C/EBPβ isoform-specific regulation of migration and invasion in triple-negative breast cancer cells

The transcription factor C/EBPβ is a master regulator of mammary gland development and tissue remodelling during lactation. The CEBPB-mRNA is translated into three distinct protein isoforms named C/EBPβ-LAP1, -LAP2 and -LIP that are functionally different. The smaller isoform LIP lacks the N-terminal transactivation domains and is considered to act as an inhibitor of the transactivating LAP1/2 isoforms by competitive binding for the same DNA recognition sequences. Aberrantly high expression of LIP is associated with mammary epithelial proliferation and is found in grade III, estrogen receptor (ER) and progesterone (PR) receptor-negative human breast cancer. Here, we show that reverting the high LIP/LAP ratios in triple-negative breast cancer (TNBC) cell lines into low LIP/LAP ratios by overexpression of LAP reduces migration and matrix invasion of these TNBC cells. In addition, in untransformed MCF10A human mammary epithelial cells overexpression of LIP stimulates migration. Knockout of CEBPB in TNBC cells where LIP expression prevails, resulted in strongly reduced migration that was accompanied by a downregulation of genes involved in cell migration, extracellular matrix production and cytoskeletal remodelling, many of which are epithelial to mesenchymal transition (EMT) marker genes. Together, this study suggests that the LIP/LAP ratio is involved in regulating breast cancer cell migration and invasion. This study together with studies from others shows that understanding the functions the C/EBPβ-isoforms in breast cancer development may reveal new avenues of treatment.


INTRODUCTION
The CCAAT/enhancer-binding protein family of basic region leucine-zipper (bZIP) transcription factors consists of six members (designated α to ζ) that are widely expressed and have functions involved in cell proliferation, differentiation, metabolism, and senescence 1,2 . C/EBPs regulate gene transcription through the formation of dimers and binding to C/EBP-specific recognition sites in the genome. Expression of C/EBPβ as well as of the related C/EBPα proteins is uniquely regulated at the level of mRNA translation involving a cis-regulatory upstream open reading frame (uORF) and translation into three protein isoforms of different length from a single mRNA molecule 3 . The C/EBPβ protein isoforms LAP1, LAP2 (Liver-enriched Activating Protein; also called LAP* and LAP) originate through regular translation initiation, and they are transcriptional activators. From the same mRNA an additional isoform LIP (liver-enriched inhibitory protein) is translated from a downstream AUG codon, omitting the N-terminal transactivation domains. LIP is generally considered a competitive inhibitor of LAP, although some studies suggest LIP has transactivating functions in addition [4][5][6] . All isoforms share a highly conserved C-terminal bZIP domain containing a basic DNAinteraction region and a leucine zipper dimerization domain. Expression of LIP requires initial translation of the uORF followed by a translation re-initiation event at the LIP-AUG codon 3,7 ( Supplementary Fig. 1a). The efficiency of LIP expression is regulated by mTORC1 signalling and prospectively through other pathways yet to be identified 8 . Differential expression of LAP and LIP is involved in breast tissue development, tissue remodelling during pregnancy and lactation, with LIP expression being elevated in proliferative phases [9][10][11][12][13][14] . Genetic ablation of the Cebpb gene in mouse mammary glands results in defective mammary ductal morphogenesis, lobuloalveolar development and differentiation of mammary epithelial cells 13 . In addition, using Cebpbknockout models it was shown that C/EBPβ is required for the expansion of mammary stem cells and for regular luminal cell lineage commitment 15 . Low LIP/LAP expression ratios are observed in untransformed cells or tissues surrounding the breast tumour 16 . In contrast, much higher expression of LIP compared to LAP has been clearly associated with high histological carcinoma grading (poorly differentiated) and highly proliferative (high Ki67 staining) aggressive tumours, including the oestrogen receptor-negative (ER−), progesterone receptor-negative (PR−) tumours [16][17][18] . In addition, a high LIP/LAP expression ratio was linked to metastatic breast cancer samples that were defective in the TGFβ-induced cytostatic response, whereas a low LIP/LAP ratio was associated with an intact growth-inhibitory TGFβ response. The TGFβ cytostatic response in cells with a high LIP/LAP ratio could be restored by re-expression of LAP. Since evasion of the TGFβ cytostatic response is a critical step during breast cancer development towards a more migratory and invasive metastatic phenotype 19 these data suggest that C/EBPβ isoform deregulation (towards a higher LIP/LAP ratio) is crucially involved in promoting breast cancer metastasis 20 .
The shift in high LIP/LAP expression instead of a general higher expression with preserved isoform ratio suggest that the underlying mechanism is post-transcriptional. In line with a translational mechanism is the finding that in HER2-overexpressing breast cancer cells HER2 activates the RNA-binding factor CUGBP1, which stimulates LIP expression over that of LAP 21 . Studies with genetically modified mouse models support that high LIP/LAP expression ratios are linked to the development of breast cancer. Specific overexpression of LIP in the mammary epithelium induces hyperplasia and neoplasia, although the latter with low frequency 9 . Mice that are systemically deficient in LIP expression but proficient in LAP expression, through mutation of the uORF in the Cebpb gene, show an overall reduced tumour incidence 8,22 .
Conversely, systemic mono-or biallelic ablation of LAP with retainment of LIP increases tumour incidence in mice 23 . Altogether, the studies demonstrate that aberrant translation of the CEBPB-mRNA into its different protein isoforms plays a crucial role in breast cancer development, although we do not fully understand the underlying complexity 24 .
Here, we show that triple-negative breast cancer (TNBC) cell lines express C/EBPβ with high LIP to LAP isoform ratios in contrast to luminal A breast cancer cell lines or untransformed mammary epithelial cells that express low LIP to LAP ratios. Reversing C/EBPβ expression to low LIP/LAP ratios in TNBC cell lines by overexpression of LAP reduces the migration and invasion potential of these cancer cells, whereas overexpression of LIP stimulates the migration in untransformed epithelial cells. Using CRISPR/Cas9 mediated knockout of CEBPB and genome-wide RNA-sequencing we reveal that expression of genes associated with epithelial to mesenchymal transition (EMT) and the extracellular matrix (ECM) partially depend on LIP and/or LAP expression. Altogether, our data suggest a role for high LIP/LAP expression ratio in the regulation of breast cancer cell migration and ECM remodelling, two key characteristics that are associated with the aggressive phenotype of TNBC.  25 as well as in the untransformed mammary epithelial cell line MCF10A cultured in media supplemented with EGF that is known to stimulate LIP expression through CUGBP1 activation 26 . The immunoblots and the bar graphs of the signal quantifications show that LIP/LAP isoform ratios are high in TNBC cells, most evidently in the cell lines with overall high expression of LIP and LAP. The luminal A cells show lower LIP/ LAP rations with lower overall LIP and LAP expression. In addition, LIP/LAP isoform ratios are moderately high in HER2positive cells as has been shown before 21 with variable overall expression levels. The MCF10A cells also express moderately high LIP/LAP ratio ( Fig. 1 and Supplementary Fig. 1b for additional independent data). Thus, similar to what was observed in patient samples of hormone receptor-negative breast cancer 17,18 , TNBC cells are characterized by high levels of LIP and high LIP/LAP expression ratios.

RESULTS
To examine whether the presence of the estrogen receptor (ER) or progesterone receptor (PR) in the luminal A-derived cancer cell lines affects LIP/LAP expression we cultivated MCF-7 cells in hormone-depleted medium and treated the cells with estrogen (E) or progesterone (P) and did not observe significant changes in the LIP/LAP isoform ratio ( Supplementary Fig. 1b). This indicates that the difference in the LIP/LAP ratio between the hormone receptorpositive luminal A-derived and the hormone receptor-negative TNBC-derived cell lines is not due to the different ER or PR expression status.
C/EBPβ regulates migration-and microenvironment-related genes in TNBC cells To study the downstream transcriptional effects of the high LIP/ LAP isoform ratio, we analysed the transcriptional changes upon CRISPR/Cas9 mediated knockout (ko) of CEBPB in the TNBC cell line BT-20 by genome-wide RNA sequencing (RNA-seq) (Fig. 2a, Supplementary Fig. 2a and b and Supplementary Data 1). Differential expression analysis using the the R package DESeq2 27 revealed upregulation of 344 genes and downregulation of 196 genes in the CEBPB-ko BT-20 cells vs BT-20 wt cells ( Fig. 2a and Supplementary Data 1). That we found more genes upregulated is congruent with the loss of the predominantly expressed transcriptional repressor LIP in BT-20 cells. Grouping of the differentially expressed genes using the DAVID Functional Clustering tool (https://david-d.ncifcrf.gov/) revealed downregulated clusters involved in cell migration, actin binding and ECM remodelling (Fig. 2b). The most prominent upregulated gene clusters are antigen presentation and MHC I and MHC II expression (Fig. 2c). In a previous study we have performed transcriptome analyses using mouse embryonic fibroblasts (MEFs) derived from Cebpb-ko mice, which were reconstituted with LIP expression 28 . In this case, DAVID functional clustering analysis revealed that LIP overexpression causes downregulation of clusters involved in ECM-cell interaction, cell adhesion molecules and collagen subtypes ( Supplementary Fig. 2c) 28 . Therefore, we hypothesised that the C/EBPβ isoform ratio might regulate cell migration and ECM-cell interaction.

C/EBPβ regulates cell migration of TNBC cells
The observed downregulation of cell migration-related genes in BT-20 CEBPB-ko cells motivated us to investigate the effects of C/ EBPβ isoform expression on cell migration. Using the Boyden chamber migration assay we examined the migration of CEBPB-ko cells versus CEBPB-wt cells and observed a decrease in migration of CEBPB-ko cells after 48 h (Fig. 3a). In addition, a cell invasion assay using a 3D Boyden chamber with matrigel overlay showed a reduction in cellular invasion for the CEBPB-ko cells (Fig. 3b). To verify that LIP but not LAP drives migration, a cumate-inducible LAP construct was introduced in wt BT-20 cells to revert C/EBPβ expression from high LIP/LAP to low LIP/LAP ratios (Fig. 3c). An IncuCyte (Sartorius) life cell imaging scratch wound assay showed that cumate-induced LAP expression results in a significantly reduced migration rate (Fig. 3d). Due to some leaky induction of LAP in the absence of cumate ( Fig. 3c; compare -cumate LAP vs. EV) the cumate-uninduced LAP cells show intermediate decrease in cell migration compared to EV and cumate-induced LAP cells (Fig. 3d). A cell invasion assay using a 3D Boyden chamber with matrigel overlay showed that LAP induction in BT-20 cells reduced cell invasion ( Fig. 3e and Supplementary Fig. 3a). Similar results were obtained using the BT-549 TNBC cell line with introduced cumate-inducible LAP expression. Induction of LAP expression resulted in reduced cell migration using the Incucyte system (Supplementary Fig. 3b and c). As mentioned above transcriptome analysis showed that expression of LIP in Cebpb-ko MEFs affects genes involved ECM-cell interaction and cell migration (Supplementary Fig. 2c) 28 . We therefore examined the effects of LIP/LAP manipulation on cell migration of MEFs. MEFs derived from mice that are deficient in LIP expression (Cebpb ΔuORF ) showed a strongly reduced migration rate in scratch wound assay (Fig. 3f). In addition, re-expression of LIP in Cebpb-ko MEFs increased the migration rate while re-expression of LAP showed no effect compared to the empty vector (EV) control (Fig. 3g). Altogether, the data show that LAP inhibits migration and invasion of the TNBC cells BT-20 and BT-549, and that LIP promotes migration in MEFs and in untransformed MCF10A breast epithelial cells (Fig. 4a).

C/EBPβ regulates a subset of EMT markers
In aggressive breast cancer, TNBC in particular, cancer cells show a highly migratory behaviour that is associated with EMT 29,30 . To examine to what extent EMT markers are regulated upon induction of LIP and the associated migratory phenotype, we used the untransformed mammary epithelial cell line MCF10A as the experimental system. MCF10A-LIP overexpressing cells adopt a more spindle-shaped morphology associated with EMT (Supplementary Fig. 4) and show an increase in cell migration as was measured by Incucyte scratch assays (Fig. 4a). Analysis of transcript levels of EMT markers showed that overexpression of LIP resulted in decreased mRNA expression of the epithelial marker and cell adhesion molecule E-cadherin (E-cad), which is characteristic of EMT (Fig. 4b). Also, in line with EMT is the increase in the mesenchymal marker N-cadherin (N-cad) and in the ECM protein Fibronectin (FN1). However, the mesenchymal marker Vimentin (Vim) and the Matrix Metallopeptidase 2 (MMP2) involved in re-organisation of the ECM upon EMT show reduced expression upon LIP overexpression. To analyse to what extent expression of EMT transcripts are altered in the BT-20 cells that predominantly express LIP over LAP versus BT-20 CEBPB-ko cells we applied Gene Set Enrichment Analysis (GSEA) (https://www. gsea-msigdb.org/gsea; the Broad Institute) to our transcriptome data. The enrichment plot shows no clear positive or negative correlation in the regulation of the EMT hallmark gene set, but rather a mixed signature of upregulated and downregulated EMT genes ( Fig. 5a and Supplementary Data 1). For a more detailed insight, we generated a heat map of EMT genes that were detected in the transcriptome analysis (Fig. 5b). Within the top 15 most downregulated genes upon CEBPB-ko within the EMT gene set, a large subset of genes was found to be related to ECM organisation (Fig. 5b). These genes include Tenascin C (TNC), Collagen 5A1 (COL5A1), Fibronectin 1 (FN1), and Thrombospondin 1 (THBS1), and genes related to ECM remodelling, including  25 and in untransformed MCF10A supplemented with EGF required for proliferation and known to induced LIP. The blots below show β-actin as a low-molecular and vinculin as a high-molecular loading control. The bar graph shows the relative LIP/LAP isoform ratio quantification per cell line of the blot shown. The LAP signals of CAL-120 and MDA-MB-361 were too low for reliable quantification.
MMP2, Lysyl Oxidase Like 1 (LOXL1) and Procollagen C-Endopeptidase Enhancer 2 (PCOLCE2), which are all known to contribute to breast tumour progression via increased ECM deposition and remodelling as reviewed in 31 . From these ECMrelated genes we chose a subset for verification of the downregulation in two independent BT-20 CEBPB-ko clones by RT-qPCR (Fig. 6). Downregulation of the ECM genes THBS1, TNC, COL5A1, MMP2 in response to the CEBPB-ko could be confirmed in both knockout clones while a reduction of FN1 expression was detected only in one of the two knock-out clones. The ENCODE database (http://genome.ucsc.edu/ENCODE/) records C/EBPβ-associated fragments associated with H3K acetylation (H3K27Ac) marking transcriptional active regions for all five genes (Fig. 6b). In addition, protein levels of THBS1 and TNC were reduced in the two clones of BT-20 CEBPB-ko cells, and protein levels of FN1 and COL5A1 were reduced in one CEBPB-ko clone (Fig. 6c). Altogether, the data point to a "partial" EMT regulation by C/EBPβ.

C/EBPβ-LAP induces the expression of ECM-associated genes
The ECM is involved in cancer progression and metastasis at several levels ranging from the EMT to engraftment and outgrowth at distant sites [32][33][34] . To examine whether LIP or LAP alters expression of the ECM-related genes Tenascin (TNC), Collagen Type V alpha 1 (COL5A1), Fibronectin 1 (FN1), Thrombospondin 1 (THBS1) and (Matrix Metallopeptidase 2 (MMP2) we used the BT-20 system (Fig. 7a). CEBPB-ko in BT-20 cells results in downregulation of all five genes (Fig. 6a), suggesting that either LIP and/or LAP contribute to their activation. Re-expression of LIP in BT-20 CEBPB-ko cells does not alter the levels of the transcripts compared to the levels in the parent cells, suggesting that in absence of LAP, LIP does not further reduce the expression (Fig. 7b). Re-expression of LAP, however, strongly induces all five ECM transcripts. Note that the uninduced (-cumate) cells, because of some "leaky" induction of LAP, already show elevated LAP levels compared to EV control (Fig. 7a) and concomitant higher ECM transcript levels. These results indicate that in wt BT-20 cells the low levels of LAP relative to LIP (high LIP/LAP ratio) (Fig. 7a) is sufficient and required to activate ECM genes, which is beneficial for tumour progression and metastasis. Finally, to investigate whether the LIP/LAP ratio changes upon inducers of EMT we treated MCF10A cells with TGFβ which is known to induce EMT in these cells 35 . The induction of EMT was monitored by the upregulation of N-cadherin which was accompanied by a strong and permanent decrease of LAP levels and a slight and transient decrease in LIP levels resulting in a strong increase of the LIP/LAP ratio at the end of the treatment period (Fig. 8). These data support the hypothesis that a high LIP to LAP ratio is involved in the EMT process.

DISCUSSION
TNBC is an aggressive type of breast cancer characterised by the lack of expression of the hormone receptors ER and PR and the growth factor receptor HER2. These receptors are used as therapeutic targets for breast cancer types proficient for at least one of these factors. Their absence in TNBC makes treatment far more difficult and contributes to the poorer prognosis. Thus, a better understanding of other factors involved in TNBC is essential for the development of more effective therapies. In this study,  Statistical differences were analysed for timepoints 12 h and 24 h using a -T-test. b Expression of epithelial and mesenchymal genes relative to GAPDH in MCF10A cells containing a LIP expression construct or empty vector control (EV) (n = 3). Results are analysed using a T-test. Error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Fig. 3 LIP and LAP differentially regulate cell migration and invasion. a Representative images of Boyden chamber migration assay of BT-20 wt and two cellular clones of CEBPB-ko cells with at the right a bar graph of corresponding quantifications (wt, n = 3; CEBP-ko n = 4). One-way ANOVA analysis; error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0. 001, ****p < 0.0001. scale bar represents 100 μm. b Bar graph with representative quantification of 3D Boyden chamber-matrigel invasion assay using BT-20 wt and BT-20 CEBPB-ko clone #1 (BT-20 wt, n = 6; and BT-20 CEBPB-ko, n = 6). Statistical differences were analysed after 48 h using an unpaired T-test. Error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0. 001, ****p < 0.0001. c Immunoblot showing the expression of LAP and LIP in BT-20 cells containing a cumate-inducible LAP construct or empty vector control (EV), two cellular clones of each. β-actin was used for loading control. d Time course of Incucyte scratch wound migration assay shown as relative wound density of BT-20 cells containing a cumate-inducible LAP construct (clone #1) or empty vector control (EV, clone #1) induced with cumate (n = 4). Statistical differences were analysed for timepoints 24 and 48 h using Two-way Anova analysis and Tukey's multiple comparisons test (BT-20 EV− cumate vs BT-20 EV+ cumate were compared (ns), and BT-20 LAP− cumate vs BT-20 LAP+ cumate were compared (**)). Error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0. 001. e Bar graph shows quantification of 3D Boyden chamber-matrigel invasion assay using BT-20 cells containing a cumate-inducible LAP construct (clone #1) or empty vector control (EV, clone #1) (BT-20 EV, n = 3; BT-20 LAP, n = 4). Statistical differences were analysed after 48 h using two-way Anova analysis. Error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0.001. f Immunoblot shows the expression of LAP and LIP in wt MEFs and LIP-deficient Cebpb ΔuORF MEFs with the respective LIP/LAP ratios shown underneath. α-tubulin was used for loading control. The bar graph shows relative migration rate of a scratch wound assay (n = 5). Statistical differences were analysed by T-test. Error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. g Immunoblot shows the re-expression of LAP and LIP in Cebpb-ko MEFs as well as empty vector control (EV). β-actin was used for loading control. The bar graph shows relative migration rate of a scratch wound assay (n = 5). Statistical differences were analysed by T-test. T Error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. The immunoblots from (f) and (g) are taken from from Ackermann et al. 28 .
Initial steps of migration often require cells to lose adhesion molecules and epithelial markers and gain mesenchymal markers, in a process called EMT 29,30 . A few studies have addressed the role of C/EBPβ in breast cancer development and cancer cell migration, however, the data on isoform-specific roles are sparse and partially contradictory. For example, it was demonstrated that LIP supports anchorage-independent growth of breast epithelial cells through suppression of anoikis, a form of programmed cell death that occurs in anchorage-dependent cells when they detach from the surrounding ECM, which is a prerequisite for metastasis formation 36 . LIP was furthermore shown to upregulate the expression of the chemokine receptor CXCR4 and the pro-invasive CDH3/P cadherin in breast cancer cells, both known drivers of metastasis 5,37 . In addition, it was demonstrated that overexpression of LIP in the SCp2 mouse mammary epithelial cell line drives EMT using in vitro assays 38 . The downregulation of all C/EBPβ isoforms through microRNA-155 was shown to induce an EMT phenotype and migration and invasion in mouse mammary epithelial cells through a pathway that involves TGFβ signalling 39 . This observation is in line with another publication showing that shRNAmediated downregulation of C/EBPβ expression induces the formation of metastasis in a mouse 4T1 mammary carcinoma cell line transplantation model 40 . The cell types in the latter two studies express low LIP/LAP ratios and the complete knockout or suppression of C/EBPβ may largely reveal effects through loss of the anti-migratory functions of LAP. In contrast, it was demonstrated that LAP induces the EMT and an invasive phenotype when overexpressed in untransformed MCF10A mammary epithelial cells through relocation of E-cadherin to the nucleus, mimicking loss of E-cadherin function 41 . The reason of these partially contradictory findings is not known, yet our data from migration and invasion assays support a pro-migratory role of LIP and suggest an anti-migration and anti-invasion function for the LAP isoform. The observation that the LIP/LAP ratio strongly increases upon TGFβ-induced EMT in MCF10A cells clearly points to an important role of LIP in the EMT process. In further support of a pro-migratory function of LIP is that we observed an increased migration of MCF10A cells by exogenous LIP expression, which is accompanied by a downregulation of E-cadherin and an upregulation of the mesenchymal marker N-cadherin and the migration and invasion promoting ECM protein Fibronectin, but not of the mesenchymal marker Vimentin and the ECM remodelling metalloprotease MMP2, suggesting a partial induction of EMT upon LIP overexpression. Furthermore, GSEA of the differentially regulated genes in the BT-20 CEBPB-ko cells revealed a mix of upregulated and downregulated EMT-related genes. This suggests that LIP may contribute to the EMT process likely in collaboration with other factors. Recent studies pointed out a greater plasticity between the epithelial and mesenchymal states, where cancer cells are observed to be in a spectrum of intermediary phases [32][33][34]42 . Single-cell migration often favours complete EMT to escape the tumour and detach from surrounding epithelial cells, whereas collective cell migration often is characterised by cells undergoing a partial EMT where the cancer b a THBS1  ANPEP  INHBA  BASP1  TNC   MSX1  FNO2  ITGB1  SNAI2  CXCL1  P3H1  LUM  PLOD1  GPC1  TGFBI  MATN3  ADAM12  COPA  SERPINH1  TTMP1  COL16A1  PFN2  MAGEE1  TNFRSF12A  EMP3  CTHRC1  ECM2  FBN1  PMP22  THBS2  PLOD2  NID2  COL4A1  CCN2  CALU  SAT1  NT5E  PVR  FAP  COL1A1  LGALS1  COL4A2  TAGLN  COLGALT1  CALD1  BMP1  CD44  OXTR  TNFRSF11B  TPM4  FGF2  MCM7  SLC6A8  TGM2  PLOD3  SDC1  PPTB  TTGB5  THY1  PDGFRB  MYL9  GLIPR1  SGCB  TPM1  MEST  CAP2  VEGFA  FERMT2  COL12A1  IL6  PTX3  COL7A1  FLNA  PMEPA1  VIM  FBLN2   IL32  COL6A2  LRRCL5  TD2  GADD45A  LAMC2  SPOCK1  TGFBR3  FFFMP2  FLN  MYLK  DKK1  RGS4  DST  LRP1  TNFAIP3  GREM1  TIMP3  CD59  DAB2  ABI3BP  LAMC1  GJA1  ECM1  ITGA5  BDNF  LOX  RHOB  ITGA2  IGFBP3  GEM  IL15  LAMA3  PCOLCE  CAPG  FUCA1  LAMA2  FAS  COL1A2  ITGB3  CCN1  COL5A2  PDLIM4  CRLF1  MMP14  MATN2  TGFB1  AREG  MGP  QSOX1  NOTCH2  SDC4  VEGFC  SCG2  ACTA2  FBLN1  APLP1  JUN  COL6A3  SERPINF1  PRSS2  LOXL2  MFAP5  GADD45B  CDH2  ITGAV  PLAUR  WIPF1  SLIT2  HTRA1  TPM2  cells retain some capacity of cell-cell adhesion and migrate in a collective manner 42 . This collective cell migration is often found in epithelial cancers, where the leader cells acquire a more mesenchymal phenotype by becoming motile and breaking down the surrounding ECM, leading the way for the follower cells 43 . Although only to be clarified by future studies, this may explain the mix of upregulated and downregulated EMT-genes upon modulation of LIP expression in our analyses. Studies have revealed an essential role of the ECM in tumour initiation, metastasis and the immune responses 33,44,45 . It has been reported that HER2-positive and TNBC tumours show an increased collagen content and ECM stiffness compared to luminal A and luminal B subtypes and that this is linked with the increased attraction of tumour-associated macrophages, TGFβ signalling and worse prognosis 46 . Although cancer-associated fibroblasts (CAFs) are the main producers of ECM, several studies have shown that interference with ECM production in cancer cells results in a significant delay in tumour progression. The ECM not only directs the cell migration by providing directional cues, but also provides chemical and physical signals to the leader cells, changing their migratory behaviour 47,48 . For example, it was reported that breast cancer cells produce Tenascin (TNC) as a metastatic niche component to prime to the lungs for breast cancer metastasis 49 . The matrix metallopeptidase 2 (MMP2) that is known to break down the basement membrane was shown to promote cancer invasion 50,51 . Fibronectin 1 (FN1) was shown to be expressed by leading invasive cancer cells and was linked to poor metastasisfree and overall survival in breast cancer 52,53 . Circulating Thrombospondin 1 (THBS1) is a marker for aggressive breast cancer 54 and exposure to stromal THBS1 promotes breast cancer progression 55 . Our GSEA revealed a downregulation of ECMrelated genes upon CEBPB-knockout in BT-20 cells, and LAP reexpression in BT-20 CEBPB-ko cells results in upregulation of prometastatic factors FN1, TNC, THBS1, COL5A1 and MMP2. Although this seems at odds with the inhibition of cell migration and invasion by LAP it points to a possible function of low LAP levels in transcriptional activation of these pro-metastatic genes. These results suggest that exclusively considering LIP as a competitive inhibitor of LAP maybe too simple and that we have to differentiate between LIP and LAP functions in tumour biology as Spike and Rosen describe in their recent review "C/EBPβ isoform-specific regulation is more complicated than we may think" 24 . LAP and LIP can dimerise with all available C/EBPα-ζ proteins in the cell with variable effects on target gene regulation.

CEBPB-ko
In addition, they have been shown to interact with other transcription factors like for example NFκB 56,57 , c-Jun 6 or YY1 11,58 , further diversifying C/EBPβ functions. Nevertheless, there is strong and accumulating evidence that C/EBPβ-LAP and -LIP specific regulation is crucially involved in the development and/or progression of TNBC, and therefore a better understanding of their oncogenic functions is important in order to address potential suitability as a pharmacological target. Previously, we demonstrated in a screening approach using FDA-approved drugs that the LIP/LAP ratio is druggable also apart from known mTORC1 inhibitors 59 . Therefore, the LIP/LAP ratio could potentially be an interesting clinical target in TNBC. Taken together, we provide evidence that the LIP/LAP ratio regulates a subset of EMT and ECM-related genes and the migration/invasion potential of breast cancer cells. The  3). b-f Expression of extracellular matrix-associated genes in CEBPB-ko clone #7 upon cumate-induced LIP and LAP expression relative to GAPDH. Differences analysed using a two-way ANOVA analysis and Tukey's multiple comparison test. Error bars represent SD, *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001.
using breast cancer mouse models with additional mutations altering expression of LIP and LAP separately in tumour cells.

Cell lines
For generation of cell lines stably expressing cumate-inducible C/EBPβ-LIP and C/EBPβ-LAP constructs or an EV construct, HEK293T cells were transfected for lentivirus production. 4.5 × 10 6 HEK293T cells were plated in 10-cm culture dishes. Twenty-four hours later, transfection was performed using the calcium phosphate method. The next morning, medium was changed and the morning thereafter, virus was harvested and BT-20 cells and BT-549 cells were infected with a cumate-inducible human C/EBPβ-LIP and C/EBPβ-LAP constructs or an EV construct using a standard protocol. Two days after infections cells were selected with puromycin 1 µg/ml for BT-20 and 0.5 µg/ml for BT-549. For overexpression of LIP in MCF10A cells, cells were infected with PLVX-IRES-Neo containing human C/EBPβ-LIP and selected using 1 mg/ml G418.

DNA constructs
For overexpression of human C/EBPβ-LIP and C/EBPβ-LAP, the coding sequence was amplified from MCF-7 genomic DNA and the start codon of LAP was changed to an optimal Kozak sequence (ATCCATGGAAGTG to AGCCATGGAAGTG). The amplified PCR product was cloned into the pCDH-CuO-MCS-IRES-GFP-EF1α-CymR-T2A-Puro SparQTM All-in-one Cloning and

Immunoblot analysis
For protein extraction, the cells were washed twice with ice-cold 1 × PBS and lysed in 50 mM Tris pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, supplemented with protease and phosphatase inhibitors followed by sonication. Equal amounts of total protein were separated by SDS-PAGE, transferred to a PVDF membrane using either the Trans-Blot Turbo System (#170-4273, BIO-RAD) following the manufacturer's protocol or tank blotting in 2.5 mM TRIS base, 20 mM glycine, 15% methanol.

Cellular migration assays
Prior to seeding for Incucyte scratch migration and invasion assays, cells were serum starved (in 1% serum). Cell migration and invasion was live monitored in the Incucyte Zoom using 96-well imagelock plates. BT-20, BT-549 and MCF10A cells were seeded at densities of 35,000 cells/well, 20,000 cells/well, and 20,000 cells/well respectively. Twelve hours post seeding a scratch wound was induced using the woundmaker, followed by washing of cells and incubation in low-serum medium. Cells were imaged every 2 h and using the incucyte zoom2018 software the Relative Wound Density was calculated. For Incuyte scratch invasion assays, wells were coated with 50 μl of 100 μg/ml matrigel (growth factor reduced, Corning # 354230) diluted in medium and embedded in 1 mg/ml matrigel in medium. For the MEF wound healing assay, the backside of each well of a six-well plate was marked with three lines. 2 × 10 6 cells were seeded in prepared six-well plates 24 h prior to the assay. To induce a wound in the cell monolayer, tips of a 200 µl pipet were used to scratch in a 90°angle to the markings along the plate. Cells were washed once with PBS and medium was replaced. Pictures were taken every 4 h at the intersection between the scratch and the drawn line for 28 h. Pictures were quantified with imageJ software.

Transwell cellular invasion assays
Prior to seeding for migration and invasion assays, cells were serum starved (in 1% FBS). Cells were seeded at densities of 20,000 cells/well in serum containing 1% medium in transwell permeable supports (6.5 mm insert, 8 μm polycarbonate membrane, corning #4322) and attracted to medium containing 10% FBS in the lower compartment. After 24 or 42 h, cells were fixed in 4% PFA and stained using 0.5% crystal violet. For imaging, three random images using the 5× objective were taken and average scores were plotted. For invasion assays, transwells were coated with 250 μg/ml matrigel in medium containing 1% FBS.

Transcriptome analysis
RNA was isolated from three pools of BT-20 WT cells and three BT-20 CEBPB-ko clones, using the RNeasy Plus Mini Kit (QIAGEN). Quality and quantity of RNA was determined using Agilent's Bioanalyzer 2100 in combination with the RNA 6000 nano chip (Agilent). Library preparation was done using Illumina's TruSeq RNA v2 kit following the manufacturer's description. The libraries were quality checked and quantified using Bioanalyzer 2100 in combination with the DNA 7500 kit (Agilent). Sequencing was done on a NextSeq500 in 75 bp, single-end sequencing, high-output mode. Sequence information was extracted using bcl2fastq v1.8.3 (Illumina), Sequencing resulted in around 50 million reads per sample. The quality of the reads was checked using FastQC (v. 0.11) and filtering and trimming of low-quality reads were performed using Trimmomatic (v. 0.33). Trimmed reads were aligned to the GRCh38/hg38 genome (genome annotation from ensemble release 92) using STAR aligner v. 2.6.0b 60 . Identification of differentially expressed genes (DEGs) was done using the R package DESeq2 27 . The resulting p-values were adjusted using Benjamini and Hochberg's approach for controlling the false discovery rate 61 . Genes were regarded as differentially expressed if adjusted p-values were <0.05. Functional clustering analysis was performed using the DAVID database (https://david.ncifcrf.gov/, version 6.7) at default settings with medium stringency (downregulated genes Fig. 2d) or highest stringency (Fig. S1A). GSEA was performed using the GSEA desktop application and compared to the Hallmark gene sets. The phenotype label was set as CEBPB WT vs CEBPB-ko. The t-statistic mean of the genes was computed for each hallmark gene set using a permutation test with 1000 replications.

Statistics and reproducibility
For analysis of differentially expressed genes (transcriptome) we refer to the specific sections above. Other statistical differences were calculated by Student's t-test or two-way ANOVA as indicated using GraphPad Prism 8. Data are presented as the mean ± standard deviation (SD). Statistically significant differences are indicated with *p < 0.05, **p < 0.01, ***p < 0.001.