Tissue fluidification promotes a cGAS–STING cytosolic DNA response in invasive breast cancer

The process in which locally confined epithelial malignancies progressively evolve into invasive cancers is often promoted by unjamming, a phase transition from a solid-like to a liquid-like state, which occurs in various tissues. Whether this tissue-level mechanical transition impacts phenotypes during carcinoma progression remains unclear. Here we report that the large fluctuations in cell density that accompany unjamming result in repeated mechanical deformations of cells and nuclei. This triggers a cellular mechano-protective mechanism involving an increase in nuclear size and rigidity, heterochromatin redistribution and remodelling of the perinuclear actin architecture into actin rings. The chronic strains and stresses associated with unjamming together with the reduction of Lamin B1 levels eventually result in DNA damage and nuclear envelope ruptures, with the release of cytosolic DNA that activates a cGAS–STING (cyclic GMP-AMP synthase–signalling adaptor stimulator of interferon genes)-dependent cytosolic DNA response gene program. This mechanically driven transcriptional rewiring ultimately alters the cell state, with the emergence of malignant traits, including epithelial-to-mesenchymal plasticity phenotypes and chemoresistance in invasive breast carcinoma.

The process in which locally confined epithelial malignancies progressively evolve into invasive cancers is often promoted by unjamming, a phase transition from a solid-like to a liquid-like state, which occurs in various tissues. Whether this tissue-level mechanical transition impacts phenotypes during carcinoma progression remains unclear. Here we report that the large fluctuations in cell density that accompany unjamming result in repeated mechanical deformations of cells and nuclei. This triggers a cellular mechano-protective mechanism involving an increase in nuclear size and rigidity, heterochromatin redistribution and remodelling of the perinuclear actin architecture into actin rings. The chronic strains and stresses associated with unjamming together with the reduction of Lamin B1 levels eventually result in DNA damage and nuclear envelope ruptures, with the release of cytosolic DNA that activates a cGAS-STING (cyclic GMP-AMP synthase-signalling adaptor stimulator of interferon genes)-dependent cytosolic DNA response gene program. This mechanically driven transcriptional rewiring ultimately alters the cell state, with the emergence of malignant traits, including epithelial-to-mesenchymal plasticity phenotypes and chemoresistance in invasive breast carcinoma.
The mechanical properties of cells and tissues are pivotal regulators of cell behaviour and fate in physiology and pathology, including during carcinogenesis 1 . Normal epithelial tissues frequently evolve into solid or jammed masses that are densely packed with cancer cells. To become malignant, a certain degree of fluidity is required for a tissue to be able to proliferate, migrate and disseminate. A recently discovered process by which cells can acquire migratory behaviour is cellular unjamming, a phase transition characterized by collective and cooperative cellular motion akin to fluid flow [2][3][4][5][6] . Whether and how unjamming impacts the acquisition of heritable changes that influence tissue state and malignant progression remains unclear.
Ductal adenocarcinoma in situ (DCIS), a precursor of invasive breast cancer, is a remarkable case in point. Firstly, DCISs typically grow at high cell density within the confinement of the mammary duct lumina (for example, comedonic growth) 7 . These conditions might expose DCIS to overcrowding and compressive mechanical stresses that impact their physical state favouring a transition to a solid (jammed) and kinetically arrested state 2,4,8 . Consistently, nearly 70% of DCISs are indolent, quasibenign lesions 9 . This suggests that packing and extreme confinement exert tumour-suppressive functions. However, 30% of these cancers overcome the caging imposed by the crowded cellular landscape of packed DCIS, by undergoing a solid-to-liquid ( jammed-unjammed) phase transition, which facilitates the acquisition of cell locomotion and progression to invasive ductal carcinoma (IDC) 2 .
We hypothesize that this material-like phase transition is an adaptive response to mechanical challenging conditions that, in addition to promoting collective dissemination of early lesions, as previously shown 2 , would also coincidentally result in a long-term, cGAS-STING-mediated, transcriptional-dependent phenotype switch in invasive breast carcinoma. Article https://doi.org/10.1038/s41563-022-01431-x velocity correlation length L C and the root mean square amplitude of the velocity fluctuations v rms (Extended Data Fig. 2g and Supplementary Video 2). The synergic increase in collective motility observed under these conditions also resulted in robust induction of CytoDR genes (Extended Data Fig. 2h).
We also studied HaCat keratinocyte cells. These cells undergo flocking after induction of RAB5A 6 , which is greatly enhanced following the addition of EGF to quiescent, serum-starved cells 16 (Extended Data Fig. 3a-c and Supplementary Video 3). EGF addition promoted robust flocking (Extended Data Fig. 3a-c) but it was insufficient to induce CytoDR genes. CytoDR gene induction required the concomitant expression of RAB5A (Extended Data Fig. 3a-c).
Together, these results indicate that endocytic-mediated tissue fluidization via flocking can transcriptionally rewire cell collectives toward a cytosolic DNA response in several normal and tumorigenic epithelia.

Tissue fluidification activates a cGAS-STING pathway
cGAS is an innate immune sensor of DNA that recognizes cytosolic DNA, resulting in the activation of STING. STING, in turns, activates TANK binding kinase 1 (TBK1) to phosphorylate the transcription factor interferon regulatory factor 3 (IRF3), which translocates to the nucleus to induce the expression of type I/III interferon and interferonstimulated genes 17 .
To determine the involvement of the cGAS-STING axis in the activation of CytoDR due to RAB5A-mediated tissue fluidification, we used pharmacological and molecular genetic loss-of-function approaches targeting each component of the cGAS-STING-TBK1-IRF3 pathway. We silenced cGAS, STING or IRF3 or treated cells with the cGAS inhibitor, RU.521, the STING antagonist, H-151, or the TBK1/IKK inhibitor, MRT67307, which impairs the phosphorylation of IRF3 (ref. 18 ). All these treatments robustly hampered the upregulation of CytoDR genes induced by tissue fluidification in MCF10.DCIS.com model tissues (Fig. 2a,b and Supplementary Fig. 1a). We also targeted key transcription factors acting downstream of the cGAS-STING axis, IRF9, STAT1 and STAT2, which robustly reduced CytoDR gene upregulation ( Fig. 2c and Supplementary Fig. 1a). Immunoblotting of cellular lysates of densely packed monolayers revealed that IRF3 and both the total and phosphorylated levels of STAT1, were elevated (Fig. 2d), consistent with this pathway being activated by RAB5A-mediated fluidification of MCF10.DCIS.com cell collectives. cGAS is activated by cytosolic DNA derived from invading microbes 19 , damaged mitochondria 20 , ruptured nuclei and micronuclei [20][21][22] , or self-DNA from engulfed tumour cells 23 . Nuclear damage frequently arises as a consequence of mechanically induced deformation 24,25 . We found no evidence of an altered number of micronuclei ( Supplementary Fig. 1b,c). Fluidification-via-flocking is, instead, accompanied by large fluctuations in cell density 12 and area ( Fig. 2e-g and Supplementary Video 4), which might result in increased nuclear deformation. We developed an automated image analysis pipeline to monitor nuclear shape changes over time to verify this conjecture. In control and RAB5-expressing MCF10A and MCF10.DCIS.com monolayers, tissue fluidification-via-flocking resulted in larger and faster deformations (Supplementary Video 5), which were measured by estimating the mean squared nuclear strain MSS(τ) ≡ ⟨⟨Δa 2 n (τ|t)⟩ n ⟩ t for different delay times τ and extracting the corresponding strain rate γ 0 ≅ MSS(τ)/τ (Fig. 2h,i). In previous expressions, Δa n (τ|t) ≡ [A n (t + τ) − A n (t)] / ⟨A n (t)⟩ t , where A n (t) is the projected area of the n-th nucleus at time t and the symbols ⟨⋅⟩ n and ⟨⋅⟩ t indicate averages performed over all the segmented nuclei and over time, respectively.
We also noticed that the expression of RAB5A resulted in a significant reduction of the mRNA levels of Lamin B1, but not of Lamin A/C, and of the protein levels measured by immunoblotting (Fig. 2j,k) and immunofluorescence ( Fig. 2l and Extended Data Fig. 4a).

Tissue fluidification induces a cytosolic DNA response
The expression of the small G protein RAB5A, a pivotal regulator of endosome biogenesis upregulated in human breast cancer and associated with decreased disease-free survival 10 , is sufficient to overcome kinetic and proliferation arrest in densely packed epithelia 2,6 . RAB5A does so by triggering a mechanically driven phase transition from a solid (or jammed) and immobile state to a flocking-fluid, hyper-motile state that is analogous to animal flocking 2,6,[11][12][13] . Molecularly, this is mediated by the endocytic function of RAB5A, which promotes the internalization of epidermal growth factor receptor (EGFR) into endosomal platforms for the prolonged activation of ERK1/2 and the actin nucleation promoting complex WAVE2. This, in turn, enhances lamellipodia that drive coordinated cell locomotion 2 . In breast carcinoma, tissue fluidification-via-flocking promotes collective motility and local invasiveness of DCIS 2 . We posit that this mechanically driven solid-to-fluid transition might also rewire the transcriptional state of early indolent lesions promoting a phenotypic switch that impacts tumour progression.
To address this possibility, we examine the transcriptional profile of densely packed epithelial monolayers formed by quasi-normal MCF10A cells and the respective oncogenic variant MCF10.DCIS. com cells. Both cell lines were engineered to express RAB5A in a doxycycline-inducible fashion to levels like those found in human breast cancer 2,10 . MCF10.DCIS.com cells express oncogenic T24-H-RAS and are used as models for the progression of DCIS to IDC 14 .
As expected, densely packed MCF10A and MCF10.DCIS.com monolayers are jammed and kinetically arrested 2,6 . Induction of RAB5A promoted the reawakening of collective motion via flocking 2,6 . This was accompanied by robust alterations in the transcriptional profile ( Fig. 1a and Extended Data Fig. 1a-c). Unexpectedly, gene set enrichment analysis (GSEA) revealed the interferon-stimulated gene signature (ISG) as the most significantly enriched in deregulated genes (Fig. 1b,c). Noticeably, innate immune responses are also promoted by free endogenous DNA present in the cytosol, which is recognized as nonself 15 . We thus verified that RAB5A expression boosted a cytosolic DNA response (CytoDR) program (Fig. 1c). Determination of the mRNA levels of the selected most upregulated genes confirmed the effect of RAB5A-fluidification, and highlighted the massive increase in the expression of a number of these genes (Fig. 1d,e). The upregulation of ISG was also detected in fluidized MCF10A monolayers ( Supplementary  Fig. 1a,b) and in MCF10.DCIS.com cells grown as tumoroid (Extended Data Fig. 2a). In all these conditions, we have previously shown that RAB5A expression is sufficient to promote a solid-to-liquid transition via flocking and persistent rotational collective motion 2,6 .
RAB5A upregulated CytoDR genes only mildly in sparse cells (Extended Data Fig. 2b), suggesting that this response is an emergent property of epithelial cell collectives and associated with tissue fluidification. To further explore this property, we correlated the expression of CytoDR genes and flocking motion (measured using the average migration speed v m of the entire cell collectives) as a function of cell density. We found that above a critical density, which corresponds to a condition where cells form a system-spanning inter-connected cluster, there is a sharp increase in v m (Extended Data Fig. 2c and Supplementary Video 1) and a concurrent elevation of CytoDR genes (Extended Data Fig. 2d). By contrast, CytoDR gene expression is diminished once RAB5A-expressing cells from compressed but flocking monolayers are replated sparsely (Extended Data Fig. 2e).
The induction of flocking motion via exposure to a hypotonic solution, which promotes tissues fluidification independently from RAB5A expression 6 , was sufficient to increase CytoDR gene expression (Extended Data Fig. 2f). Importantly, the concomitant expression of RAB5A and hypotonic treatment synergically activated flocking fluid motility, as revealed by the increase in typical quantities that measure collective motility, including the average migration speed v m , the  IF   I2  7 IF  I4  4  IF   I4  4 L  IF  I6 IF  IT  1 IF  IT  3 IS  G  1 5 O  A S L   IF   I2   7  IF   I4  4  IF   I4  4 L  IF  I6 IF  IT  1 IF  IT  3 IS  G  1 5 O  A S L   IF   I2  7 IF  I4  4  IF   I4  4 L IF  I6 IF  IT  1 IF  IT  3 IS  G  1   The increased mechanical stress and reduced Lamin B1 levels might compromise nuclear integrity and result in more frequent ruptures of the nuclear envelope (NE), with the release of DNA into the cytoplasm that, in turn, can trigger cGAS activation. We verified this possibility in multiple ways. Firstly, we expressed cGAS-fused to EGFP (EGFP-cGAS) and monitored its localization and distribution. In dense, kinetically arrested MCF10.DCIS.com monolayers and in sparsely seeded cells EGFP-cGAS displayed a primarily cytoplasmic diffuse staining, as expected 26,27 (Fig. 3a). Conversely, a focalized perinuclearly restricted localization was seen after RAB5A induction in flocking-fluid monolayers, but not in sparsely seeded cells (Fig. 3a,b), similarly to what occurs after nuclear envelope ruptures 28 .
Thirdly, we monitored nuclear envelope ruptures through real-time analysis of the dynamics of the 3NLS-EGFP sensor. 3NLS-EGFP displayed a nuclear restricted expression in control cells, but a cytoplasmic distribution in fluidized RAB5A-expressing monolayers, indicative of NE ruptures (Fig. 3d,e and Supplementary Video 6).
Finally, we performed correlative-light electron microscopy (CLEM) tomography and immune EM to directly visualize the presence of NE ruptures. In RAB5A, but not control cells, EGFP-cGAS accumulated at sites of condensed chromatin, immediately adjacent to the region where both the inner and outer NE membranes were ruptured ( Fig. 3f and Extended Data Fig. 4b). Immunofluorescent staining of Lamin A/C also revealed that nuclei in RAB5A-fluidized monolayers undergo large deformation and possibly ruptures as indicated by the accumulation of cGAS around distorted nuclei and at the apex of nuclear invagination (Extended Data Fig. 4c).

Tissue fluidification triggers mechano-protective responses
The large fluctuations in tissue density, cell area and nuclear shape suggest that RAB5A-fluidized epithelial collectives are subjected to persistent and chronic mechanical strain and stress. These stresses can compromise tissue integrity 31 and cause nuclear rupture and DNA damage 24,25 . Both individual cells and epithelial sheets, however, can adapt to acute stress by mounting a nuclear mechano-protective response that preserves them from widespread genomic damages 32 . These responses include increases nuclear rigidity and size, elevation in chromatin compaction 32,33 , and the remodelling of peri-nuclear cytoskeletal actin with the formation of nuclear actin rings 28,34 . We hypothesized that endocytic unjamming-via-flocking exerts prolonged mechanical stress in epithelial ensembles that react by mounting a mechano-protective strategy, which, eventually, fails resulting in DNA damage. We set out to investigate this possibility.
Firstly, we investigated how nuclei respond to motility-induced fluctuation in the local cell density ρ in jammed and fluid monolayers 12,35 . We considered how the instantaneous nuclear strain rate γ N = , which we estimated as the divergence of the velocity field from PIV analysis (Fig. 4a-c). In all cases, a significant correlation is found between γ N and γ C , indicating that the nucleus systematically deforms in response to compressive and tensile strains imposed on the cell by the relative motion of its neighbours. RAB5A-fluidized monolayers undergo larger density fluctuations than control-jammed monolayers (Fig. 4b,c). Furthermore, the nearly linear relation between nuclear and cell strain rates is characterized by markedly different slopes in the two cases: in response to the same variation in the cell density, nuclei of RAB5A-fluidized monolayers deform significantly less (Fig. 4d,e), that is, they are stiffer. Using a simple mechanical model, described in ref. 36 , the slope of the γ N versus γ C curve can be used to estimate the ratio E N /E CY between the effective elastic moduli E N and E CY characterizing the mechanical response to in-plane compressive/tensile stresses of the two main cellular compartments, nucleus and cytoplasm, respectively. We found that the ratio E N /E CY is about twice as large in RAB5A-fluidized monolayers, indicating a significant increase in nuclear stiffness compared with controls (inset of Fig. 4d, e). We corroborated this finding by probing the mechanical properties of RAB5A-fluidized nuclei and monolayers through several orthogonal approaches. We quantify nuclear elasticity using atomic force microscopy (AFM)-based force indentation of the nuclear surface through the cell cortex. Nuclear force indentation showed an approximately twofold increase in the stiffness in RAB5A-fluidized monolayers as compared with control (Extended Data Fig. 5a). We also probed the rheological properties of the cytoplasm, which may contribute to the difference in Young's modulus obtained by AFM indentation. We found that the mean square displacement and coefficient of diffusion of genetically encoded, cytoplasmic fluorescence particles (GEM) 37 were not significantly different in control-arrested versus flocking-fluid RAB5A monolayers (Extended Data Fig. 5b,c). Thus, the increased stiffness of RAB5-fluidized cells is likely the result of the increased nuclei rigidity.
Next, we subjected control and RAB5A monolayers plated on a deformable polydimethylsiloxane (PDMS) membrane to a predefined 22% biaxial stretch and measured the ensuing cell and nuclear deformations. The cell nuclei of flocking fluid monolayers deformed about twofold less than control cells (Extended Data Fig. 5d). Exploiting the same model used to interpret motility-induced deformations, the results of this stretching experiment can be expressed in terms of the ratio E N /E CY between the effective elastic moduli associated to the nucleus and the cytoplasm (Methods). This ratio was found to be about twice as large in RAB5A monolayers (7.1) compared with controls, (4.5) (Extended Data Fig. 5d). Similarly, the nuclear deformation, obtained by determining the rate of aspect ratio changes, of RAB5A-expressing cells that are forced into a restricted 6 μm-wide channel was two-time less than in control cells (Extended Data Fig. 6e and Supplementary Video 7), consistent with the increased nuclear stiffness.
The increased rigidity of RAB5A nuclei might impact the nuclear shape at a steady state. Thus, we measured nuclear shape variations in control and fluidized monolayers by determining the dimensionless parameter excess of the perimeter of the projected nuclear shape (EOP) through immunofluorescent analysis of SUN2, an inner nuclear membrane protein 38 . EOP values of a highly folded, presumably floppy, and soft object are expected to be close to 1, whereas EOP of a rigid object tends to be close to 0. The EOP values of RAB5A fluidized monolayers were significantly closer to 0 with respect to control-jammed ones (Extended Data Fig. 5f,g). We also estimated the nuclear envelope shape fluctuations assuming that rigid nuclei should display reduced fluctuation with respect to softer or floppy ones. We monitored nuclei in live cells expressing a mini-EGFP-Nesprin1, which encompasses the Calponin actin-binding domain and the c-terminus of the NE protein Nesprin1 (ref. 39 ), at high frame rates. The amplitude of NE fluctuations was next calculated by measuring the standard deviation of the NE from its mean position. The NE fluctuations were significantly reduced in RAB5A fluidized cells (Extended Data Fig. 5h and Supplementary Video 8).
Finally, we employed Brillouin microscopy to probe nuclear mechanical properties. We detected a clear Brillouin frequency shift that corresponds to a significant increase in the longitudinal elastic modulus, M, indicative of increased nuclear stiffness, of RAB5A-expressing cell nuclei as compared with control (Extended Data Fig. 5i-k).
We also showed that nuclear-projected areas are nearly 25% larger in RAB5A-fluidized monolayers as compared with control ones (Extended Data Fig. 6a,b). Nuclei of RAB5A-expressing cells appeared      To probe the heterochromatin state, we initially examined the nuclear levels of H3K27me3. RAB5A-fluidized DCIS cells display a small but significant increase in H3K27m3-heterochromatin marks ( Fig. 4f,g), which were enriched at the nuclear periphery (Fig. 4h). Analysis of the top 100 upregulated genes revealed among the top transcription factors, EZH2, a histone H3 lysine 27 N-methyltransferase, and SUZ12, a key component of the polycomb repressor complex-2 (PRC2) (Extended Data Fig. 6c). These enzymes deposit H3K27m3 in response to nuclear mechanical stress 40 . Additionally, a pre-ranked GSEA showed enrichment in genes that can be targeted by PRC2 (Extended Data Fig. 6d). Consistently, silencing of EZH2 or SUZ12 abrogated the increase in H3K27me3-heterochromatin marks (Fig. 4g,i, and Extended Data Fig. 6e,f).
The chronic mechanical stress together with Lamin B1 reduction in RAB5A-fluidized monolayers might also elicit genome-wide structural alterations in constitutive H3K9me3 normally associated with the lamina, as a mechanism to dissipate forces 28 . We employed SAMMY-Seq and H3K9me3 ChIP-seq to verify this possibility. SAMMY-Seq is a high-throughput sequencing-based method for genome-wide characterization of chromatin accessibility, which can detect architectural rearrangements of lamina-associated heterochromatin domains 41 . RAB5A-fluidized monolayers displayed no changes in the H3K9me3-genome-wide ChIP-seq profile (Extended Data Fig. 6g), but a consistent reduction in the SAMMY-seq signal for heterochromatin regions (Extended Data Fig. 6h).
Next, we found that tissue fluidification is also accompanied by cell shape changes and perinuclear remodelling of the actin cytoskeleton. Doxycycline induction of RAB5A resulted in perturbations of the shape of cells (Fig. 4j), increased cytoplasmic polymerized actin and the formation of perinuclear actin rings (Fig. 4j,k).
RAB5A-fluidized monolayers mount a complex mechanoprotective response, which leads to decreased nuclear pliability and softness, suggesting the possibility that these monolayers are less capable of dissipating mechanical energy to prevent DNA damage 28 . Consistently, RAB5A-flocking monolayers display elevated DNA damage, as evidenced by the increase in 53BP1 and γH2AX foci ( Fig. 5a-c), and in the tail moment determined by neutral DNA comet assays 42 (Fig. 5d,e).

RAB5A induces cGAS activation in invasive carcinoma
Next, we studied whether the nuclear mechano-perturbations leading to cGAS activation and DNA damage observed in vitro are also relevant in pathological tissues. Control and RAB5A-MCF10.DCIS.com cells were injected into the mammary fat pads of immunocompromised animals to model DCIS. In these tumours, RAB5A induction increased CytoDR, as expected (Extended Data Fig. 5i), and was associated with an elevated number of γH2AX-positive cells (Fig. 6a,b), and increased levels of cGAS, which display a perinuclear dotted or crescent-like appearance (Fig. 6c,d). This pattern is similar to the one seen in human DCIS with local infiltrative areas (Extended Data Fig. 8c,d), likely reflecting its activation by cytoplasmic DNA. In human tumours, RAB5A displayed a graded increase in expression right at the margin of locally invasive foci ( Fig. 6e-g and Extended Data Fig. 7a). Notably, these marginal cells, also displayed increased γH2AX, and phosphorylated checkpoint kinase 1 (pCHK1; Fig. 6e,f and Extended Data Fig. 7a), a marker of persistent DNA damage 43 , and more relevantly of cGAS (Fig. 6g,h and Extended Data Fig. 7b-  Article https://doi.org/10.1038/s41563-022-01431-x elevated γH2AX and cGAS was also detected in patient-derived breast cancer organoids (Extended Data Fig. 7e).
Quantitative analysis of nuclear motion in living breast cancer organoids revealed that some of them displayed persistent rotational motion and all the key features of a flocking fluid (Supplementary Video 9). By combining 3D image registration with a differential analysis of the residual intensity fluctuations 13 , we decoupled the rigid body contribution associated with the global rotation of the organoid from the internal rearrangement dynamics, which is captured by the overlap parameter Q(τ). Persistently rotating organoids also displayed markedly faster internal dynamics, with a relaxation time τ * at least two times shorter than the static ones (Extended Data Fig. 8a-c and Supplementary Video 10). Next, we subjected the two static and the three rotating organoids to RNA-seq analysis. Despite the limited number of samples, we found that rotating fluid organoids display a significantly elevated expression of RAB5A (Extended Data Fig. 8d), enrichment in several genes belonging to the interferon-alpha stimulated and Interferon-related DNA damage resistance signature (Extended Data Fig. 8e-h). Whereas the analysis of the top deregulated genes pointed to PRC2 complex components, SUZ12 and EZH2, as a key altered transcription factor in rotating organoids (Extended Data Fig. 8i).

Tissue fluidification promotes EMT and chemoresistance
Chronic stimulation of cGAS-STING signalling has been shown to exert either immuno-protective or pro-tumorigenic effects. For example, by establishing an immune-suppressive tumour microenvironment, cGAS activation can promote a transition toward a mesenchymal state 44 and chemoresistance, favouring metastatic dissemination 45 . In addition, an experimentally derived interferon-related DNA damage resistance signature (IRDS), highly related to CytoDR, has been associated with resistance to chemotherapy and/or radiation across different cancers 46 . Hence, we hypothesize that endocytic-mediated tissue fluidification promotes the acquisition of chemoresistance and plastic EMT traits.
Firstly, we noticed that several mesenchymal markers, including CDH2, ZEB1, MMP13, EGF and AXIN2, were upregulated in RAB5A-expressing fluidized-via-flocking MCF10.DCIS.com (Extended Data Fig. 9a). Several canonical mesenchymal factors, including the master EMT regulators, SNAIL1 and 2 and TWIST, were unaffected, suggesting the acquisition of what has been defined as plastic EMT state (EMP) 47 . This was corroborated by the morphological analysis, which indicated that RAB5A expression leads to the acquisition of an elongated mesenchymal morphology (Extended Data Fig. 9b), and by the nuclear accumulation of ZEB1 in RAB5A-expressing MCF10. DCIS.com spheroids (Extended Data Fig. 9c). The expression of this set of genes was dependent on YAP/TAZ activity (Extended Data   Fig. 9d,e), suggesting that the EMP phenotype switch is, at least in part, a mechanoresponsive process 48,49 .
As EMP is associated with invasion and metastasis, next, we tested whether the increased migratory and invasive capacity in RAB5A-fluidized collectives is mediated by the activation of cGAS and STING. We found, however, that either cGAS or STING inhibition, while effectively reducing the upregulation of ISG (Fig. 2a,b and Extended Data Fig. 9j), and, more relevantly, of EMP genes (Extended Data Fig. 9f,g), had no impact on wound migration or collective invasion (Extended Data Fig. 9h,i and Supplementary Videos 11 and 12). This finding indicates that endocytic-mediated tissue fluidization promotes collective motion and invasion. This is accompanied by increased mechanical stresses that results in frequent NE ruptures, the release of damaged DNA and the activation of cGAS-STING, which contributes to the acquisition of mesenchymal traits.
Finally, we determined whether the elevation of CytoDR is associated with chemoresistance to anti-tumorigenic drugs. Firstly, GSEA in control and fluidized monolayers revealed the enrichment of an interferon-related DNA damage resistance signature, previously associated with resistance to chemotherapy and/or radiation 46 (Extended Data Fig. 10a). Additionally, RAB5A-cells were slightly more resistant to the microtubule stabilizer, docetaxel (Extended Data Fig. 10b-d), and the topoisomerase inhibitor, etoposide (Extended Data Fig. 10e,f). In the former case, while most control cells display a grossly defective nuclear morphology, as expected, more than 60% of RAB5A-expressing cells displayed intact and unperturbed nuclei (Extended Data Fig. 10c,d). Increased chemoresistance to docetaxel and etoposide was also detected in 3D spheroids (Extended Data Fig. 10g,h).

Outlook
The tissue-level phase transition from a solid or jammed to a liquid-like or unjammed state has been recently proposed to be a complementary or alternative gateway to cell invasion in both normal epithelia during development 3 and in solid carcinoma during malignant progression 2,4,8,50 . Indeed, the progression from an indolent, quasibenign ductal breast carcinoma lesion to invasive ductal carcinoma is associated with the acquisition of a flocking-fluid mode of collective motion induced by the upregulation of the endocytic, promigratory gene, RAB5A 2, 10 . In addition, we show here, that the altered mechanics of fluidized tissues impact nuclear integrity, and promote DNA release into the cytoplasm that results in a robust, cell-autonomous and long-lived transcriptional rewiring toward a cGAS-STING-dependent, pro-inflammatory response. This response has potential, context-dependent, far-reaching consequences in shaping the fate of tumour cells and cells of the microenvironment. Indeed, this cytosolic DNA response axis has recently been shown to induce a pro-tumorigenic phenotype, characterized Article https://doi.org/10.1038/s41563-022-01431-x by a shift toward a mesenchymal state and increased chemoresistance 44,45 , as we also found in our system. It is, however, likely that in an immune-proficient context, the same axis might trigger a pro-immunogenic, potentially anti-tumoral response. Whether this is the case is certainly a matter of future investigation. Similarly, it will be paramount to determine what are the factors or conditions that tune the mechanically driven cGAS-STING activation toward either a pro-tumorigenic or pro-immunogenic fate.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information,

Cell streaming assay
As previously shown 2 , cells were seeded in six-well plates (1.5 × 10 6 cells per well) in complete medium and cultured until a uniform monolayer had formed. RAB5A expression was induced, where indicated, 16 h before performing the experiment by adding fresh complete media supplemented with 2.5 μg ml -1 doxycycline hyclate to cells. Comparable cell confluence was tested by taking pictures by differential interference contrast imaging using a 10× objective and counting the number of nuclei per field. In the cell streaming assay, the medium was refreshed before imaging began. An Olympus ScanR inverted microscope with 10× objective was used to take pictures every 5-15 min over a 24-48 h period. The assay was performed using an environmental microscope incubator set to 37 °C and 5% CO 2 perfusion. After cell induction, doxycycline hyclate was maintained in the medium for the total duration of the time-lapse experiment.
For plasma membrane tension perturbation by osmotic treatments, an equal volume of hypotonic buffer (H 2 O + 1 mM CaCl 2 + 1 mM MgCl 2 ) was added to cell monolayer before performing experiments.

Wound healing
For wound healing experiments, confluent monolayers of MCF10. DCIS.com control and RAB5A-positive cells, were plated on 12-well plates, and were wounded by scraping with a pipette tip and then transferred immediately to the microscope stage-top incubator. For monolayer invading assay, wounds were coated with Matrigel before the time-lapse.
Time lapses were performed with a Leica Thunder Imaging System based on a Leica DMi8 inverted microscope equipped with a Leica DFC9000 GT sCMOS camera. The images were acquired with an HC PL APO 10× objective using Leica LAS X software. Wound healing maximal velocity of closure is calculated from the area covered over time extracted using a custom Fiji 51 and Matlab code.
The area covered over time is fitted with two straight lines (https://github.com/aganse/MultiRegressLines.matlab/blob/master/ regress2lines.m) the slope of the second straight line is used to estimate the maximal velocity of closure after a lag phase.

3D spheroid formation assay
MCF10.DCIS.com cells were plated on ultra-low-attachment-surface six-well plates (Corning, Cat# 3471) at a density of 5 × 10 3 cells per well. Cells were grown in serum-free condition for 10 days by adding fresh culture media every 2 days. After 7 days, 2.5 μg ml -1 doxycycline hyclate was added to the medium to induce RAB5A expression. Doxycycline was maintained in the medium for 2 days and finally spheroids were collected and processed for total RNA extraction.

Mammary fat pad tumour development in NSG mice
All animal experiments were approved by the OPBA (Organisms for the Well-Being of the Animal) of IFOM and Cogentech. All experiments complied with national guidelines and legislation for animal experimentation. All mice were bred and maintained under specific pathogen-free conditions in our animal facilities at Cogentech Consortium at the FIRC Institute of Molecular Oncology Foundation and at the European Institute of Oncology in Milan, under authorization from the Italian Ministry of Health (autorizzazione no. 604-2016). The maximal tumour size permitted by our ethical guideline is 200 mm 3 . None of the experiments exceeded this limit.

Neutral comet assay
The neutral comet assay was performed as described previously 53 following the manufacturer's protocol (Trevigen). Comet tail moment was measured using OpenComet plugin for ImageJ 54 .

Image acquisition
Time-lapse imaging of the motility of 3D organoids was performed using a Leica TCS SP8 laser confocal scanner mounted on a Leica DMi8 microscope equipped with motorized stage; a HC PL FLUOTAR 20×/0.5 NA dry objective was used. A white-light laser was used as the illumination source. Leica Application Suite X (LAS X, https://www.leicamicrosystems.com/products/microscopesoftware/details/product/ leica-las-x-ls/) software was used for all acquisitions. Image acquisition conditions were set to remove channel crosstalk, optimizing spectral detection bands and scanning modalities. ImageJ software was used for data analysis.
Image acquisition was performed using Operetta CLS, highthroughput imaging confocal microscopy system (Perkin Elmer) with Harmony software 4.9. Cells are imaged with 20× water immersion objective NA 1.0.
Confocal microscopy was performed with a Leica TCS SP5 confocal laser scanning system based on a Leica DMI 6000B inverted microscope. The images were acquired with an HCX PL APO 63X/1.4NA oil immersion objective. The software used for all acquisitions was Leica LAS AF. Laser lines: 405 nm, 488 nm, 561 nm, 633 nm.
Hypotonic-mediated cell streaming, and EGFP-3NLS leakage time lapses were performed with a Leica Thunder Imaging System based on a Leica DMi8 inverted microscope equipped with a Leica DFC9000 GT sCMOS camera. The images were acquired with an HC PL APO 63×1.4NA oil immersion objective (EGFP-3NLS time lapse) using Leica LAS X software.
Image acquisition of cGAS expression and localization on FFPE samples was performed with an Olympus BX63 full motorized wide field microscope equipped with a B/W Hamamatsu Orca_AG camera. the system is driven by Metamorph (Molecular Devices) software. We used UPlanApo 100× objective N.A.1.35.

Growth assay survival and broken nuclei discrimination
To evaluate the growth of MCF10.DCIS control empty vector mCherry-H2B or RAB5A mCherry-H2B a Harmony 4.9 (PerkinElmer) custom pipeline was implemented. After 3 days of treatment, the images were acquired. For each well (four wells each condition) composed of 89 fields, the pipeline identifies the nuclei on the Gaussian filtered (radius 3 pixels) global image of mCherry channel using the B method of the Find Nuclei module (parameters were tuned condition by condition). Then the nuclei were filtered by intensity and morphological criteria. To discriminate broken and normal nuclei, the Linear Classifier module with two classes was used; the classifier was trained using around 30 nuclei for both classes.

Image analyses
To count the number of foci per nuclei, a custom semi-automated Fiji 51,55 plugin was developed. The plugin identifies the DAPI/nuclear marker using Li (https://imagej.net/plugins/auto-threshold#li) Thresholding Schema on the filtered image (Gaussian filter with 2 pixel radius). Nuclei are then split using the watershed (https://imagej.net/plugins/ classic-watershed) method and then checked and corrected by hand. For each nucleus, the plugin identifies and counts the foci on the Foci Channel Marker (53BP1 or γH2AX) using ImageJ's Find Maxima (https:// imagej.nih.gov/ij/docs/menus/process.html#find-maxima) plugin with the noise tolerance parameter selected by hand.
To count the number of micronuclei per field of view (FOV), a custom semi-automated Fiji 51,55 plugin was developed. The plugin identifies

Nature Materials
Article https://doi.org/10.1038/s41563-022-01431-x the DAPI/nuclear marker using Huang (https://imagej.net/plugins/ auto-threshold#huang) thresholding schema on the filtered image (median filter with radius of 1 pixel). Nuclear structures were then split using the watershed (https://imagej.net/plugins/classic-watershed) method and then checked and corrected by hand. For each FOV, the plugin identifies and counts micronuclear structures using ImageJ's Analyze Particles (https://imagej.net/imaging/particle-analysis) plugin with the size parameters selected by hand.
For assessing histone methylation on lysine 27, FOVs were randomly selected based on nuclei signal, probed by DAPI staining. Images were analysed using a custom semi-automated plugin developed in Fiji 51,55 . Nuclei were identified on the DAPI channel using the StarDist plugin (https://imagej.net/plugins/stardist) with the built-in Versatile (fluorescent nuclei) neural network model. For each nuclear region of interest, the mean intensity was measured on the H3K27me3 channel and then normalized on the median of the mean intensity distribution of control cells.
For the analysis of the differential signal intensity at the nuclear periphery and central region, images were acquired, and nuclei were segmented as described above. For each nuclear region of interest, the area was reduced to shrink it 1.5 μm from the nuclear border and the mean intensity in the H3K27me3 channel was calculated in the central nuclear region. Finally, the peripheral H3K27me3 mean intensity was calculated in the area between the central region and the nuclear border.

Cell area fluctuation analysis
EGFP-E-cadherin expressing control and RAB5A-MCF10A cells were generated as described here. Cells were seeded in six-well plates (1.5 × 10 6 cells per well) in complete medium and cultured until a uniform monolayer had formed. RAB5A expression was induced, where indicated, 16 hours before performing the experiment by adding fresh complete media supplemented with 2.5 μg ml -1 doxycycline to cells. Comparable cell confluence was tested by taking pictures by differential interference contrast (DIC) imaging using a 10× objective and counting the number of nuclei per field. To monitor cell fluctuations the phase contrast channel and EGFP-E-Cad channel were merged and then the 2D image sequences were converted into 3D image. Based on the phase contrast images and the E-Cad signal, randomly selected cells are segmented and tracked semi-automatically using Segmentation Editor in Fiji (ImageJ plugin). Cell boundaries were annotated manually with the interval of a few time points and then cell boundaries at other time points are interpolated in 3D to obtain accurate cell morphological dynamics along time. As we have excellent temporal resolution, we assume that any deformation along the direction perpendicular to the cell boundary is small enough such that we can treat it as linear. Thus, we can estimate the deformation of a given cell along time. The negative and positive cell extension of the cell, as shown in Fig. 2f, can be quantitatively extracted. For a migrating cell, its surface can be described as a function of time, that is S(x, y, t). The deformation between any consecutive-time points is captured by the partial derivative of S with respect to t: Based on our linear assumption, equation (1) can be used to find a linear minimum distance mapping for the points on the boundaries at two time points.
We also need to define whether the deformation is positive (extending-maximum positive deformation (MPD)) or negative (retracting-maximum negative deformation (MPD)). A positive deformation corresponds to a boundary point moving to a position not previously occupied by the cell, and is indicated by a red arrow in Fig. 2f. A negative deformation corresponds to a boundary point moving to a position previously occupied by the cell, and is indicated by a blue arrow in Fig. 2f. To quantify the cell mobility, we focused on the following parameters:

Nuclei tracking and segmentation
Tracking and segmentation of single nuclei in sequencies of fluorescent microscopy images of confluent monolayers of mCherry-H2B cells is performed with a custom Matlab code implementing the following procedure. Images are first corrected for background intensity inhomogeneities by applying the background removing algorithm described previously in detail 56 .
Random noise in each corrected image is then reduced by applying a Wiener filter, an adaptive noise-removal filtering that preserves nuclei edges while smoothing the white noise (see equations 9.44-9.46 in ref. 57 ).
Nuclear segmentation is obtained by applying a seeded watershed transform to the spatial gradient of each filtered image 58 . The 'seeds' (that is, the pixels that are set to zero in the image before applying the watershed transform) are determined as follows. A Laplacian-ofthe-Gaussian (LoG) filter is applied to each filtered image, leading to a map L G whose local minima correspond to the candidate centres of the nuclei. Differences in the fluorescent intensity of different nuclei are corrected by dividing L G by an intensity map obtained via bicubic interpolation of the minima of L G . The resulting map L G is binarized by setting to zero (one) all pixels whose value is above (below) a fixed threshold value, k. Repeated pixel erosion operations are applied to the obtained binary mask to remove the smallest features and partially separate overlapping nuclei, leading to a final binary map L BN from which we extract the 'seeds' to be used in the seeded watershed transform: internal 'seeds' are obtained as the pixels where L BN is non-zero, while external 'seeds' as obtained the boundaries of the watershed transform of L BN .
Once the segmentation procedure on a given image is completed, we can determine the centre of mass, x i , of each nucleus in the image, its projected area A i and the angle θ n,i (modulo π) between the major axis of the nucleus and the x axis. The direction of the major axis is obtained as the direction of the eigenvector of the covariance matrix of the segmented area associated with the larger eigenvalue 59 .
To reconstruct cell trajectories, we employ the Matlab code freely available at http://site.physics.georgetown.edu/matlab/ implementing the algorithm developed by Crocker and Grier 60 . Once nuclei in different frames are linked into trajectories, the time evolution of the relevant single-nucleus parameters x i (t), A i (t) and θ n,i (t) can be determined.
The instantaneous velocity of the i-th nucleus at frame t is estimated as v i (t) = (x i (t + 1) − x i (t − 1)) /2δt, where δt is the time interval between two consecutive frames. The instantaneous mean migration velocity is computed as v cm (t) = ⟨v i (t)⟩ i , where ⟨…⟩ i denotes the average over all the nuclei in the field of view (FOV). The amplitude of the velocity fluctuations is evaluated as the root mean square velocity of the nuclei in the centre of mass reference frame v rms (t) = √ The velocity spatial correlation function is calculated as Visual inspection reveals that the described segmentation procedure is effective in identifying about 90-95% of the nuclei present in the field of view. Despite the effort to reduce multiple segmentation and nuclei merging, however, some segmentation errors occur, Article https://doi.org/10.1038/s41563-022-01431-x especially in those cases where the signal-to-noise ratio is low or partial superpositions of nuclei are frequent. To minimize the impact of segmentation errors on the analysis of nuclear features, we implemented a 'quality filter' to remove potentially flowed measurements. To this end, we compute the total instantaneous intensity J i (t) integrating the image intensity I (x, t) over the segmented area of the i-th nucleus at frame t. We then compare the instantaneous value J i (t) with its median t−1 t−11 evaluated over the previous 10 frames. If the quantity is larger than 0.1, the segmentation of the i-th nucleus at frame t is considered unreliable and the corresponding parameters are not included in the statistics. Trajectories that, after the application of this 'quality filter', lose more than 20% of frames because of this procedure are entirely excluded.

Particle image velocimetry
Particle image velocimetry (PIV) of fluorescent microscopy images of confluent monolayers of mCherry-H2B cells fluorescent images is performed by using the Matlab PIVLab software 61 .
We choose an interrogation area with size slightly larger than the typical inter-nuclear distance, typically corresponding to 14 μm. Outliers in the reconstructed velocity field, whose modulus exceeds a fixed threshold value, are identified, and replaced with the median value of the velocity over neighbouring grid points.

Nuclear deformation dynamics
To characterize nuclear shape fluctuations, we evaluate the mean square nuclear strain MSS (τ) = ⟨⟨Δa 2 where A i (t) is the projected area of the i-th nucleus at time t. To extract the key parameters characterizing nuclear deformation, we fit the model function MSS (τ) = σ w +γ 0 τ c [1 − e −τ/τ c ] to the data. This model, which includes a term σ w accounting for the random noise in determination of the projected area, describes a diffusive-like growth of the area fluctuations with a characteristic strain rate γ 0 for short delay times MSS (τ) ∼γ 0 τ, followed by a saturation to a plateau value γ 0 τ c for long times. In Fig. 2g,h, the data and the best fitting curves are reported upon the subtraction of the baseline value σ w obtained from the fitting procedure. Since MSS(τ) does not always reach a clear plateau within the time window accessible during the experiments (Fig. 2g,h), there is a relatively large uncertainty in the determination of the overall amplitude, γ 0 τ c , of the fluctuations. However, this does not affect the robustness of the estimate of γ 0 , as it characterizes the short time behaviour of the fluctuations, which is accurately sampled in our experiments.

Estimation of the relative stiffness of the nuclei
To characterize the mechanical response of cell nuclei to intracellular stresses induced by mutual cell displacements, we independently evaluate nuclear and cell deformations by measuring the instantaneous nuclear strain rate using the automated imaging segmentation pipeline described above, and the corresponding instantaneous cell strain rate, obtained after computing the divergence of the velocity fields measured by PIV analysis 36 .

Atomic force microscopy measurements
AFM measurements were carried out at 37 °C using a NanoWizard3 AFM ( JPK, Grermany) mounted on an Olympus inverted microscope. The protocol was adapted from a previous study 62 .
Prior to AFM measurements, MCF10.DCIS.com cells, control or RAB5A-induced, were seeded as a monolayer on 24 mm glass coverslips.

Cell stretching experiments
Cell stretching experiments were carried out using an automated cell stretching dish (international patent: WO 2018/149795) 63 .
The components of the device were designed using SolidWorks CAD software and 3D printed using a stereolithography-based 3D printer (Form 2, Formlabs) and a biocompatible and autoclavable dental resin (Dental SG resin, Formlabs). The printed parts were washed with isopropyl alcohol to remove eventual uncured resin and then post-cured in a UV box to complete the polymerization process. The 3D printed parts were then polished and assembled to create the lower and the upper portions of the stretching dish.
The lower portion (cell culture chamber) has four clips clamping a deformable silicone membrane (thickness, 200 μm, Silex Silicones) sandwiched between two rings. The upper portion (aperture driver) consists of stretching means movable relative to the chamber. This is coupled to a motor controller allowing to regulate the simultaneous movement of the stretching means and then apply the required stress/ strain to the membrane seeded with cells. According to previous tests, the strain field is uniform in the central region of the dish within 6 mm 2 .
Prior to the experiments, the cell culture chamber of the stretching device was coated with fibronectin (20 μg ml -1 ). Control and RAB5A-induced MCF10.DCIS cells were seeded as a monolayer.
Before imaging, the whole cell stretching dish was assembled by connecting the aperture driver of the stretching dish to the cell culture chamber. Biaxial stretching was applied directly on the stage of the microscope. Image acquisition was performed using a Confocal Spinning Disk system (Olympus) mounted on an IX83 inverted microscope provided with a motorized stage and an IXON 897 Ultra camera (Andor, 16 bit, pixel size 16 μm), and driven by CellSens Dimension software. Fluorescence images were acquired before and after application of 22% biaxial strain (reached in six stretching steps) through a 20× objective (UPlansApo, NA 0.75) using the EPI-fluorescence mode (excitation wavelength: 530-550 nm). For both control and RAB5A-expressing cells, 30 regions (field of views) were considered within the central region of the cell stretching dish. For each field of view, we evaluated the fractional change Δa = (A f − A i ) /A i in the average nuclear-projected area upon stretching. Exploiting the mechanical model in ref. 36 , we estimate the ratio between the effective elastic moduli E N and E CY of the nucleus, and the cytoplasm, respectively, as where ϵ tot = 1.28 × 1.28 ≅ 1.49 is the imposed area strain and β is the surface fraction covered by the nuclei. In our experiments, β was found to vary in the range [0.25, 0.31] and [0.17, 0.24] for MCF10.DCIS control and RAB5A-expressing cells, respectively.

Nuclear deformation through constricted channels
To evaluate nucleus deformability suspended cells were passively flowed, at a concentration of 100,000 cells ml -1 and a flow rate of 5 μl min -1 , into microchannels of 25×20 μm size with a constriction of 6×20 μm that induces substantial nuclear deformation. The microfluidic device was obtained from a micro-structured silicon mould fabricated at the clean room facilities of the Binning and Rohrer Nanotechnology Center through standard photolithography and dry etching processes. The microfluidic chip was then fabricated in PDMS (Sylgard 184) through standard replica moulding. Briefly, the PDMS precursor was mixed with the crosslinker (10:1) and poured on the silicon mould, degassed for 1 h in a vacuum bell and then cured for 3 h at 90 °C. The chip was then demoulded, treated for 1 min with oxygen plasma and irreversibly bonded to a 35 mm bottom-glass petri dish (Mattek).
Experiments were performed at 37 °C and 5% CO 2 atmosphere. Nuclear squeezing was recorded using a LEICA widefield DMI8 inverted system equipped with a HC PL Fluotar 10× NA = 0.32 (Leica, #506522) objective. The excitation source was a solid-state LED illumination at 475/28 nm (Lumencor light engine LED8). Images were each acquired for 50 ms with a sCMOS Andor Neo 5.5 camera.
Article https://doi.org/10.1038/s41563-022-01431-x To quantify nuclear deformability, we measured the speed of aspect ratio (AR) variation of control and RAB5A-expressing cells. Nuclei were labelled with mCherry-H2B. AR is a dimensionless parameter, and its rate of change provides a direct measurement of nuclear deformability, with higher values corresponding to more deformable (that is, softer) nuclei. We measured nuclei AR for several cells in the time interval needed to pass from an undeformed, just before entering into the constriction, to a completely squeezed configuration (Supplementary Fig. 6e, from A/R 0 at t 0 to A/R f at t f ). Images were analysed with ImageJ and the nuclear AR variation rates extrapolated through a robust fit in R using the 'robustbase' package 64 , where the rate of AR changes as a function of time and is the slope of the fitted curve.

Excess of perimeter and NE fluctuations
The dimensionless parameter excess of the perimeter of the projected nuclear shape (EOP) was determined through the analysis of immunofluorescent images of SUN2, an inner nuclear membrane protein 38 . As described previously 65 , to determine EOP, we first obtained values for perimeter (P) and surface area (A) from 2D projected images taken at the maximum radius of the nucleus (using SUN2 stained nuclei). Next, we introduced R 0 as the radius of the circle defined by the area A, and compute EOP as the ratio between (P − 2πR 0 ) and (2πR 0 ). EOP values of a highly folded, presumably floppy, and soft object tend to be close to 1, whereas EOP of a rigid object with a smooth surface tends to be close to 0.
Nuclear envelope fluctuations were measured following the method described in ref. 65 . Nuclear envelope images of both control and Rab5 populations were recorded using an HC PL APO 63× NA 1,40 OIL immersion objective (Leica, #506350), on a Leica DMI8 widefield Thunder Imager, equipped with a Leica DFC9000 GTC sCMOS camera.
A 488 LED illumination was used to record short time-lapses of about 2 min with high frame rate (4 fps).
The positions of each nucleus have been corrected for the natural linear and rotational motion of the cells using the Stackreg plugin available on Fiji software 66 .
Nuclear envelope fluctuations are measured as the standard deviation from its mean position. Eight separate line scans were drawn orthogonally along the surface of the nucleus. A simple macro (stack-profile_Pala)-modified from a version found on the ImageJ website 55 by Michael Schmid-was used to determine the profile of each line scan for each time point in the video. The standard deviation of the position of the nuclear envelope around the mean position of all timepoints is taken as a measure of the fluctuation of the nuclear envelope. Each point along a nucleus contributes as one measure. Obtained values are expressed in μm. N = 10 cells for both populations, for a total of n = 80 measures.
Single particle tracking was performed for 40 nM GEMs using a Perkin Elmer Spinning Disk Confocal Microscope and fluorescence was recorded with a C9100-50 (EMCCD) Camera and a 100× objective at a 20 ms image capture rate. The tracking of particles was performed with the Mosaic suite of FIJI using the following typical parameters: radius = 3, cut-off = 0.001 of fluorescence intensity, a link range of 1, and a maximum displacement of 5 px, assuming Brownian dynamics.

Extraction of the rheological parameters
For every trajectory, we calculated the time-averaged mean square displacement as defined previously 37,67 . To characterize the individual particle trajectories, we calculated apparent diffusion coefficients by fitting mean square displacement with linear (diffusive) time dependence at short time scales as shown previously 37 .

Brillouin microscopy
Control and RAB5A-MCF10.DCIS.com cells treated with doxycycline were grown in adhesion on a glass coverslip. Cells were analysed by Brillouin micro-spectroscopy as described previously 68 . Briefly, the 53C2 nm monochromatic beam of a single mode, diode-pumped, solid state laser (Spectra-Physics Excelsior) is focused with a power less than 5 mW. The investigated position inside the cell nuclei was chosen using the correlated bright field image of a custom-made inverted microscope.
The same water immersion 60× objective (UPLSAPO-60XW from Olympus) was used to acquire bright field images, to focalize the laser beam and to collect the back-scattered light, which is analysed in frequency by the TFP-2 interferometer 68 . The archived spatial resolution of the mechanical characterization was in the micrometric length scale 69 .
The Brillouin frequency was extracted fitting the spectra using a damped harmonic oscillator function. Measurements of the longitudinal elastic moduli were obtained after setting the cellular density at ρ = 1,080 kg m −3 , and the refractive index at n = 1.386 as described previously 68 .

Statistical analysis
All data are presented as scatter plots or box plots expressed as mean ± s.d. unless otherwise indicated. The number of experiments as well as the number of samples analysed is specified for each experiment and reported in the figure legends. Statistical significance was calculated, whenever we compared two distinct distributions, using a parametric two-tails unpaired student's t-test with Welch corrections for two samples with unequal variance or non-parametric two-tailed Mann-Whitney t-test as indicated. Kruskal-Wallis/Dunn's test was used for one-way data with more than two groups. Nested one-way ANOVA was used as reported for comparison of more unmatched groups. Statistical calculations were performed in GraphPad Prism 8 or Microsoft Excel. The significance of fold difference of each differential gene expression obtained by QRT-PCR was established using a two-tailed unpaired Student's t-test with Welch corrections for two samples and were with P values > 0.05 in all cases. Data collection and analysis were performed blind to the conditions of the experiments. Specifically, in all experiments involving mice, we assigned each mouse randomly to the treatment groups (injection of control or RAB5A cells into mammary fat pads). For the experiment with cells, we had two genetically distinct groups (control versus RAB5A) that were treated equally and randomly. For the CLEM experiment, we selected blindly control or RAB5A cells displaying accumulated perinuclear cGAS. No data points were excluded.

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

Data availability
RNA-seq data of MCF10A, MCF10.DCIS.com cells and organoids are deposited in the Gene Expression Omnibus (GEO) and European Genomephenome Archive (EGA), with the respective accession numbers: GSE183479 RNA-seq, GSE183539 SAMMY-seq, GSE183407 ChIP-seq, GSE205108 RNA-seq of organoids. Other data generated or analysed during this study are included in the Supplementary Information and are available from the corresponding authors upon request. Source data are provided with this paper.
Article https://doi.org/10.1038/s41563-022-01431-x Extended Data Fig. 1 | Tissue fluidification induces a CytoDR gene signature in MCF10 monolayers and has marginal effect on Interferon and Interferon receptor genes. a. Volcano plot of DEG between control empty vector and RAB5A-MCF10A monolayers. All significantly RAB5A-deregulated genes are indicated in red (upregulated) and blue (downregulated). The enrichment (log2Fold Change) is on the x-axis and the significance (Wald test -log10 p value two-sided) is on the y-axis. Labels show genes validated in Fig. 1d. b. Gene set enrichment analysis (GSEA) of DEG in RAB5A-MCF10A monolayer over control cells. Moderated t-statistic was used to rank the genes. Significantly enriched pathways (one-sided P-value < 0.05) are with the colour of the outline of the bar corresponding to the BH-adjusted P-value. P-value was calculated as the number of random genes with the same or more extreme ES value divided by the total number of generated gene sets. c. Immunoblots of lysates from doxycyclinetreated control (CTR) and RAB5A-(RAB5A)-MCF10.DCIS.com monolayers in 1 experiment out of n = 4 independent ones with the indicated antibodies. Mw is on the left. d. Heatmap of the row-normalized expression level of leading-edge genes for the interferon alpha and gamma response pathways, the interferon and interferon receptor genes in each of the 3 replicates of control empty vector (EV) and RAB5A-expressing (RAB5A)-MCF10.DCIS and MCF10A monolayers. Magnified labels on the right indicate genes validated in Fig. 1d. Normalized expression has been calculated for each gene starting from DeSeq2 regularized log-transformed expression value and then normalized by row mean values.  mRNA expression levels of RAB5A, IFI6, IFI27, IFI44, IFIT1, IFIT3, ISG15 and OASL in doxycycline-treated control (CTR) and RAB5A-HaCat monolayers treated exactly as described above with respect to control cells. Monolayers were serum starved mock-treated or stimulated with 100 ng/ml EGF: Hatched blue bar, serum starved HaCat CTR + EGF; Red bars, HaCat RAB5A-serum starved; hatched red bar, serum starved HaCat RAB5A stimulated with 100 ng/ml EGF (Serum Starved HaCat RAB5A + EGF) over control serum starved HaCat cells. Data are expressed as mean ± s.d. (n = 3 independent experiments). Values were normalized to the controls of each experiment.
Article https://doi.org/10.1038/s41563-022-01431-x Extended Data Fig. 4 | cGAS accumulation at nuclear envelope ruptures. a. Immunofluorescence images of control and RAB5A-monolayers (n = 3 independent experiments), stained with Dapi (Cyan), Lamin A/C (Green) and Lamin B1 (red). Scale bar 15 μm. b. Sequential Z axis sections of EGFP-cGASexpressing RAB5A-MCF10.DCIS.com monolayers used for 3D tomographic reconstruction shown in Fig. 3f (lower right image). The image outlined in red correspond the one shown in Fig. 3f (upper right image). Scale Bar, 500 nm. c. Immunofluorescence images of control (CTR) and RAB5A-expressing (RAB5A) MCF10.DCIS.com monolayers (n = 4 independent experiments), transiently transfected with EGFP-cGAS and stained with DAPI and anti-Lamin A/C antibody to detect nuclei and nuclear lamina, respectively. Magnified images of the selected boxed area are shown. Scale bar 20 μm.   Article https://doi.org/10.1038/s41563-022-01431-x Extended Data Fig. 8 | Human breast cancer organoids that persistently rotate display elevated RAB5A, and enrichment in ISG and PRC2-dependent genes. a. 3D rendering of 5 living breast cancer organoids. The size of each box along the x-and y-directions corresponds to 200 μm. b. Thin blue, orange and yellow curves: temporal evolution of the x, y and z components of the angular velocity associated with the rotation of the organoid depicted in the box above each panel, respectively. Thick black curves represent the angular speed, that is, the modulus of the angular velocity. Organoids are categorized as 'rotating' if their average angular speed is larger than 0.03 cycles/h. c. Symbols: overlap parameter Q(τ) obtained from 3D-DVA analysis, capturing the internal rearrangement dynamics of the organoid depicted in the box above each panel.
Continuous thin lines are best fitting curves to the data with an exponential model Q (τ) = e −τ/τ * . Vertical thick lines are drawn in correspondence with the characteristic relaxation time τ * obtained from the fitting model. d. Expression levels of RAB5A isoform in n = 3 patients-derived rotating (red) and n = 2 patients-derived non-rotating (blue) organoids. Bottom, middle and top lines of boxes show the 25th, 50th and 75th percentiles, respectively, and whiskers show 1.5 times the interquartile range from hinges. e. Heatmap showing the top 100 deregulated genes (top 50 up and top 50 down) in each of the 3 replicates of the rotating and non-rotating organoids. f. GSEA of DEG in rotating versus nonrotating organoids. Moderated t-statistic was used to rank the genes. Reported are significantly enriched pathways (one-sided P-value < 0.05; number of random genes with the same or more extreme ES value divided by the total number of generated gene sets) with the colour of the outline of the bar corresponding to the BH-adjusted P-value. g. GSEA Enrichment plot of DEG in rotating versus non-rotating organoids. The green curve corresponds to the ES (enrichment score) curve, which is the running sum of the weighted enrichment score obtained from GSEA software, while the normalized enrichment score (NES) and the corresponding one-sided P-value are reported within the graph. h. Quantitative changes in the expression of the INF-related DNA damage resistance signature genes in rotating versus non-rotating organoids. The log2Fold Change is plotted on the x-axis and the significance (Wald test p value two-sided) is defined by the colour code of the outline. i. Transcription factor enrichment analysis for overlap between the input set of differentially expressed genes in rotating versus non-rotating organoids and entries of the ChEA and ENCODE databases In the bars is the one-sided P-value (combination of Fisher's exact test and deviation from expected rank for random input gene-set).

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted