Functional antagonism between CagA and DLC1 in gastric cancer

Helicobacter (H.) pylori-induced gastritis is a risk factor for gastric cancer (GC). Deleted-in-liver-cancer-1 (DLC1/ARHGAP7) inhibits RHOA, a downstream mediator of virulence factor cytotoxin-A (CagA) signalling and driver of consensus-molecular-subtype-2 diffuse GC. DLC1 located to enterochromaffin-like and MIST1+ stem/chief cells in the stomach. DLC1+ cells were reduced in H. pylori gastritis and GC, and in mice infected with H. pylori. DLC1 positivity inversely correlated with tumour progression in patients. GC cells retained an N-terminal truncation variant DLC1v4 in contrast to full-length DLC1v1 in non-neoplastic tissues. H. pylori and CagA downregulated DLC1v1/4 promoter activities. DLC1v1/4 inhibited cell migration and counteracted CagA-driven stress phenotypes enforcing focal adhesion. CagA and DLC1 interacted via their N- and C-terminal domains, proposing that DLC1 protects against H. pylori by neutralising CagA. H. pylori-induced DLC1 loss is an early molecular event, which makes it a potential marker or target for subtype-aware cancer prevention or therapy.


INTRODUCTION
Chronic gastritis following infection with the gram-negative bacterium Helicobacter (H.) pylori is a risk factor for gastric cancer (GC) [1], especially by strains positive for the virulence factor cytotoxin-A (CagA) [2]. Among the four consensusmolecularsubtypes (CMS) [3], mutations in RHOA and RHO GTPase-activating proteins (GAPs) are major drivers of diffuse GC (CMS2). These mutations foster constitutive GTPase activities [4], conferring adverse prognosis and therapeutic resistance [5]. Thus, identifying proteins/drugs that inhibit RHO signalling remains an unmet medical need.

Correlation between DLC1 expression and clinical factors in patients with GC
To detect in situ expression of DLC1 protein, we analysed tissue microarrays (TMAs) from patients with GC. Formalin-fixed paraffin-embedded (FFPE) tissue sections from normal/nontumour stomach (NT: n = 30) and tumour (TU: n = 116) cases were stained by immunohistochemistry (IHC) with DLC1 antibody (Ab) against the N-terminus, including the SAM domain (S1a). Quantitative analyses of graded scores displayed prominent DLC1 staining in intestinal-type, differentiated tumours both in the cytoplasm and at membranes, compared with diffusetype, undifferentiated tumours with marked stroma positivity. Compared with epithelial cells of the non-neoplastic gastric mucosa, DLC1 protein was downregulated in tumour cells (*p < 0.05 vs. NT, Kruskal-Wallis test with Dunn post-tests). No differences were detected by Lauren histology (NT: n = 30; TU: n = 34 diffuse; n = 56 intestinal). DLC1 downregulation could be associated with the grade of tumour dedifferentiation (S1b). DLC1 was also reduced by UICC (S1c) and pTNM staging, as determined by tumour size (T) (S1d) and nodal status (N) (S1e). Conversely, DLC1 was upregulated in cases exhibiting metastases (S1f) (TU: n = 36 M0; n = 11 M1; *p < 0.05 vs. NT, Mann-Whitney test). No significant alterations were recorded in the stroma. Supportive data from independent patient cohorts are provided in Supplementary Results (S2, Tables S3-5).

Expression of DLC1 variants in human GC cell lines
To explore the expression of DLC1 variants in GC [7] (Fig. 1A), the DLC1 gene was examined for COSMIC/GISTIC entries on somatic mutations and copy number variations (Table S6). Common putative driver splice mutations were detected at identical or adjacent codons (aa 438/439/501/522) within the human FL DLC1v1 gene and located N-and C-terminal of the SAM domain where the DLC1v4 CDS (aa 512) begins (Fig. 1B). Cancer hotspot mutations and splice variants were similar in cancer cell lines (CCLE) and pan-cancer patient studies (from EC/GC).
To measure DLC1 mRNA in human gastrointestinal cancer and non-cancer (HEK293T) cell lines, primers were designed against N-and the C-terminal parts of the FL DLC1 cDNA. Consistent with previous data [11], RT-qPCR analyses (Fig. 1C) revealed that FL DLC1v1 was present in HEK293T, whereas truncated DLC1v4 was detectable in all cancer cell lines examined (*p < 0.05 N vs. C, 2way-ANOVA with Bonferroni post-tests, n = 3 per cell line).
RT-qPCR analyses of total RNA extracted from frozen samples of matched normal NT and TU tissues from patients with GC ( Fig. 1D) confirmed that DLC1v1 mRNA was reduced by~40% compared with controls (*p < 0.05 TU vs. NT, 2way-ANOVA with Bonferroni post-tests, n = 5 cases).
Similar results were observed at the protein level (Fig. 1E). We amplified the DLC1v4 (start codon: MKLEI, aa 513-1528) CDS lacking the SAM region when compared with FL DLC1v1 (aa 1-1528) [7]. The cDNAs were inserted into FLAG-tagged expression vectors and transfected into HEK293T or AGS cells for 48 h. Western blotting confirmed overexpression of FL DLC1v1 (170 kDa) and N-terminally truncated ΔSAM DLC1v4 (110 kDa). A smaller sized, endogenous DLC1 protein (130-170 kDa) was detected in GC cell lines (AGS, MKN45, KATO3, N87) using an Ab specific for the N-terminal part, including the SAM domain, but not in HEK293T or SNU1, indicative of cell line-specific isoforms.
H. pylori and CagA downregulate DLC1 expression in vivo As DLC1 is frequently downregulated in GC, we speculated whether H. pylori infection contributes to human or mouse DLC1/ Dlc1 gene silencing. FFPE tissue sections from endoscopic gastric biopsies of naïve and infected individuals were stained by IHC using DLC1 Ab ( Fig. 2A). After dichotomous grouping as negative (scores 0/1) vs. positive (scores 2/3) cases, DLC1 staining was correlated to the infection status (n = 9-10 per group). Quantitative analyses of the corpus region exhibited the disappearance and displacement of DLC1 + cells by leucocyte infiltration into inflammatory regions (Fig. 2B). DLC1+ cells accumulated in the periphery of inflammatory follicles. This distribution pattern proposed that H. pylori gastritis leads to downregulation of DLC1 positivity in the stomach.
To explore the impact of H. pylori virulence factors on DLC1 expression, C57BL/6J mice were infected for 1 and 6 months with the mouse-adapted CagA+ injection-proficient H. pylori PMSS1 WT strain or isogenic ΔCagE mutant, and DLC1 was visualised by IHC. The CagE mutant cannot translocate CagA into epithelial cells, as CagE is an essential protein for the functionality of the type IV secretion system (T4SS) [2]. Notably, PMSS1 WT reduced the numbers of DLC1+ cells in the stomach of infected mice when compared with uninfected controls (*p < 0.05 vs. WT, Kruskal-Wallis test, n = 2-9 mice per group or time point); however, no changes in the numbers of DLC1+ cells were observed in the absence of CagA translocation (Fig. 2C).
To determine if DLC1 localises to gastric cell types other than ECL cells, the putative stem/chief cell marker MIST1 was examined. Supportive immunofluorescent ( H. pylori and CagA downregulate DLC1 expression in vitro An internal promoter is located upstream of exon 9 to allow transcription of N-terminally truncated DLC1v2/4 mRNAs [7]. Binding motifs for stress-sensitive transcription factors were predicted to bind this promoter (e.g. p53, STAT, HSF), factors known to be also activated by H. pylori infection (Fig. 3C).
As shown in Fig. 1, AGS cells expressed DLC1v4 but no FL DLC1v1 mRNA, making this cell line suitable to study the effect of CagA on this promoter. AGS cells were infected with the CagA+ injection-proficient H. pylori strain G27 (MOI = 100) for 3 days (Fig. 3D). RT-qPCRs revealed a reduction of DLC1v4 mRNA bỹ 90% when compared with controls (*p < 0.05 vs. mock, 2way-ANOVA with Bonferroni post-tests, n = 3).

DLC1 counteracts cell phenotypes evoked by CagA
Based on the observed CagA-mediated DLC1 downregulation, we hypothesised that, vice versa, DLC1 inhibits CagA signalling. To explore the mutual antagonism of the two proteins, we examined whether DLC1 prevents CagA-mediated cytoskeleton rearrangements, detectable as spike-like cell elongations, termed the "humming bird" phenotype [2].
AGS and HEK293T were transfected with EV, CagA, or DLC1 expression plasmids or a combination thereof for 36 h, followed by fixation and staining using DLC1 or FLAG Ab for immunofluorescent microscopy (Fig. 4A). Numbers of adherent cells normalised to total cell counts were increased (HEK293T: *p < 0.05 vs. EV, 2way-ANOVA with Bonferroni post-tests, n = 3 per cell line). DLC1 was localised to the periphery of spread-out cells and in focal adhesions (DLCv1) or neurite-like extensions (DLC1v4), whereas CagA was targeted to the plasma membrane and promoted needle-like cell elongations ("humming bird"). In cells receiving both plasmids, DLC1 reduced CagA-dependent formation of stress spikes (AGS: *p = 0.0374, Fisher-Exact test; n = 3) (Fig. 4B). Similar results were obtained for AGS (S4) and HEK293T (S5) cells. Supportive data on DLC1v1/4 effects on morphologies are provided in Supplementary Results (S6-8).
Next, we examined whether DLC1v1-mediated activation of stress/growth-related kinases is translated to the nucleus towards c-FOS transcription, a cell proliferation surrogate (Fig. 5A). HEK293T, AGS and N87 cells were co-transfected for 72 h with EV, DLC1v1, or CagA expression plasmids or combinations thereof, along with pGL3 reporter containing the SRE from the human c-FOS gene promoter. Compared with CagA-transfected controls, luciferase activity was reduced by DLC1v1 to 10% (*p < 0.05 vs. EV, 2way-ANOVA with Bonferroni post-tests, n = 3 per cell line).
H. pylori triggers oxidative stress by releasing reactive oxygen species from infected host cells [12]. To determine whether DLC1v1 interferes with hypoxia-driven stress responses (Fig. 5B), cells were transfected as above with a pGL3 reporter containing the HRE from the promoter of the human HMGB2 gene [13]. DLC1v1 decreased luciferase activity to 20% when compared with CagA-transfected controls (*p < 0.05 vs. EV, 2way-ANOVA with Bonferroni post-tests, n = 3 per cell line). Overall, FL DLC1v1 protected host cells from CagA-mediated damaging and oncogenic effects. DLC1 inhibits RHOA via its RHO-GAP domain [6], whereas CagA activates RHOA [14]. We thus examined whether DLC1 inhibits CagA-driven RHOA activity. Cells were transfected as before with EV, DLC1v1/4, or CagA plasmids, along with a modified pSRE.L reporter [15] containing an RHO-driven SRE deficient for the TCFbinding site in the human c-FOS promoter (Fig. 5C). Both DLC1 isoforms reduced RHO-driven luciferase activity compared with CagA-transfected controls. Moreover, DLC1 counteracted RHOdriven reporter gene expression in presence of CagA overexpression (*p < 0.05 vs. EV, 2way-ANOVA with Bonferroni posttests, n = 3 per cell line).
GST-pulldown assays confirmed these findings using recombinant RHO-binding domain of rhotekin as bait for binding active RHOA (Fig. 5D). TsA201 cells were co-transfected with EV, DLC1v1/ 4, or CagA plasmid or combinations for 72 h. Total cell lysates were subjected to pulldown assays and Western blotting to detect active and total (input) RHOA. CagA increased G-protein-coupled RHOA activation, an event inhibited by both DLC1 isoforms when compared with input controls. Consistently, RHOA specificity was verified by C3T and G13qL for RHOA pulldown. C3T inhibited CagA-and G13qL-mediated RHOA activation (S9c).
CoIP experiments were performed to verify complex formation between the two proteins. HEK293T cells were transfected with CagA-FL combined with DLC1v1 or v4 plasmids. IP was performed on cell lysates using Abs against CagA or DLC1 (Fig. 6B, C). Western blotting corroborated the interaction between CagA-FL and both DLC1 isoforms. We observed that DLC1v4 was more efficiently precipitated than DLC1v1 (*p < 0.05 vs. bead control, 2way-ANOVA with Bonferroni post-tests, n = 3 per plasmid). To determine which CagA domains are necessary for the interaction, cells were transfected with CagA-NT or CagA-C in combination with DLC1v1 or v4 plasmids, respectively. Both N-and C-terminal parts of FL CagA interacted with the two DLC1 isoforms. Again, DLC1v4 was more efficiently precipitated than DLC1v1 (*p < 0.05 vs. bead control, 2way-ANOVA with Bonferroni post-tests, n = 3 per plasmid). PLA microscopy confirmed the proximity of the two proteins. Oligonucleotide-labelled secondary Abs generated a red fluorescent signal on colocalisation of GFP-CagA and DLC1 (Fig. 6D). All examined CagA constructs interacted with both DLC1 isoforms (S10).
Collectively, these findings suggested that DLC1 binds and neutralises CagA in vitro [11]. Supportive data on phenotypes of Fig. 1 Expression of DLC1 mRNA/protein variants in human GC cells. A Scheme of human DLC1 mRNA/protein variants. FL DLC1v1 and the N-terminally truncated DLC1v4 isoform exhibit identical C-terminal multi-domain organisation: sterile-alpha motif (SAM), serine-rich region, nuclear localisation sequence (NLS), caveolin-1 (CAV1)-binding motif (Cavbm), RHO GTPase-activating protein (GAP), and steroidogenic acute regulatory protein-related lipid-transfer (START) domains. DLC1v4 (start codon: MKLEI, aa 513-1528) used in the present study lacks the SAM region compared to DLC1v1 (aa 1-1528). B Comparison of putative driver splice mutations in the human FL DLC1v1 gene located N-and C-terminal of the SAM domain where the DLC1v4 transcript starts. Point mutation data were retrieved from cBioPortal (Table S6)

DISCUSSION
Herein, we demonstrated that the RHO GTPase inhibitor DLC1 prevented gastric inflammation in vivo and counteracted the oncogenic actions of CagA, a major virulence factor of H. pylori and a risk factor for GC [14]. Mechanistically, we provided evidence indicating mutual functional antagonism of the two proteins, wherein DLC1 inhibits CagA-dependent RHO signalling and, vice versa, CagA prevents transcription from the DLC1 promoter, thereby lowering DLC1 protein in living cells.
This cross-talk (Fig. 6E) was driven by protein-protein interaction mediated via the N-and C-terminal CagA domains and the internal/C-terminal regions of FL DLC1v1 and ΔSAM DLC1v4. The N-terminal membrane domain of CagA inhibits the signalling activity of the C-terminal EPIYA motifs [18], and the N-terminal SAM domain in DLC1 inhibits the enzymatic activity of the RHO-GAP domain [19]. Hence, steric competition between these mutual intra-and inter-molecular regulatory protein interactions determines the outcome of signalling activities for both proteins. Notably, N-terminally truncated DLC1v4 protein, retaining the C-terminal RHO-GAP and START domains such as the caveolinbinding motif (Cavbm), was recruited to caveolin-rich areas of the plasma membrane by H. pylori [11]. Moreover, this truncation variant was present after transformation, whereas FL DLC1v1 was lost in GC cells. Thus, DLCv4, lacking the auto-inhibitory N-terminal domain and/or harbouring other unidentified mutations [20], may gain aberrant oncogenic function in gastric cells, a feature which could explain the observed enrichment of DLC1 mRNA/protein in diffuse-type (CMS2) and advanced stages of GC with distant metastases [6]. Consistently, therapeutic response in patients with GC has been correlated to DLC1 expression [9,10].
Genomic alterations in DLC1 contribute to drug sensitivity and confer growth advantage upon activation of RHO-ROCK signalling, rendering tumour cells more susceptible to ROCK-inhibitors (e.g., Y-27632). Low DLC1 expression also predicts poor therapeutic efficiency of fluoropyrimidine/oxaliplatin in adjuvant chemotherapy and was associated with advanced, metastatic disease and shortened time to recurrence and overall survival. Likewise, ROCKinhibitor fasudil [21] attenuated murine GC growth in vivo. Joshi et al. have designed cell-permeable peptides derived from the SAM domain of DLC1 [22], which destroy the intra-molecular autoinhibition between the SAM and the RHO-GAP domains. Thereby, the peptides trigger a conformational reactivation of DLC1, from a "closed" towards an "open" state ( Fig. 6E), allowing recruitment and interaction with potent tumour suppressors like PTEN, culminating in decreased RHOA activation, tumour cell growth and migration in vitro, proposing DLC1 as a druggable target.
To date, DLC1 has not been functionally connected with H. pylori in patients or animal models, although gene hypermethylation was detected upon infection [23]. Similar to humans [7], at least four murine splice/protein variants have been identified [8]. Gene trapped Dlc1 gt/+ mice harbour an insertion between exon 1 and 2 of the 6.1 kb transcript and exhibit developmental defects and disease-prone phenotypes [8,24,25]. However, Dlc1 gt/+ mice do not per se progress to cancer, suggesting that cooperative damage cues or oncogenic mutations are required for transformation. Consistently, we showed that Dlc1-deficient animals exhibit gastritis and disturbed expression of neuro-entero-endocrine hormones, characterised by moderate leucocyte infiltration and enhanced proliferation of epithelial cells, indicating an early step in the transition from inflammation to malignancy (S15). The highest in situ expression of DLC1 protein both in mouse and human gastric tissue was found in ECL cells [26], known to produce a majority of the gut peptide and histidine/tryptophanderived hormones [27]. Therefore, DLC1 may play an additional role in stomach homeostasis via auto/paracrine regulation of secretory factors, in addition to its well-established function in epithelial cells as a cytoskeleton regulator. H. pylori infection correlates with deterioration of the integrity and secretory activity of ECL cells (e.g. somatostatin producing D cells [28]) and hypergastrinemia (by gastrin producing G cells [26]), resulting in altered histamine release and subsequent gastric acid secretion from adjacent parietal cells in the oxyntic mucosa and preneoplastic alterations in humans and animal models. Consistent with this sequence of events, Dlc1 gt/+ mice exhibited reduced levels of several ECL hormones, primarily somatostatin [29], which counteracts the pro-angiogenic and trophic effects of gastrin [27]. However, further in-depth studies need to clarify the causality, considering the unidentified stem cell origin of gastric adenocarcinoma and neuroendocrine tumours (carcinoids).
According to Lauren's classification, H. pylori can be associated with intestinal GC, via the "Correa" sequence of histomorphological changes, and with diffuse GC [2], in addition to sporadic mutational events causative for this subtype. The current CMS classification [3] highlights the function of RHO signalling in diffuse (CMS2) GC. This is underscored by the fact that several mutations in RHOA and RHO GAP (e.g. CLDN18-ARHGAP6 or CTNND1-ARHGAP26 fusions) genes, to which DLC1 (also termed ARHGAP7) also belongs, were identified as oncogenic drivers for diffuse GC [3,4]. Therefore, we propose that loss of the RHOinhibitor DLC1 correlates with H. pylori infection in certain GC subtypes. Despite the declining incidence of intestinal GC, that of diffuse GC continues growing and is characterised by poor prognosis and rapid metastasis. Thus, there is an urgent medical need to specifically address this GC subtype. Consistently, our data propose that novel subtype-specific treatment strategies could be developed for RHOA inhibition.

MATERIALS AND METHODS Subjects
All patients provided written informed consent. Studies were performed in accordance with the principles of the Declaration of Helsinki and approved by the Medical Ethics Committees of the Universities of Heidelberg, TU München and Kiel. Patients' tissue specimens were collected after surgical resection, stored and classified by Lauren's histological classification. Tissue microarrays (TMAs) were custom-made [30] or purchased from US Biomax (Rockville, MD).

Animals
Tissue biobanks from C57BL/6J mice infected with H. pylori strains SS1 [11] or PMSS1 [31] were generated previously. Dlc1 gt/+ mice [8] were bred in a Fig. 3 Helicobacter pylori and CagA downregulate DLC1 expression in vitro. A Partial colocalisation of DLC1 with MIST1+ stem/chief cells. Detection of DLC1 protein in murine gastric chief/stem cells. FFPE sections from stomachs of mock (control) and WT PMSS1-infected (6 months) mice (from Fig. 2C) were co-stained using DLC1 and MIST1 Abs for immunofluorescence microscopy. Quantitative analyses (left) and representative images (right) of the corpus region. Data are numbers of double-positive cells per mm 2 presented as mean (n = 2-9 uninfected control mice per time point, n.c. = no double pos. cases detectable in infected mice). Colour code: Green = DLC1; Red = MIST1; Blue = nuclei (DAPI); original magnifications ×200; Scale bar = 20 and 100 µm. White arrow/frame: Yellow = zoomed-in overlay (DLC1+/ MIST1+). Insert (top left): Representative images of PCR-amplification products and murine DLC1 protein isoforms (GC: 70 kDa vs. NT: 170 kDa) from total tissue lysates detected by Western blot using Abs against the N-terminal part of DLC1 including the SAM domain (source file: "Original Western Blots" page 4). Legend: STO = Stomach (Corpus); L = Liver (pos. control); GC = gastric cancer. B Colocalisation of DLC1 with ECL cells, but not with parietal cells. FFPE sections from stomachs of uninfected mice were co-stained using DLC1, chromogranin A (CHGA), and H + K + ATPase (HK) Abs for immunofluorescence microscopy. Quantitative analyses from images of the corpus region presented in S3. Data are numbers of single or double-positive cells per field presented as mean (*p < 0.05, Kruskal-Wallis test, n = 2-9 uninfected control mice, n ≥ 5 fields per image). C Scheme of promoters identified in the human DLC1 gene according to Low et al. [7]. Note that the promoters generate four different transcripts/protein isoforms with identical C-terminus but distinct N-termini. Legend: E = Exon. D H. pylori downregulates DLC1 mRNA. AGS cells were infected with CagA+ injection-proficient H. pylori strain G27 (MOI = 100) for 3 days. Ct-values from RT-qPCRs on total RNA were normalised to B2M and calculated as % ± S.E. (*p < 0.05 vs. mock, 2way-ANOVA with Bonferroni post-tests, n = 3). E CagA is sufficient for DLC1 promoter inhibition. Cells (HEK293T, AGS, N87) were co-transfected for 72 h with EV or CagA expression plasmid and reporter vector pGL3 containing the human proximal promoter sequences of DLC1v1 or v1. Luciferase activity was normalised to protein concentration and calculated as -fold ± S.E. (*p < 0.05 vs. EV, one-sample t test, n = 3 per cell line).

Statistics
Sample size estimates (POWER 0.8) for animal experiments were calculated with SAS (Cary, NC). Results are displayed as means ± standard error (S.E.) from at least three independent experiments (n ≥ 3) from different cell passages or individuals (mice, patients). Optical densities (ODs) of bands in gels from Western blots and PCRs were measured using automated imaging devices and quantified with ImageJ (Washington, DC). Data were normalised to house-keeping genes or proteins and calculated as -fold change or % when compared with control. Statistical analyses were performed with GraphPad Prism (La Jolla, CA). Data were tested for nonvs. parametric distribution and equal vs. unequal variance, followed by the recommended test with correction for multiple comparisons. Tests were 2-sided and unpaired if not stated otherwise in the legends to figures. P-values < 0.05 were considered significant (*).