membrane surfaces were then catapulted into single wells in 5.5 μl of lysis buffer (see ‘Generation of scRNA-seq data’ below) in a 96-well capture plate (CapturePlate 96 (D), 415190-9151-000, Carl Zeiss). The cell lysates were quickly transferred to 96-well PCR plates (Eppendorf) by centrifugation and stored in −20 °C for cDNA library generation and scRNA-seq.

Generation of scRNA-seq data

cDNA synthesis and amplification were performed using a modified protocol of the SMART-Seq v4 Ultra Low Input RNA kit for sequencing (Takara Bio). In brief, the manufacturer’s instructions were used with the following modifications: (1) 3 μl of RNAse inhibitor was added per 20 μl for the 10x reaction buffer solution; and (2) all the reaction volumes were decreased by half to maintain reactant stoichiometry. cDNA amplification of single cells was performed by PCR for 21 cycles and the amplified products were purified using AMPure XP paramagnetic beads (Beckman Coulter).

The quality and quantity of the amplified cDNA libraries were assessed using a dsDNA HS Assay kit on a Qubit fluorometer (ThermoFisher Scientific) and an Agilent High Sensitivity DNA kit on a 2100 Bioanalyzer system (Agilent Technologies). cDNA libraries with concentrations below 0.2 ng µl–1 and/or fragment size distributions not showing a peak at 2 kb as expected for the size distribution of full-length mRNAs were excluded from subsequent analyses. Sequencing libraries were generated by tagmentation using a Nextera XT DNA Library Preparation kit (Illumina) with minor modifications of the manufacturer’s instructions. In brief, 0.1–0.2 ng µl–1 of cDNA samples were used in one-quarter of the suggested volumes for all subsequent reactions. Barcodes from the Nextera XT Index kit v2 Sets AD (Illumina) were used for multiplexing, and the quality of RNA-seq libraries was assessed using a Qubit fluorometer (ThermoFisher Scientific) and a 2100 Bioanalyzer (Agilent Technologies). Sequencing was performed on MiSeq and HiSeq 2500 sequencing instruments (2× 100 bp) after quantity normalization and additional quality assessment of the individual libraries by low-pass sequencing on a MiSeq Nano flow cell.

scRNA-seq data processing

The complete workflow of scRNA-seq data processing and downstream analysis was implemented as a snakemake pipeline (publicly available at github.com/chengzhongzhangDFCI/nature2023)47. Details of individual steps are described below.

Alignment and post-alignment processing of sequencing data

Sequencing reads were aligned using STAR (v.2.7.6a) (github.com/alexdobin/STAR) to the Gencode v.25 reference (–twoPassMode basic; –quantMode: TranscriptomeSAM and GeneCounts) and sorted by genomic coordinate. For post-alignment processing, we followed the best practice of GATK (gatk.broadinstitute.org/hc/en-us/articles/360035531192-RNAseq-short-variant-discovery-SNPs-Indels-), which included adding read group information and executing SplitNCigarReads (both using GATK v.4.1.9.0). We skipped duplicate removal as the estimated fractions of duplicated reads was below 5% for all libraries.

Quality assessment of scRNA-seq data

The STAR program outputs various alignment metrics of the RNA-seq data. We report the following information in Supplementary Table 1: the percentages of unmapped reads, multi-mapped reads, reads mapped to no features according to gene annotations, and reads mapped to multiple features according to gene annotations; the number of genes (transcripts) represented by at least 1, 5 or 10 reads; and the average number of reads covering each gene. The primary metric for removing low-quality scRNA-seq libraries was the number of genes covered by ≥5 reads. For control (untreated) RPE-1 cells isolated by FACS, Look-Seq or Look-Seq2 procedures, we excluded cells with <6,000 genes covered by ≥5 reads; for cells with or related to micronucleation, including MN cells, MN sisters, MN daughters and MN nieces, isolated by either Look-Seq and Look-Seq2, we excluded cells with <4,000 genes with 5 or more reads. A total of 464 cells were included in the final analysis, 434 of which were sequenced on HiSeq (Illumina) and another 30 by MiSeq (Illumina). All of these samples are listed in Supplementary Table 1 with annotations of the experimental setup.

Single-cell gene expression analysis

Quantification of total gene expression

We calculated the TPM for each gene using RSEM (deweylab.github.io/RSEM/) with rsem-calculate-expression. We excluded genes with low expression (mean TPM in control cells ≤ 25) that displayed more cell-to-cell variability due to both transcriptional noise and technical variation.

Quantification of allele-specific expression

We assessed allelic gene expression based on allelic depths of RNA-seq reads at heterozygous sites (only single-nucleotide variants) calculated using the ASEReadCounter module of GATK (v.4.1.9.0). The list of heterozygous variants and the haplotype phase of variant genotypes on parental chromosomes were both taken from a previous study48. To calculate the average allelic fraction of transcripts of each gene, we first summed the total number of haplotype-specific (A or B) reads at all variant sites in coding (exonic) and untranslated regions and then calculated the fraction of haplotype-specific read coverage (fA and fB; fA + fB = 1). The averaging of allelic coverage at the gene level helps to improve the accuracy of allelic fraction calculation when there are multiple variants in the transcribed sequence. To eliminate allelic gene expression bias in parental RPE-1 cells (for example, from imprinting), we only assessed allelic expression of genes with roughly equal allelic contributions from both parental homologues in control RPE-1 cells (average allele fraction in control cells is within the range of 0.3–0.7). Two chromosomes required special treatment. For chromosome X transcripts originating from one normally transcribed active X and one epigenetically silenced inactive X, we included all X-linked genes. To account for the presence of a duplicated copy of chromosome 10q (60.78 Mb-qter on GRCh38) that is translocated to the q-terminus of active X, we adjusted the normal range of allele fractions of the single-copy homologue in the trisomic 10q region to be 0.20.4. As we had previously determined the allelic identities of both the active X and the extra copy of chromosome 10q, we used the transcription of the active X and the normal (single-copy) chromosome 10 homologue as reference to calibrate the transcription of the inactive X and the trisomic 10q segment.

Quantification of transcriptional changes relative to normal disomic transcription

We assessed transcriptional changes using both the total transcriptional yield (measured by TPM) and the allelic fraction of transcripts from each gene (Extended Data Fig. 1c). For the total transcription yield, we calculated the transcription ratio by normalizing the TPM in each single cell by the mean TPM in control RPE-1 cells (listed in Supplementary Table 1). To mitigate variations in the TPM ratios derived from individual genes due to transcriptional noise, we calculated the average TPM ratio in 10 Mb genomic intervals for regional transcription analysis and across entire chromosomes or chromosome arms for chromosome (or arm)-level transcription analysis. As different genes show varying degrees of transcriptional variation, we performed a weighted average to attenuate the contributions of genes with more variability that is due to either transcriptional noise49,50 or technical variability51,52. This averaging strategy is described in the Supplementary Information. A similar strategy was used to estimate the average allelic fraction. We further introduced a scaling factor for TPM ratios in each cell to eliminate global changes to TPM values due to significant upregulation or downregulation of one or a few highly transcribed genes (Supplementary Information).

The normalized haplotype-specific transcription value was calculated by multiplying the average TPM ratio by the average allele fraction. For normal disomic transcription, the average TPM ratio is 1 and the average allelic fraction is 0.5, thus the average haplotype-specific transcription is 0.5. For monoallelic transcription that is due to DNA loss or complete epigenetic silencing, we expect the TPM ratio to be around 0.5 (assuming a linear relationship between gene transcription output and copy number24) and the silenced or lost chromosome to have allelic fraction 0; for trisomies with a 2:1 allelic ratio, we expect the TPM ratio to be 1.5. In both cases, the unaltered homologue will have close to normal allelic transcription (0.5) and serve as an intrinsic control for the altered homologue.

We note that experimental perturbations (for example, nocodazole treatment) can cause gene expression changes that are independent of chromosome-specific transcriptional changes on the MN chromosome. We found that the majority of these differentially expressed genes had both low transcription level and low transcriptional variability in control RPE-1 cells. These genes were excluded from the TPM and allelic fraction calculation by the total TPM cutoff (TPM > 25).

Quantification of normal transcriptional variation

We derived a reference distribution of normal transcription of each homologous chromosome from the haplotype-specific transcription in control RPE-1 cells (Extended Data Fig. 1d). These reference distributions were used to assess whether the observed average transcription of a chromosome in a RPE-1 cell is significantly different from normal transcription. Three chromosomes required special treatment. (1) RPE-1 cells frequently acquire alterations to chromosome 12, including trisomic 12, tetrasomic 12p or 12p uniparental disomy. We manually reviewed chromosome 12 transcriptional levels (of both homologues) in the control cells and removed those with chromosome 12 alterations. (2) RPE-1 cells share an extra copy of a 10q segment that is attached to Xa. To match the expression of the single-copy homologue to the mean expression of other autosomes, we multiplied the expression of both homologues by a factor of 1.5. (3) For chromosome X, we similarly multiplied the expression of both Xa and Xi by a constant factor (about 0.6) to match the mean expression of Xa to the mean expression of an autosome. The normalization of chromosome 10q expression and chromosome X expression only affects the visualization of transcriptional changes of individual chromosomes. It does not affect the assessment of whether the observed transcriptional change is within the normal range of variation, which was done separately for each chromosome.

Estimation of transcriptional changes due to chromosomal gain and loss

In addition to normal disomic transcription, we estimated the range of normal transcription of monosomies or trisomies to assess the normality of gene transcription after chromosome mis-segregation, micronucleation or re-incorporation of MN chromosomes. For monosomies, we estimated the residual fraction of transcripts from the deleted homologue based on the RNA-seq data of bona fide monosomies—when a pair of daughter cells showed approximately 0:2 allelic ratio. For trisomies, we estimated the range of normal trisomic transcription using three strategies.

First, assuming the average transcriptional yield from each parental homologue to be similar in both trisomies and disomies, we expected the total allelic transcription yield from two copies of the duplicated homologue in a trisomic cell to be similar to the total transcription yield from both homologues in disomic cells. Therefore, we compared the observed allelic transcription yield of the duplicated homologue to the distribution of total transcription (from both homologues) in control RPE-1 cells to assess the normality of transcription of the duplicated chromosome.

Second, assuming the transcription yield of the single-copy homologue is similar to the transcription yield from either copy of the duplicated homologue, we used the allelic ratio between the duplicated homologue and the single-copy homologue in trisomic cells to assess the normality of transcription of the duplicated chromosome. For this comparison, we used the allelic ratios of the trisomic 10q segment in control RPE-1 cells to assess the normality of transcription of spontaneous trisomies.

Finally, we used the transcription data of RPE-1 cells with de novo trisomies either induced by nocodazole treatment or generated spontaneously during cell culture. To identify bona fide trisomies, we used the following three criteria: (1) we required that there was approximately proportional changes in the chromosome-wide average TPM ratio (1.5); (2) we required that the transcriptional allele fractions were consistent with the DNA allelic fractions (1/3 or 2/3); (3) importantly, we required that each trisomy is either shared by a pair of sibling cells or accompanied by a monosomy in another sibling cell, thereby indicating a de novo mis-segregation event. The last requirement is equivalent to a biological replicate and should exclude random transcriptional variation that affects individual cells. Reference monosomies and trisomies are annotated in Supplementary Table 2 (from both generation 1 and generation 2 samples).

One advantage of using de novo trisomies as a reference is that the observed transcriptional changes are not affected by long-term adaptive changes that may occur in clonal trisomies (for example, 10q). We note that even in de novo trisomies, there is a slight decrease in the expression of each DNA copy, which resulted in a transcription ratio slightly lower than 1.5.

Classification of chromosomal transcription in single RPE-1 cells

We used haplotype-specific transcription to determine whether the observed transcription yield of each chromosome in a single cell is consistent with a normal RPE-1 genome or indicates gain or loss of transcription due to chromosome mis-segregation (including micronucleation). To classify the transcriptional copy-number state based on haplotype-specific transcription yield, we first compared the average haplotype-specific transcription of every chromosome in a RPE-1 cell to the normal range of transcription (‘reference’) derived from control RPE-1 cells (Extended Data Fig. 2d). The normal transcription distribution (Extended Data Fig. 1d) reflects transcriptional variation of a single chromosome and was calculated separately for each parental chromosome. For chromosomes in which transcription levels were outside the normal range (red dots in Extended Data Fig. 2d), we then compared the haplotype-specific transcription yield to normal disomic transcription or complete DNA loss (‘nullisomic’) to determine whether the transcriptional changes were consistent with whole-chromosome gain or loss. If the transcriptional level of a chromosome did not fall within normal ranges of monosomic (1), disomic (2) or nullisomic (0) transcription states, it was classified as intermediate (1+ or 1–). For the duplicated 10q segment or any chromosome inferred to be duplicated, we only assessed whether the transcriptional level was within the normal range of disomic transcription or displayed significantly reduced transcription.

To assess whether the observed transcription yield of a chromosome is within the range of normal monosomic or disomic transcription, we used two-tailed z-tests and considered deviations with Bonferroni-corrected P values of ≥0.05 to be non-significant. For the comparison against nullisomic transcription, we did not calculate the P value as the transcription yield should be strictly zero (that is, no variation); any deviation from zero reflects technical errors (phasing errors, amplification errors, sequencing errors, among others), for which we did not have sufficient data to estimate the null distribution. We classified a chromosome as being nullisomic if the normalized transcription yield was below 0.1 based on the observations of nullisomic chromosomes in bona fide monosomies. The classification of the transcriptional states of all chromosomes with non-disomic transcription in micronucleation-related cells is summarized in Supplementary Table 2.

Identification of mis-segregated chromosomes and chromosomes in micronuclei

We identified mis-segregated chromosomes based on changes in the total and haplotype-specific transcription in all sibling cells from each experiment (family). We first used allele-specific transcription to identify homologous chromosomes with transcription levels significantly deviating from normal (monosomic for that haplotype) transcription (summarized in Supplementary Table 2). We then considered both allelic and total transcription levels across all cells in each family (MN cell, MN sister or their daughters) to determine the integer DNA copy number states of chromosomes with non-monosomic transcription and the chromosome segregation pattern in the family. We also assessed whether the observed transcriptional variation is consistent with the expected outcome of micronucleation, micronucleation-independent mis-segregation that generates reciprocal loss and gain between sibling cells or random transcriptional noise.

The identification of chromosomes that were partitioned into micronuclei is based on matching the allelic imbalance and DNA copy number states of a chromosome in all sibling cells inferred from the transcriptome data to the expected outcomes of different mis-segregation or segregation patterns of the MN chromosome (Extended Data Fig. 2a). This inference automatically determines the parental haplotype of the MN chromatid. Notably, the pattern of micronucleus-related transcriptional changes can be identified independent of the transcription level of the MN chromatid either in the MN cell or in the MN daughter cell that has re-incorporated the MN chromatid. Therefore, the inference of the MN chromatid based on the predicted patterns of transcriptional changes does not affect the assessment of transcriptional normality of the MN chromatid either in a micronucleus or after re-incorporation.

We note that the most definitive features of micronucleus-related transcriptional changes are the loss of transcription of the MN chromatid, which occurs in two scenarios: (1) in the MN sister cell or its progeny (MN nieces) owing to mis-segregation of the MN chromatid; (2) in one of two MN daughter cells that is missing the incompletely replicated MN chromatid in the MN mother cell. In both scenarios, the inference of the MN chromosome relies on the detection of near-complete transcriptional loss in a non-MN cell (MN sister or MN nieces) or in one MN daughter cell that did not re-incorporate the MN chromatid. Therefore, the inference of the MN chromosome is insensitive to both spontaneous transcriptional variability in non-MN cells (as it relies on the detection of complete transcriptional loss) and potential transcriptional changes due to the presence (MN cell) or re-incorporation (MN daughter cell) of the MN chromosome.

We further note a few special cases. First, we identified two MN cell–MN sister pairs (F84 and F206) with no chromosome displaying significant deviations from the normal range of transcriptional variation. We inferred that the MN cell in these two families contained MN chromosomes that had undergone 2:2 segregation and had normal transcription output. Under these circumstances, the MN chromosome is transcribed like a normal chromosome and therefore ‘invisible’ based on the transcriptome data. We nonetheless cannot rule out other possibilities, for example, when the micronucleus contains an acentric chromosome arm (13p, 14p, 15p, 21p or 22p), the transcription output of which cannot be assessed by RNA-seq. The inference of normal MN transcription in these two families reflects a conservative estimate of transcriptional deficiency in micronuclei. Second, in family F71, we identified chromosome 18 to have a 3:1 transcriptional ratio between the MN cell and the MN sister cell. This ratio indicated that the MN cell contained an extra chromosome 18 (due to 3:1 mis-segregation) that is being transcribed to normal levels. The extra chromosome 18 copy could be either contained in the PN or partitioned in the micronucleus; we inferred the extra chromosome 18 copy to be in the PN because we identified chromosome 1p that displays the transcriptional pattern expected for a MN chromosome with defective transcription. Third, there were nine families of MN daughters for which we did not obtain MN niece cells (most of these were collected using the original Look-Seq method). For these cases, we inferred the identity of the MN chromosome by comparing the total and allelic transcriptional imbalance between the MN daughters to the transcriptional profiles of MN daughters for which both the MN chromosome and its segregation pattern can be directly inferred from the data of MN nieces. Specifically, when the re-incorporated MN chromosome displayed normal transcription, the MN daughters showed either about 3:2 or 2:1 transcriptional ratio, which reflected the presence of an extra, normally transcribing chromosome in one MN daughter; we used this information to infer normal transcription of re-incorporated MN when the MN daughters showed the same transcriptional ratios even when no MN niece is available. When the re-incorporated MN chromosome displayed deficient transcription with transcriptional yield a, the MN daughters would show transcriptional ratios of either 2 + a:2 or 1 + a:1. When a ≈ 0, the MN daughters showed identical transcription patterns; we can nonetheless conclude that the extra MN chromatid being present in either daughter cell produced no transcriptional output and therefore must be epigenetically silenced. We inferred family F254 to correspond to this scenario.

Finally, we noted that chromosome 18 and acrocentric chromosomes (chromosomes 13, 14, 15, 21 and 22) displayed more transcriptional variability than other chromosomes. The more pronounced variability of these chromosomes is obvious from the reference distributions. Such variation was generally not shared by sibling cells and/or is inconsistent with the patterns of transcriptional changes predicted by micronucleus-related or micronucleus-independent chromosome mis-segregation events. Therefore, the variable transcription of these chromosomes does not pose a problem for the identification of MN chromosomes.

Quantification of the transcriptional yield of MN chromosomes

After identifying the MN chromatid (both the chromosome identity and the parental haplotype), we estimated the transcriptional yield of the MN chromatid based on the haplotype-specific transcription yield. For MN cells (generation 1) of 2:2 segregation, they contained a single copy of the MN chromatid in the micronucleus, we therefore directly derived the transcriptional yield of the MN chromatid from the transcriptional yield of the MN haplotype. For MN cells having undergone 3:1 (MN cell: MN sister) mis-segregations, the haplotype-specific transcriptional yield of the MN haplotype represented the combined transcription output from both the MN chromatid and its intact sister chromatid in the PN. In this scenario, we compared the transcriptional yield of the MN haplotype to the transcriptional yield of reference disomic transcription levels to assess whether the MN chromatid displayed normal or deficient transcription.

For re-incorporated MNs, if the MN chromosome was inferred to have undergone a 2:2 segregation in generation 1, then the single-copy MN chromatid is distributed to one or both MN daughter cells. In this scenario, we estimated the transcription yield of the re-incorporated MN chromatid using the combined transcriptional yield of the MN haplotype in both MN daughters (this accounts for possible fragmentation and reciprocal distribution of fragments of the MN chromatid into both daughters). We then compared the transcription yield of the MN haplotype to the range of normal transcription of a single homologue to assess transcriptional normality or deficiency. If the MN chromosome was inferred to have undergone a 3:1 segregation in generation 1, then each MN daughter contained an extra, intact copy of the MN chromosome in addition to the re-incorporated MN chromatid. In this scenario, we compared the transcriptional yield of the MN haplotype in each MN daughter cell to the ranges of both monosomic transcription and disomic transcription to assess the normality or deficiency of transcription of the re-incorporated MN chromatid.

The data of normalized transcription ratios, inferred DNA copy number states and the transcriptional yields of MN chromatids and haplotypes are summarized in Supplementary Table 2.

Same-cell correlative live-fixed imaging

For the same-cell correlative live-fixed imaging experiments using MDC1-expressing cells (Fig. 3a,b), cells were seeded on 35-mm ibiTreat Grid-500 dishes (Ibidi) with a gridded imaging surface. Live-cell imaging was performed using wide-field fluorescence microscope as described in the ‘Live-cell imaging’ section. At the end of live-cell imaging, cells were immediately fixed by incubation with methanol for 10 min at −20 °C. A snapshot of the last imaging frame including a differential interference contrast image was taken to visualize the grids of the coverslip dish. The grid coordinate information and the last snapshot of the time-lapse images were used to locate the cells of interest after fixation and indirect immunofluorescence imaging.

For experiments using the RPE-1 RFP–NLS GFP–H2B cells (Extended Data Fig. 9a) and the modified U2OS 263 cells (Fig. 4d,e), live-cell imaging was performed as described above. At the end of the live-cell imaging, cells were fixed by incubation with methanol for 10 min at −20 °C for RPE-1 RFP–NLS GFP–H2B cells or 4% paraformaldehyde for 20 min at room temperature (modified U2OS 263 cells). Cells of interest were located according to the grid coordinates for subsequent indirect immunofluorescence analysis.

Indirect immunofluorescence and confocal microscopy of fixed cells

Cells were fixed and prepared for indirect immunofluorescence and confocal microscopy as previously described11,26.

Images were acquired on a Nikon Ti-E inverted microscope (Nikon) with a Yokogawa CSU-22 spinning disk confocal head with the Borealis modification or a Ti2 inverted microscope fitted with a CSU-W1 spinning disk. Z-stacks of 0.4–0.7 μm spacing were collected using a CoolSnap HQ2 CCD camera (Photometrics) or a Zyla 4.2 sCMOS camera (Andor) with a ×60/1.40 NA or a ×100/1.45 NA Plan Apochromat oil-immersion objective (Nikon).

The following antibodies were used for indirect immunofluorescence imaging: phospho γH2AX (Ser139) (Millipore, 05-636-I; 1:400); H3K27ac (Active Motif, 39133; 1:200); MDC1 (Abcam, ab11171; 1:1,000); MDC1 (Sigma-Aldrich, M2444; 1:1,000); phospho RNA PolII S5 (Millipore, MABE954, clone 1H4B6; 1:400); Cdk9 (Cell Signaling, 2316; 1:10); CDK12 (Abcam, ab246887; 1:400); 53BP1 (Santa Cruz, 22760S; 1:100); H3K27me3 (ThermoFisher Scientific, MA511198; 1:1,000); H3K9ac (Cell Signaling, 9649S; 1:400); H3K9me2 (Cell Signaling, 9753S; 1:400); POM121 (Proteintech, 15645-1-AP; 1:200); phospho H3T3 (Millipore, 07-424, 1:12,000); phospho H3S10 (Abcam, ab47297; 1:200); and fibrillarin (Abcam, ab4566; 1:500). Staining of Dam-methylated DNA in fixed cells was done using purified GFP-tagged m6A-Tracer protein as previously described53.

Image analysis of fixed-cell experiments

Two image analysis pipelines were used in this study. To characterize the transcriptional state and chromatin alterations in micronuclei (Fig. 1e–g and Extended Data Figs. 4 and 5), we used customized ImageJ/Fiji macros as previously described26. To characterize MN bodies or MN-body-like structures (Figs. 3 and 4 and Extended Data Figs. 9, 10 and 12), we used a Python-based analysis pipeline47 with additional preprocessing procedures performed using ImageJ/Fiji software. Both pipelines overall consisted of the following steps: (1) cells of interest were identified and their primary nuclei were segmented; (2) micronuclei or re-incorporated MN chromosomes (or chromosome bridges) were identified and segmented; (3) mean FI values for labelled proteins or DNA were quantified over the segmented regions of interest (ROIs).

Analysis of the transcription and chromatin alterations in micronuclei

Image analysis of micronuclei in immunofluorescence experiments were performed as previously described26.

Image segmentation and ROI identification

First, the three-dimension (xyz) images of primary nuclei and micronuclei were segmented using the Li or Otsu thresholding method in ImageJ/Fiji with the DNA (Hoechst) signal as input. Second, the nuclear segmentations were further refined using the ImageJ/Fiji functions Watershed and Erode to remove connecting pixels bordering abutting nuclei. Third, nuclear segmentations containing primary nuclei and micronuclei were manually selected as ROIs using the ImageJ/Fiji functions Wand Tool. ROIs from one single focal plane where primary nuclear and micronuclear DNA signal were in focus were manually selected and used for the following quantification.

FI quantification

The mean FI of labelled proteins or DNA was quantified over the selected ROIs from their corresponding microscope fluorescence channels. For quantification of nuclear proteins (for example, RNAP2-Ser5ph, RFP–NLS; Fig. 1e–g and Extended Data Fig. 4) or labelled DNA (Hoechst), the mean FI values were calculated for micronuclear ROIs and primary nuclear ROIs, respectively. These mean FI values were subtracted by the mean FI value of the non-nuclear background to obtain the background-subtracted mean FI. The background-subtracted mean FI of micronuclei were divided by the background-subtracted mean FI of the corresponding PN to obtain the MN/PN mean FI ratios. For quantification of histone modifications, including H3K27ac, H3K9ac, H3K9me2, H3K27me3 and γH2AX, the MN/PN mean FI ratios of these marks were further divided by the MN/PN mean FI ratio of DNA (Hoechst) to obtain the DNA-normalized FI ratios. To analyse micronuclei with intact or ruptured NE, micronuclei with MN/PN mean FI ratios of NLS below 0.1 relative to the PN were considered ruptured and above 0.3 were considered intact. Micronuclei with MN/PN FI ratios in between were excluded for analysis, as the assessment of NE integrity is not definitive.

In addition, the background-subtracted mean FI of RNAP2-Ser5ph are shown as exact FI values without normalizing to the background-subtracted mean FI of the corresponding PN (Extended Data Fig. 4b).

Analysis of generation 2 re-incorporated MN chromosomes

Analysis of incorporated MN chromosomes were performed primarily using an automated script written in Python47. Further details are available upon request.

Image segmentation and ROI identification

Step 1, all candidate primary nuclei within the three-dimensional images were identified and segmented either using the Li thresholding method with DNA (Hoechst) signal as the input or the Otsu or Li thresholding method with RNA Pol2S5 signal as the input. The nuclear segmentations were further refined using binary mask operations similar to the procedures described above for the micronuclei analysis pipeline.

Step 2, a smaller cropped three-dimensional image (z-stack) was generated for each segmented PN object to minimize variability in the fluorescence signal across the entire image. Only PN objects located within the middle 50% of our images were analysed to minimize the uneven illumination due to the large field of view of the camera (2,048 × 2,048 pixels). From these cropped three-dimension images, a single focal plane in which the MDC1 or m6A-Tracer (hereafter m6T) signal was in focus was selected. This single focal plane was determined as the focal plane with the largest standard deviation (s.d.) in the FI distribution of all pixels (which is used as an estimator of the strongest overall signal) from the MDC1 or m6T channel. The cropped xy images and segmentations for each candidate PN objects were then analysed.

Step 3, primary nuclei that contained potential re-incorporated MN chromosomes were located using the presence of large foci of MDC1 or m6T. To identify MDC1 or m6T large foci for each candidate PN, the FI of all pixels within the corresponding nuclear segmentation were quantified to generate a nuclear FI distribution. Positive pixels were selected if their FI > 2 s.d. above the mean for the nuclear FI distribution. These positive pixels were subject to an area size filter (300 pixels) to remove small noise pixels so that only connected-positive pixels larger than the size filter were kept to generate the final ROIs for MDC1 and m6T. For nuclei with multiple valid MDC1 and m6T foci (for example, from two or more MN chromosomes), all foci were analysed together per each nucleus. Additionally, to increase detection accuracy of m6T foci from cells with a variable m6T expression, candidate nuclei of interest were manually screened using ImageJ/Fiji. The xy coordinates of these candidate nuclei were supplemented as additional inputs for the analysis pipeline and used for locating valid nuclei containing m6T foci according to the above criteria (FI > 2 s.d. and > 300 pixels) using Python.

Step 4, segmentations for other objects (nuclear or subnuclear structures) that were used for the analysis were generated. Specifically, the ROIs for the primary nuclei were defined by excluding the MDC1 and m6T segmentations as well as the nucleoli segmentations from the original nuclear segmentation (see step 1). The nucleoli segmentations were generated using the lower 10% of the primary nuclear FI distribution of RNAP2-Ser5ph. This 10% (percentile) cutoff was determined by comparing with the nucleoli segmentations using fibrillarin-positive signals (which are pixels for which FI > 3 s.d. above the mean for its total nuclear FI distribution; Extended Data Fig. 9f): the highest overlap with the nucleoli segmentations using the fibrillarin-positive signal was achieved using the lower 10% of the nuclear RNAP2-Ser5ph FI as the cutoff for nucleoli segmentations. ROIs or the γH2AX-positive areas were defined by γH2AX-positive pixels for which FI > 3 s.d. above the mean for the nuclear FI distribution. The ROI area occupancy ratio of the γH2AX-positive pixels within the m6T foci was used to define different levels of γH2AX in re-incorporated m6T micronuclei (Fig. 4c and Extended Data Fig. 10e).

Step 5, a randomized control ROI was segmented by randomly picking a smaller area (at a size similar to the MDC1 or m6T ROI) within the primary nuclear ROI generated above for each cell containing a MDC1 or m6T foci. The random picking process was performed using our Python-based analysis pipeline.

To validate the accuracy of MDC1 and m6T foci identification, the ROI segmentations of a random subset of cells were manually examined. The mis-identification rate of our automated pipeline using random subsets of cells was typically lower than 10%. Additionally, for the m6T dataset after quantification (see below), outliers were also manually examined. The mis-identification rate for the outliers of m6T dataset was 25%. These mis-identified m6T foci (n = 27) were mostly m6T-positive micronuclei immediately next to the primary nuclei and were distributed near-symmetrically at the top and bottom of the measurement distribution. These images were excluded during the analysis.

FI quantification

The mean FI of labelled proteins or DNA was quantified over the segmented ROIs above from their corresponding microscope channels. All mean FI values were then background subtracted by the corresponding mean FI of the non-nuclear background. The background-subtracted mean FI of MDC1 or m6T ROIs (see step 3 above) and the background-subtracted mean FI of the randomized control ROI (see step 5 above) were normalized to the background-subtracted mean FI of the corresponding primary nuclear ROI (see step 4 above) to obtain the normalized mean FI ratios of labelled proteins or DNA.

For quantification of histone modifications, including H3K27ac, H3K9ac, H3K9me2, H3K27me3, γH2AX, H3S10ph and H3T3ph, the normalized mean FI ratios of these marks were further divided by the normalized mean FI ratio of DNA (Hoechst) to obtain the DNA-normalized FI ratios. This controlled for signal enrichment due to chromosome compaction.

Analysis of re-incorporated fragments from chromosome bridge resolution

Quantification of RNAP2-Ser5ph of incorporated bridge segments after bridge resolution and cell division (Extended Data Fig. 12a) was performed in a similar manner to the re-incorporated micronuclei as described above, with ROIs for MN-body-like structures from bridge segments identified using MDC1-positive signal (FI > 2 s.d. and >300 pixels).

Analysis of incorporated MN chromosomes for the correlative live-fixed imaging

For quantification of RNAP2-Ser5ph and H3K27ac in incorporated MN chromosomes marked by MDC1 foci (Fig. 3a,b), the analysis was performed in a similar manner as described above except for the following differences: (1) The MN-body segmentations were manually drawn along the MDC1-enriched pixels; and (2) the control (or PN) segmentations were manually defined as a large PN region excluding nucleoli. ROIs for MN bodies and controls were manually selected over these segmentations for a single Z-plane where MN bodies were in focus. The mean FI MN body-to-control ratios were obtained by dividing the background-subtracted mean FI of the MN body ROIs to the background-subtracted mean FI of the control ROIs. Note that the time-lapse images were analysed manually to assign the re-incorporated daughters and rupture events, and the low sample size allowed for the manual quantification analysis. In addition, note that some daughter cells with MN bodies that were included in the analysis had new micronuclei in the generation 2 samples, independent of the detected re-incorporated MN chromosome.

Graphical data from the imaging analyses were plotted and statistical analyses were performed using GraphPad Prism (v.9.4.0; GraphPad Software).

Analysis of live-cell imaging data for MS2-marked nascent transcription

To quantify the MS2-marked transcription level (Figs. 2e,f and 4d and Extended Data Fig. 8), an automated script written in Python was used with assists using ImageJ/Fiji.

Image segmentation and LacO and MS2 foci tracking

MN cells with the LacO and MS2 (LacO/MAS) reporter or control cells without MN were manually identified from the image series, and image series of interest were divided into three parts: generation 1 interphase, mitosis, and generation 2 interphase.

For time frames covering the generation 1 interphase (for both control cells and for MN cells), all primary nuclei were segmented using the cellpose package54, and all LacO/MS2 foci (in both MN and PN) were segmented using the Yen segmentation method55 for each time frame. The primary nuclei and LacI foci of interest in the first time point were identified by finding the object segmentation with the shortest distance to a user-provided xy centroid coordinate of the nucleus and the LacO/MS2 focus, respectively. For the following time points, the same nuclei and LacO/MS2 foci were automatically identified by finding the object segmentation for which the distance was the shortest to the identified nuclei and the LacO/MS2 foci segmentations from the previous time point (or time points). The identification of the LacO/MS2 foci and their corresponding primary nuclei was manually evaluated to assist the tracking of the correct LacO/MS2 foci. The xy centroid coordinates of the segmentations were used for estimating the object moving distance above.

For time points around mitosis, the nuclei and LacO/MS2 foci were manually tracked in ImageJ/Fiji to accurately identify the partitioning of LacO/MS2 foci into daughter cells during mitotic exit.

For time frames after mitosis (generation 2), daughter primary nuclei were segmented and tracked as described for the first interphase. For LacO/MS2 foci segmentation tracking, we used a combination of several criteria for technical reasons. Because long-term binding of LacI to LacO can impair DNA replication, we terminated LacI gene expression at around 18 h after mitotic shake-off. This led to a loss of LacI signal in a subset of daughter cells during the generation 2 interphase, particularly evident at later time points. For these cells that had lost the LacI signal, we quantified the FI distribution for the nuclear MS2 signal and identified positive MS2 pixels for which FI > 3 s.d. above the mean for the nuclear MS2 FI distribution. The enrichment of such nuclear MS2-positive foci (if present) was then used for the LacO/MS2 foci segmentation tracking for the subsequent time points. For time points in which LacI foci persisted, we tracked the LacO/MS2 foci as described for generation 1 interphase. If no LacO/MS2 foci could be identified during generation 2 interphase, time points were annotated as having no MS2 expression, and therefore no segmentation was performed.

Additionally, we manually examined the nuclei and LacO/MS2 foci tracking because the estimation of object movement using minimal centroid moving distance could lead to incorrect tracking when objects swap positions between time points. For these time points, additional xy pixel centroid coordinates for the nuclei and LacO or MS2 foci were obtained using ImageJ/Fiji and supplemented the automated object tracking.

Fl quantification for ROIs

ROIs for the LacO or MS2 foci and the corresponding PN were selected from their segmentation as described above. ROIs from one single focal plane where the LacI signal was in focus were used for FI quantification. Based on these ROIs, the mean FI of the MS2 signal for LacO/MS2 foci and matching PN pairs and of the non-nuclear background was quantified. The background-subtracted mean FI of LacO/MS2 foci was divided by the background-subtracted mean FI of the primary nuclear areas (excluding the LacO/MS2 foci) to obtain the normalized MS2 level.

For time points that were annotated as having no MS2 expression in generation 2, a value of 1.7 was assigned because this value is the minimal detectable normalized MS2 signal for the positive MS2 foci in the controls (see details below) for the purpose of plotting the graphs. To obtain this value, we analysed 23 control cells over two cell-cycles for which the LacO/MS2 focus was located within the PN for both generations 1 and 2. We quantified the normalized MS2 level during the generation 2 interphase for time points at which the cells had lost the LacI signal after we stopped LacI expression. These control cells maintained MS2 reporter transcription, and their LacO/MS2 foci were detected and segmented from MS2-positive pixels (FI > 3 s.d. above the mean for the nuclear MS2 FI distribution, as described above). The lowest of the normalized mean FI for all detected MS2-positive foci (n = 477) from all imaged time points above was 1.7, defining 1.7 as the minimum detectable mean MS2 signal in the control experiments. Therefore, 1.7 was used as the normalized MS2 signal when no positive MS2 foci could be detected. Note that this is a conservative estimate because the actual MS2 level can be lower as measured for some MN bodies in which the LacI signal was present and can be used for segmentation. In other words, this should underestimate the degree of MS2 signal loss for MN chromosomes that are in a normal generation 2 daughter cells.

SDS–PAGE and western blotting

Lysis of RPE-1 Dam and control RPE-1 cells (Extended Data Fig. 10c) was performed after trypsinization and washes with PBS by adding an equal volume of a 2× lysis buffer (100 mM Tris-HCl pH 6.8, 4% SDS and 12% β-mercaptoethanol). Whole-cell lysates were denatured at 100 °C for 10 min, Laemmli–SDS sample buffer (Boston BioProducts) was added, and the samples were subjected to SDS–PAGE on NuPAGE 4–12% Bis-Tris gradient gels (Novex Life Technologies). The proteins were then transferred onto a nitrocellulose membrane (Millipore). The membranes were blocked using Odyssey blocking buffer (LI-COR) and were incubated with primary antibodies for 1 h at room temperature or overnight at 4 °C. The primary antibodies and dilutions used were anti-mCherry rabbit 1:1,000 (ab167453, Abcam) and anti-GAPDH mouse 1:5,000 (ab9485, Abcam). After washes with PBS-T, we incubated the membranes with the fluorescent secondary antibodies IRDye 680RD donkey anti-rabbit 1:5,000 (926-68073, LI-COR) and IRDye 800CW donkey anti-mouse 1:5,000 (926-32212, LI-COR) for 1 h at room temperature. Membranes were visualized using a ChemiDoc MP imaging system (Bio-Rad). Note that the images shown in Extended Data Fig. 10c were cropped to show the bands at the protein size. The full scan (uncropped) blots are shown in Supplementary Fig. 1.

FACS

RPE-1 megaDam cells (see the section ‘Cell culture and cell line construction’) were analysed by FACS for mCherry expression using a LSR Fortessa flow cytometer (BD) (Extended Data Fig. 10b). Cells were stained with DAPI for dead-cell exclusion and live cells were analysed for their percentage of mCherry-positive cells (excluding autofluorescent cells by gating PE relative to FITC). Data were recorded using FACSDiva (v.8.0; BD) software, and FlowJo (v.10.7.1; BD) was used for data analysis. Examples of the gating strategy are shown in Supplementary Fig. 2.

Genomic analysis of bridge clones

DNA sequencing

Genomic DNA was purified using a DNeasy Blood and Tissue kit (Qiagen) and was then fragmented on a Covaris M220 instrument according to the manufacturer’s protocol. Libraries were prepared using Swift S2 Acel reagents on a Beckman Coulter Biomek i7 liquid handling platform from approximately 200 ng of DNA with 14 cycles of PCR amplification. DNA libraries were quantified on a Qubit 2.0 Fluorometer (Life Technologies) and fragment size distributions were evaluated on a Agilent TapeStation 2200 (Agilent Technologies). Pooled libraries were further evaluated with low-pass sequencing on an Illumina MiSeq and then sequenced to approximately 5× mean genome coverage on an NovaSeq 6000 instrument (Illumina) with 2× 150 bp paired-end configuration in the Molecular Biology Core Facilities at Dana-Farber Cancer Institute. Haplotype-specific DNA copy number was calculated using the same workflow as previously described11,48. DNA rearrangements shown in Extended Data Fig. 11 were taken from previous analyses11.

RNA-seq

RNA extraction, library preparation and sequencing were conducted at Azenta Life Sciences. In brief, total RNA was extracted from fresh-frozen cell pellet samples using a RNeasy Plus Universal mini kit (Qiagen). RNA samples were quantified using a Qubit 2.0 Fluorometer (Life Technologies), and RNA integrity was evaluated using a TapeStation 4200 (Agilent Technologies). An ERCC RNA Spike-In Mix kit (4456740, ThermoFisher Scientific) was added (but not used) and sequencing libraries were prepared using a NEBNext Ultra RNA Library Prep kit for Illumina (NEB). The quality of the sequencing libraries were validated on a Agilent TapeStation (Agilent Technologies), and the concentration of the libraries were quantified using a Qubit Fluorometer and by quantitative PCR (KAPA Biosystems). The samples were sequenced on an Illumina instrument (4000 or equivalent) with 2× 150 bp paired-end configuration with an average of around 60 million reads per sample.

Bulk RNA-seq data were aligned using STAR (v.2.7.10a) with the same parameters as single-cell RNA-seq data processing. As the estimated fraction of duplicate reads in bulk RNA-seq data was above 5%, we followed all steps of post-alignment processing (including duplicate removal) as described in the best practice of GATK. All post-alignment processing was carried out using GATK (v.4.2.6.1). The remaining steps of RNA-seq data processing were identical to the processing of scRNA-seq data.

We first generated feature counts from analysis-ready RNA-seq bam files using featureCounts from Subread 2.0.1 (subread.sourceforge.net) and then calculated total TPM47. We performed a similar global TPM normalization step for each sample by scaling the TPM values by a constant factor to match the median expression of genes that are transcribed bi-allelically and have mean TPM between 1 and 1,000 (6,683 total). After global normalization, we calculated allelic transcription of each gene using the same procedure as for the single-cell transcriptome analysis. To assess the transcriptional yield of each gene copy, we further divided both the total and allelic transcriptional levels by the DNA copy number in 250 kb local intervals; the DNA copy number was determined from whole-genome DNA sequencing data generated on the same culture. To quantify transcriptional changes relative to normal transcription, we normalized the transcription yield (both total and allelic) in bridge and control clones by the transcriptional level in the parental RPE-1 sample. The final TPM ratio (gene level) was then used to assess transcriptional changes (both total and haplotype-specific) independent of copy-number variation.

ATAC-seq

The preparation of nuclei, transposition and amplification by PCR were performed as previously described56. In brief, cells were trypsinized and washed twice with PBS. Then, 10,000 cells in 5 μl PBS were transposed in 42.5 μl of transposition buffer (33 mM Tris acetate buffer, 66 mM potassium acetate, 10 mM magnesium acetate, 0.1% NP-40, 16% DMF, 0.004× protease inhibitor cocktail and ddH2O to 42.5 µl) and 2.5 μl of TDE1 Illumina Tn5 transposase. The transposition reaction was conducted for 30 min at 37 °C, followed immediately by DNA purification using a ZYMO DNA Clean and Concentrator 5 kit (Zymo Research). Cycle-determining quantitative PCR was conducted to amplify libraries and stop amplification before saturation. The amplified libraries were purified using a ZYMO DNA Clean and Concentrator 5 kit and quantified by using a Qubit 2.0 Fluorometer. The libraries were normalized and pooled based on quantitative PCR analysis and were subsequently sequenced on a NovaSeq S1 instrument (Illumina) with 2× 50 bp paired-end configuration or a NextSeq instrument (Illumina) with 2× 38 bp configuration at the Bauer Core Facility of Harvard University.

Reads were trimmed to remove adapter sequences and then aligned to hg38 using Bowtie2 (ref. 57) with the following parameters: -X2000–rg-id. Chromatin accessibility peak calling was conducted as previously described58. In brief, we first performed peak-calling on each sample using MACS2 (ref. 59) with the following parameters/options: –nomodel,–nolambda,–keep-dup all,–call-summits.

We then combined and merged overlapping peaks (within 400 bp) called from all samples to create a unique list of peaks (259,036 total). The fragment count within each peak was calculated using the getCounts function from chromvar60, and then normalized using the preprocessCore normalize.quantiles function61.

To assess changes in chromatin accessibility in the bridge clones, we first divided the quantile-normalized fragment count60 for every peak by the local DNA copy number (250 kb bins) to account for DNA gain or loss, which was almost exclusively restricted to chromosome 4.

To account for technical variation during library preparation, we applied a permutation approach to generate a reference ATAC profile for each individual clone based on the ATAC profiles in control clones. First, for each ATAC-seq peak, we generated a replicate set of 50 peaks with similar GC content and average accessibility in the control samples (ten control RPE-1 subclones) using the getBackgroundPeaks(<normalized.counts>, bias = <gc.bias>) command from chromvar60.

Here <gc.bias> was calculated for each peak region (300 bp) and <normalized.counts> denotes the ATAC fragment counts in the ten control subclone samples. Assuming the replicate peaks are subject to similar technical variation, we then used the ATAC-seq densities of replicate peaks as the null distribution for the peak of interest to perform intra-sample background normalization by random permutations.

During each permutation, we randomly selected 1 out of 50 replicate peaks for each peak in a given genomic interval to create a random reference ATAC profile. By generating a sufficient number of reference ATAC profiles through permutations, we could assess the statistical deviation of the observed ATAC profile in each genomic interval from the null distribution generated by permutations. To enable a sufficient number of permutations, we only considered intervals with at least 10 peaks per Mb (with a maximum of 5010 ≈ 9.8 × 1016 permutations). We performed around 106 permutations for each interval (lower than the number of all possible permutations) to identify outliers with P values on the order of 10−6.

The above-described permutation sampling was performed on the fragment counts in each sample in 1, 5 or 10 Mb intervals. Based on the null distributions derived from random permutations, we then calculated the fold change of the observed ATAC density relative to the mean of the null distribution. The re-centered fold change of ATAC signal is shown in Extended Data Fig. 12b. We further estimated the likelihood of the observed average ATAC density of each interval in each sample based on the null distributions generated by permutations (an example is shown in Extended Data Fig. 12d). Shown in Figs. 5b and 12c are the average fold change of ATAC signals across all 12 bridge clones.

Our permutation sampling directly accounts for GC bias. It also accounts for non-uniform peak density across the genome. Additionally, in our analysis, we primarily focused on clonal or near-clonal changes that are more likely generated by the initial formation and resolution of bridges than subclonal changes that are more likely to have arisen downstream. We therefore focused on intervals with a significant reduction in the average ATAC fold change (<0.70).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Read more here: Source link