The BRD4S-LOXL2-MED1 interaction at the forefront of cell cycle transcriptional control in triple-negative breast cancer

Triple-negative breast cancer often develops resistance to single-agent treatments, which can be circumvented with targeted combinatorial approaches. Here, we demonstrate that the simultaneous inhibition of LOXL2 and BRD4 cooperate to reduce triple-negative breast cancer proliferation in vitro and in vivo. Mechanistically, we reveal that LOXL2 interacts in the nucleus with the short isoform of BRD4 and MED1 to control cell cycle progression at the gene expression level via sustaining the formation of BRD4-MED1 nuclear transcriptional foci. Indeed, the pharmacological or transcriptional repression of LOXL2 provokes downregulation of cell cycle gene expression, G1-S cell cycle arrest, and loss of BRD4-MED1 foci. Our results indicate that the BRD4S-LOXL2-MED1 interaction is fundamental for the proliferation of triple-negative breast cancer. Therefore, targeting such interaction holds potential for the development of novel triple-negative breast cancer therapies.


Introduction
Bromodomain-containing protein 4 (BRD4) is an epigenetic reader well known in cancer for its role in the regulation of super-enhancers assembly (1, 2) and oncogenes transcriptional activation (3)(4)(5). Several BRD4 inhibitors, known as BET inhibitors (BETi), have been tested in multiple cancer models, and more than twenty clinical trials are currently ongoing to evaluate their anticancer efficacy (6). In breast cancer, the inhibition of BRD4 has shown promising preclinical results (7), sparking a particular enthusiasm for the treatment of the triple-negative breast cancer (TNBC) subtype, for which no targeted or efficient anticancer therapy has been developed so far (8)(9)(10). However, due to its heterogeneous and aggressive nature, TNBC commonly develops resistance to single-agent approaches (11), including BETi (10), which might be limited with the use of anticancer combinatorial approaches.
Lysyl Oxidase Like 2 (LOXL2) is a member of the lysyl oxidase family of copper-dependent amine oxidases (12) that catalyzes the oxidative deamination of peptidyl lysine residues. In the extracellular matrix, LOXL2 activity promotes collagen, elastin (13), and tropoelastin (14) crosslink, a phenomenon that is associated with the accumulation of extracellular matrix, fibrosis, and inflammation, which are all typical hallmarks of cancer (15)(16)(17). Intracellularly, LOXL2 can localize in the nucleus and promote the oxidation of nuclear proteins such as TAF10 (18) and Histone-3 (19,20), leading to transcriptional repression and heterochromatinization, respectively.
Recently, it has been found that LOXL2 has a pivotal role in different solid cancers, including liver, pancreas, lung, and breast (21)(22)(23). LOXL2 inhibition efficiently reduces TNBC cell proliferation (24), sensitizes TNBC cells to common anticancer treatments such as DNA damaging agents (25), and inhibits the formation of TNBC distal metastasis (21). The mechanism by which LOXL2 can support cancer proliferation is not yet entirely understood and may vary among cancer types and subtypes.
In the present study, we addressed the question of whether the simultaneous inhibition of BRD4 and LOXL2 could be proposed as a novel strategy for the treatment of TNBC. Not only we discovered the two treatments synergize, but also that in the nucleus LOXL2 selectively interacts with the short isoform of BRD4 to promote the expression of cell cycle genes, thus controlling TNBC proliferation.

The simultaneous inhibition of BRD4 and LOXL2 impairs TNBC proliferation
Given the fact that BRD4 and LOXL2 are both promising targets for the treatment of TNBC, we investigated whether the combination of their inhibition may cooperate in hindering TNBC proliferation. For this, we sought to perform a drug synergism analysis in three TNBC cell lines, MDA-MB-468, MDA-MB-231, and BT-549, which express different levels of BRD4 and LOXL2 (Fig. 1A). First, we verified that the LOXL2 inhibitor PXS5382 (24) efficiently reduced the nuclear catalytic activity of LOXL2 by checking the levels of oxidized Interestingly, when the two treatments were given simultaneously we observed either an additive (in MDA-MB-468 and BT-549) or a synergistic (in MDA-MB-231) effect ( Fig. 1B   and fig. S1C).
As the combinatorial treatment in MDA-MB-231 cells showed a synergistic effect, we orthotopically implanted these cells into the mammary glands of immunodeficient mice (NOD SCID). When tumor volumes reached approximately 100mm3, MDA-MB-231 bearing mice were randomly divided into 4 groups, and treated with either vehicle, PXS5382, (S)-JQ1, or the combination of PXS5382 and (S)-JQ1 (combo). While PXS5382 or (S)-JQ1 alone were sufficient to delay tumor growth in vivo, the combo treatment showed clear superior effects (Fig. 1, C and D). Since MDA-MB-231 was derived from the pleural effusion of a TNBC patient, this cell line is known for its metastatic capacity when orthotopically implanted (27). Indeed, histopathologic analysis of lung sections from vehicletreated mice showed tumor metastatic foci ( Fig. 1E and fig. S1D). In contrast, such metastatic foci were observed neither in PXS5382-treated nor in combo-treated mice, confirming the potential anti-metastatic effect of LOXL2 inhibition (21). (S)-JQ1 treatment alone also showed great anti-metastatic potential, however, one metastatic nodule could still be found in the analyzed samples. Importantly, no significant toxicity was observed in combo-treated mice compared to vehicle-treated, PXS5382-treated, or (S)-JQ1-treated mice  . S1J). Taken together, these findings suggest that the simultaneous inhibition of LOXL2 and BRD4 can delay TNBC tumor growth in vitro and in vivo, and this novel strategy is promising for potential clinical application.

LOXL2 does not control BRD4 expression
Since LOXL2 may induce chromatin compaction via H3K4 oxidation (25) . S2, A and B). However, LOXL2wt overexpression was followed by minimal changes at the BRD4 protein levels ( fig. S2C). Similarly, when downregulating LOXL2 in either MDA-MB-231 or BT-549 TNBC cell lines, which normally express medium to high levels of LOXL2, we could observe a decrease of H3K4 oxidation at the BRD4 promoter followed by a mild increase of BRD4 gene expression ( fig. S2, D to G).
Again, almost no changes were detected at the BRD4 protein levels ( fig. S2, H . S3B). A similar result was observed when analyzing the TCGA proteomics data of human breast tumor samples (29) (fig. S3C). These data suggested that the synergism observed with the simultaneous inhibition of BRD4 and LOXL2 is not mediated by a transcriptional co-regulation mechanism driven by BRD4 and LOXL2 antithetic chromatin-associated roles.
We, therefore, asked whether breast cancer cells overall show increased protein levels of either LOXL2 or BRD4. By analyzing the CPTAC proteomics dataset (30), we stratified breast cancer samples into the different subtypes depending on the expression of common markers (VIM, HER2, PGR, and ESR1), and compared their LOXL2 and BRD4 protein levels with those of adjacent normal breast tissues. Interestingly, LOXL2 protein levels were significantly increased in every subtype of breast cancer, while BRD4 showed a mild increase only in the TNBC subtype (Fig. 1F). Additionally, when comparing the CCLE-associated BETi (31) sensitivity we could observe that cell lines expressing high levels of LOXL2 were less sensitive to BRD4 inhibition than LOXL2-low cell lines (Fig. 1G). A similar trend was shown when stratifying the cell lines based on LOXL2 protein levels ( fig. S3D). These results suggested the possibility of a functional relationship between BRD4 and LOXL2, which we further investigated to unveil the molecular mechanism underlying the observed small molecules' cooperation.

LOXL2 interacts with the short isoform of BRD4 via its bromodomains but in an acetylation-independent manner
In the attempt of finding a possible functional connection between BRD4 and LOXL2, we wondered whether they may physically interact. To test this hypothesis, we extracted nuclei from MDA-MB-231 cells and performed a BRD4 pull-down (PD) experiment. Samples were then analyzed by western blot, and the results showed that LOXL2 was a BRD4 nuclear interactor ( Fig. 2A). Given the fact that there are no efficient LOXL2 antibodies to perform endogenous LOXL2 PD, we carried out the complementary experiment by transiently overexpressing LOXL2wt in MDA-MB-231 and performing Flag PD instead. To our surprise, the results showed that LOXL2 selectively interacted with the short isoform of BRD4 (BRD4S) (Fig. 2B). The interaction was retained also when overexpressing LOXL2m, suggesting that the catalytic activity of LOXL2 is not required for its interaction with BRD4S ( Fig. 2B). Comparable results were obtained when transfecting LOXL2wt or LOXL2m into HEK-293-T cells, in which LOXL2 expression is undetectable ( fig. S4A).
Flag-PD confirmed that LOXL2 interacted with BRD4S. BRD4S harboring mutations in the AcK binding pocket of both bromodomains still retained the ability to bind to LOXL2. In addition, both bromodomains either alone or together as well as the specific C-terminal domain of BRD4S interacted with LOXL2. Finally, although BRD4L was expressed at lower levels, we could not detect interaction with LOXL2, confirming results from Fig. 2b (Fig.   2D). Finally, we performed a docking analysis of BRD4_BD1/LOXL2 and BRD4_BD2/LOXL2, respectively. A collection of structures (table S1) from the Protein Data Bank (38) (PDB) and Interactome3d (39), and three independent software (ZDOCK (40), Autodock VINA (41), and ProteinFishing (42)) were used to propose binding models.
Results were energetically minimized and ranked based on buried surface, FoldX (43) interaction energy, and FoldX (43) stability. A filtering step reduced the number of reliable docks to 7 (table S2). For those, we performed computational mutagenesis to exclude models incompatible with experimental data that showed that the mutations N140F of BD1 and N433F of BD2 still allowed LOXL2 binding (Fig. 2D). Such analysis reduced the candidate models to 4 (Fig. 2E). With these remaining models, we performed computational mutagenesis to highlight putative key-binding residues (table S3), which may be genetically perturbed to further dig down into the molecular basis of the interaction. In all the 4 proposed models, we observed that histidines 626 and 628 of LOXL2, which once mutated abrogate its catalytic activity (19,44), did not participate in the interaction with BRD4_BDs ( fig.   S4D), corroborating the pull-down experiment presented in Fig. 2B showing that LOXL2m retained the interaction with BRD4S. Two of the 4 proposed binding models, were very similar between BD1 and BD2 and implicated the interaction of LOXL2 with the BD1/BD2 AcK binding pockets (models 2 and 4 in Fig. 2E). In these models, it is observable how the mutations N140F and N433F did not abrogate LOXL2 binding despite the strong involvement of the residues in the interaction ( fig. S4E). For the remaining models (1 and 3) the interaction between BD1 or BD2 with LOXL2 did not involve the AcK binding pockets, making the N140F and N433F mutations irrelevant. We, therefore, reasoned that if the interaction between LOXL2 and BRD4_BDs would involve the AcK binding pocket most probably (S)-JQ1 treatment would abrogate it. Therefore, we treated MDA-MB-231 cells with (S)-JQ1 and performed a BRD4 PD experiment to investigate whether the treatment would impair BRD4-LOXL2 interaction. Results showed that the interaction is importantly reduced in presence of the treatment (Fig. 2F). In parallel, we observed that the presence of (S)-JQ1 invalidated the docking models 2 and 4 while did not perturb 1 and 3 ( Fig. 2G), indicating that, most probably, the docking models 2 and 4 are the most accurate. Overall, these results indicated that the interaction between LOXL2 and BRD4S involves the AcK binding pocket of both BRD4S bromodomains, but that the residues N140 (BD1) and N433 (BD2), which are directly responsible for interacting with acetylated proteins, are dispensable. Additionally, the fact that the specific C-terminal domain of BRD4S mediates the interaction with LOXL2 may explain why BRD4L cannot engage in this interaction.
Nonetheless, we cannot discard that the unstructured C-terminal domain of BRD4L may in addition function as a destabilizer.

LOXL2 and BRD4S control the expression of DREAM target genes
Given the fact that LOXL2 and BRD4 were found to interact in the nucleus, we hypothesized that they may share transcriptional-associated functions. To dissect the possible transcriptional role executed by the BRD4S-LOXL2 interaction, we have performed and integrated ATAC-seq, RNA-seq, and BRD4-ChIP-seq comparing shControl (C) and shLOXL2 (KD) conditions in MDA-MB-231 cells ( fig. S5A). The ATAC-seq experiment indicated that upon LOXL2 downregulation chromatin became much more relaxed, as expected due to its role in controlling chromatin compaction (Fig. 3A). Therefore, we initially expected that LOXL2 downregulation would induce gene expression changes towards upregulation. However, such a hypothesis was disproved since such changes were equally distributed between upregulated and downregulated genes (Fig. 3B). In addition, the increased accessibility observed upon LOXL2 downregulation was not followed by transcriptional activation (Fig. 3C). These data, therefore, showed that even though the loss of LOXL2 induced important chromatin changes, these were mostly not functional, or buffered against, as they did not correlate with gene expression changes. We then characterized the functional effect of LOXL2 downregulation on gene expression, independently of whether chromatin at that region was opening or not. When performing Gene Set Enrichment Analysis (GSEA), we observed that the downregulation of LOXL2 induced the upregulation of processes involved in cell morphology, secretion, membrane trafficking, and cell differentiation, with cell-cell junction being one of the most significant pathways ( fig. S5B). These results are in agreement with the role of LOXL2 in reshaping the extracellular matrix (45) and regulating epithelial to mesenchymal transition (46)(47)(48), thus corroborating that the dataset we produced was of high quality. On the other hand, when we performed the same analysis on genes downregulated following LOXL2 KD we found that there was a significant enrichment of terms associated with cell cycle and, specifically, with DNA duplication (S-phase) and mitotic completion (M-phase) (Fig. 3D), suggesting a possible novel role of LOXL2 in controlling cell cycle progression. Of note, other downregulated GO terms, even if not directly, were still clearly associated with cell cycle regulation. Clear examples were DNA conformational changes that may relate to DNA duplication or chromatin condensation into chromosomes, and organelle fission, which refers to the process of mitochondrial fission that happens during the G2-M phase transition and is required to segregate mitochondria into the two daughter cells (Fig. 3D).
We then asked how BRD4S and BRD4L bound across the genome in control conditions and whether their localization is affected by LOXL2 downregulation. We performed ChIPsequencing (ChIP-seq) using two antibodies, one recognizing only BRD4L (Ab1) and one recognizing both isoforms (Ab2). We reasoned that the signal overlap between the two antibodies are chromatin regions bound either by BRD4L or BRD4L and S together, while the signal coming only from the Ab2 should correspond to BRD4S preferentially bound regions (Fig. 3E). We retrieved in total 2774 peaks for Ab1 and 3288 peaks for Ab2 and around 20% of the peaks were located at promoter regions ( fig. S5C). With these identified promoters, we looked at overlaps with gene sets (GSs) in the Molecular Signatures Database Fisher_DREAM targets GS caught our attention because it comprises genes that regulate cell cycle progression (DREAM target genes) (50), which according to our RNA-seq, were deeply affected by LOXL2 downregulation (Fig. 3D). Interestingly, the majority of the DREAM target genes retrieved in our Ab2-ChIP-seq were downregulated in the LOXL2 KD condition (Fig. 3G), suggesting a functional interaction between BRD4S and LOXL2 transcriptionally controlling cell cycle progression.
We, therefore, tried to understand whether BRD4S and BRD4L relocalize differently on chromatin when LOXL2 is downregulated. For both antibodies, we observed a big increase in signal upon LOXL2 downregulation ( fig. S5E), which agreed with the increased chromatin accessibility ( fig. S5F). However, the increased BRD4 binding was not significantly followed by gene expression upregulation ( fig. S5G), suggesting that it might just be a nonfunctional consequence of the increased chromatin accessibility. We then focused the analysis on DREAM target genes. We observed that in the absence of LOXL2 the promoters of the DREAM target genes were not anymore exclusively captured with Ab2, but with both antibodies, thereby indicating either a BRD4S-BRD4L co-binding or exclusive BRD4L binding ( Fig. 3H and fig. S5H). We hypothesized that the increased signal observed with Ab1 and Ab2 in the KD condition may principally depend on BRD4L binding nonspecifically wherever chromatin becomes more accessible following the downregulation of LOXL2.
Indeed, the overlap between the chromatin loci identified with the two antibodies considerably increased in the LOXL2 KD condition ( fig. S5I). Altogether, these results suggested that LOXL2 and BRD4S may interact at the promoter of DREAM target genes to transcriptionally control cell cycle progression. Interestingly, the binding of BRD4L to DREAM target genes promoters in the absence of LOXL2 was followed by DREAM target genes' transcriptional repression rather than upregulation.

LOXL2 repression destabilizes the transcriptional control of early cell cycle genes
Given that 1) LOXL2 interacted with BRD4S, 2) that promoters of DREAM target genes were preferentially bound by BRD4S, and 3) that the downregulation of LOXL2 induced the transcriptional repression of DREAM target genes, we sought to investigate what was the underlying connection bringing BRD4S-LOXL2 interaction at the forefront of cell cycle transcriptional control. We first reasoned that if the downregulation of LOXL2 decreases the expression of DREAM target genes, it should also induce cell cycle arrest. We confirmed by qPCR that LOXL2 downregulation reduced the expression of a set of selected DREAM target genes in MDA-MB-231 cells ( fig. S6A). When comparing the cell cycle profile of MDA-MB-231 cells treated with either C or KD, the latter was arrested in the G1 phase of the cell cycle ( Fig. 4A). We then performed high-throughput (HT-) immunofluorescence (IF) using an antibody recognizing H3 serine 10 phosphorylation (H3S10p), a typical marker of mitotic entry, and quantified its signal in C and KD treated MDA-MB-231 cells. We observed that the percentage of positive H3S10p cells dramatically decreased when LOXL2 was downregulated, indicating that the absence of LOXL2 drastically hampered mitotic entry ( Fig. 4B and fig. S6B). Thus, we wondered whether the role of LOXL2 in the regulation of cell cycle progression was dependent or independent of its catalytic activity. To answer this question, we treated MDA-MB-231 cells either with DMSO or PXS5382 and observed that LOXL2 inhibition considerably decreased the expression of DREAM target genes ( fig. S6C), similarly to what was observed for the LOXL2 KD condition. This result indicated that the catalytic activity of LOXL2 is required for cell cycle transcriptional control. As a consequence, cells treated with PXS5382 were mostly arrested in the G1-S phase of the cell cycle, as observed by FACS (Fig. 4C). In addition, we performed a time-lapse experiment transducing MDA-MB-231 cells with vectors encoding for the protein SLBP, which accumulates in the nucleus during G1 and starts being degraded in G2 (51), tagged with the mTurquoise2 fluorophore, and the Histone H1, which allows nuclear identification, tagged with the Maroon1 fluorophore. When comparing DMSO vs PXS5382 treated cells, we observed that while DMSO-treated cells were able to progress into the cell cycle, first acquiring and then progressively losing the accumulation of nuclear SLBP-mTurquoise2, PXS5382-treated cells failed, retaining much longer the nuclear mTurquoise2 fluorescence, indicative of G1-S arrest (Fig. 4D, fig. S6D and movie S1). For this reason, early cell cycle genes often have E2F consensus sequences at their promoters, while late cell cycle gene promoters include B-MYB and FOXM1 consensus sequences (50,53). Given the fact that when downregulating or inhibiting LOXL2 we observed a G1-S cell cycle arrest, we speculated that the BRD4S-LOXL2 interaction may be required for the transcription of early cell cycle genes. Supporting this hypothesis, when we conducted motif analysis on promoter regions recognized only with Ab2 we indeed observed enrichment in E2F consensus sequences (table S4 and fig. S6E).
We, therefore, wondered whether BRD4S and LOXL2 interact with the MuvB complex, B-Myb, and/or FOXM1. We performed Lin9 (MuvB subunit), b-MYB, and FOXM1 PD experiments using wild type MDA-MB-231 cells and showed that BRD4S and LOXL2 interacted with all the three factors, however, both proteins were stronger associated with Lin9 and B-MyB and milder with FOXM1 (Fig. 4E), again indicating that the BRD4S-LOXL2 interaction may be required for interphase cell cycle progression (G1-S-G2), rather than mitotic (G2-M). Additionally, we could barely detect BRD4L as either interacting with Lin9, b-MYB, or FOXM1, in agreement with the ChIP-seq results indicating that DREAM target gene promoters were preferentially bound by BRD4S (Fig. 3F). Interestingly, our PD experiments also showed that MED1 preferentially interacted with Lin9 and b-MyB, addition, other BRD4 functionally related genes appeared as differentially essential in LOXL2 low-expressing cells, such as the de novo purine synthesis genes PAICS and GMPS (55). Finally, BRD4 itself, the BRD4 target oncogene MYC (4,5,56,57), and the rest of the subunits of the Mediator complex showed a similar trend ( Fig. 4F and fig. S6F), indicating that cells with reduced LOXL2 levels were more susceptible to the loss of BRD4 functional partners.

LOXL2 catalytic activity is required for the stability of BRD4-MED1 transcriptional foci
It is very well known that the BRD4-Mediator interaction supports the formation of nuclear transcriptional foci (58) that decorate super-enhancer and boost the expression of target genes (1, 59). Given that BRD4 and MED1 colocalize at nuclear transcriptional foci (58), we investigated whether in our system we could detect such foci and if the downregulation of LOXL2 could impair their formation. We performed HT-IF in MDA-MB-231 cells either treated with C or KD, and immunostained for BRD4 or MED1. When quantifying the number of foci per area of the nucleus, we excitingly observed that in the KD condition there was a dramatic reduction of both BRD4 and MED1 foci ( fig. S7A). Similarly, when comparing cells treated either with DMSO or PXS5382 we observed that the latter considerably reduced the number of BRD4 foci (Fig. 5, A and B) as well as the overlap between BRD4 and MED1 foci (Fig. 5, C and B). Consequently, the distance between a MED1 spot and the nearest BRD4 spot significantly increased (Fig. 5D). Given the fact that it has been recently discovered that BRD4S is crucial for the formation of BRD4-Mediator transcriptional foci (60), we wondered whether the downregulation of cell cycle genes observed with LOXL2 chemical or transcriptional inhibition could be the result of the destabilization of LOXL2-BRD4S-MED1 interaction. When performing a BRD4 PD we observed that the PXS5382 treatment did not impair the interaction between LOXL2 and BRD4, as expected from the docking results ( fig. S7B). We, therefore, performed MED1 PD in DMSO and PXS5382 treated cells. Interestingly, we observed that in the DMSO condition MED1 was able to interact with LOXL2 but the PXS5382 treatment abolished this interaction (Fig. 5E). On the other hand, even though (S)-JQ1 treatment impaired the interaction between LOXL2 and BRD4 as previously observed (Fig. 2F), it only mildly affected the interaction between BRD4 and MED1 (Fig. 5F). Therefore, the BRD4S-LOXL2-MED1 triangular interaction is only partially affected by solely inhibiting either BRD4 or LOXL2, thus explaining why the combinatorial treatment showed synergy in the tested TNBC models (Fig. 5G). Overall, these data revealed a novel mechanism by which the interaction between BRD4S and LOXL2 is required for the formation of BRD4-MED1 transcriptional foci and the gene expression regulation of early cell cycle genes (Fig. 5H).

Discussion
15% of breast cancers are classified as TNBC, they present a very aggressive phenotype and have an unfavorable prognosis. Given the fact that there is no targeted therapy to treat them, the standard regimen for TNBC treatment relies on the use of conventional chemotherapeutic agents. However, this strategy in most cases fails to arrest TNBC proliferation, whose molecular basis is largely unknown. In this study, we report an unprecedented mechanism controlling TNBC proliferation. While investigating whether the simultaneous inhibition of two promising TNBC targets -BRD4 and LOXL2-cooperates in arresting tumor proliferation, we discovered not only that such cooperation exists, but also that they are physical and functional partners. We showed that, in the nucleus, LOXL2 binds to the short isoform of BRD4 and that together they promote the expression of genes involved in the progression of the cell cycle interphase. For the very first time, we report LOXL2 as a transcriptional activator rather than a repressor, as previously described (18). We furthermore discovered that LOXL2 also interacts with MED1 and that LOXL2 catalytic activity favors the formation of BRD4-MED1 transcriptional foci. We hypothesize that the reduced formation of BRD4 and MED1 foci, which are known to decorate super-enhancer clusters, is the cause of the transcriptional repression of cell cycle genes observed either with LOXL2 downregulation or chemical inhibition. Interestingly, while LOXL2 has never been associated with cell cycle transcriptional control, BRD4 has been previously linked to the regulation of early cell cycle gene expression (61,62), however, the molecular mechanism remained mostly unexplored. In addition, our work sheds further light on the divergent roles executed by different BRD4 isoforms, confirming the prooncogenic function of BRD4S (63).
In this study, we provide in vivo data demonstrating that the simultaneous inhibition of LOXL2 and BRD4 can pave the way for the development of future TNBC anticancer approaches. Interestingly, the essentiality analysis that we conducted also revealed that LOXL2-low expressing cells are very sensitive to the loss of the oncosuppressor TP53, and conversely, they survive better if MDM2, which is required to induce TP53 proteasomal degradation, is absent ( Fig. 4F and fig. S6F). This evidence is in line with previous results indicating that the loss of LOXL2 in TNBC enhances DNA damage and may indicate that cells with low levels of LOXL2 require a resilient mechanism to guarantee genome integrity and avoid apoptosis. Excitingly, also BRD4 has been multiple times associated with DNA damage response (64,65), mechanistically being responsible either to insulate chromatin at the DNA damaged sites to allow repairing (66), or required to prevent the accumulation of R-loops and protecting against transcription-replication collision (67). The role of LOXL2 and BRD4 in DNA damage and their cooperation in controlling the transcriptional regulation of cell cycle progression described in this study may suggest that their simultaneous inhibition in tumor cells could act as a double-edged sword. In line with this, combining LOXL2 and BRD4 inhibition with DNA damaging agents could be expected to further improve the cancer treatment outcome. Although beyond the scope of this study, we believe that this approach may be suitable for the treatment of tumors where both BRD4 and LOXL2 are highly expressed.

RNA extraction and qPCR
RNA was extracted using the PureLink RNA mini kit (Invitrogen) and converted into cDNA using the High Capacity RNA-to-DNA kit (Applied Biosystems) following the manufacturer's instructions. qPCR was performed using Power SYBR Green PCR Master Mix (Applied Biosystems) in a 7900HT thermocycler (Applied Biosystems). For RNA-seq, samples were sequenced using the Illumina HiSeq 2500 system.  In the same way, ΔΔGs smaller than -2.0 kcal/mol should be considered invalidating for the relative model ( Stained cells were FACS analyzed using LSRFortessa and FlowJo V10 (BD Biosciences).

Small molecule treatment and synergism analysis
Triple-negative breast cancer cell lines were seeded at day 1 in 96-well plates in triplicate:

Western blot and pull-down experiments
Whole-cell extracts were obtained using SDS lysis buffer (2% SDS, 50 mM Tris-HCl, and

ATAC-seq sample preparation
Three biological replicates of ATAC-seq samples were prepared as previously described (70).
Briefly, 50.000 MDA-MB-231 cells infected with shControl (C) or shLOXL2 (KD) were collected and treated with transposase Tn5 (Nextera Tn5 Transposase, Illumina Cat #FC-121-1030). DNA was purified using AMPure XP beads to remove big fragments (0.5x beads; >1kb) and small fragments (1.5x beads; <100bp). Samples were then amplified using NEBNexthigh-Fidelity 2x PCR Master Mix (New England Labs Cat #M0541) with primers containing a barcode to generate the libraries, as previously published (71). Each replicate was amplified with a combination of the forward primer and one of the reverse primers containing the adaptors listed in the primer's table below. The number of cycles for library amplification was calculated as previously described (70). DNA was again purified using MinElute PCR Purification Kit (Qiagen) and samples were sequenced with Illumina HiSeq 2500.
Read alignment was offset as previously described (71). Peaks were called using MACS2 and 2µl of Proteinase K and incubated at 55ºC for 1h and 65ºC O/N for de-crosslinking (inputs were also incubated for de-crosslinking).
DNA was purified with MinElute PCR purification kit (Qiagen) and eluted in nuclease-free water. Genomic regions of interest were detected by qPCR using the primers listed in the primers list. The NEBNext Ultra DNA Library Prep Kit for Illumina was used to prepare the libraries and samples were sequenced using Illumina HiSeq 2500 system.
Statistical significance of the distributions of ATAC-seq signals in ChIP-seq peaks was performed using a two-sample Kolmogorov-Smirnov test.

RNA-seq sample preparation
Three biological replicates of RNA samples were prepared by using 1x10 6  After 4 cycles of (S)-JQ1 treatment (Day 26), mice were euthanized and tumors were excised and measured. In addition, the lung of each mouse was dissected and fixed with 4% paraformaldehyde, then embedded in paraffin. Each lung sample was serially sectioned at 5µm of thickness, stained with H&E, and the 3rd, 5th, 15th, 25th, and 35th slides were used to count metastatic nodules.

Primers
All primers used in this manuscript are listed below:

Statistics
The statistical tests used are described within each figure legend.