Droplet Hi-C enables scalable, single-cell profiling of chromatin architecture in heterogeneous tissues

Chromatin organization plays a crucial role in both embryonic development^1,2 and the progression of various diseases^3,4. During development, the chromatin structure influences which genes are turned on or off by the distal regulatory elements^5,6,7,8. Abnormal chromatin organization can lead to aberrant gene expression or silencing, contributing to cancer development and metastasis⁴. Indeed, components of the nuclear machinery that regulates chromatin organization are among the most frequently mutated genes in human cancers^9,10,11,12. Thus, a comprehensive analysis of the chromatin organization is crucial for understanding the gene regulatory programs involved in both normal development and disease pathogenesis.

Analyzing chromatin organization in primary tissues and tumor biopsies presents unique challenges owing to the complexity of chromatin dynamics and technical limitations. First, the biospecimens are often heterogeneous, containing multiple cell types. Current techniques such as bulk Hi-C could not resolve specific chromatin organization patterns relevant to particular cell types, especially in tumors in which cancer cells coexist with stromal and immune cells. Second, the data obtained from bulk chromatin organization assays are complex, and relating changes in chromatin organization to specific functional outcomes can be difficult. Addressing these challenges requires the development of more sensitive single-cell chromatin structure profiling assays and advanced computational tools for data analysis.

Great strides have been made in recent years in single-cell Hi-C technologies^1,2,3,4. In general, single-cell Hi-C methods can be categorized into low-throughput microwell-based approaches and high-throughput combinatorial-indexing-based approaches. In the case of microwell-based single-cell Hi-C methods, cells or nuclei are individually dispensed into microwells, and library construction is carried out in each microwell in parallel, frequently with the help of automated liquid handlers. Notably, several single-cell Hi-C methods have been developed, such as single-cell Hi-C^13,14,15single-nucleus Hi-C¹⁶ and diploid chromatin conformation capture (Dip-C)¹⁷and third-generation sequencing-based scNanoHi-C¹⁸. Recently, additional microwell-based single-cell Hi-C methods have emerged that combine Hi-C analysis with assays for one or more additional molecular modalities in a single cell. For example, Methyl-HiC¹⁹ and single-nucleus methyl-3C sequencing (sn-m3C-seq)²⁰ capture chromatin interactions and DNA methylation patterns from the same cell. Hi-C and RNA sequencing employed simultaneously (HiRES)²¹ jointly performs Hi-C and RNA sequencing (RNA-seq) to explore the functional relationship between three-dimensional genome organization and transcriptome dynamics. In general, microwell-based techniques are limited in scale and difficult to adopt broadly owing to lengthy procedures and high cost. On the other hand, combinatorial-indexing-based single-cell Hi-C methods use combinatorial barcoding strategies to achieve high throughput and scalability. Previously, a single-cell combinatorial indexed Hi-C (sci-Hi-C)²² method allowed the generation of single-cell Hi-C libraries from a few thousand cells, albeit with limited genomic coverage in each cell. Building upon the same strategy, genome architecture and gene expression by sequencing (GAGE-seq)²³ was developed to simultaneously profile chromatin interactions and gene expression in single cells, to achieve high throughput, multimodality and high coverage per cell. However, the lengthy and largely manual combinatorial indexing procedure still poses a challenge for its general adoption.

Cancer cells often show large-scale structural variations (SVs), such as deletions, insertions, translocations, inversions, duplications and extrachromosomal DNA (ecDNA), all of which have been implicated in cancer initiation and progression²⁴. First reported in 1965 as double minutes (DM), ecDNA has recently been shown to be prevalent in human cancers and nearly always harbors oncogenes^25,26. Unlike the kilobase-sized circular DNA (eccDNA) found in healthy somatic tissues²⁷ecDNA varies in size from dozens of kilobases to megabases, making it 100 to 1,000 times larger²⁸. The ecDNA is characterized by high amplification and a circular structure²⁹. Bulk Hi-C can detect SVs and ecDNAs in tumor tissues^30,31but it struggles to resolve tumor heterogeneity and evolution during therapy. Moreover, bulk DNA sequencing cannot distinguish ecDNAs from homogeneously staining regions (HSRs).

Here we introduce a highly scalable, generally accessible, droplet-based single-cell Hi-C method, Droplet Hi-C, which combines an in situ chromosomal conformation capture (3C) assay with commercially available droplet microfluidics, to simultaneously capture the three-dimensional genome structure from tens of thousands of individual cells in a single experiment. We showed the utility of Droplet Hi-C data in resolving cell-type-specific chromatin architecture in complex tissues such as the mouse brain. We further showed that Droplet Hi-C could be used to identify aberrant chromatin structure in cancer cells. We used Droplet Hi-C to identify ecDNA and mapped their chromatin interactome in tumor cells at single-cell resolution. Finally, we extended this method to enable the simultaneous capture of transcriptome and chromatin architecture in single cells.

Development of Droplet Hi-C

Droplet Hi-C is based upon an in situ Hi-C procedure³². It captures spatial proximity genome-wide between chromatin fibers in formaldehyde-cross-linked cells or nuclei through restriction digestion and ligation in situ (Fig. 1a). After SDS treatment to remove histone proteins, DNA fragmentation and capture is then carried out in a commercially available microfluidic platform (that is, the 10x Genomics single cell assay for transposase-accessible chromatin (ATAC) kit), in which cell-specific DNA barcodes are added to the DNA fragments. This is followed by sequencing library construction and next-generation sequencing (Fig. 1a). The whole procedure lasts about 10 h from fixed cells or nuclei to final sequencing libraries, and 8 samples can be processed in parallel, enabling the profiling of 40,000 or more cells simultaneously (Fig. 1a–c). The throughput of Droplet Hi-C surpasses plate-based single-cell Hi-C methods by an order of magnitude, offering the shortest experimental duration and hands-on time to date (Fig. 1b). Given the widespread use of commercial microfluidic systems, ease of the experimental procedure and relatively low cost (Fig. 1c), Droplet Hi-C has the potential to be rapidly adopted by the research community.

To test the performance of Droplet Hi-C, we first applied it to a mixture of the human HeLa S3 cell line and the mouse embryonic stem cell (mESC) line. Equal numbers of cells from the two cell lines were combined after cross-linking and subjected to Droplet Hi-C. The results showed that 1,773 human and 3,489 mouse high-quality cells (with a total of>1,000 read pairs) were recovered after shallow sequencing, along with 284 potential doublets (Extended Data Fig. 1a). We further carried out Droplet Hi-C with a mixture of three human cell lines (K562, GM12878 and HeLa S3). After quality control and clustering using Higashi³³we obtained 3,709 high-quality cells with a median of 108,439 unique read pairs in each cell (duplication rate, 23.9%), including 1,604 HeLa S3 cells (median, 101,460 read pairs), 1,359 GM12878 cells (median, 109,518 read pairs), 606 K562 cells (median, 111,255 read pairs) and 140 cells with chromatin interaction patterns reminiscent of mitotic chromosomes (median, 119,573 read pairs; Extended Data Fig. 1b and Supplementary Table 1). By performing genome-wide correlation analysis with bulk in situ Hi-C references, each distinct population can be confidently assigned to one cell type (Extended Data Fig. 1c–f). The mitotic population is determined based on the similar contact frequency decay by distance pattern compared with HeLa S3 cells arrested at prometaphase³⁴and the signature contact maps enriched with frequent interactions along the diagonal (Extended Data Fig. 1d,e). On the chromosome level, cell-type-specific chromatin structures can be observed, such as the distinct Hi-C pattern of chromosomal rearrangements on chromosome (chr) 11 in HeLa S3 cells (Extended Data Fig. 1e). When focusing on a region encompassing the K562-specific upregulated gene LGR4 (ref. ³⁵), we observed an active A compartment at the LGR4 genomic locus that is specific to the K562 cells (Extended Data Fig. 1g). The differences in compartment scores as well as insulation scores among different cell lines can also be faithfully reproduced as in the published datasets (Extended Data Fig. 1g,h). We also performed Droplet Hi-C on a mixture of diploid cells (GM12878 and WTC-11) to show that single-cell analysis of chromatin conformation enables separating different cells with normal karyotypes. Indeed, clustering of the 3,669 cells passing quality control threshold uncovered three populations (Extended Data Fig. 1i). Two of them correspond to the GM12878 cells (n = 2,236) and the WTC-11 cells (n = 1,202), based on their high correlation with bulk in situ Hi-C data (Extended Data Fig. 1j). The third population, consisting of 231 cells, appear to be cells in mitosis based on the contact maps and enrichment of long-range contacts characteristic of mitotic chromosomes³⁴ (Extended Data Fig. 1k,l). In summary, these results show that Droplet Hi-C can accurately distinguish cell-type-specific chromatin organization.

Droplet Hi-C reveals chromatin structure in mouse cortex

To show the feasibility of applying Droplet Hi-C to primary tissues, we used cortex tissues from 8-week-old mice and generated 6,235 high-quality single-cell chromatin profiles with a median of 175,021 unique read pairs per cell (with 22,456 cis-long contacts (>1 kb) and 9,954 trans contacts per cell, at 58% duplication rate) (Extended Data Fig. 2a–d). Despite the sparsity in contact signals, intrachromosomal interactions were enriched in the single-cell contact map (Extended Data Fig. 2e). When comparing Droplet Hi-C mouse cortex data with those of Dip-C³⁶ and sn-m3C-seq³⁷decay of contact probability by distance, chromatin compartments and topologically associating domain (TAD) boundaries showed consistent patterns across methods (Extended Data Fig. 2f–o). Droplet Hi-C yields a higher cis-long (>1 kb) interaction ratio than Dip-C, but lower than sn-m3C-seq (Extended Data Fig. 2g,h). Although Droplet Hi-C used Tn5 transposase for chromatin fragmentation in the nucleus, it showed minimal open chromatin biases (Extended Data Fig. 2k). Overall, these results indicated that Droplet Hi-C can reliably and robustly detect chromatin structures in complex tissue.

Chromatin interactions within the gene body have been shown to be positively associated with gene expression levels³⁸which can be used to cluster and annotate single-cell Hi-C data. We calculated the chromatin contacts on the gene body region of each gene as gene associating domain (GAD) scores³⁹and generated a cell-by-gene GAD score feature matrix. We then integrated the resulting matrix with the previous single-nuclei RNA-seq data of the mouse cortex⁴⁰ to perform co-embedding and predict cell identity for each Droplet Hi-C cell (Extended Data Fig. 3a and Supplementary Table 2). With this strategy, we successfully resolved 20 cell groups (5 non-neuronal, 9 glutamatergic and 6 GABAergic) (Fig. 1d,e and Extended Data Fig. 3b–d). GAD score variations across neuron types aligned with the expression levels of known cell type markers (Extended Data Fig. 3e–i). As validation, we integrated Droplet Hi-C data with sn-m3C-seq data and yielded comparable cell-type annotations (Extended Data Fig. 3j,k). Chromatin organization features, such as compartments, domains and loops, were distinctly visible in the aggregated single-cell profiles (Fig. 1f–h). For example, the glutamatergic neuron marker gene Satb2 showed specific long-range interactions with the region nearby Gm32949 gene and had elevated expression as well as an H3K27ac mark in ITL6GL (intratelencephalic projecting neuron, cortical layer 6) and PTGL (excitatory neuron, cortex PT) glutamatergic neurons (Fig. 1h). By contrast, in other cell types, this gene remained silenced as evidenced by the presence of the repressive H3K27me3 mark. To ensure accurate chromatin organization feature analysis, we quantified the minimum number of cells required. Down-sampling mouse cortex ITL23GL (intratelencephalic projecting neuron in cortical layer 2/3) neurons showed a notable drop in compartment score and insulation score correlations below 400 cells or fewer than 10 million long-range interactions (Supplementary Fig. 1a–c), which aligns with previous studies on large chromatin features such as compartments⁴¹. More reads are probably needed to confidently call and analyze chromatin loops.

To systematically probe the differences in chromatin structure across different cell types and elucidate their relationships with chromatin state and gene expression, we first examined differences in A and B compartments at 100 kb resolution. Compartment scores correlated with transcriptional activity, with 895 genomic regions showing variable compartment scores linked to chromatin states (Fig. 2a,b). We also identified cell-type-specific TAD boundaries. For example, Pdgfrawhich is essential for oligodendrocyte differentiation⁴²resides in an oligodendrocyte precursor cell (OPC)-specific chromatin domain boundary (Fig. 2c). We also found that genes with higher expression variation across cell types are more likely to be associated with variable TAD boundaries³³ (Fig. 2d,e).

Next, we examined chromatin loops at 10 kb resolution. We observed that active chromatin mark H3K27ac was enriched at loop anchors in each cell type, while repressive chromatin mark H3K27me3 was relatively depleted from the loop anchors (Fig. 2f). GREAT analysis⁴³ of VIPGA (CGE-derived neurogliaform cell Vip+)- and OGC (oligodendrocyte)-specific loop anchors showed enrichment for genes involved in neurotransmitters and myelination, respectively, corresponding to cell-type-specific functions (Fig. 2g).

Lastly, we developed an approach to identify multi-way (≥3) chromatin interactions (chromatin hubs)^44,45,46 in mouse cortical cells at single-cell resolution. To do this, we first calculated all genomic loci (10 kb bins) involved in multi-way interactions in each cell as well as their frequencies among all the cells in each cell type. Next, we defined chromatin hubs as the 10 kb bins with higher-than-expected contact probability in each cell type (Fig. 2h and Methods). Around 5% of the genomic bins were identified as chromatin hubs, with most being cell type specific (Extended Data Fig. 3l,m) and enriched at super-enhancers and marker genes from the corresponding cell type (Fig. 2i,j). These findings show the ability of Droplet Hi-C to robustly analyze chromatin architecture and cell-type-specific gene regulation in complex tissues.

Droplet Hi-C detects chromosomal aberrations in cancer cells

We hypothesized that single-cell Hi-C analysis could detect genomic variations in individual cancer cells and better resolve chromatin structure aberrations and ecDNA heterogeneity in them. As a proof of concept, we applied Droplet Hi-C on two colorectal cancer cell lines, COLO320DM and COLO320HSR, both carrying similar copies of the MYC amplicon. In COLO320DM, the MYC gene resides on ecDNA, whereas in COLO320HSR, MYC is part of HSRs. We inferred the copy number of each 1 Mb bin in individual cells and pseudo-bulk profiles, finding elevated copies of the MYC locus in both lines, as expected (Fig. 3a). Variations in genome-wide copy numbers were also observed (Fig. 3a). By using the inferred DNA copy number to correct bias in the contact matrices, we identified distinct SVs between these cell lines, including a duplication on chr6 (Fig. 3b).

It is challenging to differentiate MYC on ecDNA from HSR using whole-genome DNA sequencing alone. Tools such as AmpliconArchitect^47,48 sometimes misclassified the HSR as a circular amplicon. Although genome-wide interaction patterns have been proposed as ecDNA indicators⁴⁹aggregate Hi-C profiles showed similar interchromosomal interaction patterns for the MYC locus in both COLO320DM and COLO320HSR (Fig. 3c). Adjusted normalized interchromosomal interaction frequency (_adjnTIF), a method for identifying ecDNA⁴⁹also failed to distinguish the two forms of MYC amplicons (Fig. 3c).

We hypothesized that ecDNAs and HSRs show distinct three-dimensional genome architecture, which can be captured by single-cell Hi-C. In COLO320DM, we observed uniform interactions within ecDNA (Extended Data Fig. 4a), while COLO320HSR showed two deletions uniformly present in all cells (Extended Data Fig. 4a). Interaction hotspots for ecDNA and HSR across the genome (Methods) revealed that MYC ecDNA interacted more evenly with other chromosomes, while MYC HSR favored certain chromosomes (Fig. 3d). To quantify interchromosomal contact uniformity, we devised the hub index, which is the Gini coefficient of interchromosomal contacts for the 1 Mb MYC bin in each single cell. MYC ecDNA showed a significantly lower hub index than MYC HSR, indicating that ecDNAs are more evenly distributed (Fig. 3e). We also inferred MYC copy numbers (Fig. 3f) and measured the trans-to-cis contacting ratios (Fig. 3g), both showing significant differences between COLO320DM and COLO320HSR (Fig. 3f,g). COLO320DM had approximately 47% more MYC copies, consistent with a previous report⁵⁰ (Fig. 3f), and a higher trans-to-cis contacting ratio than COLO320HSR (Fig. 3g). These findings suggest that ecDNA and HSR show distinct chromosomal interaction patterns in individual cells and across cell populations, further supporting that single-cell Hi-C data could be used to distinguish ecDNAs from HSRs.

To investigate whether ecDNAs cluster into ‘hubs’, we shuffled the interchromosomal contacts to generate a uniform background distribution and found that COLO320HSR had a significantly higher hub index, consistent with the tandem repeat aggregation of HSR (Extended Data Fig. 4b). Although COLO320DM had a lower hub index, it was higher than the random background, suggesting that ecDNA tends to aggregate into hubs as well⁴⁴ (Extended Data Fig. 4b).

These differences in chromosomal interaction patterns provide a way to distinguish ecDNA from HSR using single-cell Hi-C data. To this end, we developed a multivariate logistic regression model to predict ecDNA based on the hub index, estimated copy number and trans-to-cis contact ratios (Extended Data Fig. 5a and Methods). We trained the model to classify MYC (chr8: 127–128 Mb) as ecDNA in COLO320DM and HSR in COLO320HSR cells. The model achieved 0.99 specificity, 0.97 precision, 0.89 accuracy and 0.71 sensitivity in MYC ecDNA prediction. Applying the model genome-wide, we identified two consecutive bins (chr8: 126–128 Mb) as ecDNA in COLO320DM, with 72.7% of cells predicted to be ecDNA positive, compared with only 20.5% in COLO320HSR cells (Extended Data Fig. 5b). In a control mouse cortex dataset, no ecDNA was detected. To enhance the accuracy of ecDNA identification in single-cell Hi-C data, we developed a deep learning model using convolutional neural networks (Fig. 3h, Extended Data Fig. 5c and Methods). This model achieved 0.80 sensitivity, 0.99 specificity, 0.93 accuracy and 0.99 precision in MYC ecDNA prediction (Extended Data Fig. 5d). Applying this deep learning model, we identified MYC ecDNA in 96% of COLO320DM cells and HSR in 99.54% of COLO320HSR cells (Fig. 3i and Supplementary Table 3). Given its superior performance, we used the deep-learning-based ecDNA caller model for all our subsequent analyses.

Droplet Hi-C uncovers ecDNA heterogeneity and evolution

Previous studies showed that glioblastoma (GBM) cells carrying the extrachromosomal oncogenic variant of the epidermal growth factor receptor (EGFRvIII) can develop drug resistance after treatment with tyrosine kinase inhibitors such as erlotinib (ER), which coincides with the loss of EGFR ecDNA⁵¹. To further characterize the dynamics of EGFR ecDNA abundance and heterogeneity during the acquisition of drug resistance, we used Droplet Hi-C to profile both untreated GBM39 cells and GBM39 cells under more than 30 days of erlotinib treatment (GBM39-ER) (Fig. 4a). With 9,204 cells profiled, we resolved five different clusters of cells based on the chromatin architecture (Fig. 4b and Extended Data Fig. 6a). After erlotinib treatment, a notable evolutionary shift was observed in the GBM39 cells. Subpopulation cluster 2 (C2) nearly vanished, subpopulation cluster 0 (C0) expanded and several new subpopulations including clusters 1, 3 and 4 (C1, C3 and C4) emerged (Fig. 4b–d).

At the sample level, elevated interchromosomal contacts were observed at the EGFR locus and, surprisingly, the MYC locus before treatment. After erlotinib treatment, interchromosomal contacts at the EGFR locus greatly diminished, but interchromosomal contacts at the MYC locus increased, along with the emergence of interchromosomal contacts at a new locus on chr12 encompassing MDM2 (Extended Data Fig. 6b). We applied the ecDNA caller to the Droplet Hi-C data to identify candidate ecDNAs in both GBM39 and GBM39-ER cells (Supplementary Table 3). In GBM39, we detected both EGFR ecDNA (ecEGFR) and MYC ecDNA (ecMYC) (Extended Data Fig. 6c, left). We also detected a low-frequency ecDNA on chr18 (1.5%, ecChr18) (Extended Data Fig. 6d). In GBM39-ER cells, we discovered ecDNA harboring the MDM2 locus (ecMDM2) and also ecMYC (Extended Data Fig. 6c, right). In addition, we identified MYC HSR in GBM39-ER cells by ecDNA caller (Extended Data Fig. 6e) and validated it using fluorescence in situ hybridization (FISH) (Extended Data Fig. 6f).

At the cluster level, we calculated the percentage of cells containing each ecDNA in every cluster, revealing the dynamics of ecDNA-harboring cancer cell populations during development of drug resistance (Fig. 4e). Specifically, ecEGFR disappeared nearly completely after erlotinib treatment (C1, C3, C4) as previously reported⁵¹whereas ecMDM2 appeared only in erlotinib-resistant cells and was dominant in GBM39-ER subgroup C3. In addition, ecMYC showed low frequency in GBM39 subgroup C2 and increased in transition state clusters C0 and C3. We also plotted the contact maps at 25 kb resolution for these three ecDNAs in every cluster. We observed that the structure and size of the ecMYC amplicon were altered, unlike ecEGFR and ecMDM2, suggesting a more rapid transformation at the ecMYC during drug treatment (Fig. 4e).

We also found coexistence of distinct ecDNA species in some cancer cells (Fig. 4f, top, and Extended Data Fig. 6g). In C0, a small population of cells (4.7%) harbored both ecMYC and ecEGFR, which can be validated by FISH imaging (Fig. 4f, bottom). These cells may potentially become drug resistant during erlotinib treatment, given that the gain of ecMYC may increase cellular fitness. To explore this hypothesis, we focused on the ecMYC and characterized its internal structure in GBM39 cells before and after erlotinib treatment by plotting the inferred copy number in each single cell, as the simultaneous increase in copy number can indicate that two adjacent genomic regions are co-present in the same ecDNA. Although the copy number varied among different cells, the boundaries of ecMYC in GBM39 were almost identical (Fig. 4g, left). After treatment, the ecMYC boundaries as well as the internal structure showed huge variability at single-cell resolution (Fig. 4g, right). When performing DNA FISH with a dual-color MYC break-apart probe set in GBM39-ER cells, while some cells showed co-amplification of both 5′ MYC and 3′ MYCmany cells showed only 5′ MYC amplification (Fig. 4h), which also indicated that ecMYC boundaries showed intricate transformations in GBM39-ER cells. Boundary variations in ecMYC did not show specificity to any cluster (Fig. 4g). We next selected the same number of cells from GBM39 and GBM39-ER within the transition cluster (C0) and compared the aggregated contact maps at ecMYC regions to find an intermediate ecMYC structure. Both contact maps retained their resemblance to the GBM39 and GBM39-ER cells rather than showing a similar intermediate structure, despite being from the same cluster (Ext ended Data Fig. 6h). Therefore, we reasoned that for GBM39 cells, erlotinib treatment probably induced the formation of a unique cell population containing distinct ecMYC, rather than selectively enriching a pre-existing cell population harboring ecMYC with growth advantage upon treatment.

Droplet Hi-C detects heterogeneity in primary tumor samples

The ecDNA is not only a biomarker for poor prognosis, but also a crucial driver in GBM⁵². Droplet Hi-C has successfully revealed ecDNA dynamics in GBM39 cells treated with drug. We further applied it to an isocitrate dehydrogenase-wildtype GBM tumor sample (Fig. 5a), clustering cells based on chromatin structure and segregating malignant from nonmalignant cells (Fig. 5b). In malignant cells, we detected expected chromosomal aberrations, such as deletions in chr10 and amplifications in chr7 (Fig. 5c), and the EGFR locus identified as ecDNA-like based on chromatin contact patterns and copy number variations (CNVs) (Fig. 5c–e and Extended Data Fig. 7a,b). Our ecDNA caller confirmed enrichment of ecEGFR in malignant cells (Fig. 5f and Supplementary Table 3). DNA FISH on frozen tissue sections further showed dispersed EGFR amplicons in the interphase nuclei (Fig. 5f, right bottom). Droplet Hi-C also identified SVs specific to the malignant population (Fig. 5g and Extended Data Fig. 7c), including translocation of the tumor suppressor gene IKZF1 to ecEGFR without its promoter, leading to its decreased transcription (Fig. 5g). The VOPP1 locus on ecDNA showed the strongest interaction with IKZF1suggesting their fusion, but it had no discernible effect on VOPP1 gene transcription (Fig. 5g). We observed notable changes in chromatin architecture in malignant cells at the compartment level, correlating with gene expression variations (Extended Data Fig. 7d). For example, a compartment switch from an inactive B to an active A compartment at the SOX2-OT locus coincided with increased transcription in the malignant cells. Droplet Hi-C effectively detected CNVs, ecDNAs, SVs and chromatin architecture specific to malignant cells in the primary GBM sample.

Previous studies using single-cell RNA-seq identified four major cellular states in malignant glioma cells: neural progenitor-like (NPC-like), OPC-like, astrocyte-like (AC-like) and mesenchymal-like (MES-like). However, the chromatin structure heterogeneity underlying these cell states remains unclear⁵³. To further recapitulate different cellular states and identify intra-tumoral heterogeneity of chromatin architecture, we used k-nearest neighbors to assign malignant cells to one of the four GBM cellular states based on co-embedding with single-nucleus RNA-seq (snRNA-seq) data generated from the same sample (Fig. 5h,i). Correlation analysis using compartment scores showed a similar cell state hierarchy compared with transcriptome data and previous studies, in which the MES-like state showed higher similarity to the AC-like state, while the OPC-like and NPC-like states were more similar (Extended Data Fig. 7e)⁵⁴.

Chromatin compartment analysis showed that state-specific gene expression aligned with chromatin architecture. For example, metastasis suppressor CSMD1 was in an active compartment in the OPC-like and NPC-like states and in an inactive compartment in the MES-like and AC-like states, mirroring gene expression differences observed from snRNA-seq data (Extended Data Fig. 7f,g). Comparing compartment differences between MES-like and OPC-like, we identified 1,782 switched compartments (Extended Data Fig. 7h), which are significantly associated with differential gene expression (Extended Data Fig. 7i,j) and state-specific biological functions (Extended Data Fig. 7k). In addition to compartment changes, we also observed insulation score alterations and loss of domain boundaries in malignant cells (Extended Data Fig. 7l). Such disruption can rewire enhancer–promoter interactions, contributing to the GBM progression⁵⁵.

We also characterized ecDNA heterogeneity across different GBM cellular states. While the fraction of cells containing ecEGFR was similar across states (Fig. 5j), AC- and MES-like cells had higher ecEGFR copy numbers (Fig. 5k), corresponding to increased EGFR gene expression and chromatin accessibility (Fig. 5l,m). Furthermore, by defining ecDNA boundary, we found that expression levels of genes on the ecDNA varied with cellular states, and that such correlation could not be fully explained by ecDNA copy number, indicating a complex regulatory mechanism governing ecDNA genes expression among cellular states (Fig. 5n). Consistent with previous reports⁵³the expression level of EGFR gene showed the highest correlation with AC-like state scores (Fig. 5n). Other genes on the ecEGFR, such as SEC61G and LANCL2showed preferential expression in the MES-like state, or the AC- and MES-like states, respectively (Fig. 5o).

To further show the utility of Droplet Hi-C in clinical tumor samples, we applied it to bone marrow mononuclear cells from a patient with myelodysplastic syndrome and secondary acute myeloid leukemia (AML) before and after treatment with azacitidine (a DNA methyltransferase inhibitor) and venetoclax (an inhibitor of the anti-apoptotic protein BCL2) (Extended Data Fig. 8a). Pretreatment bone marrow mononuclear cells harbored a large (~5 MB) ecDNA containing the MYC gene. After treatment, the ecMYC disappeared (Extended Data Fig. 8b,c), consistent with the patient’s remission. Besides, the proportion of malignant cells also diminished greatly after treatment (Extended Data Fig. 8c). The long-range interactions between MYC and blood enhancer cluster, which activate MYC expression in AML⁵⁶were detected pretreatment but disappeared after therapy, suggesting a role in tumor progression (Extended Data Fig. 8d,e). Thus, in this patient’s tumor, ecDNA appears to promote MYC expression not only through increased DNA copy numbers but also through cancer-specific enhancer–promoter interactions.

In summary, Droplet Hi-C enables the detection of ecDNA dynamics in tumors before and after treatment. Beyond ecDNA, it also identifies SVs and chromatin architecture, shedding light on the regulatory programs driving tumor progression and drug resistance. Droplet Hi-C offers substantial potential for understanding tumor evolution and treatment response.

Joint profiling of chromatin architecture and transcriptome

Single-cell joint profiling of chromatin organization and transcriptome helps investigate the relationships between gene expression and chromatin architecture. We modified the Droplet Hi-C protocol to work with the 10x Genomics Chromium Single Cell Multiome kit, creating Paired Hi-C, which simultaneously profiles RNA and Hi-C from the same single nuclei (Fig. 6a and Supplementary Fig. 2). This protocol uses milder SDS treatment to reduce RNA degradation and a lower formaldehyde concentration for better complexity of chromatin contact profiles and revised the cDNA library preparation to recover more cDNA molecules. Paired Hi-C offers higher throughput and efficiency compared with other microwell- or combinatorial-indexing-based single-cell joint Hi-C and RNA-seq assays (Supplementary Fig. 2).

We first validated Paired Hi-C by performing a species-mixing experiment with HeLa S3 and mESC, successfully separating human and mouse cells based on Hi-C (HeLa S3: 3,928, mESC: 3,301, total: 7,677) and RNA (HeLa S3: 3,999, mESC: 3,293, total: 7,677) data (Extended Data Fig. 9a,b). Applying Paired Hi-C to the adult mouse cortex sample, we obtained 12,361 joint single-cell profiles with 42,210 read pairs per cell for the Hi-C modality (duplication rate, 66.2%), and a median of 3,914 unique molecular identifiers (UMI) per cell and 1,746 genes per cell for the RNA modality (duplication rate, 74%; Extended Data Fig. 9c). We successfully identified 20 cell types using the Paired Hi-C RNA data, including 9 excitatory neuron types, 6 inhibitory neuron types and 5 non-neuronal types (Fig. 6b and Supplementary Table 4). Cell-type annotation was validated by integration with other snRNA-seq or single-cell RNA-seq (scRNA-seq) datasets and expression patterns of marker genes (Extended Data Fig. 9d–f).

Although single-cell three-dimensional genome features were relatively limited in Paired Hi-C, compartment scores strongly correlated with Droplet Hi-C and sn-m3C-seq profiles from the same cell types (Fig. 6c,d and Extended Data Fig. 9g). We detected concordant changes of compartment and gene expression at cell-type-specific marker genes (Fig. 6c–e). For example, Erbb4 gene was in the A compartment in PVGA (MGE-derived neurogliaform cell etc+) and VIPGA GABAergic neurons, but they were placed in the B compartment in OPC non-neuronal types and ITL5GL (intratelencephalic projecting neuron, cortical layer 5) glutamatergic neurons, correlating with higher Erbb4 expression in GABAergic neurons (Fig. 6c). We also used transcriptome information from Paired Hi-C and co-embedded it with Droplet Hi-C. The results agree well with the cell annotations based on the Droplet Paired-Tag⁴⁰ (Extended Data Fig. 9h). As such, we can augment the number of chromatin contacts for each cell type by combining Paired Hi-C and Droplet Hi-C.

We also applied Paired Hi-C to human peripheral blood mononuclear cells (PBMCs) and obtained 7,585 high-quality single-cell profiles. The transcriptome data were benchmarked against other methods such as 10x Genomics Multiome and DOGMA-seq⁵⁷showing comparable performance (Extended Data Fig. 9i,j). Using the 10x Genomics Multiome RNA as the reference, we performed label transfer to predict cell types in the Paired Hi-C RNA modality, and filtered out cells whose identities could not be accurately assigned (Extended Data Fig. 9k). Marker gene expression profiles confirmed the accuracy of annotation (Extended Data Fig. 9l). We performed co-embedding of Droplet Hi-C with Paired Hi-C profiles to compensate for the Hi-C sparsity in Paired Hi-C data, and analyzed the chromatin architecture differences between the T cells and monocytes. T cells showed strong chromatin loops between COG6 and FOXO1 gene loci, correlating with higher transcription of these genes compared with monocytes (Extended Data Fig. 9m,n).

Paired Hi-C enables joint analysis of chromatin structure and gene expression at single-cell resolution, offering insights into the regulation of gene activation or repression. We applied it to erlotinib-treated GBM39 cells to study how changes in chromatin structure, particularly ecDNA, affect gene expression. Paired Hi-C linked ecDNA structure with gene expression in single cells, showing that gene expression generally correlates with ecDNA copy number (Fig. 6f). Genes at the 5′ end of the variable ecMYC boundaries such as CASC8 and PCAT1 showed a strong copy number–expression correlation (Fig. 6f,g), while genes at the 3′ end, such as the CCDC26 gene, showed high expression despite low copy numbers (Fig. 6f,g). This indicates a complex relationship between ecDNA structure and gene expression.

In addition, we observed different trans-interaction patterns of ecMYC in GBM39 and GBM39-ER cells that were associated with distinct transcriptional responses. Genes with GBM39-specific ecMYC interactions were more highly expressed in GBM39 cells, while GBM39-ER-specific ecMYC interacting genes were more highly expressed in GBM39-ER cells (Extended Data Fig. 10a). The higher expression levels were probably not caused by the elevation in the copy number of interacting regions (Extended Data Fig. 10b). This observation is consistent with a potential regulatory role of ecDNA in global transcription as suggested previously⁴⁹.

In the sample from the patient with GBM, ecEGFR showed higher copy numbers and gene expression levels in AC-like and MES-like cells, possibly driving changes in cellular states under erlotinib treatment. We classified GBM39 and GBM39-ER cells into four GBM cellular states using Paired Hi-C RNA-seq data and found a reduction in the differentiated-like population after erlotinib treatment (Fig. 6h). As found in the GBM tumor sample, EGFR expression was higher in AC-like and MES-like cells (Fig. 6i). Interestingly, MDM2 expression was enriched in the progenitor states (OPC- and NPC-like), which became more prevalent after treatment (Fig. 6i). These results highlight the unique interplay among ecDNA dynamics, cellular states and drug resistance.

Erlotinib treatment also caused dramatic shifts in chromatin architecture and gene expression changes in GBM39 cells. A total of 1,066 A compartments switched to B compartments, and 2,796 B compartments shifted to A compartments, with correlated changes in gene expression (Extended Data Fig. 10c,d). Genes in compartments that changed from A to B showed decreased expression, while those in compartments that shifted from B to A showed increased expression (Extended Data Fig. 10d).

In summary, Paired Hi-C can directly link chromatin reorganization and structural alterations, such as ecDNA, to gene expression. This tool should offer valuable insights into gene regulatory mechanisms in development and disease, particularly in understanding tumor progression and drug resistance.

Recent advances in single-cell Hi-C methods have greatly expanded their applications. However, existing techniques still face challenges in analyzing heterogeneous tissues and tumor biopsies. Our Droplet Hi-C method, using a commercially available microfluidic platform, provides high-throughput single-cell Hi-C assays with minimal hands-on time and lower costs. Applying Droplet Hi-C to the adult mouse cortex, we identified cell-type-specific chromatin structures and correlated them with epigenetic modifications. In addition, we detected changes in chromatin organization, CNVs, SVs and ecDNAs in cancer cells and patient tumor samples, particularly during drug treatment. We further developed Paired Hi-C, which profiles both chromatin architecture and transcriptome in single cells, enabling the study of gene expression in relation to three-dimensional genome structure. Although the Hi-C complexity in Paired Hi-C is lower than in Droplet Hi-C, combining data from both methods allowed us to resolve and annotate cell types while improving chromatin contact resolution.

Our droplet-based single-cell Hi-C methods offer advantages in scalability, speed, cost and adaptability. However, improvements are needed, particularly in the library complexity of Paired Hi-C. In addition, while Droplet Hi-C has a lower ratio of cis-long (>1 kb) interactions than some in situ Hi-C methods (Extended Data Fig. 2h), it performs similarly to or better than other single-cell Hi-C methods that do not enrich ligated DNA fragments (Extended Data Fig. 2h). While cis-short interactions (<1 kb) provide limited information on chromatin contacts, they are valuable for analyzing CNVs and SVs in cancer cells. Biotin enrichment of cis-long interactions could improve chromatin architecture capture efficiency and reduce sequencing costs.

The ecDNA is prevalent in cancer, often carrying oncogenes^25,26. Current single-cell ecDNA detection methods fall into two categories: computational tools using whole-genome sequencing data from a next-generation sequencing⁴⁷ or a third-generation sequencing platform^58,59 and Circle-seq^60,61,62which retains circular DNA by degrading linear DNA. However, distinguishing ecDNAs from HSRs remains challenging. Here we develop an ecDNA detection algorithm that uses single-cell Hi-C data to reliably identify ecDNAs and effectively distinguish them from HSRs. Our deep-learning-based ecDNA caller can accurately detect ecDNA in single cells and calculate its proportion in the population. By analyzing intra-ecDNA contacts, we can pinpoint ecDNA boundaries. For instance, we identified ecMYC boundary variations in GBM39-ER cells.

Pan-cancer studies have revealed ecDNAs in many cancer types and their association with poor clinical outcomes^{63,64,65,66,67}. Furthermore, tumors with ecDNA show unique therapeutic vulnerabilities^68,69. Thus, given the independent prognostic value of ecDNA and its potential to be targeted, recent calls have been made to include ecDNA profiling in clinical specimens⁷⁰. Our Droplet Hi-C and ecDNA caller offer a reliable toolkit detecting ecDNA in clinical samples, as shown in GBM and AML tumors. This toolkit is particularly valuable for heterogeneous samples, in which ecDNA structures and copy numbers evolve during treatment. Importantly, our ecDNA caller can distinguish between amplifications on ecDNA and HSRs, which may have important therapeutic implications. In our study, we observed ecEGFR disappearance, ecMDM2 emergence and ecMYC structural changes in GBM39 cells following treatment with erlotinib, a tyrosine kinase inhibitor commonly used in the treatment of EGFR-positive tumors. The poor response to erlotinib in GBM may be linked to the ecDNA evolution. Similarly, in a sample from a patient with AML, ecMYC was no longer detectable after treatment with azacitidine and venetoclax. Our methods hold promise for advancing ecDNA research in tumor diag nosis and treatment.

Experimental protocol for Droplet Hi-C

A brief description of the Droplet Hi-C experimental procedure is provided below.

Fixation

Cells were cross-linked in 1% formaldehyde, which was diluted from 37% formaldehyde with 1× PBS (pH = 7.4), and incubated at room temperature for 10 min. After cross-linking, the reactions were quenched in 200 mM glycine and incubated at room temperature for 5 min. Quenched reactions were spun down at 1,000 × g for 5 min at 4 °C and resuspended using 1% bovine serum albumin (BSA) in 1× PBS (pH = 7.4) to wash twice. One million cells were aliquoted into each tube. The cells were spun once again at 1,000 × g for 5 min, the supernatant was removed, and the pellet was flash frozen in liquid nitrogen and finally stored indefinitely at −80 °C.

In situ Hi-C

Cell pellets were lysed with pre-cold 300 μl lysis buffer (10 mM Tris–HCl, pH 8.0 (Thermo Fisher Scientific, 15568025), 10 mM NaCl (Sigma, S5150), 0.2% IGEPAL CA-630 (Sigma, I8896) and 1× protease inhibitor (Roche, 5056489001)) on ice for 45 min, centrifuged at 1,000 × g for 5 min at 4 °C to collect nuclei, and washed once with 200 μl lysis buffer. The nuclei were resuspended with 50 μl 0.5% SDS and incubated at 62 °C for 10 min on Thermomixer. Then, we added 145 μl of nuclease-free H₂O and 25 μl 10% Triton X-100 (Sigma, 93443) to quench SDS and incubated samples at 37 °C for 15 min on Thermomixer at 300 rpm. To the samples were added 27 μl of 10× CutSmart Buffer and three restriction enzymes, including 50 U DpnII (NEB, R0543L), 62.5 U MboII and 7.5 U NlaIII (NEB, R0125L); then, the samples were incubated at 37 °C for 90 min on Thermomixer at 550 rpm. The enzymes were deactivated at 65 °C for 20 min and then cooled down to room temperature. The nuclei were collected by centrifugation at 1,000 × g for 5 min at 4 °C and washed once with 200 μl ligation buffer (100 μl 10× T4 DNA ligase buffer (NEB, B0202S), 5 μl 20 mg ml⁻¹ BSA (NEB, B9000S), 865 μl H₂O); the ligation reaction was performed with 200 μl ligation buffer and 20 μl T4 DNA ligase (NEB, M0202L) at 37 °C for 40 min on Thermomixer at 300 rpm.

Single-cell Hi-C sequencing library construction

The ligated nuclei pellets were resuspended in 1 ml 1% BSA in PBS (diluted from 10% BSA in PBS (Sigma, A1595) using 1× PBS (pH = 7.4)) in each tube, 1 μl 1,000× 7-AAD (Invitrogen, A1310) was added and the nuclei were sorted by fluorescence-activated nuclei sorting with an SH800 cell sorter (Sony) for the isolation of single nuclei (Supplementary Fig. 3). The nuclei were collected in collection buffer (5% BSA in PBS) at 5 °C and immediately centrifuged for 5 min at 1,000 × g and 4 °C to collect the nuclei. The nuclei were washed once with 1× Nuclei Buffer (diluted from 20× Nuclei Buffer (10x Genomics, PN-2000207)), resuspended in 10 μl 1× Nuclei Buffer and counted on an RWD C100-Pro fluorescence cell counter with DAPI staining. According to the desired targeted nuclei recovery number, 3,000–15,000 nuclei were aliquoted into PCR tubes, to which was added Transposition Mix following the user’s guide for the Chromium Next GEM Single Cell ATAC Reagent Kits v1.1 or the Chromium Next GEM Single Cell ATAC Reagent Kits v2 (10x Genomics). The incubation time for tagmentation is 60 min for both the v1.1 and v2 kits. We also extended the index PCR elongation time from 20 s to 1 min. The double-sided size selection was changed to 1.14× SPRIselect to remove only small fragments.

All sequencing experiments were performed with an Illumina NextSeq 2000 or NovaSeq 6000 sequencer.

Experimental protocol for Paired Hi-C

A brief description of the Paired Hi-C experimental procedure is provided below.

Fixation

Cells were cross-linked in 0.6% formaldehyde, which was diluted from 37% formaldehyde with 1× PBS (pH = 7.4), and incubated at room temperature for 10 min. After cross-linking, the reactions were quenched in 200 mM glycine and incubated at room temperature for 5 min. The quenched reactions were spun down at 1,000 g for 5 min at 4 °C and resuspended using 1% BSA in 1× PBS (pH = 7.4) to wash twice, and then 1 million cells were aliquoted into each tube. The cells were spun once again at 1,000 × g for 5 min, the supernatant was removed, and the pellet was flash frozen in liquid nitrogen and finally stored indefinitely at −80 °C.

In situ Hi-C

Cell pellets were lysed with pre-cold 300 μl lysis buffer (10 mM Tris–HCl (pH 8.0; Thermo Fisher Scientific, 15568025), 10 mM NaCl (Sigma, S5150), 0.2% IGEPAL CA-630 (Sigma, I8896), 1× protease inhibitor (Roche, 5056489001), 1 U μl⁻¹ RNaseOUT (Invitrogen, 10777-019) and 1 U μl⁻¹ SUPERaseIn inhibitor (Invitrogen, AM2694)) on ice for 45 min, then centrifuged at 1,000 × g for 5 min at 4 °C to collect nuclei and washed once with 200 μl lysis buffer. The nuclei were resuspended with 50 μl 0.5% SDS, which was diluted from 10% SDS using 1× PBS (pH = 7.4) with 1 U μl⁻¹ RNaseOUT and 1 U μl⁻¹ SUPERaseIn inhibitor, and incubated at 37 °C for 60 min on Thermomixer at 300 rpm. Then, we added 197 μl quenching buffer (137.59 μl of nuclease-free H₂O, 25 μl 10% Triton X-100, 27 μl 10× CutSmart, 1 U μl⁻¹ RNaseOUT and 1 U μl⁻¹ SUPERaseIn inhibitor) and incubated samples at 37 °C for 15 min on Thermomixer at 300 rpm. The samples were centrifuged at 1,000 × g for 5 min at room temperature, the supernatant was removed and the pellets were resuspended with 250 μl digestion buffer (1× CutSmart Buffer, 1 U μl⁻¹ RNaseOUT, 1 U μl⁻¹ SUPERaseIn inhibitor, 50 U DpnII, 62.5 U MboII and 5 U NlaIII), and then incubated at 37 °C for 90 min on Thermomixer at 550 rpm. The enzymes were deactivated at 65 °C for 20 min and then cooled to room temperature. The nuclei were collected by centrifugation at 1,000 × g for 5 min at 4 °C and washed once with 200 μl ligation buffer (827.5 μl H₂O, 100 μl 10× T4 DNA ligase buffer, 5 μl 20 mg ml⁻¹ BSA, 25 μl SUPERaseIn inhibitor and 12.5 μl RNaseOUT), and the ligation reaction was performed with 200 μl ligation buffer and 20 μl T4 DNA ligase at 37 °C for 40 min on Thermomixer at 300 rpm.

PBMCs (Allcells, LP, CR, MNC, 100 M) were cross-linked in 1% formaldehyde. Then, 1 million cell pellets were lysed with 200 μl of high-salt lysis buffer 1 (50 mM HEPES (pH 7.4), 2 mM EDTA (pH 8.0), 140 mM NaCl, 0.25% Triton X-100, 0.5% IGEPAL CA-630, 10% glycerol and 1× proteinase inhibitor cocktail) and incubated on ice for 10 min. After this, the cells were pelleted at 500 g for 2 min at 4 °C and then resuspended in 200 μl of high-salt lysis buffer 2 (10 mM Tris–HCl (pH 8), 3 mM EDTA, 200 mM NaCl, 1× proteinase inhibitor cocktail). The solution was incubated on ice for 10 min. Following this, the cells were then pelleted at 500 g for 2 min at 4 °C and then resuspended in 200 μl of 1× T4 DNA ligase buffer containing 0.2% SDS. The cells were then incubated at room temperature for 1 h. To quench the reaction, 200 μl of 1× T4 DNA ligase buffer and 10 μl 10% Triton X-100 were added to the tube. Finally, the cells were spun at 500 g for 4 min at 22 °C. The pellets were resuspended with 250 μl digestion buffer (1× T4 DNA ligase buffer, 1 U μl⁻¹ RNaseOUT, 1 U μl⁻¹ SUPERaseIn inhibitor, 80 U DpnII) and then incubated at room temperature for 2 h. The nuclei were collected by centrifugation at 500 g for 4 min at 4 °C and washed once with 200 μl ligation buffer (857.5 μl H₂O, 100 μl 10× T4 DNA ligase buffer, 5 μl 20 mg ml⁻¹ BSA, 25 μl SUPERaseIn inhibitor and 12.5 μl RNaseOUT), and the ligation reaction was performed with 200 μl ligation buffer and 20 μl T4 DNA ligase at room temperature for 40 min.

Single-cell sequencing library construction

The ligated nuclei pellets were resuspended in 1 ml 1% BSA in PBS (diluted from 10% BSA in PBS using 1× PBS (pH = 7.4)) with 1 U μl⁻¹ RNaseOUT and 1 U μl⁻¹ SUPERaseIn inhibitor in each tube, 1 μl 1,000× 7-AAD (Invitrogen, A1310) was added, and the nuclei were sorted by fluorescence-activated nuclei sorting with an SH800 cell sorter (Sony) for the isolation of single nuclei. The nuclei were collected in collection buffer (5% BSA in PBS with 5 U μl⁻¹ Recombinant RNasin (Promega, PAN2515)) at 5 °C and immediately centrifuged for 5 min at 1,000 × g and 4 °C to collect the nuclei. The nuclei were washed once with 1× Nuclei Buffer (diluted from 20× Nuclei Buffer (10x Genomics, PN-2000207) with 1 mM DTT and 1 U μl⁻¹ Recombinant RNasin), resuspended in 10 μl 1× Nuclei Buffer and counted on an RWD C100-Pro fluorescence cell counter with DAPI staining. According to the desired targeted nuclei recovery number, 3,000–15,000 nuclei were aliquoted into PCR tubes, to which was added Transposition Mix following the user’s guide of Chromium Next GEM Single Cell Multiome ATAC + Gene Expression (10x Genomics). The tagmented nuclei were loaded onto a Chromium Next GEM Chip J for droplet generation with a Chromium X microfluidic system (10x Genomics). The gel bead-in-emulsions (GEMs) were collected, reverse transcription and cell barcoding were performed using a thermal cycler and the reaction was quenched following the user’s guide of the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression (10x Genomics). After GEM cleanup and pre-amplification, the SPRIselect products were eluted into 100 μl elution buffer (Qiagen, 19086) instead of 160 μl. For scHi-C library construction, we extended the index PCR elongation time from 20 s to 1 min, and the double-sided size selection was changed to 1.55× SPRIselect to remove only small fragments. For scRNA-seq library construction, we used 0.8× SPRIselect to purify cDNA twice instead of 0.6× SPRIselect to keep short cDNA fragments.

All sequencing experiments were performed with an Illumina NextSeq 2000 or NovaSeq 6000 sequencer.

Processing of Droplet Hi-C data

Custom scripts for demultiplexing, mapping and extracting single-cell contacts from Droplet Hi-C data were publicly available at https://github.com/Xieeeee/Droplet-Hi-C. First, depending on the indexing primers used, Droplet Hi-C fastq files were demultiplexed using Illumina bcl2fastq (v2.19.0.316) or 10x Genomics cellranger-atac mkfastq (v2.0.0). After demultiplexing, a custom script was used to extract the cellular barcode sequence from Read2, and barcodes were aligned to the white list provided by 10x Genomics using Bowtie (v1.3.0)⁷². Aligned barcode sequences were appended to the beginning of the read name to record the cellular identity of each read. Next, sequencing adapters were detected and trimmed with Trim-Garole (v0.6.10)⁷³. Cleaned reads were then mapped to the human (hg38) or mouse (mm10) reference genome using BWA-MEM (v0.7.17)⁷⁴with arguments ‘-SP5M’ specified. Finally, after mapping, valid contacts were parsed, sorted and dedupli cated by Pairtools (v0.3.0)⁷⁵with barcode information stored in a separated column in the pairs file. Contacts were balanced and stored in cool format using Cooler (v0.8.10) for visualization and downstream analysis⁷⁶. High-quality cells were selected based on the number of total unique contacts per cell in each library.

Analysis of Droplet Hi-C data

Here we describe the general analysis strategy and workflows for our Droplet Hi-C data. Dataset-specific manipulations in the paper, if any, will be indicated.

Visualization of Hi-C contact matrices

Bulk, pseudo-bulk or single-cell contact matrices in cool format were visualized and plotted using Cooler along with Matplotlib (v3.5.1)⁷⁷.

Signal enrichment calculation

Density plots and signal enrichment over Center for Epigenomics of the Mouse Brain Atlas candidate cis-regulatory elements (cCREs) were generated using bamCoverage from deepTools (v3.5.3). Peaks that overlapped with ENCODE blacklist (v2) regions were removed during the calculation of enrichment⁷⁸.

Embedding and clustering of single-cell Droplet Hi-C dataset on cultured cells

We used Higashi to infer low-dimensional cell embeddings for all our Droplet Hi-C datasets without imputations, including a human cell line mix (HeLa–GM12878–K562; Extended Data Fig. 1b), GM12878–WTC-11 (Extended Data Fig. 1i), GBM39–GBM39-ER (Fig. 4b) and the sample from the patient with GBM (Fig. 5b). For visualization, the L2 norm of cell embeddings was projected to a two-dimensional space with uniform manifold approximation and projection (UMAP). To identify cells with similar identity, we performed Leiden clustering using igraph (v0.9.9) and leidenalg (v0.8.8)^79,80.

Contact frequency by distance analysis

We used the ‘expected_cis’ module from cooltools (v0.5.1) to calculate the average contact frequency as a function of genomic separation⁸¹. To plot the curve, we performed smoothing and aggregation among chromosome regions.

Imputation of single-cell chromatin contacts on the adult mouse cortex Droplet Hi-C dataset

For the adult mouse cortex Droplet Hi-C data, we used the scHiCluster (v1.3.4) to perform contact matrix imputation for individual cells at three different resolutions: 100 kb (for cell embedding and visualization), 25 kb (for domain boundary calling) and 10 kb (for cell embedding and loop calling)⁸². In brief, contacts from individual cells were first masked for ENCODE blacklist regions⁷⁸. scHiCluster then performed linear convolution and random walk with restart to impute the sparse single-cell contact matrices for each chromosome. Considering storage efficiency, we output the imputed results for the whole chromosome at 100 kb, for the 10.05 Mb window at 25 kb and for the 5.05 Mb window at 10 kb resolution for each cell. For cancer cells or tumor samples, we did not perform imputation owing to CNVs and SVs.

Embedding and annotation of single-cell Droplet Hi-C dataset on the adult mouse cortex

For the imputed adult mouse cortex datasets and PBMC datasets, we adopted a previously described method to co-embed Hi-C data with gene expression data (snRNA-seq) generated from the same tissue and to assign cell-type identities³⁹. Briefly, single-cell GAD (scGAD) score R_ij is calculated as the raw number of interactions at the gene body region for gene i in cell j with the 10 kb imputed matrix. This method has been implemented in scHiCluster as the ‘gene_score’ module. By this, the single-cell chromatin interaction profiles can then be represented as a cell-by-gene scGAD score matrix, which serves as the input for integration with snRNA-seq data.

To perform integration, we began by normalizing and selecting the top 2,000 variable genes in reference snRNA-seq data. These genes were subjected to dimension reduction by principal component analysis (PCA) using Scanpy (v1.7.2), retaining 30 principal components⁸³. The derived PCA model was then applied to transform the scGAD score matrix. Subsequent integration was conducted with a tailored version of Seurat integration using canonical correlation analysis⁸⁴. For the visualization of co-embedded data, we calculated the nearest neighbor graph (k = 25) and UMAP embedding with Scanpy.

To annotate cell identities within the Droplet Hi-C dataset, we identified the 15 nearest snRNA-seq neighbors for each cell via PyNNDescent (v0.5.6) using Euclidean distances⁸⁵. These distances were scaled and converted into standardized scores so that the total score is added up to 1, and the neighbor with the minimal distance would have the highest score. The identity of each Droplet Hi-C cell was inferred by nominating the cell type label that garnered the highest standardized score among its top nearest neighbors.

To validate the cell type annotations, co-embedding of Droplet Hi-C and sn-m3C-seq datasets was performed using scHiCluster, which facilitates the low-dimensional representation of chromatin contact information. For each cell, imputed matrices at 100 kb resolution were binarized and flattened. The binarized data from individual cells in both datasets were then concatenated into a larger matrix. PCA was applied to this matrix on a per-chromosome basis. The PCA results for each chromosome were concatenated for a second round of PCA to generate the final cell embeddings. Batch effects were corrected using harmonypy (v0.0.9)⁸⁶. After batch correction, Leiden clustering was used to identify co-embedding clusters. The validity of the original annotation was verified by calculating the overlap coefficients (O_i,j) between the clusters and the original annotations from Droplet Hi-C (A) and sn-m3C-seq dataset (B): ({O}_{i,j}=min ([frac{{A}_{i}cap {C}_{k}}{{A}_{i}}],,max [frac{{B}_{j}cap {C}_{k}}{{B}_{j}}]))where i indicates the query cell type from Droplet Hi-C, j indicates the reference cell type from sn-m3C-seq and k indicates the co-embedding cluster.

Analysis of A and B compartments

On sample levels, we used the ‘eigs_cis’ module from cooltools to perform eigen decomposition on balanced contact matrices and calculate the compartment score at 100 kb resolution⁸¹. The orientation of the resulting eigenvectors was adjusted to correlate positively with the CpG density of the corresponding 100 kb genomic bins, thereby determining the sign of the compartments (A and B). Similarity between samples is calculated as the Spearman’s correlation coefficient (SCC) of compartment scores among all autosomes (Extended Data Fig. 1c,j). To illustrate chromatin interactions within and between compartments, we used the ‘saddle’ module from cooltools to calculate the average observed versus expected contact frequency, categorized by compartment scores.

To generate pseudo-bulk profiles for distinct cell types or clusters within a sample, we adopted a uniform approach outlined before to minimize bias and facilitate direct comparisons. Initially, the raw contact matrices at the sample level were normalized for distance effects. Next, Pearson correlation coefficients were computed on the distance-normalized matrices, and PCA was then performed on the correlation matrix. The model fitted at the sample level was used to transform the raw contact matrices of various cell types or clusters. To compare the similarity between cell types or clusters, we calculated the SCC of the compartment score using the compartment that intersected with variable gene regions (Fig. 1e and Extended Data Fig. 7e). To identify differential compartments, we used dchic (v2.1)⁸⁷selecting only those with an adjusted P value < 0.01 and a Manhattan distance greater than the 2.5th percentile of the standard normal distribution for further downstream analysis.

For joint analysis of compartment and cell-type-specific histone modifications, we used the recently published adult mouse cortex Droplet Paired-Tag dataset⁴⁰. We retained only cell types that contain>100 cells and were shared between datasets to ensure comparability among the analyses. For all differential compartments, we calculated the Pearson correlation coefficients between the compartment score and the signal of H3K27ac or H3K27me3 (counts per million mapped reads (CPM)) within the same 100 kb genomic bins among various cell types.

Analysis of chromatin domains

On sample levels, we used the ‘insulation’ module from cooltools to compute the insulation score for raw contact matrices at 25 kb resolution. To delineate cell-type-specific variable domain boundaries in mouse cortex data, we used TopDom (v0.0.2) on 25 kb resolution imputed contact matrices for each individual cell⁸⁸. We defined the ‘boundary probability’ for a given bin as the proportion of cells within a particular cell type that designate the bin as a domain boundary. The presence or absence of a domain boundary is summarized by n cell types into an n × 2 contingency tables. We then computed the chi-square statistics and P value for each bin, and the domain boundaries were classified as variable if they showed a false discovery rate (FDR) <0.001 and a boundary probability difference>0.05.

For correlation analysis of gene expression surrounding variable domain boundaries, we first identified genes within a 100 kb window centered on the variable domain boundaries. We then calculated Pearson correlation coefficients to quantify the relationship between boundary probabilities and expression levels (reads per kilobase per million mapped reads (RPKM)) of these genes from Droplet Paired-Tag data among shared cell types. The genes were further categorized into ‘housekeeping’, ‘constant’ and ‘variable’. In brief, the housekeeping gene list was taken from a previous report⁸⁹. The top 2,500 variable genes from Droplet Paired-Tag RNA modality were selected as the variable group. Genes with a similar expression level as the top 2,500 variable genes but were not within the ‘housekeeping’ or ‘variable’ group were treated as ‘constant’.

Analysis of chromatin loops

For loops calling on sample levels, we adopted the ‘dots’ module in cooltools, which implements the principle of the commonly used HiCCUPS loop calling strategy, to call loops on the 10 kb raw contact matrices⁹⁰.

A modified version of the SnapHiC workflow implemented in scHiCluster is used to perform loops calling on the 10 kb imputed contact matrices for mouse cortex cell types⁹¹. To compare histone modification enrichment at loop anchors, we calculate H3K27ac or H3K27me3 CPM among all cell types at all loop anchors identified. When histone modification profiles and loop anchors are from the same cell types, the enrichment is classified as ‘matched’; otherwise, it is classified as ‘unmatched’.

Gene Ontology (GO) annotation of loop anchors was performed using rGREAT (v1.26.0) in R⁹². The GO biological process was selected for annotations. The result is used to generate plots in Fig. 2g.

Multi-way chromatin interaction analysis

Multi-way interactions are extracted with ‘pairtools parse2’ function in Pairtools, which is designed to rescue complex chromatin ligation events. After deduplication, only trans interactions and cis interactions with genomic distance>10 kb were r etained. Read pairs overlapping with the ENCODE mm10 blacklist regions were removed. For each pair-end read on autosomes supporting multi-way contacts, we collected the 10 kb genomic bins containing their anchors. Finally, we defined one pair-end read to support multi-way contacts, if it contacts ≥3 unique 10 kb bins (can be both intrachromosomal and interchromosomal).

To define contact hubs for multi-way interactions in each mouse brain cell type, we followed the below strategy: first, the reference genome (mm10 for mouse cortex) was segmented into consecutive 10 kb bins. Next, a binary indicator was assigned to each 10 kb bin in each single cell, depending on whether it is involved in a multi-way contact. Then, for each 10 kb bin, we calculated the frequency of single cells involved in multi-way contacts among all cells belonging to the same cell type. We excluded the top 1% of bins with the highest frequency or bins with sequence mappability scores <0.8 when calculating the mean and standard deviation for frequency distribution, then converted the frequency distribution into a Z-score. Finally, a 10 kb bin is defined as in multi-way contact hub if its Z-score is>1.96. Custom scripts to analyze the multi-way interaction hub are available at https://github.com/HuMingLab/Multiway.hub.

Enrichment of chromatin features at multi-way chromatin hubs

After identifying the multi-way chromatin hubs in each mouse cortex cell type, we performed enrichment analysis against cCREs, super-enhancers and cell-type specifically expressed genes. Specifically, cCREs are calculated and converted into 10 kb genomic bins from single-nucleus ATAC using sequencing (snATAC-seq) data⁷¹. Super-enhancers are called using HOMER ‘findPeaks -style super’ on the identified cCREs within each cell type with default parameters⁹³and converted into 10 kb genomic bins. For each cell type A involving a multi-way contact hub, we calculated its overlap with cCREs or super-enhancers from another cell type (B) and created a 2 × 2 contingency table. We then performed Fisher’s exact test to evaluate the significance of enrichment and reported the log₂ odds ratio as the enrichment score. We enumerated all pairs of cell types and compared the log₂ odds ratio in matched (AA) versus unmatched (ABwhere B ≠ A) cell types. We repeated the same enrichment analysis for cell-type-specific marker genes from published datasets⁴⁰except that we used the 10 kb bin overlap with the marker gene transcription start site for enrichment analysis.

Infer CNV with Droplet Hi-C data

Copy number is inferred from Hi-C data using the ‘calculate-cnv’ module in NeoLoopFinder (v0.4.3)⁹⁴. For each single cell, the output residuals from the generalized additive model in ‘calculate-cnv’ were directly used to estimate the copy number ratio. On bulk or pseudo-bulk level, an additional hidden Markov model-based segmentation is performed using the ‘segment-cnv’ module on the copy number ratio to determine the boundaries of copy number ratio segments. We assume all samples used in this paper for CNV calculation are diploid; therefore, the inferred copy number is equal to 2 × copy number ratio. A genome-wide CNV heatmap is plotted at 10 Mb resolution, where chromosome level CNV is plotted at 1 Mb resolution, and the regional CNV profile is plotted at 10 kb resolution in R using pheatmap (v1.0.12). The inferred copy number is further used to assist in identifying ecDNA candidates and the associated significant trans-interactions, and to correct the bias in contact matrices for SV identification.

Identify SV from Droplet Hi-C data

SV was identified using the ‘predictSV’ module in EagleC (v0.1.9)⁹⁵. In short, CNV effects on contact matrices at 5 kb, 10 kb and 50 kb resolutions were removed using the ‘correct-cnv’ module in NeoLoopFinder. ‘predictSV’ then uses a deep learning model to predict SV at each resolution on the corrected matrices and combines all results to obtain a uniform, high-resolution SV list at 5 kb resolution. The ‘annotate-gene-fusion’ module is applied on the final SV list to annotate gene fusion events.

Metrics to define chromatin interactions

We compiled a suite of metrics to assess the cis- and trans-interaction patterns at specific genomic loci. These metrics include the following: (1) contact evenness or ‘hub index’, quantified as the Gini coefficient for trans-interactions among all chromosomes excluding chromosome Y given a 1 Mb genomic bin, and was derived using ineq (v0.2-13) in R; (2) trans-to-cis contacting bin ratio (R_i), which compares the quantity of interacting bins within the same chromosome (N_C) to those among different chromosomes (N_T) for a given bin i: ({R}_{i}=frac{{N}_{T}}{{N}_{C}}). This ratio represents the trans-interaction tendency while it is not confounded by CNV; and (3) copy-number-adjusted trans-chromosomal interaction (_adjnTIF), which is also used to measure a genomic locus’s interaction activity, was calculated as described before⁴⁹.

Develop ecDNA callers for identifying ecDNA candidates

We derived two ecDNA callers to identify ecDNA candidates; one is based on the logistic regression model, and the other one is based on the convolutional neural networks. Both models predict genome-wide 1 Mb bins that contain ecDNA, as well as cells with presence of ecDNA population-wide. Detailed methodologies for both algorithms are delineated as follows.

Logistic regression model-based ecDNA caller

We trained a multivariate logistic regression model using the glm function in R with inferred copy number, hub index and trans-to-cis contacting bin ratio as predictive variables. The training dataset comprised the well-defined ecMYC in COLO320DM (chr8: 127–128 Mb) and EGFR ecDNA in GBM39 (chr7: 55–56 Mb) as positive data, with the same loci in COLO320HSR and GBM39-ER as negative control. We established a probability threshold for classifying ecDNA presence. Specifically, in COLO320DM/COLO320HSR data, this threshold was determined to be 0.5. In GBM39/GBM39-ER data, the threshold was determined to be 0.95.

Deep-learning-model-based ecDNA caller

Data preprocessing

To train the deep-learning-based ecDNA caller, we also selected the well-defined ecMYC in COLO320DM (chr8: 127–128 Mb) and EGFR ecDNA in GBM39 (chr7: 55–56 Mb) as positive data, with the same loci in COLO320HSR and GBM39-ER as negative control. We used all autosomes and chromosome X to create a Hi-C contact matrix at 1 Mb resolution (3,044 × 3,044) for each cell. We then randomly selected 90% of cells as the training data and kept the remaining 10% of cells as the validation data. For each 1 Mb bin, we retained its local 5 Mb neighborhood region (including the center bin itself), and used both cis- and trans-contacts (that is, a 5 × 3,044 matrix) as its feature in the neural network model.

Model architecture

Our proposed model consists of two sequentially placed convolutional modules that extract features from the binarized Hi-C contact matrices. Each convolutional module consists of a multi-channel convolutional layer (8 channels in the first module, 16 channels in the second module), a batch normalization layer, a rectified linear unit activation function⁹⁶ and a max-pooling⁹⁷ layer sequentially. Convolution kernels scan along the direction of rows in each layer. Large convolutional kernel sizes (5 × 45 in the first module and 1 × 45 in the second layer) are set to improve pattern capture because of sparsity. Max-pooling reduces the size of the matrix to half on its second dimension to keep the most important features and thus improves learning efficiency and propagation speed. Strides and paddings of each layer were designed to balance computational efficiency and information retention and must be compatible with the matrix shapes and max-pooling layers. To enhance robustness and prevent overfitting, a dropout layer⁹⁸ with probability of 0.5 is inserted between convolutional modules.

Subsequently, a two-layer fully connected (dense) network integrates information from multiple sources and makes ternary predictions. The first fully connected layer with a 223 hidden size receives (1) the flattened output of the convolutional modules, (2) the flattened 5 × 5 small contact matrix of the corresponding 5 Mb region (whose diagonal entries denote intra-bin contacts) from binarized 5 × 3,044 Hi-C contact matrices, (3) the hub index (used to measure inequality and heterogeneity of interactions with different chromosomes for certain bins and calculated by Python package ‘pygini’) of the center bin calculated from interaction counts aggregated per chromosome and (4) L1 normalized row means of binarized 5 × 3,044 Hi-C contact matrices. The output passes through a batch normalization layer, a Gaussian error linear unit activation function⁹⁹ and a dropout layer with probability of 0.5. Finally, the second fully connected layer with hidden size 64 produces predictions with a subsequent softmax activation function, in which each prediction contains the three probabilities of ‘none’, ‘ecDNA’ and ‘HSR’ sequentially.

Model training and validation

With the package PyTorch¹⁰⁰the training process spanned 40 epochs using the mini-batch training strategy (batch size = 32). To ensure robust optimization, we applied the AdamW^101,102 optimization algorithm with 0.001 weight decay and 0.001 learning rate, and implemented the ‘hard’ bootstrapped cross-entropy loss with parameter b equal to 0.99, which is calculated from one-hot-encoded class labels and predictive softmax¹⁰³ probabilities. To efficiently reduce false positives, we compensated the difference in quantities of different labels and in addition forced the model to favor negative prediction (‘none’) using biased class weights, which is incorporated into the loss calculation. Thus, the model receives a greater loss as a penalty when generating any positive prediction (‘ecDNA’ or ‘HSR’). We mainly used confusion matrix and subsequently derived binary and ternary precision, sensitivity, specificity and accuracy as our evaluation metrics, and binary results (‘none’ or ‘ecDNA’ only) from logistic regression as the baseline performance.

Prediction of ecDNA

The model scans each Hi-C contact matrix from beginning to end without strides to generate SoftMax probabilities on each 1 Mb bin (except the first two bins of chromosome 1 and the last two bins of chromosome X as no padding to the Hi-C matrix is applied) on each single cell. Argmax (the arguments of the maxima) function is used to determine the final predicted class, which is either 0 (‘none’), 1 (‘ecDNA’) or 2 (‘HSR’). Subsequently, results are aggregated to calculate the proportion of cells with ecDNA and/or HSR among the entire cell population.

Identify significant Between ns-interactions of ecDNA candidate loci

To identify genomic regions that preferentially interact with ecDNA, we first quantify the number of interactions (N_i) of 500 kb genomic interval i on different chromosomes with ecDNA candidate loci, treating each chromosome separately. The observed interaction frequency (P_i) is the proportion of N_i relative to all contacts on the same chromosome: ({P}_{i}=,frac{{N}_{i}}{{sum }_{g=1}^{n}{N}_{g}}). The expected interaction frequency for interval i was calculated as the ratio of the inferred copy number (CN_i) to the total copy number on the same chromosome: ({E}_{i}=,frac{{rm{CN}}_{i}}{{sum }_{g=1}^{n}{rm{CN}}_{g}})based on the null hypothesis that the interaction frequency for genomic regions with ecDNA is weighted only by its underlying copy number. To identify significant interactions, observed-versus-expected P value was calculated based on the binomial distribution model. Multiple testing correction was done by Bonferroni adjustment. Regions with adjusted P value < 0.05 were selected as significant interacting regions.

ecDNA hub analysis

To generate a background distribution for ecDNA hub analysis, we shuffled chromosome identities for all trans interactions in each cell and calculated the hub index using R package ineq based on the shuffled contact matrices. P value was calculated using the Wilcoxon signed-rank test.

Analysis of GBM cellular states

To classify single cells within GBM 10x Genomics Multiome or Paired Hi-C RNA datasets into one of four predefined malignant cellular states, we adopted a previously described two-dimensional visualization technique⁵³. Briefly, we quantified the gene enrichment score (SC_j(i)) for each cell (i) against one of the four gene sets (G_j) associated with the particular cellular state. This score was calculated as the relative averaged expression (Exp) of G_j in cell i compared with a group of genes (G_jcont) with a similar level of expression as the control: ({rm{SC}}_{j(i)}=frac{{sum }_{g=1}^{{N}_{j}}{rm{Exp}}_{g(i)}}{{N}_{j}}-frac{{sum }_{g=1}^{{N}_{j}rm{cont}}{rm{Exp}}_{g(i)}}{{N}_{{j}rm{cont}}})where N_j and N_jcont are the number of genes in G_j and G_jcont. After obtaining scores for all four cellular states, the cells were stratified into OPC/NPC versus AC/MES categories using the differential score (D=max left({rm{SC}}_{rm{OPC}},,{rm{SC}}_{rm{NPC}}right)-max left({rm{SC}}_{rm{AC}},,{rm{SC}}_{rm{MES}}right)). For further refinement, OPC/NPC cells were assigned an identity value (C={log }_{2}(left|{rm{SC}}_{rm{OPC}}-{rm{SC}}_{rm{NPC}}right|+1))and AC/MES cells were similarly categorized with (C={log }_{2}(left|{rm{SC}}_{rm{AC}}-{rm{SC}}_{rm{MES}}right|+1)). The distribution of cellular states was then plotted in the two-dimensional representation with D on the y-axis and C on the x-axis.

For cellular state identification in Droplet Hi-C data, co-embedding with the reference snRNA-seq was performed using the scGAD score. After determining the top 15 nearest snRNA-seq neighbors and their scaled distance-based similarity score (D_m) for each Droplet Hi-C cell, the Hi-C gene enrichment score (HSC_j(i)) was computed as the sum of the neighbor-weighted enrichment scores: ({rm{HSC}}_{jleft(iright)}=,{sum }_{g=1}^{15}{rm{SC}}_{j(g)}x{D}_{m(g)}).

Processing and analysis of Paired Hi-C data

Preprocessing and analysis of Hi-C modality in Paired Hi-C are identical as described for Droplet Hi-C. For RNA modality, fastq files were demultiplexed by cellranger-arc, but preprocessed using cellranger (v6.1.2). After obtaining the cell-by-gene matrix, clustering and visualization were performed as described for the 10x Genomics Multiome RNA dataset. As barcodes in the same gel bead for RNA and Hi-C modalities are different, we performed manual pairing to match cell barcodes based on the 10x Genomics Multiome barcodes white list provided in cellranger-arc (10x Genomics).

Integration of the Paired Hi-C RNA dataset with reference datasets was performed by Seurat. First, normalization of gene counts was performed, and the top 2,000 shared variable genes among datasets were selected as integration features. Subsequent canonical correlation analysis allowed projecting all nuclei into a unified embedding space. Anchors (pairs of corresponding cells from distinct datasets) were then discerned through mutual nearest neighbor searching. Anchors with low confidence were excluded, and the shared neighbor between anchors and query cells are computed. Louvain clustering was applied to the shared neighbor graph to discern co-embedded clusters. We calculated overlap coefficients as described in a previous section to compare clustering results from different datasets.

Analysis of gene expression levels and copy numbers of ecDNA

To calculate the correlation between gene expression level and inferred copy number of genes on ecDNA, we first refined the range of ecDNA at 10 kb resolution. This is based on the observation of the increased intra-ecDNA interaction frequency than interactions with regions on linear genome, irrespective of their genomic distance. Specifically, we enumerated the local interactions within each 10 kb segment of the 1 Mb ecDNA candidates. Subsequently, the contact numbers were smoothed among adjacent bins. By calculating the differential contact numbers between neighboring bins, we determined the changing points of interaction with an average difference cutoff among the entire 1 Mb region. The outermost local maxima and minima were designated as ecDNA boundaries.

With the ecDNA boundaries established, we categorized genes with gene bodies residing within ecDNA as ‘ecDNA genes’. In the case of ecMYC, we further classified genes on GBM39 ecMYC as ‘shared genes’ and genes on non-overlapping regions between GBM39 and GBM39-ER as ‘ecMYC variable genes’, considering the greater ecDNA size in GBM39-ER at the pseudo-bulk level. The SCC between gene expression levels (RPKM) and the average inferred copy number over all 10 kb segments at gene body among all cells was calculated, for a comprehensive representation of the interplay between gene expression and copy number.

Ethical approval

The GBM specimen collection was approved by the Institutional Review Board (IRB) at the University of Minnesota (IRB number STUDY00012599). The AML and myelodysplastic syndrome specimen collection was approved by the IRB at the University of California, San Diego (IRB number 131550). Before sample collection, a dedicated research coordinator obtained informed consent from each patient, in accordance with the Declaration of Helsinki and appropriate Ethics Committee approval from each partner institution.

Reporting summary

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

Raw and processed sequencing data generated in this study have been submitted to GEO (accession number GSE253407)¹⁰⁴. Datasets for bulk in situ Hi-C on cultured cells were downloaded from the 4DN Data Portal with the following accession numbers: K562 (4DNFIIX5BNC9 and 4DNFI244AS29)¹⁰⁵GM12878 (4DNFIIS73OJN and 4DNFI3O82QVV)³² and HeLa S3 at prometaphase (4DNFIW458FJD)³⁴. Other external datasets were downloaded from NCBI GEO with the following accession numbers: in situ Hi-C on WTC-11 (GSE106690)¹⁰⁶sci-Hi-C on cell line mixture (GSE84920)²²Dip-C on the adult mouse cortex (GSE162511)¹⁷HiRES on mouse embryos (GSE223917)²¹single-cell Hi-C on mouse Th1 cells (GSE48262)¹³single-nuclei Hi-C on K562 (GSE80006)¹⁶Droplet Paired-Tag dataset on the mouse frontal cortex (GSE152020)⁴⁰10x Genomics Multiome dataset on the mouse cortex (GSE210749)¹⁰⁷10x Genomics Multiome dataset on COLO320DM and COLO320HSR (GSE160148)⁴⁴ and DOGMA-seq dataset on PBMCs (GSE156478)⁵⁷. The BICCN whole mouse brain sn-m3C-seq datasets³⁷ and BICCN MOp 10x snRNA-seq data¹⁰⁸ were downloaded via the NeMO archive (https://assets.nemoarchive.org/dat-sig83t9; https://assets.nemoarchive.org/dat-ch1nqb7). The 10x Genomics Multiome dataset on PBMCs was downloaded from the 10x Genomics dataset portal (https://www.10xgenomics.com/en/datasets). The human reference genome (GRCh38/hg38, https://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips)¹⁰⁹ and the mouse reference genome (GRCm38/mm10, https://hgdownload.soe.ucsc.edu/goldenPath/mm10/bigZips)¹¹⁰ were from UCSC. Source data are provided with this paper.

Scripts and code are available via GitHub at https://github.com/Xieeeee/Droplet-Hi-C (ref. ¹¹¹). The ecDNA callers are available via GitHub at https://github.com/HuMingLab/ecDNAcaller (ref. ¹¹²). Scripts to analyze the multi-way interaction hub are available via GitHub at https://github.com/HuMingLab/Multiway.hub (ref. ¹¹³).

1. Du, Z. et al. Allelic reprogramming of 3D chromatin architecture during early mammalian development. Nature 547232–235 (2017).

   Article  CAS  PubMed  Google Scholar

2. Ke, Y. et al. 3D chromatin structures of mature gametes and structural reprogramming during mammalian embryogenesis. Cell 170367–381. e320 (2017).

   Article  CAS  PubMed  Google Scholar

3. Xu, J. et al. Subtype-specific 3D genome alteration in acute myeloid leukaemia. Nature 611387–398 (2022).

   Article  CAS  PubMed  PubMed Central  Google Scholar

4. Xu, Z. et al. Structural variants drive context-dependent oncogene activation in cancer. Nature 612564–572 (2022).

   Article  CAS&nbs p; PubMed  PubMed Central  Google Scholar

5. Dixon, J. R. et al. Chromatin architecture reorganization during stem cell differentiation. Nature 518331–336 (2015).

   Article  CAS  PubMed  PubMed Central  Google Scholar

6. Zheng, H. & Xie, W. The role of 3D genome organization in development and cell differentiation. Nat. Rev. Mol. Cell Biol. 20535–550 (2019).

   Article  CAS  PubMed  Google Scholar

7. Ghosh, R. P. & Meyer, B. J. Spatial organization of chromatin: emergence of chromatin structure during development. Annu. Rev. Cell Dev. Biol. 37199–232 (2021).

   Article  CAS  PubMed  PubMed Central  Google Scholar

8. Gorkin, D. U., Leung, D. & Ren, B. The 3D genome in transcriptional regulation and pluripotency. Cell Stem Cell 14762–775 (2014).

   Article  CAS  PubMed  PubMed Central  Google Scholar

9. Morgan, M. A. & Shilatifard, A. Chromatin signatures of cancer. Genes Dev. 29238–249 (2015).

   Article  CAS  PubMed  PubMed Central  Google Scholar

10. Papaemmanuil, E. et al. Genomic classification and prognosis in acute myeloid leukemia. N. Engl. J. Med. 3742209–2221 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar

11. Kandoth, C. et al. Mutational landscape and significance across 12 major cancer types. Nature 502333–339 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar

12. Blum, A., Wang, P. & Zenklusen, J. C. SnapShot: TCGA-analyzed tumors. Cell 173530 (2018).

    Article  CAS  PubMed  Google Scholar

13. Nagano, T. et al. Single-cell Hi-C reveals cell-to-cell variability in chromosome structure. Nature 50259–64 (2013).

    Article  CAS  PubMed  Google Scholar

14. Nagano, T. et al. Cell-cycle dynamics of chromosomal organization at single-cell resolution. Nature 54761–67 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar

15. Stevens, T. J. et al. 3D structures of individual mammalian genomes studied by single-cell Hi-C. Nature 54459–64 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar

16. Flyamer, I. M. et al. Single-nucleus Hi-C reveals unique chromatin reorganization at oocyte-to-zygote transition. Nature 544110–114 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar

17. Tan, L., Xing, D., Chang, C.-H., Li, H. & Xie, X. S. Three-dimensional genome structures of single diploid human cells. Science 361924–928 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar

18. Li, W. et al. scNanoHi-C: a single-cell long-read concatemer sequencing method to reveal high-order chromatin structures within individual cells. Nat. Methods 201493–1505 (2023).

    Article  CAS  PubMed  Google Scholar

19. Li, G. et al. Joint profiling of DNA methylation and chromatin architecture in single cells. Nat. Methods 16991–993 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

20. Lee, D.-S. et al. Simultaneous profiling of 3D genome structure and DNA methylation in single human cells. Nat. Methods 16999–1006 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

21. Liu, Z. et al. Linking genome structures to functions by simultaneous single-cell Hi-C and RNA-seq. Science 3801070–1076 (2023).

    Article  CAS  PubMed  Google Scholar

22. Ramani, V. et al. Massively multiplex single-cell Hi-C. Nat. Methods 14263–266 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar

23. Zhou, T. et al. GAGE-seq concurrently profiles multiscale 3D genome organization and gene expression in single cells. Nat. Genet. 561701–1711 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar

24. Aaltonen, L. A. et al. Pan-cancer analysis of whole genomes. Nature 57882–93 (2020).

    Article  Google Scholar

25. Cox, D., Yuncken, C. & Spriggs, A. Minute chromatin bodies in malignant tumours of childhood. Lancet 28655–58 (1965).

    Article  Google Scholar

26. Spriggs, A. I., Boddington, M. M. & Clarke, C. M. Chromosomes of human cancer cells. Bro. With. J. 21431–1435 (1962).

    Article  CAS  PubMed  PubMed Central  Google Scholar

27. Møller, H. D. et al. Circular DNA elements of chromosomal origin are common in healthy human somatic tissue. Nat. Common. 91069 (2018).

    Article  PubMed  PubMed Central  Google Scholar

28. Paulsen, T., Kumar, P., Koseoglu, M. M. & Dutta, A. Discoveries of extrachromosomal circles of DNA in normal and tumor cells. Trends Genet. 34270–278 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar

29. Wu, S. et al. Circular ecDNA promotes accessible chromatin and high oncogene expression. Nature 575699–703 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

30. Dixon, J. R. et al. Integrative detection and analysis of structural variation in cancer genomes. Nat. Genet. 501388–1398 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar

31. Harewood, L. et al. Hi-C as a tool for precise detection and characterisation of chromosomal rearrangements and copy number variation in human tumours. Genome Biol. 18125 (2017).

    Article  PubMed  PubMed Central  Google Scholar

32. Rao, S. S. et al. A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell 1591665–1680 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar

33. Zhang, R., Zhou, T. & Ma, J. Multiscale and integrative single-cell Hi-C analysis with Higashi. Nat. Biotechnol. 40254–261 (2022).

    Article  CAS  PubMed  Google Scholar

34. Abramo, K. et al. A chromosome folding intermediate at the condensin-to-cohesin transition during telophase. Nat. Cell Biol. 211393–1402 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

35. Salik, B. et al. Targeting RSPO3-LGR4 signaling for leukemia stem cell eradication in acute myeloid leukemia. Cancer Cell 38263–278.e266 (2020).

    Article  CAS  PubMed  Google Scholar

36. Tan, L. et al. Changes in genome architecture and transcriptional dynamics progress independently of sensory experience during post-natal brain development. Cell 184741–758.e717 (2021).

    Article  CAS  PubMed  Google Scholar

37. Liu, H. et al. Single-cell DNA methylome and 3D multi-omic atlas of the adult mouse brain. Nature 624366–377 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar

38. Zhang, C. et al. tagHi-C reveals 3D chromatin architecture dynamics during mouse hematopoiesis. Cell Rep. 32108206 (2020).

    Article  CAS  PubMed  Google Scholar

39. Shen, S., Zheng, Y. & Keleş, S. scGAD: single-cell gene associating domain scores for exploratory analysis of scHi-C data. Bioinformatics 383642–3644 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar

40. Xie, Y. et al. Droplet-based single-cell joint profiling of histone modifications and transcriptomes. Nat. Struct. Mol. Biol. 301428–1433 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar

41. Lafontaine, D. L., Yang, L., Dekker, J. & Gibcus, J. H. Hi-C 3.0: improved protocol for genome-wide chromosome conformation capture. Curr. Protoc. 1e198 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

42. Zhu, Q. et al. Genetic evidence that Nkx2.2 and Pdgfra are major determinants of the timing of oligodendrocyte differentiation in the developing CNS. Development 141548–555 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar

43. McLean, C. Y. et al. GREAT improves functional interpretation of cis-regulatory regions. Nat. Biotechnol. 28495–501 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar

44. Hung, K. L. et al. ecDNA hubs drive cooperative intermolecular oncogene expression. Nature 600731–736 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

45. Le, D. J., Hafner, A., Gaddam, S., Wang, K. C. & Boettiger, A. N. Super-enhancer interactomes from single cells link clustering and transcription. Preprint at bioRxiv https://doi.org/10.1101/2024.05.08.593251 (2024).

46. Di Giammartino, D. C., Polyzos, A. & Apostolou, E. Transcription factors: building hubs in the 3D space. Cell Cycle 192395–2410 (2020).

    Article  PubMed  PubMed Central  Google Scholar

47. Turner, K. M. et al. Extrachromosomal oncogene amplification drives tumour evolution and genetic heterogeneity. Nature 543122–125 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar

48. Deshpande, V. et al. Exploring the landscape of focal amplifications in cancer using AmpliconArchitect. Nat. Common. 10392 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

49. Zhu, Y. et al. Oncogenic extrachromosomal DNA functions as mobile enhancers to globally amplify chromosomal transcription. Cancer Cell 39694–707.e7 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

50. Liang, Z. et al. Chromatin-associated RNA dictates the ecDNA interactome in the nucleus. Preprint at bioRxiv https://doi.org/10.1101/2023.07.27.550855 (2023).

51. Nathanson, D. A. et al. Targeted therapy resistance mediated by dynamic regulation of extrachromosomal mutant EGFR DNA. Science 34372–76 (2014).

    Article  CAS  PubMed  Google Scholar

52. Purshouse, K. et al. Oncogene expression from extrachromosomal DNA is driven by copy number amplification and does not require spatial clustering in glioblastoma stem cells. eLife 11e80207 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar

53. Neftel, C. et al. An integrative model of cellular states, plasticity, and genetics for glioblastoma. Cell 178835–849.e821 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

54. Chaligne, R. et al. Epigenetic encoding, heritability and plasticity of glioma transcriptional cell states. Nat. Genet. 531469–1479 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

55. Rahme, G. J. et al. Modeling epigenetic lesions that cause gliomas. Cell 1863674–3685.e3614 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar

56. Lancho, O. & Herranz, D. The MYC enhancer-ome: long-range transcriptional regulation of MYC in cancer. Trends Cancer 4810–822 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar

57. Mimitou, E. P. et al. Scalable, multimodal profiling of chromatin accessibility, gene expression and protein levels in single cells. Nat. Biotechnol. 391246–1258 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

58. Fan, X. et al. SMOOTH-seq: single-cell genome sequencing of human cells on a third-generation sequencing platform. Genome Biol. 22195 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

59. Chang, L. et al. Single-cell third-generation sequencing-based multi-omics uncovers gene expression changes governed by ecDNA and structural variants in cancer cells. Clin. Transl. Med. 13e1351 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar

60. Chamorro González, R. et al. Parallel sequencing of extrachromosomal circular DNAs and transcriptomes in single cancer cells. Nat. Genet. 55880–890 (2023).

    Article  PubMed  PubMed Central  Google Scholar

61. Møller, H. D., Parsons, L., Jørgensen, T. S., Botstein, D. & Regenberg, B. Extrachromosomal circular DNA is common in yeast. Proc. Natl Acad. Sci. USA 112E3114–E3122 (2015).

    Article  PubMed  PubMed Central  Google Scholar

62. Koche, R. P. et al. Extrachromosomal circular DNA drives oncogenic genome remodeling in neuroblastoma. Nat. Genet. 5229–34 (2020).

    Article  CAS  PubMed  Google Scholar

63. Kim, H. et al. Extrachromosomal DNA is associated with oncogene amplification and poor outcome across multiple cancers. Nat. Genet. 52891–897 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar

64. Wu, S., Bafna, V., Chang, H. Y. & Mischel, P. S. Extrachromosomal DNA: an emerging hallmark in human cancer. Annu. Rev. Pathol. 17367–386 (2022).

    Article  CAS  PubMed  Google Scholar

65. Verhaak, R. G., Bafna, V. & Mischel, P. S. Extrachromosomal oncogene amplification in tumour pathogenesis and evolution. Nat. Rev. Cancer 19283–288 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

66. Wu, S., Bafna, V. & Mischel, PS Extrachromosomal DNA (ecDNA) in cancer pathogenesis. Curr. Opin. Genet. Dev. 6678–82 (2021).

    Article  CAS  PubMed  Google Scholar

67. Curtis, E. J., Rose, J. C., Mischel, P. S. & Chang, H. Y. Extrachromosomal DNA: biogenesis and functions in cancer. Annu. Rev. Cancer Biol. 8135–153 (2024).

    Article  Google Scholar

68. Chowdhry, S. et al. Abstract 1520: Replication stress and the inability to repair damaged DNA, the potential “Achilles’ heel” of ecDNA+ tumor cells. Cancer Res. 821520 (2022).

    Article  Google Scholar

69. Bei, Y. et al. Passenger gene coamplifications create collateral therapeutic vulnerabilities in cancer. Cancer Discov. 14492–507 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar

70. Noorani, I., Mischel, P. S. & Swanton, C. Leveraging extrachromosomal DNA to fine-tune trials of targeted therapy for glioblastoma: opportunities and challenges. Nat. Rev. Clin. Oncol. 19733–743 (2022).

    Article  CAS  PubMed  Google Scholar

71. Li, Y. E. et al. An atlas of gene regulatory elements in adult mouse cerebrum. Nature 598129–136 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

72. Langmead, B., Trapnell, C., Pop, M. & Salzberg, S. L. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 10R25 (2009).

    Article  PubMed  PubMed Central  Google Scholar

73. Martin, M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet J. 1710–12 (2011).

    Article  Google Scholar

74. Li, H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. Preprint at https://arxiv.org/abs/1303.3997 (2013).

75. Open2C et al. Pairtools: from sequencing data to chromosome contacts. PLOS Comput. Biol. 20e1012164 (2024).

76. Abdennur, N. & Mirny, L. A. Cooler: scalable storage for Hi-C data and other genomically labeled arrays. Bioinformatics 36311–316 (2019).

    Article  PubMed Central  Google Scholar

77. Hunter, J. D. Matplotlib: a 2D graphics environment. Comput. Sci. Eng. 990–95 (2007).

    Article  Google Scholar

78. Amemiya, H. M., Kundaje, A. & Boyle, A. P. The ENCODE blacklist: identification of problematic regions of the genome. Sci. Rep. 99354 (2019).

    Article  PubMed  PubMed Central  Google Scholar

79. Traag, V. A., Waltman, L. & van Eck, N. J. From Louvain to Leiden: guaranteeing well-connected communities. Sci. Rep. 95233 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

80. Csardi, G. & Nepusz, T. The igraph software package for complex network research. InterJournal Complex Syst. 16951–9 (2006).

    Google Scholar

81. Open2C et al. Cooltools: Enabling high-resolution Hi-C analysis in Python. PLOS Comput. Biol. 20e1012067 (2024).

82. Zhou, J. et al. Robust single-cell Hi-C clustering by convolution- and random-walk-based imputation. Proc. Natl Acad. Sci. USA 11614011–14018 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

83. Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 1915 (2018).

    Article  PubMed  PubMed Central  Google Scholar

84. Liu, H. et al. DNA methylation atlas of the mouse brain at single-cell resolution. Nature 598120–128 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

85. Dong, W., Moses, C. & Li, K. Efficient k-nearest neighbor graph construction for generic similarity measures. In Proc. 20th Int. Conf. on World Wide Web 577–586 (Association for Computing Machinery, 2011).

86. Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 161289–1296 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar

87. Chakraborty, A., Wang, J. G. & Ay, F. dcHiC detects differential compartments across multiple Hi-C datasets. Nat. Common. 136827 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar

88. Shin, H. et al. TopDom: an efficient and deterministic method for identifying topological domains in genomes. Nucleic Acids Res. 44e70 (2015).

    Article  PubMed  PubMed Central  Google Scholar

89. Hounkpe, B. W., Chenou, F., de Lima, F. & De Paula, E. V. HRT Atlas v1.0 database: redefining human and mouse housekeeping genes and candidate reference transcripts by mining massive RNA-seq datasets. Nucleic Acids Res. 49D947–D955 (2020).

    Article  PubMed Central  Google Scholar

90. Durand, N. C. et al. Juicer provides a one-click system for analyzing loop-resolution Hi-C experiments. Cell Syst. 395–98 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar

91. Yu, M. et al. SnapHiC: a computational pipeline to identify chromatin loops from single-cell Hi-C data. Nat. Methods 181056–1059 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

92. Gu, Z. & Hübschmann, D. rGREAT: an R/bioconductor package for functional enrichment on genomic regions. Bioinformatics 39btac745 (2022).

    Article  PubMed Central  Google Scholar

93. Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38576–589 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar

94. Wang, X. et al. Genome-wide detection of enhancer-hijacking events from chromatin interaction data in rearranged genomes. Nat. Methods 18661–668 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar

95. Wang, X., Luan, Y. & Yue, F. EagleC: a deep-learning framework for detecting a full range of structural variations from bulk and single-cell contact maps. Sci. Adv. 8eabn9215 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar

96. Nair, V. & Hinton, G. E. Rectified linear units improve restricted Boltzmann machines. In Proc. 27th Int. Conf. on Machine Learning (ICML-2010) 807–814 (2010).

97. Krizhevsky, A., Sutskever, I. & Hinton, G. E. ImageNet classification with deep convolutional neural networks. Adv. Neural Inf. Process. Syst. 25(2012).

98. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I. & Salakhutdinov, R. Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res. 151929–1958 (2014).

    Google Scholar

99. Hendrycks, D. & Gimpel, K. Gaussian error linear units (GELUs). Preprint at https://arxiv.org/abs/1606.08415 (2016).

100. Paszke, A. et al. Pytorch: an imperative style, high-performance deep learning library. Adv. Neural Inf. Process. Syst. 32(2019).

101. Kingma, D. P. & Ba, J. Adam: a method for stochastic optimization. Preprint at https://arxiv.org/abs/1412.6980 (2014).

102. Loshchilov, I. & Hutter, F. Decoupled weight decay regularization. Preprint at https://arxiv.org/abs/1711.05101 (2017).

103. Bridle, J. S. Probabilistic interpretation of feedforward classification network outputs, with relationships to statistical pattern recognition. In Neurocomputing: Algorithms, Architectures and Applications 227–236 (Springer, 1990).

104. Chang, L. et al. Droplet Hi-C enables scalable, single-cell profiling of chromatin architecture in heterogeneous tissues. GSE253407. Gene Expression for All https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE253407 (2024).

105. Ray, J. et al. Chromatin conformation remains stable upon extensive transcriptional changes driven by heat shock. Proc. Natl Acad. Sci. USA 11619431–19439 (2019).

     Article  CAS  PubMed  PubMed Central  Google Scholar

106. Bertero, A. et al. Dynamics of genome reorganization during human cardiogenesis reveal an RBM20-dependent splicing factory. Nat. Common. 101538 (2019).

     Article  PubMed  PubMed Central  Google Scholar

107. Bravo González-Blas, C. et al. SCENIC+: single-cell multiomic inference of enhancers and gene regulatory networks. Nat. Methods 201355–1367 (2023).

     Article  PubMed  PubMed Central  Google Scholar

108. Yao, Z. et al. A transcriptomic and epigenomic cell atlas of the mouse primary motor cortex. Nature 598103–110 (2021).

     Article  CAS  PubMed  PubMed Central  Google Scholar

109. Schneider, V. A. et al. Evaluation of GRCh38 and de novo haploid genome assemblies demonstrates the enduring quality of the reference assembly. Genome Res. 27849–864 (2017).

     Article  CAS  PubMed  PubMed Central  Google Scholar

110. Ch urch, D. M. et al. Modernizing reference genome assemblies. PLoS Biol. 9e1001091 (2011).

     Article  CAS  PubMed  PubMed Central  Google Scholar

111. Xie, Y. Droplet Hi-C. GitHub https://github.com/Xieeeee/Droplet-Hi-C (2024).

112. Sun, J. & Hu, M. ecDNAcaller. GitHub https://github.com/HuMingLab/ecDNAcaller (2024).

113. Mishra, S. & Hu, M. Multiway hub. GitHub https://github.com/HuMingLab/Multiway.hub (2024).

Download references

Author notes

1. These authors contributed equally: Lei Chang, Yang Xie.

Authors and Affiliations

1. Department of Cellular and Molecular Medicine, University of California, San Diego, La Jolla, CA, USA

   Lei Chang, Yang Xie, Brett Taylor, Zhaoning Wang, Ethan J. Armand, Jie Xu, Melodi Tastemel, Audrey Lie, Zane A. Gibbs, Hannah S. Indralingam & Bing Ren

2. Biomedical Sciences Graduate Program, University of California, San Diego, La Jolla, CA, USA

   Yang Xie & Brett Taylor

3. Medical Scientist Training Program, University of California, San Diego, La Jolla, CA, USA

   Brett Taylor

4. Department of Quantitative Health Sciences, Lerner Research Institute, Cleveland Clinic Foundation, Cleveland, OH, USA

   Jiachen Sun, Shreya Mishra & Ming Hu

5. Systems Biology and Bioinformatics PhD Program, Case Western Reserve University School of Medicine, Cleveland, OH, USA

   Jiachen Sun

6. Bioinformatics and Systems Biology Program, University of California, San Diego, La Jolla, CA, USA

   Ethan J. Armand

7. Moores Cancer Center, UC San Diego, La Jolla, CA, USA

   Tuyet M. Tan, Rafael Bejar & Bing Ren

8. Department of Neurosurgery, University of Minnesota, Minneapolis, MN, USA

   Clark C. Chen

9. Department of Medicine, University of California, San Diego School of Medicine, La Jolla, CA, USA

   Frank B. Furnari

10. Center for Epigenomics, Institute for Genomic Medicine, School of Medicine, University of California, San Diego, La Jolla, CA, USA

    Bing Ren

Contributions

This study was conceived by B.R. and L.C. L.C. designed and carried out the Droplet Hi-C, Paired Hi-C, part of the 10x Genomics Multiome experiments and DNA FISH. Y.X. performed the bioinformatics analysis. B.T. contributed to ecDNA-related experimentation and data analysis. Z.W. dissected mouse cortex samples for this study, generated 10x Genomics Multiome data from GBM samples and performed next-generation sequencing on the NextSeq 2000 platform. M.H. and J.S. developed the ecDNA caller algorithm. S.M. and M.H. performed multi-way interaction analysis. E.J.A. helped perform ecDNA hub analysis and Hi-C contact analysis. J.X. and M.T. generated HeLa S3 bulk Hi-C datasets. Z.A.G. provided WTC-11 cells. A.L. and H.S.I. contributed to experiments during paper revision. R.B. and T.M.T. provided the samples from patients with AML, and C.C.C. provided the samples from patients with GBM. F.B.F. provided the GBM39 cell line and related reagents, and contributed to ecDNA-related experimentation and data analysis. B.R., L.C. and Y.X. wrote the paper with input from all the authors.

Corresponding author

Correspondence to Bing Ren.

Competing interests

B.R. is a cofounder of Epigenome Technologies, Inc. and has equity in Arima Genomics, Inc. The other authors declare no competing interests.

Peer review information

Nature Biotechnology thanks Caleb Lareau and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions