Abstract
Oncogenic extrachromosomal DNA elements (ecDNA) play an important role in tumor evolution, but our understanding of ecDNA biology is limited. We determined the distribution of single-cell ecDNA copy number across patient tissues and cell line models and observed how cell-to-cell ecDNA frequency varies greatly. The exceptional intratumoral heterogeneity of ecDNA suggested ecDNA-specific replication and propagation mechanisms. To evaluate the transfer of ecDNA genetic material from parental to offspring cells during mitosis, we established the CRISPR-based ecTag method. ecTag leverages ecDNA-specific breakpoint sequences to tag ecDNA with fluorescent markers in living cells. Applying ecTag during mitosis revealed disjointed ecDNA inheritance patterns, enabling rapid ecDNA accumulation in individual cells. After mitosis, ecDNAs clustered into ecDNA hubs, and ecDNA hubs colocalized with RNA polymerase II, promoting transcription of cargo oncogenes. Our observations provide direct evidence for uneven segregation of ecDNA and shed new light on mechanisms through which ecDNAs contribute to oncogenesis.
ecDNAs are vehicles for oncogene amplification. The circular nature of ecDNA affords unique properties, such as mobility and ecDNA-specific replication and segregation behavior. We uncovered fundamental ecDNA properties by tracking ecDNAs in live cells, highlighting uneven and random segregation and ecDNA hubs that drive cargo gene transcription.
See related commentary by Henssen, p. 293.
This article is highlighted in the In This Issue feature, p. 275
Introduction
Tumor evolution drives intratumor heterogeneity, which is a source of therapy failure and resistance (1, 2). Genomic instability and chromosomal structural variations, including oncogene amplification, play a critical role in driving tumor evolution (3, 4). Focal amplifications in cancer may occur on extrachromosomal DNA (ecDNA) elements, are observed in most glioblastomas and at high frequencies in many cancer types (5–8), and contribute to accelerated tumor growth and poor patient survival. ecDNAs are 50-kb to 5-Mb genomic elements containing genes and/or regulatory sequences (9, 10). Acentromeric and atelomeric features of ecDNAs suggest uneven ecDNA segregation during mitosis, leading to discordant ecDNA inheritance and promoting rapid ecDNA accumulation in a subpopulation of cancer cells (11). Clonal variability of ecDNAs in neuroblastoma cells has been observed and provided evidence toward this hypothesis (12). There is limited knowledge of ecDNA behavior during DNA replication and of the ecDNA mobility in proliferating cancer cells. FISH and 4′,6-diamidino-2-phenylindole (DAPI) staining of fixed cells have been used to demonstrate that oncogenes can reside extrachromosomally in cancer (5, 6, 8, 9). These static readouts of the number of ecDNA copies in single cells are unable to record behavioral patterns. Genome engineering technologies based on the CRISPR-associated RNA-guided inactive endonuclease Cas9 have been leveraged to visualize DNA in living cells. In recent studies, this technique allowed real-time tracing of the dynamic reorganization of genomic DNA during mitosis (13), programmable three-dimensional (3-D) genome interactions (14), and chromosome translocation induced by genome editing (15). Although originally designed for labeling of repetitive sequences (13), recent advances in CRISPR-based live-cell imaging techniques have also enabled visualization of nonrepetitive chromosome loci (16, 17). These techniques collectively help advance our ability to visualize genome organization during both physiologic and pathologic cell states, as well as in response to CRISPR-guided perturbations. ecDNA sequences are indistinguishable from their parental chromosomal DNA, barring ecDNA-specific breakpoint sequences, which provide an opportunity for live-cell ecDNA imaging. Here, we report a CRISPR-based DNA tracking system (ecTag) that leverages DNA breakpoint junctions to label ecDNA elements with multiple fluorescent molecules. We applied this technology to understand ecDNA spatiotemporal dynamics and the mechanisms by which ecDNA contributes to intratumoral heterogeneity.
Results
ecDNA Shows Increased Intratumoral Copy Number Variability
Although ecDNA has been nominated as a key factor contributing to intratumoral heterogeneity resulting from suspected unequal segregation (5, 7, 18), there is a paucity of direct, experimental evidence supporting this assumption. The proposed model of ecDNA inheritance, in contrast to canonical inheritance of linearly amplified DNA on chromosomes, can account for a high degree of intratumoral multiplicity of ecDNA copy number (Fig. 1A). We hypothesized that the number of ecDNA copies across single cells would be highly variable, whereas the copy number of genes amplified linearly on chromosomes is expected to be identical at the single-cell level. To evaluate the distribution of the number of ecDNA copies per cell, we performed interphase FISH on four glioblastoma (GBM) tumor tissue samples (SM006, SM012, SM017, and SM018) and a pair of primary and recurrent GBM neurosphere lines derived from the same patient (HF3016 and HF3177). We have previously found the GBM oncogene EGFR to be focally amplified in all four GBMs and both neurosphere lines (8, 19). As a control, we included probes mapping to chromosome 7, which was broadly amplified at a low level in all six specimens. We evaluated the fit of EGFR-containing ecDNA copy number distributions to discrete probability distributions using the Akaike information criterion. We observed that a Gaussian distribution showed the best fit in most cases, demonstrating that the number of ecDNA copies varied widely across cells and implicating uneven segregation of ecDNA (Fig. 1B and C; and Supplementary Fig. S1A–S1D). These findings are a slight deviation from previous observations implying that uneven ecDNA segregation derives a binomial ecDNA distribution across cells (12). The copy numbers of chromosome 7 appeared more evenly distributed (Fig. 1B and C, bottom) but not stable, possibly as a result of subclonal chromosome 7 tetraploidy, which is common in GBM, with cells residing in different stages of the cell cycle and noise levels of the assay. We calculated the median absolute deviation (MAD) of the distribution of copy numbers, which is a metric that represents variability independent of copy number level, and found the MAD of EGFR ecDNA copies to be significantly higher than the MAD of chromosome 7 copies (average MAD 10.25 vs. 1.61; Fligner–Killeen test, P < 1.5e-08 in all samples). Representative images of EGFR-containing ecDNA copies in cells containing identical numbers of chromosome 7 reflect the impact of ecDNA on intratumoral heterogeneity (Fig. 1B and C, top).
To expand our observation, we assessed FISH images across a collection of different cancer cell lines and of genes that were recently shown to reside on either linear or circular amplicons by whole-genome sequencing (9). In total, we compared FISH signals from 7 genes on ecDNA and 16 linearly amplified genes. Although we observed considerable variability in single-cell copy number of both linear and ecDNA amplicons, ecDNA MADs (median MAD 43 ± 33.18) were significantly higher than linear amplicon MADs (median MAD 1.48 ± 1.18; Fig. 1D; Supplementary Fig. S2). The difference in copy number distribution between ecDNA and linear amplicons corroborates previous circumstantial evidence that ecDNA segregates unevenly (7, 12).
Intratumoral heterogeneity, which impairs treatment response, is marked by genomic variability as well as intercellular diversity in gene and protein expression (20, 21). To examine whether the aptitude of ecDNA for enhancing intratumoral diversity ultimately affects the heterogeneity of functional protein expression, we determined the association of EGFR copy number with EGFR protein expression at the single-cell level (Fig. 1E and F). In all samples, we observed a positive correlation between EGFR ecDNA copy number and EGFR protein expression (Fig. 1E and F, bottom) but not β-actin protein expression (Supplementary Fig. S3A and S3B; Fig. 1G). The moderate correlation implies additional factors may determine expression levels.
CRISPR-Based Labeling Enables Live-Cell ecDNA Tracking
To understand how ecDNA heterogeneity is derived, we developed ecTag, a CRISPR-based DNA labeling method to study ecDNAs in live cells. ecDNAs are formed by DNA breakage followed by end-to-end ligation of DNA segments, resulting in one or more ecDNA breakpoint junctions (refs. 22–25; Fig. 2A). The sequences covering the breakpoint sites are unique and cannot be detected in the parental linear chromosomes. ecDNA-specific breakpoint sequences provide an opportunity for the design of single-guide RNAs (sgRNA) to label or target ecDNA. ecTag was developed by employing Casilio (17, 26), a hybrid technique that combines dead Cas9 (dCas) labeling and Pumilio RNA binding to recruit multiple fluorescent protein molecules at a prespecified sgRNA target locus (Fig. 2A). sgRNAs designed with programmable Pumilio/FBF (PUF) RNA binding sites (PUFBS) achieve target DNA binding through a spacer sequence mapping the ecDNA breakpoint and also enable recruitment of fluorescent molecules conjugated with PUF (Fig. 2A). We evaluated ecTag to label ecDNA-specific breakpoints. To find targetable breakpoint sequences, we analyzed whole-genome sequencing data of the HF3016 neurosphere line to reconstruct the structure of the ecDNA element (ref. 27; Supplementary Fig. S4A–S4E). This identified four unique ecDNA structures, harboring an EGFR fragment (exon1), the full EGFR coding sequence, and the noncoding genes CCAT and CCDC26. We labeled these four lesions as ecEGFRx1, ecEGFR, ecCCAT1, and ecCCDC26, respectively. We designed four primer pairs (Supplementary Fig. S4D, yellow arrows) that allowed extraction of an ecDNA breakpoint fragment from each ecDNA on agarose gels (Supplementary Fig. S5A). We then performed Sanger sequencing to define the precise breakpoint sequence (Supplementary Fig. S5B; ref. 28).
To validate that the target breakpoints were specific to the ecDNA and were extrachromosomal, we used dual FISH, in which two DNA bacterial artificial chromosome (BAC) library FISH probes corresponding to the DNA sequence on either side of the breakpoint were labeled with two different fluorescents (Methods; Supplementary Table S1; Supplementary Fig. S6A). With this approach, a merged or adjacent fluorescent signal marked an extrachromosomal breakpoint (Supplementary Fig. S6B). We also included a second neurosphere line, HF3177, which was derived from the recurrent GBM from the same patient from whom HF3016 was established; therefore, both cell lines were likely to share the same ecDNA amplifications. We have previously found that the PC3 prostate cancer cell line contains ecDNAs (9), but very different in sequence from those in HF3016/HF3177, and therefore we used PC3 as a negative control. The dual FISH analysis showed breakpoints colabeled with two-color probes outside of chromosomes (Supplementary Fig. S6B). We counted breakpoint-positive metaphase cells and found that breakpoints were shared between primary (HF3016) and recurrent (HF3177) cell lines but in different ratios (Supplementary Fig. S6C, left and middle). We observed some signal in PC3 cells, which is likely due to nonspecific binding of the probes (Supplementary Fig. S6C, right). Only signals labeled with both color probes were counted as derived from ecDNA (Supplementary Fig. S6D). The distribution of the number of breakpoint signals per cell confirmed the random segregation pattern (Supplementary Fig. S6E). We observed that a subset of neurosphere cells presenting with homogeneously staining regions stained with one of the two-color probes (Supplementary Fig. S6F). The variability of breakpoint quantities, which may reflect the evolution of the ecDNA structure (8), was confirmed by breakpoint PCR (BP-PCR; Supplementary Fig. S7). These results showed that ecDNA-specific breakpoints can be leveraged to visualize ecDNA in single cells through fluorescence microscopy.
To engineer the ecTag live-cell ecDNA tracking system, we cloned breakpoint-specific sgRNAs with 15 or 25 PUFBS repeats. Cotransfection of sgRNAs, catalytically inactivated Cas9, and Clover-PUF fusion protein–expressing plasmids allowed the enrichment of fluorescent signals at the targeted ecDNA breakpoint loci (Fig. 2A). To validate the targeting efficiency of breakpoint-specific sgRNAs, we performed an in vitro cleavage assay on HF3016 and HF3177 cells and confirmed on-target efficiency of sgRNA (Supplementary Fig. S8).
To verify the cell-type specificity of the four selected ecDNA breakpoints (ecEGFRx1, ecCCAT1, ecEGFR, and ecCCDC26), the on-target HF3016 and HF3177 cell lines and PC3 control cell line were cotransfected with three components: (i) breakpoint-specific sgRNA (15 PUFBS repeats), (ii) dCas9, and (iii) Clover-PUF fusion protein–expressing plasmid. Each component was prepared as an individual plasmid. We included two positive control sgRNAs: (i) sgRNAs labeling an intronic region on chromosome 7 (Chr7) or the chr 3q29 gene MUC4 as representative linear DNA controls, and (ii) a nonrepetitive EGFR-targeting sgRNA. As a negative control, we included an sgRNA targeting the yeast gene GAL4. Cells expressing the sgRNA target region showed an abundance of nuclear spot signals (Fig. 2B, representative images). The spot signals did not result from fluorescent molecules specifically aggregating in the nucleolus (Supplementary Fig. S9A). We confirmed that ecTag transfection did not affect the cell proliferation rate (Supplementary Fig. S9B) or result in nuclear atypia, which is a characteristic of HF3016 neurosphere cells (Supplementary Fig. S9C). The fraction of ecTag-targeting cells for ecDNA-sgRNAs in HF3016 and HF3177 was 39.5% and 52.8%, respectively, compared with an off-target 6% in PC3, indicating that ecDNA-targeting sgRNAs were specific to HF3016 and HF3177 cells. The control sgRNAs (nonrepeat region of Chr7-, MUC4-, and EGFR-targeting sgRNA) showed comparable fractions of ecTag-targeting cells in HF3016, HF3177, and PC3 (38.2%, 40%, and 33.8%, respectively). The Gal4-targeting sgRNA used as negative control was detected in 1.9% in HF3016, 2.1% in HF3177, and 7.9% in PC3. The population of ecTag-targeting cells for ecDNA-sgRNAs without dCas9 in HF3016 (3.6%) demonstrated the minimal amount of nonspecific signals generated by sgRNA-Clover coupling without dCas9. As the ecDNA-targeting sgRNAs map to a specific ecDNA breakpoint, not all ecDNA molecules are being tagged. Thus, the ecEGFRx1 or ecEGFR breakpoint regions may not be part of all EGFR-containing ecDNAs (Supplementary Fig. S4E), and these sgRNAs mark fewer signal foci compared with EGFR DNA FISH (Fig. 1C). The copy number distribution pattern of ecTag signal foci per cell was comparable to those observed using FISH (Supplementary Fig. S10A–S10C). Two-color imaging of DNA-FISH and ecTag ecDNA labeling demonstrated that the spot signals derived from the ecTag accurately mapped the specific ecDNA breakpoint with 81.2% normalized targeting efficiency and 67.3% on-target efficiency, on average (Fig. 2C and D; Supplementary Fig. S11A–S11D). The targeting efficiency of applying ecTag using guides designed to map chromosomal controls (Chr7 and MUC4) was determined by measuring signal foci paired with each control DNA probe out of the number of control DNA probe signal foci (Fig. 2E; Supplementary Fig. S11E and S11F). The ecTag-targeting efficiency was also confirmed by performing EGFR DNA FISH on EGFR-targeting ecTag-transfected HF3016 cells (Supplementary Fig. S11G). We found that, on average, EGFR-targeting ecTag was able to detect 42% of FISH probe–labeled EGFR. The lower specificity observed in this experiment can be explained by multiple ecDNAs containing a partial EGFR region that were unable to be captured by a single sgRNA mapping to a nonrepetitive single region. The number of ecTag signals accompanying the DNA FISH signal showed a broad distribution across cells but not in chromosomal controls (Chr7 and MUC4), demonstrating that ecTag is able to recapitulate the previously observed distribution pattern of ecDNA (Fig. 2F).
ecDNA Spatiotemporal Tracking Shows Uneven Segregation of ecDNA during Mitosis
Centromeres provide attachment sites for spindle microtubules to enable chromosome segregation, and the acentromeric character of ecDNA therefore implies unequal segregation in mitosis (5, 29). Using our ecTag-based ecDNA tracing system, we sought to evaluate the distribution of ecDNA following cell division. We monitored mitosis in HF3016 neurosphere cells with fluorescent labels attached via ecTag to ecDNAs. We found that the fluorescent signal was diluted during cytoplasmic division, which may be explained by the level of DNA compaction during metaphase (30) or potential ecDNA clustering during mitosis (31), resulting in an inability of sgRNAs to bind target sequences (Fig. 3A). Once the telophase finished and the two daughter cells entered interphase, the fluorescent signals reestablished, again visualizing ecDNA molecules. We found that daughter cells often inherit different numbers of ecDNAs (Fig. 3A; Movie 1). We quantified fluorescent signals in offspring cells and observed that Chr7- and MUC4-derived signals showed uniform segregation (Supplementary Fig. S12A and S12B), reflected by a Pearson correlation of 1 or near 1, whereas the same analysis of ecDNA inheritance showed a marginal and nonsignificant correlation (Fig. 3B). To visualize the segregation of ecDNA over the full time window of a cell division, we established a ecTag-stable PC3 model through serial transduction and drug selection of each of the Casilio-based ecTag components (dCas9, Clover, and sgRNA; Supplementary Fig. S13A–S13C). Time-lapse images of the ecTag-stable PC3 model confirmed that ecTag signals are diluted as the cells enter into metaphase; thus, not only the ecDNA-specific signal foci but also the entire Clover spread across nuclei was not visible during metaphase (Fig. 3C). Together, both the postmitosis ecDNA distribution in the transient ecTag H3016 cells and the ecTag-stable PC3 cells demonstrate, unequivocally, that ecDNA segregates unevenly.
To obtain better insight into the discordant inheritance pattern of ecDNA, we determined the distribution of single-cell ecDNA and linear DNA copy numbers every 2 days, relative to the 2-day doubling time of HF3016 cells (Supplementary Fig. S14). ecDNA copy number continued to be highly variable over several cell doublings, in comparison to the Chr7 and MUC4 controls (Fig. 3D). In contrast, the average copy number of controls Chr7 and MUC4—linear DNA amplicons on canonical chromosomes—remained stable, indicating that the DNA replication of these regions is tightly regulated by cell cycle (7, 18). To rule out the possibility that the distribution pattern of ecDNA copies was caused by cellular heterogeneity, we created three single-cell clones from the well-established PC3 cell line, which contains circularly amplified MYC. We performed MYC/chromosome 8 (Chr8) DNA FISH on three single-cell clones and the PC3 instance from which they were derived (Supplementary Fig. S15). We observed that MYC DNA copy numbers were highly variable (MAD = 7.413–14.826) in original PC3 (PC3-NCI) as well as in all isogenic clones (PC3-C4, PC3-C5, and PC3-C7), whereas Chr8 copy numbers were stable (MAD = 0), in support of uneven ecDNA segregation. Our results suggest that, unlike linear DNA amplicons, ecDNA frequencies continuously fluctuate over time, emphasizing the rapid mode of tumor evolution that ecDNA elements are able to direct in comparison to linear amplifications (8).
The live-cell ecDNA tracking experiments provided a spatiotemporally dynamic feature of ecDNA within a single cell but also suggested that ecDNAs showed a propensity to physically cluster together, not involving the nucleolus (Fig. 3E). We observed that ecDNA signals were significantly larger than controls (Supplementary Fig. S16A), implying a contribution by Clovers from multiple ecDNA molecules and indicating hubs of multiple ecDNAs. We performed stepwise photobleaching, a method to reveal the number of objectives within a molecule complex (32), on Chr7 and ecDNA foci. High laser intensity results in loss of fluorescent molecules, and a longer time to signal intensity decrease indicates a higher number of fluorescent molecules. The resulting intensity drop of ecDNA signals is significantly less in comparison to Chr7, indicating that ecDNA signals consist of a higher number of Clover molecules than Chr7 (Supplementary Fig. S16B). To rule out the possibility that ecDNA clustering is an artifact explained by the high number of ecDNAs in the nucleus, we compared ecDNA interactions in hubs to interactions between telomeres, as there is a similarly large number of telomeres in each nucleus (Supplementary Fig. S17A–S17F). To achieve this, we used an sgRNA targeting repetitive telomere sequences. The distance between two interacting signal foci was measured from the core of one focus to another, every 30 minutes for 7 hours (Supplementary Fig. S17A). We observed that ecDNA signal foci were stably contacting each other and over longer time periods compared with the temporal and dynamic interactions taking place between telomeres (Supplementary Fig. S17A–S17E). To exclude the possibility that the circular nature of ecDNAs resulted in ecTag-induced adhesion, we evaluated ecTag in HF3016 transfected with a dsRED-expressing plasmid (Supplementary Fig. S17F). The interactions of dsRED-targeting ecTag signal showed a dynamic fluctuation pattern similar to what we observed with respect to telomeres (Supplementary Fig. S17A), confirming previous similar observations that plasmids are not likely to form aggregates (33). Next, we employed a dual-color ecTag labeling system to determine whether larger ecTag signals were contributed by multiple individual ecDNA elements (Supplementary Fig. S18A). Two sgRNAs were designed mapping to the same breakpoint to visualize ecDNA aggregation using green and red fluorescent molecules. We observed complete merging of yellow signals or closely assembled two colors indicating ecDNA clustering (Supplementary Fig. S18B).
In live-cell imaging, ecDNA hubs were observed to take place in more than 50% of cells within 48 hours of live-cell imaging (Fig. 3F), expanding previous observations in anaphase cells (31) and suggesting functional relevance. We used multicolor ecTag labeling to tag two different ecDNAs and observed ecDNA containing different cargo genes in close proximity as well as completely merged (Supplementary Fig. S19), suggesting that ecDNA hubs can form independent of ecDNA content. The similar ecDNA signal size distribution in HF3016 (primary) and HF3177 (recurrent) suggested that the ability for forming ecDNA hubs was preserved across time points (Supplementary Fig. S20A and S20B).
ecDNA Hubs Associate with RNA Polymerase II Activity
Recent studies have shown that ecDNA drives high levels of oncogene expression (9, 34, 35). We hypothesized that the generation of ecDNA hubs enhances transcriptional activity. First, we examined whether ecDNA hubs were associated with colocalization of nuclear bodies, including Cajal bodies and promyelocytic leukemia protein (PML) nuclear bodies (36, 37). Cajal and PML bodies have been reported to contain hyperphosphorylated RNA polymerase II (RNAPII), nominating them as sites of active mRNA transcription (38, 39). ecTag-transfected cells were stained using Cajal/PML marker protein antibodies, and a secondary antibody conjugated with red fluorescent molecule was used to capture the primary marker protein antibody (Supplementary Fig. S21A and S21B). We compared colocalization between nuclear bodies and ecDNA, using MUC4 and Chr7 colocalization as a control. Colocalization was defined as two different fluorescent signal foci partially or completely overlapping. The fraction of cells containing ecDNAs colocalizing with nuclear bodies was substantially higher than the fraction of cells containing MUC4 or Chr7 colocalizing with nuclear bodies (Cajal bodies: 45%, 6%, and 0% for ecDNAs, MUC4, and Chr7, respectively; PML bodies, 67%, 10%, and 17% for ecDNAs, MUC4, and Chr7, respectively; Supplementary Fig. S21C and S21D). Because the amount of colocalized signal is linearly correlated to the number of target copies and the amount of ecDNAs is much greater than the Chr7/MUC4 control, we normalized the number of colocalized signal loci per cell by the level of ecTag signal (0.07–0.24 in Cajal body and 0.09–0.16 in PML body, Mann–Whitney test, P < 3e-03 in all samples; 0.03 in Cajal body and 0.05 in PML body, ecDNAs vs. Chr7, Mann–Whitney test, P = 0.5 in Cajal body and P = 0.72 in PML body, MUC4 vs. Chr7, Mann–Whitney test; Supplementary Fig. S21E and S21F). This showed that the interaction between ecDNA and nuclear bodies is not due to the abundance of ecDNAs. We also analyzed the size of colocalized area normalized by the total ecTag signal size. This analysis also showed that a significantly higher proportion of ecDNA signal is merged with nuclear bodies compared with Chr7 (0.02–0.06 in Cajal body and 0.03 to 0.06 in PML body, Mann–Whitney test, P < 2e-03 in all samples, ecDNAs vs. Chr7), whereas the proportion of MUC4 area merged with nuclear bodies (0.01 in Cajal body and PML body, Mann–Whitney test, not significant, P = 0.24 in Cajal body and P = 0.64 in PML body, MUC4 vs. Chr7) showed no significant differences (Supplementary Fig. S21G and S21H). Although we observed interactions between ecDNA and nuclear bodies at rates significantly higher than the controls (Supplementary Fig. S21A–S21H), there was no significant linear correlation between ecDNAs and the number of nuclear bodies (Supplementary Fig. S22A and S22B). This suggests that the relative increase of ecDNA with Cajal/PML body signal is due to the higher abundance of ecDNA and implies that ecDNA is not actively being trafficked toward the nuclear bodies. We measured the average grayscale intensity of nuclear body signals overlapping with ecTag signal foci over a 12-hour window, which demonstrated fluctuations indicating unstable interaction between ecDNAs and nuclear bodies (Supplementary Fig. S23A and S23B), suggesting that ecDNAs generate their own clusters independently from those nuclear bodies.
To determine whether ecDNA signal regions are being actively transcribed, ecTag-transfected cells were stained with RNAPII antibody (Fig. 4Ai and Supplementary Fig. S24A). More than half of the cells (59.9%) contained ecDNAs colocalizing with RNAPII (Fig. 4Aii). The number of colocalized loci per cell normalized by total ecTag signal showed that a significantly higher number of ecDNA loci colocalized with RNAPII compared with Chr7 (0.07–0.42, Mann–Whitney test, P < 8e-03 in all samples). A significantly higher number of ecDNA loci colocalized with RNAPII compared with MUC4 was observed in two ecDNA cases (Mann–Whitney test, P < 0.002, in ecEGFR and ecCCDC26). We also found significantly higher MUC4-RNAPII signal (0.14, Mann–Whitney test, P = 0.04, MUC4 vs. Chr7) in comparison to Chr7, reflecting the active transcription of a linear chromosomal gene (Fig. 4Aiii). In addition to the increased number of overlapping loci, we detected a significantly higher total ecDNA area to be merged with RNAPII (0.01–0.08, Mann–Whitney test, P < 8e-03 in all samples, ecDNAs vs. Chr7) compared with Chr7 (Supplementary Fig. S24B). The number of ecEGFRx1 and ecEGFR ecDNA breakpoints showed a linear correlation with RNAPII count, suggesting that the colocalization event of ecDNAs with RNPII is not random (Fig. 4B). ecDNA hubs colocalizing with RNAPII were found to be significantly larger than ecDNA without colocalization (Fig. 4C, Mann–Whitney test, P < 0.05 in all samples, ecDNAs vs. Chr7). Chr7 and MUC4 signals did not show significant differences in size, suggesting that clustering and recruitment of functional transcriptional machinery are specific to ecDNA (Fig. 4C). As an alternative active transcription marker, we assessed correlation between ecDNA and 5-ethynyl uridine (EU) incorporation (Supplementary Fig. S25A). ecDNA signals colocalizing with the EU incorporated region were significantly larger compared with ecDNA signals without EU colocalization. In contrast, the colocalization between ecDNA signals and Ki-67, which is abundant in the nucleus without an association with transcriptional activity, was unaffected by ecDNA signal size (Supplementary Fig. S25B).
EGFR gene expression was also positively correlated with the size of ecEGFR signals, as cells with a comparable number but physically larger ecEGFR-ecTag foci expressed more EGFR mRNA (Fig. 4D). The number of ecEGFR signal foci did not correlate with EGFR mRNA abundance, potentially as a result of the ecTag tagging only the subset of EGFR-containing ecDNAs containing the target breakpoint (Supplementary Fig. S26A). We did not observe the same gene expression–ecTag signal size correlation for ecEGFRx1, which maps to ecDNAs containing only EGFR exon 1 (Supplementary Fig. S26B), and other ecDNA breakpoint tags (Supplementary Fig. S26C and S26D) or control loci (Supplementary Fig. S26E and S26F). In summary, these results support that ecDNA hubs can serve as sites of active transcription. Clustering of ecDNA may drive increased transcriptional activity of its cargo gene, with a spatial interaction advantage provided by forming ecDNA hubs, thus exposing multiple ecDNA molecules to transcriptional machinery simultaneously.
Discussion
Here, we take advantage of a CRISPR dCas9 technique that enables ecDNA-specific fluorescent tagging to interrogate undiscovered ecDNA biology, including inheritance pattern and dynamic behavior. By doing so, we extend previous in situ single time point observations, which are limited in the level of advance they are able to provide. The CRISPR-based genome visualization shown here expands beyond single time point observations and to live-cell tracking to demonstrate the longitudinal development of extrachromosomal DNA dynamics.
Previous approaches have used three or more sgRNAs to map loci of interest (13, 40), which is not feasible when tagging a unique breakpoint sequence. The Casilio system (26), which leverages the RNA binding domains of the PUF proteins fused with the fluorescent molecule, enabled us to label a single nonrepetitive target locus using a single sgRNA to create ecTag. Applying ecTag for tracing ecDNA in the process of cell mitosis visualized unequal ecDNA segregation during cell division despite the technical challenges of live-cell imaging, such as limited time frames (∼48 hours) as a result of laser-induced cellular stress. The limited accessibility of DNA during mitosis, in which all genomic components including ecDNA are condensed, restricted the ability to visualize DNA from metaphase to telophase. Future technological developments are needed to overcome this limitation. Our results show that ecDNA, through random segregation during mitosis, enhances intratumoral diversity at the genomic level, thus allowing ecDNA accumulation over the course of just a few cell cycles. This observation was reflected by fluctuations in ecDNA copy number distribution over subsequent cell cycles, the rules of which have been further explored (41). The basis for ecTag is the use of a single sgRNA, and its use is therefore not restricted to labeling of ecDNA and includes labeling of other DNA as well. This includes tagging of small circular DNAs, also referred to as extrachromosomal circular DNA (eccDNA). eccDNAs are typically less than 1 kb (42), creating challenges in the use of, for example, DAPI or FISH staining for visualization. eccDNAs derive through random ligation of genomic DNA fragments, and ecTag could be used to target eccDNA-specific breakpoints, in combination with an available protospacer adjacent motif sequence. eccDNAs have been implicated in the immune response, marking them as interesting targets for further discovery (43).
We observed that ecDNAs tend to cluster into hubs, leading to increased transcriptional activity. ecDNA hubs recruited RNAPII, leading to active mRNA expression of cargo genes, highlighting an additional mechanism that explains the exceptionally high levels of ecDNA gene expression reported earlier (34, 35). Our results compound recent discoveries of the wide-open chromatin accessibility of ecDNA (34), the topologic advantage of ecDNA for communicating with regulators (35, 44), ecDNA-driven oncogenic genome remodeling (45), and the function of ecDNAs as a mobile enhancer (46), and they add to mounting evidence that the advantage of tumor cells for maintaining ecDNA extends beyond simple dosage effects on cargo gene transcription. The presence of ecDNA hubs has been independently and in parallel shown by others (47). Nuclear condensates resulting from phase separation have been associated with transcriptional processes, and future work may help to understand how ecDNA hubs relate to biomolecular condensation (48–50). The 3-D topologic orientation of genomic loci on linear chromosomes enables physical interaction between distal regulatory elements and gene promoters (51). The circularization of oncogenes on ecDNAs increases enhancer interactions in ways restricted by insulators when on linear chromosomes (35). The ecDNA hubs described here create additional interaction opportunities between ecDNA oncogene promoters and enhancers. Such ecDNA hubs can serve as transcriptional hotspots by sharing transcriptional machinery. Our results on ecDNA hubs reflect that the physical assembly of multiple ecDNA molecules may be required for optimal transcriptional activity. A comprehensive understanding of ecDNA clustering behavior will help explain the biological roles of ecDNA contributing to gene expression, cell proliferation, and cell motility in cancer. Taken together, our observations build on genome engineering technologies to provide new insights into ecDNA biology, a factor contributing to intratumoral heterogeneity. Defining mechanisms of ecDNA replication and assembly will be needed to understand how ecDNA can be leveraged for cancer therapeutics.
Methods
Human Tumor Specimens
Human glioma resection specimens (SM006, SM012, SM017, and SM018) were obtained from St. Michael's Hospital. All tissue donations were approved by the Institutional Review Board of The Jackson Laboratory and the clinical institutions involved. This work was performed in accordance with the Declaration of Helsinki principles. The molecular subtypes of these glioma tissues were profiled in Johnson and colleagues (19). SM006 = classical; SM012 = proneural + mesenchymal; SM017 = mesenchymal + classical; SM018 = mesenchymal.
Cell Cultures and Cell Lines
Patient-derived glioblastoma spheroids (HF3016 and HF3177) were obtained with written informed consent from patients, with the protocol approved by the Henry Ford Hospital Institutional Review Board. These cells were cultured in neurosphere medium (NMGF): 500 mL DMEM/F12 medium (Invitrogen, 11330) supplemented with N-2 (Gibco, 17502–048), 250 mg BSA (Sigma, A4919), 12.5 mg gentamicin reagent (Gibco, 15710–064), 2.5 mL antibiotic/antimycotic (Invitrogen, 15240–062), 20 ng/mL EGF (Peprotech, 100–15), and 20 ng/mL bFGF (Peprotech, 100–18B). The molecular subtypes of two neurosphere lines were determined in DeCarvalho and colleagues (8). HF3016 and HF3177 = proneural. Human prostate cancer cell line PC3 was a gift from Dr. Paul Mischel at the University of California, San Diego, and cultured in F12-K (ATCC, 30–2004) with 10% FBS (VWR, 97068–085). All cultured cells were tested for Mycoplasma contamination before use with the MycoAlert Mycoplasma Detection Kit (Lonza).
FISH Analysis
For patient tissues, the slightly thawed tissues were transferred to a positively charged glass slide by pressing against the surface of the specimen. The tissue slides were then immediately transferred into Carnoy's fixative (3:1 methanol/glacial acetic acid, v/v), incubated at room temperature for 30 minutes and then air-dried. For interphase cell prep, neurospheres were dissociated into single cells and fixed in Carnoy fixative for 20 minutes, briefly washed with fixative, and resuspended in fixative. Desired amounts of cells were then dropped onto the glass slide and air-dried. A hybridization buffer (Empire Genomics) mixed with EGFR-Chr7 probe (EGFR-CHR07–20-ORGR; Empire Genomics) was applied to the slides, and the slides were denatured at 75°C for 5 minutes. The slides were then immediately transferred and incubated at 37°C overnight. The posthybridization wash was with prewarmed 0.4× saline sodium citrate (SSC) at 75°C for 1 minute, followed by a second wash with 2× SSC/0.05% Tween-20 for 2 minutes at room temperature. The slides were then briefly rinsed by water and air-dried. The VECTASHIELD mounting medium with DAPI (Vector Laboratories) was applied and the coverslip was mounted onto a glass slide. Tissue images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (100×). As excitation laser, 405 nm, 488 nm, and 561 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
Evaluation of Copy Number Distribution Fitness
We evaluated fitness of EGFR-containing ecDNA copy number distribution to one or more discrete probability distributions. We first used descdist function in R package, fitdistrplus (v. 1.1.3), to plot the skewness-kurtosis graph for the choice of distributions (52). Acknowledging limited data points (n < 100) for model fitness, results showed that our data were most likely sampled from a cell population following Gaussian, negative binomial distribution or Poisson distribution. Next, we estimated model fitness by fitting our data to each of these three distributions. Resulting relative likelihood of non-Gaussian models—based on Akaike information criterion values—was significantly (≥10-fold) lower for 8 of 12 fitted models for either negative binomial or Poisson distribution compared with those from Gaussian distribution (Supplementary Fig. S1C and S1D).
Pan-Cancer FISH Analysis
We collected FISH images of four genes presented on ecDNA (MDM2, PDGFRA, EGFR, and MYC) in six cell lines (CA718, GBM6, GBM39, HK359, MB411FH, and PC3) from GBM, medulloblastoma, and prostate cancer cell lines (Supplementary Fig. S1, red bar plots). We also collected FISH images of 16 linearly amplified genes (TADA2A, CCND3, NFKBIA, RORC, EGFR, CCDN1, SBDS, HMGA2, BMP5, ARID5B, ERCC2, BRF2, IRF4, KCND3, TNFRSF13B, and IGFBP1) in 16 cell lines (BP474, COLO205, DU145, EKVX, GSC11, H23, H322, HCC827, HCC1569, HK259, HOP62, OVCAR5, RPMI8226, SKBR3, SN12C, and SW620) from breast, colon, prostate, lung, GBM, ovary, hematopoietic, meduloblastoma, and kidney cancer cell lines (Supplementary Fig. S1, blue bar plots). All images used here were obtained from https://figshare.com/s/6c3e2edc1ab299bb2fa0 and https://figshare.com/s/ab6a214738aa43833391.
ImmunoFISH Analysis
GBM tumor tissue specimens were embedded in Tissue-Tek Optimal Cutting Temperature compound and stored at −80°C until sectioning. Frozen 10-μm tissue sections were obtained on slides using a Cryostat (Leica, CM3050S), and slides were stored at −80°C until processing for ImmunoFISH analyses. For neurospheres, 1.2 × 105 cells were plated with 10% FBS-containing NMGF media into a glass-viewing area of a confocal dish (VWR, 75856–740). The next day, cells were fixed with 4% paraformaldehyde (PFA) at room temperature for 10 minutes and briefly rinsed with 1× PBS. To allow permeabilization, fixed cells were incubated with 0.5% PBST (0.5% Tween-20/1× PBS) for 20 minutes. Fixed cells were then dehydrated by incubating dishes in gradually increasing concentration (70%/80%/90%/100%) of EtOH and air-dried. The dehydrated dishes were stored at −20°C with foil.
FISH analysis was performed on the dehydrated tissue slides or dishes with cells as described above. After the last washing step with water, the slides were then briefly rinsed with washing buffer (0.025% Tween-20/1× PBS) and incubated with blocking buffer (5% BSA/0.3% Triton X-100/1× PBS) at room temperature for 1 hour. Without the washing step, EGFR antibody (#4267, Cell Signaling Technology, 1:100) or β-actin antibody (#4970, Cell Signaling Technology, 1:200) diluted in antibody diluent (1% BSA/0.3% Triton X-100/1× PBS) was applied to the slide and incubated at 4°C overnight (or room temperature for 3 hours). Then slides were washed with washing buffer three times, and the secondary antibody conjugated with Alexa 555 (ab150086, Abcam, 1:1,000) was applied. After a 1-hour incubation, the slides were washed three times and counterstained with VECTASHIELD mounting medium with DAPI (Vector Laboratories). Images were scanned under Leica STED 3×/DLS Confocal with an oil-immersion objective (100×). As excitation laser, 405 nm, 488 nm, and 561 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
Immunofluorescence Staining
HF3016 cells (1.2 × 105) were plated with 10% FBS-containing NMGF media into a glass-viewing area of a confocal dish (VWR, 75856–740). The next day, cells were transfected with ecTag plasmids (83.3 ng dCas9, 83.3 ng sgRNA, and 83.3 ng Clover) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. At 48 hours posttransfection, the dishes were briefly rinsed with 1× PBS three times and fixed with 4% PFA at room temperature for 15 minutes. The fixed cells were then rinsed with 1× PBS and permeabilized by incubating with washing buffer at room temperature for 5 minutes. The dishes were then incubated with blocking buffer at room temperature for 1 hour, and immediately the primary antibody [Coilin, ab87913 (1:200); PML, ab96051 (1:500); RNAPII (1:200), ab193468; Ki-67, ab16667 (1:250); Abcam] diluted in antibody diluent was applied and incubated at room temperature for 1 hour (at 4°C overnight for Ki-67). The dishes were then washed with washing buffer three times, and the secondary antibody conjugated with Alexa 555 (ab150086, Abcam, 1:1,000) was applied. After a 1-hour incubation, the slides were washed three times and counterstained with VECTASHIELD mounting medium with DAPI. Images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (100×). As excitation laser, 405 nm, 488 nm, and 561 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
EU Incorporation
ecTag-transfected cells were prepared as described above (immunofluorescence staining section in Methods). After 48 hours of transfection, EU incorporation was then performed following the manufacturer's protocol (Click-iT RNA Alexa Fluor 594 lmaging Kit, C10330; Invitrogen).
Breakpoint-Specific PCR
Genomic DNA was isolated from each cell line using the QIAamp DNA Mini Kit (Qiagen 51304) or Quick DNA Mini Prep Plus Kit (Zymo Research, D4068). Breakpoint-specific PCR was performed in an automated thermal cycler (C1000 Touch Thermal Cycler; BioRad). Each reaction mixture was prepared with the AccuPrime Taq DNA Polymerase system (Invitrogen, 12339016). The PCR protocol was denaturation at 94°C for 5 minutes followed by 30 cycles comprising denaturation at 94°C for 30 seconds, primer annealing at 61°C for 30 seconds and DNA elongation at 68°C for 1 minute, further extension at 72°C for 5 minutes, and rapid cooling to 4°C. PCR products were analyzed by 1% agarose gel electrophoresis and visualized under UV illumination after SYBR-Safe staining (Invitrogen, S33102). For Sanger sequencing, the PCR amplicons were extracted using the Nucleospin Gel and PCR Clean-up kit (Macherey-Nager, 740609). The bidirectional Sanger sequencing was performed by EtonBio.
Metaphase Dual FISH
Neurosphere cell cultures and PC3 were synchronized at metaphase by treating with 80 ng/mL Colcemid (Roche, 10–295–892–001) overnight. Cells were washed with PBS and incubated with 0.075 mol/L KCl at 37°C for 15 minutes. Samples were then fixed in Carnoy's fixative (3:1 methanol/acetic acid, v/v) according to standard cytogenetic procedures. Metaphase cells were dropped onto glass slides and baked at 56°C for 4 to 9 hours. A hybridization buffer (Empire Genomics) mixed with two probes (red probe/gold probe = 1:1.5; see Supplementary Table S1 for the list of BAC probes used for dual FISH) was applied to the slides, and the slides were denatured at 75°C for 5 minutes. The slides were then immediately transferred and incubated at 37°C overnight. The posthybridization wash was with prewarmed 0.4× SSC at 75°C for 1 minute followed by a second wash with 2× SSC/0.05% Tween-20 for 2 minutes at room temperature. The slides were then briefly rinsed with water and air-dried. The VECTASHIELD mounting medium with DAPI (Vector Laboratories) was applied, and the coverslip was mounted onto a glass slide. Images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (100×). As excitation laser, 405 nm, 488 nm, and 561 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
FISH Validation for ecTag
HF3016 cells (1 × 105) were plated with 10% FBS-containing NMGF media into a glass viewing area of a confocal dish (VWR, 75856–740). The next day, cells were transfected with ecTag plasmids (100 ng dCas9, 200 ng sgRNA, and 100 ng Clover) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. The next day, cells were fixed with 4% PFA at room temperature for 10 minutes and briefly rinsed with 1× PBS. To allow permeabilization, fixed cells were incubated with 0.5% PBST for 20 minutes. Fixed cells were then dehydrated by incubating dishes in gradually increasing concentration (70%/80%/90%/100%) of EtOH and air-dried. The dehydrated dishes were stored at −20°C with foil. A hybridization buffer (Empire Genomics) mixed with the red probe from the set of dual FISH probes was applied to the slides, and the slides were denatured at 68°C for 2 minutes. The slides were then immediately transferred and incubated at 37°C overnight. The posthybridization wash was with prewarmed 0.4× SSC at 50°C for 1 minute followed by a second wash with 2× SSC/0.05% Tween-20 for 2 minutes at room temperature. The slides were then briefly rinsed with water and air-dried. The VECTASHIELD mounting medium with DAPI (Vector Laboratories) was applied, and the coverslip was mounted onto a glass slide. Images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (100×). As excitation laser, 405 nm, 488 nm, and 561 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
PC3-Derived Isogenic Cell Subcloning
Using the PC3 parental cell line, cells were plated in serial dilutions in 96-well plates in such a way as to obtain the last two columns of the plates with 1 and 0.5 cells/well, respectively. A few hours after plating to allow attachment of cells, all wells in the last two columns of the plates were examined manually under a microscope to confirm the presence of no more than a single cell per well, and wells with single cells were marked. Single-cell clones were allowed to grow and used for analyses. The metaphase cells of these PC3-derived clones were prepared as described above (the metaphase dual FISH section in Methods). The metaphase slides then were used for MYC/Chr8 FISH analysis.
Cloning
The backbone vectors for cloning were a gift from Dr. Albert W. Cheng. The 5′-phosphorylated duplexed oligos for the guide sequences of each breakpoint were purchased from IDT. The sgRNA spacer sequences were digested via BbsI (NEB, R3539) and then were cloned into sgRNA-PUFBS expression vectors (single color: pAC1373-pX-sgRNA-25xPUFBSa, dual color: pCR8-sgRNA-15xPUFBSa and pCR8-sgRNA-15xPUFBSc). Guide sequences were then cloned into the digested vector using T4 ligase (Roche, 4898117001).
ecTag Transfection for ecTag-Targeting Cell Counting
The 1 × 105 cells of HF3016, HF3177 were plated with 10% FBS-containing NMGF media into a glass viewing area of a confocal dish (VWR, 75856–740). Then, 1 × 105 PC3 cells were plated with regular media (F12K with 10% FBS) into a glass viewing area of a confocal dish (VWR, 75856–740). The next day, cells were transfected with ecTag plasmids (75 ng dCas9, 150 ng sgRNA, and 75 ng Clover) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. The next day, cells were fixed with 4% PFA at room temperature for 10 minutes and briefly rinsed with 1× PBS. The dishes were then air-dried, and the VECTASHIELD mounting medium with DAPI (Vector Laboratories) was applied and the coverslip mounted onto a dish. Images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (40×). As excitation laser, 405 nm and 488 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
In Vitro sgRNA Test
The in vitro targeting efficiency of sgRNAs was tested using a Guide-it Complete sgRNA Screening System (Takara, 632636) according to the manufacturer's instructions. Breakpoint-PCR amplicons were used as targeting templates and incubated with appropriate sgRNA and recombinant Cas9 (rCas9) at 37°C for 1 hour. The cleaved fragments generated by sgRNA targeting were perceived on agarose gel.
Live-Cell Imaging
Neurospheres were plated into 100-mm tissue culture dishes at 1 million cells per dish with NMGF medium containing 10% FBS. The next day, cells were transfected with ecTag plasmids (1,250 ng dCas9, 1,250 ng sgRNA, and 1,250 ng Clover) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the cells were trypsinized and resuspended in Dulbecco's phosphate-buffered saline (DPBS), including 10% FBS. To isolate ecTag-transfected cells, the Clover-positive cells were then sorted by FACSAria Fusion (BD Biosciences) with a 130-μm nozzle. To recover the sorted cell fitness, the ecTag-transfected cells were then replated into a 384-well plate at 5,000 cells per well with NMGF, including 10% FBS, and incubated at 37°C overnight. The next day, cells were stained with CellMask Deep Red Plasma membrane stain (Invitrogen, C10046) and NucBlue Live ReadyProbe (Invitrogen, R37605). Live-cell imaging was performed on an Opera Phenix High-Content Screening System (PerkinElmer) under 5% CO2 and 37°C. To avoid evaporation of medium, empty wells around the samples were filled with DPBS. All the images were acquired by taking six to eight different focal planes and shown as a maximum-intensity projection. To track ecDNA dynamics, images were acquired every 30 minutes for 24 or 48 hours with 20× magnification. Time-lapse movies and snapshot images were generated and analyzed by Harmony High-Content Imaging and Analysis Software (PerkinElmer).
Size Threshold Analysis of a Single ecTag Signal
In the live-cell ecDNA tracking experiment (Fig. 3D), we noticed saturated fluorescence for some cells. All plated cells were captured using the same laser settings throughout 48 hours, and the images were acquired with 20× magnification. This led to larger single signal foci, as shown in the Chr7 and MUC4 images in Fig. 3D. ecTag signal dissection performed using sgRNAs tagged with fewer PUFBS repeats (×15) and imaged as fixed cells with a higher magnification (100×) detected the upper single signal focus size in four ecDNA groups compared with Chr7 and MUC4 (Chr7 = 0.551 μm2; MUC4 = 0.680 μm2; Supplementary Fig. S18).
Stepwise Photobleaching Analysis
PC3 cells stably expressing the ecTag system targeting Chr7 or ecDNA were fixed in 4% PFA. Microscopy imaging was performed on Dragonfly equipped with a Zyla camera (Andor) and 488-nm laser. Each dot was bleached using a 100% 488-nm laser, and images were acquired every 100 ms for 60 seconds. To measure spot intensity, each spot was cropped, and the average intensity of the cropped region was calculated. Background spot intensity, obtained from an area outside spot loci, was subtracted.
Live-Cell Imaging of Nuclear Bodies
HF3016 cells were plated into 384-well plates at 4,000 cells per well with NMGF medium containing 10% FBS. The next day, cells were transfected with 10 ng dCas9, 20 ng sgRNA, 20 ng mRuby, and 25 ng reporter plasmid (Cajal body, #36906; PML body, #118360; Addgene) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. The next day, cells were stained with NucBlue Live ReadyProbe (Invitrogen, R37605) at 48 hours posttransfection. Live-cell imaging was performed on an Opera Phenix High-Content Screening System (PerkinElmer) under 5% CO2 and 37°C. To avoid evaporation of medium, empty wells around the samples were filled with DPBS. All the images were acquired by taking six different focal planes and analyzed as a maximum-intensity projection. Images were acquired every 30 minutes overnight with 40× magnification. Image analysis for measuring distance between two signal foci was performed in Fiji.
Live-Cell Imaging of Telomeres and dsRED Plasmids
For the imaging of telomeres, 1.2 × 105 HF3016 cells were plated with 10% FBS-containing NMGF media into one of the wells of a glass-bottom dish (Greiner Bio-One, 627870). The next day, cells were transfected with ecTag plasmids (100 ng dCas9, 200 ng Telomere-sgRNA-PUFBSa, and 100 ng Clover-PUFa) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. The next day, cells were stained with 1.0 μg/mL Hoechst 33342 (Invitrogen, H3070) at 48 hours posttransfection. Live-cell imaging was acquired on Dragonfly Confocal (Andor) using an iXon EMCCD camera with Leica HC PL APO 63×/1.47NA OIL CORR TIRF objective mounted on a Leica DMi8 inverted microscope under humidified 5% CO2 and 37°C. Clover images were acquired with a 150-mW solid-state 488-nm laser and 525/50-nm BP emission filter. Z-series covering the full nuclei was acquired with a 0.3-μm step size every 30 minutes. For the imaging of dsRED plasmids, 1 × 105 HF3016 cells were plated with 10% FBS-containing NMGF media into one of the wells of a glass-bottom dish (Greiner Bio-One, 627870). The next day, cells were transfected with dsRED-expressing plasmid (80 ng) and ecTag plasmids (80 ng dCas9, 160 ng dsRED-sgRNA, and 80 ng Clover) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. The next day, cells were stained with NucBlue Live ReadyProbe (Invitrogen, R37605) at 48 hours posttransfection. Live-cell imaging was acquired on Dragonfly under 5% CO2 and 37°C. The images were acquired by taking 74 and 61 different focal planes for the telomere imaging and dsRED imaging, respectively, and analyzed as a maximum-intensity projection. Image analysis for measuring the distance between two signal foci was performed in Fiji.
Dual-Color ecDNA Labeling Experiment
HF3016 cells (3 × 105) were plated with 10% FBS-containing NMGF media into a glass-viewing area of a confocal dish (VWR, 75856–742). The next day, cells were transfected with ecTag plasmids (400 ng dCas9, 200 ng sgRNA-PUFBSa, 200 ng sgRNA-PUFBSc, 200 ng Clover-PUFa, and 200 ng mRuby-PUFc) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. At 48 hours posttransfection, the dishes were briefly rinsed with 1× PBS three times and fixed with 4% PFA at room temperature for 15 minutes. The fixed cells were then rinsed with 1× PBS and counterstained with VECTASHIELD mounting medium with DAPI. Images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (100×). As excitation laser, 405 nm, 488 nm, and 561 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
Neurosphere Doubling Time Test
HF3016 cells were plated in individual wells of a 96-well plate and viable cells were quantified using CellTiter-Glo 3D Cell Viability Assay (Promega) in triplicate wells as per the manufacturer's instructions. Luminescence readings, which represented viable cells, were taken on a Cytation 3 Cell Imaging Multi-Mode Reader (BioTek).
Test for the Dynamic Change of ecDNA Copy Number Distribution
To image ecTag-transfected cells every 2 days (day 2, day 4, and day 6), three sets of cells were prepared. Then, 1.2 × 105 HF3016 cells were plated with 10% FBS-containing NMGF media into a glass viewing area of a confocal dish (VWR, 75856–740). The next day, cells were transfected with ecTag plasmids (83.3 ng dCas9, 83.3 ng sgRNA, and 83.3 ng Clover) with Lipofectamine 3000 (Invitrogen, L3000015). After 24 hours, the media were changed to fresh media. At 2 days posttransfection, the dishes of the first set were briefly rinsed with 1× PBS three times and fixed with 4% PFA at room temperature for 15 minutes. The fixed cells were then rinsed with 1× PBS and counterstained with VECTASHIELD mounting medium with DAPI. Images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (100×). As excitation laser, 405 nm and 488 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software. The second set of cells was processed after 4 days posttransfection and the third set of cells was processed after 6 days posttransfection.
EGFR RNA FISH
Fluorescence (Quasar 670 Dye)-conjugated EGFR DesignReady probe was purchased from Biosearch Technologies. RNA FISH was performed on HF3016 cells transfected with ecTag-labeling components using the manufacturer's protocol. Images were scanned under Leica STED 3×/DLS Confocal with oil-immersion objective (100×). As excitation laser, 405 nm, 488 nm, and 647 nm were used. Z-stack acquired at a 0.3- to 0.5-μm step size was performed and all analysis conducted based on maximum intensity projection images of the 3-D volume of the cells. Images were acquired and processed by LAS X software.
Image Analysis and Data Availability
Macro scripting of FIJI (ImageJ 1.53c) was used for automated image analysis. Speckle inspector function in the BioVoxxel plugin was used for counting the copy number of fluorescent signals. Colocalization analysis was done using the JACoP plugin (Pearson correlation test), the Image Calulator, and the Analyze Particle function of Fiji. Fiji macro codes used for image processing and analysis are available on GitHub (https://github.com/yehyehyeh/EDTB_imageanalysis).
Statistical Analysis
All sample sizes and statistical methods are indicated in the corresponding figure or figure legends. All data were tested for normality using the D'Agostino–Pearson omnibus test. According to the results of the normality test, all data in this study that were not normally distributed were then run through the Mann–Whitney U test (for two groups). The homogeneity of variances between groups was determined by the Fligner–Kileen test. All statistical tests are two-sided. All plots are shown with median and upper and lower quartiles. All statistical tests were performed in GraphPad Prism 8 (GraphPad Software) or R version 4.0.2 (R Development Core Team).
Authors' Disclosures
E. Yi reports grants from American Brain Tumor Association during the conduct of the study, as well as a patent for US20210062250A1 (extrachromosomal DNA labeling) pending to The Jackson Laboratory, which includes E. Yi and R.G.W. Verhaak as inventors. K.C. Johnson reports grants from American Cancer Society during the conduct of the study. S. Das reports grants from CIHR; other support from Alkermes, Medicenna, Xpan Medical, and Subcortical Surgery Group; and personal fees from Cancer Care Ontario outside the submitted work. P.A. Clow reports grants from National Human Genome Research Institute and National Science Foundation during the conduct of the study, as well as a patent for live cell imaging of non-repetitive genomic loci pending. A.W. Cheng reports grants from National Cancer Institute, National Human Genome Research Institute, and National Science Foundation during the conduct of the study, as well as a patent for WO2016148994 issued. R.G.W. Verhaak reports grants and other support from Boundless Bio during the conduct of the study; personal fees from Stellanova, Inc. outside the submitted work; and a patent for US20210062250A1 pending to The Jackson Laboratory. No disclosures were reported by the other authors.
Authors' Contributions
E. Yi: Conceptualization, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. A.D. Gujar: Conceptualization, validation, writing–review and editing. M. Guthrie: Validation, methodology, writing–original draft. H. Kim: Resources, formal analysis, investigation. D. Zhao: Validation, investigation. K.C. Johnson: Resources, investigation, writing–original draft. S.B. Amin: Data curation, investigation, writing–original draft, writing–review and editing. M.L. Costa: Validation, investigation. Q. Yu: Software, methodology. S. Das: Investigation, writing–review and editing. N. Jillette: Methodology. P.A. Clow: Formal analysis, methodology. A.W. Cheng: Resources, supervision, methodology. R.G.W. Verhaak: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.
Acknowledgments
We gratefully acknowledge the contribution of the Microscopy Service and the Single Cell Biology Laboratory at the Jackson Laboratory for expert assistance with the work described in this publication. This work was supported by the following NIH grants: R01 CA237208, R21 NS114873, R21 CA256575, and R33 CA236681 (R.G.W. Verhaak), Cancer Center Support Grant P30 CA034196 (R.G.W. Verhaak; A.W. Cheng), and R01 HG009900 and CCF 1955712 (A.W. Cheng); Department of Defense W81XWH1910246 (R.G.W. Verhaak); and grants from the Musella Foundation, the B*CURED Foundation, and the Brain Tumour Charity (R.G.W. Verhaak). E. Yi is supported by a basic research fellowship from the American Brain Tumor Association (BRF1800014). K.C. Johnson is the recipient of an American Cancer Society Fellowship (130984-PF-17-141-01-DMC).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.