Investigating cell cycle-dependent gene expression in the context of nuclear architecture at single-allele resolution

Life at the cellular level is the result of the coordinated efforts of molecular machinery inside a cell brought about by timely expression of genes at mRNA and protein levels as posited by the central dogma of molecular biology (Crick, 1970). A cell achieves this by regulating genes at various stages of their expression – from the transcription of genes to the translation of their respective mRNAs and eventual degradation of the gene products (Day and Tuite, 1998; Wray et al., 2003; Zhao et al., 2017). Many of these effectors of gene regulation can be linked to two important factors: nuclear architecture and the cell cycle (Misteli, 2001; Sadasivam and DeCaprio, 2013). The cell cycle encompasses the complete gamut of processes by which a single cell divides into two daughter cells. This implies that almost all cellular events or attributes have to be coordinated with the cell cycle to ensure proper cell division – even the nuclear architecture (Cooper, 2000; Finn and Misteli, 2019; Probst et al., 2009). Nuclear architecture is the organization of the genome inside a cell nucleus with respect to nuclear compartments such as the nuclear lamina or heterochromatin. It can play a key role in determining the gene expression pattern in a cell and hence the cellular identity, as proposed by the topological model of gene regulation (TMGR) (Cremer and Cremer, 2001). The TMGR hypothesizes that: (1) spatial positioning of a gene with respect to chromosomal locations such as centromeres can affect the transcription of that gene, and (2) differentiating cells develop a pattern of gene positions with respect to nuclear compartments (such as the nuclear lamina, heterochromatin or interchromatin compartments) that defines patterns of gene expression and hence the eventual identities of those cells (Cremer and Cremer, 2001). There have been a number of studies supporting the TMGR but most of them lacked either single-cell resolution or the ability to interrogate gene position and expression at the same time in the same cells (Ernst et al., 2011; Fullwood et al., 2009; Morey et al., 2007, 2009). Single-cell-level studies on nuclear architecture-dependent gene expression have been elusive because they require the combination of immunofluorescence-based detection of proteins with fluorescent in situ hybridization (FISH)-based detection of gene positions and mRNAs while also preserving the 3D nuclear architecture of the cell. The existing methods for RNA FISH and 3D DNA FISH are incompatible because the steps involved in either can affect the detection of the other adversely.

Here we present a simple and reliable way to combine DNA FISH with single molecule RNA FISH (smFISH) and immunofluorescence while also preserving the 3D nuclear architecture of a cell. We further link it with a microscopy-based cell cycle staging method developed previously (Dhuppar and Mazumder, 2018) to study cell cycle-dependent changes in nuclear architecture and gene expression on a cell-by-cell basis. In particular, we investigate how the positioning of the cyclin-A2 (CCNA2) gene with respect to the nuclear lamina correlates with the known cell cycle-dependent expression of the gene.

The ease of the technique presented here, combined with the depth of the analyses, makes this method a useful tool in addressing various questions pertaining to the cell cycle, nuclear architecture and gene expression at a single-cell resolution.

Genome organization has previously been shown to be involved in the regulation of gene expression. Most of these studies either rely on bulk biochemical assays for measuring expression (Thomson et al., 2004) or involve sequential FISH-based labeling and imaging of mRNA and the gene of interest – this relies heavily on the ability to mark and identify the same cells between the two steps (Clemson et al., 2009; Lomvardas et al., 2006; Morey et al., 2009). Recent years have seen the development of smFISH methods that yield absolute transcript counts on a cell-by-cell basis, unlike standard RNA FISH where only relative intensities are measured (Raj et al., 2006, 2008; Zenklusen et al., 2008). Although there have been previous studies investigating mRNA localization inside the cell (Buxbaum et al., 2015) or transcription in the context of nuclear architecture that combined DNA and RNA FISH (Clemson et al., 2009; Hall et al., 2002, 2014; Jiang et al., 2013; Lomvardas et al., 2006; Morey et al., 2009; Pageau et al., 2007), fewer attempts have been made to combine 3D DNA FISH with smFISH. The difficulty in combining DNA FISH with RNA FISH or immunofluorescence stems from the fact that DNA FISH involves harsh treatments, such as acid- or formamide-based DNA denaturation, which can adversely interfere with the subsequent detection of proteins or RNAs inside a cell (Morey et al., 2007; Zhang et al., 2007). Furthermore, the ethanol dehydration involved in such procedures destroys the 3D architecture of the cells – which are more than 50% water – by flattening them out. The organelle most affected by such dehydration would be the cell nucleus, whose water content can be as high as 85% (Century et al., 1970). This makes it difficult to combine the three assays while also minimizing distortions to nuclear architecture. Additionally, the existing protocols for 3D DNA FISH involve steps such as nitrogen freeze-thaw cycles, which add another layer of difficulty to the problem (Bienko et al., 2013; Bolland et al., 2013).

We started out by first standardizing a protocol for 3D DNA FISH devoid of conventional difficulties. We realized that the major and perhaps the only factor affecting the 3D nuclear architecture of a cell in a conventional 2D DNA FISH protocol is ethanol dehydration. We discovered that the loss in efficiency of probe hybridization due to omission of ethanol dehydration can be compensated for by having longer hybridization times (>40 h). The resulting protocol, in fact, is largely similar to a conventional 3D DNA FISH protocol if nitrogen freeze-thaw cycles are removed.

Once the protocol for 3D DNA FISH was standardized, the next step was to determine the order in which the three techniques can be performed for least interference among them. We observed that the order that gives the best result is the following: immunofluorescence followed by refixation with 4% paraformaldehyde, followed by DNA FISH and then smFISH for RNA. The order is especially important with regard to smFISH, where 20-nucleotide long oligomers are used as the hybridization probes. If not performed in the correct order, the high formamide concentration in the DNA FISH hybridization buffers can destroy all smFISH RNA signals.

Once standardized, we used the above technique to study how gene position inside the nucleus can be correlated with gene expression across the cell cycle in HeLa cells. We selected the CCNA2 gene for our study due to its known cell cycle-dependent gene expression. We found that most cells had three copies of the CCNA2 gene, which was consistent with the fact that HeLa cells are hypertriploid (Fig. 1A). We also observed that mRNA and protein expression levels of CCNA2 were strongly correlated (Fig. 1A) and that the activity of two adjacent alleles, as inferred from the colocalization of smFISH and DNA FISH signals, could be very different – at least in 2D-projected images of nuclei (Fig. 1B).

To convert the above observations into quantifiable, comparable metrics we performed a quantitative 3D image analysis, as described below.

Cells were first analyzed to quantify their DNA content in order to determine the cell cycle stage for each cell using a previously-developed module for imaging-based cell cycle determination (Dhuppar and Mazumder, 2018). Briefly, the workflow ran thus: the average-projected images from the DAPI channel were passed to a module that quantified the DNA content of individual nuclei in a field. Once DNA content was quantified, the 3D stack for each nucleus in a field was sent to another module that eliminated out-of-focus planes from the 3D stacks and segmented the nucleus in 3D, based on texture analysis (Fig. 2A). All 3D analyses were done in a cell-by-cell manner in order to reduce both the size of the images and the subsequent computer processing requirements. DNA FISH spots were also segmented in 3D, using another previously reported module (Dhuppar and Mazumder, 2018). Fig. 2B shows an example of one such 3D segmented nucleus and the segmented CCNA2 alleles within (see also Movie 1). The nuclear volumes obtained from the above analysis compared well with those reported previously for HeLa cells (Fig. 2C) (Monier et al., 2000).

The 3D segmented nuclei and alleles thus obtained were then used to calculate distances from the centroid of an allele to the nuclear lamina and nuclear centroid in order to define a closeness metric, as discussed below.

The TMGR suggests that, upon differentiation, cells adopt the expression patterns of their differentiated states by defining a unique and reproducible pattern of gene positions with respect to different nuclear compartments such as the nuclear lamina, heterochromatin or interchromatin compartments (Cremer and Cremer, 2001). However, other studies have suggested that genes can make long extrusions out of their chromosome territories for their transcription (Morey et al., 2007; Osborne et al., 2004). This looping-out of genes for expression could be to avoid nuclear compartments such as the nuclear lamina or heterochromatin, which are generally associated with gene silencing (Finlan et al., 2008; Reddy et al., 2008; Wang et al., 2018). To determine which of the two models – the static TMGR model or the dynamic extrusion-based model – is true in the case of the CCNA2 gene, we calculated the distances from the centroids of the three copies of CCNA2 to the nuclear lamina and nuclear centroid, and then correlated them with expression of the CCNA2 gene.

We observed that, although expression of CCNA2 mRNA and CCNA2 protein peaked in the G2/M phases of the cell cycle (Fig. 3A,B), the mean of the absolute distances between the centroids of the gene copies and the nuclear lamina or the nuclear centroid did not change substantially across the cell cycle (Fig. 3C,D). Furthermore, we defined a closeness metric for the gene copies that could vary continuously between 0 and 1 – taking a value of 0 if the gene copy is at the nuclear centroid and 1 if it is at the lamina. We ranked the gene copies such that copy 1 corresponded to the gene copy with highest closeness value. We observed that the mean value of closeness in a cell did not correlate with the expression of CCNA2 in terms of mRNA number or protein level (Fig. 3E). In fact, the mean closeness for the CCNA2 gene did not vary much across the cell cycle, even though the expression was highly cell cycle-dependent (Fig. 3E).

The above observations strongly support the TMGR mode of gene regulation in the case of the CCNA2 gene in the differentiated HeLa cancer cell line. However, the analysis is incomplete without information on the activity of the gene at a single-allele level. The following section discusses the position-dependent activity of CCNA2 gene at a single-allele level.

We measured the activity of a gene copy by determining the colocalization between smFISH and DNA FISH signals at a position in the nucleus in 3D. A cuboid just enveloping the gene spot was constructed and then was searched for the mRNA signal. If the cuboid contained at least one mRNA spot, then that gene copy was deemed as actively expressing. The expression metric thus defined was found to have high precision when verified against synthetic 3D images of similar spots (Fig. S1).

We observed that there was no obvious correlation between the position of a CCNA2 allele and its expression across the cell cycle. However, there was a marginal increase in the percentage activity of the CCNA2 copy farthest from nuclear periphery (Fig. S1C). As shown in Fig. 4, the expression of a gene copy depended little on its position relative to the lamina, or on the cell cycle stage. Gene copies could express in any cell cycle stage irrespective of subnuclear position – though overall mRNA levels are cell cycle dependent. We also investigated the number of CCNA2 copies actively expressing in a cell and the corresponding total mRNA count. We found that the cells with higher mRNA counts, on average, had more gene copies with active transcription (Fig. S2). Interestingly, we found that cells with very low mRNA counts can have active expression from multiple gene copies, and cells with active expression from a single gene copy can have high mRNA counts. This illustrates the stochastic nature of gene activation via transcriptional bursts (Fig. S2) (Raj et al., 2006).

We also observed that two of the three alleles always remained close to the nuclear lamina (Fig. 4). This is striking, given that individual bona fide lamina-associated domains (LADs) can be randomized between two cell cycles (Nagano et al., 2017) even if the overall nuclear organization of chromosome territories can remain the same (Gerlich et al., 2003).

The above observations support the TMGR mode of action for CCNA2 gene regulation, where the average position of the gene relative to the nuclear lamina varies little within the cell cycle even though expression of the gene – both at mRNA and protein levels – is highly cell cycle-dependent.

Nuclear architecture derives from the non-random organization of the eukaryotic genome in terms of its arrangement of chromosome territories and genes with respect to each other and to other nuclear landmarks, such as the nuclear lamina or nucleoli (Misteli, 2005). Nuclear architecture is involved in cell fate decisions during differentiation (Morey et al., 2007; Stadhouders et al., 2018, 2019) and has been shown to be tissue-specific (Parada et al., 2004). It is also implicated in the regulation of gene expression via association with different nuclear compartments such as the nuclear periphery, heterochromatin, euchromatin and interchromatin compartments (Holwerda and Laat, 2012; Sexton et al., 2007). Most of these studies are based on bulk biochemical assays that efficiently capture population-level changes in organization or expression but miss out on the cell-to-cell variability within a population. In fact, according to recent reports, there is a substantial level of heterogeneity in genome organization within a population, and at any given time only a small fraction of cells within a population harbor the interactions captured at the population level (Finn and Misteli, 2019; Finn et al., 2019). This underscores the need for single-cell studies to capture nuclear architecture-dependent regulation of DNA processes in their entirety.

The existing methods to study 3D nuclear architecture and gene expression at single-cell resolution are incompatible with one another. Here we report a simple and efficient method to combine DNA FISH with smFISH and immunofluorescence that also preserves the 3D nuclear architecture of a cell. We further combined this with an imaging-based cell cycle staging process developed previously (Dhuppar and Mazumder, 2018) to study gene regulation in the context of nuclear architecture across the cell cycle for a cell cycle-regulated gene, CCNA2. DNA FISH methods that use extensive dehydration can cause flattening of the cell nucleus and loss of three-dimensional information. This is addressed by newer methods of 3D DNA FISH. Compared to other 3D DNA FISH techniques, our method is simpler and allows for multiplexing with smFISH for RNA and immunofluorescence for proteins. Although undoubtedly each of these methods have been performed many times in isolation by many groups, combining all three in the same cell provides direct insight into nuclear architecture-dependent gene expression.

We observed that the cell cycle dependence of CCNA2 gene expression is hardly reflected in the average positioning of the CCNA2 gene relative to the nuclear periphery (Fig. 3E). Interestingly, in contrast to previous studies associating positioning at the nuclear periphery with gene silencing (Finlan et al., 2008; Reddy et al., 2008; Wang et al., 2018), we observed that nuclear positioning of a CCNA2 gene copy – which more often than not stays close to the nuclear periphery – matters little for its activity (Fig. 4; Fig. S1). Also, the number of CCNA2 copies actively expressing in a cell does not reflect the overall mRNA count in that cell. There can be cells with very high mRNA count but fewer than two copies actively expressing, and vice versa. This demonstrates the stochastic nature of transcriptional bursts and gene regulation via the modulation of burst frequency (Li et al., 2018; Raj et al., 2006). This might also suggest a possible decoupling between the activities of different alleles within a nucleus where each is agnostic of the others' transcriptional status, unlike as reported recently in bacteria (Wang et al., 2019). The above observations further emphasize the importance of studying the heterogeneity in genome organization within a population at a single-cell resolution.

Finally, our method can potentially be combined with DNA FISH for chromosome territories. Thus, the experimental and analysis modules used in this study can easily be adapted to study the regulation of gene expression mediated by megabase-pair looping seen in differentiating stem cells (Chambeyron and Bickmore, 2004; Morey et al., 2009) or by short excursions of genes in differentiated cells to shared sites of transcription, also called ‘transcription factories’ (Osborne et al., 2004, 2007), on a cell-by-cell basis. Taken together, the technique presented here, along with the analysis modules, is a powerful tool to study nuclear architecture and nuclear architecture-dependent gene regulation across the cell cycle at single-cell resolution.

HeLa cells were grown on ethanol-sterilized, poly-D-lysine-treated coverslips. The cell line was procured from the National Centre for Cell Science (NCCS), the national cell line repository of India. DMEM/F12 (Gibco) supplemented with 10% FBS (Gibco) without any antibiotic was used for cell culture. Cells were tested negative for bacterial contamination and mycoplasma. They were passaged every ∼4 d and were allowed to grow for at least 24 h before starting an experiment. Experiments were performed in conformity to the norms set out by the Institutional Biosafety Committee of TIFR Hyderabad.

Cells were fixed with 4% paraformaldehyde (PFA) in nuclease-free (NF) PBS for 15 min at room temperature, followed by permeabilization with 0.3% Triton X-100 in NF PBS for 10 min at room temperature. The cells were then washed twice with PBS and were prepared for immunofluorescence. The order of assays was always as follows: immunofluorescence followed by re-fixation for 15 min with 4% PFA in NF PBS at room temperature, followed by DNA FISH, followed by RNA FISH. The re-fixation post immunofluorescence is important to preserve the signals.

For immunofluorescence, cells were first blocked for non-specific binding using 1% NF BSA (Ambion) for 30 min at room temperature. Then the cells were incubated with the primary antibodies in 1% NF BSA for 60 min at room temperature. After two washes with NF PBS, the cells were incubated with the secondary antibodies in 1% NF BSA for 60 min at room temperature, followed by two washes with NF PBS.

For DNA FISH, cells were first washed with 2×NF SSC (Ambion, Thermo Fisher Scientific). After which the DNA was denatured with 70% formamide in 2×NF SSC at 80°C for 15 min. While DNA in the cells was being denatured, DNA hybridization mixture was prepared: 9 µl hybridization buffer (Empire Genomics)+1 µl probes. The mixture was heated at 80°C for 3 min to denature the probes. A slide was wiped clean with RNaseZap (Ambion, Thermo Fisher Scientific) and left to dry. After DNA denaturation, the DNA FISH hybridization mix was transferred to the slide, and the coverslip was placed over it carefully. The cells were then incubated for hybridization for more than 40 h at 37°C in a dark, heavily-humidified chamber. After hybridization the cells were washed once with 2×SSC at room temperature followed by two 10-min washes with 50% formamide in 2×NF SSC at 37°C.

For smFISH, cells were first rinsed with 2×NF SSC followed by two brief washes with 10% formamide in 2×NF SSC at room temperature. smFISH hybridization mix was prepared as follows: 9 µl hybridization buffer (Stellaris)+1 µl formamide+0.5 µl RNA FISH probes. The slides were wiped clean with RNaseZap and were allowed to dry completely. The hybridization mix was then transferred onto the slide and the coverslip was carefully placed over it. The cells were then incubated overnight at 37°C in a dark, heavily-humidified chamber. The next day, cells were washed once with 2×SSC at room temperature followed by two 5-min washes with 10% formamide in 2×NF SSC at 37°C. Finally, the cells were DNA stained with 1 µg/ml Hoechst 33342 dye for 10 min.

CCNA2 protein was labeled with a rabbit antibody (1:500) from Abcam (ab181591). For detection, goat anti-rabbit IgG antibody (1:500) with Alexa Fluor 647 dye was used. Pre-labeled CCNA2 DNA FISH probes were procured from Empire Genomics (CCNA2-20-Re). The fluorophore was spectrally similar to Alexa Fluor 594. The hybridization buffer was supplied along with the probes.

For RNA smFISH the probes were designed using Stellaris Probe Designer by Biosearch technologies, and probes and hybridization buffer were ordered from the same source. The probes were labeled with Quasar 570 fluorophores. The probe sequences are available in Table S1 and have been described previously (Dhuppar and Mazumder, 2018).

Ambion ultrapure 50 mg/ml BSA (AM2616) was procured from Invitrogen. All other reagents (including PBS, SSC, water and RNaseZap) were Ambion RNase-free products from Invitrogen, Thermo Fisher Scientific.

Imaging was always performed with Vectashield as a mounting medium. Images were acquired at 14-bit resolution using a fully-automated Olympus IX83 microscope on a Retiga 6000 camera (QImaging). All images were take using a 60× oil objective with 1.42 NA. 31 planes with 300 nm step size were taken for every field. Appropriate filter cubes from Chroma Technologies (49309 ET – Orange#2 FISH and 49310 ET – Red#2 FISH) and Olympus were used to prevent any bleed through across the spectrum of fluorophores.

The 3D stacks were first average projected and analyzed for DNA content using an automated MATLAB (MathWorks) routine developed previously (Dhuppar and Mazumder, 2018). Thereafter, the 3D stack corresponding to each nucleus in the field was sent one-by-one to another module for texture-based removal of the out-of-focus light followed by 3D segmentation of the nucleus (Fig. 2). After segmentation, the boundary voxels and the centroid of each cell nucleus were used to find relevant distances. A closeness metric was defined for relative distances of the alleles from the lamina and the centroid thus: ratio of the distance of an allele from the centroid to the sum of the distances from the lamina and the centroid (Fig. 3). Its value lies always between 0 and 1: unity if the allele is on the lamina and zero if it is at the centroid. The 3D segmentation was done on a cell-by-cell basis to reduce the size of the images and hence make the computation feasible with a desktop computer.

For position-dependent analysis of gene expression, only nuclei with exactly three clearly identified copies of the CCNA2 gene were used (446 cells) – they formed ∼73% of a total population of 611 cells.

The physical size of a pixel on the camera was 4.54 µm, corresponding to an area of 20.6 µm². Using a 60×1.42 NA objective without binning, this translates to 5.73×10⁻³ µm² on the sample. With stack separation of 300 nm, the voxel volume comes to 1.72×10⁻³ µm³. Distances in 3D calculated found similarly. Images for the purpose of visualization were prepared using ImageJ (National Institutes of Health).

All image analyses were performed using MATLAB (MathWorks). For graphs, MATLAB and Python 3 were used. All code and programs used in this study are available on https://github.com/shuppar.

S.D. thanks Dr Sitara Roy for many useful suggestions on data representation and visualization.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.246330.reviewer-comments.pdf