Image-based 3D genomics through chromatin tracing

Abstract

Correct organization of higher-order genome folding is essential for the regulation of gene expression, DNA replication and other genomic functions. Technological advances in high-throughput sequencing-based methods have allowed for systematic profiling of the fundamental architectural features of chromatin organization at the genome level. However, how chromatin is folded in 3D space at single-cell and single-chromosome-copy resolution in intact cells and tissues has been a long-standing question owing to a lack of appropriate methodology. Recent advances in chromatin labelling, imaging and automated fluidics technologies have led to the development of chromatin tracing, enabling direct mapping of the 3D chromatin folding trajectory in situ at the single-cell and single-molecule level. Within nearly a decade of its development, chromatin tracing has been applied at different genomic scales and to a spectrum of cell types and model organisms, improving our understanding of the structures, mechanisms and functions of chromatin organization in various biological and medical areas. In this Primer, we introduce the experimental principles, data analysis procedures and current applications of chromatin tracing. We describe how chromatin tracing can be combined with multimodal imaging and genetic screening technologies and provide a perspective on the limitations of current chromatin tracing approaches and the direction of technological developments for filling major gaps in discoveries.

Introduction

In mammalian cells, metres of genomic DNA are packaged to fit inside the 10-μm scale cell nucleus while maintaining the highly organized architecture required for essential biological functions such as gene expression regulation and DNA replication timing^1–3 (Fig. 1a). Understanding genomic DNA folding in the nucleus is fundamental to understanding its function. At the first level of the genomic DNA folding architecture, DNA is wrapped around histone proteins to form individual nucleosomes, initially discovered through electron microscopy⁴, producing an approximately sevenfold linear compaction of the genomic DNA^5,6. Electron microscopy also directly observed the formation of DNA loops^7,8. On the other end of the spectrum, in situ hybridization techniques revealed that during interphase of the cell cycle, individual chromosomes occupy distinct nuclear zones known as chromosomal territories^9,10, which are non-randomly positioned in the nucleus¹¹. However, understanding how the 3D genome is organized at intermediate scales has been a long-standing question owing to a lack of available techniques.

Fig. 1 |. Technologies for mapping 3D chromatin organization.

a, Levels of 3D genome organization from entire chromosomes to A/B compartments, topologically associating domains (TADs), chromatin loops and nucleosomes. b, Timeline of key technological development for mapping 3D chromatin organization, including sequencing methods (blue) and imaging methods (orange). c, How chromatin tracing solves the diffraction and fluorescence colour limitations of conventional DNA fluorescence in situ hybridization (FISH). 3C, chromosome conformation capture; 4C, circular chromosome conformation capture; 5C, chromosome conformation capture carbon copy; Capture-C, chromosome conformation capture with capture probes; ChIA-Drop, chromatin interaction analysis by droplet-based and barcode-linked sequencing; ChIA-PET, chromatin interaction analysis by paired-end tag sequencing; CTCF, CCCTC-binding factor; Dip-C, diploid chromatin conformation capture; DNA MERFISH, DNA multiplexed error-robust fluorescence in situ hybridization; DNA seqFISH+, DNA sequential fluorescence in situ hybridization plus; EM, electron microscopy; GAGE-seq, genome architecture and gene expression by sequencing; GAM, genome architecture mapping; Hi-C, high-throughput chromosome conformation capture; HiChIP, high-throughput chromosome conformation capture with chromatin immunoprecipitation; Hi-M, microscopy-based chromosome conformation capture; HiRES, Hi-C and RNA-seq employed simultaneously; Hyb, hybridization round; IGS, in situ genome sequencing; LiMCA, linking mRNA to chromatin architecture; Micro-C, micrococcal nuclease chromosome conformation assay; MINA, multiplexed imaging of nucleome architectures; MUSIC, multinucleic acid interaction mapping in single cells; OligoFISSEQ, Oligopaint fluorescence in situ sequencing; ORCA, optical reconstruction of chromatin architecture; scHi-C, single-cell high-throughput chromosome conformation capture; scSPRITE, single-cell split-pool recognition of interactions by tag extension; SPRITE, split-pool recognition of interactions by tag extension. Part c adapted from ref. 121, CC BY 4.0.

Advances in genomic technologies, particularly sequencing technologies, within the first decade of the twenty-first century have generated multiple new methods for mapping chromatin organization at intermediate scales. First, new methods that measure chromatin contacts or proximity based on pairwise proximity ligation (such as chromosome conformation capture (3C)¹², high-throughput chromosome conformation capture (Hi-C)¹³ and micrococcal nuclease chromosome conformation assay (Micro-C)^14–16) or non-ligation approaches (such as split-pool recognition of interactions by tag extension (SPRITE)^17,18 and genome architecture mapping (GAM)^19,20) have enabled the genome-wide mapping of higher-order chromatin structures at different genomic scales^21–23, including A/B compartments¹³, subcompartments²⁴, topologically associating domains (TADs)^25–28, sub-TADs^29,30 and chromatin loops²⁴ and hubs^17–20. The addition of capture-enrichment strategies enabled dissection of chromatin interactions associated with specific factors (for example, chromatin interaction analysis by paired-end tag sequencing (ChIA-PET)³¹ and high-throughput chromosome conformation capture with chromatin immunoprecipitation (HiChIP)³²) or genomic loci (such as chromosome conformation capture with capture probes (Capture-C)^33,34), which has provided valuable information related to the regulatory mechanisms underlying 3D genome organization. Sequencing-based approaches also enabled the discovery of lamina-associated domains³⁵ and nucleolus-associated chromatin domains^36,37. Despite the important findings, all sequencing-based 3D genomics methods are incapable of directly tracing the 3D folding path of chromatin and require the dissociation of cells from tissues, along with cell lysis during sample preparation, resulting in loss of spatial information in the context of both cellular architecture and the tissue microenvironment. Most of the 3D genome information obtained using sequencing-based methods are cell-population-averaged results, with the exception of findings obtained from recent single-cell versions of the methods, such as single-cell Hi-C (scHi-C)^38–42, single-cell split-pool recognition of interactions by tag extension (scSPRITE)⁴³ and diploid chromatin conformation capture (Dip-C)^44,45. Combining the sequencing-based methods with the multimodal profiling of RNAs, proteins and other biomolecules within the same single cells remains challenging. Therefore, the development of alternative approaches able to compensate for these limitations and complement sequencing-based methods is needed. Imaging-based 3D genomics methods have been proved to be effective (Fig. 1b).

In contrast to sequencing-based methods, imaging-based methods directly visualize the 3D positions of genomic loci and are able to construct the 3D folding path of chromatin by linking the positions of imaged genomic loci along the same DNA molecule⁴⁶. We use the term chromatin tracing here to refer to methods that use this strategy to construct 3D chromatin folding paths. Currently, most chromatin tracing methods depend on multiplexed fluorescence in situ hybridization (FISH)⁴⁶. FISH is a cytogenic technique using fluorescent DNA probes to target specific chromosomal locations within the nucleus and has long been used to locate specific DNA sequences^47–50. The simultaneous imaging and unambiguous identification of numerous genomic regions on the same chromosome using conventional FISH techniques have been prohibited by the limited number of distinct colour channels (usually ≤5) and diffraction limit (250–550 nm) of fluorescence microscopy⁵¹. Although super-resolution microscopy can improve the resolution to 5–50 nm (ref. 52), it cannot reveal the folding trajectory of chromatin molecule⁵³. These challenges have been addressed with the development of multiplexed FISH, which enables the sequential labelling, imaging and distinguishing of many genomic loci on the same chromosome effectively at super-resolution⁴⁶ (Fig. 1c). Multiplexed FISH has also been used in combination with a barcoding system, further expanding the number of targeted loci to genome-wide coverage^54–59. Chromatin tracing by multiplexed FISH is intrinsically an in situ single-cell single-molecule approach that is compatible with intact tissue preparation, captures multiway chromatin interactions and can be combined with multimodal imaging of other biomolecules. These features make chromatin tracing via multiplexed FISH an important strategy for mapping 3D genome architecture, complementing sequencing-based methods (Table 1).

Table 1 |.

Technological features of major sequencing-based and fluorescence in situ hybridization-based methods for mapping 3D chromatin organization

Methods	Resolution limit	Coverage	Multiway interactions	Single-trace trajectory	Single-cell information	Spatial information	Multimodality
Sequencing-based methods
3C12	1 kb	One vs one	No	No	No	No	No
4C154	2 kb	One vs all	No	No	No	No	No
5C155	1 kb	Many vs many	No	No	No	No	No
Capture-C33,34	1 kb	Many vs all	No	No	No	No	No
Hi-C13,24	1 kb	All vs all	No	No	No	No	No
Micro-C14–16	200 bp	All vs all	No	No	No	No	No
ChIA-PET31,156	500 bp	All vs all	No	No	No	No	Yes (one protein)
HiChIP32	500 bp	All vs all	No	No	No	No	Yes (one protein)
GAM19,20	40 kb	All vs all	Yes	No	No	No	No
SPRITE17,18	25 kb	All vs all	Yes	No	No	No	No
scHi-C38,40,42	100 kb	All vs all	No	Inferred for hapLoid ceLL	Yes	No	No
Dip-C44	10s of kb	All vs all	No	Inferred for dipLoid ceLL	Yes	No	No
scSPRITE43	10s of kb	All vs all	Yes	No	Yes	No	No
ChIA-Drop157	10s of kb	All vs all	Yes	No	Yes	No	Yes (one protein)
HiRES158	10s of kb	All vs all	No	Inferred for dipLoid ceLL	Yes	No	Yes (transcriptome)
MUSIC159	10s of kb	All vs all	No	No	Yes	No	Yes (transcriptome)
GAGE-seq160	10s of kb	All vs all	No	No	Yes	No	Yes (transcriptome)
LiMCA161	10s of kb	All vs all	No	Inferred for dipLoid ceLL	Yes	No	Yes (transcriptome)
FISH-based methods
Conventional FISH162	~2 kb	One to severaL Loci	Yes (when >2 targets)	No	Yes	Yes	Yes (one to severaL transcripts and/or proteins)
Super-resolution FISH163–165	2.5 kb	One to severaL regions	Yes (when >2 targets)	No	Yes	Yes	Yes (one to severaL transcripts and/or proteins)
Chromatin tracing54–59,69,70,72,77,80	2.5 kb	Tens to ~100,000 Loci	Yes	Yes	Yes	Yes	Yes (one to >10,000 transcripts and/or one to tens of proteins)

Capture-C, chromosome conformation capture with capture probes; ChIA-Drop, chromatin interaction analysis by droplet-based and barcode-linked sequencing; ChIA-PET, chromatin interaction analysis by paired-end tag sequencing; Dip-C, diploid chromatin conformation capture; FISH, fluorescence in situ hybridization; GAGE-seq, genome architecture and gene expression by sequencing; GAM, genome architecture mapping; Hi-C, high-throughput chromosome conformation capture; HiChIP, high-throughput chromosome conformation capture with chromatin immunoprecipitation; HiRES, Hi-C and RNA-seq employed simultaneously; LiMCA, linking mRNA to chromatin architecture; Micro-C, micrococcal nuclease chromosome conformation assay; MUSIC, multinucleic acid interaction mapping in single cells; scHi-C, single-cell high-throughput chromosome conformation capture; scSPRITE, single-cell split-pool recognition of interactions by tag extension; SPRITE, split-pool recognition of interactions by tag extension; 3C, chromosome conformation capture; 4C, circular chromosome conformation capture; 5C, chromosome conformation capture carbon copy.

Chromatin tracing can also be done with in situ sequencing^60,61. For simplicity, this Primer mainly discusses the experimental principles, data processing pipeline, major applications, reproducibility and data sharing of chromatin tracing by multiplexed FISH. For chromatin tracing by in situ sequencing, we only briefly describe the unique features in its experimental design. We also discuss the limitations of current chromatin tracing technologies and potential mitigation strategies. Finally, we provide an outlook of potential development priorities related to chromatin trancing technology and its application over the next 5–10 years.

Experimentation

The basic procedure of chromatin tracing experiments involves labelling numerous genomic loci along chromatin, imaging their spatial locations and distinguishing their DNA sequences and then linking the positions of the targeted loci following the order of their sequences on the genomic map to reconstruct 3D chromatin structure. In chromatin tracing by multiplexed FISH, numerous genomic loci are labelled with a library of single-stranded DNA probes through base pairing after DNA denaturation in a primary FISH step (Fig. 2a). The spatial locations of the labelled loci are then visualized in a series of sequential secondary FISH steps using dye-labelled single-stranded DNA FISH probes, which simultaneously read out the DNA sequences (genomic identities) of the loci. In chromatin tracing by in situ sequencing, the labelling of loci is achieved either through a primary FISH procedure as mentioned earlier or through Tn5-transposase-mediated insertion of artificial DNA sequences into the genome. Image-based in situ sequencing then locates the FISH probes or inserted sequences.

Fig. 2 |. Technological improvement related to multiplexed fluorescence in situ hybridization.

a, Comparing probe preparation and signals of conventional and oligo-pool-based multiplexed fluorescence in situ hybridization (FISH). Conventional FISH probes are made from randomly fragmented cloning products, producing probes with variable lengths and hybridization quality, and probes targeting different loci need to be prepared separately. Multiplexed FISH probes are designed in silico and prepared using array-based synthesis in one pool, with uniform probe length and hybridization quality. The two-tiered design of multiplexed FISH probes allows imaging selected loci in different hybridization rounds. b, Methods used for switching signals between hybridization rounds in multiplexed FISH. Signals from the previous hybridization round need to be extinguished before hybridizing new secondary probes at different targeted loci in the next hybridization round. Commonly used signal removal methods include photobleaching the fluorophores on the secondary probes, washing off the fluorescent secondary probes with buffer containing high concentration of formamide, chemical cutting of a disulfide bond linking the fluorescent dye to the secondary probe and displacing the fluorescent secondary probes with non-fluorescent competitive oligos through strand displacement. c, Automatic flow and imaging platform required for multiplexed FISH. In each hybridization round, the flow system sequentially pumps hybridization, washing, imaging and signal removal buffers through the sample; the microscope takes images while the sample is in imaging buffer. Both the flow system and the microscope are automatically controlled by a computer system using pre-set parameters. BAC, bacterial artificial chromosome; TCEP, tris(2-carboxyethyl)phosphine.

In this section, we first discuss the experimental procedures of different chromatin tracing implementations, followed by a more in-depth discussion of how chromatin tracing is combined with other biomolecular imaging methods and genetic screening.

Chromatin tracing by multiplexed FISH

Conventional FISH uses fluorescent probes made from randomly fragmented cloning products to illuminate targeted genomic loci, which induces high variability in labelling efficiency and accuracy^62–64. Two important technical improvements to FISH probes facilitated multi plexed FISH-based chromatin tracing (Fig. 2a). The first is the use of probes designed in silico, taking into account template sequence and the desired probe length, melting temperature, guanine and cytosine (GC) content and the potential secondary structure, followed by array-based oligonucleotide syntheses and amplification using in vitro transcription and reverse transcription reactions^46,65. This probe preparation procedure enables strict control of the probe sequence at the base pair level, reducing off-target effects during hybridization⁵³. Multiplexed FISH probes may also be referred to as Oligopaint probes owing to their generation using array-based synthesis^66–68.

The second technical improvement is the use of a two-tiered design for hybridization of probes with genomic DNA. First, a library of primary probes is hybridized to all genomic targets; each primary probe contains a genomic targeting sequence and one or more overhanging readout sequences. Second, fluorescent-labelled secondary probes (also called readout probes) hybridize to the readout sequences on the primary probes. The genomic targeting sequences in primary probes are usually 28–42 nt in length and must be exactly complementary to their targeted genomic sequence with high melting temperature (>65 °C)^46,55,69. By contrast, the secondary probe binding sequences are designed to have minimal homology with the genomic sequence⁶⁵, are usually lower in melting temperature and shorter in sequence (15–20 nt)^46,55,69. Such design allows fast (<30 min) binding of the secondary probes to the overhanging sequences on the primary probes at low temperature without affecting the binding of the primary probes to the genomic DNA. The fluorescent signals from the secondary probes are imaged and then extinguished through photobleaching^46,70, formamide stripping^55,56, chemical cleavage⁷¹ or strand displacement^69,72, before the sequential hybridization with the next secondary probes (Fig. 2b). This strategy allows visualization of individual chromatin loci located in close proximity to each other along the genome using sequential imaging or simultaneous imaging utilizing different fluorescent channels, allowing for spot centre fitting with nanoscale accuracy. These steps show how multiplexed FISH overcomes molecular crowding, bringing the chromatin tracing resolution below the diffraction limit of light microscopy.

In real practice, formamide is added during the probe hybridization steps to reduce the effective melting temperature to easily achievable temperatures, such as 37–47 °C for primary probes and room temperature (25 °C) for secondary probes, while maintaining hybridization specificity⁷³. During secondary probe hybridization, formamide can be replaced by its non-toxic alternative, ethylene carbonate, without affecting secondary FISH signal quality^{55–57,71,74,75}. However, whether ethylene carbonate or other non-toxic denaturation chemicals are stable enough to replace formamide during the long (1–5 days) primary probe hybridization remains to be tested.

Implementation of multiplexed FISH requires specific equipment comprising three major components: an automated fluidic system used to pump buffers for hybridization, washing, imaging and signal removal in and out of the sample chamber, an automated microscope system capable of high-precision 3D fluorescence imaging with long-term stability and a computer system that automates and controls both the fluidic system and the microscope system using custom preset parameters (Fig. 2c). Details of the multiplexed FISH platform have been described before⁷⁶.

Depending on whether each targeted genomic locus is linked to the same or a combination of multiple readout sequences, chromatin tracing by multiplexed FISH can be categorized as sequential or barcoded.

Sequential chromatin tracing.

The first application of multiplex FISH for chromatin tracing used a sequential format to detect the spatial organization of TADs and A/B compartments along individual chromosomes in human IMR90 cells⁴⁶. Approximately 1,000 primary probes containing the same readout sequence targeted the 100 kb-sized central regions of each TAD, as previously defined by Hi-C experiments. The TADs were sequentially imaged two at a time using two different fluorescent channels, and their fluorescent spot centres were connected one after another based on their order on the genomic map to reconstruct the chromatin folding path at the TAD-to-chromosome scale (Fig. 3a). A later study used a similar strategy, sequentially imaging consecutive 30 kb segments rather than discrete TAD centres, to trace chromatin at sub-TAD scale within regions spanning multiple TADs, successfully visualizing TAD boundaries⁶⁹. Two finer-scale implementations of chromatin tracing, termed microscopy-based chromosome conformation capture (Hi-M) and optical reconstruction of chromatin architecture (ORCA), achieved chromatin tracing at 2.5–10 kb resolution within select regions of hundreds of kilobases in Drosophila embryos, reaching the scale of promoter-enhancer looping^72,77.

Fig. 3 |. Chromatin tracing methods based on multiplexed fluorescence in situ hybridization.

a, Hybridization and imaging procedures of sequential chromatin tracing based on the initial chromatin tracing study⁴⁶. In each hybridization round, two targeted loci are detected using secondary probes carrying two different fluorophores (orange and blue). In this cartoon, 30 targeted loci on the chromatin are imaged using 15 hybridization rounds. The centre positions of imaged signal spots are computationally linked based on the genomic coordinates of the targeted loci to reconstruct the 3D chromatin folding trajectory. b, Hybridization, imaging and decoding procedures of two barcoded chromatin tracing methods, DNA multiplexed error-robust fluorescence in situ hybridization (DNA MERFISH) and DNA sequential fluorescence in situ hybridization plus (DNA seqFISH+), based on their initial reports^54,55. Binding of secondary probes and decoding of the signals at four representative targeted loci (A, B, C and D) are shown. In DNA MERFISH, targeted loci are encoded with 100-bit binary barcodes. The binary digits or bits, 1 and 0, represent the presence or absence of the corresponding readout sequence on the primary probes. Two bits, one from the first 50 bits and another from the last 50 bits, can be imaged simultaneously in each hybridization round using two secondary probes carrying different fluorophores (orange and blue). In total, 50 hybridization rounds are required to capture the full 100-bit barcodes. In DNA seqFISH+, targeted loci are encoded with 5-digit pseudocolour barcodes, with 16 pseudocolour choices at each digit. Each pseudocolour represents a unique secondary probe binding in 16 hybridization rounds. In total, 80 hybridization rounds are required to capture the full 5-digit pseudocolour barcodes. Only one true colour channel is shown in this cartoon (orange). c, Combining chromatin tracing with multimodal imaging of RNA and protein molecules as well as genetic screening. Chromatin tracing can be integrated with spatial profiling of RNA molecules using similar multiplexed RNA FISH methods^54,58,70,72. To combine chromatin tracing with multiplexed protein imaging, oligo-conjugated antibodies are often used to allow sequential imaging of different protein molecules in multiple hybridization rounds as in multiplexed FISH^55–57. To combine chromatin tracing with pooled genetic screening, each single-guide RNA (sgRNA) is expressed together with a barcode RNA and the types of sgRNAs expressed in each cell can be identified through decoding the co-expressed barcode sequence using barcode amplification by rolling circle and fluorescence in situ hybridization (BARC-FISH)⁹⁶. CMV, cytomegalovirus promoter; Hyb, hybridization round; LTR, long terminal repeat; U6, U6 promoter. Part b adapted from ref. 55, Springer Nature Limited.

The first application of chromatin tracing in mammalian tissue was achieved using an integrative approach known as multiplexed imaging of nucleome architectures (MINA), which traced the chromatin folding of Chr19 in mouse fetal liver⁷⁰. MINA combines multiscale chromatin tracing of discrete TADs and consecutive fine segments of 5 kb within the same cells, facilitating mapping of 3D structures of the same chromosome at different scales.

Finally, a subgroup of sequential multiplexed FISH methods, collectively known as volumetric chromatin tracing, uses single-molecule localization microscopy (SMLM)⁷⁸ to image tens to thousands of single-molecule localizations for each targeted genomic segment. This enables the reconstruction of occupied volumes and overlapping fractions between targeted genomic segments⁷⁹. Representative volumetric chromatin tracing methods use Oligopaint stochastic optical reconstruction microscopy (OligoSTORM) or Oligopaint DNA-based point accumulation for imaging in nanoscale topography (OligoDNA-PAINT), which enabled ≤20 nm resolution volumetric imaging of genomic DNA through stochastic blinking of fluorophores or transient binding of fluorescently labelled oligos, respectively^80,81.

Theoretically, sequential chromatin tracing can cover C × N genomic loci, in which C is the number of available fluorescent colour channels for imaging genomic loci and N is the number of secondary probe hybridization rounds. Assuming each hybridization takes 30 min to complete, a 100-loci, two-colour sequential multiplexed FISH chromatin tracing experiment will require 50 rounds of hybridization, taking ~25 h. Adding the imaging time, usually 30–60 s for each field-of-view (FOV) with z-stacking, data collection may take several days. Therefore, although sequential multiplexed FISH is suitable for tracing individual whole chromosomes at the TAD-to-chromosome scale or for tracing select genomic regions at finer scales, it is not sufficiently high throughput for genome-wide chromatin tracing, which requires imaging thousands of loci. This drawback has led to the development of barcoded chromatin tracing methods, allowing parallel identification of thousands of genomic loci with relatively few hybridization-imaging rounds^54–59.

Barcoded chromatin tracing.

In barcoded chromatin tracing, each targeted genomic locus is assigned a unique binary barcode, which is physically encoded onto primary probes through a combination of readout sequences. The binary digits, 1 and 0, in each barcode represent the presence or absence of the corresponding readout sequence on the primary probes, which in turn dictates the presence or absence of the fluorescent signal in the corresponding imaging round (Fig. 3b). Identification of targeted loci requires reading all digits in their barcodes, which relies on the signals of multiple, rather than just one, rounds of imaging in barcoded chromatin tracing. Theoretically, 2^L − 1 genomic loci can be identified in total, in which L is the number of binary digits (bits) of each barcode. In reality, DNA multiplexed error-robust FISH (DNA MERFISH), the first barcoded chromatin tracing method, only uses barcodes with equal Hamming weights to control the error rate of decoding⁵⁴. In this way, C(L, W) barcodes could be generated in total, in which C is the combinatorial number, L is the number of binary digits and W is the weight, or number of 1 bits, of the barcode. DNA MERFISH offers considerable error robustness as a genomic locus would need to both lose a fluorescent signal in one round and gain a fluorescent signal in another round to be wrongly decoded as a different genomic locus; having two error events on the same genomic locus is unlikely. Additionally, barcode shuffling has been performed to ensure an approximately equal number of loci imaged per bit for each chromosome and to maximize the genomic distance between loci with a 1 bit at the same code digit. This enabled DNA MERFISH to map 1,041 genomic loci across the genome at ~2.5 Mb resolution in human IMR90 cells within 50 hybridization rounds using two-colour imaging⁵⁴.

In parallel, a second barcoded chromatin tracing method was developed, known as DNA sequential FISH plus (DNA seqFISH+)^55,56. In this approach, each colour channel is first expanded to a much larger palette of pseudocolours through sequential hybridization and imaging, with each targeted locus assigned a barcode with a unique order of pseudocolours. Theoretically, S^L loci could be distinguished at maximum, in which S is the number of pseudocolours and L is the barcode length. DNA seqFISH+ also uses an error correction scheme by adding one extra digit (L + 1 digits in pseudocolour barcode in total)⁸². Experimentally, DNA seqFISH+ uses padlock circularization^83,84 and bis-succinimide ester-activated polyethylene glycol crosslinking to stabilize primary probes and further reduce the chance of signal loss in extensive sequential FISH. Through these strategies, DNA seqFISH+ has achieved 3,660-loci genome-wide chromatin tracing at 1 Mb resolution in mouse embryonic stem cells⁵⁵ and mouse cerebral cortex tissues⁵⁶ within 80 hybridization rounds using three-colour imaging.

Both DNA MERFISH and DNA seqFISH+ have allowed efficient imaging of thousands of genomic loci at megabase resolution using no more than 100 hybridization rounds. By contrast, a three-colour sequential chromatin tracing approach takes more than 200 hybridization rounds to image approximately 650 genomic loci⁵⁴, indicating that barcode-based methods can save a significant amount of hybridization and imaging time when tracing thousands of genomic loci. However, the distance between adjacent genomic loci is likely to be smaller than the diffraction limit of conventional microscopy when those loci are within 1 Mb intervals⁵¹, prohibiting simultaneous imaging in the same barcode digit in barcoded chromatin tracing. Therefore, when fine-scale chromatin tracing is needed, sequential chromatin tracing remains the best option. On the basis of this principle, a two-layered strategy is used in the latest application of DNA seqFISH+ for tracing more than 100,000 loci across the genome at 25 kb resolution in mouse brain — the whole genome is first separated into two thousand 1.5 Mb segments that can be identified by DNA seqFISH+ in 36 hybridization rounds using three-colour imaging, followed by each segment being further separated into 25 kb segments examined using sequential chromatin tracing in 60 hybridization rounds⁵⁷.

Chromatin tracing by in situ sequencing

In situ sequencing-based chromatin tracing methods include Oligopaint fluorescent in situ sequencing (OligoFISSEQ)⁶¹ and in situ genome sequencing (IGS)⁶⁰. OligoFISSEQ uses primary FISH probes to target genomic sequences, as in multiplexed FISH, followed by either sequencing-by-ligation or sequencing-by-synthesis to readout barcode sequences on the primary probes in situ. In comparison to FISH-based readout, both sequencing-by-ligation and sequencing-by-synthesis can achieve comparable barcode multiplicity using much shorter readout sequences⁶¹.

By contrast, IGS uses Tn5 transposase to randomly incorporate DNA-sequencing adapters carrying amplicon-specific unique molecular identifiers (UMIs) into a fixed genome. The adapters, UMIs and their linked genomic sequences are amplified in situ with rolling circle amplification. The spatial positions and sequences of inserted UMIs are then detected using sequencing-by-ligation in situ, followed by dissociation of the amplicons and PCR amplification to obtain a library suitable for paired-end ex situ sequencing. The obtained ex situ sequencing reads are aligned to a known genome to identify the genomic sequence linked to each UMI, which is then matched to the spatial position of the UMI initially recorded in the images obtained by in situ sequencing. IGS can be thought of as bridging sequencing and imaging technologies with the potential to link sequence-level variations such as single-nucleotide polymorphisms (SNPs) with spatial organization. Through IGS, the whole genome of human fibroblast and 2–4 cell mouse embryos has been traced with a resolution of several megabases⁶⁰.

Application of in situ sequencing technologies for fine-scale chromatin tracing remains to be examined, and neither technique has been applied to tissue sections. The long ligation or polymerase reaction time needed for sequencing-by-ligation and sequencing-by-synthesis, as well as the limited commercial availability of reagents, may limit the throughput and adaptation of the techniques. These factors may have contributed to the wider and more advanced adaption of FISH-based chromatin tracing. However, chromatin tracing by in situ sequencing has its advantages, as mentioned here, and is worth continued development and optimization.

Combination with other imaging approaches

Chromatin tracing is routinely combined with nuclear staining using 4′,6-diamidino-2-phenylindole (DAPI) or SYTOX to define nuclear regions⁷⁰. In some cases, cell envelope stain such as dye-labelled wheat gem agglutinin is also included to facilitate cell segmentation in the image analysis. Chromatin tracing is also often combined with imaging of other biomolecules, particularly RNA and proteins, in multimodal or multi-omic imaging (Fig. 3c).

Image-based spatial transcriptomics.

Combining chromatin tracing with RNA imaging facilitates the building of structure–function connections between DNA folding and transcriptional activity. Gene expression information also aids in distinguishing cell types and states in complex tissue microenvironments. The technical homology in DNA and RNA multiplexed FISH approaches makes FISH-based chromatin tracing and FISH-based spatial transcriptomics compatible⁸⁵.

As RNA molecules are prone to degradation, most combinations of chromatin tracing with RNA FISH map RNA molecules first, followed by RNA digestion and probing of DNA. This two-step approach was first introduced with ORCA, which imaged 29 RNA transcripts sequentially by single-molecule FISH before performing chromatin tracing⁷². A similar procedure was adopted by subsequent implementations, including DNA MERFISH^54,58 and DNA seqFISH+ (refs. 55–57), for integrating genome-wide chromatin tracing and larger-scale transcriptome profiling. However, the two-step approach is experimentally laborious as it requires dissembling the sample from the microscope for RNA digestion and DNA denaturation before DNA probe hybridization. In addition, more data analysis steps are required to achieve accurate registration of images from the separated RNA and DNA imaging steps based on their DAPI or other shared fiducial marker signals. To simplify the process, a one-step approach was developed using MINA, in which RNA MERFISH probes and chromatin tracing probes were simultaneously hybridized to their targets after DNA denaturation and then imaged in the same microscopy session, thereby simplifying experimental and data analysis procedures^70,86. The resulting high correlation between transcript copy number obtained by MINA and fragments per kilobase of transcript per million mapped reads (FPKM) detected with bulk RNA-seq indicates that RNA can be maintained during the harsh DNA denaturation steps, with RNase inhibitor added afterwards and maintained throughout the hybridization process. However, MINA requires BLASTing of probes against both genome and transcriptome sequences to avoid potential cross-hybridization; chromatin tracing probes should not bind to transcripts or cross-hybridize with RNA probes. This requirement may limit the number of targeting sequences available for primary probe design targeting certain genes or genomic regions. The one-step strategy cannot be used for intron RNA FISH, as intron FISH probes will bind to the corresponding DNA locus and generate artificial transcription burst signals^54,87.

In other methods, such as Hi-M, RNA FISH signals are amplified via tyramide signal amplification, which can be stably sustained following the DNA FISH procedure⁷⁷, although this approach is only able to image one RNA species. Another strategy involves the use of acrydite-modified RNA FISH probes that can be covalently linked to a polyacrylamide gel⁸⁸. In this procedure, RNA FISH hybridization is performed first, followed by embedding of the cell sample into a polyacrylamide gel, which occurs before chromatin tracing primary probe hybridization. This procedure also permits imaging of DNA and RNA targets in the same microscopy session and is compatible with intron RNA FISH.

Protein imaging.

Chromatin tracing can be integrated with immunofluorescence staining, which enables the investigation of the spatial relationship between genome loci and subnuclear landmarks and structures (such as the nuclear lamina, nuclear speckles and nucleoli), epigenetic marks and regulators, and transcriptional machinery. This approach can also be used for examining the influence of the cell cycle on 3D chromatin structure when cell cycle markers are imaged.

Chromatin tracing combined with immunofluorescence staining was successfully used to distinguish the active and inactive X chromosomes (Xa and Xi, respectively) in a female human cell line via the immunofluorescence staining of macroH2A.1, which is enriched on Xi, and performing allele-specific chromatin tracing analyses of the two X chromosomes⁴⁶. The primary macroH2A.1 antibody and fluorescent secondary antibody were applied after DNA FISH probe hybridization, and immunofluorescence signals were subsequently imaged and photobleached before starting sequential readout imaging of the chromatin tracing signals. In a later sub-TAD scale chromatin tracing study, immunofluorescence staining of the CDK inhibitor geminin was performed before primary DNA probe hybridization with a post-fixation step between staining and hybridization to stabilize the immunofluorescence signals⁶⁹. These strategies have been adopted in other studies combining chromatin tracing with conventional immunofluorescence staining^54,70,89. However, the number of proteins targeted using conventional immunofluorescence is limited by the number of colour channels available, and photobleaching of the immunofluorescence signal is time-consuming. Prior assessments are needed to verify that each antibody is still able to target its respective antigen following the HCl treatment and heat denaturation steps of the primary DNA FISH, or whether the antibody can be fixed to sustain the harsh conditions involved in FISH.

Oligo-conjugated antibodies are used in DNA seqFISH+ to improve the protein target throughput of chromatin tracing combined with protein profiling^55–57. The main principle is to conjugate the primary antibodies with DNA oligos carrying readout sequences that can bind to secondary probes, similar to chromatin tracing primary probes^90,91. After primary antibody staining, the oligo-conjugated antibodies are fixed on sample through bis-succinimide ester-activated polyethylene glycol crosslinking. In this way, many proteins can be sequentially imaged in multiple hybridization rounds, as in multiplexed FISH. Using oligo-conjugated antibodies, DNA seqFISH+ has demonstrated successful combination of chromatin tracing with sequential immunofluorescence imaging of up to 65 oligo-conjugated antibodies⁵⁷. However, in this approach, prior assessment is required to ensure that each antibody still retains its target binding affinity and specificity after oligo conjugation.

Combination with genetic screening

Identification of factors regulating 3D chromatin folding is essential for understanding the molecular mechanisms of chromatin organization and, more importantly, for providing potential drug targets involved in disease-causing aberrant chromatin folding. Previously known 3D genome regulators were discovered by perturbing one candidate gene at a time^92,93. Development of sequencing-based methods for genetic screening of 3D genome regulators has been challenging owing to the high cost and low throughput of such approaches. In comparison, imaging-based methods can provide highly cost-efficient 3D genome phenotyping for high-throughput screening. Along this line, a FISH-based small interfering RNA (siRNA) screening method, known as high-throughput imaging position mapping (HIPMap), has been developed, enabling automated FISH imaging in 384-well plates containing different siRNA perturbations for systematic identification of factors regulating the subnuclear positioning of one genomic locus⁹⁴. A recent publication reported a plate-based siRNA screening method known as high-throughput DNA or RNA labelling with optimized Oligopaints (HiDRO), which uses Oligopaint images of two adjacent TADs as the readout to discover regulators of the TAD organization⁹⁵. Another approach, Perturb-tracing, has combined chromatin tracing with CRISPR screening technology⁹⁶. Instead of using multiwell plates to separate different siRNA perturbations, as in HIPMap and HiDRO, Perturb-tracing developed a cellular barcoding and in situ decoding method, termed barcode amplification by rolling circle and FISH (BARC-FISH), which enables direct identification of the genetic perturbation at the single-cell level in a pooled CRISPR screen by imaging. Using chromatin tracing for phenotyping in the pooled screen, Perturb-tracing has enabled scalable and high-content discovery of factors regulating higher-order 3D chromatin folding at multiple scales, including TADs, chromatin compartments, chromosomal territories and the whole nuclear genome organization.

Results

Chromatin tracing data analysis includes four major components: image processing, quality control (QC), downstream analyses and validation of results (Fig. 4). Here, we focus on introducing the analysis of multiplexed FISH data collected using diffraction-limited epifluorescence or confocal microscopes, which are currently the major chromatin tracing data types. The analysis pipeline introduced here can largely be adapted to other types of chromatin tracing data such as the ones generated via in situ sequencing methods.

Fig. 4 |. Chromatin tracing data analysis pipeline.

a, Raw signals of 4′,6-diamidino-2-phenylindole (DAPI) staining and chromatin tracing with sequential or barcoded strategy. b, Nuclear segmentation based on DAPI staining and removal of noise signals outside the nuclear region. c, Identification of real signal spot centres for spots with signal intensity profiles that can be fitted to a 3D Gaussian function. d, Drift correction between hybridization rounds based on signals of fiducial markers. e, Chromatic shift correction in multicolour imaging. f, Correction of spectral crosstalk between different colours and the signal intensity difference between hybridization rounds. g, Decoding of the genomic sequence identities of fitted foci in barcoded chromatin tracing. h, Separation of fitted foci belonging to the different copies of the same chromosome in the same cell nucleus. i, Linking fitted foci on the same chromatin based on genomic order to reconstruct chromatin trace. j, Quality control based on, for example, detection efficiency and spatial-versus-genomic distances. k, Downstream analysis plots, for example, averaged and single-cell interloci distance maps. l, Results validation by comparing the chromatin tracing data with high-throughput chromosome conformation capture (Hi-C) data. Hyb, hybridization round. Part e adapted from ref. 119, Springer Nature Limited. Part k adapted from ref. 53, CC BY 4.0.

Image processing

Raw imaging quality check.

High-quality raw images are essential for successful downstream analyses. A complete set of chromatin tracing data usually contains three components: chromatin tracing images collected with z-stacking and fiducial markers (such as fluorescent beads or anchoring genomic loci) in each FOV, nuclear staining images (for example, with DAPI) for nuclear segmentation and chromatic shift calibration images if more than one colour channel is used for chromatin tracing. To reconstruct a complete and super-resolved chromatin trajectory in 3D, chromatin tracing images are usually collected with a z-step size smaller than the diffraction limit of microscopy (for example, 200 nm z-step size) and with 10 μm in depth to cover at least one layer of nuclei⁹⁷. For sequential chromatin tracing, two (diploid cells) or multiple (polyploid cells) chromatin tracing signal spots are generally observed within each nucleus in each imaging round, given that the nucleus is completely covered by the z-stacking, whereas tens to hundreds of signal spots should be observed for barcoded chromatin tracing, depending on the barcoding scheme (Fig. 4a). The intensity of each chromatin tracing signal spot depends on the number of probes hybridized at each targeted locus, yet should be bright enough to be distinguished from the surrounding background and roughly fit a Gaussian profile as a diffraction limited spot⁹⁸.

Nuclear segmentation.

As the genome usually resides in the nucleus, nuclear segmentation can help filter chromatin tracing signals and remove spots outside the cell nuclei (Fig. 4b). Nuclei stained by DAPI or SYTOX can be computationally segmented using marker-controlled watershed segmentation⁹⁹. More precise nuclear region segmentation in tissue samples can be achieved by machine-learning-based segmentation tool kits such as ilastik¹⁰⁰ and CellPose^101–103, which were successfully applied in several chromatin tracing studies^55–58.

Foci fitting.

Processing of chromatin tracing images starts from fitting the spatial coordinates of the signal spots. First, local intensity maxima are identified in the images to determine potential spot locations. Then, the intensity profile surrounding each maximum is fitted to a 3D Gaussian function to identify the centre position of the signal spot with nanoscale accuracy, as well as the centre signal intensity and the fitting quality, which can help filter fitted spots⁴⁶ (Fig. 4c). Similar 3D Gaussian fitting is also used for determining the positions of fiducial beads and fiducial marker genomic loci in drift correction and chromatic shift calibration. Recent studies have started using alternative algorithms, such as a 3D radial centre algorithm¹⁰⁴, to achieve faster foci fitting⁵⁹. The spatial coordinates of fitted chromatin tracing signal spots are then subjected to sample drift correction and chromatic shift correction before being used for later analyses.

Drift correction.

Fluidic exchange, thermal drift of the imaging sample assembly and limited positioning accuracy of the sample stage of the microscope can cause small spatial shift between images of the same FOV in different hybridization rounds. To correct the sample drifts, fiducial beads attached to sample coverslips^46,54,69 or fiducial marker genomic loci^55,56,72 are imaged in each FOV with z-stacking and in every hybridization round, and their 3D positions are subsequently used for cancelling sample drift (Fig. 4d).

Chromatic shift calibration and correction.

Multicolour imaging is commonly used in chromatin tracing to increase genomic throughput and to speed up data collection. However, the chromatic aberrations caused by differences in the refraction indices of fluorescent light of different wavelengths can lead to imperfect alignment of spatial positions measured in different colour channels. Such chromatic shift is often non-negligible in chromatin tracing. Therefore, chromatic shift calibration is usually required in multicolour chromatin tracing experiments. This can be achieved through labelling and comparing signal positions of the same set of multicolour beads⁸⁹ or genomic loci⁵⁴ in different colour channels that are imaged simultaneously (Fig. 4e).

Signal intensity correction.

For barcoded chromatin tracing, its decoding depends on the analyses of signal intensities among different hybridization rounds and often requires additional intensity corrections (Fig. 4f). One type of intensity correction is spectral crosstalk correction, which adjusts the intensity of one channel to remove emission bleed-through from another channel^54,59. Another type of intensity correction is intensity uniformity correction, which normalizes the average intensity in different hybridization rounds⁵⁹.

Decoding.

For sequential chromatin tracing, fitted and corrected signal spots obtained from the steps mentioned earlier are ready for trace linking. For barcoded chromatin tracing, the spot identities need to be decoded first (Fig. 4g). Briefly, the 3D positions of the spots from all imaging rounds of each FOV are compared to identify the same spots reappearing in different imaging rounds. These spots are then subjected to barcode matching with the appearance of signal as a 1 bit and the absence of signal as a 0 bit. To find the best matching barcode, an iterative scoring strategy taking into account signal intensities, spatial matching precision and chromosomal territory organization has been used^54,59.

Homologous chromosome separation.

A challenging aspect of chromatin tracing data analysis is the separation of the signal spots belonging to the two homologues of the same chromosome, particularly when the two homologues are spatially adjacent to each other (Fig. 4h). Some standard spatial clustering algorithms, including K-means and DBSCAN, have been used to group the signal spots in several early chromatin tracing studies^55,56,69,77. To achieve accurate separation of homologues close to each other, improved K-means and DBSCAN algorithms have been proposed to take into account the spatial positions, genomic coordinates and clustering fraction of assigned signal spots to refine the initial clustering results^54,57. In addition, algorithms based on polymer physics theory¹⁰⁵ have been used to tackle the uncertainty in the number of homologues as a result of DNA replication, aneuploidy or incomplete inclusion of nuclei in tissue sections⁵⁸.

Trace linking.

To construct chromatin folding traces in sequential chromatin tracing, signal spots in each round of imaging are linked to the closest spot in the next imaging round — which visualizes the next locus on the genomic map — within certain maximal distance limits (Fig. 4i). In TAD-to-chromosome-scale tracing, 2 μm is often set as the maximal limit to confine linked TADs in individual chromosomal territories, as adjacent TAD distances rarely exceed such a limit⁴⁶. Shorter distance limits are required for finer-scale tracing experiments^69,72. A recently published trace linking algorithm uses a variable distance limit based on the genomic distance between adjacent loci¹⁰⁵. Occasionally, not every targeted locus can be detected during the initial trace reconstruction, and further trace refinement is required to increase the trace integrity. Briefly, the distance limit is loosened, and the original fitted signal spots that have not been used for initial trace linking are revisited and used to fill gaps in territories determined by the initial traces where they fall⁷⁰. For barcoded chromatin tracing, a trace refinement step is not necessary as the iterative scoring step during decoding has already maximized the recovery efficiency of all detected signal spots.

Quality control

The most commonly used QC metric for chromatin tracing is detection efficiency — the proportion of expected genomic loci that are detected along each detected chromatin trace (Fig. 4j). Sequential chromatin tracing has been shown to achieve a detection efficiency of ~90% in human IMR90 cell culture even after more than 200 rounds of hybridization and washes⁵⁴. For barcoded chromatin tracing, the detection efficiency of DNA MERFISH reached ~80% in IMR90 cell culture⁵⁴ and 10–60% in mouse brain tissue⁵⁸. Similarly, the detection efficiency of DNA seqFISH+ was ~50% in mouse embryonic stem cells⁵⁵ and decreased to ~20% in mouse brain tissues^56,57. In general, tissues provide lower detection efficiencies compared with cell cultures, likely because of the incomplete inclusion of whole nuclei in tissue sections, high fluorescent background, optical crowding owing to smaller nuclear size and reduced hybridization efficiency owing to variable tissue preparation and fixation conditions⁵⁷.

With barcoded methods, the false-positive rate is another important QC measure that reflects decoding authenticity. Only a subset of barcodes from the entire barcode pool is used for encoding targeted genomic loci, whereas the remaining barcodes are referred to as blank barcodes. After decoding, the barcodes that have been matched to at least one signal foci are called on-target barcodes. As in previous multiplexed FISH-based spatial transcriptome profiling methods^65,87, the false-positive rates in chromatin tracing signal decoding are calculated as the median on-target blank barcode counts divided by median on-target total barcode counts. For DNA seqFISH+, the false-positive rate was reported as 1.2% in mouse embryonic stem cells⁵⁵ and 4.3% in mouse brain tissues⁵⁷.

For genome-wide chromatin tracing, other QC metrics include whether intrachromosomal distances are in general shorter than inter-chromosomal distance on a whole-genome interloci spatial distance matrix, consistent with the chromosomal territory organization^54–56,59. Furthermore, in a spatial-versus-genomic distance plot for pairs of intrachromosomal loci, larger chromosomes with longer genomic distances should in general reach larger intrachromosomal spatial distances than smaller chromosomes^55,56,59.

Downstream analyses

Downstream analyses transfer the spatial positions of genomic loci and chromatin traces to biological meaningful interpretations. Although the choice of analyses is highly dependent on the biological question, here we summarize some commonly performed downstream analyses and discuss their biological implications.

Spatial distance, contact and looping.

Every chromatin tracing study calculates the interloci spatial distance, which is defined as the Euclidean distance between the fitted centre positions of targeted genomic loci. Spatial distances between all loci pairs from one chromosome copy in one cell can generate pairwise spatial distance matrix that reflects chromatin structure at the single-chromosome-copy and single-cell resolution, whereas the mean/median value of spatial distances in all cells can generate a mean/median spatial distance matrix that reflects chromatin folding architecture at the population-averaged level.

The spatial distance information directly obtained from chromatin tracing experiments could be transformed to spatial contact frequency, similar to the results stemming from Hi-C experiments. The key step is to select a distance threshold below which two genomic loci are considered as being in contact. In sub-TAD-scale tracing, two segments are often considered as being in contact when their spatial distance is less than 150 nm (ref. 69). Elevated contact frequency or decrease in spatial distance between pairs of genomic loci in comparison to adjacent genomic regions has been used as a metric to call chromatin loops such as promoter–enhancer loops^70,72,106.

Chromatin compaction.

The interloci spatial distances of a chromatin trace overall directly reflect the extent of chromatin compaction^59,96. Alternatively, the compaction level can be measured with the radius of gyration of the chromatin trace using the following equation^96,107:

$$\text{Radius}\text{of}\text{gyration}=\sqrt{\frac{1}{n}\sum _{i=1}^n{d}_i^2},$$ (1)

in which n is the number of detected loci in the given trace and d_i is the spatial distance between locus i and the centroid of the whole trace.

A/B compartment analyses.

Initially revealed by ensemble-averaged Hi-C results, the interphase genome is segregated into A and B compartments corresponding to transcriptionally active and inactive chromatin regions, respectively¹³ (Fig. 1a). Identification of the A/B compartments using ensemble-averaged chromatin tracing data was performed in an initial chromatin tracing study⁴⁶ and the approach has been used in subsequent analyses at the compartment level^{54–57,59,70,88,89,108}. Briefly, the mean spatial distances matrix is first normalized to the expected spatial distances at each genomic interval — calculated by fitting a power-law function to the spatial-versus-genomic distance plot — to obtain a normalized distance matrix. A Pearson correlation matrix is then constructed with the Pearson correlation coefficient between each pair of rows or columns of the normalized distance matrix, followed by the principal component analysis of the Pearson correlation matrix. Depending on the positive and negative value attributes of the coefficients of the first principal component, the traced loci along the chromosome can be sorted into two compartments. The compartment showing higher active epigenetic marks or gene density is generally classified as the A compartment, whereas the other is classified as the B compartment. Individual chromosomes can be spatially segregated into two arms by the centromere region, and the first principle component may capture this organization rather than the A/B compartments⁴⁶. Therefore, in some later studies, A/B compartment identification was performed separately for each arm^54,58.

The polarization index quantifies the degree of polarized separation of A/B compartments along single-chromosome traces in single cells⁴⁶. It is calculated as the geometric mean of the non-shared proportions of compartments A and B using the following equation:

$$\text{Polarization}\text{index}=\sqrt{\left(1-\frac{V_s}{V_A}\right)\left(1-\frac{V_s}{V_B}\right)},$$ (2)

in which V_A, V_B and V_S represent the volumes of 3D convex hulls constructed by the A compartment, the B compartment and their overlapping space, respectively. A polarization index of 1 indicates that the A and B compartments are completely separated from each other in a side-by-side manner, whereas a polarization index of 0 indicates that A and B compartments completely overlap, or one is wrapped around by the other.

The local A/B density ratio quantifies the purity of loci found in compartments A and B that surround a genomic locus in a single cell⁵⁴. To calculate the local A/B density ratio, the spatial distances of all detected loci surrounding a specific genomic locus within a distance threshold are converted to density weights based on a Gaussian probability density function. The local densities of compartment A and compartment B loci surrounding that specific locus are then calculated as the sums of corresponding density weights, and the ratio between the two densities is calculated as the local A/B density ratio. Higher local A/B density ratio reflects purer compartment A loci within the surround region.

TAD and single-cell domain analyses.

The spatial distance matrix generated from chromatin tracing data shows block-like structures representing population-averaged TADs or single-cell domains. Generally, when calling ensemble-averaged TAD and sub-TAD boundaries from the mean or median distance matrix, an insulation score is calculated based on the difference between intradomain and interdomain mean or median distances within a sliding window moving along the diagonal of the matrix. The local maxima of insulation scores are then called and defined as boundary sites^54,88,109. By contrast, when calling single-cell domains from the single-trace distance matrix, a ratio between spatial distances in sliding windows upstream and downstream of a locus is calculated, followed by calling of local maxima points of the ratio values as domain boundaries^54,69,89.

Presentation of results.

The 3D chromatin trace conformations are often collapsed into distance matrices for data presentation. The mean or median spatial distance matrix can help visualize population-averaged structures such as loops, domains and compartments, whereas the distance matrices of single traces can help reveal the variations of these structures. In addition, 3D trace trajectories are often plotted to provide direct visualization of the 3D structures of single chromatin molecules (Fig. 4k). Downstream analysis of chromatin tracing data can further take advantage of clustering and dimensionality reduction visualization strategies developed in the single-cell sequencing analysis to link chromatin folding features (such as super-enhancer interactions and the local A/B density ratio) with biological functions (such as transcriptional activity and tumour progression states) at the single-cell level^59,110.

Validation of results

Comparison of chromatin tracing data with data obtained using Hi-C or Micro-C is common practice for validating results. High-quality chromatin tracing results can achieve a correlation coefficient >0.8 between the spatial distances and Hi-C contact frequencies on a log–log plot and will generate similar A/B compartment and TAD patterns as seen in Hi-C results^46,70 (Fig. 4l). In addition to Hi-C, chromatin immunoprecipitation sequencing results of CCCTC-binding factor (CTCF) and cohesin subunits have been used to validate TAD boundaries identified by chromatin tracing⁶⁹. Chromatin immunoprecipitation sequencing data can also help validate imaging-based epigenetic profiles generated by combined chromatin tracing and multiplexed immunofluorescence^55–57.

Compared with chromatin tracing, which targets a preselected panel of genomic loci, sequencing-based 3D genomics methods by default have much less biased genome-wide coverage and can achieve very high resolution at the population-averaged level. However, at the single-cell and single-chromosome-copy level, the resolution of sequencing-based methods can be limited by data sparcity¹¹¹. By contrast, in high-quality chromatin tracing data (>80% detect efficiency), each targeted locus is visualized on almost every imaged allele, allowing for single-cell and single-molecule mapping of chromatin structure at the resolution defined by the genomic intervals of the targeted loci. Although consistencies between chromatin tracing and Hi-C results have been repeatedly shown in multiple studies^46,54,69, the harsh DNA denaturation procedure of chromatin tracing might still affect the accuracy of fine-resolution interactions detected by chromatin tracing, particularly when samples are not sufficiently fixed. This can cause discrepancies between chromatin tracing spatial distances and Hi-C contact frequencies, particularly at the scale of promoter–enhancer contacts¹¹². In addition to single-cell information and genomic resolution and coverage, other factors can also affect the choice of imaging-versus-sequencing methods for different research questions, including the requirement of spatial and/or multiway interaction information, the accessibility of required platform, the complexity of data analysis and the throughput of assay (Table 2). The fixation or crosslinking step in both chromatin tracing and sequencing-based experiments may cause artificial protein puncta or missed transient chromatin interactions^113,114. Live-imaging approaches allow visualization of dynamic chromatin interactions in live cells through engineering fluorescently labelled protein arrays on DNA^115–118. However, the live-imaging approaches are largely limited by their target-number throughput and currently are mainly used to study a few selected chromatin regions and interactions at a time. Given the pros and cons of different methods, it is often beneficial to combine results from different methods to obtain a comprehensive understanding of chromatin structures.

Table 2 |.

Comparison of chromatin tracing methods and sequencing-based methods in studying 3D chromatin structure

Chromatin tracing methods (multiplexed FISH-based)	Sequencing-based methods
Advantages
Direct measurements of 3D coordinates of and physical distances between genomic loci Intrinsic single-cell and single-molecule/trace methods Maintains spatial information in cells and tissues Maintains multiway interaction information Resolution is predefined and not affected by sample size Easy to combine with multimodal imaging of multiple RNA and protein molecules High throughput in single-cell data collection (tens of thousands of cells per sample)	Unbiased mapping of chromatin interactions genome-wide Can reach <1 kb resolution at the population-averaged level Many open-source integrative data analysis packages and pipelines Commercially available assay kits and sequencing services High throughput in multisample data collection (parallel sequencing of multiple samples is common)
Disadvantages
Targeted loci are preselected, which may cause bias Reaching very fine (<2 kb) resolution is prohibited by FISH probe length and density DNA denaturation procedure may disrupt very fine chromatin structures Analysis depends on customized scripts, lack of integrative data analysis packages Lack of commercially available assay kits and automatic flow and imaging platforms or services Low throughput in multisample data collection (parallel sample imaging platforms are still lacking)	Indirect inference of 3D conformation from contact events between genomic loci Single-cell methods have not reached kilobase resolution. High resolution depends on sequencing depth and requires averaging results from a large number of cells Loss of spatial information (such as cell–cell neighbouring information and cell morphology) Multiway interactions are only maintained in some methods Low throughput when combined with protein detection (one protein at a time) Low throughput in single-cell data collection (usually several hundreds to a few thousands of cells per sample)

FISH, fluorescence in situ hybridization.

Applications

Chromatin tracing is broadly applicable to directly map the folding organization of the genome with true 3D coordinates, to help discover upstream regulatory mechanisms of 3D genome and to elucidate 3D genome functions in health and disease^{22,53,79,85,119–121}. Here, we summarize key applications in the field using chromatin tracing.

Mapping 3D genome organization

Physical view of chromatin architectures at single-copy and single-cell resolution.

Ensemble Hi-C studies have discovered important 3D genome architectures such as the presence of A/B compartments and TADs^13,25. However, whether these structures stably exist along single copies of chromatin in single cells was a long-standing question. It was unclear whether A/B compartments physically exist as aggregates in single copies of chromosomes through simultaneous multiway interactions or whether they are merely statistical trends of transiently interacting A–A and B–B region pairs. To address this question, an initial TAD-to-chromosome-scale chromatin tracing study showed that compartment structures physically exist along single chromosomes in single cells, with A and B compartments organized in a spatially polarized pattern⁴⁶. This finding was confirmed by later single-cell Hi-C studies^39–42,44.

To address whether TADs are stable structural units in single cells, another chromatin tracing study imaged a region of several megabases on Chr21 at the sub-TAD scale (30 kb resolution) and showed that globular TAD-like domains do exist in single cells⁶⁹. However, substantial cell-to-cell variation in domain boundaries was observed, indicating that the population-averaged TAD structures are not stably present in single chromatin copies. The physical presence of the highly variable TAD-like structures was further demonstrated by subsequent chromatin tracing studies at other genomic regions in both cell lines^54,89 and mouse brain tissues⁵⁶, and heterogeneity in other domain features, such as genomic size, physical size and insulation/intermixing degree, has been further observed⁵⁴. Notably, depletion of cohesin, a well-known regulator of population-averaged TADs, does not diminish single-cell domains but eliminates their preferential boundary positions at CTCF sites⁶⁹. This indicates that cohesin is not required for the establishment and maintenance of the single-cell domains. Instead, it is required for preferential positioning of a subset of single-cell domain boundaries at CTCF sites.

Another key application that leverages the single-cell and single-copy precision of chromatin tracing is the study of promoter and enhancer interactions. For long-range promoter–enhancer interactions across several TADs, it was unclear whether the TAD boundaries are nested together through multiway interactions in the same copy of chromatin, bringing the enhancer and promoter together through boundary stacking, or whether multiple TADs are merged as one without boundaries, allowing interaction of the promoter and enhancer. Chromatin tracing of a 750 kb region at 5 kb resolution in mouse embryos revealed interactions between the Pitx1 gene promoter and its distal enhancer, Pen, with stacked intervening TAD boundaries¹²². Fine-scale chromatin tracing of a 210 kb region across an engineered Sox2 gene region in mouse embryonic stem cells demonstrated that CTCF-bound genomic sequences function as an insulation element and partially dampen transcription by forming domain boundaries between the promoter and enhancer of the Sox2 gene¹⁰⁹. Fine-scale regional chromatin tracing combined with large-scale whole-chromosome tracing⁷⁰ or with genome-wide chromatin tracing^57,59 further enables investigation of the interplay between multiple layers of chromatin architecture within a single cell.

Profiling of ultra-long-range chromatin interactions.

Generation of high-resolution contact maps using Hi-C data requires high-sequencing depths and millions of input cells to achieve unbiased sampling of pairwise chromatin contacts at different scales, particularly at large scales or between chromosomes where contacts are rare. By contrast, chromatin tracing enables the measurement of absolute distances between preselected genomic loci in every single cell regardless of their genomic distances and can generate distance map of the targeted loci matching the resolution of Hi-C results using tens to hundreds of cells^46,69. Chromatin tracing is therefore an efficient approach for profiling interactions of targeted genomic regions across multiple scales. As an example, chromatin tracing revealed that within a chromosome, A–A domain contacts are more frequent than B–B domain contacts at long genomic distances (>75 Mb), whereas the converse is true at shorter genomic distances (up to 75 Mb)⁵⁴. In addition, genome-wide chromatin tracing has demonstrated that trans-chromosomal A–A domain contacts are more frequently observed than trans-chromosomal B–B contacts⁵⁴. These results together suggest the existence of different molecular mechanisms underlying short-range and long-range domain interactions.

Chromatin structure regulation

Chromatin tracing offers a new lens to study the regulation of 3D genome organization and to investigate models of 3D genome regulators. Several chromatin tracing studies have investigated Polycomb targeting regions, such as the Hox gene cluster and Sox2 gene region^72,107,123, demonstrating that not all Polycomb chromatin regions are physically compacted or bound by single-cell domain boundaries^72,107. Chromatin tracing performed at histone 3 lysine 9 trimethylation (H3K9me3)-modified AACS1 and Nanog gene regions confirmed the relationship among epigenetic memory establishment, chromatin compaction and stem-cell fate commitment¹²⁴. In the study of Xa and Xi structures in female human cells, Xi was known to adopt a significantly more compact structure along with multiple heterochromatin modifications to achieve stable chromosome-wide transcriptional silencing in somatic cells^125–127. Chromatin tracing confirmed this compaction difference at the whole-chromosome scale⁴⁶. However, single-cell domains were shown to exist on both Xa and Xi at comparable frequencies and with similar sizes, despite the drastically different epigenetic profiles on the two chromosomes, and these domain structures are highly tolerant to perturbations of multiple epigenetic components and transcription^56,89. These results demonstrate that the increased compaction of Xi occurs at a length scale above single-cell domains, and the presence of single-cell domains is independent of several major epigenetic marks⁸⁹.

Perturb-tracing recently discovered dozens of novel regulators of the 3D genome across several length scales. Perturb-tracing identified CHD7 — previously known as a chromatin remodeller involved in opening up local chromatin at the nucleosome level^128–130 — as a chromatin compactor over genomic distance >3 Mb, and that it globally suppresses gene expression⁹⁶. A new computational pipeline was recently designed to call candidate regulators of the 3D genome based on a BART algorithm^59,131,132. These developments further enhance the study of molecular mechanisms underlying multiscale 3D genome architectures.

Development and disease

Cell type differentiation during normal development depends on correct regulatory interactions along the genome. Chromatin tracing identified chromatin structures that emerge and disappear in specific cell types or at various differentiation states during development in multiple organisms. In Caenorhabditis elegans embryogenesis, TAD-to-chromosome-scale chromosome tracing of ChrV in early embryos showed the presence of an unconventional A/B compartment requiring stretching by the nuclear lamina¹⁰⁸. During Drosophila embryonic development, sub-TAD-scale tracing of a 250 kb region across the snail and escargot genes on Chr2L found spatial compaction of chromatin into TADs after the midblastula transition during zygotic genome activation⁷⁷. At a later developmental stage, body segment-dependent TAD structures in post-fertilized Drosophila embryos were observed⁷². In mammalian development, chromatin tracing in mouse fetal liver cells demonstrated cell-type-specific folding schemes of the A/B compartment profile and the lamina and nucleolar association of TADs⁷⁰. At fine resolution, a novel promoter–enhancer interaction in the cis-regulatory region of Scd2 was identified, which is enriched in fetal liver hepatocytes and may explain the hepatocyte-specific expression of the gene. Application of chromatin tracing to human pluripotent stem cells has demonstrated that naive and primed X chromosomes do not exhibit the chromatin compaction states typically observed with somatic X chromosomes⁸⁸.

In disease contexts, chromatin tracing has been applied to study alterations in chromatin structure caused by mutations or deletions in disease-related genes. MECP2-dependent radial organization of active and inactive chromatin in the mouse brain, specifically in the nucleus of neurons, may provide a potential functional link with MECP2 mutations in Rett syndrome⁵⁸. In cancer, fine-scale tracing of the AXIN2 gene locus, an important variable chromatin modules locus, found that a 5 bp insertion or deletion identified in patients with chronic lymphocytic leukaemia facilitates enhancer activation and causes aberrant gene expression¹³³. In addition to enhancer and promoter interactions, chromatin tracing has also been used to study the relationship between large-scale genome organization and disease. The first single-cell 3D genome atlas of cancer was generated through genome-wide chromatin tracing in an autochthonous lung adenocarcinoma mouse model⁵⁹, identifying a structural bottleneck of 3D genome in early cancer progression. Conformations of the 3D genome in early cancer cells are significantly altered and become more homogeneous, which suggests 3D genome importance early in cancer progression⁵⁹. The work further developed new computational pipelines to nominate novel candidate progression driver and suppressor genes and identified that a Polycomb groups protein, RNF2, can partially drive 3D genome changes during cancer progression.

Modelling, simulation and prediction

Extensive theoretical and computational studies have been conducted to understand the fundamental biophysical mechanisms underlying chromatin folding and genome organization¹³⁴, and chromatin tracing data have supported testing of existing models. Whole-chromosome chromatin tracing data have revealed the relationship between spatial distance and genomic distance, which was shown to follow power-law scaling as predicted, but with a scaling factor close to one-fifth at genomic distances larger than 7 Mb, which deviates from the expected one-third scaling factor expected from a previous fractal-globule polymer model⁴⁶. Revised models based on chromatin tracing results are starting to emerge^107,135.

Computational simulation of chromatin folding can aid in the interpretation of chromatin tracing findings. For instance, polarized A/B compartment organization observed through chromatin tracing, along with the distribution of chromatin regions with varying compartment strengths, cannot be replicated using polymer models that only include A–A and/or B–B intrachromosomal interactions⁷⁰. This suggests that extrachromosomal interactions are crucial to achieving the observed organization. In addition, recent studies have used chromatin tracing and polymer modelling to predict the formation of epigenetic memory, which is essential for transcription regulation and cell fate determination^124,136. When leveraging machine-learning and deep-learning models, chromatin tracing data can be used to predict complex biological processes without considering specific molecular or biophysical mechanisms, such as predicting transcriptional activity using a convolutional neural network model¹³⁷ or predicting lung cancer cell states using a support vector machine model, which demonstrates the potential of using chromatin tracing data as a new type of biomarker for disease diagnosis and prognosis⁵⁹.

Reproducibility and data deposition

Since its introduction in 2016, chromatin tracing has been applied in many studies (Supplementary Table 1) and generated thousands of terabytes of raw data. To allow other researchers to explore and validate published data, easily accessible data formatting and sharing is crucial. However, a major challenge for sharing raw chromatin tracing data is the large file size. In contrast to text-based sequencing data sets which are several gigabytes in size, a chromatin tracing data set is usually terabytes in size, which is too large for generic data repositories. Currently, the most common format for sharing chromatin tracing data is the 4D Nucleome Consortia (4DN) FISH Omics Format-Chromatin Tracing (FOF-CT), which reports detected genomic loci identities and their spatial positions, as well as other auxiliary information in a text-based format. The 4DN FOF–CT format is small in data size, compatible with all chromatin tracing methods and good for various downstream analyses. Currently, no common platform exists that allows sharing of raw chromatin tracing data with justifiable cost.

Transparency in data analysis procedures is another key practice to ensure reproducibility. A detailed metadata file describing the sample preparation procedure, imaging parameters, analysis processes and data organization should be shared with the data to help other researchers understand the technical context and structure of each data set. Custom data analysis scripts should also be shared in a public repository and/or a publicly accessible server. Furthermore, it is important to report QC measures, such as detection efficiency and the false-positive rate, to help researchers identify any data sets that do not meet the minimal quality requirement for re-analysis.

Finally, as a relatively new technology still undergoing extensive development and continuous improvement, reproducibility of chromatin tracing may depend on many method details. Unlike ordinary biological assays, reproducing chromatin tracing methods require not only consistency in experimental procedures but also design of correct probes and setting up an appropriate imaging environment. To allow other researchers to design probes specific to the method condition, it is important for original method developers to clearly report probe design parameters, such as probe layout, targeting and readout sequence length, melting temperature and GC percentage range as well as the number of probes per targeted locus. Probe selection criteria following BLASTing against the targeted genome and/or transcriptome must also be reported. As almost all published chromatin tracing studies are performed on custom-built image acquisition platforms combining a microscope with an automatic fluidic system, it is also important to share the key parameters of both, including the imaging modality, laser type and wavelength, filter configurations and objective lens and camera specifications for the microscope and flow speed, volume and time for the fluidic system. This facilitates the same imaging and sequential hybridization conditions to be repeated on other customized imaging platforms.

Limitations and optimizations

Despite the remarkable scientific achievements through chromatin tracing since its development, further improvement of the technology is required to unlock its full potential. Here, we elaborate on the major limitations of current chromatin tracing methods and discuss potential strategies to improve the performance and application of the technology.

Resolution, coverage and throughput

A major limitation for current chromatin tracing approaches is achieving fine resolution, genome-wide coverage and high sample throughput simultaneously. The latest two-layer seqFISH+ technique has achieved genome-wide chromatin tracing at a resolution of 25 kb, yet it requires 96 secondary probe hybridization rounds⁵⁷. Large numbers of hybridization rounds are time-consuming and increase the potential for tissue deformation and primary probe loss⁵⁴. Furthermore, a resolution of 25 kb is not sufficient for resolving many enhancer–promoter interactions, and sequential chromatin tracing at selected genomic regions remains the only strategy for successful chromatin tracing at very fine scales (2–10 kb resolution)^{70,72,109,122,123,133,137}.

One important factor limiting chromatin tracing resolution is the minimum number of probes required for sufficient signal intensity above the background at each target locus. In one study, a minimum of 20 non-overlapping 40 nt targeting sequences were required to target each locus, with each targeting sequence linking to two readout sequences on each primary probe to double the signal intensity during imaging⁷². Considering the gaps between targeting sequences (as not all consecutive sequences meet the criteria for probe design), 2–5 kb is the finest resolution that has been achieved by chromatin tracing by multiplexed FISH. To achieve finer-scale chromatin tracing, technical improvements are needed to reduce the minimum number of targeting probes at each locus while not affecting signal-to-background ratio and signal specificity, likely requiring combined multiplexed FISH methods and signal amplification technologies^138–143. The harsh DNA denaturing procedure before FISH probe hybridization may induce artificial structure changes at very fine scales, requiring the development of potentially less-disruptive DNA labelling technologies such as deactivated CRISPR-based approaches¹⁴⁴. Computational algorithms, such as integrative modelling of genomic regions⁸⁰, are also being developed to predict fine-resolution chromatin structure from low-resolution chromatin tracing results combined with high-resolution Hi-C data^145–147.

Sample throughput decreases as the genomic coverage increases; a longer time is needed for each replicate or imaging area, as more hybridization-imaging rounds are needed to decode more loci. Reducing the total number of hybridization-imaging rounds by using a more efficient, less sparse coding strategy may be useful to improve throughput. However, the maximum number of probed loci that can be simultaneously imaged is restricted by the resolution limit of the microscopy approach used⁵¹. Although super-resolution microscopy can bypass the diffraction limit and has already been applied in some chromatin tracing studies^69,80, the slow image acquisition speed, limited FOV size and high instrument cost compromise its routine usage in most laboratories. In addition, the finest structural resolution achievable with SMLM is limited by the density of FISH probes along the DNA⁵², limiting the accuracy of SMLM at very fine scales⁵³. Expansion microscopy (ExM), enabling nanoimaging with conventional diffraction limited microscopes, may be combined with barcoded chromatin tracing¹⁴⁸, although it is currently unclear whether chromatin structure is preserved when ExM is combined with DNA FISH, and the longer imaging time of ExM owing to the expanded sample volume reduces throughput.

Increasing the speed of signal switching between hybridization rounds is another potential solution for increasing throughput. Recent developments in multiplexing methods based on transient adapters such as FLASH-PAINT¹⁴⁹ and fluidic-free thermal probes such as Thermal-plex¹⁵⁰ have considerably reduced the signal switching time from tens of min to <1 min. Application of these new technologies in chromatin tracing may increase sample throughput significantly in the future.

Finally, increasing data acquisition speed can also help improve throughput. This is largely dependent on the development of microscopy techniques that allow larger FOV imaging with the same optical resolution, faster image capture, transfer and storage speed as well as faster stage movement. Notably, vibrational probes for spectral multiplexing¹⁵¹ can expand the available imaging colours from 5 (common in chromatin tracing fluorescent microscopes) to more than 10, which may further speed up data collection at the imaging step by two to three times.

FISH efficiency

For chromatin tracing based on multiplexed FISH, high-quality data depend on high FISH efficiency, as missing FISH signals lead to drop-out values in the data. Similar to Hi-C results, the drop-out values in chromatin tracing data can be averaged out in the population-averaged analysis, yet they will affect single-cell and single-molecule analyses. Currently, the detection efficiency along chromatin traces is used as the main measurement of FISH efficiency. There are no universally established thresholds of detection efficiency for data inclusion in single-cell and single-molecule analyses. Despite the application of primary probe stabilization, signal amplification and error-robust decoding strategies, detection efficiency in tissue samples is still lower than 50% for both DNA MERFISH and DNA seqFISH+ methods^56,57,59, limiting high-resolution downstream analyses. Further experimental enhancements are necessary to increase FISH efficiency, particularly in tissue samples. Finally, although some studies have started using linear interpolation to fill in missing values in distance matrices based on their non-missing neighbours^88,89, more sophisticated computational algorithms are needed to better impute missing values in chromatin tracing data.

Computational tools

There is a lack of computational tools specifically for chromatin tracing probe design and data analysis. Most probe designs and data analyses in reported chromatin tracing studies were performed with custom-written scripts. The high dependency on customized scripts causes the lack of generalizability between methods and brings challenges to researchers who wish to adopt chromatin tracing but are not experienced with programming. Optimization of the algorithm is still required at many steps during image processing, including barcode generation and assignment, signal decoding, trace linking and homologous chromosome segregation. Substantial future work is needed to improve downstream data analyses, particularly in relation to multiway chromatin interactions and multimodal data integration.

Automatic imaging platform

The lack of accessible suitable equipment is another factor hindering the wide adoption of chromatin tracing. To our knowledge, all currently published chromatin tracing studies have been performed using custom-built imaging platforms combining microscopy with automatic fluidic systems. Although some programmes controlling the system have been made publicly available⁵⁴, instrumentation requires expertise in fluidic engineering, microscopy and programming to assemble all components. By contrast, commercial platforms, such as MERFSCOPE and GenePS, are emerging and have allowed successful performance of spatial multiplexed RNA FISH experiments. Well-validated commercial platforms suitable for chromatin tracing experiments remain unavailable.

Application in clinical samples

To our knowledge, all existing chromatin tracing methods used on tissue sections have been established on fresh-frozen or fresh-fixed tissues, and there have been no reports of successful application of chromatin tracing on formalin-fixed paraffin-embedded (FFPE) tissues, which is commonly used for preserving clinical samples. Given that conventional FISH protocols using FFPE samples have existed for years^152,153, applying chromatin tracing on FFPE samples should, in theory, be achievable. The challenge is likely that clinical FFPE sample preparation procedures may vary markedly between hospitals and physicians, and sample qualities may vary drastically between patients and even between different samples from the same patient. Therefore, chromatin tracing methods developed for clinical FFPE samples must be sufficiently robust to provide consistent results, despite sample variations. Efficient clinical use of chromatin tracing also requires that the aforementioned issues related to throughput, data sharing, data analyses and instrument accessibility are addressed.

Outlook

Since its introduction, chromatin tracing has significantly improved our understanding of chromatin organization and has opened up many new lines of investigations in diverse biomedical areas. Increasingly, studies are using chromatin tracing to study 3D genome organization directly in complex tissue microenvironments, at the single-molecule and single-cell level by default, obtaining 3D genome information that cannot be easily obtained with other methods, such as true multiway chromatin interactions and multimodal relationship between the 3D genome, spatial transcriptome, proteome, metabolome and cell and tissue morphology. We expect chromatin tracing to be broadly applied in studies of development, ageing and disease, as well as in many organisms and tissue types. We expect further development with human patient samples, as the single-cell 3D genome depicted by chromatin tracing has the potential to be developed into a new type of biomarker for disease diagnosis, prognosis and prediction of treatment response.

There remains significant room for technological improvement. Effort is required to increase the coverage, resolution and throughput simultaneously to achieve efficient genome-wide fine-scale chromatin tracing. There is also an urgent need for chromatin tracing methods that can be stably applied to FFPE samples for clinical studies. Moreover, in vivo screening of chromatin organization regulators in the context of different cell types will require integrated approaches able to combine chromatin tracing with imaging-based screening technologies, as well as spatial imaging of the transcriptome and/or proteome. To avoid incomplete inclusion of the nucleus within a thin tissue section, it is essential to develop methods for 3D thick tissue chromatin tracing. This requires not only optimized strategies for tissue permeabilization to allow probes to efficiently reach deep into the specimen and tissue clearing to reduce autofluorescence of thick sample but also the use of advanced microscopy (such as light-sheet microscopy and/or adaptive optics) as well as image-computation methods to allow high-throughput, distortion-free 3D imaging. Finally, as a significant portion of our genome consists of repetitive elements — which are largely omitted in current chromatin tracing studies — future works probing different classes of repetitive elements in combination with the tracing of non-repetitive genome will shed new light on the spatial organization of the repetitive genome in diverse contexts.

To allow broad adoption of chromatin tracing, several developments are needed. First, commercial automated multiplexed FISH imaging platforms must be modified to permit chromatin tracing, ideally with multimodal imaging capacities. Second, user-friendly, cross-method and open-source computational tools are needed to allow researchers without computer programming backgrounds to design chromatin tracing probes, process raw images and perform downstream analyses. Third, community efforts are needed to maintain and update existing data sharing formats (such as FOF–CT) and platforms (such as the 4DN data portal) and establish new ways to manage large chromatin tracing data sets as the technique further improves. Fourth, standard practices related to analysed chromatin tracing data handling and downstream analyses need to be further established. Finally, reducing the time and cost of producing probe oligos and fluorophores is another key factor that can promote the broader adoption of chromatin tracing technologies.

In summary, chromatin tracing complements sequencing-based 3D genomics technologies and is bound to continuously expand the field of 3D genomics. As an interdisciplinary approach, we expect chromatin tracing to promote advances in multiple fields including genome biology, cell biology, biophysics, bioengineering, microscopy, computational biology and medicine.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s43586-024-00354-y.

Glossary

A/B compartments

Two large-scale chromatin compartments originally identified by high-throughput chromosome conformation capture, enriched with transcriptionally active (A) or inactive (B) chromatin, respectively.

Boundary stacking

The phenomenon in which more than two topologically associating domain boundaries are nested together, bringing distant genomic regions into close proximity.

Chromatin contacts

Physical contacts between different regions of chromatin within the nucleus, which can be mapped to understand genome organization.

Chromatin tracing

A series of methods for constructing the 3D folding path of chromatin through imaging and linking the spatial positions of numerous DNA loci based on their genomic coordinates.

DNA denaturation

The process of separating double-stranded DNA into single strands by breaking the hydrogen bonds between complementary bases, thereby opening up the DNA for single-stranded probe hybridization.

Hamming weight

In barcoded chromatin tracing, this refers to the number of ‘1’ bits in a binary barcode.

Melting temperature

The temperature at which 50% of DNA are denatured from double-stranded DNA to single-stranded DNA in an equilibrium state.

Molecular crowding

An imaging issue that happens when the density of signal spots is too high to be resolved by fluorescence microscopy, as the spatial distance between adjacent signal spots is smaller than the resolution limit.

Padlock circularization

After primary hybridization, the 5′ and 3′ ends of primary probes are ligated with the help of a splint oligo, so that the primary probes form circles that are ‘locked’ on the genomic DNA.

Readout sequences

Short DNA sequences on primary probes that bind to fluorescently labelled secondary probes during sequential hybridization rounds in multiplexed fluorescence in situ hybridization.

Sequencing-by-ligation

A next-generation sequencing method that uses DNA ligase to extend a complementary strand to generate sequencing reads.

Sequencing-by-synthesis

A next-generation sequencing method that uses DNA polymerase to extend a complementary strand to generate sequencing reads.

Spot centre fitting

The process of determining the exact centre position of a fluorescent signal spot in microscopy images, typically using Gaussian fitting algorithms for high accuracy.

z-stacking

Creating z-stack images along the z-axis (depth) of a sample with a constant step size between adjacent z-stack images.