Search
Close this search box.

ARTseq-FISH reveals position-dependent differences in gene expression of micropatterned mESCs – Nature Communications

Simultaneous visualisation of mRNAs and (phosphor)proteins

The ARTseq-FISH workflow comprises a series of consecutive steps (Fig. 1a, d, Supplementary Note 1 and Supplementary Figs. 116). In short, fixed cells are stained with DNA-barcoded antibodies against (phospho-)protein targets of choice (Supplementary Data 1, 2). The DNA barcode of each antibody, known not to influence antibody binding efficiency38,39, contains a 10 nt polyadenylation (poly-A) linker and a specific 36 nt sequence that is complementary to a corresponding target-specific single-stranded DNA (ssDNA) padlock probes (PLPs). On the other hand, cellular mRNAs are directly targeted by one specific ssDNA PLP with a target sequence of 36 nt (Supplementary Fig. 1a, b), resulting in DNA/RNA hybridisation. Therefore, the circularising of the PLPs that hybridise to either proteins or RNAs require T4 DNA ligase and SplintR ligase, respectively40,41,42 (Fig. 1a and Supplementary Figs. 46). Next, rolling circle amplification (RCA) generates rolling circle products (RCPs) that amplify the original (phospho-)protein or mRNA target (Fig. 1a and Supplementary Fig. 8). To decode the original target amplified by the RCPs, we first hybridise target-specific bridge probes (BrPs), which are each recognised by unique combinations of fluorescently labelled readout probes (RoPs) (Fig. 1a, Supplementary Figs. 1c, d,  9 and Supplementary Data 1), allowing the detection of individual spots for each mRNA and protein (Fig. 1b, c and Supplementary Figs. 17, 18). ARTseq-FISH provides similar relative protein abundance compared to classic immunofluorescence (Supplementary Fig. 19). We determined a detection efficiency of ~47% (for Nanog mRNA) by comparing mRNA quantification using ARTseq-FISH with conventional single molecule RNA FISH (smFISH)27 (Supplementary Fig. 20 and Supplementary Data 3). Furthermore, ARTseq-FISH enhances the signal ~4-fold with respect to conventional smFISH (Supplementary Fig. 20c, d), resulting in a higher signal-to-noise ratio. On average, ~0.03–0.5 spots per cell were detected as false positives from unspecific PLP (protein detection only), BrP, and RoP hybridisation (Supplementary Fig. 3), in particular in the presence of true hybridisation signal (Supplementary Fig. 18b). Similar to existing methods, we are unable to quantify the degree of false positives coming from unspecific binding of the antibody or PLP, and the target detection efficiency is highly antibody33,34 and PLP-dependent40,41,42,43 (Supplementary Fig. 10). Lastly, we specifically employ one PLP per mRNA target, since we are interested in relative mRNA differences. However, should more rigorous quantification of mRNA be necessary, at least three PLPs per target are likely required40,41,42,43.

Fig. 1: Experimental outline of ARTseq-FISH.
figure 1

a The experimental workflow of ARTseq-FISH enables the detection of different molecular species (mRNA, proteins and phosphoproteins) simultaneously. Each antibody conjugated to oligonucleotides, targets its associated (phospho-)protein in situ. Target-specific PLPs bind to these oligonucleotides as well as cellular mRNA, respectively. Then the PLPs are circularised and amplified into RCPs. A set of target-specific BrPs hybridise to RCPs, and then RoPs labelled with fluorophores bind to one of the four flanking sequences on the BrP. b Representative raw data of simultaneously detecting Beta-catenin mRNA, protein, and phosphoprotein with the same resolution. Scale bar, 20 µm, 1 µm. c The relative mean pixel intensity of the hybridisation signal’s point spread functions (n = 2000) around the local maxima. Comparison of the average observed signal of ARTseq-FISH to smFISH and a 200 nm fluorescent bead. d The sequential stripping and rehybridisation of new known RoPs allows subsequent decoding of stacked signals. Source data are provided as Source Data sheet tab Fig. 1c.

Automated quantification of mRNAs and proteins in cells

Multiplexed, simultaneous detection and quantification of mRNAs and (phospho-)proteins over a large area is achieved through serial stripping and rehybridisation of RoPs following a colour barcoding scheme27 (Fig. 1d, Methods, Supplementary Notes 1, 2, Supplementary Figs. 1416 and Supplementary Data 4). The barcode of each target is determined by the presence of a given colour (green, red, or far red) in a sequence of specific hybridisation rounds. To correctly decode the detected signal and translate this into mRNA or protein counts for individual cells, we developed an automated image analysis pipeline that returns the abundance of each target per cell.

The pipeline (Fig. 2a, more detailed description in Supplementary Notes 2, SN2 Figs. 13, software packages used44,45,46,47,48,49) initiates the analysis by identifying the local maxima per colour channel50 by denoising and sharpening the image in six steps (Supplementary Notes 2SN2 Figs. 34) and classifying the local maxima as a true hybridisation signal. This is done by using a support vector machine (SVM) model trained on data of true point spread functions, which allows for signal deconvolution (Supplementary Notes 2SN2 Figs. 56). Single-cell analysis requires the reconstruction of single cells in 3D from a collection of images (Fig. 2a and Supplementary Notes 2SN2 Figs. 715). However, consistent segmentation is challenging because a segmentation algorithm that maps onto one nucleus might not map onto others (Supplementary Notes 2SN2 Fig. 8), or a global image filter that cleanly separates the nuclei in one image might not be optimal for another (Supplementary Notes 2SN2 Fig. 10). Therefore, an iterative approach was implemented, repeatedly segmenting nuclei using different global image filters and segmentation parameters51 (Supplementary Notes 2SN2 Figs. 9 and 13). We collected the segmentation labels and used an SVM model to filter out any failed segmentations. The remaining labels were subsequently used to reconstruct the nuclei in 3D by clustering their centroids with a node-based graph (Supplementary Notes 2SN2 Figs. 14, 15). After interpolating any missing nucleus segmentations across the z-axis, we obtained a 3D bounding box. Another SVM model subsequently assessed if this bounding box contained a single nucleus. Before assigning mRNA and protein counts to a cell, we adjusted for any drift, by using the segmented nuclei as unique markers to align the images. Once a global shift was found, we refined the alignment by moving all hybridisation signals across a narrow range of shifted pixels until the signals overlapped (Supplementary Notes 2SN2 Figs. 16, 17). Ultimately, we decoded the overlapping signals and assigned each hybridisation signal to a particular protein or mRNA in a specific cell (Fig. 2b and Supplementary Fig. 2, Supplementary Notes 2SN2 Figs. 18, 19). Cytoplasmic targets were assigned to a specific cell based on the shortest distance to the nucleus (Supplementary Notes 2SN2 Fig. 23).

Fig. 2: Computational outline of ARTseq-FISH.
figure 2

a A condensed summary of the automated computational pipeline that converts the raw microscope images into a single-cell dataset, providing the abundance of each target per cell. First, the hybridisation signal is detected, followed by an iterative AI-based segmentation approach, the segments are clustered, and the cells are reconstructed in 3D. Next, an image drift correction is applied to align the hybridisation between rounds. In the final step, these hybridisation signals are decoded and assigned to the targets as well as individual cells. b Reconstructed images show all the detected molecules in the cells cultured on micropattern 48 h after leukaemia inhibitory factor (LIF) withdrawal. The spots are subdivided into three classes: mRNA (yellow), protein (blue) and phosphorylated protein (brown). Scale bar, 20 µm.

Expression profiles of mRNAs and (phospho)proteins in mESCs

We utilised ARTseq-FISH to explore how the expression of mRNAs and (phospho-)proteins is affected during the early stages of differentiation on micropatterns. LIF/serum cultured mESCs are heterogenous (Supplementary Fig. 22), due to their pluripotent nature, enabling them to make different cell fate decisions upon LIF withdrawal52,53,54. Therefore, we plated mESCs directly on a 750-µm-diameter micropattern and withdrew LIF from the serum medium for up to 48 h to allow for cells to exit pluripotency. Notably, cells cultured under this condition are not the same as the cells primed to epiblast-like cells (EpiLCs)12. After fixation, we performed ARTseq-FISH on the micropatterned mESCs. Using our analysis pipeline, we visualised the locations of unique mRNA and (phospho-)protein targets (Fig. 3a) and quantified the abundance of 67 targets in individual cells (Fig. 3a–d). When discussing the protein targets detected using the respective antibodies (Supplementary Data 2), it is important to note that these measurements represent the total amount of protein detected, which includes both the unmodified and modified forms. Conversely, phosphorylated proteins were detected with specific antibodies (*protein or phos-protein in figures).

Fig. 3: ARTseq-FISH profiles the distribution of mRNAs and proteins in individual mESCs on micropatterns.
figure 3

a Reconstructed images show the detected spots in Fig. 2b assigned to individual cells and the identity of these targets at a specific location. Scale bar, 20 µm, 5 µm and 0.5 µm. bd Quantification of proteins (b), mRNAs (c) and phosphorylated proteins in single cells (d). Forty-eight cells were analysed per target. e Quantification of spots per cell for negative controls in three different channels (the three fluorophores are AT: ATTO 488, Tm: TAMRA, and CY5). Negative controls were performed without PLPs and 579 cells were analysed. When adding true signal to negative control images in silico for the ATTO 488 channel, the number of detected spots decreased >6-fold (shown; merged image with the smallest fold change reduction in false positives) for the green channel (bottom). f The scatter plot shows the average correlation between the total count of individual targets detected within the image for four biological replicates at 48 h differentiation of mESCs. g The correlation of RNA targets per cell between ARTseq-FISH and scRNAseq performed on serum/LIF cultured mESCs. scRNAseq data were analysed by Seurat88. 455, 503, 470, 482, 454, 466, 488, 408, 408, 408, 180, 241, 200, 200 and 233 non-micropatterned cells were analysed for Smad2, Tcf3, E-cadherin, Cdk4, CyclinD, Smad1, Klf4, Oct4, Nanog, Sox2, p53, Beta-catenin, Stat3, CyclinE, Yap respectively. Source data are provided as Source Data sheet tabs Fig. 3a–g.

Among protein targets, lineage-associated markers i.e. N-Cadherin, FOXA2, GATA6, showed higher abundance than pluripotency-related markers, i.e. NANOG, OCT4 and SOX2. Although we cannot rule out that these differences are due to dissimilarities in antibody efficiency, this is expected in mESCs that are exiting pluripotency (Fig. 3b)10,12,15. Notably, while RB shows low spot counts at a protein level, its mRNA and Phos-RB display higher spot counts (Fig. 3b–d), indicating that the difference in spot count between total RB and Phos-RB is caused by differences in antibody specificity. As mentioned previously, similar to other omics methods that rely on antibodies33,34, ARTseq-FISH is dependent on the binding efficiency of primary antibodies (for proteins) and PLPs (for mRNAs). Therefore, we cannot compare absolute counts between species. However, strong correlation of target counts between four different biological replicates (Fig. 3f) demonstrates the high reproducibility of ARTseq-FISH (Supplementary Note 2 and Supplementary Fig. 21).

In addition, after filtering out the low abundant RNA targets in single-cell RNA sequencing (scRNAseq) data, ARTseq-FISH and scRNAseq mRNA quantification showed a high correlation (r2 = 0.73) (Fig. 3g). Finally, to ensure true signals were detected with the image analysis software since no fixed intensity threshold is implemented, we tested the software on three samples: (i) samples that did not include any readout probes, i.e. negative control (Fig. 3e, top panel); (ii) samples that included readout probes, i.e. true signal; and (iii) an in silico merged image which combined samples i and ii. The ATTO 488 channel was the noisiest channel. For the merged images with ATTO 488 the number of false positives decreased at minimum sixfold (Fig. 3e, bottom panel). This demonstrates that in the presence of true signal, the false positive rate for the noisiest channel is <1%. Together these data indicate that ARTseq-FISH is a reproducible technique that should be implemented when comparing relative spatial or temporal changes in expression levels of mRNA and proteins.

Expression gradients across micropatterned mESCs

ARTseq-FISH provides single-cell quantification (Figs. 13) of each molecular species while retaining spatial information. As confirmed by immunofluorescence, ARTseq-FISH data show three patterns of molecular distribution across mESCs on micropatterns: (i) increased expression in cells located at the centre (i.e. p53); (ii) evenly distributed throughout (i.e., phosphorylated RNA Pol II); and (iii) increased expressed in cells at the periphery (i.e. BRACHYURY) (Fig. 4a–f and Supplementary Fig. 23). Consistently, targets are classified into these three groups (Fig. 4e and Supplementary Fig. 24, purple, white, and orange respectively). Overall, lineage, pluripotency as well as cell cycle associated proteins are more highly expressed in cells localised towards the edge of the micropattern. Specifically, BRACHYURY and GATA6 are mostly enriched in cells at the edge of the micropattern, as are the NANOG and SOX2 (Fig. 4e and Supplementary Fig. 25a). Yet, BRACHYURY and GATA6 are not necessarily highly expressed in the same cells as NANOG and SOX2 (Supplementary Fig. 25e). Surprisingly, some mRNAs (i.e. Stat3 and Cdk2) appear localised to the centre although their (phospho-)protein counterparts are localised to the edge. This could either be because cells at the edge are translationally more active and thereby require less mRNA to produce a higher amount of protein. Alternatively, cells could be in the process of up- or downregulating their mRNA, and differences in mRNA and protein half-life could lead to a lack of correlation between the two species55. Lastly, when calculating the mean fold change in expression level in cells located in the centre versus edge, cells >125 μm from the edge were defined as being in the centre of the micropattern (Supplementary Fig. 24), which is possibly an oversimplification. For example, Smad2 and Rb mRNA and protein are enriched in cells at the edge, while our analysis suggests that the phosphorylated proteins are more homogeneously expressed across the micropattern (Fig. 4e). However, when quantifying the average number of spots per cell at seven positions across the micropattern, a reduction in Phos-RB and Phos-SMAD2 protein at the centre (~375 μm from the edge) of the micropattern becomes evident (Supplementary Fig. 25b–d). Consequently, the localisation analysis likely masks more subtle changes in gene expression between cells at the edge and centre of the micropattern.

Fig. 4: ARTseq-FISH reveals gradients of mRNA and protein expression levels within micropattern-differentiated mESCs.
figure 4

a Schematic illustration of three representative spatial distributions of molecules on micropatterns. Created with BioRender.com. b Immunofluorescence results of p53, Phos-Pol II (Phosphorylated RNA Pol II), and BRACHYURY within a micropattern-differentiated mESCs. Scale bar, 100 µm. c Average number of spots per cell at different positions on the micropattern (edge to centre) for p53, Phos-Pol II and BRACHYURY proteins from three biological replicates (150, 125 and 158 cells and 1 micropattern were per replicate). d Pearson correlation between relative mean fluorescence in (b) and relative spots per cell with respect to their position on the micropattern (c). e Mean fold change of the expression level of individual targets between the edge and centre (purple and orange respectively) of the micropattern across four biological replicates (125–158 cells and 1 micropattern analysed per replicate). *: phosphorylated protein. Error bars represent the SEM of four biological replicates. f UMAP analysis of expression levels of 57 markers of individual cells after 48 h of LIF withdrawal, with the colour representing if cells are localised towards the edge (orange) or centre (purple) of the micropattern (left). The same UMAP with single cell abundance of p53, Phos-Pol II and BRACHYURY proteins indicated in blue. 158 cells and 1 micropattern were analysed. g The spots per cell (y-axis) with respect to the position of the cell on the micropattern (x-axis) of selected targets. The edge of the micropattern serves as the reference value for the relative position of the cells. Cells are found at the edge at 0 µm and near the centre at 375 µm. 158 cells were analysed per target. Source data are provided as Source Data sheet tabs Fig. 4c–g.

To more closely examine the relationship between target expression levels and cellular location, we quantified the levels of some targets in individual cells and plotted these against cellular position on the micropattern. This analysis revealed that targets that are more highly expressed in cells at the edge (i.e. BRACHYURY), display more of an exponential trend (Fig. 4g), rather than linear. Conversely, some targets that are classified as localised towards the centre (i.e. Phos-RB) or the edge (i.e. Smad1 mRNA) appear to be most abundant in cells located ~150 μm from the edge, rather than 375 μm from the edge, the latter being the true centre. Interestingly, the single-cell abundance of individual targets is highly heterogeneous (Fig. 4g). While this could be a technical consequence of ARTseq-FISH, the heterogeneity of p53, Phos-PolII, and BRACHYURY is also visible with regular immunofluorescence (Fig. 4b), and mRNA levels have previously been reported to become more heterogeneous as cells exit pluripotency56. Furthermore, we have shown that the mean gene expression values are reproducible across replicates when quantified by ARTseq-FISH (Fig. 3f). Therefore, this heterogeneity is likely biological and indicates that although cellular location on the micropattern does generate trends in gene expression profiles, both mRNA and protein synthesis is still heterogenous after 48 h of LIF withdrawal.

Position-dependent expression profiles emerge over time

To investigate how mRNA and protein abundance changes over time, we examined the spatial distribution of molecules in micropatterned cells at different time points (0, 12, 24 and 48 h) following LIF withdrawal (Fig. 5a). To better understand the changes, we grouped targets into different categories (Supplementary Data 5) and quantified the relative changes in target abundance across the micropattern in 2D (Fig. 5b) over the 48-h LIF withdrawal period. At the initial time point (0 h), both RNA and protein levels across the different categories showed a relatively homogeneous distribution. Interestingly, we noticed a specific region, ~150 μm from the edge of the micropattern, where mRNAs related to cell cycle class 2 (genes that promote cell cycle progression) were elevated (Fig. 5c, cell cycle class 2). Moreover, this region exhibited slightly increased levels of mRNAs and proteins related to (LIF) signalling at 0 h (Fig. 5c, signalling). While we considered that this might be an artefact of initial uneven cell seeding, cells were seeded on the micropattern 24 h before LIF withdrawal (i.e. at T = −24 h). Therefore, these subtle initial expression differences could be indicative of the first changes in position-based expression profiles. This is reinforced by two distinct areas on the micropattern emerging where both mRNA and protein levels tended to increase over time: these were found at positions 2 and 6 of the micropattern, located ~150 μm from either side of the edge (Fig. 5b).

Fig. 5: Spatial-temporal quantification of mRNAs and proteins after LIF withdrawal from mESCs on micropatterns.
figure 5

a Schematic of the experiment where ARTseq-FISH was performed at 0, 12, 24 and 48 h after LIF withdrawal. Created with BioRender.com. b Schematic of the definition of centre and edge on the micropattern. Created with BioRender.com. c Heatmap showing the abundance of different classes of targets (see Supplementary Data 5 for full list of targets) across 1 micropattern at 0, 12, 24 and 48 h after LIF withdrawal. The expression level of targets is normalised to the maximum abundance across all four time points. d Relative single-cell expression levels of the targets shown in (c), at the edge (top) and the centre (bottom) of the micropattern at 0, 12, 24 and 48 h after LIF withdrawal. 0–24-h data includes two biological replicates; 48 h includes three biological replicates (for the edge 393 cells and the centre 238 cells were analysed in total). For each replicate and time point, 1 micropattern was analysed. The line represents the mean. Shaded regions represent 95% confidence intervals. e Single cell NANOG protein expression at the edge and centre of micropattern at 0, 12, 24 and 48 h after LIF withdrawal detected by ARTseq-FISH (left). 0–24-h data includes two biological replicates; 48 h includes three biological replicates (edge and centre include 393 and 238 cells in total, respectively). For each replicate and time point, 1 micropattern was analysed. Quantification of live cell time-lapse of NANOG-GFP positive mESCs for five different positions at the edge and the centre of 1 micropattern over 46.5 h of LIF withdrawal (right). The line represents the mean. Shaded regions represent 95% confidence intervals. f Pearson correlation and hierarchical clustering of pluripotency and lineage-related markers at the edge of the micropattern (3 biological replicates, 1 micropattern each, 111 cells total) at 48 h after LIF withdrawal. g (i) UMAP clustering of expression levels of 57 markers of individual cells within a single micropattern experiment at different time points (0, 12, 24 and 48 h) after LIF withdrawal. (ii) UMAP clustering of expression levels of 57 markers of single cells at different positions (centre, edge or between centre and edge) within the micropattern (purple, orange and white, respectively). One biological replicate and 1 micropattern (0, 12, 24 and 48 h include 269, 139, 141 and 158 cells, respectively). Source data are provided as Source Data sheet tabs Fig. 5d–g.

To further explore whether proteins are expressed differently based on cellular location on the micropattern, we quantified gene expression profiles for individual cells situated either at the true edge or at the true centre of the micropattern (Fig. 5b). Once again, we categorised targets based on their functions and plotted the relative single-cell changes over the 0, 12, 24 and 48 h of LIF withdrawal. At 0 h, cells at the micropattern’s edge contain higher levels of proteins associated with cell cycle 1 (lengthening G1) than at the centre (Fig. 5d, 0 h cell cycle 1 top compared to bottom). Subsequently, at the edge after 12 h, we observed a decrease in proteins associated with cell cycle 1 (lengthening G1) and an increase in proteins related to cell cycle 2 (cell cycle progression) (Fig. 5d). These findings suggested that within the first ~12 h of LIF withdrawal, cells at the edge alter the expression of genes associated with cell cycle progression, potentially leading to higher proliferation rates during this time. In contrast, at the micropattern’s centre, the levels of cell cycle 1 (lengthening G1) increase already from 0–12 h, while cell cycle 2 genes (cell cycle progression) initially remain constant (Fig. 5d, bottom). These results indicated that the G1 phase of cells located at the centre might start increasing 0–12 h after LIF withdrawal.

When analysing genes associated with pluripotency and particular lineages, we only observed slightly different gene expression profiles at the centre compared to the edge of the micropattern. Proteins that exhibit the most significant differences in expression profiles are SOX17, BRACHYURY, (Fig. 5d, right) and NANOG (Fig. 5e, left). In particular, SOX17 levels increase mostly between 0–12 h at the edge of the micropattern, while at the centre, there is a more gradual increase over the course of 48 h of LIF withdrawal. BRACHYURY levels increase gradually at the edge of the micropattern from 12 to 48 h of LIF withdrawal and remain relatively constant at the centre. Lastly, NANOG levels increase from 0 to 12 h and decrease again from 12 to 24 h of LIF withdrawal at the edge of the micropattern. In contrast, at the centre of the micropattern, NANOG levels increase from 12 to 24 h of LIF withdrawal.

To validate these findings, we performed time-lapse microscopy on micropatterned NANOG-GFP mESCs57,58 for 46.5 h after LIF withdrawal (Fig. 5e, right). When quantifying the mean NANOG-GFP intensity for different positions at the edge and the centre of the micropattern, we observed distinct NANOG-GFP expression changes. Specifically, cells at the edge of the micropattern show an initial increase in NANOG-GFP expression (lasting ~12–15 h) followed by a gradual decrease in NANOG-GFP expression. Conversely, the centre of the micropattern displays a decrease in NANOG-GFP expression from 15 to 20 h post-LIF withdrawal, followed by a continuous increase in NANOG-GFP expression (Fig. 5e, right). While the overall trend at the edge is consistent with ARTseq-FISH NANOG data (Fig. 5e, left, orange), the initial decrease observed in the time-lapse experiment for cells at the centre of the micropattern is not visible by ARTseq-FISH (Fig. 5e, left, purple). This could be because ARTseq-FISH reveals expression levels of cells that are fixed after 0, 12, 24 and 48 h post-LIF withdrawal and do not represent a true time course, which might conceal rapid changes in expression levels. Furthermore, the time-lapse experiment was not analysed at a single-cell level. Alternatively, it is possible that the differences between the time-lapse experiment and ARTseq-FISH are cell-type specific. Therefore, we sought to confirm the expression of another pluripotency marker (OCT4) by immunofluorescence at 0 and 48 h of LIF withdrawal, which shows similar behaviour as ARTseq-FISH (Supplementary Fig. 26).

Interestingly, at the single-cell level, we found a negative correlation between pluripotency-associated proteins (i.e. OCT4 and SOX2) and lineage-associated markers (i.e. FOXA2, GATA6 and SOX17) (Fig. 5f) in cells at the edge of the micropattern. This indicates that cells at the edge are not uniform in their protein expression levels. When we project the data on a 2D UMAP, we observe the most prominent separation of cellular expression profiles occurring between 0 and 12 h post-LIF withdrawal (Fig. 5g). Furthermore, while there is some separation in cells located at the edge and centre, this separation is subtle, and predominantly visible at 48 h post-LIF withdrawal (Fig. 5g).

Cellular location on the micropattern impacts proliferation

In light of the gene expression profiles (Fig. 5c) indicating differences in protein signals between the micropattern’s edge and centre, we proceeded to investigate the translational activity of cells across the micropattern. To do this, we exposed micropatterned mESCs to the puromycin analogue O-propargyl-puromycin (OPP), which is incorporated into newly synthesised proteins59. Through click chemistry, we fluorescently labelled the nascently translated polypeptides (Fig. 6a). Our data revealed a gradual decrease in protein synthesis across the entire micropattern over time (Fig. 6b). Notably, the signal in the micropattern’s centre decreased more readily, particularly after 24–48 h (Fig. 6b, c, light and dark green). These findings are in line with ARTseq-FISH data, showing more prominent protein detection at the micropattern’s edge after 48 h of LIF withdrawal (Fig. 6d). These data suggest that spatial organisation plays a role in the dynamics of translational activity in micropatterned mESCs.

Fig. 6: mESCs display position-dependent differences in protein synthesis and proliferation rates.
figure 6

a Representative images showing the protein synthesis of the micropatterned mESCs at 0, 12, 24 and 48 h after LIF withdrawal. Images from different time points are set as the same maximum and minimum grey value Scale bar, 120 µm. b Quantification of relative fluorescent intensity of the composite image of 2–9 micropatterns of the protein synthesis in mESCs at 0, 12, 24 and 48 h after LIF withdrawal (individual intensity profiles of micropatterns in Source Data). c Representative images showing the protein synthesis of the micropatterned mESCs at 48 h after LIF withdrawal (left). Scale bar, 120 µm. Quantification of relative fluorescent intensity of the protein synthesis in mESCs across 10 micropatterns at 48 h after LIF withdrawal (right). d Pie chart showing the percentage of mRNAs, proteins, and phosphoproteins of targets that are more abundant at the centre (purple) and targets that are more abundant at the edge (orange) of the micropattern. e, f Heatmap of DAPI intensity across the micropattern and normalised DAPI intensity across the micropattern at 0, 12, 24 and 48 h after LIF withdrawal. g Schematic shows the ‘donut’ shape of the micropatterned mESCs. h Representative images showing EdU staining of the micropatterned mESCs at 0, 12, 24 and 48 h after LIF withdrawal. Images from different time points are set as the same maximum and minimum grey value. Insets: an image of EdU staining at 48 h after LIF withdrawal with adjusted contrast. Scale bar, 120 µm. i Quantification of relative fluorescent intensity of EdU staining in mESCs of composite image of 3–10 micropatterns (individual intensity profiles of micropatterns in Source Data) at 0, 12, 24 and 48 h after LIF withdrawal. j, Heatmap of Ki-67 expression level at z-projection across 1 micropattern at 0, 12, 24 and 48 h after LIF withdrawal. Source data are provided as Source Data sheet tabs Fig. 6a–j.

To gain insights into why translation might be reduced in the centre of the micropattern, we next tried to determine if cells proliferate differently depending on their position on the micropattern. To this end, we first quantified cell density across the micropattern as a function of DAPI intensity. Interestingly, on our 750 µm diameter micropatterns, cell density shows two radially symmetric peaks over time, ~100 µm from the edge of the micropattern (Fig. 6e, f), indicating that the colonies are not flat but somewhat ‘donut’ shaped (Fig. 6g). To determine if these differences in cell density were caused by increased proliferation, we performed EdU staining after 0, 12, 24 and 48 h of LIF withdrawal and quantified cell proliferation from the incorporation of EdU into newly synthesised DNA60 (Fig. 6h). The results indicate that there are fewer cells in the S phase at the centre of the micropattern than towards the edge and therefore likely display reduced proliferation already after 12 h of LIF withdrawal (Fig. 6i). Interestingly, at the very edge of the micropattern there still appear to be cells incorporating EdU after 48 h of LIF withdrawal. To confirm that actively cycling cells are more abundant at the edge of the micropattern after LIF withdrawal, we performed Ki-67 staining61 (Fig. 6j and Supplementary Fig. 27), an indirect marker for proliferating cells. The data show that at 0 h after LIF withdrawal, Ki-67 staining is mostly homogenous across the micropattern, indicating comparable cell cycling at the centre and edge of the micropattern (Fig. 6i). Notably, at 0 h there is reduced Ki-67 signal at the very centre of the micropattern (micropattern position 1 and 8), consistent with quantified EdU signal (Fig. 6i, light blue). From 0 to 12 h, there appears a slight increase in Ki-67 intensity (i.e. cell cycling) at the centre of the micropattern, which subsequently drops from 24 to 48 h post-LIF withdrawal, demonstrating a decrease in cellular proliferation or a state of cell cycle arrest (Fig. 6j and Supplementary Fig. 27). Caspase-3 and 7-AAD staining, as well as nuclear morphology analysis, indicate that this lack of proliferation could be connected to increased apoptosis at the centre of the micropattern62 (Supplementary Fig. 28). Conversely, at the edge of the micropattern cells show more consistent Ki-67 intensities over time (Fig. 6j), although at 48 h the cells at the very edge (micropattern position 1 and 8), do not show increased cycling as would be expected from the EdU staining (Fig. 6j compared to Fig. 6i), this could be because Ki-67 is less sensitive, in particular for cells with a short G1 phase63. Nevertheless, these combined assays indicate that cells positioned more towards the edge of the micropattern proliferate more readily than cells at the centre.

Cellular location on the micropattern impacts the cell cycle

Given the distinct expression profiles of cell cycle-related genes observed between cells at the micropattern’s edge and centre using ARTseq-FISH (Fig. 5d), we aimed to quantitatively analyse cell cycle progression on a single-cell basis. To this end, we conducted time-lapse microscopy on micropatterned Fucci mESCs64,65, which enabled the visualisation of cell cycle dynamics in live cells (Fig. 7a). Single-cell tracking and measurement of Azami Green (Az1) and Kusabira Orange-2 (KO-2) fluorescence over time delineates the individual stages of the cell cycle (Fig. 7b, c). We performed single-cell tracking of 35 cells across the micropattern (20 at the centre and 15 at the edge) over a 45-h period following LIF withdrawal. When assessing the length of the G1 phase at different time intervals post-LIF withdrawal, intriguing differences emerge between cells at the micropattern’s edge and those at the centre. Specifically, cells at the micropattern’s centre that enter the G1 phase within the initial 5 h of LIF withdrawal consistently exhibit a relatively short G1 phase of ~3 h (Fig. 7d, top and Supplementary Fig. 31a, left). Conversely, cells that enter the G1 phase after this initial 5-h period show a notably extended G1 phase (Fig. 7e). Markedly, cells at the micropattern’s edge display a G1 phase duration similar to cells at the centre (Fig. 7d, bottom and Supplementary Fig. 31a, middle). However, a subset (about 25%) of cells at the edge already exhibit a prolonged G1 phase within the first 5 h of LIF withdrawal (Supplementary Fig. 31a, right). Similarly, after the initial 5-h period, just over half (~60%) of cells closer to the edge continue to cycle, while the remaining portion (about 40%) display an extended G1 phase (Fig. 7f). These single-cell analyses show that the G1 phase of cells at the centre gradually elongates, while >half of the cells at the edge continue cycling until ~30 h post-LIF withdrawal. These data suggested that the cell cycle pace is more heterogeneous at the edge of the micropattern than at the centre.

Fig. 7: Live cell imaging reveals differences in cell-cycle progression.
figure 7

a Schematic illustration of the Fucci mESCs demonstrating different colours at different cell cycle stages. Created with BioRender.com. b, c Kusabira Orange-2 (KO-2) (G1) and Azami Green (Az1) (S/G2/M) intensity of a tracked cell in the centre (b) and edge (c) of the micropattern. d Average KO-2 intensity throughout the G1 phase for cells at the centre (top) and edge (bottom) of the micropattern that enters G1 within 5 h of LIF withdrawal. The average is 12 cells for the centre and 6 cells for the edge. The line represents the mean. Error bars represent standard deviation. e, f Single cell KO-2 (G1) traces of cells that enter G1 after 5 h of LIF withdrawal at the centre (e) and edge (f) of the micropattern. 7 (centre) and 9 (edge) cells were analysed respectively. g Quantification of the fraction of G1, early S, and late S/G2/M mESCs at the centre (left) and the edge (right) of the micropattern over time after 45 h of LIF withdrawal. 1961 cells at the edge and 2190 cells at the centre were analysed in total. h Quantification of mean EdU intensity of mESCs for ten different positions at the edge and the centre of 1 micropattern at 0, 12, 24 and 48 h after LIF withdrawal. The line represents the mean. Shaded regions represent 95% confidence intervals. Source data are provided as Source Data sheet tabs Fig. 7b–j.

We next performed a more comprehensive population-based analysis spanning the entire micropattern over the course of the 45-h LIF withdrawal. By comparing the proportion of cells in G1 versus S/G2/M phases for cells at the micropattern’s edge and centre over time, we again observe a subtle yet discernible difference (Fig. 7g). Specifically, in the centre, there was a gradual decline in the fraction of cells in S/G2/M phase, beginning ~6 h post-LIF withdrawal (Fig. 7g, left). Instead, cells at the edge of the micropattern show a relatively unchanged distribution across the cell cycle stages over time, with the proportion of cells in the S/G2/M phase only declining after ~40 h of LIF withdrawal (Fig. 7g. right). Since these data align with the single-cell analysis, the latter is likely representative of the full population. Furthermore, these findings are consistent with the Ki-67 staining results (Fig. 6j and Supplementary Fig. 27), and the cell cycle stages quantified from DAPI intensity differences66,67,68 (Supplementary Figs. 29, 30). Although differences in the timing of these changes might arise because the Fucci mESCs are a separate cell line64,65, the fraction of cells in the late S/G2/M phase (Fucci mESCs) follow a similar trend to the EdU staining performed on mESCs (Fig. 7h). Lastly, single-cell tracking of both NANOG-GFP and Fucci mESCs suggests that cell migration is minimal (Supplementary Fig. 31b, c, respectively). While the potential for rapid cell migration that eludes cell tracking exists, the prevailing data indicates that cells remain either at the centre or the edge of the micropattern (Supplementary Fig. 31c). Collectively, the data indicate that cells closer to the micropattern’s edge exhibit increased proliferation compared to those at the centre.