Implementation of FRET-FISH
FRET-FISH is based on the hybridization—in fixed cells—of oligonucleotide (oligo) DNA FISH probes coupled to two fluorescent dyes with overlapping spectra targeting two proximal DNA sequences, so that if the two sequences are closer than the Förster distance, R0, FRET is detected (Fig. 1a). We reasoned that a FISH probe consisting of multiple oligos carrying alternating FRET donor (D) and acceptor (A) dyes, targeting a given genomic locus, should enable probing chromatin compaction at that locus by measuring FRET efficiency. Because FRET efficiency is influenced by the dipole orientation and the molecular distance between D and A dyes, we empirically tested three different FRET-FISH probe designs, all targeting a region of 20 kilobases (kb) encompassing the human MYC gene locus, to check which probe design yields the highest FRET efficiency (Fig. 1b and Supplementary Table 1). The first design is essentially identical to the iFISH probe design that we previously described16 and consists of primary oligos with a target (T) sequence complementary to the genomic DNA target (60 nucleotides (nt) instead of 40 nt as in iFISH) and left (L) and right (R) adapter sequences that are needed for PCR during the production of the probes and serve as docking sites for fluorophore-conjugated detection oligos (L* and R*, respectively) (Fig. 1b, Design 1). In the second design, the L and R sequences of each primary oligo are extended with a left and right stabilizing sequence (LSS and RSS, respectively), where the 3’ 6 nt of the RSS in one oligo are complementary to the 5’ 6 nt of the LSS in the next primary oligo along the linear genomic target (Fig. 1b, Design 2). We reasoned that this design should stabilize the proximity between D and A dyes, thus enhancing FRET efficiency. Lastly, in the third design, the stabilizing sequence is added to the L* and R* detection oligos, so that the 3’ 6 nt of an L* oligo can anneal to the 5’ 6 nt of the R* oligo bound to the next primary oligo along the linear genomic target (Fig. 1b, Design 3). In all three cases, we designed each probe to contain 134 primary D-A oligo pairs with a minimum distance of 5 nt between the 3’ of the T sequence of a primary D oligo and the 5’ of the T sequence of the next primary A oligo along the linear genomic target. We used Cy3 and Cy5 as D and A dyes, respectively, since they are widely used in FRET experiments due their relatively high brightness and lower price compared to other fluorophores.
To test each probe design, we hybridized HAP1 human chronic myeloid leukemia cells inside custom-designed 9-well silicone-coated coverslips to minimize technical variability between samples and to compare all three probe designs within the same experiment (Supplementary Fig. 1a and Methods). We calculated the FRET efficiency by dividing the signal intensity detected in the A channel (Cy3 excitation and Cy5 emission) by the sum of the signal intensities in the D (Cy3 excitation and Cy3 emission) and A channels (Methods). In two independent experiments, all the three designs produced readily detectable FRET signals, with Design 3 yielding the highest FRET efficiency (39.5% ± 6.8%, mean ± s.d.), most likely because the presence of complementary annealing sequences in the detection oligos stabilizes the primary D-A oligo pairs that are in physical proximity (Fig. 1c and Supplementary Fig. 1b). Surprisingly, in both replicate experiments, Design 2 yielded the lowest FRET efficiency (24.8% ± 5.2%, mean ± s.d.), possibly because the stabilization sequence in the primary oligos hindered the energy transfer between the D and A dyes or because it acted as a quencher (Fig. 1c and Supplementary Fig. 1b). Independently of the probe design, the FRET efficiency was consistently lower in control samples in which only D or A primary oligos were hybridized, while detection oligos targeting both were included, demonstrating the specificity of our approach (Fig. 1c and Supplementary Fig. 1b). Importantly, the FRET efficiency distributions were similar between controls, indicating that the influence of cross-excitation was similar in all the designs tested. We obtained similar results using 6-well chambered coverslips where two designs were compared side-by-side in each experiment (Supplementary Fig. 1c-f). Altogether, these results demonstrate that proximity between in situ hybridized oligos carrying alternating FRET acceptor and donor dyes can be detected by measuring FRET efficiency in fixed cells.
Optimization of FRET-FISH probe design to measure local chromatin compaction
Prompted by these results, we then sought to further optimize the FRET-FISH probe design to measure local chromatin density at selected genomic loci. We reasoned that the three probe designs that we tested might not allow detecting changes in chromatin compaction, given that in those probes D and A primary oligos bind very close along the linear genome. Instead, we designed probes consisting of D and A primary oligos separated by larger linear genomic distances, so that their physical distance in the nucleus would be sensitive to the extent of local chromatin compactness. We tested three different spacing (S) distances (50, 150, and 300 nt) between consecutive oligos, as in a previous study on purified nucleosomes it was shown that FRET dyes positioned at similar distances yield detectable FRET signals17. We also modified the probe design so that each probe would consist of alternating groups (G) of 1, 2 or 4 D oligos followed by groups of 1, 2 or 4 A oligos (Fig. 1d-i). We reasoned that, since the oligo hybridization efficiency in DNA FISH most likely does not reach 100%, having consecutive groups of multiple D and A oligos would maximize the chances of multiple D-A pairs to be in sufficient spatial proximity to yield a FRET signal. We named the different designs G1-S50, G1-S150, G2-S50, G2-S300, G4-S50, and G4-S300 (Fig. 1d-i). As a proof-of-concept, we targeted the mouse Ogt gene locus located on chromosome (chr) X, by hybridizing female mouse embryonic fibroblasts (MEFs), reasoning that we might detect differences in chromatin compaction since this locus is known to escape female X chromosome inactivation, albeit at a low (~6%) frequency18. All the six probe designs yielded detectable FRET signals in dozens of single cells analysed, however the distributions of the FRET efficiency differed depending on the probe design (Fig. 1d-i). Design G1-S150 and G2-S50 showed a major FRET efficiency peak on the right, most likely corresponding to Ogt alleles within compacted chromatin, flanked by a leftward lower FRET efficiency tail presumably corresponding to Ogt alleles in a less compacted state (Fig. 1d, e). Of note, the FRET efficiency consistently peaked around 40% in all the probe designs tested, and this finding was recapitulated in a different cell line (NIH3T3 fibroblasts) (Supplementary Fig. 2a-f). We also examined the distribution of FRET acceptor intensities across all the probe designs and did not detect any major differences (Supplementary Fig. 2g-l), demonstrating that the ability of FRET-FISH probes to detect chromatin compaction differences at the targeted loci depends on the generation of FRET. Lastly, we tested a different pair of FRET dyes, AlexaFluor 488 (AF488) and AlexaFluor 594 (AF594), which are characterized by higher quantum yield and lower fluorescence signal degradation over time compared to Cy3 and Cy5. Using the same Ogt probe design, G1-S150, together with AF488 and AF594 dyes resulted in clearly detectable FRET with reduced crosstalk and bleed-through compared to the use of Cy3 and Cy5 (Fig. 1j and Supplementary Fig. 3). Moreover, the same probe design and dye combination yielded two clearly distinct FRET efficiency peaks indicative of different underlying chromatin compaction states (Fig. 1k). Therefore, we adopted the G1-S150 probe design and AF488 (FRET donor) and AF594 (FRET acceptor) dyes as standard FRET-FISH settings for all subsequent experiments.
Next, we sought to validate FRET-FISH by comparing chromatin compaction measurements at selected loci with chromatin accessibility measured at the same loci by ATAC-seq14. To this end, we designed FRET-FISH probes against six genes (Atp2b3, Ddx3x, Kdm5c, Magix, Pbdc1, and Tent5d) on mouse chrX, including three genes (Atp2b3, Magix, and Tent5d) that are constitutively inactivated on one chrX copy and three genes (Ddx3x, Kdm5c, and Pbdc1) that frequently escape inactivation18 (Supplementary Table 1). In three independent experiments, we measured FRET at each of the six targeted loci in female MEFs, which yielded reproducible FRET efficiency bimodal distributions, indicating that these loci most likely exist in two distinct chromatin compaction states (Fig. 2a-f and Supplementary Fig. 4a-r). Of note, the FRET efficiency distributions were unique to each gene and there was no obvious difference between constitutive active and escapee genes.
We then retrieved ATAC-seq read counts from MEFs, matching all the six gene loci probed by FRET-FISH (Supplementary Table 1 and Supplementary Methods). We compared the mean FRET efficiency measured at each locus with the corresponding ATAC-seq read counts and found an inverse correlation between the two metrics (Pearson’s correlation coefficient, PCC: –0.75), demonstrating that local chromatin accessibility and density are indeed associated (Fig. 2g). To further validate FRET-FISH, we assessed whether local chromatin compaction measured by FRET-FISH correlates with contact frequencies measured by Hi-C15 at the same loci. To this end, we retrieved publicly available Hi-C data from MEFs19 and correlated the total number of contacts within the genomic windows encompassing the six gene loci probed by FRET-FISH with the corresponding mean FRET efficiencies measured by FRET-FISH (Supplementary Table 2 and Supplementary Methods). The correlation between FRET-FISH and Hi-C was considerably stronger (PCC: –0.98) than between FRET-FISH and ATAC-seq, suggesting that Hi-C can be used to probe chromatin compaction genome wide (Fig. 2h). While FRET-FISH and ATAC-seq showed poor concordance for the Ddx3x gene locus, we instead observed a high agreement between FRET-FISH and Hi-C at this locus (Fig. 2h). Importantly, the correlation between FRET-FISH and Hi-C decreased when we only considered Hi-C contacts over short distances (< 5 kb) within the genomic windows encompassing the six genes probed by FRET-FISH (Fig. 2i). This suggests that FRET is not only generated by consecutive D and A oligos in our FRET-FISH probes, but also by D-A pairs separated by higher genomic distances that are in physical proximity in the nucleus. Of note, the correlation between ATAC-seq and Hi-C (PCC: 0.78) was, in absolute terms, lower than between FRET-FISH and either method (Supplementary Fig. 4s), suggesting that Hi-C might be better suited than ATAC-seq to probe chromatin density genome wide. Altogether, these results demonstrate that FRET-FISH is a valid method allowing reproducible measurements of local chromatin compaction at selected genomic loci, and that global and local chromatin accessibility and density are closely intertwined.
FRET-FISH detects chromatin compaction changes along the nuclear radius
To further validate FRET-FISH, we examined whether our method can also distinguish between different chromatin density states along the nuclear radius. In mammalian cells, the genome is radially organized and is characterized by distinct radial patterns of different chromatin types and genomic features20,21. In particular, highly condensed heterochromatin is enriched in proximity to the nuclear lamina and around nucleoli21. We therefore reasoned that FRET-FISH should yield different FRET efficiency distributions for the same locus, depending on where the locus is radially positioned in a cell. To test this, we assessed the FRET efficiency distribution in four arbitrary concentric nuclear layers drawn in 2D segmented nuclei (Supplementary Methods). At three loci examined (Apt2b3, Kmd5c and Magix), the FRET efficiency progressively decreased from the nuclear periphery inwards, in line with our expectation that a locus positioned at the nuclear periphery would have a more compact chromatin compared to the same locus positioned more centrally (Fig. 2j-l). These results are consistent with a radial gradient model of genome organization21 and further highlight the ability of FRET-FISH to detect local chromatin density differences with single-allele resolution.
FRET-FISH can detect local effects of global chromatin condensation changes
Next, we sought to test whether FRET-FISH could detect local changes in chromatin density associated with global changes in chromatin condensation. To this end, we treated female MEFs with a combination of sodium azide and 2-deoxy-d-glucose, which causes intracellular ATP depletion and, in turn, increases the intracellular pool of polyamines and divalent cations that neutralize the negative charges on DNA, leading to chromatin condensation22 (Supplementary Methods). The same treatment also results in drastic reduction of gene expression23. To confirm this effect in our experimental setup, we quantified by fluorescence microscopy nascent transcripts, in which we incorporated the fluorescent uridine analog 5-ethynyluridine (EU) and observed a drastic reduction in the total amount of nuclear fluorescence upon treatment of MEFs with sodium azide and 2-deoxy-d-glucose (Supplementary Fig. 5a, b). To test whether these drastic global changes in chromatin condensation are reflected at the level of individual genes, we performed FRET-FISH in female MEFs that were either treated with sodium azide and 2-deoxy-d-glucose or not, using four of the six probes targeting different genes (Atp2b3, Kdm5c, Magix and Pbdc1) on mouse chrX (Supplementary Table 1). In two independent experiments, ATP-depleted cells showed a significantly higher FRET efficiency at all the four gene loci examined (Fig. 3a-d and Supplementary Fig. 5c-f), demonstrating that FRET-FISH can reproducibly detect local changes in chromatin density that mirror global changes in chromatin condensation.
To further assess the ability of FRET-FISH to detect local changes in chromatin compaction resulting from global changes in chromatin condensation, we performed FRET-FISH in female MEFs cultured for varying passage numbers, which we hypothesized might result in increased global chromatin compaction based on previous studies24–26. Indeed, MEFs that underwent more than 10 passages showed a significantly higher (P = 1.8\(\times\)10−70, Wilcoxon test, two-tailed) global chromatin compaction—as determined based on the mean fluorescence intensity per nucleus of the DNA intercalator Hoechst 33342—compared to cells that underwent less passages (Supplementary Fig. 5g and Supplementary Methods). We then assessed whether FRET-FISH would detect a corresponding increase in local compaction at the level of individual gene loci. In two independent experiments, the height of the right peak of the FRET efficiency bimodal distributions measured at the Atp2b3, Kdm5c, and Magix loci (which most likely corresponds to a higher local compaction state) progressively increased upon prolonged cell passaging, mirroring the global increase in chromatin condensation (Fig. 3e-j and Supplementary Fig. 5h-m). Altogether, these results confirm that FRET-FISH can reproducibly detect local chromatin density changes, with single-allele resolution.
FRET-FISH detects chromatin compaction changes during the cell cycle
Lastly, we tested whether FRET-FISH could detect changes in local chromatin density in different phases of the cell division cycle, during which chromatin undergoes dramatic condensation changes reaching the highest compaction during mitosis27. To this end, we classified the same cells, in which we previously measured the compaction at the Atp2b3, Kdm5c, and Magix loci, as either G1 or non-G1 based on the fluorescence intensity of the DNA intercalator dye, Hoechst 33342 in the cell nucleus (Supplementary Fig. 6a and Supplementary Methods). In the case of Magix, we also managed to identify mitotic cells with detectable FRET signals. As expected, the FRET efficiency measured by FRET-FISH in these cells was significantly higher (P = 0.008, Wilcoxon test, two-tailed) compared to cells in other phases (Supplementary Fig. 6b). Notably, for all the three genes tested in two independent experiments, the FRET efficiency was substantially lower in non-G1 compared to G1 cells, which was reflected in higher left peaks in the bimodal FRET efficiency distributions in non-G1 cells (Fig. 3k-s and Supplementary Fig. 6c-h). These results are consistent with prior observations based on the ATAC-see method, which allows visualizing global DNA accessibility in single cells28. We then further separated G1 cells into HoechstHigh (top quantile) and HoechstLow (all remaining G1 cells) and found that the FRET efficiency measured at the same three loci was significantly higher in HoechstHigh cells, which most likely represent cells that just exited mitosis (Fig. 3k-m). These results further highlight the sensitivity of FRET-FISH and its ability to reproducibly detect local DNA density differences, at the level of individual alleles.