Enterobacteria impair host p53 tumor suppressor activity through mRNA destabilization

Increasing evidence highlights the role of bacteria in the physiopathology of cancer. However, the underlying molecular mechanisms remains poorly understood. Several cancer-associated bacteria have been shown to produce toxins which interfere with the host defense against tumorigenesis. Here, we show that lipopolysaccharides from Klebsiella pneumoniae and other Enterobacteria strongly inhibit the host tumor suppressor p53 pathway through a novel mechanism of p53 regulation. We found that lipopolysaccharides destabilize TP53 mRNA through a TLR4-NF-κB-mediated inhibition of the RNA-binding factor Wig-1. Importantly, we show that K. pneumoniae disables two major tumor barriers, oncogene-induced DNA damage signaling and senescence, by impairing p53 transcriptional activity upon DNA damage and oncogenic stress. Furthermore, we found an inverse correlation between the levels of TLR4 and p53 mutation in colorectal tumors. Hence, our data suggest that the repression of p53 by Enterobacteria via TLR4 alleviates the selection pressure for p53 oncogenic mutations and shapes the genomic evolution of cancer.

Over the last decades, the human microbiome has emerged as an important regulator of the physiopathology of cancer and it has been estimated that infections could be the main driver for over 20% of cancers [1]. Recent evidence highlights the role of the bacterial microbiota in cancer initiation, progression and response to therapy [2]. Enrichment of specific bacterial species has been found in cancer patients by metagenomics study, linking bacterial dysbiosis to cancer [3,4]. Moreover, the ability for oncogenic bacteria to act as cancer drivers and to initiate tumorigenesis has been demonstrated for Helicobacter pylori [5] as well as for genotoxin-producing bacteria [6,7]. Several other bacterial species have also been experimentally associated to increased cancer risk, such as Fusobacterium nucleatum [8]. Overall, the essential role of the bacterial microbiota in cancer is now well established. However, despite recent advances, the underlying molecular mechanisms by which our microbiota influences cancer remain elusive.
Bacteria produce multiple toxins, metabolites and proinflammatory molecules which directly or indirectly target host signaling pathways involved in all the cancer hallmarks and thereby interfere with the host defense against tumorigenesis, such as the p53 pathway. The tumor suppressor p53, encoded by the TP53 gene, is a transcription factor which acts as a cell signaling hub to integrate various cell stresses into an appropriate cell response [9]. Upon stress, phosphorylation events lead to the dissociation of p53 from its ubiquitin-ligase MDM2, preventing its degradation by the proteasome and leading to its accumulation in the nucleus. p53 controls transcription programs regulating multiple cell functions including apoptosis, cell cycle arrest and DNA damage response, and acts as the main barrier against tumorigenesis. Additionally, p53 has been long known to be involved in host defense against viral infection, due to its role in the DNA damage response, and hence it is specifically targeted by oncogenic viruses [10,11]. Interestingly, known oncogenic bacteria such as H. pylori [12][13][14], Chlamydia trachomatis [15] and several Mycoplasma species [16] also interfere with the p53 pathway. Altogether, these studies suggest that p53 inactivation could be a more general hallmark of microbes with tumorigenic potential.
Klebsiella pneumoniae is an opportunistic pathogen of the digestive and upper respiratory tract. Metagenomics and clinical epidemiologic data suggest that K. pneumoniae could play a role in cancer initiation and progression. K. pneumoniae has been shown to be enriched in the gut of colorectal cancer patients (CRC) [17]. Moreover, hypervirulent strains of K. pneumoniae, which cause pyogenic liver abscesses, have been associated with increased risk of CRC [18,19]. Interestingly, several hypervirulent K. pneumoniae strains harbor the pks locus encoding for the genotoxin colibactin which could be involved in K. pneumoniae tumorigenic potential [20]. However, the mechanisms that link K. pneumoniae and cancers are still poorly understood. In this study, we investigate the impact of K. pneumoniae on the host p53 pathway. We find that lipopolysaccharide (LPS) from K. pneumoniae and other Enterobacteria inhibits p53 through the TLR4-NF-κB pathway, and we uncover a novel mechanism of p53 regulation whereby p53 inhibition occurs at the mRNA level.

Klebsiella pneumoniae inhibits p53
To investigate the effect of K. pneumoniae on the host p53 pathway, we infected human immortalized fibroblasts (BJ hTERT) with live bacteria. We observed a robust downregulation of p53 protein level, concomitant with the activation of NF-κB, as demonstrated by Ser536 phosphorylation of the NF-κB subunit p65 (RelA) (Fig. 1a). Interestingly, p53 downregulation was bacteria-cell contact independent and was recapitulated when treating the cells with supernatant from K. pneumoniae culture (KpSN), independently of the K. pneumoniae strains used ( Fig. 1b and Supplementary Fig. S1a-b). Pathway analysis of gene expression changes in human fibroblasts exposed to KpSN, measured by RNA-seq, highlighted a robust activation of the canonical immune and inflammatory pathways involved in the host response to bacteria, as expected ( Fig. 1c-d). Notably, the analysis of genes downregulated upon KpSN revealed a significant inhibition of the p53 signaling pathway and DNA damage response (Fig. 1d). These data suggest that K. pneumoniae-mediated p53 protein decline is sufficient to impair the biological function of p53.    Klebsiella pneumoniae impairs the p53 response to DNA damage and oncogenic stress To confirm that K. pneumoniae-induced downregulation of p53 impacts its transcriptional activity, we investigated the effect of KpSN on p53 level and activity upon doxorubicin-induced DNA damage, which is well known to stabilize p53 and activate its target genes. First, we confirmed that upon DNA damage, the stabilization of p53 is impaired in KpSN treated cells (Fig. 2a). Then, we investigated how it affects p53-dependent transcription. The set of genes regulated by p53 is highly dependent on the cell context [21]. Therefore, we defined a gene set of 161 high confidence p53 targets in our model. We intersected genes upregulated upon DNA damage in a p53-dependent manner in BJ-hTERT fibroblasts (which were not induced by doxorubicin upon p53 shRNA according to our RNA-seq data) with the p53 core targets gene set defined by Fisher et al. [22] (Fig. 2b). Then, we compared how KpSN affected doxorubicin-induced activation of our gene set of 161 high confidence p53 targets. Our analysis revealed an overall inhibition of p53 targets by KpSN upon DNA damage (Fig. 2c). Using RT-qPCR, we investigated the expression of well characterized p53 target genes PMAIP1 (NOXA), CDKN1A (p21), TP53I3 (PIG3), SESN1, as well as ZMAT3 (Wig-1) and TP53INP1 which have been recently highlighted as major effectors of p53mediated tumor suppression [23,24] (Fig. 2d-e). Interestingly, both PMAIP1 and CDKN1A were upregulated by KpSN in a p53independent manner. However, KpSN partially prevented a robust p53-mediated induction of CDKN1A upon doxorubicin and strongly impaired the induction of TP53I3, SESN1 and TP53INP1.
Notably, while we observed a diminished ZMAT3 induction upon DNA damage, ZMAT3 was repressed by KpSN in a p53independent manner (Fig. 2e). We then confirmed the KpSNinduced impairment of the p53 response to DNA damage at protein level (Fig. 2f).
To investigate the biological consequences of K. pneumoniaeinduced inhibition of p53, we used BJ fibroblasts with an inducible oncogenic Ras system (BJ hTert HRasV12 ER-Tam ) as a model of oncogenic stress. In these cells, activation of oncogenic Hras by 4-hydroxytamoxifen (4-OHT) induces a strong p53-and p21dependent senescence [25]. We monitored senescence by increase in β-galactosidase activity ( Fig. 3a-b) and by cell replication arrest revealed by EdU incorporation assay ( Fig. 3c-d).
We found that KpSN impairs p53-dependent induction of senescence upon oncogene activation, as well as prevents accumulation of p53 and p21 (Fig. 3e).
Overall, our data show that secreted factors from K. pneumoniae downregulate p53 protein level upon DNA damage and oncogenic stress, which impairs its transcriptional activity and its tumor suppressive function.     . c p53 protein stability in BJ hTert cells exposed to KpSN (orange) or 100 ng/mL LPS (pink) was investigated using cycloheximide (CHX) chase assay. Cells were pretreated 4 h with KpSN before CHX treatment. Lower panels show the densitometry quantification of Western blots bands normalized to β-actin level. d RT-qPCR for TP53 mRNA level of BJ hTert cells upon 8 h exposure to KpSN, assessed by. e, RT-qPCR for TP53 mRNA level of BJ hTert cells upon 8 h exposure to indicated dose of LPS. f RT-qPCR for TP53 mRNA level upon KpSN or 100 ng/mL LPS (8 h) in PBMC-derived macrophages, polarized into M1 phenotype. The different data point symbols indicate different blood donors. g Luciferase reporter assay using TP53 promoter-Luc construct transfected in BJ hTert cells. Transfection rate was normalized by co-transfection of Renilla luciferase vector. Endogenous TP53 mRNA level was measured in the same samples. h Luciferase reporter assay using a construct expressing the luciferase flanked by p53 UTRs, transfected in BJ hTert cells. Transfection rate was normalized by co-transfection of Renilla luciferase vector. Endogenous TP53 mRNA level was measured in the same samples. i Upper panel: structure of the TP53 gene with the two alternative promoters P1 and P2. Lower panel: mRNA level of P1 and P2 p53 isoforms in BJ hTert cells upon treatment with KpSN or 100 ng/mL LPS (8 h). *p < 0.05; ***p < 0.01; NS non-significant.  lipid A, completely rescued the downregulation of host p53 (Fig.  4c). To confirm the involvement of lipid A, we generated a K. pneumoniae deletion mutant for the lpxM gene, which encodes a myristoyl transferase that catalyzes the final step of lipid A biosynthesis ( Supplementary Fig. S2b). Deletion of lpxM did not affect bacterial growth nor the amount of LPS secreted ( Supplementary Fig. S2c, d). However, K. pneumoniae ΔlpxM was unable to affect p53 protein level in host cells, neither upon direct infection nor upon treatment with bacterial supernatant, while the complemented strain restored the downregulation (Fig. 4d).
Altogether, these results demonstrate that K. pneumoniae inhibits p53 through the lipid A of its LPS.
To further examine which host pathway is involved, we performed similar experiments while inhibiting the classical LPS signaling pathway, including its receptor TLR4 or the major cellular mediator NF-κB. Both TLR4 inhibitor and Tet-inducible knockdown of the NF-κB subunit p65 completely rescued the inhibition of host p53 (Fig. 4e, f). As the TLR4-NF-κB axis is a canonical signal transduction pathway activated by LPS from most Gram-negative bacteria, we investigated if the effect on p53 is shared with other bacterial species (Fig. 4g and Supplementary Fig. S2e). Indeed, culture supernatants from other Enterobacteria (Salmonella typhimurium, Shigella flexneri, Escherichia coli) also strongly repressed p53. In contrast, p53 was activated by Gram-positive Streptococcus pneumoniae. Importantly, we did not observe p53 repression by LPS from commensal Gram-negative Bacteroides fragilis, which is known to have a penta-acetylated lipid A structure acting as TLR4 antagonist [26]. However, p65 was phosphorylated to a similar extent, suggesting that besides the canonical NF-κB pathway, other transduction pathways are involved in p53 repression.
Next, we explored if the inhibition of p53 by LPS/TLR4 could play a role in tumor initiation or progression, using data from cancer patients deposited in the Cancer Genome Atlas (TCGA). We hypothesized that inhibition of p53 by bacterial signaling could alleviate the selection pressure for p53 mutation in TLR4 expressing tumors. Hence, we investigated the relationship between the TLR4 expression level and the p53 mutation rate in early-stage colorectal adenocarcinoma (grade I-II). We observed a robust correlation (p < 0.0001) between TLR4 mRNA level and absence of a p53 mutation in the tumors (Fig. 4h). Moreover, comparison between tumors with high TLR4 (Z-score > 0.75) and low TLR4 (Z-score < −0.75) revealed a striking inversion of the p53 mutation rate (Fig. 4i). While these data could be explained by a transcriptional control of TLR4 by wild-type p53, as previously suggested by Menendez et al. [27], we did not observe the same correlation in other tumor contexts known to have no or low bacterial load (such as breast and liver cancers) or with a lower abundance of Gram-negative bacteria (such as lung cancer) [28] ( Supplementary Fig. S3a, b). These data suggest that activation of the TLR4-NF-κB by LPS from the microbiota could contribute to tumorigenesis by restricting p53 tumor suppressive function and shed a new light on the known pro-tumorigenic effect of LPS [29].
p53 inhibition occurs at the mRNA level p53 has been extensively described to be regulated solely at the protein level, through modification of its stability, conformation and/or protein interactions [9]. To ensure a fast response to stress, p53 is constantly produced and targeted to proteasomal degradation through MDM2-dependent polyubiquitination. Several bacterial toxins have been described to enhance p53 degradation through AKT-mediated activation of MDM2 [30]. Calpain-cleavage of p53 by bacteria has also been reported [31]. Surprisingly, we found that downregulation of p53 by KpSN did not involve enhanced proteasomal degradation. While p53 was indeed stabilized upon proteasome inhibition by MG132 or MDM2 inhibition by Nutlin, we did not observe any prevention of the p53 decline induced by KpSN or by K. pneumoniae direct infections ( Fig. 5a-b). Similarly, doxorubicin-induced DNA damage (which leads to disruption of the p53-MDM2 complex and p53 stabilization) did not rescue KpSN-induced p53 downregulation (Fig. 2a). Inhibition of AKT or calpain also both failed to rescue p53 protein level (Supplementary Fig. S4a). Moreover, analysis of p53 protein stability by cycloheximide chase assay did not reveal any significant change in p53 stability when cells were exposed to KpSN or LPS (Fig. 5c).
Since our data showed that p53 protein stability was not affected, we investigated the TP53 mRNA by RT-qPCR and found that KpSN strongly decreased TP53 mRNA level in human immortalized fibroblasts (BJ hTert) (Fig. 5d-e), normal primary fibroblasts (HNF) and in A375 melanoma cells (Supplementary Fig.  S4b-c). This effect was recapitulated in a dose-dependent manner upon treatment of cells with K. pneumoniae LPS (Fig. 5e and Supplementary Fig. S4d). We also observed the repression of TP53 mRNA level in M1 and M2 monocytes-derived macrophages upon treatment with KpSN or LPS ( Fig. 5f and Supplementary Fig. S4e). Finally, inhibition of TLR4 or deletion of K. pneumoniae lpxM prevented TP53 mRNA downregulation ( Supplementary Fig. S4f, g).
TP53 gene transcription has previously been described to be repressed by phosphorylated STAT3 upon LPS [32]. However, we did not observe a rescue of TP53 mRNA in STAT3-depleted cells ( Supplementary Fig. S4h-i). Taken together, our results suggest a novel mechanism of p53 regulation by LPS-TLR4-NF-κB through STAT3-independent regulation of TP53 mRNA.
To identify the mechanism of TP53 mRNA inhibition by K. pneumoniae, we tested whether TP53 mRNA expression was affected at the promoter level or post-transcriptionally through TP53 mRNA UTRs. We performed dual luciferase reporter assays using a construct where the luciferase expression is under the control of the TP53 promoter (TP53 promoter-Luc) and another construct constitutively expressing the luciferase mRNA flanked by TP53 untranslated regions (Luc-TP53 UTRs). Inhibition of endogenous TP53 was controlled in the same sample by RT-qPCR. We did not observe any change in luciferase expression of TP53 promoter-Luc upon treatment with KpSN or LPS (Fig. 5g), suggesting that the promoter activity of TP53 is not affected. However, both KpSN and LPS induced a robust decrease in luciferase from the Luc TP53 UTRs plasmid (Fig. 5h), indicating that KpSN and LPS might regulate TP53 mRNA stability through its UTRs. To find out which UTR was involved, we investigated the regulation of different p53 isoforms (Fig. 5i). TP53 gene encodes two families of p53 isoforms through an alternative promoter P2 located in intron 4, which regulates the expression of short isoforms (Δ133 and Δ160 p53 isoforms), while the full length isoforms are regulated by its main promoter P1 [33]. P1 and P2 isoforms share their 3ʹ-UTR but have different 5ʹ-UTR sequence and different promoter regulation. Using RT-qPCR with isoformspecific primers, we found that both isoform families are similarly repressed by KpSN and LPS, suggesting a regulation of the TP53 3ʹ-UTR (Fig. 5i).

LPS-induced downregulation of ZMAT3 destabilizes TP53 mRNA
We then investigated the expression of genes known to regulate TP53 mRNA by binding its 3ʹUTR (Fig. 6a) [34]. Interestingly, we observed that several genes encoding factors stabilizing TP53 mRNA (such as ZMAT3 or CPEB1) [35,36] were repressed upon exposure to KpSN, while factors known to inhibit its translation (such as Tia1) [37] where activated. Strikingly, ZMAT3, which encodes the RNA binding protein Wig-1, was one of the top repressed genes in our RNA-seq data (Fig. 1c). Wig-1 has been shown to bind to an A/U-rich element (ARE) on TP53 3ʹUTR to stabilize TP53 mRNA [35]. Moreover, ZMAT3 forms a regulatory feedback loop with p53 since it is a p53 target gene crucial for p53 tumor suppression [23,24]. As we already showed that ZMAT3 repression upon KpSN is p53 independent (Fig. 2e), we tested if Wig-1 was inhibited at earlier time points (8 h) than other p53 target genes. We observed the downregulation of both isoforms of Wig-1 at the same time frame as TP53 mRNA but before other p53 targets such as PIG3 (Fig. 6b and Fig. 2f). Moreover, Wig-1 was repressed at mRNA and protein level in M1 and M2 monocytesderived macrophages (Fig. 6c, d and Supplementary Fig. S5a-b). Analysis of RNA-seq data from PBMC-derived macrophages from 60 different blood donors infected by Salmonella thyphimurium [38] confirmed the robust inhibition of both TP53 and ZMAT3 upon direct infection of macrophages by another Enterobacterial species (Fig. 6e).
To confirm the role of Wig-1 in p53 downregulation, we validated the binding of Wig-1 to TP53 mRNA by RIP-qPCR in our model (Fig.  6f). Then, we established fibroblasts stably overexpressing wild type Wig-1 (BJ hTert pLV Wig-1) or a zinc-finger mutant unable to bind TP53 mRNA [35] (BJ hTert pLV Wig-1*H88A). Overexpression of wt Wig-1, but not of the H88A mutant, partially rescued both TP53 mRNA and p53 protein levels (Fig. 6g-h), demonstrating that repression of ZMAT3 gene by KpSN participates in the destabilization of TP53 mRNA. Finally, ZMAT3 repression also affected the stability of several others mRNA, previously described to be either stabilized (such as EIF4B and RRM1) [39] or destabilized by Wig-1 (such as FAS) [40], suggesting a broader effect than p53 inhibition ( Supplementary  Fig. S5c, d). We then investigated the mechanism of ZMAT3 repression. We found that knock-down of p65 rescued Wig-1 downregulation, confirming the NF-κB involvement in ZMAT3 repression by KpSN (Fig. 7a). LPS-induced repression of genes through NF-κB has been recently shown to be regulated by histone deacetylases (HDACs) [41]. Using two different pan-HDACs inhibitors, vorinostat and belinostat, we could rescue the downregulation of ZMAT3 mRNA upon KpSN as well as Wig-1 and p53 proteins (Fig. 7b,  c), confirming HDACs-dependent repression of ZMAT3. Altogether, our results demonstrate that activation of the TLR4-NF-κB pathway by bacterial LPS leads to destabilization of TP53 mRNA through its 3′ UTR, mediated by inhibition of RNA-stabilizing factors such as Wig-1.
In conclusion, our data reveal a novel mechanism of p53 inhibition during the cell response to K. pneumoniae and other Enterobacteria, through NF-κB-dependent destabilization of TP53 mRNA, p53 protein downregulation and impairment of its transcriptional activity and tumor suppressive function.

DISCUSSION
In this study, we discovered that K. pneumoniae strongly inhibit p53 and prevent its anti-tumorigenic functions. Interestingly, we found that K. pneumoniae LPS was sufficient to inhibit p53. Moreover, our data point out to a more general mechanism of TP53 mRNA repression by TLR4 agonist LPS from Enterobacteria, which elicits the canonical LPS-TLR4-NF-κB pathway. Our results shed a new perspective on published data about p53 response to bacteria, where inhibition of p53 downstream of microbeassociated molecular patterns (MAMPs) and pattern recognition receptors (PRRs) signaling has not been investigated [15,42].
The mechanisms controlling p53 protein stability have been extensively characterized and accepted as the major level of p53 regulation [43]. Only few studies addressed the regulation of TP53 mRNA during the host response to stress [44][45][46]. Our findings reveal that TP53 mRNA serves as an important regulatory platform to turn off p53 activity upon infection and NF-κB activation. The NF-κB and p53 pathways have long been known to be tightly intertwined and to antagonize each other at multiple levels [47,48]. p53 activation impairs NF-κB function, including in response to LPS [49,50], and conversely, loss of p53 increases NF-κB activity [51]. Therefore, p53 inhibition in innate immune cells and fibroblasts exposed to MAMPs could be crucial to ensure a proper NF-κB-dependent inflammation and innate immune response. Supporting this hypothesis, p53 KO mice infected by K. pneumoniae display an enhanced bacterial clearance by neutrophils and macrophages [52]. Additionally Western blot of BJ hTert cells exposed to KpSN upon HDAC inhibition. Right panels: densitometric quantification of p53 and Wig-1 Western blot bands normalized to β-actin level. *p < 0.05; ***p < 0.01; NS non-significant.
key level of regulation to coordinate the switch from immune cell proliferation to activation during innate immune response [38].
Our finding that RNA binding factors such as Wig-1 are among the top differentially expressed genes is consistent with this idea. Importantly, Wig-1 forms a positive feedback loop with p53 and it is an important p53 target gene [23,24]. Thus, our data suggest that TP53 mRNA destabilization and Wig-1 downregulation cooperate during acute inflammation to robustly activate NF-κB to fight bacterial infection, simultaneously weakening tumor barrier (Fig. 8). Importantly, CRC patients have elevated circulating LPS and systemic inflammation [53]. Moreover, CRC-associated dysbiosis is usually enriched in Gram-negative bacteria [54], which suggests that LPS secretion is an important factor linking bacterial dysbiosis and tumorigenesis. We found that commensal bacteria expressing a TLR4 antagonist LPS, such as B. fragilis, do not downregulate p53, which could contribute to their protective role in colitisassociated colorectal cancer [55]. In addition, colonization of the gut by Enterococcus faecalis, a Gram-positive bacterium, induces inflammatory response and colitis similarly to Enterobacteria colonization but does not induce tumorigenesis [56]. Altogether, it suggests that activation of the TLR4-mediated signaling by LPS plays a critical role in promoting cancer not only by inducing inflammation but also by inhibiting p53.
Chronic inflammation linked to bacterial dysbiosis could cause persistent impairment of p53 tumor suppressor function in innate immune cells, which might create a cellular context favorable for oncogenesis. Supporting this idea, mouse models with genetic inactivation of p53 specifically in myeloid cells are more prone to tumorigenesis [57]. Additionally, inactivation of p53 in the tumor microenvironment have profound consequences on the tumor cells themselves in a non-cell autonomous fashion [58,59]. As tumor progression is associated with disruption of the epithelial layer and invasion of the tumor by bacteria, MAMPs signaling and downstream inhibition of p53 in cancer-associated fibroblasts and tumorassociated macrophages could fuel tumor development. Accordingly, several studies have shown that LPS activation of TLR4 accelerates tumor growth [29,60]. Interestingly, our analysis of cancer patient data demonstrated an inverse correlation between TLR4 expression and p53 mutation rate in CRC. Our results support the idea that TLR4 activation disables p53 function in the absence of mutation. Moreover, it suggests that the cancer-associated bacterial microbiota exerts a selection pressure on cancer cells and drives tumor evolution. In line with this hypothesis, recent metagenomics analysis of lung cancer microbiota have shown significant differences in bacterial populations associated with p53 wild-type versus p53-mutant tumors [61]. Further analyses of metagenomics data from cancer patients in connection with tumor genome profiling are required to understand the connection between the microbiota and host genomic alterations.

Bacterial strains
All bacterial strains used in this study are listed in Supplementary Table S1. K. pneumoniae SHG4 and SGH10 strains were a gift from Y.-H. Gan [62,63], IA565 strain was a gift from G. Huffnagle [64]. Enterobacteria were grown either in DMEM 10% FBS, Luria Broth or low salt Luria Broth at 37°C. S. pneumoniae was grown on agar blood plates at 37°C with 5% CO 2. B. fragilis was grown on agar blood plates at 37°C under anaerobic conditions. When stated, 100 μg/mL gentamycin, 50 μg/mL hygromycin, 50 μg/mL apramycin was added to the culture.
K. pneumoniae lpxM deletion mutant (ΔlpxM) was constructed using a previously established allelic exchange method with a few modifications [65]. Briefly, the apramycin resistance cassette with extensions homologous to regions adjacent to lpxM was generated by PCR using the pIJ773 plasmid [65] as a template (primers: forward 5′-CTACACTATCCCATTATCTTGATTAA GCAGTCGATCTGCGGATTGGGCATGATTCCGGGATCGTCGACC-3′; reverse 5′-CGAGTAAGCACGGTAGAGATAAAAAAGCCTCCTGACGGAGGCTTTTTTTATGT AGGCTGGAGCTGCTTC-3′). The PCR product was then electroporated into K. pneumoniae cells carrying the pSIM18 plasmid [66]. Replacement of the gene by the apramycin resistance cassette flanked by two Flp recombinase target sequences was confirmed by PCR. The resulting strain was then transformed with the pFLP-hyg plasmid (Addgene #87831; [65]) and incubated for 24 h at 30°C, allowing excision of the cassette by the Flp recombinase. Plasmid pFLP-hyg was then eliminated at 42°C, and the cassette excision was verified by sequencing. The pBAD33-gent plasmid producing lpxM was constructed by standard restriction/ligation cloning by inserting an XbaI-lpxM-HindIII fragment (obtained by PCR using the primers: forward 5′-GATCCTCTAGAGGATTGGGCATGGAAACGAAAAAAAA T-3′; reverse 5′-GATCCAAGCTTTTATTTCTTTTTCGTGAACAGCTCTTTGCG-3ʹ) into the XbaI-HindIII digested pBAD33-gent plasmid (Addgene #65098; [67]

Feedback loop
Oncogenic or genotoxic stress Fig. 8 Schematic representation of p53 inhibition by LPS. Upon oncogenic or genotoxic stress (left panel), p53 is stabilized, and induces ZMAT3/Wig-1 expression. In turn, Wig-1 stabilizes TP53 mRNA, leading to a robust p53 accumulation, p53 activity and tumor suppression. Upon LPS signaling (right panel), activation of the canonical TLR4-NF-κB pathway inhibits ZMAT3 transcription via HDACs, preventing TP53 mRNA stabilization and ultimately impairing p53 tumor suppressive function. Additionally, p53 is unable to inhibit NF-κB, allowing a robust inflammatory response.

Human cell lines and primary cells
Human immortalized fibroblasts BJ hTert, BJ hTert shp53, BJ hTert HRasV12 ER-Tam and BJ hTert HRasV12 ER-Tam shp53 cells were obtained from R. Agami [68]. Human melanoma cell line A375 was from ATCC. Primary human normal fibroblasts (HNF) from dermal biopsies were obtained from A. Falk (Karolinska Institute). Cell authentication was performed by STR profiling analysis at Eurofin Genomics. Mycoplasma contamination was tested monthly using MycoAlert Mycoplasma Detection Kit (Lonza) according to the manufacturer's instructions. All experiments were performed within 10 passages from frozen stocks.
For experiments on PBMC-derived macrophages, monocytes were isolated from anonymous buffy coats of healthy blood donors (Karolinska University Hospital) using Ficoll gradient and centrifugation. Briefly, blood was diluted in PBS and layered on to Ficoll-Paque (GE Healthcare) and centrifuged at 1,200 rpm for 20 min. The interface layer containing the monocytes was collected and monocytes were further washed twice in PBS. Monocytes were then incubated 2 h for adhesion and unattached cells were washed with PBS. For differentiation into M1 and M2 macrophages, monocytes were cultured in RPMI 1640, 2 mM I-glutamine, 10% FBS, streptomycin/penicillin (Sigma-Aldrich) supplemented with increasing concentration of Granulocyte-Macrophage Colony-Stimulating (Sigma; up to 400 ng/ml) or Macrophage Colony-Stimulating Factor human (Sigma-Aldrich; up to 40 ng/ml) for 7 days.

Transient infections of BJ hTert cells and bacteria cell free supernatant preparation
For transient bacterial infections, overnight DMEM cultures of K. pneumoniae were diluted in fresh medium (DMEM, 10% FBS) and 2.10 5 BJ hTert cells seeded in 6-wells plate were infected with a multiplicity of infection (m.o.i.) of 0.1 to 10 as indicated. After 2 h, cells were washed 3-6 times and incubated in cell culture medium with 200 µg/mL gentamicin for an extra 6 h.
To prepare bacteria cell free supernatant, Enterobacteria strains were cultivated in DMEM 10% FBS at 37°C until OD reaches 2. S. pneumoniae and B. fragilis were grown on Blood agar plate overnight at 37°C and respectively incubated with 5% Co2 or under anaerobic conditions. Supernatant was obtained by centrifugation at 4000 rpm for 15 min. The supernatant fraction was then passed through a 0.22 µm pore size filter (Millipore). Supernatant fractionation was done using concentrators with 3 kDa and 100 kDa cut-off (Pierce). Heat treatment was performed at 75°C for 15 min.

Cell treatments
Unless stated otherwise, KpSN and other bacterial supernatants were used to supplement culture medium to a final concentration of 10% v/v for 8 h (for RT-qPCR) or 10 h (for Western blots). Purified K. pneumoniae LPS (Sigma-Aldrich) was used at 100 ng/mL for the same duration. Doxorubicin (Sigma-Aldrich) was used at 0.2 μM. For treatments with TAK242, Nutlin, MG132, AKT1/2 kinase inhibitor, calpain inhibitor III, vorinostat (Sigma-Aldrich) and belinostat (VWR), cells were pretreated 2 h with the inhibitor before co-treatment with KpSN or LPS. For treatment with polymyxin B, medium containing KpSN or LPS was incubated 24 h at 37°C with 30 μg/ mL polymixin B sulfate (Sigma-Aldrich) before addition to the cells. For cycloheximide chase assay, cells were pre-treated 4 h with KpSN or LPS, and then treated with 10 μM cycloheximide (Sigma-Aldrich) for the indicated time.

RT-qPCR
Total RNA extraction and cDNA synthesis were performed using Aurum total RNA and iScript cDNA synthesis kits (Bio-Rad) according to supplier instructions. Relative mRNA levels were measured by RT-qPCR using SsoAdvanced Universal SYBRGreen SuperMix (Bio-Rad). GAPDH and RPL13A were used as housekeeping genes. Error bars represent standard deviation from mean of at least three independent experiments. The sequences of RT-qPCR primers used are detailed in Supplementary Table S2.
For detection of K. pneumoniae lpxM, a polyclonal anti-lpxM antiserum was produced by immunizing rabbits with a synthetic peptide, GEIEPYKR-KELFTKKK (Singapore Advanced Biologics).

LPS quantification
Cell-free supernatants preparations were serially diluted in pyrogen-free water, and reactogenic LPS was detected using the Pierce LAL chromogenic endotoxin quantification kit (LifeTechnologies), according to the manufacturer's instructions but halving the suggested volumes. Reactogenic LPS units for each sample were normalized to the OD 600 of the culture. All data are presented as the mean and standard error of the mean (SEM) for at least three sets of biological replicates.

Measurement of oncogene-induced senescence
BJ hTert HRasV12 ER-Tam and BJ hTert HRasV12 ER-Tam shp53 cells were seeded in 12-wells plates on glass coverslips, then treated the following day with 4-OHT and KpSN. The culture medium was renewed every 3 days, and KpSN was added each day to the culture medium. Senescence was measured 8 days after the start of the experiments. For β-galactosidase assay, cells were fixed and stained 24 h using the Senescence Cells Histochemical Staining kit (Sigma-Aldrich) according to manufacturer's instructions. For EdU incorporation Assay, 10 μM EdU was added to the culture medium 24 h before fixation, then EdU was labeled using the Click-iT EdU Alexa Fluor 488 Imaging Kit (Thermo Fisher Scientific). For both assay, quantification was performed by counting positive cells in n ⩾ 200 cells in three independent experiments.

RIP-qPCR
RNA immunoprecipitation (RIP) was performed on 2.10 7 BJ hTert pLV Wig-1-Flag cells using the Magna RIP Kit (Merck-Millipore) according to the supplier instructions. Wig-1-Flag immunoprecipitation was performed using a monoclonal anti-Flag antibody (F3165, Sigma-Aldrich) and mouse IgG as control. Isolated RNA was then analyzed by RT-qPCR.

RNA sequencing
BJ hTert and BJ hTert shp53 were treated for 8 h with KpSN and 0.2 μM doxorubicin in three independent biological replicates. Total RNA was extracted using the mirVana miRNA isolation kit (Ambion). RNA-seq was performed by Vertis Biotechnologie AG. Random-primed cDNA libraries were prepared according to Illumina protocol. The cDNA pool was paired-end sequenced on an Illumina NextSeq 500 system using 2 × 150 bp read length. Quality control was performed using FastQC. nf-core/rnaseq pipeline (v1.4) was used to process the reads [71]. HISAT2 (v2.1.0) was used to align the raw RNA-seq fastq reads to the human reference genome (GRCh38.97). Read quantification was computed using featureCounts (v1.6.4). Differential analysis was performed with the R package DESeq2. False positive discovery (FDR) < 0.05 was set as cut-off for differentially expressed genes (DEGs). DEGs were plotted using R package ggplot2. Gene ontology (GO: biological process) and pathway enrichment analysis was performed using g:Profiler [72] and visualized using Cytoscape (v3.6.1) [73] as described by Reimand et al. [74]. The complete lists of differentially expressed genes and enriched GO terms upon KpSN treatment are provided in Supplementary Table S3. To define a high confidence gene set of p53 targets in BJ hTert cells, we intersected genes upregulated upon doxorubicin treatment, repressed upon p53 shRNA and p53 core targets genes defined by Fisher et al. [22]. Complete list of the 161 high confidence p53 target genes is provided in Supplementary Table S4.

Statistical analyses
Unless otherwise stated, statistical significance was calculated using twotailed Student t-test from at least three independent experiments.

DATA AVAILABILITY
The datasets generated in the study are available in GEO database (https://www.ncbi. nlm.nih.gov/geo/) under accession number GSE174531.