Authors: Yanyu Zhu (1Department of Bioengineering, Stanford University, Stanford, CA 94305, USA), Ashwin Balaji (2Department of Chemistry, Stanford University, Stanford, CA 94305, USA; 3Biophysics PhD Program, Stanford University, Stanford, CA 94305, USA), Mengting Han (1Department of Bioengineering, Stanford University, Stanford, CA 94305, USA), Leonid Andronov (2Department of Chemistry, Stanford University, Stanford, CA 94305, USA), Anish R. Roy (2Department of Chemistry, Stanford University, Stanford, CA 94305, USA), Zheng Wei (4Computational Biology Program, Public Health Sciences Division and Translational Data Science IRC, Fred Hutchinson Cancer Center, Seattle, WA, 98109, USA), Crystal Chen (5Department of Chemical Engineering, Stanford University, Stanford, CA 94305, USA), Leanne Miles (1Department of Bioengineering, Stanford University, Stanford, CA 94305, USA), Sa Cai (6Department of Materials Science and Engineering, Stanford University, Stanford, CA 94305, USA), Zhengxi Gu (1Department of Bioengineering, Stanford University, Stanford, CA 94305, USA), Ariana Tse (6Department of Materials Science and Engineering, Stanford University, Stanford, CA 94305, USA), Betty Chentzu Yu (4Computational Biology Program, Public Health Sciences Division and Translational Data Science IRC, Fred Hutchinson Cancer Center, Seattle, WA, 98109, USA), Takeshi Uenaka (7Institute for Stem Cell Biology & Regenerative Medicine and Department of Pathology, Stanford University School of Medicine, Stanford, CA 94305, USA), Xueqiu Lin (4Computational Biology Program, Public Health Sciences Division and Translational Data Science IRC, Fred Hutchinson Cancer Center, Seattle, WA, 98109, USA), Andrew J. Spakowitz (2Department of Chemistry, Stanford University, Stanford, CA 94305, USA; 5Department of Chemical Engineering, Stanford University, Stanford, CA 94305, USA), W.E. Moerner (2Department of Chemistry, Stanford University, Stanford, CA 94305, USA; 8Sarafan ChEM-H, Stanford University, Stanford, CA 94305, USA), Lei S. Qi (1Department of Bioengineering, Stanford University, Stanford, CA 94305, USA; 8Sarafan ChEM-H, Stanford University, Stanford, CA 94305, USA; 9Chan Zuckerberg Biohub, San Francisco, CA 94080, USA; 10Lead Contact)
Categories: Article
Source: Cell
Authors: Yanyu Zhu, Ashwin Balaji, Mengting Han, Leonid Andronov, Anish R. Roy, Zheng Wei, Crystal Chen, Leanne Miles, Sa Cai, Zhengxi Gu, Ariana Tse, Betty Chentzu Yu, Takeshi Uenaka, Xueqiu Lin, Andrew J. Spakowitz, W.E. Moerner, Lei S. Qi
3D genome dynamics are crucial for cellular functions and disease. However, real-time, live-cell DNA visualization remains challenging, as existing methods are often confined to repetitive regions, suffer low resolution, or require complex genome engineering. Here, we present Oligo-LiveFISH, a high-resolution, reagent-based platform for dynamically tracking non-repetitive genomic loci in diverse cell types, including primary cells. Oligo-LiveFISH utilizes fluorescent guide RNA oligo pools generated by computational design, in vitro transcription, and chemical labeling, delivered as ribonucleoproteins. Utilizing machine learning, we characterized the impact of gRNA design and chromatin features on imaging efficiency. Multi-color Oligo-LiveFISH achieved 20-nm spatial resolution and 50-ms temporal resolution in 3D, capturing real-time enhancer and promoter dynamics. Our measurements and dynamic modeling revealed two distinct modes of chromatin communication, and active transcription slows enhancer-promoter dynamics at endogenous genes like FOS. Oligo-LiveFISH offers a versatile platform for studying 3D genome dynamics and their links to cellular processes and disease.
The three-dimensional (3D) organization and dynamic communication of the eukaryotic genome play crucial roles in regulating gene transcription and cellular functions^1^. Aberrations in 3D genome architecture are linked to cancers, neurological disorders, and developmental abnormalities^2-5^. However, 3D genome communication and its relationship with gene expression remain unclear due to the lack of robust approaches for real-time labeling and tracking non-repetitive genomic regions in live cells.
Recent imaging (e.g., multiplexed DNA fluorescence in situ hybridization, FISH) and sequencing (e.g., Hi-C) methods have revealed chromatin organizations including kilobase (kb)-scale loops^6-8^, megabase (Mb)-scale topologically associating domains (TADs)^9-12^, and chromosomal compartments^10, 13, 14^. Beyond these static structures, chromatin communication (i.e., dynamic interaction between DNA elements) are essential for genome functions such as stimulus-mediated transcription, DNA damage repair, and epigenetic remodeling^15-17^. However, these techniques capture cells at fixed time points, limiting our understanding of how real-time chromatin communication influences transcription and cellular processes.
Imaging specific chromatin regions in living cells requires methods that recruit sufficient fluorescent labels to achieve a high signal-to-noise ratio (SNR) while preserving the integrity of chromatin organization. Additionally, high spatial and temporal resolution is essential for capturing chromatin dynamics^17^. For instance, distinguishing different models of enhancer-promoter interaction requires a spatial resolution of 30 nm or better^17^. Early live-cell imaging systems addressed this by integrating large artificial arrays (e.g. LacO) into specific genomic regions, which were bound by LacI-GFP for the recruitment of sufficient fluorophores over the background for visualization^15, 18, 19^. However, this system requires generating stable cell lines via laborious genomic insertion of large arrays at each locus, which is challenging in many cell types. Furthermore, introducing large exogeneous tags (~10kb) may disrupt native chromatin structure and function^16, 18^.
CRISPR-based imaging methods offer programmable platforms for visualizing chromatin DNA in live cells but are limited to repetitive regions^20-24^. We recently developed a reagent-based live-cell FISH (LiveFISH) approach, delivering in vitro assembled CRISPR fluorescent ribonucleoproteins (fRNPs) containing fluorophore-labeled CRISPR-associated RNAs (crRNAs) and nuclease-dead dCas9 protein^25^. While LiveFISH improves SNR and detects chromatin dynamics in diverse cell types, it has not been demonstrated for imaging non-repetitive regions. This limits our ability to study 3D chromatin organization, as many key genomic elements such as enhancers and promoters are non-repetitive. However, efforts to tag non-repetitive loci face challenges^26-32^. Previous approaches require stable cell lines, finetuning expression of multiple constructs, and cannot be applied to primary cells.
We report Oligo pool-based LiveFISH (Oligo-LiveFISH) for flexibly imaging both repetitive and non-repetitive genomic loci and studying chromatin dynamics in live cells (Figure 1A). We exploited approaches for designing effective gRNA pools to image non-repetitive loci by combining in vitro transcribed (IVT) crRNAs with a universal fluorescently labeled trans-acting crRNA (tracrRNA). We characterized key parameters affecting SNR and established gRNA design principles for visualizing diverse genomic loci across cell types. We integrated Oligo-LiveFISH with 3D super-localization microscopy and tracked non-repetitive regions at high spatial (20-nm) and temporal (50-ms) resolution, revealing highly sub-diffusive chromatin motion consistent with fractional Brownian motion. Dynamic modeling revealed two distinct modes of chromatin 1D cis-communication, predominated at short distances (up to 300 kb), and 3D trans-communication, significant over long distances (over one megabase). Oligo-LiveFISH imaging results are consistent with Hi-C data while providing additional insights into dynamic chromatin communication at the single-cell level. Applying Oligo-LiveFISH to study enhancer-promoter dynamics, we found that stimulating FOS transcription with zinc led to reduced 3D distance, increased confinement, and slower dynamics between the FOS enhancer and promoter.
Oligo-LiveFISH is designed to utilize a pool of fluorescent ribonucleoproteins (fRNPs) for imaging. These fRNPs consist of three a nuclease-dead dCas9 protein, a crRNA containing a sequence complementary to the target DNA, and a universal trans-acting crRNA (tracrRNA) that pairs with the crRNA to form the gRNA structure recognized by dCas9^34^. To systematically compare the efficacy of the crRNA:tracrRNA duplex with a single guide RNA (sgRNA), we also designed fRNPs containing dCas9 and fluorescently labeled sgRNAs (Table S1).
To flexibly image non-repetitive regions, we assembled a fRNP pool to tile the region of interest (Figure 1A, Table S1-S2). A single-stranded DNA oligo pool encoding crRNA or sgRNA was chemically synthesized, amplified to generate double-stranded DNA, and in vitro transcribed to produce a crRNA pool or an sgRNA pool. The crRNA pool was annealed with a universal dye-labeled tracrRNA to form crRNA:tracrRNA duplexes, and the sgRNA pool was chemically conjugated with a dye of choice. The purified, dye-labeled crRNA:tracrRNA or sgRNA pools were assembled with purified dCas9 protein as fRNPs and electroporated into live cells.
Compared to LiveFISH that used dye-labeled crRNA annealed with an unlabeled tracrRNA^25^, we employed a universal, fluorescently labeled tracrRNA (5’-Cy3-tracrRNA) to avoid dye labeling the crRNA pool. For comparison, we annealed 5’-Cy3-tracrRNA and 5’-A488-crRNA targeting a 29-kb region on chromosome 3 (Chr3q29, 529 repeats) near the ACAP2 gene (ACAP2-rep) in human U2OS osteosarcoma cells. The A488-crRNA signal colocalized with the Cy3-tracrRNA signal (Figure 1B), confirming that dye labeling preserved the tracrRNA secondary structure for annealing and assembly with dCas9. We compared imaging quality using in-house IVT crRNA and commercially synthesized crRNA and observed comparable SNR under confocal microscope (Figure 1C-D, microscope B in six microscope setups, Table S3).
To confirm specificity, we validated Oligo-LiveFISH with FISH. Since harsh FISH conditions disrupt Oligo-LiveFISH signals, we developed a sequential Oligo-LiveFISH-FISH approach (Figure S1A, Methods). The FISH probe labeled with a different dye targeted a neighboring region of the Oligo-LiveFISH target (Table S1). The colocalization of FISH and Oligo-LiveFISH signals confirmed imaging specificity of Oligo-LiveFISH using dye-labeled tracrRNA (Figure 1E-F) or dye-labeled crRNA (Figure 1G, S1B). Specificity was 96.9% for the ACAP2-rep site using dye-labeled tracrRNA (Figure 1F-G) and 95.9%, 95.0%, and 92.5% for ACAP2-rep, Chr13, and ChrX repetitive sites using dye-labeled crRNA (Figure 1G, S1B). This experiment also confirmed that Oligo-LiveFISH could be multiplexed to image multiple genomic loci simultaneously using a pool of compatible probes (Figure 1G)^21^. Specificity was further confirmed by colocalization of telomere signals labeled by Oligo-LiveFISH and the telomeric-binding protein TRF1-HaloTag (Figure S1C).
We investigated how RNA chemistry affects imaging efficiency. We tested three methods to label gRNA: (1) uridine labeling by incorporating Cy3- or Cy5-UTP with unlabeled UTP during gRNA synthesis; (2) 5’-end labeling using the heterobifunctional, zero-length carbodiimide crosslinker EDC (1-ethyl-3-(3-(dimethylaminopropyl)-carbodiimide), imidazole, and amine-containing Cy3 (Figure S1D-E); or (3) 3’-end labeling by using azido-modified adenosine nucleotides and yeast poly(A) polymerase^35^ (Figure 1H). While method (1) showed inefficient chromatin labeling, likely due to RNA secondary structure disruption by dye-UTP incorporation, both methods (2) and (3) that labeled the gRNA termini exhibited effective chromatin imaging. We primarily chose Method (3) for its simplicity and potential to add multiple dyes per RNA at the 3’ end. As an example, crRNA labeled using Method (3) and annealed with unlabeled tracrRNA or 5’-Cy5-tracrRNA robustly and specifically labeled ACAP2-rep loci (Figure S1F). Moreover, two-color Oligo-LiveFISH using dye-labeled crRNA:tracrRNA and IVT sgRNA labeled at the 3’ end achieved 100% colocalization (Figure 1H, S1G).
We expanded Oligo-LiveFISH to non-repetitive regions using repetitive regions as references for characterizing imaging efficiency and specificity. We designed a crRNA:tracrRNA pool to tile a non-repetitive region upstream of ACAP2 (ACAP2-nonrep, Figure 1I-J). Our pipeline for designing and synthesizing the pool included a computational algorithm that identifies all possible protospacer adjacent motif (PAM, NGG for Cas9) sites in or near the target region and removes potential off-target guides (Figure 1I, Methods). To synthesize crRNAs, we PCR amplified the DNA oligo library followed by IVT to generate a crRNA pool, which was annealed with a universal dye-labeled tracrRNA and mixed with dCas9 to assemble the fRNP pool. Using 24 Cy3-crRNA:tracrRNAs, we observed robust labeling of ACAP2-nonrep in U2OS cells, which colocalized with the downstream ACAP2-rep, indicating high imaging specificity (Figure 1J). While repetitive regions generally exhibited higher SNR than non-repetitive regions, our results demonstrate that Oligo-LiveFISH with dye-labeled crRNA:tracrRNA pools can robustly image non-repetitive regions in live cells (Data S1).
Controls confirmed that dye-crRNA:tracrRNA without dCas9, dCas9 without dye-crRNA:tracrRNA, or dCas9 with non-targeting crRNA produced no observable signals (Figure S2A-B). Moreover, ACAP2-nonrep probes can independently label chromatin without co-delivering ACAP2-rep probes (Figure S2C). Specificity was further validated by colocalization of crRNA:Cy3-tracrRNA with dCas9-EGFP in both non-repetitive and repetitive imaging (Figure S2D-F). Interestingly, dye-RNA achieved higher SNR than dCas9-EGFP (Figure S2E) and simplified multiplexed imaging (Figure 1G). In addition to producing crRNA:tracrRNA from multi-well plates, we tested a more economical approach by synthesizing pooled DNA oligos and amplifying the entire library, which demonstrated efficient labeling (Figure S2G).
To optimize Oligo-LiveFISH, we systematically characterized how imaging efficiency depends (1) the number of distinct crRNA:tracrRNA probes tiling a non-repetitive region; (2) biochemical crRNA features such as spacer length, GC content, and secondary structure; and (3) the accessibility and openness of the target chromatin region (Figure 2A). We developed a method to calculate SNR specifically for punctate signals such as DNA loci (Methods). Utilizing this enhanced SNR metric, we quantified the dependence of imaging efficiency on the crRNA:tracrRNA and various chromatin features.
To determine the optimal number of probes, we designed 192 unique crRNA:tracrRNA probes tiled across the ACAP2-nonrep region and divided them into 8 subgroups (P1-P8) of 24 probes each (Figure 2A). We then evaluated imaging efficiency using combinations of 12, 24, 48, 96 and 192 probes. To test the effect of crRNA length, we also designed two sets of probes for each one with a longer spacer (18-20 nt) and one with a shorter spacer (13-17 nt). These crRNA:tracrRNAs targeting the non-repetitive region were co-delivered with differently labeled sgRNAs targeting a repetitive region, which served as a reference to calculate SNR (Figure S3A). The measured SNRs using 24 probes are shown in Figure 2B and SNRs for other probe numbers are shown in Figure S3B. We also defined a ‘detection ratio’ as the fraction of DNA loci with SNR above 4, and found that as few as 24 probes generated detectable signals (Figure 2B). Interestingly, the mean SNR and detection ratio for each subgroup correlated positively with the mean intensity of H3K27ac and ATAC-seq signals at the targeted sites, indicating that chromatin accessibility influences imaging efficiency (Figure 2C, S3C-F).
We developed an elastic-net regularized generalized linear model to identify determinant features of Oligo-LiveFISH imaging efficiency^36^. We split 19 samples into a training set (17 samples) and an evaluation set (2 samples). We examined 15 features, including gRNA probe number, GC content, spacer length, secondary structures of crRNA and annealed crRNA:tracrRNA, distances to the nearest H3K27ac and ATAC peaks, and various chromatin features such as the intensities of ATAC-seq, H3K27ac, H3K4me1, H3K18ac, H3K56ac, H3K9ac, H3K9me3, and H3K4me3 (Figure 2D).
We performed correlation analysis to remove correlated features and selected 8 features for the machine learning model (Figure S4A). Our analysis revealed that the most critical features for achieving higher SNR (Figure 2E) and detection ratio (Figure 2F) are the number of gRNA species and chromatin accessibility. Predicted SNR and detection ratio correlated well with experimental results (Figure 2G-H). For comparison, we also trained the model using all 15 features without removing correlative features. Although the performance (mean squared error, MSE = 28.691) was less robust than when using the 8 selected features (MSE = 20.739), the results still indicated that gRNA number and chromatin accessibility are the most significant determinants of imaging efficiency (Figure S4B-C).
We calculated the signal-to-background-noise ratio (SBN, Figure S4D, Methods, Data S1), a measurement often misconstrued as SNR^37-39^, and tested different thresholds to define the detection ratio. Results consistently indicated that probe number and chromatin accessibility are the most critical features (Figure S4E-L). Additionally, SNR varied significantly with probes targeting chromatin of different accessibility but showed no significant difference with crRNAs of different lengths for the same probe set (Figure 2B, S3B). This is consistent with the machine learning results, highlighting chromatin accessibility as a more important determinant. Therefore, to better image non-repetitive regions, we designed crRNA probes near the ATAC-seq and H3K27ac peaks. These genome imaging principles may also benefit gRNA pool design in CRISPR-based gene perturbation^40^. To facilitate computational gRNA design, we developed a user-friendly website (Methods).
To test the generality of Oligo-LiveFISH, we visualized non-repetitive genomic loci on different chromosomes. Since the repetitive regions on Chr13 and ChrX were confirmed by FISH (Figure 1G), we designed probes to image non-repetitive regions downstream of these regions, targeting the TUBGCP3 gene (Chr13, Figure 3A-B) and the PLCXD1 gene (ChrX, Figure 3C-E), respectively, following the gRNA design principles summarized above. Our results demonstrated specific imaging of these non-repetitive regions in U2OS and HeLa cells. We tracked both repetitive and non-repetitive TUBGCP3 sites (~285kb apart) and observed that the inter-locus distance varied between 100 and 300 nm (Figure S5A, Data S1).
Moreover, we imaged the non-repetitive MUC4 gene and its upstream repetitive region (~320 kb apart) on Chr3 (Figure S5B). Most cells showed only one ChrX locus in U2OS and HeLa cells using Oligo-LiveFISH (Figure 3D), consistent with our FISH results (Figure 1G, 3F) and the monosomic nature of ChrX in these cell lines revealed by karyotype^41^, indicating loss of one X chromosome. The specificity (Figure 1E-G, 3F), multiplexing ability across chromosomes (Figure 1G, 3A-E), and compatibility with primary cells (vide infra) demonstrate the potential of Oligo-LiveFISH for cytogenetic diagnosis, such as preimplantation genetic testing for aneuploidy (PGT-A) in living cells.
We next imaged the enhancer and promoter of FOS, a gene critical for modulating the immediate early neuronal response and stimulating gene expression in cell proliferation and differentiation^42, 43^ (Figure 3G-H). Recent time-course 5C mapping studies suggested reorganization between its enhancer and promoter^44^. We applied Oligo-LiveFISH to simultaneously track FOS promoter and enhancer dynamics (Figure 3I-J). Similarly, we imaged the promoter and enhancer of the oncogene MYC in U2OS cells (Figure 3K-M). RNA FISH confirmed that Oligo-LiveFISH labeling of the MYC promoter did not affect gene expression (Figure 3N).
We quantified the number of bound fRNPs per target at equilibrium using single-molecule bleaching (Figure 3O, Methods). For ACAP2-rep sites (529 repeats), 194 ± 110 fRNPs were bound per locus (Figure 3P). In contrast, tiling 237 crRNA probes across the non-repetitive FOS promoter showed only 44 ± 44 bound fRNPs per locus (Figure 3P). This corresponds to bound fractions of 36.7% for the repetitive site and 18.5% for the non-repetitive site. This difference arises because each specific probe in the non-repetitive case has only one available binding site. Although non-repetitive loci were dimmer, we show next that their localization precision is sufficient to capture real-time enhancer-promoter dynamics across various genes.
One unique advantage of Oligo-LiveFISH compared to other CRISPR-based imaging methods is that it bypasses the need to construct stable cell lines and can thus be applied to primary cells. We applied Oligo-LiveFISH to label non-repetitive regions in primary human T cells. The round-shape, suspension primary T cells are more challenging to image compared to adherent, flat U2OS and HeLa cells. Using an optimized Oligo-LiveFISH protocol for suspension cells (Methods), we imaged repetitive and non-repetitive ACAP2 regions in primary T cells (Figure 4A-B), achieving 98.8% specificity in imaging ACAP2-nonrep (Figure 4C). Unlike heterogeneous aneuploid U2OS cells (Figure 1E-G, 3E), which showed a wide range of Chr3 copy numbers, diploid T cells mostly exhibited two loci (Figure S5C).
In both U2OS cells (Figure 4D) and activated T cells (Figure 4E-G), we observed a fraction of loci displaying two distinct lobes, which we termed “doublets”, likely representing newly replicated sister chromatids. The 2D distance between the two lobes for both ACAP2-rep and ACAP2-nonrep fluctuated around 500 nm (Figure 4E-G). Approximately 8% of loci in activated T cells and 4% in U2OS cells appeared as doublets (Figure 4H), indicating it is a rare, but non-negligible phenomenon. To confirm active replication, we performed flow cytometry to measure cell cycle and proliferation (Figure S5D-E). Cells stained with Vybrant DyeCycle Ruby exhibited higher total DNA content in activated primary T cells and U2OS cells than in resting T cells, with lower CellTrace dye intensity further confirming active proliferation (Figure 4I). This indicates that the observed doublets are replicated sister chromatids, suggesting the potential of Oligo-LiveFISH for studying dynamic chromatin replication and segregation.
We applied Oligo-LiveFISH to image chromatin regions in induced neurons (iNs) derived from human embryonic stem cells. Imaging at 6 and 24 hours after electroporating fRNP targeting ACAP2-rep regions showed labeled loci along with dendrites and elongating axons in iNs (Figure S5F), indicating that Oligo-LiveFISH does not disrupt neuronal differentiation. Moreover, we simultaneously imaged the FOS promoter and enhancer in iNs (Figure 4J). Cell viability assays in primary T cells and iNs demonstrated that Oligo-LiveFISH is compatible with live-cell tracking over various timescales (Figure S5G-L). Collectively, these results confirm that Oligo-LiveFISH can reliably capture chromatin dynamics in primary T cells and neurons, which were difficult to image using previous approaches.
We next show that the SNR provided by Oligo-LiveFISH is sufficient for detailed quantitative analysis. Extracting such information at a single-cell level provides microscopic insights into biological processes^16, 45-47^. Conventional light microscopy is limited by the diffraction limit (DL), preventing it from resolving objects spaced below ~250 nm. However, for sparse emitters, by fitting the point-spread-function (PSF) of the microscope with a suitable function (e.g., 2D Gaussian), the localization uncertainty can be much lower than the diffraction limit (Figure 5A)^48, 49^. This method, termed super-localization (SL) microscopy, provides subpixel positional information about the emitter^50^. Here we show that SL microscopy and Oligo-LiveFISH can be integrated (termed SL-Oligo-LiveFISH) to track chromatin regions with a localization precision of 10-30 nm in both 2D and 3D. Comparing DL and SL images of ACAP2-rep, the SL image exhibits much lower localization uncertainty (Figure 5B). Localization precision further improves as more photons are recorded for each PSF. Despite higher localization uncertainty for non-repetitive regions compared to repetitive regions (Figure S5M), both achieved sufficient precision (10-30 nm) for analyzing enhancer-promoter looping interactions^51^.
Besides spatial resolution, temporal resolution is crucial for tracking chromatin dynamics as chromosome shapes constantly change during biological processes spanning from milliseconds (e.g., local diffusion) to days (e.g., differentiation) (Figure 5C)^20, 24, 52^. Using 2D SL-Oligo-LiveFISH, we tracked the non-repetitive FOS promoter at time resolutions of 8.2 s (Figure 5D), 2.2 s (Figure S5N), 50 ms (Figure S6A), and as fast as 18 ms per frame (Figure 5E, S6B, Movie S1). Even at 18 ms per frame, localization precision (29.0 ± 3.4 nm) is sufficient for tracking rapid chromatin dynamics. To characterize motion driven by thermal fluctuations, we calculated the mean-square displacement (MSD) as a function of time lag and fitted MSD with the equation based on fBM that accounts for dynamic and static localization errors (Methods)^53^, which suggested sub-diffusive behavior of the FOS promoter (Figure S6C).
A detailed understanding of the real-time dynamics of paired loci across genomic distances in mammalian cells remains elusive. We applied multi-color Oligo-LiveFISH to capture chromatin dynamics to reveal its physical organization, focusing on two sites on Chr3: site 1 at 195.46 Mb and site 2 at 195.49 Mb (Table S1). We quantified the 2D spatial dynamics between these sites (28 kb apart) at 50 ms per frame. Representative trajectories of two Chr3 alleles in the same cell are shown in Figure 5F. Movement in the x and y directions reveals highly correlated motion between the two sites (Figure 5G) with the 2D inter-locus distance fluctuating around ~75 nm (Figure 5H). As a control, tracking beads and nuclear motion at the same temporal resolution precludes the possibility that the correlated motion is due to stage drift or nucleus movement (Figure S6D-F).
To test whether fRNP binding affects intrinsic chromatin dynamics, we performed MSD and velocity autocorrelation (VAC) analyses. Both loci exhibited subdiffusive motion (Figure 5I), with MSD fitting aligned with previous reports^53, 54^. VAC analysis for both loci revealed characteristic negative peaks at τ=δ for all δ, indicating negative correlations in DNA motion due to both polymer elasticity and medium viscoelasticity, confirming subdiffusive behavior of chromatin at this time scale (Figure 5J). Here δ is the measurement time between locus position to estimate velocity and τ is the time lag over which the correlation of velocities is examined (Methods, Data S1). Moreover, rescaling τ by δ results in a universal curve for all values of δ, which quantitatively agrees with fBM analytical theory.
The VAC results and the power-law behavior of MSD indicate that the chromatin move subdiffusively by an fBM model in a viscoelastic nuclear environment^54, 55^. This observed fBM dynamics using Oligo-LiveFISH aligns with those observed using imaging tags in yeast^54^ and bacterial^55-58^ chromosomes, suggesting that fBM is likely a universal mode of chromatin dynamics. Importantly, fBM facilitates recurrent interactions for proximal loci, leading to stronger regulatory responses^59, 60^.
Extracting 2D dynamics from a single z-plane can lose dynamical information because DNA loci inherently move in 3D. To capture fast DNA dynamics in 3D, we employed a double-helix point spread function (DHPSF) microscope (Methods, microscope C, Table S3)^49, 53, 61^. This technique involves PSF engineering to produce two spots, such that the angle of the two spots encodes the depth position of the source over an axial range of 2 μm (Figure 5K, S6G-H), which allows simultaneous 3D tracking with high spatial and temporal resolution.
We applied the two-color 3D DHPSF microscope to track sites 1&2 to confirm the 2D tracking results with the same temporal resolution of 50 ms. The images (Figure 5L) and trajectories (Figure 5M, S6I) of a representative loci pair indicates robust tracking of both sites in 3D for ~1500 frames. The motion in x,y,z directions indicate a correlated behavior of the two loci in 3D (Figure S6J), validated by a control measurement using fixed bead tracking (Figure S6K-L). A closer look indicates a 3D inter-locus distance of ~ 80 nm (Figure 5N).
We compared several biophysical parameters in 2D and 3D. Statistical analysis of the distance distributions indicated a mean separation of 75.1 ± 57.5 nm per localization for the 2D projections (Figure 5O). When accounting for z positions, the true 3D inter-locus distance between sites 1&2 is larger, as expected, at 101.3 ± 44.1 nm (Figure 5O). Similar VAC curves are observed in 3D (Figure S6M-O), confirming the subdiffusive behavior of chromatin with either 2D or 3D methods, while fixed bead motion controls show no correlation (Figure S6N).
In all, we used multiple microscopes (Table S3) to demonstrate that Oligo-LiveFISH is compatible with different microscopes and to confirm biophysical findings with different imaging modalities.
We expanded our approach to simultaneously image three sites (sites 2-4, Figure 6A) using confocal microscope B (Table S3) with A488-, A565- and A647-labeled fRNPs respectively at 24-s intervals (time-lapse mode). Additional dynamics of chromatin regions far apart (> 200kb) can be explored at this longer timescale. Sites 2&3 are 284 kb apart (proximal pair) and sites 3&4 are 2.3 Mb apart (distal pair). A representative image of three locus groups from 3 alleles in one U2OS cell is shown in Figure 6B-C.
On average, loci with smaller genomic distances have smaller spatial distance (Figure S7A)^12^. Surprisingly, we found that a fraction of loci groups demonstrated the opposite behavior, with the distal pair 3&4 in proximity (Figure 6B-C). In the example cell shown, in allele 1, the proximal pairs 2&3 were spatially closer, while in allele 2 the distal pairs 3&4 were spatially closer (Figure 6C-D).
To systematically investigate this phenomenon, we assigned the loci group to a “canonical” group if the mean distance of the proximal pair (2&3) was shorter than the distal pair (3&4) (Figure 6E). Trajectories where the distal pair was closer in 3D space were assigned to a “variant” group (Figure 6E). Among 67 trajectory groups analyzed, 1/3 belonged to the variant group, indicating that the variant behavior is a significant phenomenon of intrinsic chromatin organization (Figure 6E). A Gaussian mixture model (GMM) was further applied and number of clusters, k, was selected as 2 to minimize the Bayesian information criterion (BIC) (Figure S7B)^62, 63^. The resulting bimodal distribution identified by GMM-BIC aligns with the canonical and variant groups we identified, supporting the existence of two distinct chromatin organization states (Figure 6E-F, S7B). Using 600 nm as the cutoff for spatial proximity, 53% of proximal pair 2&3 and 27% of distal pair 3&4 were in proximity (Figure 6F). We counted the number of variant loci groups in cells with three Chr3 alleles and found similar fractions (Figure S7C). Interestingly, every cell has at least one canonical Chr3 locus (Figure S7C). Overall, the spatial proximity of 2&3 (canonical group) and 3&4 (variant group) occurs with approximate probabilities of 2/3 and 1/3, respectively.
Next, we quantified chromatin organization at the single-cell level for each group separately. In the canonical group, larger genomic distances exhibited greater mean spatial inter-locus distances
Finally, we analyzed chromatin communication over time. Chromatin loci do not move independently, as biological processes such as chromatin segregation, recombination, and transcription involve dynamic communication between chromatin loci for orchestrated action^66^. While motion of loci on the same chromosome is primarily coupled through the elastic properties of chromatin, this stress communication is not instantaneous and depends on genomic separation^65^. Using velocity cross-correlation analysis (VCC), we characterized the dynamic coupling between two sites at various rescaled time lags (τ∕δ, Figure 6I, Data S1). By fitting experimental VCC data to a theoretical model that treats chromatin as a Rouse polymer in a viscoelastic environment^65^, we extracted the relaxation time τR, the timescale for stress communication between sites 1&2 (28 kb) through the intervening segment, to be 0.44 s (Figure 6J, Movie S2).
Similarly, the communication time between sites 2&3 in the canonical group was 89 s (Figure 6K, S7L), close to the predicted value of 76 s (Methods, Table S4). Based on the communication time between any two chromatin sites determined via VCC, we calculated the genomic communication distance, which is the minimum length between two chromatin sites to transmit stress. The communication distance between sites 2&3 in the canonical group is 305 kb, nearly matching their actual genomic distance (284 kb). In contrast, sites 2&3 in the variant group exhibited a communication time of 384 s, larger than the predicted value and consistent with their greater spatial distance.
Surprisingly, while polymer theory predicts that the communication between sites 3&4 (2.3 Mb apart) would take about 2.3 hours if stresses were communicated solely through the chromatin, our measured communication time is < 300s in both groups, much shorter than the prediction. This corresponds to a genomic communication distance of ~500 kb, significantly shorter than their actual genomic distance (Table S4). This short genomic communication distance suggests that DNA communication between sites 3&4 occurs more efficiently via an alternative mechanism, likely involving bridging proteins and noncoding RNA complexes that create shortcuts to reduce the effective communication distance.
To better understand our dynamic tracking results, we compared them with bulk Hi-C data for these sites. The close spatial proximity of sites 3&4 in the variant group, along with their fast communication time, are consistent with the architectural stripe near site 3 observed in Hi-C (Figure 6L). Oligo-LiveFISH, however, captures dynamic chromatin interactions at the single-cell level, revealing heterogeneity among cells and identifying distinct chromatin dynamics that bulk Hi-C or multiplexed DNA FISH cannot resolve.
We speculated that the variant group primarily contributes to the architectural stripe in Hi-C. To test this, we calculated pairwise spatial contact probabilities for the four sites from Hi-C and converted these probabilities into genomic communication distances using previously reported simulations^67^ (Figure 6L-M, Table S4, Methods, Data S1). The resulting genomic communication distances closely match the effective genomic lengths derived from Oligo-LiveFISH dynamic tracking.
Taken together, we propose two modes of DNA dynamic 1D cis-communication and 3D trans-communication (Figure 6N). In 1D cis mode, genomic DNA communicate primarily via the elastic behavior of the polymer segment. This mode dominates in the canonical group up to several hundred kilobases and can be accurately predicted by the Rouse polymer model. However, 1D cis-communication alone is too slow for distal pairs to communicate over megabases as our calculations indicate that two sites 2.3 Mb apart would require over 2 hours to communicate via 1D cis-communication and this communication time increases rapidly with genomic distance with a power of ~2. Such delays would be impractical for many biological processes like DNA damage repair. In contrast, 3D trans mode allows sites to circumvent the long chromatin polymer and accelerates communication via bridging complexes, which we speculate could involve protein and RNA complexes. Thus, cells modulate chromatin interaction dynamics by orchestrating both1D cis- and 3D trans-communication via DNA polymer and bridging complexes.
Enhancer-promoter interactions play a central role in gene transcription^7, 36^. Although sequencing-based methods provide static measurements indicating physical proximity between enhancers and promoters^7, 68, 69^, how enhancers dynamically communicate with their target promoters and how this communication relates to gene transcription remains largely unknown (Data S1) ^37, 70-72^. Therefore, we investigated the interplay between FOS gene transcription and the dynamics of its promoter and enhancer using Oligo-LiveFISH.
FOS gene activation by environmental stimuli such as zinc (ZnSO4) has been reported^73, 74^. We performed quantitative real-time PCR (qRT-PCR) and observed that exposing U2OS cells to 400 μM ZnSO4 for 3.5 hours robustly induced FOS transcription, which was verified by RNA FISH on the single-cell level (Figure 7A-E). Notably, we verified that Oligo-LiveFISH probes targeting the FOS promoter and enhancer do not perturb gene transcription (Figure 7E).
Using 3D two-color SL-Oligo-LiveFISH, we captured the dynamics of the FOS promoter and enhancer before and after Zn stimulation with high temporal (50 ms) and spatial resolution. Representative trajectories revealed that, following Zn treatment, both loci exhibit slower mobility and increased confinement, with a concomitant decrease in the inter-locus distance (Figure 7F-K). We calculated the mean enhancer-promoter distance per trajectory and confirmed the distance reduction after Zn addition (Figure 7L). Moreover, dynamical tracking enabled us to calculate distance variability along each trajectory, which showed reduced fluctuation upon Zn stimulation, indicating a more stable enhancer-promoter interaction upon transcription (Figure 7M).
MSD analysis indicated that both FOS promoter and enhancer exhibited slower dynamics following Zn stimulation (Figure 7N-O). Additionally, the MSCD between the promoter and enhancer, which describes how the distance between the loci evolves over time, decreases significantly after Zn stimulation, indicating increased spatial confinement during active transcription (Figure 7P). This confinement implies less freedom for the enhancer and promoter to explore the surrounding context, consistent with a more steady-state communication. Consistently, the VAC results verified the loci undergo more confined movement after Zn stimulation (Figure S7M-N)^54, 55^. These findings demonstrate that, during active transcription, the FOS promoter and enhancer exhibit a shorter separation, increased confinement, and slower dynamics.
Oligo-LiveFISH offers a reagent-based approach that bypasses laborious genetic engineering for high-resolution non-repetitive genomic loci imaging in live cells. Our platform integrates computational gRNA design, pooled in vitro transcription, fluorophore labeling and purification. Since mammalian genomes may contain potential off-target sites when using pooled gRNAs, particularly considering mismatches, we applied a strict computational design pipeline to filter out gRNAs that might generate detectable off-target signals.
Synthesizing and purifying individual fluorescent crRNAs is not cost-effective^29^. Oligo-LiveFISH addresses this issue by employing a common fluorescently labeled tracrRNA, exploring RNA chemistry to label and purify RNA pools with dyes and delivering a pool of pre-assembled fRNPs. A multiplexed in vitro transcribed gRNA library generated from a DNA library significantly reduces the cost and time, allowing paralleled studies of many genomic loci across different cell types.
Oligo-LiveFISH offers an opportunity to advance study the dynamic structure-function relationship in the genome over various timescales^46, 47^. We tracked multiple sites on the same chromosome spanning various genomic distances and found that some distal pairs are unexpectedly close in space and communicate faster than predicted. By combining imaging and dynamic modeling, we propose two modes of chromatin communication. Polymer theory accurately predicts the dynamics between genomic sites up to ~300kb via 1D cis-communication, whereas communication beyond 1Mb is accelerated via 3D trans-communication. Various structural proteins in the genome can form loops, bridging structures, or compartments, which can aid 3D trans-communication. For example, TADs boundaries are usually delimited by CCCTC-binding-factor (CTCF) which stalls cohesin^13, 75^. CTCF forms clusters that couple with cohesin to create complexes^75^, which may facilitate 3D trans-communication^47^. Future studies should investigate the generality and structural origins of these phenomena.
It remains unclear how enhancers dynamically communicate with promoters to regulate transcription due to challenges in imaging these regulatory regions. Using Oligo-LiveFISH to study FOS transcription under Zn stimulation, we observed slower dynamics between enhancer and promoter during active transcription. This slowdown may arise because, during active transcription, the binding of transcription factors and RNA polymerases to DNA create a more crowded environment. In turn, this enhanced confinement may foster more stable enhancer-promoter communication. Looking forward, integrating Oligo-LiveFISH with perturbation approaches like CRISPRi/a and epigenetic editing can infer causal relationships linking enhancer-promoter dynamics to transcription and epigenetic regulation^36, 40, 76, 77^.
The concentration of delivered fRNPs is crucial for Oligo-LiveFISH, and too high a concentration can produce excess free dye and high background noise. We recommend testing various fRNP concentrations to find the optimal level. Additionally, we applied filtering during gRNA probe design and post-imaging analysis to eliminate non-specific localizations. Due to the lack of effective real-time RNA imaging methods, it remains challenging to simultaneously track endogenous transcription along with enhancer and promoter dynamics in live cells. We monitored enhancer and promoter dynamics before and after Zn stimulation for FOS activation. Due to the stochastic nature of transcription, not every Zn-stimulated cell was actively transcribing during the observation window. To address this, we analyzed many cells to assess average dynamics under both conditions. Future advancements in real-time RNA imaging would expand the ability to simultaneously track chromatin dynamics and transcription to study their relationship in single cells.
Requests for reagents, resources, and further information should be directed to and will be fulfilled by the lead contact, Lei S. Qi (slqi@stanford.edu).
All resources and materials reported in this paper will be shared by the lead contact upon request.
Microscopy data reported in this paper will be shared by the lead contact upon request. Custom scripts used for image analysis are available at Github.com (https://github.com/QilabGitHub) and upon request. The plasmids used in the study will be deposited to Addgene (https://www.addgene.org/Stanley_Qi/). We developed a website to aid the design of Oligo-LiveFISH gRNA probes available at https://oligo-livefish.org/.
The U2OS and HeLa cells were cultured in DMEM with GlutaMax (Thermo Fisher Scientific, 10569-044) supplemented with 10% FBS (Sigma-Aldrich, F0926). A human embryonic stem cells (hESCs), H1 (WiCell, #WA01), expressing human NGN2 under a doxycycline-inducible system in the AAVS1 safe harbor locus were used. To generate the line, H1 cells were transfected employing Lipofectamine Stem (Thermo Fisher, #STEM00003) in combination with Alt-R S.p. HiFi Cas9 Nuclease V3 (IDT, #1081060), a customized sgRNA targeting the human AAVS1 safe harbor locus (located between exons 1 and 2 of the PPP1R12C gene) obtained from Synthego (sequence: ACCCCACAGTGGGGCCACTA), the pUCM-AAVS1-TO-hNGN2 plasmid (a gift from Michael Ward; Addgene #105840), and pCE-mp53DD (a gift from Shinya Yamanaka; Addgene #41856). After selection with 2 μg/mL puromycin for one week, mCherry-positive hESCs were isolated via fluorescence-activated cell sorting (FACS) and plated in a serial dilution series. Individual clones with confirmed integration of TetO-Ngn2/CAG-rtTA at the AAVS1 locus (validated by PCR genotyping) were selected. Selection markers (mCherry and puromycin resistance) were removed by transfecting GFP-Cre plasmid, followed by sorting GFP-positive cells and subcloning. Clones that were mCherry-negative and puromycin-sensitive were chosen for subsequent experiments. NGN2-H1 cells were cultured according to standard stem cell culturing protocols^85^. In brief, cells were grown in mTeSR Plus medium with the accompanying supplement (STEMCELL Technologies, #100-0276) and the media was changed every 2 days. Cells were passaged when colonies were mature. When passaging the NGN2-H1 cells, a 6-well plate was coated with iMatrix-511 (amsbio, #AMS.892012) and incubated for 1 h at 37°C in a CO2-regulated tissue culture incubator. Cells were detached using ReLeSR (STEMCELL Technologies, #100-0483) according to the vendor’s protocol and ReLeSR activity was quenched using mTeSR Plus medium. The iMatrix-511 solution was removed from each respective well of a 6-well plate and approximately 80,000 – 100,000 cells were seeded into each well. The passaged NGN2-H1 cells were grown in mTeSR Plus that was supplemented with 10μM of Rock inhibitor, Y-27632 (dihydrochloride) (STEMCELL Technologies, #72302) for 12-24 hours, followed by a media change with fresh mTeSR Plus (without rock inhibitor). All cells were cultured at 37°C and 5% CO2 in a humidified incubator.
Primary human T cells were isolated as previously described^86^. Briefly, buffy coats derived from anonymous healthy blood donors were purchased from the Stanford Blood Center. Primary human T cells were isolated using the EasySep Human T Cell Isolation kit (STEMCELL Technologies, #17951) according to the manufacturer’s protocol with Ficoll-Paque PLUS (GE Healthcare, #17144002) and SepMate-50 tubes (STEMCELL Technologies, #85450). Isolated T cells were immediately cryopreserved at 2-4 × 10^6^ cells per vial in FBS supplemented with 10% DMSO (Sigma). Cryopreserved T cells were thawed and activated on the same-day using Human T-Activator CD3/CD28 Dynabeads (Gibco, #11132D) at a 1 bead-to-cell ratio. T cells were cultured in 0.22 μm sterile-filtered RPMI with L-Glutamine and 25 mM HEPES, supplemented with 10% FBS, 100 U mL^−1^ penicillin, and 100 μg mL^−1^ streptomycin (Gibco). Human recombinant IL-2 (STEMCELL Technologies, 78036) was provided at 100 U mL^−1^. Dynabeads were magnetically removed on day 3 of culture.
To design dense crRNA/sgRNA targeting non-repetitive regions for imaging, we retrieved DNA sequences from human genome hg38 at multiple loci, including regions on Chr3, Chr8, Chr13, Chr14, and ChrX (region coordinates in Table S1). The enhancer regulating MYC and FOS were predicted enhancers by the active-by-contact (ABC) model. Home script was used to design all the possible crRNAs/sgRNAs with different spacer lengths, from 13 bp to 20 bp. First, all these crRNAs/sgRNAs had the NGG PAM sequences and started with G for higher in vitro transcription efficiency. Additionally, the GC content was between 35% and 80%. Finally, these crRNAs/sgRNAs were filtered if they contained TTTT; 10×G; CTCGAG; GCTNAGC; CCANNNNNNTGG. To ensure no imaging signal caused by off-target crRNAs/sgRNAs, we set up a pipeline to remove crRNAs/sgRNAs which possibly target other regions with a short interval distance. Cas-OFFinder^82^ was used to search all potential off-target sites in human genome hg38 with ≤ 1 mismatch in spacer length < 16 bp, with ≤ 2 mismatches in spacer length ≥ 16 bp. We observed that some short crRNAs/sgRNAs had many off-target loci. Thus, we removed these crRNAs/sgRNAs with >100 off-target loci. Next, we defined off-target regions, which contained ≥ 2 off-target loci with interval distance < 2000 bp. All the crRNAs/sgRNAs involved in these off-target regions were filtered. Finally, to control the interval distance between tilling crRNAs/sgRNAs, we removed crRNAs/sgRNAs which were too close to the previous crRNAs/sgRNA (<10 bp).
To examine the occupancy of histone modification (HM) profiles and chromatin accessibility at probe targeting site, the genome-wide profiles of HMs in U2OS and HeLa cell line were generated from Gene Expression Omnibus (GEO) database ^78^. The sample ID in GEO database for each HM and ATAC-seq in U2OS cells are listed in the Key Resources Table. Data downloaded from GEO were processed with Bowtie (60) and MACS2 (61) to generate the sequence depth normalized bigwig file. Then, we retrieved the enrichment signal of each HM at each probe targeting site.
The DNA library was either purchased from IDT or Twist Bioscience. The DNA libraries from Twist Bioscience are in the form of oligo pools. Each DNA oligo contains a unique forward adaptor sequence for PCR, a T7 promoter, a spacer sequence, and a common crRNA/sgRNA backbone (Table S1). The library was amplified by PCR (Q5 Hot Start High-Fidelity 2X Master Mix, NEB, M0494S) using a unique forward primer and a general reverse primer. The DNA libraries from IDT are in the form of ssDNA plates and we mix a subset of them each time to form the sub libraries. Each DNA oligo contains a T7 promoter, a spacer sequence, and a common crRNA/sgRNA backbone. The library was amplified by PCR using the library as the forward primer and a general reverse primer. The PCR product was purified by agarose gel electrophoresis (Monarch DNA Gel Extraction Kit, NEB, T1020S) and the purified PCR products were used as templates for in vitro transcription.
The universal 5’-dye-tracrRNAs and unlabeled tracrRNA were synthesized by IDT. The unlabeled crRNA, and sgRNA were synthesized in vitro by T7 RNA polymerase (HiScribe^™^ T7 Quick High Yield RNA Synthesis Kit, NEB, E2050S). The IVT RNA was treated with RNase-free DNaseI to digest the DNA template for 20 min and purified using RNA Clean & Concentrator^™^-25 (Zymo Research, R1018). The concentration of IVT RNA was measured based on absorption (Nanodrop). crRNA and tracrRNA were annealed in RNA folding buffer (20 mM HEPES, pH 7.5 and 150 mM KCl) with a molar ratio 1-5:1 with the component without dye in access, or with an equal molar ratio if both tracrRNA and crRNA are labeled with dye. The crRNA:tracrRNA complex was incubated at 95°C for 5 min, 70°C for 10 min, gradually cooled down to room temperature and supplemented with 1mM MgCl2, then kept at 40°C for 5 min and gradually cooled down. The sgRNA was incubated at 70°C for 10 min, gradually cooled down to room temperature and supplemented with 1mM MgCl2, then kept at 40°C for 5 min and gradually cooled down. We included an A-U flip (Table S1) in the original Cas9 crRNA:tracrRNA backbone sequence based on previously reported improvement with the sgRNAs for imaging applications^20^.
To label the 3’ end of IVT RNA with fluorescent dye, yeast poly(A) polymerase (PAP, RNT-006-S, Jena Bioscience) and click chemistry were applied^35^. Briefly, 1-10 μM RNA purified from IVT was mixed with 2 mM 2’-Azido-2’-dATP (Jena Bioscience, NU-976S), 1200 U yeast poly(A) polymerase, and 1xPAP reaction buffer (PAP, RNT-006-S, Jena Bioscience) at 37°C overnight in an RNase free environment followed by purification using RNA Clean & Concentrator^™^-25 (Zymo Research, R1018). The click reaction was then performed with the purified 3’-azido-modified RNA, 1 mM DBCO-dye (Click Chemistry Tools, CCT-A130-1 for Cy5; A140-1 for Cy3), and 1U/μL RNase inhibitor (M0314S, NEB) at 37°C for 1 hour, followed by purification using RNA Clean & Concentrator^™^-25 (Zymo Research, R1018).
To label the 5’ end of IVT RNA with fluorescent dye, 120 mM GMP solution, instead of H2O was added to the IVT system. 13.5 μL purified IVT RNA was dissolved in 1.5 μL 10x PBS buffer (1xPBS final concentration). To label the RNA with −Cy3, 0.6 μL 0.8 M Cy3 amine (BP-22558, BroadPharm) was added to 9.4 μL 0.1 M imidazole (pH = 6) to form Cy3 amine/imidazole solution. 1.44 mg (1-ethyl-3-(3-dimethylamino) propyl carbodiimide, hydrochloride) (EDC) was weighed into 1.5 mL tube, followed by adding 15 μL RNA solution (diluted in 1xPBS) and immediately adding 10 μL Cy3 amine/imidazole solution. The solution was vortexed and spun down. 40 μL 0.1 M imidazole (pH = 6) was further added to the system, followed by briefly vortexing and incubating at 37 °C overnight. The system was purified by RNA Clean & Concentrator^™^-25 (Zymo Research, R1018) for three times. To label the RNA with Alex Fluor-647 (AF-647), 2 μL ethylenediamine was added to 98 μL 0.1 M imidazole (pH = 6) to form ethylenediamine/imidazole solution. 1.44 mg (1-ethyl-3-(3-dimethylamino) propyl carbodiimide, hydrochloride) (EDC) was weighed into 1.5 mL tube, followed by adding 15 μL RNA solution (diluted in 1xPBS) and immediately adding 10 μL ethylenediamine/imidazole solution. The solution was vortexed and spun down. 40 μL 0.1 M imidazole (pH = 6) was further added to the system, followed by briefly vortexing and incubating at 37 °C overnight. The system was purified by RNA Clean & Concentrator^™^-25 (Zymo Research, R1018) once and elute in 60 μL H2O. The RNA solution was further added 6 μL 3M CH3COONa (pH= 5.5) and 200 μL EtOH prechilled at −80 °C and stored at −80 °C overnight. The solution was centrifuge in 4°C for 20 min with 13000 rpm to form the RNA precipitate. The RNA precipitate was washed with 70% EtOH (−80°C) three times, air dried and dissolved in H2O. 18 μL NH2-RNA solution was added by 2 μL 1M NaHCO3 (pH = 8.5) and 2 μL Alex Fluor-647 NHS ester (Invitrogen, A37573, dissolved in DMSO), and incubated at 37 °C overnight. The system was purified by RNA Clean & Concentrator^™^-25 (Zymo Research, R1018) for three times.
RNP delivery by electroporation was performed using Neon Transfection System 10 μL kit (Thermo Fisher, MPK1096). For delivery of fluorescently labeled RNP complexes, 0.5-10 pmol of dye-labeled crRNA:tracrRNA or sgRNA were mixed with equal molar amount of dCas9-EGFP (GenScript, T2211048) or x1.2-1.5 molar amount of dCas9 (IDT, 1081067), and then incubated at room temperature for 10 min to allow for fRNP assembly. The assembled fRNP complexes were transfected into 2-4 x 10^5^ suspended cells using the standard protocol for the Neon Transfection System 10μL kit. Electroporation was performed in U2OS cells at 1400V/15ms/4 pulses, in activated T cells at 1400V/10ms/3 pulses, in hESC iNs at 1300V/20ms/3 pulses, and in HeLa cells at 1400V/15ms/4 pulses. For U2OS and HeLa cells, the transfected cells were immediately plated in either 8-well μ-plates (ibidi, 80827-90) or 24-well μ-plates (ibidi, 82426) containing prewarmed culture medium. Hoechst 33342 (Thermo Fisher, H3570) was added to cells at 0.1 μg/mL to stain the nucleus 6-8 hours after transfection or right before imaging. The growth medium was changed to imaging medium [BrightCell MEMO Photostable Media (Sigma, SCM144) containing 10% FBS] for live-cell imaging. For the Zn stimulation of FOS gene, 3.5 h before imaging, 400 μM ZnSO4·7H2O (Sigma, 83265-250ML-F) were added to the medium.
The Oligo-LiveFISH procedure was optimized for suspension cells. For suspension primary human T cell imaging, transfected T cells were recovered in pre-warmed complete medium for 3 hours. Cells were then stained with 0.1 μg/mL Hoechst 33343, washed once, and then transferred to 8-well μ-plates (ibidi, 80827-90) or 24-well μ-plates (ibidi, 82426) containing prewarmed imaging medium. Plates were pre-treated with 1 ug/ml poly-D-lysine (PDL) (Thermo Fisher Scientific, A3890401) for 1 hour at 37°C to accelerate cell attachment. Cells were incubated for another 3 hours to facilitate attachment before imaging.
Plasmid pSLQ14825 (pHR-PGK-TRF1-HaloTag-PYL1) was cloned using standard molecular cloning strategies. TRF1 was cloned from plasmid pHR-U6-sgTel-CMV-puro-P2A-TRF1-mCherry, HaloTag was cloned from plasmid pHR-TRE3G-dCas9-HaloTag, and PYL1 was cloned from plasmid pHR-PGK-PYL1-sfGFP-Coilin^87^.
To further confirm the specificity of Oligo-LiveFISH, we labeled telomere simultaneously using two (1) transfecting a plasmid (pSLQ14825) expressing TRF1-HaloTag driven by PGK promoter; (2) delivering Oligo-LiveFISH probes with crRNA targeting telomere annealed with 5’-dye-tracrRNAs. Telomeres are composed of tandem repeats bound by a six-protein complex known as shelterin and TRF1 is one component of the shelterin. Plasmid pSLQ14825 were transfected to U2OS cells with Lipofectamine 2000 (Thermo Fisher Scientific, 11668030) according to vendor’s protocol. 24 hours later, fRNPs of Oligo-LiveFISH were delivered to the cells by electroporation using Neon Transfection System as above and the transfected cells were immediately plated in 24-well μ-plates (ibidi, 82426) containing prewarmed culture medium. Before imaging, 200 nM Janelia Fluor 503 HaloTag Ligand (Promega, HT1010) was added to cells for 30 minutes and Hoechst 33342 was added to cells at 0.1 μg/mL to stain the nucleus. The growth medium was changed to imaging medium [BrightCell MEMO Photostable Media (Sigma, SCM144) containing 10% FBS] for live-cell imaging.
The FISH probes (Empire Genomics) were designed to target a neighboring region of the (Oligo-)LiveFISH probes (Table S1-S2). Because the harsh condition in FISH disrupts the signals of (Oligo-)LiveFISH probes, we developed a sequential protocol in which the (Oligo-)LiveFISH signals were recorded first in the microscope and FISH signals were taken again by finding the same cells using a gridded coverslip. Specifically, the (Oligo-)LiveFISH experiment is conducted as above. The cells were cultured at 37°C in growth medium on gridded glass coverslips Grid-50 (ibidi, 10187) after the delivery of the (Oligo-)LiveFISH probes by electroporation. The coverslip has imprinted grid with a repeat distance of 50 μm to find the same cell. 4-6 h after electroporation, the growth medium was removed, and the cells were washed once by 1xPBS solution. The chamber was incubated with 4% formaldehyde solution for 10 min at room temperature and then washed twice with 1xPBS solution. Hoechst 33342 was added to cells at 0.1 μg/mL to stain the nucleus. The samples were imaged under microscope with both fluorescent channels (for (Oligo-)LiveFISH signals) and bright field channels (for grid coordinates). After imaging, the medium was changed to 70% EtOH at −20°C overnight. The coverslip was air dried, washed with 1xPBS once, incubated with 0.1 mg/mL RNase A (Thermo Fisher Scientific, EN0531) in 1xPBS solution for 1 hour at 37°C, washed with 1xPBS, 70% EtOH, 100% EtOH sequentially and then air dried. 2 μL FISH probe and 13 μL hybridization buffer (Empire Genomics) were mixed and added to the cellular area on the coverslip, which was further assembled with a 24 mm × 60 mm coverslip (Globe Scientific, 1419-10). For the three-color FISH experiment, 1.5 μL FISH probe of each chromosome and 10.5 μL hybridization buffer were mixed to make the final 15 μL solution. The assembled coverslip was wrapped by aluminum foil and heated for 10 min at 95°C and then incubated in a humidified chamber at 37°C overnight. The sample was washed twice with 50% wash buffer (50% formamide and 0.1% TritonX-100 in 2x SSC buffer), incubated in 50% wash buffer for 10-20 min in dark, washed once with 2x SSC at RT, and mounted onto slides with VECTASHIELD PLUS Antifade Mounting Medium (Vector Laboratories, H-1900-10) for imaging. The colors of FISH probes and (Oligo-)LiveFISH probes are different to exclude the possibility that the FISH signal is the remaining signal from (Oligo-)LiveFISH probes.
NGN2-H1 hESCs were differentiated into neurons based on established protocols^85, 88, 89^ with slight adaptations. In brief, Accutase (STEMCELL Technologies, #07922) was used to detach NGN2-H1 cells. The singularized cells were seeded onto Matrigel-coated 6-well plates at a density of approximately 400,000 cells per well. “Pre-differentiation” medium was prepared to achieve the final working KnockOut DMEM/F12 medium (Gibco, #12660012), 1x MEM NEAA (Gibco, #11140050), 1x N-2 supplement (Gibco, #17502048), 1x GlutaMAX (Gibco, #35050061), 1μg/mL doxycycline, 1μg/mL Laminin (Sigma-Aldrich, #L02020). The pre-differentiation medium was supplemented with Rock inhibitor on Day 1. The cells underwent medium change with pre-differentiation medium without Rock inhibitor on Days 2 and 3. A 24-well glass bottom imaging plate was coated with 0.1 mg/mL of poly-D-Lysine (Thermo Fisher Scientific, A3890401) for 24 h at 37°C, rinsed thoroughly with sterile double distilled water, coated with 200 μL of 1 μg/mL of laminin per well, and incubated at 4°C overnight. On day 4, the cells were pushed into neuron maturation. First, cells were detached with Accutase according to the vendor’s protocol and diluted in complete “maturation” Neurobasal-A (without phenol-red, Gibco, #12349015), 1x B-27 supplement (Gibco, #17504044), 1x MEM NEAA, 1x GlutaMAX, 1 μg/mL doxycycline, 1 μg/mL laminin, 10 ng/mL BDNF (PeproTech, #450-02-10UG), 10 ng/mL NT3 (PeproTech, #450-03-10UG) in a 4 ratio of Accutase to medium. The cells were pelleted at 300 xg for 2 minutes, rinsed once with DPBS, counted, centrifuged at 300 xg for 2 minutes, and resuspended in electroporation buffer R from the Neon Transfection System 10 μL kit (Thermo Fisher, MPK1096). Meanwhile, 500 μL of maturation medium was added to each well of a 24-well plate. The cells were electroporated at 1.5-2x 10^5^ cells per electroporation condition, washed once and seeded directly into their respective wells within the PDL/laminin-coated 24-welled plate. Hoechst 33342 was added to cells at 0.1 μg/mL to stain the nucleus 6 hours after transfection or right before imaging.
Primary human T cells were thawed and split into two groups, with one group activated with Human T-Activator CD3/CD28 Dynabeads at a 1 bead-to-cell ratio. On day 3 of culture, 5 x 10^5^ T cells and U2OS cells were stained with Vybrant DyeCycle Ruby Stain (Invitrogen, V10309) according to the manufacturer’s protocol in 1 mL RPMI with L-Glutamine and 25 mM HEPES, supplemented with 10% FBS, 100 U mL^−1^ penicillin, and 100 μg mL^−1^ Streptomycin (Gibco). After staining, cells were immediately analyzed via flow cytometry, which was conducted on a CytoFLEX S (Beckman Coulter). For all samples, a minimum of about 1 x 10^4^ events was collected within the final gated population-of-interest. Data was analyzed using FlowJo v10 (BD Biosciences). In the cell cycle assay, cells stained with Vybrant DyeCycle Ruby Stain exhibited higher total DNA content in activated primary T cells and U2OS cells than in resting T cells, indicating these actively replicating cells possessed more total DNA content than the non-replicating resting T cells.
U2OS cells and primary human T cells thawed same-day were stained with CellTrace Far Red dye (Invitrogen, C34572) according to the manufacturer’s protocol. Immediately after staining, freshly thawed primary human T cells were split into two groups, with one group activated with Human T-Activator CD3/CD28 Dynabeads (Gibco) at a 1 bead-to-cell ratio. Cells were incubated for 5 days at 37 °C and 5% CO2. Dynabeads were magnetically removed before analysis. Cells were washed twice with FACS buffer (DPBS supplemented with 2% FBS and 1 mM EDTA (Gibco)) prior to flow cytometry, which was conducted as described above^86^. In the cell proliferation assay, lower CellTrace dye intensity in activated primary human T cells and U2OS cells indicated active proliferation compared to resting T cells.
smiFISH (single molecule inexpensive FISH) probes were designed for RNA FISH staining of human FOS and MYC mRNA. Primary smiFISH probes were designed via Oligostan^90^ and all the sequences of smiFISH probes used in this study are described in Table S2. The primary oligos for making smiFISH probes were ordered from IDT. Equimolar mixture of 48 probes for each gene were made in Tris-EDTA pH 8.0 (TE) buffer at a final concentration of 20 μM. FLAP-Cy5 with two Cy5 dyes labeled on the 5’ and 3’-end of the oligonucleotide (/5Cy5/AA TGC ATG TCG ACG AGG TCC GAG TGT AA/3Cy5Sp/) were applied as secondary probes^90^. 10 μL primary probes mixtures were added with 2.5 μL 100 uM secondary FLAP-Cy5 probes, 5 μL 10x NEB buffer (1M NaCl, 500 mM Tris-HIC, 100 mM MgCl2, pH 7.5) and RNase free water were added to make a final volume of 50 μL. The mixture undergoes a hybridization process at 85 °C for 3min, 65 °C for 3min, and 25 °C for 5 min, then is placed on ice.
For RNA FISH staining, cells were washed with 1xPBS once, fixed with fixation buffer A (3.7% formaldehyde in 1xPBS) at room temperature for 10 min, quenched with 50 mM glycine at room temperature for 5 min, washed with 1xPBS twice, and permeabilized in 70% ethanol at 4°C for 1 hour to overnight. Permeabilized cells were incubated with wash buffer A (10% formamide in 2xSSC buffer) at room temperature for 5 min and incubated with hybridization system [10% formamide and 0.12 μM FISH probes in Stellaris RNA FISH Hybridization Buffer (LGC Biosearch Technologies, SMF-HB1-10)] at 37°C for 4 to 16 hours. Cells were then washed twice with wash buffer B (15% formamide in 2xSSC buffer) at 37°C for 30 min, stained with 2 ug/mL Wheat Germ Agglutinin (WGA)-CF488A (Biotium, 29022-1) and 0.1 μg/mL Hoechst 33342 for 10 min, washed with PBST buffer (0.1% Tween-20 in 1xPBS) once, washed with 2xSSC buffer once, and mounted onto slides with VECTASHIELD PLUS Antifade Mounting Medium (Vector Laboratories, H-1900-10) for imaging.
Oligo-LiveFISH was performed on T cells and iNs as previously described. Cells without Oligo-LiveFISH were incubated on Poly-D-lysine-treated chambers or chambers without treatment (for T cells) at standard growth conditions. After 6 hours of incubation, T cells and iNs were stained with propidium iodide at a 1000 dilution for 10 min. T cells were washed with FACS buffer and analyzed by flow cytometry. For iNs, medium was either fully exchanged or not before flow cytometry and both conditions were tested. To assess cell viability with imaging, cells were stained for 10 min with 2 μL Annexin V Conjugates for Apoptosis Detection (Alexa Fluor 647, Thermo Fisher, A23204) or propidium iodide at a 1000 dilution in 24-well imaging chambers containing 500 μL growth medium before imaging in Nikon confocal microscope (Microscope B) or Incucyte S3 (Sartorius) at different timescales. For each frame in Microscope B, the cells were exposed to 365 nm (exposure time 20ms), 549 nm (exposure time 300 ms), and 640 nm laser (exposure time 30ms) at 186 mW/cm^2^, 877 mW/cm^2^, 1073 mW/cm^2^. For Incucyte S3 (Sartorius) imaging, cells were imaged every 20 min for phase contrast, green fluorescence (excitation 441-481 nm, emission 503-544 nm, acquisition time 100 ms) and red fluorescence (excitation 567-607 nm, emission 622-704 nm, acquisition time 300 ms) channels. The phase contrast channel was used to determine the total number of cells with a segmentation adjustment value of 0. The red channel was employed to identify dead cells using adaptive segmentation (Incucyte Base Analysis Software). A minimum cell area threshold of 20 μm^2^ was applied to filter out small debris.
ZnSO4 of gradient concentration was applied to U2OS cells for 3.5 hours. RNA from U2OS cells was isolated with a RNeasy kit (Qiagen,74136), followed by cDNA synthesis with iScript cDNA Synthesis Kit (Bio-Rad,1708891). Gene expression of FOS and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was assessed by real-time PCR with iTaq Universal SYBR Green Supermix (Bio-Rad,1725124), according to manufacturer’s specification. Dissociation curves were used to confirm specificity of the PCR. Data were normalized using GAPDH as a reference gene. Relative expression levels compared to normal growth cells were calculated by the change-in-cycling-threshold (ΔΔCt) method as ΔΔCt = ΔCt (Zn+) − ΔCt (Zn−), where ΔCt = Ct (FOS mRNA) − Ct (GAPDH mRNA). The primers used in RT-PCR are in Table S2^74^. Experimental values are expressed as mean ± S.E.M. of three biological replicates, and each biological replicate was measured in triplicate (technical replicates).
Six different microscopes were used for imaging in this study (Table S3) to show the flexibility of the Oligo-LiveFISH labeling approach.
Widefield microscopy was performed with a Leica dMi8 microscope equipped with a Leica-DFC9000GT-VSC10166 camera, a SOLA Light Engine, and a live cell imaging environment control system. Images were taken using LAS X system with Z stacks at 0.3 μm step size. The magnified camera pixel size is 0.103 μm/pixel with the HC PL APO 63x oil objective (NA = 1.40). For live cell imaging, the cells were kept at humidified 37 °C and 5% CO2.
Single-color live-cell 2D tracking of loci and some of the loci bleaching data were acquired using a custom-built single-molecule inverted widefield microscope. A 100 mW 561 nm continuous-wave (CW) laser was used to excite the sample. The laser beam was magnified twice by a pair of lenses before entering the backport of the microscope body (Olympus IX71). A Köhler lens was placed near the backport to ensure widefield illumination after the excitation light was collimated by the objective (Olympus, UPlanSAPo 100x/1.4 oil). Fluorescence was separated from the excitation light using a multi-band dichroic mirror (Semrock, Di01-R405/488/561/635-25x36) and then focused by a tube lens (f=180 mm) to an intermediate image plane. The fluorescence emission was further relayed by a pair of lenses (f=90 mm for both) in a 4f optical configuration, as previously reported^91^. This emission pathway enables 3D imaging experiments using dielectric phase masks, but for these experiments, no phase masks were inserted into the emission pathway, enabling standard 2D imaging. The emission light was further filtered with a combination of a 561-notch filter (Chroma, ZET561NF) and a 605/70 band-pass filter (Chroma, ET605/70M) placed in the 4f emission pathway. The fluorescence light was finally focused by the last 4f lens onto an EMCCD camera (Andor, iXon897) with a pixel size of 160 nm in sample coordinates. To image the nucleus, the samples were excited with a 405 nm laser (Coherent Cube), and the fluorescence was collected and filtered using the same dichroic mirror mentioned earlier, along with a 440 long-pass filter (Chroma, HQ440LP). For live-cell 2D tracking, samples were imaged at 37°C using a stage-top incubator (INUB-PI-F1, Tokai Hit) and illuminated at approximately 3-10 W/cm^2^. The camera exposure time was set either 18 ms (Figure 5E) or 50 ms (Figure S6A).
Loci bleaching data were acquired using a custom-built single-molecule widefield microscope^92^, consisting of an inverted Nikon Diaphot 200 frame with an oil-immersion objective 60x/1.35 NA (Olympus UPLSAPO60XO) and an EMCCD camera (Andor iXon Ultra 897). 642 nm and 560 nm 1W CW lasers (MPB Communications Inc.) were used for excitation of Cy5 and Cy3, accordingly, and a 405 nm 50 mW CW diode laser (Coherent OBIS) was used for Hoechst 33342. All laser beams were expanded, co-aligned in free space and coupled into a square-core multi-mode fiber with a shaker for speckle reduction (Newport F-DS-ASQR200-FC/PC). The output tip of the fiber (200 × 200 μm^2^ core size) was imaged with a 10x/0.25 NA objective and magnified to achieve a square sample illumination region of 47.6 × 47.6 μm^2^ with a constant intensity in the image plane of the main objective. The fluorescence was split from the excitation light with a multi-band dichroic mirror (ZT405/488/561/640rpcv2, Chroma). The fluorescence light was then filtered with a combination of the ZET635NF and ET685/70M filters (Chroma) for Cy5, with a combination of the FF01-609/57-25 and 561LP EdgeBasic filters (Semrock) for Cy3, or with the FBH450-40 filter (Thorlabs) for Hoechst 33342. The image of the sample was focused with a tube lens (f = 400 mm) on the EMCCD camera, providing a pixel size of 117 × 117 nm^2^ in sample coordinates. The sample was mounted on two stacked piezo stages (PI U-780.DOS for coarse and P-545.3C8S for fine movement). To correct axial (Z direction) drifts, an infrared beam (Thorlabs S1FC808) was directed toward the sample through the main objective. This beam was aligned so that its reflection from the coverslip-imaging buffer interface moves laterally when the sample moves axially. During imaging, the Z position of the fine stage was directed to move proportionally to the shift of the reflected beam, compensating for Z drifts.
Live-cell two-color 2D tracking of loci labeled with AZ488 and Cy3 was performed with the Microscope C with the following modifications. The 647 nm notch filter was replaced with a 488 nm notch filter (Semrock, NF03-488E-25). The emission pathway dichroic was replaced with a 560 nm long-pass (Semrock, FF560-FDi01). Bandpass filters were selected appropriately for each dye (Semrock, FF03-525/50-25; Chroma, ET605/70M). The double-helix phase masks were removed to produce the open aperture point-spread function. Each cell was first imaged with 6 W/cm^2^ intensity of 405 nm with 50 exposure time and 272 EM gain to locate the nucleus. Loci were then imaged simultaneously with 27 W/cm^2^ intensity for 488 nm and 29 W/cm^2^ intensity for 561 nm with 50 ms exposure time and 272 EM gain until the loci photobleached. After imaging cells, 100 nm Tetraspeck beads were imaged to fully sample the microscope field-of-view to compute the transformation function to register localizations in each channel. Two-color imaging with microscope A4 allows simultaneous observation of the two chromatin sites even when they are only ~70 nm apart.
Confocal microscopy was performed at the Stanford University Cell Sciences Imaging Facility. Most images were taken with a Nikon TiE inverted spinning disk confocal microscope (SDCM) equipped with a Photometrics Prime 95B sCMOS camera and a CSU-X1 confocal scanner unit with microlenses. The magnified camera pixel size of SDCM is 0.183 μm/pixel with 60x PLAN APO IR water objective (NA = 1.27). The Hoechst, gRNA-A488 (or dCas9-GFP and Fluorescein), gRNA-Cy3 (or gRNA-A565) and gRNA-Cy5 (or gRNA-Atto647 and gRNA-AF647) fluorescence were excited using 405 nm, 488 nm, 561 nm, and 642 nm lasers lines and collected with a 445/20, 525/30, 605/15 and 676/29 emission filters (Semrock) respectively. Figure 3N, 4B, 4J, 7D,7E, S1C, and S5G-I were taken using the Nikon Ti2 Crest SDC microscope equipped with a Photometrics Kinetix camera, a Perfect Focus (PFS) focus lock system, and 60x Plan Apo oil objective (NA = 1.40), and 365 nm, 488 nm, 561nm, and 640 nm lasers. For both SDC microscopes, the pinhole size is 50 μm. Z-series images were taken using a piezo Z-axis stage (Mad City Labs) and the NIS Elements software with Z stacks at 0.3 μm step size except the three-color tracking in Figure 6, which used 0.2 μm step size for better axial position. When imaging 4 color channels (Figure 6), this microscope captures 5 z slices for 30 frames, with each frame completed every 24 s due to filter changes. For other situations, the time between position measurements is changed appropriately.
Live-cell 3D tracking of loci labeled with Cy3 and AlexaFluor647 was performed with a home-built 3D single-molecule imaging microscope. CW 647 nm (MPB Communications), CW 561 nm (Coherent Sapphire), and CW 405 nm (Coherent OBIS) lasers were coaligned and focused onto the fiber input of 200-μm diameter circular-core multimode fiber de-speckler (Newport, F-DS-AFS200-FC/PC). The fiber output was imaged with a 10x objective (Newport, LIO-10) and relayed to the sample plane through the back port of the microscope body (Olympus, IX71) via 3 achromatic lenses, a quad-pass dichroic (Chroma, ZT405/488/561/640rpcv2), and the objective lens (Olympus, UPlanSAPo 100x/1.4 oil), producing a 35-μm diameter flat, circular illumination profile. Fluorescence was collected with the objective and focused with the tube lens before passing through a 561-notch filter (Chroma, ZET561NF), a 647-notch filter (Chroma, ZET647NF) and entering a two-channel 4f system. The first 4f lens (f=90mm; Qioptiq, G322389000) was positioned one focal length away from the intermediate image plane after the tube lens. After this lens, a 640 nm long-pass dichroic (Semrock, FF640-FDi02-t3-25x36) was used to split emission into two color channels. A double-helix phase mask (2012-2.4μm-DH2.7-615-SQ, fabricated in-house^93^; 2018-2.4μm-DH2.7-686-DHO, Double-Helix Optics) was placed one focal length after the first lens in each arm of the emission pathway, followed by bandpass filters (Chroma, ET605/70M; Chroma, ET700/75M) for each channel. A second 4f lens (f=120mm; Edmund Optics, 32921) was placed 120 mm after the phase masks in each channel. Finally, emission from each arm of the emission pathway was reflected off the two faces of a knife-edge mirror to form an image of each channel onto separate portions of an EMCCD camera (Andor, iXon 897 Ultra) with a pixel size of 124 nm in the 605/70 channel and 117 nm in the 700/75 channel. Each cell was first imaged with 6 W/cm^2^ intensity of 405 nm with 50 ms exposure time and 272 EM gain to locate the nucleus. Loci were then imaged simultaneously at 29 W/cm^2^ for 561 nm and 37 W/cm^2^ intensity for 647 nm with 50 ms exposure time and 272 EM gain until the loci photobleached. After imaging cells, 100 nm Tetraspeck beads (Invitrogen, T7279) were imaged to fully sample the microscope field-of-view to compute the transformation function to register localizations in each channel (see Calibration Bead Sample Preparation for details). A calibration z-stack of 100 nm Tetraspeck beads was also acquired simultaneously in each color channel (3.2 μm range, 50 nm steps).
To assess the impact of microscope stability and sample drift on observed loci motion, 100 nm Tetraspeck beads were imaged with 561 nm and 647 nm excitation in two color channels with Microscopes A4 (2D) and C (3D), allowing us to see the same bead in both color channels at the same time. Excitation power and white light background were selected to produce similar signal and background levels as the chromosomal loci. Beads were imaged with the same camera parameters as the chromosomal loci. Beads were then localized and registered onto each other.
To assess the drift of typical nuclei of live U2OS cells, cells were stained with 0.1 μg/mL Hoechst 33342 and then mounted on Microscope A4. Nuclei were imaged with ~6W/cm^2^ 405 nm. Nucleus images were then thresholded and converted to binary format to isolate the nucleus. The center of mass of the binary nucleus images was then calculated and plotted against time as a measure of live-cell nuclear drift.
CellVis sample chambers (D29-20-1.5H) were incubated with 1M potassium hydroxide for 10 minutes, washed with water, incubated with 0.44% polyethyleneimine for 10 minutes, washed with water, and dried with nitrogen. 100 nm Tetraspeck beads (Invitrogen, T7279) were diluted in hot agarose (1% w/v) and placed onto the prepared sample chamber. The agarose-bead mixture was then allowed to cool and solidify before the chamber was mounted on the coverslip for imaging.
Image processing was performed in Fiji (ImageJ)^80^, PYME and MATLAB. A single microscope plane showing fluorescence of labeled genomic loci or the projection of adjacent Z planes showing maximum loci fluorescence is shown in figures. The contrast levels were kept the same for each channel and merged. Line scans were performed using the “Analyze/Plot Profile” function in Fiji. To compare the fluorescence intensity in different channels with different backgrounds, the relative fluorescence intensity was calculated by setting the background fluorescent intensity as 1. For Figure 1G, 3B, 3I, 3L, 3M, 4D, 4J, S1D, S5F, S5N, the Oligo-LiveFISH fluorescent signal only within the nucleus is shown. To quantify the specificity, the Oligo-LiveFISH loci and FISH dots were manually counted, and the number of co-localized loci were divided by the total number of loci detected by Oligo-LiveFISH to obtain the specificity. The doublet loci, which occur in less than 5% of all loci, were excluded for localization and tracking in analysis in Figure 6. Figure 5A was generated by MATLAB simulation. To quantify the mRNA intensity for each single cell, sum projection of all 15 Z planes was generated and the background intensity of the empty plate was subtracted. The final total mRNA intensity per cell was obtained by integrating the intensity within cell outline using the “Analyze/Measure” function in Fiji. All the Box-and-whisker plots (Figure 2B, 3N, 7E, S3B, S7F) were plotted with the bar indicating the median and the whiskers extend to 10% and 90%. The boxes show the 25% to 75% interquartile range with individual data points shown for the lowest and highest 10% of each dataset.
To characterize the signal-to-noise ratio (SNR) of Oligo-LiveFISH labeling, we performed a two-color Oligo-LiveFISH experiment targeting a repetitive genomic region in one color (G) and its nearby non-repetitive genomic region in a second color (R) in confocal microscope. We assumed that the repetitive loci were labeled with 100% specificity (from FISH results) and used it as the marker for the corresponding non-repetitive loci when they were imaged simultaneously. Each channel was imaged as a z-stack and a single z plane showing the highest fluorescence of the labeled loci in each channel was used to calculate the signal in that channel by a custom MATLAB code. In practice, we first selected the position of the locus (x0, y0) approximately based on the image of the repetitive color channel (G). A 15-pixel × 15-pixel region of interest (ROI) centered on (x0, y0) was selected for both repetitive (G) and non-repetitive (R) loci. For each channel, the central 7-pixel × 7-pixel region around the peak of the loci (x1, y1) was removed and the mean signal count of the remaining 176 pixels was used as the background value. This background value was subtracted from the signal counts of all pixels of the initial 15-pixel × 15-pixel ROI. The integrated signal I was calculated as a sum of these background-subtracted pixels and the signal amplitude A was calculated as (Equation 1)A=I2πσ2 using the integral of a 2D Gaussian I=∫−∞∞∫−∞∞Ae−((x−x0)22σx2+(y−y0)22σy2)dxdy=2πAσxσy=2πAσ2 where σ is the standard deviation width of the image of the locus assuming a symmetrical Gaussian σ=σx=σy. The signal-to-noise ratio (SNR) was then calculated using the following canonical (Equation 2)SNR=Amplitudeofsignal(Amplitudeofsignal)+(STDofbackground)2=AA+s2=I2πσ2I2πσ2+s2
The signal-to-background-noise ratio (SBN) was calculated using the following (Equation 3)SBN=AmplitudeofsignalSTDofbackground=As=12πσ2s
We used σ=1.5 pixels for all non-repetitive loci and σ=1.4 pixels for all repetitive loci based on a manual measurement of the image profiles of several loci. Note that σ indicates the Gaussian width of the image of a locus, while s signifies the pixel-based standard deviation of the signal counts of the 176 background pixels from the periphery of the ROI.
The centers of the DNA loci were obtained by fitting the raw images in the ThunderStorm plugin^94^ in Fiji. To obtain the position of the loci from confocal First, the z projection images were filtered with a wavelet filter with a b-spline order of 3 and a scale of 2. The coarse localizations were found as local maxima with an 8-neighborhood connectivity and a threshold of 2·std (Wave.F1). These localizations were weighted least squares-fitted with the integrated Gaussian model using a radius of 5 pixels and an initial sigma of 1.0 to get the (x,y) position of the chromosome loci. A 7-pixel × 7-pixel region around the fitting position was selected and its intensity with different z slices were examined. The z slice with maximum intensity was selected as the z position of that locus. For two-color 2D tracking using widefield microscope, localizations were fitted to an integrated Gaussian model with maximum likelihood estimation using a radius of 8 and an initial sigma of 1.6. Tetraspeck bead data for channel registration were similarly localized with ThunderSTORM (Wavelet Filter: b-spline order 3, scale of 2; Localization Detection: local maxima, 8-neighborhood connectivity, threshold of 3·std (Wave.F1); Fitting: integrated Gaussian model, maximum likelihood estimation, radius 8, initial sigma 1.6.), and a 2D locally weighted mean quadratic transformation is computed mapping the reflected channel onto the transmitted channel. Locus localizations from the nucleus are then cropped.
Raw images of two-color 3D DHPSF tracking experiments were analyzed with the software Easy-DHPSF v.2^7^.The software uses a model function of 2 Gaussians separated by a distance, with the axial position given by the angle formed between the line connecting both Gaussians and the horizontal. Z-stacks of the same Tetraspeck bead in each channel produced only by axial sample translation are fitted with the model function to produce a calibration curve mapping angle to z position for each color channel. Templates of the DHPSF at 8 evenly spaced angles throughout the PSFs full angular range are then chosen from the raw loci tracking data. These templates are then used to identify DHPSF signal in the raw data via phase correlation in the Fourier domain. Candidate DHPSFs are then fit with the model function using non-linear least squares fitting. Tetraspeck bead data for channel registration are similarly localized, and a 3D locally weighted mean quadratic transformation is computed mapping the reflected channel onto the transmitted channel. Chromosomal locus localizations are then filtered for lobe separation (4.5 pixels < lobe separation < 10 pixels) and then registered. Z localizations are scaled by 0.7 to account for depth-dependent focal shift. For the 3D FOS gene enhancer-promoter tracking, chromosomal locus localizations are filtered for lobe sigma (120 nm-300 nm) and lobe separation (700 nm-1500 nm). Locus localizations from the nucleus are then cropped. Only trajectories that last longer than 100 frames were used to calculate MSD, MSCD and VAC.
If two localizations with distance < 100 nm are detected in the cropped localizations, the average position of the two localizations was assigned as the localization in that frame. If the two localizations have distance > 100 nm or if more than 3 localizations are detected in the cropped localization, these localizations are removed.
To acquire single-molecule bleaching data, loci were initially imaged at a relatively low excitation power density (2 W/cm^2^ for Cy5 and 1.4 W/cm^2^ for Cy3) to prevent any potential photobleaching for several seconds. This data served to estimate the initial locus brightness before photobleaching. Subsequently, to observe stepwise bleaching of individual fluorophores, the power density was increased to 19 W/cm^2^ for Cy5 or to 71 W/cm^2^ for Cy3, and the loci were imaged until complete photobleaching occurred. Nuclei labeled with Hoechst 33342 were imaged separately at 1 W/cm^2^. To quantify the number of fRNP bound to a single locus, we employed an approach similar to previous methods^95^. Initially, an 11-pixel × 11-pixel region of interest (ROI) centered on the image of a given locus was cropped from both the low and high excitation power movies. For each frame, the background was estimated as a mean value of the 40 edge pixels of this ROI and was subtracted from the values of all pixels in the ROI. After background subtraction, a smaller 5-pixel × 5-pixel ROI was cropped from the center of the 11-pixel ×11-pixel ROI. The number of photons in each frame (the brightness) was estimated as the sum of the values of all pixels in the 5-pixel × 5-pixel ROI, considering the photoconversion gain of the camera. Plotting the brightness against time, an intensity curve was generated with discrete steps, as shown in Figure 3O. Adapting from a previous approach^96^, a semi-automated maximum likelihood change-point finding algorithm was deployed to determine the time points of each step. As each step likely represents the photobleaching of a single molecule, the difference of the averaged brightness values before and after the step was computed to estimate the brightness of a single emitter. Multiple estimates for single-molecule brightness in a given locus were further averaged to reduce noise. The brightness of the locus in the low excitation power dataset was estimated by averaging the brightness within several frames at the beginning of the low power movie. This low power brightness estimate was scaled up to the high excitation power case by using the ratio between the laser powers from the low and high power conditions. This initial locus brightness was then divided by the estimate for the brightness of a single molecule, thereby yielding the number of fRNPs bound to an individual locus. MATLAB was used to perform all calculations in this analysis.
The ensemble-averaged mean-square displacement (MSD) was calculated (Equation 4)MSD(τ)=〈(R→(t+τ)−R→(t))2〉=1N∑i1m∑m[R→i(mt+τ)−R→i(mt)]2
Here, N is the number of trajectories over which the ensemble average is taken. The index m runs from 1 to a specific value given the lag time τ=tE×n, providing the time average of each trajectory. Here tE is the exposure time and n is the number of frames spanning a time lag. The final MSD is an average over the ensemble and over time^57, 97^. We used the following expression for fitting the experimental MSD curves, which accounts for both static localization errors due to finite localization precision and dynamic errors due to motion blur^54^.
The expression was developed for anomalous diffusion consistent with fractional Brownian motion (fBM). Here d is the dimensionality, D∗ is the effective diffusion coefficient, α∈(0,2] is the power law exponent of MSD with α<1 indicateing subdiffusion, and 2σ2is the static error term including the spreading of the PSF due to the movement of the locus. The second to last term describes the dynamic error effects. We fitted the first 50 lags of MSDs to Eq. 5 in MATLAB with free parameter D∗,α, and σ by minimizing the square error in the log space.
Mean square change in distance (MSCD) ^98^ was calculated as (Equation 6)MSCD(τ)=〈(ΔR→(t+τ)−ΔR→(t))2〉=1N∑i1m∑m[ΔR→i(mt+τ)−ΔR→i(mt)]2
In Equation 6, ΔR→(t)=R→i(t)−R→j(t) is the displacement between the two loci in channel i and channel j.
For velocity autocorrelation (VAC), each curve is calculated as (Equation 7)cvδ(τ)=〈v(t+τ)⋅v(t)〉
Here δ is the time interval during which displacement occurred for calculating the (Equation 8)v(t)=1δ(r(t+δ)−r(t))
And the τ is the time lag over which the correlation of velocities is examined. The normalized discrete velocity autocorrelation function for fractional Brownian motion (fBM) is (Equation 9)cvδ(τ)cvδ(0)=∣τ−δ∣α+∣τ+δ∣α−2∣τ∣α2δα
The blue line in Figure 5J and Figure S6N is the theoretical prediction of fBM using Eq. 9 and α=0.38. For velocity cross-correlation (VCC) of two sites i and j, each curve is calculated as (Equation 10)cvvδ(τ)=〈vi(t+τ)⋅vj(t)〉
δ is the time interval during which displacement occurred for calculating the velocity for each (Equation 11)vi(t)=1δ(ri(t+δ)−ri(t))
And the τ is the time lag over which the correlation of velocities is examined.
We applied a viscoelastic coarse-grained Rouse polymer model developed before^65^ to fit Oligo-LiveFISH tracking data. Briefly, the polymer chain is discretized into monomer segments which experience hydrodynamic drag and are connected into a linear chain by Hookean springs. The spring coefficient between the monomers is k=3kBTb−2, where kB is the Boltzmann constant, T is the temperature and b is the Kuhn length. In our viscoelastic Rouse polymer model, each monomer exhibits a drag coefficient ξ and a viscoelastic response to motion that is mediated by the memory kernel (Equation 12)K(t)=(2−α′)(1−α′)∣t∣−α′ α′ refers to the power-law exponent associated with the MSD of a single particle in viscoelastic medium and equals to 2 α, where α is the MSD power-law exponent for the polymer as defined in Eq. 5. By fitting the experimental MSD using Eq.5, we got α=0.43 and 0.48 for site 1 and 2 and we chose a mean α=0.45. Therefore, we assign α′=2α=0.9. The dynamic motion of polymer is governed by the fractional Langevin equation^65^: (Equation 13)ξ∫0tdt′K(t−t′)∂R→(n,t′)∂t=k∂2R→(n,t)∂n2+F→(n,t) where the left-hand term indicates contributions from viscoelastic drag described by the memory kernel K(t−t’), the first right-hand term denotes Gaussian springs with spring constant k and, F→(n,t) denotes Brownian force that obeys the fluctuation-dissipation theorem.
For a finite polymer chain of N Kuhn segments, the Rouse time τR is given by (Equation 14)τR=[N2b2ξkBT]1α′
The Rouse relaxation time τR(Δn), which refer to as communication time in the text, between two monomer segments with distance Δn is given by (Equation 15)τR(Δn)=[(Δn)2b2ξkBT]1α′ and in the intermediate regime where δ is large enough for stress to communicate between two sites but small enough that it has not reached the chain ends, τR(Δn), can be approximated as (Equation 16)δτR(Δn)=δτR(∣Δn∣N)−2α′
Therefore, communication time τR(Δn) increases with genomic distance with a power of ~2 (exact 2∕α′=2.22,α′=2α=0.9).
By generating a series of theoretical VCC using τ(Δn) as a variable and finding the VCC which minimizes the mean squared error (MSE) with respect to the experimental VCC, we find τR(Δn) for different sites listed in Table S4. An MSE is calculated for each δ and τR(Δn) separately and these MSEs are averaged to obtain the final MSE to be minimized. For sites 1 & 2, τR(Δn) was examined from 0.01 s to 10 s with 0.01s increments. For sites 2 & 3 and 3 & 4, τR(Δn) was examined from 1 s to 5000 s with 1s increments.
The predicted τR(Δn) from polymer theory can be calculated using τR(Δn=28kb)=0.44s as a reference, assuming that Rouse intersegment relaxation time follows Eq.16. For example, τR(Δn=284kb)=(284kb28kb)20.9×0.44s=76s, which is very close to the 89 s by fitting experimental VCC. On the other hand, the effective genomic length Δn can also be calculated using Eq. 15 by inputting experimental τR(Δn) as known variable (Table S4).
Spatial interaction between sites were generated with Juicebox using vanilla coverage square root (VC_SQRT) normalized Hi-C signal (GSM4194450) at 50 kb resolution. Hi-C was plotted using Juicebox. The relationship between genome-averaged contact probability P(s) and genomic distance was adopted from a previously simulations with steady-state extrusion^67^. The spatial contacts from Hi-C were normalized to match the contact probability P(s) using sites 1&2 as a reference.
A Gaussian Mixture Model (GMM) was applied to identify clusters in Figure S7B by assuming that all data points are drawn from a probability density function composed of k independent Gaussian components^62, 63^. We fit the GMM parameters using the built-in MATLAB function “fitgmdist”, which uses an expectation maximization algorithm to determine maximum likelihood parameters based on the observed data (1000 max iterations). We repeated this for k=1 to k=10 and selected the final model (k=2) based on the minimum value of the Bayesian Information Criterion (BIC), which introduces a penalty for each additional term to prevent overfitting. For each point, the posterior probability of being a member of a cluster was calculated using the MATLAB function “posterior” and the data points were assigned to the clusters with higher membership probability (>0.5).
We used experimentally calculated SNR as predictors in a linear regression (Equation 17)SNR=β0+xβT+ε x: = (number of probes, GC content, crRNA length, crRNA structure, distance to H3K27ac peak, ATAC, H3K4me3, H3K4me1) (Equation 18)β:=(β1,…,βk)T,k=8
The regression model is fitted using the glmnet R package (version 4.1-8) by elastic net regularization with the α set to 0.5, and the penalty parameter λ which is given by the model with the minimum mean squared error in cross-validation curve. We first examined 15 features with nonzero coefficients the number of distinct gRNA species (12, 24, 48, 96 or 192) , the GC percentage of the spacer sequence, the average length of crRNA, the average secondary structure scores of crRNAs, the average secondary structure scores of crRNA:tracrRNA duplexes, the average distance of the target regions to the closest H3K27ac peak, the average distance of the target regions to the closest ATAC peak, average H3K27ac signal, average ATAC signal, average H3K4me1 signal, average H3K18ac signal, average H3K56ac signal, average H3K9ac signal, average H3K9me3 signal, and average H3K4me3 signal. We performed correlation analysis and removed the highly correlative variables and the selected 8 the number of distinct gRNA species, the average GC percentage in the spacer sequence, the average length of crRNA, the average secondary structure scores of crRNAs, the average distance of the target regions to the closest H3K27ac peak, average ATAC signal, average H3K4me1 signal, and average H3K4me3 signal. The average secondary structure scores of crRNAs were calculated by RNAfold [--gquad] on the crRNA sequences and the crRNA:tracrRNA duplexs. The model was selected by 10-fold cross-validation.
We randomly selected two samples as test set and remaining 17 samples as training set. We used the training set for parameter selection and model training. Finally, we used the model to predict the variable and calculate the square error according to the real value. To evaluate the performance of the model, we compute the squared errors 100 times, and the average square error is smaller for the model trained by selected variables compared to the model trained by all variables.
The statistical values are represented as mean ± 1S.D. unless otherwise specified. For Figure 2B, 3N, 7E, 7L-M, S3B, S5L, the p values were calculated using two-tailed unpaired t test with Welch’s correction in Graph Pad. n.s., not significant (p > 0.05); *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001. For Figure S3C-F, the Peason correlation coefficient r and its significance level (p value) were calculated in Graph Pad. For Figure 2G, 2H, S4C, S4K, S4L, the Peason correlation coefficient r and its significance level (p value) were calculated in R.