Multilayered control of mRNA delivery to cell protrusions drives localized mini-cytoplasm establishment for stress-induced cell activation

Abstract

In response to stress, cells exploit plasticity to self-activate and form protrusions that explore the environment and guide migration. While protrusion initiation is well characterized, their maturation and functional composition remain poorly understood. Here, using a new pooled genetic approach (ReGenT-seq) to interrogate chromatin factors controlling epidermal progenitor activation, we unexpectedly identified Cbx3 as a critical determinant of protrusion formation. Beyond its nuclear activities, we revealed a cytoplasmic moonlighting function of CBX3 that governs the subcellular localization of specific mRNAs within polarizing protrusions of migrating epidermal progenitor cells, as well as in invadopodia of squamous carcinoma cells. CBX3-dependent mRNA transport promotes the localized establishment of multiple organelles, including the endoplasmic reticulum and lysosomes, within maturing protrusions, and regulates focal adhesion dynamics. Together, our findings uncover a multilayered gene regulatory mechanism controlling cell motility through protrusionogenesis and identify a chromatin-to-cell-protrusion mRNA transport pathway that represents a potential therapeutic target for enhancing wound healing and limiting cancer invasion.

Here the authors show that, under stress, epithelial cells reveal an unexpected role for a chromatin factor in directing mRNA to cell protrusions, enabling local organelle assembly and cell movement, with implications for wound healing and cancer invasion.

Introduction

Under homeostatic conditions, epithelial stem cells preserve tissue integrity through interactions with other cellular components and by tightly controlling the balance between proliferation and differentiation^1–4. However, their intrinsic plasticity enables them to respond to multiple and severe stresses, including tissue injury and the acquisition of carcinogenic mutations, through cell activation (CA). Despite their heterogeneity, both insults converge on CA through shared transcriptional programs^5–9. In the skin, for example, epidermal stem cells (EpSCs) undergo dramatic changes following injury. They elicit remarkable plasticity, adopting behaviors that are absent from their homeostatic repertoire, such as cell migration. This newly acquired feature is essential for rapid and efficient wound healing^10,11. As a consequence of CA, these cells acquire a specific expression signature that partially mirrors that of activated EpSCs within precancerous and cancerous skin lesions, including the expression of Keratin 6^12,13. There is a striking convergence between epidermal cells during wound repair and in cancer, at both transcriptomic and chromatin levels, leading to similar patterns of lineage deregulation. This lineage infidelity constitutes a functional component of cell plasticity. However, while this activated state becomes permanently fixed in cancer, it is transient during wound healing and is released once tissue repair is complete¹⁴.

Through CA, diverse external stimuli and stress signals induce cell migration and cancer cell invasion, processes that require the establishment of cell protrusions^15–17. The initiation of protrusions has been extensively characterized and involves multiple molecular components, including actin polymerization, the Arp2/3 complex, and local depletion of actin-membrane links^18–20. In contrast, much less is known about the later stages of protrusion maturation and how their heterogeneous composition influences function.

Understanding the key molecular mechanisms governing CA offers the opportunity to identify new therapeutic targets that are effective in both wound healing and epidermal cancer contexts. This goal can be addressed using pooled genetic screens, which are a powerful tool for unbiased interrogation of gene function. However, the range of biological applications suitable to this approach has been limited by the available readout. Indeed, most studies identified genes that influence susceptibility to external perturbations or that intrinsically impact cell viability^21,22.

Epigenetic factors allow cells to adapt to stress through the onset of transcriptional programs that drive specific cellular phenotypes. Indeed, they represent key regulators of cell plasticity, a requirement for CA in diverse cellular contexts^23–26.

Over the past two decades, there has been growing interest in protein moonlighting (the ability of certain proteins to perform more than one distinct biological function²⁷), although this concept has only recently been appreciated in the context of epigenetic factors²⁸. However, it remains unclear whether epigenetic factors exert diverse functions across multiple layers of gene expression control, including cytoplasmic roles that regulate RNA availability within subcellular compartments. Understanding protein moonlighting, and thus the full complexity of protein functions, not only provides insight into fundamental questions in molecular biology, molecular evolution, and synthetic biology, but it is also essential for the development of more specific, efficient, and targeted therapeutic strategies.

In this study, we developed and applied ReGenT-seq (REciprocal GENetic Test), an integrated pooled genetic screening approach based on two complementary transcriptional reporter systems (one conferring drug resistance and the other drug susceptibility) to identify key epigenetic regulators of progenitor/stem CA. Because even the top-ranking hits from genetic screens can include false positives, arising from assay bias or stochastic variation, our method integrates two reciprocal sub-screens to effectively minimize this limitation. ReGenT-seq enables high-stringency genetic screening for defined transcriptional outputs, allowing robust identification of gene hits. Using this approach in cultures of epidermal progenitor cells (EPCs), we uncovered a network of chromatin activators and repressors that regulates CA by controlling lineage identity, transposable elements and motility-related genes. The functional characterization of the top hit Cbx3, a member of the chromobox (CBX) gene family^29,30, reveals an unexpected and predominant cytoplasmic role controlling the maturation of cell protrusions in both migrating EPCs and invading cancer cells. Cbx3 is required for protrusion biogenesis, particularly for establishing a complete and functional mini-cytoplasm within these structures. Indeed, CBX3 controls the subcellular compartmentalization of selected mRNAs required for the local establishment of multiple organelles, as well as other protrusion features, such as focal adhesions, thereby supporting protrusion maturation and function.

Results

Integrating reciprocal screens of epigenetic factors that control cell activation

To systematically identify epigenetic regulators governing epidermal cell plasticity that drive CA, we developed ReGenT-seq, a novel approach that combines multiple reciprocal pooled genetic screens (Fig. 1A–C). Specifically, we employed an optimised, previously well-characterised shRNA library targeting 615 chromatin regulators, comprising 5049 shRNAs³¹ (Supplementary Fig. 1A–C) and engineered two reporters that convert CA into antibiotic blasticidin (BL) resistance and into diphtheria toxin (DT) susceptibility, through DT receptor expression (Fig. 1B). To model CA, these reporters were based on Krt6b (K6) promoter transcriptional activity. We chose Krt6b for several reasons: (1) keratin 6 expression marks epidermal stress-related activation, both in wound healing and epidermal tumors^10,13; (2) Krt6b mRNA is not expressed under homeostatic conditions but is induced exclusively during wound healing, in contrast to Krt6a mRNA, which is also expressed under homeostasis in HF lineages (Supplementary Fig. 1D); (3) in vitro, Krt6b promoter is induced by multiple stress stimuli related to in vivo CA (Supplementary Fig. 1E, F). In our screening strategy, shRNAs that modulate K6 promoter activity through the targeting of epigenetic regulators influence cell viability upon drug treatment (BL or DT). BL administration positively selects EPCs expressing BL resistance, whereas DT treatment selectively depletes EPCs expressing the DT receptor (Fig. 1B). Intersection of the two sub-screens conducted with the same shRNA library but different reporters (K6-BL or K6-DT receptor) ensures robust identification of bona fide hits, which represent CA-positive or CA-negative regulators (see pink and violet curve lines in Fig. 1B representing the reciprocal intersection of the two sub-screens). The two readout systems were transduced into cultured EPCs (Supplementary Fig. 1G, H) as these cells represent an excellent in vitro model to study cell-stress-related states rather than epidermal homeostasis, as evident by comparative analysis of their expression profiles (Supplementary Fig. 2A–D). Validation using shRNAs targeting the AP1 transcription factor, a well-established activator of epidermal CA^14,32 that promotes Krt6 expression (Supplementary Fig. 3A–C), demonstrated that our screening strategy allows the identification of CA regulators. ReGenT-seq hits emerge from two independent pooled genetic sub-screens (3 biological replicates and 2 time points each), and display opposing selection patterns, being positively selected in one and negatively selected in the other (Fig. 1B). Thanks to our screen’s performance and reproducibility across replicates and two time points (Supplementary Fig. 3D–I) we obtained a total of 123 depleted and 105 enriched gene-hits in K6-BL resistance sub-screening, while 240 depleted and 96 enriched gene-hits in the K6-DT receptor sub-screening (Fig. 1D). However, applying our stringent approach based on their reciprocal intersection results in the identification of 11 negative and 6 positive regulators of epidermal CA (anti-CA and pro-CA, respectively) (Fig. 1E andSupplementary Data 1). The pronounced reduction in gene hits following reciprocal integration of the sub-screens (Fig. 1B–E and Supplementary Fig. 3G) suggests that our method effectively filters out genes that do not affect K6 promoter activity itself, but instead reflect unrelated effects, such as altered cell viability or involvement in blasticidin resistance or DT receptor-mediated toxicity. Consequently, ReGenT-seq substantially reduces the false discovery rate. Consistently, all tested shRNAs were validated by a single shRNA approach (Supplementary Fig. 3J).

Fig. 1. Identification of chromatin factors controlling cell activation through ReGenT-seq.

A ReGenT-seq workflow. Epidermal progenitor cells (EPCs) were transduced with two cell activation (CA)-reporter systems driven by the Krt6b promoter, conferring either blasticidin resistance or diphtheria toxin receptor (DToxinRec). Hygromycin-selected cells were subsequently transduced with a shRNA library targeting 5049 chromatin regulators and selected with neomycin. Cells were treated with blasticidin or diphtheria toxin (DT) or left untreated. Genomic DNA was isolated at 3 and 4 weeks post-transduction for library preparation. B Reciprocal integration strategy combining two complementary sub-screens to identify bona fide positive (pro-CA) and negative (anti-CA) regulators of cell activation. C Computational workflow of ReGenT-seq analysis. Blasticidin and DTR sub-screens are shown in green and yellow, respectively. D Ranked gene scores from both screens. ROAST gene set analysis (FDR < 0.25; Wu et al.⁹²) was used to integrate multiple shRNAs per gene. Top overlapping genes are highlighted in orange/blue. Gene score is defined as −log(P-value) ×  (number of enriched − depleted shRNAs). P-values are reported in Supplementary Data 1. n = 3 independent replicates per experimental group and time point. E Identification of CA regulators based on the intersection of sub-screens as in (C). Bars represent gene scores as defined in (D). F Viability screening workflow. Pooled genetic screens were performed using the same shRNA library as in ReGenT-seq. Genomic DNA was collected at 0, 2, and 3 weeks after selection for library preparation. G Venn diagrams showing overlap between depleted (blue) and enriched (red) hits in viability screens 1 and 2 (VS1 and VS2). Statistical significance of overlap was assessed using Fisher’s exact test. n = 3 independent replicates per experimental group and time point. H Protein–protein interaction network of ReGenT-seq hits. Node color intensity reflects CA score, while arrows indicate impact on viability (up: positive, down: negative). Gene ontology analysis revealed enrichment for “cellular response to stress”; corresponding factors are highlighted with yellow outlines.

Cell viability can be modulated upon CA through alterations in multiple cellular processes, including cell proliferation. Therefore, using the same epigenetic factor library employed in ReGenT-seq, we conducted a viability screen to gain further functional insight into the roles of chromatin regulators. With this approach, the enrichment or depletion of each shRNA in EPCs over time allows the identification of chromatin factors that regulate cell viability (Fig. 1F and Supplementary Fig. 4A–D). Integration of two identical but independent screens (3 biological replicates and 2 time points each) showed that 65% of the CA-epigenetic factors affect cell viability (Fig. 1G, Supplementary Fig. 4E and Supplementary Data 2). In addition, we observed no apparent correlation between the classical viability screening and ReGenT-seq, confirming that our newly developed method was not biased by general effects on proliferation.

We built the epigenetic factor network integrating ReGenT-seq and viability screen hits and protein–protein interaction (PPI) data (Supplementary Fig. 4F). This analysis revealed that epigenetic regulators of cell viability typically act coordinately as components of multi-protein complexes, whereas regulators of CA are organized into sparse modules within known complexes. (Supplementary Fig. 4F, G). In addition, only the functionality of these latter factors correlates with their propensity to form epigenetic complexes, highlighting their promiscuity (Supplementary Fig. 4H). Focusing on the chromatin factors identified by ReGenT-seq, we found that many are associated with the gene ontology (GO) term “cellular response to stress” (Fig. 1H).

Among the 17 ReGenT-seq hits (Fig. 1H), several genes have already been identified as cell plasticity regulators in other cellular contexts, including somatic cell reprogramming. For instance, Ehmt2, the top hit as a negative regulator of CA, is known to be a cell plasticity repressor^25,33, while Ncor1, one of our top positive CA regulators, is known to facilitate early cell reprogramming³⁴. Thus, the reciprocal approach implemented in ReGenT-seq constitutes a stringent and unbiased strategy applicable to pooled genetic screening with a transcriptional readout, enabling the identification of chromatin factors that govern CA.

Cell activation through regulation of illegitimate gene expression

Among the top hits identified as CA-positive and -negative regulators (Fig. 1H), we found several epigenetic factors known to inhibit transcription. In particular, ReGenT-seq identified the well-characterized transcriptional repressors Ehmt2 and Kmt5c (responsible for H3K9 mono-/di-methylation and H4K20 di-/tri-methylation, respectively^25,33,35), as well as Cbx3. Although CBX3 was originally characterized as a modulator of transcriptional silencing in heterochromatin-like complexes, it has also been directly associated with transcriptional activation and RNA splicing^29,30. To decipher the specific role of these three chromatin factors in activated EPCs, we performed RNA-seq following shRNA-mediated knockdown (KD) of each factor (Supplementary Fig. 5A–C). Interestingly, comparison of differentially expressed genes (DEGs) resulting from individual gene silencing revealed a strong correlation between CBX3 and KMT5CKD cells, which share a large common gene signature comprising both upregulated and downregulated genes, whereas EHMT2 KD cells did not show this overlap (Fig.2A, Supplementary Fig. 5D, and Supplementary Data 3). This strong similarity in transcriptional control between Kmt5c and Cbx3 is consistent with the ReGenT-seq results, which identified Cbx3 and Kmt5c as positive regulators, while Ehmt2 as a negative regulator of CA.

Fig. 2. Chromatin factors regulate cell activation through control of illegitimate gene expression and motility-related genes.

A Transcriptional programs dependent on Cbx3 and Kmt5c (pro-CA factors, green) are similar and differ from those dependent on Ehmt2 (anti-CA factor, red). Venn diagrams show the overlap between differentially expressed genes (DEGs) in CBX3, KMT5C and EHMT2 knockdown (KD) versus control cells from RNA-seq experiments, n = 5 replicates per experimental group. Green indicates activators, red indicates repressors of cell activation (CA) as identified by ReGenT-seq. B De-repression of RNAs with repetitive elements is more pronounced in EHMT2 KD in comparison to CBX3 and KMT5C KDs. Volcano plots showing differentially expressed (DE) repetitive DNA elements in CBX3, KMT5C and EHMT2 KD vs. control cells. Analysis was performed using DESeq2 on top of repeat expression levels quantification by TETranscripts. Significant differentially expressed elements (|FC| > 1, FDR < 0.05) are indicated in red (upregulated) and blue (downregulated). C Bar plot showing the percentage of up-regulated transposable elements (TEs) families, for each KD condition. D Heatmap showing hierarchical clustering of pairwise TEs deregulation similarity among the three KD conditions. E Heatmap showing lineage deregulation similarity among the three KDs. Pairwise similarity is calculated as Pearson’s correlation of NES values obtained by GSEA analysis of neuron-related gene sets from the Descartes Atlas. F Heatmap showing GSEA analysis (Molecular Function and Biological Process) for each KD condition. G GSEA enrichment plots from RNA-seq data showing downregulation of “epithelial-to-mesenchymal transition” (EMT) and “regulation of chemotaxis” in KMT5C KD versus control cells. H GSEA enrichment plots and GO analysis of upregulated and downregulated genes in CBX3 KD EPCs, indicating positive regulation of migration and negative regulation of proliferation by CBX3. For all panels, GSEA was performed using a running-sum statistic with permutation-based testing; P values were adjusted using the Benjamini–Hochberg method.

Although CBX3, EHMT2, and KMT5C all function as transcriptional repressors, EHMT2 KD resulted in the most pronounced de-repression (Supplementary Fig. 5E), with a marked impact on repetitive DNA sequences (Fig. 2B and Supplementary Fig. 5F). By contrast, KMT5C and CBX3 regulate a more restricted set of RNAs containing repetitive elements from similar classes, whereas EHMT2 exerts broad repression across repetitive RNAs, including those derived from transposable elements (Fig. 2C, D). These differences in transcriptional repression activity are once again consistent with ReGenT-seq results. The strong silencing activity of EHMT2 suggests that it could serve as a powerful lineage keeper to preserve lineage identity, preventing the expression of genes associated with alternative cell lineages. This interpretation is supported by in vivo data indicating that lineage infidelity represents a positive functional feature of cell plasticity during wound healing, as well as during metastatization^14,36. To test this hypothesis, we performed GSEA analysis on DEGs from CBX3-, KMT5C- and EHMT2-depleted versus control cells, using gene signatures of developing cell types³⁷, as epidermal cells re-activate embryonic gene programs during wound healing³⁸. This analysis further highlights the similarity between the pro-CA factors Cbx3 and Kmt5c, compared to the anti-CA factor Ehmt2, in terms of lineage deregulation (Fig. 2E, F, and Supplementary Fig. 5G). Specifically, EHMT2 KD resulted in a marked de-repression of genes associated with neuronal development (Supplementary Fig. 5G), a finding supported also by GSEA of Biological Process and Molecular Function GO terms (Fig. 2F). In parallel, this analysis showed that EHMT2 KD cells activate piRNA-associated pathways to cope with the highest induction of RNAs with transposable elements in comparison to CBX3- and KMT5C- KD cells (Fig. 2F).

Overall, our transcriptomic data pointed towards a scenario where EHMT2 acts as a negative regulator of cell plasticity by maintaining a strong transcriptional repression, thereby preventing the expression of specific plasticity traits, such as lineage infidelity. On the other hand, KMT5C and CBX3, identified as positive regulators of cell plasticity, regulate a largely overlapping transcriptional program.

Balancing cell fate between activated phenotypes: CBX3 promotes migration and invasion while restraining cell proliferation

Following an injury, activated EpSCs display remarkable plasticity, engaging in multiple behaviors that are not characteristic of their homeostatic repertoire. For instance, EpSCs respond to injury with enhanced proliferation and migration in two distinct areas proximal to the wound, known as the proliferative hub and the migratory leading edge, respectively^10,11.

GO enrichment and GSEA analysis of DEGs from both KMT5C- and CBX3-KD versus control EPCs predicted that these two chromatin factors, identified as hits by ReGenT-seq, promote CA primarily by favoring cell migration (Fig. 2G, H). Based on previous studies, this result was anticipated for Kmt5c³⁵ (Supplementary Fig. 6A), whereas it was more unexpected for Cbx3, which has been predominantly studied as a heterochromatin-associated protein. For these reasons, we focused subsequent analyses on defining the role of Cbx3 in CA and elucidating the underlying molecular mechanisms.

GO term and GSEA analyses of CBX3 KD cells revealed that, in addition to its role in promoting expression of genes associated with motility (i.e., epithelial-mesenchymal transition, EMT), CBX3 negatively regulates the expression of genes linked to proliferation (Fig. 2H). Here, this RNA-seq analysis, performed on CBX3 KD and control EPCs, captures global transcriptional changes reflecting both direct and indirect effects of CBX3 on gene expression. Consistent with this, the GO term and GSEA results align with our pooled genetic screen, which identifies Cbx3 as a negative regulator of cell viability (Fig. 1F, G). To further validate these predictions, we assessed the functional impact of CBX3 perturbation. Wound-healing and proliferation assays using CBX3-KD and CBX3-overexpressing EPCs confirmed that it promotes cell migration, while restraining proliferation (Fig. 3A and Supplementary Fig. 6B–E).

Fig. 3. CBX3 contributes to cell activation by promoting cell migration in EPCs and invasion in SCC cells.

A Scratch assay validating the role of Cbx3 in cell migration. Two shRNAs targeting Cbx3 were compared with the control. The wound area is shown in black (left panels). Quantification of wound closure (%) is shown as the mean of n = 4–6 independent experiments. Error bars represent ± SEM. B CBX3 levels in skin sections under homeostatic conditions (left) and during wound healing (right), showing the proliferative hub and leading edge at 1 week post-wounding. C Quantification of CBX3 levels in basal epidermal cells within the proliferative hub and leading edge. n = 9 images from 3 mice. D Correlation between CBX3 expression and basal cell elongation. Line thickness indicates variability (±SEM). Pearson’s correlation coefficient (r) and P-value are shown. n = 9 images from 3 mice. E Keratin 6 levels in basal and suprabasal cells within the proliferative hub and leading edge. Median Keratin 6 level in basal cells of the proliferative hub is shown as a reference. n = 414–553 cells from 3 mice. Statistical significance was determined using a t-test with Benjamini–Hochberg false discovery rate (FDR) correction. F, G CBX3 and Ki-67 levels in human skin from acute wounds, chronic venous ulcers, and pyoderma gangrenosum-associated ulcers. CBX3 is expressed as the edge/hub fluorescence intensity ratio, while Ki-67 is shown as the ratio of positive cells between regions. n = 5–11 samples. Representative immunofluorescence images are shown in (G). H, I SCC sections showing CBX3 overexpression in deep versus superficial tumor regions. Pseudocolouring based on cell detection highlights IFE, HF, necrosis, stroma, and tumor cells in the upper panel of (H). DAPI is shown in the lower panel of (H), and CBX3 in (I). J CBX3 (green) and Keratin 6 (red) expression in superficial and deep SCC sections. K Virtual slide generated using QuPath showing CBX3 (left) and Keratin 6 (right) distribution within SCC regions. Stroma and suprabasal epidermal cells are shown in gray and yellow; basal cells are color-coded based on protein intensity. L, M Quantification of CBX3 and Keratin 6 in superficial and deep SCCs. n = 3–4 images from 3 mice. N CBX3 overexpression enhances SCC invasion. Invasion assay results are shown as % translocated nuclei (left) and representative immunofluorescence images (right). n = 10 images from 2 independent experiments. Error bars represent ± SEM. Box plots show median, IQR, and whiskers (1.5× IQR). Statistical significance was determined using unpaired two-sided Student’s t test unless otherwise stated. Dotted white lines separate epidermis from dermis (B, G). Scale bars: 25 μm (N), 50 μm (B, G), 100 μm (J), 1000 μm (I).

In order to assess the potential in vivo relevance of Cbx3 in coordinating migration and proliferation in injured skin, we performed histological analysis of murine skin during wound healing. Following the identification of the proliferative hub and the migratory leading edge during re-epithelialization (Supplementary Fig. 7A–F), we found that CBX3 expression in basal cells is higher at the leading edge relative to the proliferative hub (Fig. 3B–D). Migratory basal cells also displayed an increased expression of the stress marker Keratin 6 (Fig. 3E and Supplementary Fig. 7G).

To evaluate the potential clinical relevance of CBX3 in the context of acute versus chronic wounds, we compared its expression in human skin samples from acute injuries in the healing phase with those from chronic venous ulcers and pyoderma gangrenosum-associated ulcers. Firstly, consistent with our observations in murine acute wounds (Fig. 3B–D), we found that human acute wounds display well-defined epidermal regions required for effective healing¹¹. These comprise highly proliferative regions marked by densely packed, Ki-67-positive epidermal cells with nuclei predominantly oriented perpendicular to the basement membrane, alongside distinct migratory regions that are largely Ki-67-negative and exhibit nuclei aligned more parallel to the basement membrane (Fig. 3F, G). In agreement with our findings in mouse skin (Fig. 3B–D), basal epidermal cells in human acute wounds exhibited significantly higher CBX3 expression in migratory regions compared with areas of enhanced proliferation (Fig. 3F, G). By contrast, this spatial correlation between CBX3 expression and migratory versus proliferative compartments was lost in both types of chronic wounds analyzed (Fig. 3F, G). This likely reflects the altered architectural organization of chronic wounds, in which the proliferative compartment predominates, while the migratory edge is reduced or occasionally absent. These observations are consistent with the prevailing view that keratinocytes in chronic wounds often display hyperproliferation but impaired migratory capacity^39,40. Importantly, CBX3 expression correlates with migratory behavior across wound types (Fig. 3F, G), further supporting a critical role for Cbx3 in regulating migration in activated epidermal cells during wound healing.

Since we found Keratin 6, the transcriptional readout of our ReGenT-seq, elevated in SCC relative to homeostatic conditions (Supplementary Fig. 8A), we evaluated CBX3 and Keratin 6 expression in basal epidermal cells of superficial and deep areas of invasive squamous cell carcinomas SCC (Supplementary Fig. 8B–E). This analysis revealed that the highest expression of CBX3 and Keratin 6 is present in basal cells of the deep SCC areas, where the invasive SCC features are elicited (Fig. 3H–M and Supplementary Fig. 8F, G), with respect to the far non-tumoral regions. Intriguingly, we also detected rare cells in the deepest regions of SCC that exhibited cytoplasmic localization of CBX3 (Supplementary Fig. 8H). This suggested an unexpected non-nuclear function of the CBX3 chromatin factor in invading cells.

In clinics, metastasis is generally associated with poor prognosis. Together with evidence that CBX3 is overexpressed in multiple epithelial cancers and its expression is associated with poor prognosis in numerous carcinomas⁴¹, our data prompted us to hypothesize that CBX3 plays a functional role in cancer cell invasion. To test this hypothesis, we performed in vitro scratch and invasion assays using cutaneous SCC cells. The results showed that CBX3 overexpression markedly promotes cell invasion (Fig. 3N and Supplementary Fig. 8I, J).

Overall, these data indicate that CBX3 favors migratory over proliferative cell state in activated EPCs. This fate switch between cell phenotypes is likely to be functional in vivo during re-epithelialization of a cutaneous injury. In addition, CBX3 overexpression promotes invasion of SCC cells, a mechanism that could be at the basis of metastatization in carcinomas with poor prognosis.

CBX3-dependent cell migration extends beyond its nuclear multi-layered gene regulation

Cbx3 belongs to the heterochromatin protein 1 (HP1) family, encoded by a class of genes known as the chromobox (CBX) genes⁴². HP1 proteins were initially characterized as key mediators of heterochromatin-dependent silencing of repetitive DNA^43,44, and selected gene promoters⁴⁵. Subsequent studies demonstrated that CBX3 localizes within the transcribed region of active genes⁴⁶ and more recent work has established its role as a mediator of co-transcriptional RNA processing²⁹. In line with this function, CBX3 has been shown to influence RNA splicing by coupling mRNA maturation to chromatin through recognition of methylated H3K9⁴⁷, as well as to regulate splicing fidelity by limiting splicing noise⁴². Mechanistically, exogenously tagged CBX3 has been reported to bind RNA within intronic regions, thereby modulating splicing outcomes⁴⁸.

Within this complex regulatory landscape, we performed an in-depth molecular analysis to dissect the multilayered nuclear mechanisms by which CBX3 governs gene expression and epidermal cell fate decisions, summarised in Fig. 4A. To understand its transcriptional role, we first assessed the genomic direct targets of endogenous CBX3 by ChIP-seq (Supplementary Fig. 9A–E). As previously observed²⁹, CBX3 binds within gene bodies (Supplementary Fig. 9C), consistent with gene activity (Fig. 4B, Supplementary Fig. 9F, G, and Supplementary Data 4). However, based on adjusted P-values from GO enrichment analysis of MSigDB Hallmark gene sets, the overlap between ChIP-seq peaks and genes deregulated in CBX3-depleted cells suggests that CBX3 plays only a modest direct role in regulating migration-associated gene expression, while exerting a stronger influence on other gene sets (Supplementary Fig. 9H). After the identification of CBX3 DNA direct targets, since it is involved in co-transcriptional RNA processing, we next identified its RNA direct targets by seCLIP (Fig. 4C–G, Supplementary Fig. 10A–F, and Supplementary Fig. 11A, B). We found that endogenous CBX3 predominantly binds spliced, mature mRNAs in multiple sites, mostly at the level of a GAAGAAGA consensus that is likely to adopt a single strand conformation (Fig. 4D–F and Supplementary Fig. 10E). In addition, CBX3 binds multicopy RNAs, including the spliceosomal snRNAs U1 and U6, as well as snoRNAs involved in rRNA processing, and 7SK RNA, which is required for proper regulation of RNA Polymerase II transcription⁴⁹ (Supplementary Fig. 11B). Considering that CBX3 GAAGAAGA motif contains the GAAGAA hexamer, an exonic splicing enhancer (ESE) essential for splicing⁵⁰, we assessed the effect of CBX3 depletion on splicing (Fig. 4H, I, and Supplementary Fig. 12A–E). Consistently with previous work³⁰, CBX3 KD led to splicing deregulation, including de novo junction formation and differential alternative splicing events (ASE) affecting both direct and indirect CBX3 RNA targets (Fig. 4H, I, and Supplementary Fig. 12A). To further characterize the direct role of CBX3 in splicing, we performed GO term enrichment analysis on genes that were both differentially spliced upon CBX3 depletion and bound at the RNA level by CBX3 (as identified by seCLIP). This revealed that these splicing alterations primarily affect genes related to cell proliferation and RNA processing, rather than cell migration (Fig. 4I).

Fig. 4. Multilayered control of gene expression by nuclear CBX3 activities.

A Schematic representation of the nuclear regulatory activities of CBX3 examined in Fig. 4. B CBX3 chromatin binding across gene bodies increases with gene expression levels. The box plot shows the average CBX3 ChIP-seq signal (n = 2 replicates) within gene bodies. Genes are grouped into RNA-seq expression quartiles (n = 5 replicates) (Q1–Q4). A random size-matched set of non-expressed genes is included as a control. C Distribution of CBX3 seCLIP peaks across annotated genic regions (n = 2 replicates). D Cbx3 RNA-binding motif identified by Homer de novo motif discovery. E Top CBX3 RNA-binding motif in seCLIP peaks and occurrence frequency. F 6-mer enrichment in ssRNA/dsRNA structures. The x-axis shows average unpairedness probability; the y-axis shows the Wilcoxon rank-sum estimate comparing observed vs. randomized sequences. Nonsignificant 6-mers are gray, ssRNA-enriched are dark orange, and dsRNA-enriched are blue. The top five ssRNA-enriched 6-mers are indicated. G IGV snapshot of CBX3 RNA-binding profiles (orange). Size-matched input (SMI, yellow), RNA-seq (gray), and CBX3 peaks are shown. H Differential junction usage in Cbx3 KD vs. scramble (P -value< 0.05). De novo junctions (not in GENCODE M23) were quantified per gene and analyzed using Poisson regression (glm, R; family = “poisson”). Log2-transformed junction counts (+1 pseudocount) are shown. Cbx3-bound RNAs are indicated on the right. n = 3 replicates per condition. I GO enrichment analysis of genes associated with ASEs and splicing noise was performed using Enrichr. J CBX3-dependent regulation of RNA nuclear export. Box plots show RNA nuclear export differences in CBX3 KD vs. control from subcellular RNA-seq, analyzed using edgeR GLM with quasi-likelihood F-test (QLF). Genes are grouped by FDR (<0.05) and log2FC. n = 3 replicates per group. The bar plot shows CBX3-bound and unbound differentially exported genes. K Overlap between CBX3 RNA binding and RNA regulatory outputs. Enrichment of seCLIP-bound mRNAs in ASEs, noisy splicing, and nuclear export categories is shown. Significance was assessed using Fisher’s exact test. For all box plots, median, IQR, and 1.5× IQR are indicated. Statistical significance was assessed using unpaired two-sided Student’s t test unless otherwise stated.

Although our data clearly show that CBX3 regulates RNA splicing, it was surprising to find that most CBX3-bound RNAs were in the mature form, as it preferentially binds RNA exons (Fig. 4C). This observation, together with the finding that CBX3 does not impact on RNA stability in mammalian cells^29,51, led us to hypothesize a role for CBX3 in modulating RNA nuclear export. To test this hypothesis, we performed RNA-seq on nuclear and cytoplasmic RNA isolated from CBX3 KD and control cells (Supplementary Fig. 12F). We found that CBX3 functions to retain transposable element-containing RNAs within the nucleus (Supplementary Fig. 12G). However, despite high reproducibility between replicates (Supplementary Fig. 12F), CBX3 was found to exert only a modest fine-tuning effect on mRNA nuclear export; this is supported by the statistically significant but low fold changes observed in differentially exported RNAs between CBX3- KD and control cells (Fig. 4J).

Collectively, our multi-omics analysis identifies RNA splicing regulation as the primary direct nuclear function of CBX3 in transcriptionally active genes (Fig. 4K and Supplementary Data 5). However, our findings did not support that its direct nuclear gene regulatory functions were sufficient to drive cell migration. Together with the observation that endogenous CBX3 predominantly binds mature RNA and has only a limited effect on RNA nuclear export, these findings suggested the existence of an additional CBX3-dependent layer of gene expression regulation. In particular, we hypothesized a role for CBX3 in RNA localization, given that GA-rich RNA sequences similar to its binding motif have been implicated in RNA transport to cell protrusions^52,53.

CBX3 governs cell protrusion maturation by regulating mRNA availability for local organelle organization and focal adhesion turnover

Cell protrusions are dynamic structures essential for environmental exploration and directed cell motility. Indeed, migrating cells extend protrusions at the leading edge prior to substrate adhesion.

Although CBX3 is predominant in the nucleus, we observed a precise cytoplasmic distribution pattern. In particular, CBX3 is enriched in the front cytoplasm of polarized cells, as defined by the Golgi apparatus distribution, compared with the rear, and it is also detected at high levels within leading cell protrusions⁵⁴ (Fig. 5A, B). This observation is consistent with previous mass spectrometry data on cell protrusions performed by Dermit et al.⁵⁵ across 6 different cell lines. Our re-analysis of these data confirmed that CBX3 is enriched in cell protrusions compared with cell bodies (Supplementary Fig. 13A, B), further supporting the hypothesis that CBX3 directs RNA subcellular distribution within cell protrusions. To test this, we performed RNA-seq on fractionated nuclear, cytoplasmic and protrusion extracts from CBX3 KD and control EPCs. This optimized protocol^55,56 allowed segregation of transcripts from these three different compartments (Fig. 5C, D, and Supplementary Fig. 13C, D). As previously observed⁵⁵, GSEA revealed a very strong enrichment of the RNAs of ribosomal proteins within cell protrusions (Supplementary Fig. 13E). Differential RNA representation analysis identified distinct transcript signatures of each cell compartment (Supplementary Fig. 13F). Their GO enrichment analysis revealed that RNAs derived from genes involved in signaling pathways are enriched in the nucleus and cytoplasm, whereas RNAs associated with genes encoding nuclear protein complexes are unexpectedly enriched in cell protrusions. In contrast, transcripts from genes involved in diverse metabolic processes are distributed across multiple cellular compartments (Supplementary Fig. 13G). We next investigated the role of Cbx3 in RNA localization within cell protrusions. A stringent differential RNA representation analysis comparing protrusions from control versus CBX3 KD cells surprisingly revealed that a reduction of CBX3 strikingly decreases the availability of specific RNAs within cell protrusion (641 downregulated genes with a 2.2 average fold change, while only a small number of genes were upregulated) (Fig. 5E, Supplementary Fig. 13H, and Supplementary Data 6). In addition, CBX3-dependent mRNAs localized to protrusions were associated not only with cell migration-related GO terms, such as “positive regulation of wound healing,” “collagen-containing extracellular matrix,” and “focal adhesion” (Fig. 5F), but also with processes involved in organelle organization. Notably, these enriched GO terms did not include mitochondria or peroxisomes, but instead prominently featured the Golgi apparatus, lysosomes, endosomes, secretory granules, and, strikingly, the endoplasmic reticulum (ER) with an adjusted P-value of 2.2 × 10⁻⁹⁷ (Fig. 5F, G). Collectively, these unexpected findings suggest a model in which the functional establishment of specific organelles represents a defining feature of certain complex cell protrusions, reflecting a maturation process that may depend on the availability of organelle-related mRNAs under CBX3 control. Interestingly, transcriptomic comparison between whole-cell (Fig. 2A) and compartment-specific (Fig. 5C–G) analyses of CBX3 KD versus control cells indicated that certain cellular processes, such as cell proliferation, are regulated by CBX3 independently of mRNA compartmentalization, whereas CBX3 governs cell migration largely through the control of mRNA availability within cell protrusions (Fig. 5H and Supplementary Fig. 14A). Moreover, the integration of these transcriptomic data with ChIP-seq and seCLIP datasets revealed that CBX3 directly operates at both chromatin and mRNA subcellular localization levels for a subset of migration-associated genes, which are more likely to impact on cell phenotypes (Supplementary Fig. 14B).

Fig. 5. CBX3 marks a subset of frontal cell protrusions and controls their RNA composition.

A, B Polarization of CBX3 in cell protrusion and cytoplasm. A Co-staining of CBX3 and Wheat Germ Agglutinin (WGA) in EPCs. B CBX3 quantification within the Golgi-facing and the Golgi-opposite sides is shown in cell protrusions (Protru.) and cytoplasm (Cytopl.). The horizontal line indicates the median, the box represents the interquartile range (IQR), and whiskers denote the 1.5× IQR. Statistical significance was assessed using a paired two-sided Student’s t test. n = 32–46 cells from 2 independent experiments. C Schematic of the transwell-based protrusion induction assay and subsequent fractionation to isolate nucleus, cytoplasm, and cell protrusions. FBS fetal bovine serum, GFs growth factors. D PCA of gene expression (2000 top variable genes) in control vs. CBX3 knockdown (KD) EPCs from cell body, nucleus, cytoplasm, and protrusion fractions. n = 3 independent replicates per group. E CBX3 KD leads to a strong and specific decrease in RNA abundance in cell protrusions. Heatmaps show all DEGs in protrusion versus nucleus and cytoplasm (left and right, respectively), in CBX3 KD and control (scr) cells. Differential subcellular expression analysis was performed using the edgeR GLM testing framework with quasi-likelihood F-test (QLF). DEGs present in both compartments or restricted to one are highlighted in red. F GO enrichment analysis of downregulated genes in CBX3-KD versus control cells was performed using Enrichr. Adjusted P values (Benjamini–Hochberg correction) are reported. G Analysis of DEGs (P < 0.05) across subcellular compartments in CBX3-KD cells, focusing on selected GO terms related to cell organelles (as in F). For each GO term, the ratio between the number of DEGs and the total number of genes annotated to that term is shown. H Comparison of CBX3-dependent gene expression regulation at the whole-cell level and within cell protrusions. In the outer circular plot, the difference between the number of up- and down-regulated DEGs in whole cells, protrusions or both is shown. The inner circle indicates the mean log(FC) of DEGs for each GO term. In the right panel, DEGs within each GO term are scored (greyscale) based on whether they are direct CBX3 chromatin (ChIP-seq) or RNA (seCLIP) targets. Scale bar: 25 μm (A).

Pairwise integration of the omics datasets revealed that CBX3 directly mediates multilayered expression control across gene categories linked to distinct cellular functions (Supplementary Fig. 14C). Specifically, CBX3 regulates focal adhesion genes at multiple levels, including co-transcriptional RNA processing, splicing, nuclear export, and mRNA localization within cell protrusions. In contrast, actin cytoskeleton genes are primarily regulated at the level of co-transcriptional RNA processing; E2F-associated genes, known regulators of the G1/S transition, are regulated at the levels of splicing and nuclear export; and ER-associated genes are predominantly regulated through control of mRNA availability within cell protrusions (Supplementary Fig. 14C).

To validate this unexpected Cbx3-dependent protrusion scenario, we firstly performed smFISH in classical 2D cultured conditions. By comparing CBX3 KD and control EPCs, we confirmed that the availability of organelle-related mRNAs in cell protrusions is Cbx3-dependent (Fig. 6A, B, and Supplementary Fig. 14D). We then performed an histological analysis of cell protrusions to evaluate their molecular and morphogenic features in CBX3 KD and control EPCs. This analysis confirmed that, in 2D cultures, CBX3 is present only in a subset of protrusions (Fig. 6C). Notably, CBX3 localization within cell protrusions correlates with protrusions size, as well as with the presence of wheat germ agglutinin (WGA) which marks cellular membranes including those of the Golgi apparatus, and Phospho-FAK (Tyr925), a marker associated with focal adhesion turnover required for cell migration⁵⁷ (Fig. 6C–F and Supplementary Fig. 14E). Strikingly, these molecular and physical features are all CBX3-dependent in cell protrusions, as well as focal adhesions per se in polarized cells (Fig. 6D–H). We next performed histological analysis using markers of the ER and lysosomes, P4HB and LAMP1, respectively,^58,59 and validated their presence in a subset of cell protrusions in 2D cultures. Remarkably, CBX3 KD leads to a substantial decrease of P4HB and LAMP1, specifically within cell protrusions (Fig. 6I, J). By contrast, consistently with our transcriptomic data (Fig. 5E–G), mitochondria in cell protrusions were unaffected by CBX3 KD (Supplementary Fig. 14F, G).

Fig. 6. CBX3 controls cell protrusion maturation by regulating mRNA availability for focal adhesions and organelles.

A Representative smFISH image showing localization of CBX3 and mRNAs (Pdia3, Hspa5, and P4hb) directly bound by CBX3 (as identified by seCLIP). Cell boundaries (red lines) were defined by Phalloidin staining. B Quantification of the same smFISH experiments shown in (A), reporting differential RNA localization in CBX3 knockdown (KD) versus control cells. Signal in protrusions relative to the cell body is expressed as mean grey value; 7–15 cells were analyzed per probe. C Representative immunofluorescence images of CBX3, WGA, and p-FAK in EPCs. Orange boxes indicate CBX3-positive and WGA-positive protrusions, while yellow boxes highlight protrusions negative for both markers. Orange arrows indicate CBX3-, WGA-, and p-FAK-positive cell protrusions. Asymmetric localization of the Golgi (red arrow) marks cell polarity. Separate and merged channels are shown. D PCA of protrusion features (AR = aspect ratio; Circ. = circularity; Solid. = Solidity; Feret = Feret’s diameter; Round = roundness; Perim. = perimeter; area) and CBX3, p-FAK, WGA intensities in KD and scramble control cells. Box plots show distributions along PCA dimension 1. PCA data are colored by CBX3 intensity or represented by directional arrows for each descriptor. E Values of CBX3, p-FAK, WGA intensities, and protrusion shape descriptors in CBX3 KD relative to scramble controls. Statistical significance was assessed using unpaired two-sided Student’s t tests; P-values for each shRNA are shown. F Correlation between WGA and CBX3 intensities in p-FAK-positive protrusions from scramble and shRNA1/2 cells. Pearson’s r and P-value are indicated. n = 416–575 protrusions. G Quantification of p-FAK levels in scramble, shRNA1 and shRNA2 cells. Each dot represents the mean p-FAK intensity per cell. n = 22–28 cells. H Representative images and quantification of vinculin-positive protrusion area in migrating cells. Each dot represents the mean per cell. n = 33–36 cells. I, J Quantification of P4HB and LAMP1 in p-FAK-positive protrusions, marking ER and lysosomes, respectively, in scramble and shRNA cells. Each dot represents the mean per cell. n = 27–34 (P4HB) and 33–48 (LAMP1). For all box plots, dots represent data from two independent experiments. The horizontal line indicates the median, boxes represent the interquartile range (IQR), and whiskers denote 1.5× IQR. Statistical significance was assessed using unpaired two-sided Student’s t tests unless otherwise stated. Scale bars: 12.5 μm (A), 25 μm (C, H), 12.5 μm (I, J).

Finally, to assess functional conservation, we performed representative experiments in spontaneously immortalized human keratinocytes derived from primary epidermal cells, demonstrating that CBX3 overexpression recapitulates the murine phenotype (Supplementary Fig. 15).

Altogether, our data reveal pronounced heterogeneity among cell protrusions within individual cells under 2D culture conditions, closely linked to cell polarity and migration behavior. We identified a distinct subset of cell protrusions characterized by CBX3 enrichment, the presence of multiple organelles (including the ER and lysosomes) and elevated focal adhesion turnover. We refer to these protrusions as mature, in contrast to CBX3-negative early protrusions. Our data further suggest that CBX3-positive protrusions represent an exploratory state that precedes productive cell migration. Importantly, CBX3 emerges as a critical regulator of these defining protrusion features, while others, such as mitochondria localization, are CBX3-independent. Mechanistically, CBX3 controls the local availability of mRNAs required for organelle establishment and focal adhesion turnover within large, mature cell protrusions.

The establishment of mature protrusions in EPCs and invasive protrusions in cutaneous SCC cells requires CBX3

As shown earlier, EPCs in standard 2D culture conditions exhibit a spectrum of distinct cell protrusions. By contrast, in the 3D culture setting used to direct cell protrusion formation (Fig. 5C), all the protrusions are mature (Fig. 7A–C and Supplementary Fig. 16A). This homogeneous context enables a clearer assessment of the role of Cbx3 in the establishment of newly induced mature protrusions. Within hours after induction, murine and human epidermal cells formed mature protrusions, while CBX3 reduction markedly decreased protrusion number. Conversely, CBX3 overexpression in human keratinocytes increased protrusion formation. (Fig. 7D, E, Supplementary Figs. 15A, and 16B–E).

Fig. 7. The induction of mature protrusions in EPCs and invasive protrusions in cutaneous SCC cells requires CBX3.

A Confocal microscopy strategy to visualize 3D-induced mature protrusions and the cell body of EPCs. B Cross-sections of confocal Z-stack images showing 3D-induced protrusions in EPCs. Protrusions induced through 3-μm pores were stained for p-FAK, LAMP1, P4HB, vinculin, and WGA. Cell boundaries (brown outlines) were defined by F-actin staining. C Quantification of immunofluorescence images from 2D protrusions (Fig. 6) and 3D-induced mature protrusions (Fig. 7B), expressed as percentage of positive protrusions relative to total protrusions for each marker. D Representative F-actin and CBX3 staining in cell body (upper) and mature protrusions (lower) of scramble, shRNA1, and shRNA2 cells. Cell boundaries (brown outlines) are indicated. Bright-field overlaid with green channel shows pores in the transwell filters. E Quantification of induced mature protrusions per cell in KD versus control cells. Data are shown as mean ± SEM from 3 to 5 images. F Schematic representation of the invasive protrusion assay. SCC cells were seeded on a Matrigel layer and allowed to invade for 4 h. G Representative images of LAMP1, F-actin, and PH4B staining of SCC invasive protrusions. H Quantification of invasive protrusion assay in control vs. CBX3-overexpressing SCC cells (left). Central line shows mean, whiskers indicate ± SEM. n = 6 images. Representative images are shown (right). I H&E and immunofluorescence staining of CBX3 (green) and E-cadherin (red) in invasive and noninvasive human SCC samples. J Quantification of CBX3 levels in invasive versus noninvasive human SCC samples. n = 18–25 images from 5 to 7 biological replicates. K Schematic model of CBX3-dependent protrusion maturation in EPCs and SCC cells, including the proposed molecular mechanism. For in vitro data, graphs represent one of two independent experiments. Box plots show median, IQR and 1.5× IQR. Statistical significance was assessed using unpaired two-sided Student’s t test unless otherwise stated. Scale bars: 25 μm (B, D), 12.5 μm (G), 50 μm (H), 100  μm (I).

Given that CBX3 levels are higher in migrating EPCs during wound healing compared with homeostasis, and in deep invasive epidermal areas of SCC respect to superficial areas, and since CBX3 overexpression enhances the invasive capability of SCC cells (Fig. 3B–N), we wondered whether this phenotype results from CBX3-dependent control of protrusion maturation in invading SCC cells. To induce invasive protrusions in SCC cells, we add Matrigel before cell plating on a 3 μm-pore transwell. This setting allowed us to evaluate the invasive capacity of cell protrusions through the extracellular matrix toward nutrient gradients prior to traversal of the transwell membrane pores (Fig. 7F). Indeed, even with a twofold longer induction period, Matrigel reduced protrusion formation to approximately half of that observed under Matrigel-free conditions (Supplementary Fig. 16F). In this setting, we found that the genesis of invasive protrusions, which shares the features with mature protrusions (Fig. 7G), is strongly dependent on Cbx3 (Fig. 7H).

Lastly, we evaluated the in vivo human relevance of our findings by comparing cutaneous SCCs classified by pathologists as invasive or noninvasive SCC. Consistently with our observations in mice and with our results from functional assays, CBX3 levels are higher in invasive SCC (Fig. 7I, J), suggesting that CBX3 might be functional in the maturation of invasive protrusions in human SCC cells. This is in line with previous findings showing that CBX3 mRNA is highly expressed in multiple human carcinoma and that its elevated levels predict poor prognosis in many solid tumors, including SCCs⁴¹.

Thus, our data demonstrate that CBX3 is a key regulator of mature protrusions in EPCs in response to environmental cues, as well as of invasive mature protrusions in SCC cells.

Taking together, our study uncovers a Cbx3-dependent multilayered control of gene expression, including an intracellular mRNA transport from chromatin to cell protrusions, which regulates cell motility through protrusionogenesis, in both EPCs and squamous carcinoma cells (Fig. 7K).

Discussion

To adapt to stress in challenging contexts (such as wound healing, carcinogenic pressure, or the acquisition of a mutational landscape), EpSCs exit homeostasis and undergo CA. Through their plasticity, EpSCs induce cell protrusions that are required to explore the surrounding environment. While the initiation phase of cell protrusion has been extensively studied^18–20, it is also well established that specific RNAs are transported to cell protrusions for local translation, a process that plays a key role in regulating protein localization within these structures^55,56. However, the heterogeneity, functional composition, and maturation of cell protrusions remain largely undefined. Our study demonstrated that CBX3, previously characterized as chromatin and splicing factor^29,30, exerts an unexpected function in controlling the delivery of specific mRNAs to a subset of cell protrusions. These CBX3-positive protrusions, mainly located at the leading edge of polarized cells, are large and mature outward cell extensions, characterized by high focal adhesion turnover and the presence of multiple membranous organelles. These include the ER, lysosomes, and mitochondria, as well as likely Golgi outposts, endosomes, and secretory granules. Notably, this specific composition (characteristic of mature protrusions) is also observed in invasive protrusions of SCC cells, consistent with our in vivo histological observations in murine and human wound healing and cutaneous carcinomas. Importantly, most of these features are directly under CBX3 control. Specifically, CBX3 governs the transport of mRNAs to cell protrusions, enabling focal adhesion dynamics and the establishment of the ER, lysosomes, and likely other membranous organelles. These findings demonstrate that CBX3 represents a crucial factor to build a mini cytoplasm with the required organelles in cell protrusions.

Considering the known actin-based protrusion types, our data indicate that CBX3 marks a subset of lamellipodia and podosomes-like structures in epidermal cells (2D and 3D culture systems, respectively) as well as invadopodia in SCC cells^16,17,60.

Seeking chromatin factors regulating CA in EPCs, we developed ReGenT-seq and identified Cbx3 as a top hit. Our newly implemented approach builds on classical pooled genetic screens^22,31, but uniquely integrates two parallel sub-screens employing two distinct yet complementary transcriptional readout systems. Unlike previous functional genetic screens that relied on a single transcriptional readout system with a drug-based selection⁶¹, ReGenT-seq uses both survival and death reporter genes. Reciprocal integration of results from these two reporters (Fig. 1B) enabled highly stringent detection of gene hits. Our data further suggest that integrating two closely related and complementary pooled sub-screens, as implemented in ReGenT-seq, improves bona fide hit detection while minimizing false positives. As with genetic screens relying on a single-gene readout, a potential limitation of our approach is that the transcriptional activity of the Krt6b promoter may not be exclusively indicative of stress. However, under culture conditions, its induction is most consistent with, and likely reflects, a stress-associated response, despite reports of in vivo expression in other differentiated epithelial cell states.

Before focusing on Cbx3, we also examined transcriptomic alterations following Ehmt2 and Kmt5c KDs. These three screening hits are chromatin factors with transcriptional repressive functions involving the methylation of K9 in Histone 3 and K20 in Histone 4^24,33,35. Nevertheless, ReGenT-seq identified Cbx3 and Kmt5c as pro-CA factors, while Ehmt2 as anti-CA. Interestingly, in comparison with the two pro-CA factors, the anti-CA EHMT2 exhibited the strongest repressive activity toward repetitive DNA sequence expression, including transposable elements, as well as illegitimate lineage-associated gene expression. Our data complement recent studies highlighting the importance of endogenous retrovirus silencing for EpSC activity during tissue repair, as well as the role of transcriptional deregulation in lineage identity^14,62. Together, these findings suggest that an ideal pro-CA chromatin factor maintains a finely tuned balance of transcriptional derepression, enabling simultaneous expression of non-epithelial lineage genes (lineage infidelity) while preventing uncontrolled, widespread activation of repetitive DNA elements. This balance appears essential for EPCs to successfully engage their intrinsic plasticity programs.

A growing, yet still limited, number of proteins with both nuclear and cytoplasmic functions exert multilayered control over gene expression by simultaneously affecting two distinct regulatory steps, including chromatin accessibility, transcription, RNA splicing and RNA translation (for example, in Sofiadis et al.⁶³). However, very few factors are known to impact nearly all these regulatory layers while also controlling RNA subcellular distribution. This unique property has been described for proteins, such as TDP-43 and FMRP proteins, which are well characterized in neurons, where they mediate the transport of target mRNAs to distal axonal compartments^64–66. In this specific cellular context these proteins play crucial roles, including regulation of axonogenesis and axon growth, and their dysregulation is implicated in multiple neurological diseases^67–69. In our study, we identified and characterized CBX3 as a protein with comparable multifunctional features in the context of epithelial cells, where it supports the establishment of cell protrusions. Accordingly, our data reveal a functional analogy between the molecular mechanism underlying axonogenesis in neurons and protrusionogenesis in non-neuronal cells. This conserved regulatory logic operates not only in cancer cells during invasion, but also in epidermal stem/progenitor cells during wound healing, highlighting protrusionogenesis as a fundamental, CBX3-dependent cellular program. Our in-depth analysis further delineated the specific impact of CBX3 on multiple layers of gene regulation, particularly influencing genes functionally associated with cell protrusion maturation. Specifically, CBX3 directly regulates focal adhesion-related genes at multiple levels of gene expression control, including mRNA subcellular localization to cell protrusions. In contrast, genes linked to cell proliferation are directly regulated by CBX3 primarily at the levels of RNA splicing and nuclear export. Notably, CBX3 exerts a particularly strong influence on genes involved in ER establishment within cell protrusions by controlling their mRNA subcellular distribution.

It is intriguing to consider why a single gene evolved the ability to control multiple layers of gene expression, rather than distributing this complexity across multiple regulators⁷⁰. Our findings suggest that critical stress responses might rely on conserved single-gene regulators, while distributed control is suited for more complex regulated processes, emphasizing flexibility over precision. In addition, for processes critical to survival, centralizing control in one master regulator could ensure rapid, synchronized responses without the need to coordinate multiple regulators.

Humans and mice possess three HP1 paralogs encoded by the CBX5, CBX1, and CBX3 genes. While HP1 proteins were originally identified in the context of heterochromatin, as denoted by their name, it is known that this family of proteins exerts a broad range of functions^42,71. Our analysis reveals a previously unrecognized extranuclear function of CBX3 in directing the localization of specific mRNAs to cell protrusions. Intriguingly, CBX3 and CBX5 genomic loci are highly conserved in humans and mice. Both genes share bidirectional promoters with the heterogeneous nuclear ribonucleoproteins (hnRNPs) HNRPA2B1 and HNRPA1, respectively, suggesting a functional link. The hnRNPs assist the maturation of pre-mRNAs and stabilize the mRNA during its nuclear export⁷². They can both be found in cytoplasmic stress granules^73,74, non-membrane-bound cellular compartments, often referred to as membraneless organelles, induced under various stress conditions. Here, they function to protect mRNAs, proteins, and stalled translation complexes but also to promote the translation of specific proteins, mitigating cell stress by limiting global cellular translation^75–77. However, the role of stress granules in translational control remains unclear, although recent evidence suggests that they preferentially prioritize stress-responsive mRNAs⁷⁸. In this scenario, it is tempting to speculate that CBX3 may participate in RNA granule-associated mechanisms involved in cell protrusion establishment.

In addition to CBX3-dependent mRNA delivery to protrusions, we identified CBX3-independent localization of mRNAs encoding chromatin- and splicing-related protein complexes (Supplementary Fig. 13F, G). This result suggests at least two potential hypotheses. Cell protrusions could represent a hub for the storage of mRNA, thereby preventing premature translation (as previously suggested by Mardakheh et al.⁵⁶) and enabling the rapid availability of these mRNAs for translation in the cytoplasm under certain conditions. Indeed, these proteins are not found by mass spectrometry (Supplementary Fig. 13A). Alternatively, these mRNAs may exert local functions or serve a structural role in their RNA form⁷⁹.

Initially, it was surprising to discover that CBX3 primarily regulates the delivery of mRNAs to cell protrusions that are not enriched in these compartments. Indeed, the mRNAs required for specific organelles (including the ER, lysosomes, the Golgi apparatus and secretory granules) are enriched in the cytoplasm but depleted in cell protrusions where they are present at low levels. This apparent paradox can be explained by considering that the limited availability of these mRNAs in cell protrusion represents a rate-limiting factor for cell protrusion maturation, through local translation, as these mRNAs encode proteins critical for the organelles. This agrees with observations from Mardakheh et al.⁵⁶. Intriguingly, the mRNAs related to organelles involved in energy metabolism, such as mitochondria and peroxisomes, are not regulated by CBX3. This suggests that CBX3 does not control local energy metabolism and that these organelles may instead be required during early stages of protrusion formation, being already localized to developing protrusions in preparation for subsequent maturation steps.

At first glance, it may appear counterintuitive that a protein, such as CBX3, originally identified as a functional component of heterochromatin, would also be involved in RNA processing and transport. This apparent discrepancy became more coherent when we considered our findings from an evolutionary perspective, particularly the idea that some modern chromatin repressors may have originated from ancestral defense mechanisms against transposable elements. Mechanisms that initially evolved to silence transposable elements may have subsequently been co-opted to fulfil broader regulatory functions, including gene repression during development, differentiation, and stress responses. Such an evolutionary trajectory, shaped by long-term host–parasite co-evolution, likely drove the emergence of increasingly sophisticated layers of epigenetic regulation. Indeed, from genomics studies, we know that many genes located near transposable elements are subject to epigenetic regulation, hinting that they might have served as loci for recruiting silencing machinery that was later expanded to neighboring regions. CBX3 evolution may fit within this hypothetical scenario. Indeed, the co-transcriptional silencing mechanism to repress transposable elements requires the ability to somehow interact nascent RNAs and also contribute towards the establishment of a heterochromatin state⁸⁰, two features that characterize CBX3. Related to this, we found that CBX3 KD leads to increased expression RNAs containing transposable elements, whereas CBX3 binds these transcripts and mainly retains them in the nucleus. The identification of CBX3-dependent RNA transport from nuclear export to cell protrusion is also interesting in the context of recent findings on protein complexes responsible for transposable elements silencing and RNA nuclear export that share some proteins⁸¹.

Taken together, our work revealed a multi-layered control of gene expression to regulate cell motility, and in particular a chromatin-to-protrusion mRNA delivery that represents a potential therapeutic target in regenerative medicine, to enhance tissue repair, as well as in oncology to tackle cancer cell invasion and metastasis.

Methods

Plasmids

The pLMN-shRNA library targeting 615 chromatin factors³¹ used in ReGent-Seq and Viability screens was kindly supplied by Professor Johannes Zuber (Research Institute of Molecular Pathology (IMP), Vienna BioCenter (VBC), Vienna, Austria). The two reporter systems (K6-Blast resistance gene and K6-DTR gene) for ReGenT-seq were generated in the pLenti CMV Hygro DEST vector (w117-1) (Addgene 17454). Krt6b promoter region (chr15:101,510,655-101,512,725) was amplified from EPCs genomic DNA (prim 1, prim 2). Blasticidin resistance mRNA sequence was amplified from the plentiCas9-BLAST vector (Addgene 52962), while DTR (diphtheria toxin receptor) sequence from pATF5-DTR-GFP vector⁸², a gift from Professor Michael R. Green (prim 3 and 4 for blasticidin resistance, prim 5 and 6 for DTR). Retroviral vectors expressing individual shRNAs for validation of the ReGenT-seq hits were generated in the pMSCV backbone⁸³. 97-mer oligonucleotides specific for each shRNAs (prim 101–108) were PCR amplified (prim 109–110). To generate the plasmid overexpressing CBX3 (pLenti-Cbx3-Blast) we used pLenti-Cas9 Blast vector (Addgene 52962) in which Cas9 sequence was removed, we extracted RNA from EPCs by Rneasy mini kit (Qiagen), and retrotranscribed it into cDNA through iSCRIPT cDNA Synthesis Kit (BioRad). cDNA was amplified using prim 99–100. For each cloning, PCR was performed using the high-fidelity AccuPrime (Thermo) Taq. Lentiviral vectors expressing individual shRNAs for CBX3 KD in human cells were generated in the pLKO.1 puro backbone (pLKO.1 puro scramble, shRNA1 and 2) (prim 123–126). All new shRNAs were designed using the Design Hairpins tool of the BROAD institute (https://portals.broadinstitute.org/gpp/public/seq/search). Digestions of vectors and PCR products were performed at 37° for 4 h. The digested vector was ligated to the digested PCR product using T4 DNA ligase (NEB). DNA was mixed with Turbo Electrocompetent cells (NEB) or calcium competent Escherichia Coli. Transformed bacteria were pre-grown at 37 °C for 1 h, then plated on ampicillin plates and incubated at 37 °C, O/N. Plasmids were purified using NucleoBond Xtra Midi (Macherey Nagel). All cloned sequences were checked by Sanger sequencing (Eurofins) using appropriate primers. shRNA plasmids to knockdown c-Jun and validate the ReGenT-seq reporter system were purchased from Horizon Discovery (shRNA1, TRCN0000055205; shRNA2, TRCN0000042693). shRNA scramble (TRC lentiviral pLKO.1 puro Empty Vector control, ID:RHS4080) was used as a control. Oligos for cloning are listed in Supplementary Data 7.

Cell culture, lentiviral, and retroviral production and transduction

EPCs were isolated from the dorsal skin of adult mice as previously described⁸⁴. SCC cells were isolated from a murine SCC13. Briefly, after collection, samples were chopped in 1mm2 pieces and incubated 1 h at 37 °C in a digestion mix: 5 mg pronase (Merck Millipore) (in 250 μL H₂O, 2.5 μl of 1 M Tris buffer (pH 8), 0.5 μl of 0.5 M EDTA (pH 8.0) and 5 mg Collagenase D (Merck Millipore), dissolved in 2 ml complete medium. After incubation, the tissue was mechanically dissociated and filtered through a 70 μm cell strainer. Isolated EPCs and SCC were plated on confluent mitomycin C-treated J2 feeder layers. Co-cultures of EPCs and J2 were cultured in high glucose, glutamine-free, calcium-free DMEM medium supplemented with 10%FBS, penicillin and streptomycin (50 μg ml⁻¹), glutamine (2 mM), sodium pyruvate (1 mM), hydrocortisone (0.5 μg ml⁻¹), insulin (5 μg ml⁻¹), cholera toxin (8.4 ng ml⁻¹), epidermal growth factor (10 ng ml⁻¹). Versene was employed to remove J2 feeder cells. J2 cells were provided by Prof. Fiona Watt and cultured in high-glucose DMEM medium supplemented with 10% FBS, penicillin and streptomycin (50 μg ml⁻¹), glutamine (2 mM). EPCs adapted to grow in J2 absence were used in ReGenT-seq experiments and the following experiments. HaCaT were cultured in high-glucose DMEM supplemented with 10% FBS, penicillin and streptomycin (50 μg ml⁻¹), glutamine (2 mM). Primary vulvar epidermal cells were isolated from a patient undergoing surgery. Ethical approval protocol #0002319 was obtained from the Comitato Etico Territoriale (CET) Interaziendale AOU Città della Salute e della Scienza di Torino. The tissue was disinfected in Betadine and subsequently in 70% ethanol and then washed in PBS supplemented with penicillin and streptomycin. The tissue was minced into small pieces (approximately 2–3 mm). The pieces were then transferred to a culture plate and cultured in FAD medium (1:1 [v/v] F-12 [Ham]:DMEM, 5% FBS, 0.4 μg/ml hydrocortisone, 5 μg/ml insulin, 8.4 ng/ml cholera toxin, 10 ng/ml EGF, 24 μg/ml adenine, 100 U/ml penicillin, and 100 μg/ml streptomycin). Non-adherent tissue fragments were removed, and the FAD medium was replaced. Colonies were allowed to grow for several days. Fibroblasts were removed using 0.05% trypsin–EDTA for 3 min. Keratinocytes were then detached with trypsin at 37 °C for 15 min. After the first passage, the ROCK inhibitor (Y‑27632, Sigma-Aldrich) was added to the FAD medium for keratinocyte growth. Experiments were performed at passage three. Lentiviral and retroviral virions were produced in Lenti-X 293T and Retro-X 293T (GP2-293) packaging cells, respectively, in high-glucose DMEM supplemented with 10% FBS, penicillin and streptomycin (50 μg ml⁻¹), glutamine (2 mM). Packaging cells transfection was performed using Attractene Transfection Reagent (Invitrogen). Lenti-X 293T cells were co-transfected with transfer and packaging plasmids (pxPAX2 plasmid, Addgene 12260 and pMD2.g plasmid, Addgene 12259). Similarly, RetroX 293T cells were used to produce the retroviral library by co-transfection with the VSV-G plasmid (Addgene 14888). Viral supernatants were 0.45-μm filtered and used to transduce EPCs in the presence of polybrene. Efficiency of transduction was assessed by Flow Cytometry.

Epidermal differentiation was achieved by disaggregating murine EPCs and placing them in suspension as previously described⁸⁵. Gene expression analysis was performed by RT-qPCR to assess the expression of epidermal stem/progenitor cell markers and epidermal differentiation markers.

Murine and human skin samples

Full-thickness wounds were performed with a circular 2 mm biopsy punch in the tail skin of B6CBAF1/J 6–8 weeks old mice as previously described¹³. Acute and chronic (venous and pyoderma gangrenosum ulcers) wound samples were confirmed by a pathologist for their diagnosis. The UVB irradiation protocol to induce SCCs was performed as previously described¹³. Briefly, the DMBA-UVB two-stage-induced carcinogenesis protocol was used. Mouse skin was treated with 50 μL of 120 μg/mL of DMBA dissolved in acetone. Subsequently, UVB irradiation (180 mJ/cm²) was started and continued for 3 times a week till the end of the experiments when SCC samples were collected. Maintenance, care and experimental procedures have been approved by the Italian Ministry of Health, in accordance with Italian legislation (authorization no. 427/2023-PR). Cutaneous human SCC samples were categorized by pathologists as invasive or noninvasive prior to histological analysis. The institutional review board of the Hokkaido University Graduate School of Medicine approved the human study described (ID: 13-043, 14-063, and 15-029). The study was carried out according to the Declaration of Helsinki Principles. The participants provided written informed consent.

RNA extraction and quantitative RT-PCR analysis

RNA was prepared using the RNeasy Mini Kit (Qiagen). cDNA synthesis was performed using SuperScript II Reverse Transcriptase (Thermo). cDNA synthesis was performed using the iScript cDNA Synthesis Kit (BioRad). Real-time qPCR was performed on ABI 7900HT (Applied Biosystems) with Platinum SYBR Green qPCR SuperMix-UDG w/ROX (Thermo). For quantification analysis, the comparative Threshold Cycle (Ct) method was used, and gene expression levels were normalized to GAPDH or ACTB. Relative mRNA quantification was performed using the 2^−ΔΔCt method. Oligos for RT-PCR are listed in Supplementary Data 7.

Nucleus/cytoplasm fractionation

For nucleus and cytoplasm fraction isolation, freshly harvested cells were lysed in a hypotonic Cytoplasmatic buffer on ice for 20 min (10 mM HEPES pH 7.9, 60 mM KCl, 1 mM DTT, 1 mM EDTA, 0.075% NP40, 1:100 Protease-inhibitor Mix M (Serva), and 1:100 RNAse inhibitor (Thermo) and centrifuged (16,000 × g, for 5 min at 4 °C). The supernatant was retained as a cytoplasmic fraction. The nuclei pellet was washed in a Washing Buffer (10 mM HEPES pH 7.9, 60 mM KCl, 1 mM DTT, 1 mM EDTA, 1:100 Protease-inhibitor Mix M, and 1:100 RNAse inhibitor), centrifuged (16,000 × g for 15 min at 4 °C) and then resuspended in a hypertonic nuclear buffer (20 mM Tris pH 8.0, 1.5 mM MgCl₂, 420 mM NaCl, 0.2 mM EDTA, 25% glycerol and 1:100 Protease-inhibitor Mix M). Pellet was dispersed with a pipette and passed through a 27 G needle 5 times. After 10 min-incubation on ice, nuclei were centrifuged (16,000 × g for 15 min at 4 °C) to collect them.

Protein extraction and Western blotting

For whole-cell extracts, cells were lysed on ice in Lysis Buffer (50 mM Tris 1 M HCl pH 7.5, 100 mM NaCl, 1% NP40, 0.1% SDS, 0.5% DOC) with protease and RNAse inhibitors and insoluble proteins were separated via centrifugation (16,000 × g for 20 min at 4 °C). For cytoplasmic and nuclear protein extraction (See “Nucleus/cytoplasm fractionation”). Protein amounts were quantified using Pierce BCA Protein Assay Kit (Thermo). Equivalent quantities of solubilized proteins were resolved by SDS-PAGE in 4–20% Novex Tris-Glycine Mini Protein Gels, 4–12% (Thermo) and transferred to nitrocellulose membranes iBlot Transfer Stack (Invitrogen) using the iBlot 3 Western Blot Transfer System (Thermo). Membranes were blocked with 5% non-fat milk in PBS supplemented with 0.1% Tween-20 (PBS-T) for 1 h and incubated with the primary antibodies O/N. Membranes were washed with TBST and then incubated with secondary antibodies for 1 h. The membranes were washed with PBS-T, and then blots were visualized using Clarity Western ECL (Bio-Rad). Protein bands were detected using the ChemiDoc Touch Imaging System (Bio-Rad). The primary antibodies used for Western Blotting are listed in Supplementary Data 8.

Validation of ReGenT-seq reporter system

EPCs expressing the K6-Blast reporter were transduced with the c-Jun shRNA1 and 2, and the scramble shRNA, all in pLKO.1 puro vector. After the transduction, cells were treated with Puromycin (25 μg ml⁻¹, Thermo Fisher Scientific) for 5 days. Cells expressing shRNAs were split in the treated group, selected by blasticidin antibiotic (30 μg ml⁻¹), and the Untreated group, not selected. All groups were cultured for 3 weeks, maintaining Puromycin selection. For each shRNA, the viability ratio was evaluated at 2 and 3 weeks. In detail, at t0 (n week-2day) we plated 10,000 cells expressing the shRNA in 6 replicates, for each group. At t1 (n week), we counted the number of cells in the population by flow cytometry. Then, we used the viability rate (t1/t0) to calculate the viability ratio (treated group/untreated group viability rate).

ReGenT-seq

To ensure library representation (500 copies for each shRNA) and a single retroviral integration, a total of 30 × 10⁶ of EPCs expressing the K6 CA reporter systems (K6-Blast or K6-DTR) were transduced with 5% efficiency in triplicate. Since the shRNA library vector carries Neomycin resistance and GFP, 48 h after infection, cells were treated with G418 antibiotic (30 mg ml⁻¹) for 10 days and analyzed by flow cytometry for GFP expression. Based on the percentage of GFP-positive cells, an appropriate number of cells were maintained at each passage to preserve library representation. At t = 0 (day 11 after infection), replicates of G418-selected cells were split into the treated group, treated with Blasticidin antibiotic (30 μg ml⁻¹) or DT (0.05 ng ml⁻¹), and the Untreated group. t = 3w (time 3 weeks) and t = 4w (time 4 weeks) triplicates were collected and stored for genomic DNA extraction (gDNA). DNA-sequencing libraries were generated by performing two consecutive PCR. For each sample, gDNA from at least 5 × 10⁶ cells were used as template in parallel multiple PCR-1 reactions to amplify shRNA cassette. Each reaction containing gDNA, 2× XtraRTL Master Mix (GeneSpin), primers (Prim 7 or 8 or 9 as forward, Prim 10 as reverse) and 0.75 U AccuPrime (Thermo), were run using the following cycling parameters: 98 °C for 5 min; 28 cycles of 98 °C for 45 s, 59 °C for 45 s and 72 °C for 45 s; 72 °C for 5 min. The PCR-1 products were mixed and purified using AMPure XP beads (Beckman Coulter). PCR-1 was performed in duplicate. For PCR-2, the purified PCR-1 products of each replicate were used as a template in two parallel PCR reactions to be tagged with standard Illumina adapter (p7, p5). Each reaction containing DNA, 2× XtraRTL Master Mix (GeneSpin), primers (Prim 12–86 and Prim 11), were run using the following cycling parameters: 98 °C for 5 min; 28 cycles of 98 °C for 45 s, 59 °C for 45 s and 72 °C for 45 s; 72 °C for 5 min. For both PCR-2, multiple PCR-1 products were combined. PCRs primer sequences are in Supplementary Data 7. DNA of each sample were pooled together and purified on 0.6% agarose gel (MinElute gel extraction kit, Qiagen). The final library, consisting of both PCR-2 replicates, was quantified on a Bioanalyzer (Agilent). DNA libraries were sequenced on an Illumina NextSeq 500 platform, according to the manufacturer’s guidelines (Illumina). To detect shRNA enrichment or depletion, we acquired >500 reads per the 5049 shRNAs in the pool.

Data processing: following quality controls (performed with FastQC), sequencing reads were trimmed of the common portion of the vector and upstream PCR nucleotides using Cutadapt v2.6 (options: -e 0.2 -a TGCCTACTGCCTCGGACTTCAAGGGGCTAGAATTC -g TAGTGAAGCCACAGATGTA -m 22 -M 22 -n 2 --discard-untrimmed). The extracted hairpin seed sequences (22 nt) were next aligned to the reference shRNA library (i.e., the PCR amplified common portion of the hairpin up and downstream of the seed + the 22 seed nucleotides) using Bowtie v1.2.3, allowing at most 2 mismatches, discarding reads with multiple alignments (options: -a --best --strata -v 2 -m 1 -S -q), and subsequently filtering alignments to allow at most 1 nucleotide shift in the alignment position of the seed (possibly resulting from indels in the PCR product). Counts per hairpin were retrieved running Samtools (v1.9) idxstats utility on BAM files. Downstream analysis: hairpin counts were next analyzed using the edgeR package. Lowly detected hairpins (below 1 CPM in 3 or more samples) were filtered out, and normalization factors were calculated using the trimmed-mean of M-values (TMM) method (implemented in the calcNormFactors function). PCA was performed using the prcomp R function (parameters: scale = TRUE, center = TRUE), using the top 2500 variable hairpins. Following dispersion estimation (estimateDisp function, robust = TRUE), differential hairpin representation between the reciprocal ReGenT-seq selections and their respective controls (treated vs. untreated group) was obtained by fitting a generalized linear model (GLM) to all sample groups (glmFit function) and performing a likelihood-ratio test (glmLRT function), using time as a blocking variable to identify the hairpin consistently differentially represented across the two collection time points. Hairpins with FDR < 0.05 and |logFC | > 1 were considered statistically significant. To assess the effect of multiple hairpins targeting the same gene and obtain gene isoform-level results, hairpin sequences were used to query BLAST, in order retrieve their specific targeted isoform (blastn utility, run on the mouse RNA database, excluding gene models and selecting the top matching alignments with at least 21 nt match). Next, hairpin targeting the same isoform were grouped in sets, and the isoform-level effect was tested using ROAST (mroast function, as implemented in the edgeR package). Finally, for each isoform, an effect ranking score was calculated from ROAST results as: score = −log10(P-value) × (N.hairpins up − N.hairpins down). Gene isoforms with P < 0.05 and a score greater or lower than 0 were considered as significantly enriched or depleted hits, respectively. Final selection of CA regulators was obtained by crossing ranks in opposite directions for the reciprocal treatments.

Viability RNAi screening

30 × 10⁶ of EPCs were transduced with low efficiency (5%) in triplicate. Forty-eight hours after infection, cells were treated with G418 antibiotic (30 mg ml⁻¹) for 10 days. t = 0 (day 11 after infection), t = 2w (time 2 weeks) and t = 3w (time 3 weeks) were collected in triplicate and stored for genomic DNA extraction (gDNA). See “ReGenT-seq” for library preparation. Genes regulating viability were identified by detecting shRNA enrichment or depletion in t = 2w and t = 3w, comparing t = 0w, acquiring >500 reads per all 5049 shRNAs in the pool.

The same analysis procedure described for the ReGenT-seq, at hairpin- and gene-level, was employed to compare the two time points against the starting time point (2-3w vs 0). PPI network for the joint set of ReGenT-seq and viability screening gene hits was retrieved from the STRING database (https://string-db.org/), considering physical interactions only. Network centrality measures, graph-level and node-level descriptive statistics calculations and network visualization were performed with the igraph R package. Correlations between network centrality measures and CA and viability scores were assessed using the cor.test R function. Significance of network metrics correlations with CA and plasticity scores was assessed by bootstrap analysis, generating 1000 down-sampled network realizations for each set of regulators. The significance of the observed correlation difference in network centrality measures between CA and plasticity scores was assessed by a permutation test, generating, for each of the tested metrics, 1000 random network realizations by node permutation, and calculating the correlation difference on each random realization to create a null distribution for the estimation of an empirical P-value. Metrics with P < 0.05 were considered as statistically significant.

Validation of the ReGenT-seq hits

EPCs expressing one of the two reporter systems (K6-blasticidine-resistance or K6-diphtheria-toxin-receptor) were transduced with individual shRNA targeting the screening hits or scramble shRNAs. Thirty-six hours after the transduction, cells were treated with G418 antibiotic (30 mg ml⁻¹). Cells expressing shRNAs were split in the treated group selected by blasticidin antibiotic (30 μg ml⁻¹), or by DT (0.05 ng ml⁻¹), and the Untreated group not selected. For each shRNA, the viability ratio was evaluated at 2 and 3 weeks. In detail, at t1 (n week-2day), we plated 10,000 cells expressing the shRNA in 6 replicates, for each group. At t2 (n week), we counted the number of cells in the population by flow cytometry. Then, we calculated the viability score by normalizing the viability rate (t2/t1) to the maximum value obtained for each shRNA.

RNA-seq, library preparation, and analysis

Mini bulk RNA-seq was performed as previously described¹³. EPCs expressing shRNA or control vectors were sorted for GFP in Lysis Buffer (0.5% Triton X-100), in the presence of RNase inhibitors. Then, a biotinylated Oligo(dT) was bound to Dynabeads MyOne Streptavidin T1 beads (Thermo) and used to isolate mRNA from the cell lysate. Reverse transcription was performed with SuperScript II Reverse Transcriptase (Thermo). The resulting cDNA was amplified by PCR, and libraries were prepared for sequencing with the Illumina NexteraXT Illumina protocol (Illumina). Library preparation batch correction was performed using ComBat-seq⁸⁶, using the Combat_seq function from the sva R package. Gene expression counts were next analyzed using the edgeR package. Lowly expressed genes (below 1 CPM in at least 5 samples) were filtered out, and normalization factors were calculated using the trimmed-mean of M-values (TMM) method (implemented in the calcNormFactors function). PCA was performed using the prcomp R function (parameters: scale = TRUE, center = TRUE), using log-normalized expression counts as inputs. Differential expression analysis was performed using the edgeR GLM testing framework, comparing KD versus control samples (formula: ~0 + condition) by quasi-likelihood F-test (QLF). Genes with FDR < 0.05 and |log2FC| > 0.5 were considered as significantly differentially expressed. Heatmaps were generated using the ComplexHeatmap R package. For repetitive elements expression analysis, reads were realigned using STAR with options: --outFilterMultimapNmax 100 and --winAnchorMultimapNmax 100, to allow for highly multi-mapping reads inclusion. Repeats expression levels were next quantified using TEtranscripts v2.2.3 (TEcount utility with options: --format BAM --stranded no --sortByPos), and testing for differential expression the repeats-uncuding transcriptome using DESeq2.

For whole-cell bulk RNA-seq, RNA was extracted from 5 × 10⁶ EPCs, or nuclear-, cytoplasmatic-, protrusion-, fraction using the RNeasy Mini Kit (Qiagen). For library preparation, the quantity and quality of the starting RNA were checked by Qubit and Bioanalyzer (Agilent). One microgram of total RNA was subjected to poly(A) selection, and libraries were prepared using the Illumina TruSeq RNA Sample Prep Kit (Illumina) or Illumina stranded mRNA (Illumina), following the manufacturer’s instructions. Libraries were sequenced on Illumina NextSeq 500 or 1000 System (single-end 75 bp reads). Downstream analysis: following quality controls, sequencing reads were aligned to the mouse reference genome (mm10/GRCm38.p6) using STAR v2.7.1a with options: --outFilterMultimapNmax 10 --outFilterMultimapScoreRange 1 --outFilterMismatchNmax 999 --outFilterMismatchNoverLmax 0.04. Gene expression levels were quantified with featureCounts v1.6.1 (options: -t exon -g gene_name) using the GENCODE Release M23 annotation. Multi-mapped reads and reads mapping to a comprehensive annotation of ribosomal RNAs (obtained from repeatMasker and GENCODE) were excluded from quantification. Signal tracks were generated using the deepTools BamCoverage utility (parameter: --normalizeUsing CPM --binSize 1), keeping uniquely mapped reads only in the input BAM file.

Downstream analysis for ASE: differential ASE between CBX3 KD and control samples were identified using rMATS-turbo (rmats.py script with parameters: --novelSS -t paired --libType fr-firststrand --readLength 122 --variable-read-length), retaining uniquely mapped reads only in the input BAM files (command: samtools view). CBX3-dependent splicing noise analysis was performed as previously described in 30 with slight modifications. In brief, replicate BAM files for each condition were first pooled to extract and annotate all splice junctions using RegTools v1.0.0 (regtools junctions extract and regtools junctions annotate commands with default parameters). Junctions spanning more than 20 kb were filtered out as possibly originating from misalignments during reads mapping. Condition-specific junctions were then obtained by removing junctions commonly present in both KD and control samples and reassigned to replicates per condition to obtain sample-level estimates. Gene-level differential junction analysis between CBX3 KD and control samples was next carried out by (i) counting the number of unique junctions per gene per sample, (ii) filtering low-count genes, and performing Poisson regression on junction counts for each gene (glm R function, with family = “poisson” argument). Genes with P < 0.05 were considered statistically significant. Enrichment analysis for the list of splicing-regulated genes was performed using the EnrichR package. Nucleus/cytoplasm RNA-seq was performed as whole-cell bulk RNA-seq after cell fractionation. Subcellular gene expression concentration estimates were next analyzed using the edgeR package. Lowly expressed genes (below 1 CPM in at least 3 samples) were filtered out, and normalization factors were calculated using the trimmed-mean of M-values (TMM) method (implemented in the calcNormFactors function). PCA was performed using the prcomp R function (parameters: scale = TRUE, center = TRUE), using log-normalized expression counts as inputs. Differential nuclear export analysis between KD and control samples was performed using the edgeR GLM testing framework, comparing subcellular concentration ratios (i.e., nucleus to cytoplasm) between KD and control samples (i.e., testing the coefficients (expr_in_compartment_A_KD)-(expr_in_compartment_B_KD)-(expr_in_compartment_A_WT)-(expr_in_compartment_B_WT), thus obtaining a fold-change of ratios) with quasi-likelihood F-test (QLF). Genes with FDR < 0.05 were considered significantly differentially expressed.

For the comparative analysis between in vitro-cultured EPCs and in vivo transcriptomic datasets, the same data processing and differential expression analysis pipeline described above was employed. Shared gene signatures were defined by comparing in vitro EPCs, SCC stem cells, and wound-induced EpSCs to a common homeostatic stem cell reference derived from epidermal and hair follicle stem cells.

ChIP-seq

ChIP-seq was performed as previously described⁸⁷ with some modifications. Briefly, 10 × 10⁶ EPCs were cross-linked with 1% formaldehyde and then washed with PBS after quenching the reaction with glycine. Cell pellets were resuspended in Farnham Lysis Buffer (5 mM HEPES, pH 7.9, 85 mM KCl, 0.5 NP-40) with protease and RNAse inhibitors, incubated on ice and centrifuged. Sonication Buffer (10 mM Tris pH 8.0, 1 mM EDTA, 1% SDS) with protease and RNAse inhibitors was added to resuspend pellet. Shearing of DNA was performed by the Bioruptor Plus sonicator device (Diagenode). Sonicated chromatin was centrifuged, and the supernatant incubated O/N with CBX3 antibody (Sigma-Aldrich, clone 42s2, 05-690), previously coupled with Dynabeads Protein G (Thermo), in IP Buffer 56.25 mM HEPES pH 7.9, 157.5 M NaCl, 1 mM EDTA, 1.125% Triton X-100, 0.1% Na-deoxycholate). After five consecutive washes with LiCl Buffer (100 mM Tris pH 8.0, 100 mM Tris pH 8.0, 1% NP-40, 1% Na-deoxycholate) and one with TE Buffer (10 mM Tris, 1 mM EDTA), the material was resuspended in Elution Buffer (1 % SDS, 0.1 M NaHCO₃ sodium hydrogen carbonate) to be reverse crosslinked O/N, treated with proteinase K and RNase A. After DNA purification, sequencing libraries were prepared with TruSeq ChIP Library Preparation Kit (Illumina) and sequenced on Nextseq 1000 platform. Data processing: Following quality controls (performed with FastQC v0.11.2), paired-end sequencing were processed with Trim Galore! v0.5.0 to perform quality and adapter trimming (parameters: --stringency 3 (--paired for paired-end)). Trimmed reads were next aligned to the mouse reference genome (mm10/GRCm38) using Bowtie v2.3.5.1 [Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359 (2012)]. Duplicated alignments (identified by Picard MarkDuplicates, https://broadinstitute.github.io/picard) and low-quality alignments/multi-mapping reads were excluded using SAMtools v1.6. Immunoprecipitation and corresponding control (Input DNA) datasets were treated identically. Peak calling was performed for each IP against its matched Input DNA using MACS v2.1.4 (command: callpeak -g mm --nomodel –extsize 200 -q 0.05 for single-end). Input-normalized ChIP-seq fold-enrichment signal tracks were obtained using MACS v2.1.4 (command: callpeak -g mm --nomodel -f BAMPE -q 0.05 --SPMR -B and bdgcmp -m FE). For CBX3 ChIP-seq, broad peak calling was performed using epic2 (command: --genome mm10 --false-discovery-rate-cutoff 0.05 --guess-bampe). Signal summary scores in peaks and gene bodies and respective heatmaps were generated with the deepTools multiBigWigSummary, computeMatrix and plotHeatmap utilities, using input-normalized ChIP-seq fold-enrichment signal tracks. Annotation of peaks to genes and genomic regions was obtained using the HOMER annotatePeaks.pl utility.

seCLIP-seq

seCLIP-seq was performed using the RBP-eCLIP kit of ECLIPSEBIO (Cat.n.143637) with minor modifications. Briefly, 10 × 10⁶ of EPCs were washed two times with cold PBS and crosslinked at 254 nm UV with an energy setting of 400 mJoules/cm² using a UVP Crosslinker CL-3000 (Analytic Jena). While keeping the cells on ice, cells were scraped, transferred to a 15 mL falcon and centrifuged at 200 × g for 5 min at 4 °C. The supernatant was removed, and cell pellets were snap-frozen. Cell pellets were then subjected to a fractionation protocol as follows. Pellets were gently resuspended in two volumes of hypotonic buffer (Hepes 10 mM, KCl 10 mM, 1.5 mM MgCl₂, 0.1% NP40, supplemented with 1:100 protease inhibitors (Serva) and 400 U/mL of RNAse inhibitors (NEB) and incubated 5 min on ice. Samples were centrifuged for 10 min at 1000 × g at 4 °C. Supernatant was discarded (cytoplasmic fraction). Cell pellets were resuspended in 1 mL RBP-eCLIP kit eCLIP Lysis Buffer supplemented with 1:100 protease inhibitors and 400 U/mL of RNAse inhibitors and incubated 10 min on ice. To disrupt DNA before immunoprecipitation, Turbo DNAse (Thermo) was added to a final concentration of 4 U/mL and incubated for 10 min at 37 °C. DNA was additionally sheared by passing through a 26-gauge needle from a 1 ml syringe 5 times. Fractions were centrifuged for 10 min at 1000 × g at 4 °C. Supernatants were tested with fraction-specific markers by western blotting. Nuclear fractions were then subjected to the seCLIP protocol following the manufacturer's instructions. CBX3 antibodies (Sigma-Aldrich, clone 42s2, 05-690) were coupled to Dynabeads Protein G (Thermo) for 1 h at RT for each sample. Nuclear fractions were added to the antibody-bead complexes and rotated O/N at 4 °C. Biotin-labelled CBX3-RNA bands were excised from the nitrocellulose membrane corresponding to low (from 20 to 40 kDa), middle (from 40 to 130 kDa), and high (from 130 to 260 kDa) molecular weight of CBX3 protein/RNA complexes. The extracted RNA was then converted into libraries. High-throughput sequencing of seCLIP libraries was performed on the NextSeq 1000 platform.

Raw sequencing reads were processed using a bioinformatics workflow based on Blue et al.⁸⁸. In brief, following quality controls, the unique molecular identifier (UMI) was retrieved from the start of read sequences (first 10 bp) and appended to the read name using UMI-tools v1.1.4 (umi_tools extract command with options: --random-seed 1 --bc-pattern NNNNNNNNNN). UMI-extracted reads were next subjected to two rounds of trimming using Cutadapt (command: cutadapt --match-read-wildcards --times 1 -e 0.1 -O 1 --quality-cutoff 6 -m 18 -j {params.cores} -a AGATCGGAAGAGCAC -a GATCGGAAGAGCACA -a ATCGGAAGAGCACAC -a TCGGAAGAGCACACG -a CGGAAGAGCACACGT -a GGAAGAGCACACGTC -a GAAGAGCACACGTCT -a AAGAGCACACGTCTG -a AGAGCACACGTCTGA -a GAGCACACGTCTGAA -a AGCACACGTCTGAAC -a GCACACGTCTGAACT -a CACACGTCTGAACTC -a ACACGTCTGAACTCC -a CACGTCTGAACTCCA -a ACGTCTGAACTCCAG -a CGTCTGAACTCCAGT -a GTCTGAACTCCAGTC -a TCTGAACTCCAGTCA -a CTGAACTCCAGTCAC), removing adapters off the 3′ end of each read by using a tiling strategy that segments the InvRil19 adapter and (i) conservatively trims if any overlap (−O 1) is found in the first round, and (ii) trims again by using the same parameters but increasing the required minimum overlap length (−O 5) to ensure adapter dimers are properly trimmed. Trimmed reads were next aligned to the mouse reference genome (mm10/GRCm38.p6, obtained from GENCODE release M23) using STAR v2.7.1a using the following options: --outSAMunmapped Within --outFilterMultimapNmax 1 --outFilterMultimapScoreRange 1 --outSAMattributes All --outSAMtype BAM Unsorted --outFilterType BySJout --outReadsUnmapped Fastx --outFilterScoreMin 10 --outSAMattrRGline ID:foo --outStd Log --alignEndsType EndToEnd --outBAMcompression 10 --outSAMmode Full. Uniquely mapped reads were sorted twice (samtools sort), and PCR duplicates were removed using UMI-tools v1.1.4 (umi_tools dedup command with options: --random-seed 1 --method unique). Non canonical chromosomes and reads mapping to repeats were filtered out (bedtools intersect -s -split -f 0.95 -v, using repeatMasker annotation and requiring 95% overlap), and peak calling for each sample was performed using CLIPper (https://github.com/YeoLab/clipper, options: –species mm10) and the companion post-processing tools available as Docker images from: https://github.com/YeoLab/merge_peak). Peak normalization over size-matched input and merging/compression of the normalized peak set was performed for each sample using the overlap_peakfi_with_bam.pl and compress_l2foldenrpeakfi_for_replicate_overlapping_bedformat.pl Perl scripts. Read-depth normalized signal tracks were generated for each sample from the filtered BAM files using the deepTools BamCoverage utility (parameter: --normalizeUsing CPM --binSize 1). To identify a final set of reproducible binding sites for each band, two distinct approaches were employed: (i) high-stringency reproducible peak call, based on the irreproducible discovery rate (IDR) workflow, using peak entropy (obtained with make_informationcontent_from_peaks.pl Perl script) as ranking metric for IDR analysis (idr tool with options: --input-file-type bed --rank 5 --peak-merge-method max), and using the parse_idr_peaks.pl script for post-processing; (ii) lower-stringency reproducible peak call, based on selecting common peaks fulfilling the enrichment thresholds criteria in both replicates: P ≤ 0.001 and fold enrichment ≥ 2. The set of peaks were merged across bands for each analysis, obtaining the two final sets of CBX3 binding sites. Downstream analysis: annotation of peaks to transcripts and genic regions was performed using RCAS. To remove redundancy arising from peaks annotation to multiple isoforms of the same gene locus, gene isoform expression levels were first estimated from pooled SMI input samples using Salmon v1.10.0, running the salmon quant command with options: -l A. Each peak was then assigned to the top-expressed isoform with the following genic region priority: cds > 3′UTR > 5′UTR > non-coding exons > introns > no features (if no mapping to annotated genes was possible). Motif discovery was carried out using two similar approaches: (i) using the HOMER findMotifsGenome.pl utility in RNA mode (options: -rna -size given); (ii) by overrepresented k-mers analysis. For the latter, hexamer counts were generated from peak sequences and a set of background control regions, obtained by shuffling each peak location 3 times within their respective transcript. Z-scores were then calculated for each hexamer as: Z (k) = (c_k − μ_k)/σ_k, where c_k are the observed kmer counts in peaks, μ_k and σ_k are the mean and standard deviation of background kmer counts, respectively. The distribution of Z-score values was then fitted by a three-component Gaussian mixture model to discriminate between background, enriched and highly-enriched hexamers. For RNA secondary structure analysis of peaks, RNAplfold from the ViennaRNA package v2.6.4 was used to predict local unpairedness probabilities of bound transcripts and a set of control transcripts (options: -W 200 -L 150 -u 12). Control transcript sequences were generated by shuffling hexamers in the true bound transcript sequences, thus preserving their overall k-mer content. For each hexamer, the average unpairedness probabilities distributions at their locations in peaks were compared between true and shuffled transcript sequences by the Wilcoxon signed-rank test (i.e., a paired test by peak). Hexamers with significantly (P < 0.05) higher (estimate > 0) or lower (estimate < 0) average unpairedness probabilities in observed versus control transcripts were considered as enriched in single-stranded or double-stranded RNA secondary structures, respectively. Repetitive elements binding analysis was performed as described for RNA-seq.

Protrusions purification and analysis

Cell protrusions were isolated as described before⁵⁶, with some modifications. 2 × 10⁶ cells were seeded on the top of collagen-I-coated polycarbonate transwell filters with 3-μm pore size (Greiner Bio-One) in complete DMEM. Cells were allowed to adhere O/N without the addition of media to the bottom chamber of the transwells. The day after, the media on top of the filter was replaced by FBS- and epidermal growth factor- free DMEM. Protrusions were induced for 2 h by the addition of complete DMEM to the bottom chamber. Transwells were then washed with cold PBS, and protrusions were collected by shaving the bottom of the filter using a glass coverslip dipped in Lysis Buffer (50 mM Tris 1 M HCl pH 7.5, 100 mM NaCl, 1% NP40, 0.1%SDS, 0.5%DOC, 1:100 Protease-inhibitor Mix M (Serva), 1:100 SUPERase in Rnase Inhibitor (Thermo). The cell-body fraction was subsequently purified by direct addition of the Lysis Buffer to the top of the filter. Nucleus/cytoplasm fractionation of the cell body was performed as described in “Protein extraction and Western blotting.” For Immunofluorescence (IF) of cells that protrude through a transwell 50,000 cells were seeded onto 12 mm membrane, fixed, and stained as described in “Immunofluorescence.” For RNA-seq, RNA was extracted from the nucleus, cytoplasm and protrusion using RNeasy Mini Kit (Qiagen). Library preparation is described in “RNA-seq, library preparation, analysis. Subcellular gene expression concentration estimates were next analyzed using the edgeR package. Lowly expressed genes (below 1 CPM in at least 3 samples) were filtered out, and normalization factors were calculated using the trimmed-mean of M-values (TMM) method (implemented in the calcNormFactors function). PCA was performed using the prcomp R function (parameters: scale = TRUE, center = TRUE), using log-normalized expression counts as inputs. Differential subcellular expression analysis was performed using the edgeR GLM testing framework with quasi-likelihood F-test (QLF). Relative compartmental enrichment analysis in scramble (control) samples was performed by comparing the expression estimates across compartments (e.g., nucleus versus cytoplasm, protrusion versus cytoplasm, protrusions versus nucleus), and clustering the union of all significantly DEGs (FDR < 0.05) obtained by these pairwise comparisons by their specific compartmental enrichment. Differential enrichment analysis between KD and control samples was next performed by comparing subcellular concentration ratios (i.e., nucleus to cytoplasm, protrusion to cytoplasm, protrusions to nucleus) between KD and control samples (i.e., testing the coefficients (expr_in_compartment_A_KD)-(expr_in_compartment_B_KD)-(expr_in_compartment_A_WT)-(expr_in_compartment_B_WT), thus obtaining a fold-change of ratios). Genes with FDR < 0.05 were considered significantly differentially expressed.

In vitro responsiveness of the Krt6b promoter to stress-induced activation

EPCs expressing the K6-GFP reporter were subjected to different stress conditions, including hypoxia, UV exposure, hydrogen peroxide treatment, viral infection, mechanical stimulation with Yoda-1, and TNF-alpha treatment. Briefly, a total of 6500 cells were plated into a 96-well plate in quadruplicate. For the hypoxia experiment, cells were transferred to a multigas incubator, maintained a low-oxygen environment (2% O₂) and incubated for 24, 48, and 72 h. To evaluate the effect of UV exposure, cells were irradiated with 254 nm UV with an energy dose of 400 mJoules/cm² using a UVP Crosslinker CL-3000 (Analytic Jena). Cells were then incubated under standard conditions for 6, 12, 24, and 48 h. For hydrogen peroxide experiments, cells were treated with 0.5 nM H₂O₂ for 2, 6, 24, and 48 h. For the viral infection assay, cells were transduced with a pLKO.1 puro scramble virus with a 40% efficiency, and the virus-containing medium was maintained for 2, 12, and 24 h. To evaluate the effect of inflammatory signaling, cells were treated with 20 ng/µL recombinant mouse TNF-α (Bio-Techne) for 2 or 6 h. Mechanical stress was induced by treating cells with 10 µM Yoda-1 (Sigma-Aldrich) for 6, 12, or 24 h. For all experiments, cells were collected at the indicated time points, and the percentage of GFP-positive cells was quantified by flow cytometry. Untreated cells were used as the baseline control for GFP expression.

In vitro scratch wound-healing assay and transwell migration/protrusion assay

For the scratch assay, EPCs or SCC expressing shRNAs were seeded to confluence in a 48-well plate or 96-well plate. To block proliferation, cells were treated with Mitomycin C (4 μg ml⁻¹). Then, a wound was inflicted by dragging a sterile pipette tip across the monolayer, and the medium changed to remove the cell debris. Wound closure was detected in each sample by Incucyte® SX5 Live-Cell Analysis Instrument. Photos were taken at time points 0, 6, 12, 18, and 24 h and analyzed by Basic analyzer. Wound area closure (%) was calculated using the following formula: (t = 0, 6, 12, 18, 24 h − t = 0 h wound area) × 100/t = 0 h wound area. For migration assay 75,000/150,000 EPCs expressing pMSCV scramble or pMSCVshRNAs, or 90,000 HaCaT expressing pLenti-Cbx3-Blast or pLenti-Blast, or 90,000 human primary keratinocytes expressing pLKO.1 puro scramble or pLKO.1 puro shRNAs were plated into a transwell insert (Corning, 6.5 mm, 8-μm pore size; 12 mm, 3-μm pore size) in a 24 or 12-well plate, for migration and protrusions assay, respectively. Cells were seeded in serum-free medium to the upper chambers O/N. 500/1000 μL of medium containing 10% FBS was added to the lower chambers. Cells invading the downside of the chamber membrane were fixed after 8 and 3 h, respectively. Cells were stained by immunofluorescence or smFISH and analyzed by confocal microscopy.

Proliferation assay

EPCs expressing shRNAs were plated in 96-well plates at a concentration of 5000 cells per well. The number of cells seeded for each shRNA allowed for exponential growth over the course of the assay. Cells were counted by flow cytometry at t = 0, 2, 3, 4, and 5 days. The following formula was used to calculate the cumulative proliferation rate: t = 0, 2, 3, 4, and 5 days, cell count/t = 0 day cell count.

SCC cell invasion assay and invasive protrusion assay

This assay was conducted with transwell inserts (Corning, 6.5 mm, 8-μm pore size, and 12 mm, 3-μm pore size) in a 24 or 12-well plate, for Cell Invasion Assay and Invasive Protrusion Assay, respectively. The upper chambers of the transwell were precoated with Matrigel Matrix (Corning) at a density of 300 μg/ml and incubated at 37 °C for 6 h. Briefly, 150/300 μL of serum-free medium containing 75,000/150,000 SCC expressing pLenti-Cbx3-Blast or pLenti-Blast cells were seeded to the upper chambers O/N. 500/1000 μL of medium containing 10% FBS was added to the lower chambers. Cells invading the downside of the chamber membrane were fixed after 8 and 3 h, respectively. Cells were stained by immunofluorescence or smFISH and analyzed by confocal microscopy.

Immunofluorescence, in situ hybridization on cells (sm-FISH)

Immunofluorescence of paraffin-embedded human and mouse skin sections was performed as described before⁸⁹. Immunofluorescence of cultured cells was performed as described⁹⁰, with minor modifications. For staining on EPCs expressing shRNA in standard 2D cultures, cells were plated at a concentration of 100,000 cells/well on a 6-well plate with rounded glass disks O/N. For staining on EPCs in 3D transwell-induces cells protrusions, 75,000 cells were seeded onto 12 mm membrane inserts (Corning, 12 mm, 3-μm pore) O/N. Cells were fixed with PFA 4% in PBS for 15 min, washed 3 times with PBS and permeabilized with Permeabilization Buffer (Triton X-100 0.3% in PBS) and incubated in Blocking Buffer (5% bovine serum albumin and Triton X-100 0.1% in PBS) for 1 h. Primary antibodies were diluted in Blocking Buffer and incubated at 4° O/N. Secondary antibodies were incubated 1 h at RT and DAPI (1 µg ml⁻¹, Thermo) 5 min at RT. Glass disks and insert membranes were mounted using Mowiol 4–88 Reagent (Sigma-Aldrich) mounting solution. sm-FISH was performed following the established SCRINSHOT protocol⁹¹, with some modifications. Briefly, cells were plated on Superfrost Plus Gold microscopy slides (Thermo), arranged in a cell culture plate. Twenty-four hours after seeding, the slide was washed twice in PBS-DEPC and fixed with 3% PFA DEPC-PBS, and Grace Bio-Labs SecureSeal hybridization chambers (Sigma-Aldrich) were mounted. Cells were permeabilized using 0.5% Triton X100 in PBS-DEPC for 10 min at RT and incubated 30 min at RT in blocking solution 1× Ampligase Buffer (Lucigen), 0.05 M KCl (Sigma-Aldrich), 20% deionized Formamide (Sigma-Aldrich), 0.2 μg/μl BSA (NEB), 1 U/μl Ribolock (Thermo), 0.2 μg/μl tRNA (Thermo) and 0.1 μM Oligo-dT30 VN (Eurofins Genomics). Padlock probes were designed as previously described⁹¹. Hybridization of the padlock probes was performed by incubating 40 nM of probes in blocking solution. The unhybridized probes were washed out by three washes with Washing Buffer (2×SSC, Sigma-Aldrich and 10% deionized formamide, Sigma-Aldrich). The following ligation reaction (0.5 Units/μl SplintR ligase (NEB), 1× T4 RNA Ligase Buffer (NEB), 10 μM ATP, 0.2 μg/μl BSA and 1 U/μl Ribolock was then carried out. Subsequently, RCA reaction (1× Φ29 buffer, 5% glycerol, 0.25 mM dNTPs (Thermo R0181), 0.2 μg/μl BSA and 0.5 U/μl Φ29 DNA polymerase with RCA primer (TAAATAGACGCAGTCAGT*A*A, Eurofins Genomics)) took place. After three more washes with Washing Buffer, detection reaction with detection oligo (GAC TGC GTC TAT TTA GTG GAG C [FITC], Eurofins Genomics) was performed. Finally, after the removal of the hybridization chambers, the slides were washed and incubated with 1 μg/mL DAPI and mounted with ProLong Glass Antifade Mountant (Thermo). CBX3 staining was performed after smFISH following the protocol described above. The antibodies are listed in Supplementary Data 8.

Microscopy

Fluorescence images of Standard 2D cultures were acquired using the TCS SP8 Tandem Scanner Leica confocal microscope equipped with 20× dry and 40× or 63× immersion objectives (Zeiss). Four laser lines (405, 488, 561, and 633 nm) were used for near-simultaneous excitation of DAPI, Alexa448, Alexa568 and Alexa647 fluorophores. For 3D transwell-induced cell protrusions, cell migration and cell invasion, Z-stacks were acquired with Leica TCS SP8 at 200 or 400 Hz with an optimal stack distance and a 1024 × 1024 or 2048/2048 dpi resolution. Z-stack projections were generated using the LAS AF software package (Leica Microsystems) as max intensity projections. Tile-scan fluorescent images of mouse cutaneous SCC samples were obtained with Leica TCS SP8 (20× objective, 600 Hz and 1024 × 1024 dpi resolution) and processed to combine with a 10% margin in LAS X software (Leica Microsystems). Acute and chronic (venous and pyoderma gangrenosum ulcers) human wound samples were acquired with Leica TCS SP8 (20x objective, 400 Hz and 1024 × 1024 dpi resolution). H&E slides were acquired using Olympus BX41/Leica DM6 microscope equipped with a 4× or 10× objective.

Image analysis

Immunofluorescence images were analyzed using Fiji software platforms (https://imagej.net/) and Qupath (version 0.5.1; https://qupath.github.io/). For quantification of CBX3 staining, cell boundaries were defined by Phalloidin labeling. Cells were co-staining with WGA, used to define polarized cells and to manually divide the cell in the Golgi-facing or the Golgi-opposite sides. For each half of the cells, protrusions were defined by enlarging the cell boundary for −70 pixels and +70 pixels, consecutively. Mean intensity signal was measured in the protrusions, or in the cytoplasm areas without protrusions, for each marker. For quantification of proteins in the 2D protrusions and measurement of morphological parameters (area, perimeter, ferret diameter, circularity, AR, roundness, and solidity), cell boundaries were defined by Phalloidin labeling. For CBX3, WGA, vinculin, and p-FAK, p-FAK-positive protrusion areas were manually defined by using p-FAK IF staining as a reference and setting an appropriate threshold. For P4HB, LAMP1, and TOM20 IF staining, protrusions areas were defined by enlarging the cell border for −70 pixels and +70 pixels, consecutively. Mean intensity signal in each ROI was measured for each marker. Same p-FAK-positive protrusion areas were used to define morphological parameters by the Fiji command Analyze>Set Measurements. For quantification of 3D protrusions IF in the Protrusion Purification analysis and 3D Cell and Protrusion Invasion Assay, a transwell filter was used to define an up and bottom zone. For each of these two areas, the command Process>filters Gaussian blur and Process>find maxima were applied to set a threshold on the maximum value for each marker used. The output number of particles in the bottom area was normalized on the number of particles in the upper one. For quantification of the 3D protrusions IF in the protrusion purification analysis and 3D Cell and protrusion invasion assay, a transwell filter was used to define an up and bottom zone. For each of these two areas, the command Process>filters Gaussian blur and Process>find maxima were applied to set a threshold on the maximum value for each marker used. The output number of particles in the bottom area was normalized on the number of particles in the upper one. For smRNA-FISH quantification in the 2D protrusions, protrusion and cell-body areas were defined by phalloidin staining. RNA-FISH intensity signal in all protrusions was normalized to the overall cell body. For acute and chronic human wound immunofluorescence samples, 2–3 layers of basal cells were selected using Topro as a marker. A threshold for identifying proliferative cells was established using Ki-67 staining in the migration zone, based on the number of visually confirmed positive cells. This same threshold was then applied to quantify Ki-67–positive cells in the proliferation zone.

For IHC analysis, for cell detection and segmentation, nuclei was segmentate from the DAPI channel using the StarDist 2D algorithm running on Qupath with the “dsb2018_heavy_augment” model (https://github.com/stardist, https://github.com/qupath/qupath-extension-stardist). The segmented nuclear areas were expanded to generate cytoplasmic areas, and made measurements from different components of cells (nucleus, cytoplasm, and whole cell) using Qupath built-in functions. The images were then manually classified and annotated based on the expression of epithelial markers Keratin 6 and E-cadherin, as well as morphological characteristics of each tissue and cell. Classification was made into the following groups: basal and suprabasal layers of keratinocytes, including SCC cells, skin appendages, and stroma. The fluorescent intensity of the nuclei and cytoplasm in each ROI was calculated using a Groovy script on Qupath. This script calculated the sum of the data (IntD, area, and mean intensity) of each component identified through cell detection contained within the ROI areas. Positional plots from single or tile-scan images were also generated based on the cell detections using StarDist and Qupath. The classification of each cell detection was conducted using Qupath built-in object classifier, which was trained with manual annotations. All charts were created using GraphPad (Prism).

Statistical analyses

Details of statistical analysis and the number of replicates can be found in the figures and figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

G.D. designed and supervised the study. C.D., A.C., C.L.L., O.A. and L.C. performed the experiments with L.E., A.A., D.D., V.P. and F.N. assistance. Libraries were sequenced by F.A., A.L., G.P. and E.B. performed computational analysis. M.W. and K.N. provided human samples for histology. F.B. provided human samples for primary cells. C.D., A.C., A.L., S.O. and G.D. analyzed and interpreted the data. C.D. and G.D. wrote the manuscript with input from all authors.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Source data for all graphs are provided with this paper. Omic data that support the findings of this study have been deposited in the Gene Expression Omnibus (GEO) under accession code GSE276632. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains Supplementary material available at 10.1038/s41467-026-74864-6.