A significant proportion of gastric-cancers are sensitive to ATRA

The heterogeneity of gastric-cancer [7, 27] is recapitulated by cell-lines whose constitutive gene-expression profiles are available in the CCLE (Cancer-Cell-Lines-Encyclopedia) database. This database contains RNA-sequencing (RNA-seq) data from 37 gastric-cancer cell-lines (Fig. 1A). According to their gene-expression profiles, these cell-lines classify into the G-INT and G-DIFF subgroups [8].

Fig. 1

ATRA-dependent anti-tumor activity in gastric-cancer cell-lines, tissue-slice cultures of gastric-cancer specimens and LMSU/NCI-N87 mouse xenografts. A Unsupervised-hierarchical-clustering of the indicated gastric-cancer cell-lines based on the gene-expression profiles determined with the RNA-seq results (TCGA-dataset): cell-lines are classified according to transcriptomic-profile (G-DIFF=red; G-INT=blue) and histochemical-characteristics (colored-boxes). B ATRA growth-inhibitory effects in RERF-GC-1B (G-DIFF, red) and IM95 (G-INT, blue) cell-lines: the values are normalized for the vehicle-treated controls, which are taken as 100%. Each point is the mean±SD of 10 cultures. When the cell-growth results observed in ATRA-treated specimens are significantly lower than those of the vehicle-treated counterparts, p-values (two-tailed Student’s t-test) are shown in red. C Gastric-cancer cell-lines are ranked according to the ATRA-score values: the bimodal and arbitrary threshold-value of 0.55 separates the cell-lines characterized by high ATRA-sensitivity (ATRA-score >0.55) from the cell-lines showing low ATRA-sensitivity (ATRA-score <0.55). Each calculated value is representative of at least 2 independent experiments. D Surgical specimens of 13 patients (P1-P13) were challenged with vehicle (DMSO) or ATRA (1.0 μM) for 48 hours: P3-P6-P9=female-patients; P1-P2-P4-P5-P7-P8-P10-P11-P12-P13= male-patients; G-INT-cases=blue; G-DIFF-cases=red; not-determined=black. Each value = Mean ± SE of 5 histological-fields/experimental-sample. The p-values (two-tailed Student's t-test) of the comparisons between the ATRA and corresponding vehicle-treated samples are shown. When the Ki67 amounts observed in ATRA-treated specimens are lower than those of the vehicle-treated counterparts, the p-values are marked in red. E Mice were xenografted subcutaneously with 1×107LMSU or NCI-N87 cells on both flanks. One week after transplantation, 10 animals/experimental group were treated with vehicle (DMSO) or ATRA (15.0 mg/kg) intra-peritoneally once/day, 5-days/week, as indicated (arrows). The calculated volumes of the tumors are plotted. As for LMSU, each point is the Mean ± SE of 15 and 19 tumors in the case of DMSO- and ATRA-treated animals, respectively. As for NCI-N87, each point is the Mean ± SE of 19 and 22 tumors. We compared the vehicle and ATRA-treated values at each time-point and the p-values (two-tailed Student’s t-test) of each comparison are shown

To define the sensitivity of gastric-cancer cells to ATRA growth-inhibitory action, we selected 27 cell-lines and exposed them to increasing concentrations of the retinoid (0.001-10µM). Subsequently, we evaluated the anti-proliferative effects of ATRA at day-3, day-6 and day-9, as exemplified in the case of the RERF-GC and IM95 cells (Fig. 1B). Cell-lines were ranked for their sensitivity to the retinoid (Fig. 1C) using a modified version of the quantitative ATRA-score index [14, 16]. This new version of the ATRA-score is based on the AUCs determined for each retinoid-treated cell-line and it ranges from a minimal “0.00” value to a maximal “1.00” value. We calculated the ATRA-scores using the growth-curves at day-6 (Fig. S1). To obtain a bimodal distribution of the 27 gastric-cancer cell-lines based on their level of sensitivity to the retinoid, we used an arbitrary ATRA-score threshold value of 0.55, which falls above the median number of cell-lines. Indeed, this threshold value allows separation of the cell-lines into 2 unbiased groups, consisting of 14 and 13 cell-lines, which are characterized by high (ATRA-score > 0.55) and low (ATRA-score < 0.55) ATRA-sensitivity (Fig. 1C). The 2 most sensitive (GCIY/HGC-27) and the 2 least sensitive (HuG1-N/OCUM-1) cell-lines have a G-DIFF and a G-INT phenotype, respectively. This reflects the trend observed in our panel of cell-lines, which shows a respective enrichment of the G-DIFF and G-INT phenotypes in the high ATRA-sensitivity (9/14) and the low ATRA-sensitivity (8/13) groups.

To validate the results obtained with cell-lines, we evaluated the anti-proliferative effects exerted by ATRA in gastric-cancer tissue-slice cultures [16, 28] from 13 patients (Table S4, patients-characteristics). Five and 6 of these tumors classify as G-INT and G-DIFF, respectively, according to the constitutive gene-expression profiles determined by the RNA-seq analyses performed on 11 cases. We challenged tissue-slices with vehicle or ATRA for 48 h and the growth-inhibitory action of the retinoid was determined by quantitative immune-histochemical measurement of the Ki-67 proliferation-marker, as exemplified for patient-1 (G-INT) and patient-2 (G-DIFF) (Fig. S2). The results demonstrate that ATRA causes a decrease of the Ki-67 levels in 9 cases (G-DIFF = 5 cases; G-INT = 3 cases; NotDefined = 1 case) (Fig. 1D).

To support the therapeutic potential of ATRA in gastric-cancer, we performed in-vivo studies with xenografts of a G-DIFF (LMSU/ATRA-score = 0.80) and a G-INT (NCI-N87/ATRA-score = 0.37) cell-line. Mice were administered vehicle (DMSO) or ATRA intra-peritoneally for 2/3 weeks and the tumor-size was determined at different time-points (Fig. 1E). As for LMSU, an ATRA-dependent reduction of the tumor-mass is already evident following 5 days of treatment and the decrease is maximal at day-15 (Fig. 1E, upper). In NCI-N87 xenografts, the maximal effect of ATRA is of lower magnitude and it is delayed, being observed only at day-19 (Fig. 1E, lower). In these conditions, ATRA is devoid of systemic toxicity, as it exerts no significant effect on the body-weight of LMSU- and NCI-N87-transplanted mice (Fig. S3).

Involvement of RARα in the anti-proliferative action of ATRA

Six retinoid-receptors (RARα/RARβ/RARγ/RXRα/RXRβ/RXRγ) are known and the active forms of these transcription-factors consist of RAR/RXR heterodimers, in which RAR acts as the ligand-binding component [29, 30]. ATRA is a pan-RAR agonist, binding/trans-activating RARα/RARβ/RARγ with equal affinity/efficiency.

To identify the RAR isoform(s) underlying the anti-proliferative action of the retinoid, initially, we evaluated the constitutive expression of RARα, RARβ, RARγ, RXRα, RXRβ and RXRγ mRNAs in our gastric-cancer cell-lines (Fig. 2A). The CCLE RNA-seq data indicate that all the cell-lines express similar levels of RARα, RARγ, RXRα and RXRβ mRNAs. In contrast, the levels of RARβ mRNA are variable, although the majority of cell-lines expresses barely detectable amounts of the transcript. No correlation between RARβ mRNA levels and ATRA-sensitivity or G-DIFF/G-INT phenotype is evident. Moreover, no cell-line expresses detectable amounts of the RXRγ mRNA. Noticeably, the RAR/RXR expression profiles of the cell-lines recapitulate the situation of primary gastric-cancers, as indicated by the RNA-seq data of the TCGA database (Fig. 2B).

Fig. 2

RAR and RXR mRNAs expression and RAR agonists anti-proliferative effects in gastric-cancer. A The panel shows the constitutive expression levels of the mRNAs coding for the indicated RAR and RXR isoforms in our panel of 27 gastric-cancer cell-lines. The cell-lines are ranked according to their decreasing sensitivity to the anti-proliferative effects of ATRA from left to right, as indicated. The expression values of the RAR and RXR mRNAs in gastric-cancer cell-lines are calculated using the RNA-seq results of the CCLE database. The values are expressed as Log 2 [CPM (Counts Per Million)]. G-DIFF and G-INT cell-lines are marked in red and blue, respectively. B The box plots indicate the constitutive expression levels of the mRNAs coding for the indicated RAR and RXR isoforms in gastric-cancer tissues characterized by a G-DIFF (red) or a G-INT (blue) phenotype. The values are calculated with the RNA-seq results of the TCGA database. The results are expressed as the Median [CPM] values  ±  SD. C The indicated G-DIFF (HGC-27; LMSU; Hs746T) and G-INT (IM95) cell-lines were exposed to vehicle (DMSO; [Retinoid] = 0 nM) or the indicated concentrations of the pan-RAR agonist, ATRA, the RARα agonist, AM580, the RARβ agonist, CD2314, and the RARγ agonist, BMS961, for 6 days. At the end of the treatment, the growth of each cell line was evaluated with the MTS assay. Each value is the Mean  ±  SD of 5 independent cultures and the data are normalized for the growth value of vehicle-treated cells (100%). We compare each compound-treated sample with the corresponding vehicle-treated counterparts at the different concentrations of the compounds. In case of statistical significance (two-tailed Student’s t-test), the p-values are shown in red. When the p-values lack statistical significance, they are marked in black. D Three independent cultures of HGC-27 cells were exposed to vehicle (DMSO) the RARα antagonist, ER50891 (0.1 µM), the RARβ /γ antagonist, CD2665 (0.1 µM), or the RARγ antagonist, MM11253 (0.1 µM), in the absence and presence of ATRA (0.01 µM) for 9 days. At the end of the treatment, the number of viable cells was counted automatically. In all samples, cell viability was always  ≥  85%. The values indicated by the columns are the Mean  ±  SD of the 3 independent cultures considered. We compare each ATRA, ATRA + ER50891, ATRA + CD2665 and ATRA + MM11253 treated sample with the corresponding vehicle, ER50891, CD2665 and MM11253 treated counterparts. In case of statistical significance (two-tailed Student’s t-test), the p-values shown above the corresponding columns are marked in red

Subsequently, we took a pharmacological approach involving the use of RARα/RARβ/RARγ agonists. With this approach, we determined the growth of the HGC-27 (ATRA-score = 1.00), IM95 (ATRA-score = 0.88), LMSU (ATRA-score = 0.80) and Hs747T (ATRA-score = 0.59) cell-lines, which belong to the high ATRA-sensitivity group. To this purpose, we exposed the 4 cell-lines to increasing concentrations of ATRA, AM580 (RARα agonist) [31], CD2314 (RARβ agonist) [32] and BMS961 (RARγ agonist) [16] for 6 days (Fig. 2C). In all the cell-lines, AM580 is the sole RAR agonist causing a concentration-dependent growth-inhibitory action of the same order of magnitude as the one observed with ATRA. Indeed, CD2314 and BMS961 exert only marginal effects on the growth of HGC-27, LMSU, IM95 and Hs747T cells and these effects are observed only with the highest concentration(s) of the two compounds. The data obtained with the RAR agonists support the idea that RARα is the primary retinoid receptor involved in the anti-proliferative action exerted by ATRA in gastric cells.

Finally, we performed experiments aimed at validating the hypothesized role of RARα in the growth inhibitory action exerted by ATRA, with the use of a specific RARα antagonist (ER50891), a RARβ/γ antagonist (CD2665) and a selective RARγ antagonist (MM11253) (Fig. 2D). To conduct these studies, we exposed HGC-27 cells to ATRA, ER50891, CD2665 and MM11253 as well as the combinations of ATRA + ER50891, ATRA + CD2665 and ATRA + MM11253 for 9 days. In these experimental conditions, ATRA causes the expected growth inhibitory effect, while ER50891, CD2665 and MM11253 are devoid of any anti-proliferative action on HGC-27 cells. Remarkably, the RARα antagonist, ER50891, is the sole compound suppressing the growth inhibitory action of ATRA. Indeed, exposure of HGC-27 cells to ATRA + CD2665 and ATRA + MM11253 results in the same level of growth inhibition observed with ATRA alone.

In conclusion, the data obtained in HGC-27 cells with a pharmacological approach based on the use of RAR agonists and antagonists are consistent with the idea that activation of RARα is necessary and sufficient to mediate the anti-proliferative activity of ATRA in sensitive gastric-cancer cells.

Networks of genes whose constitutive expression is associated with ATRA-sensitivity

As a first step towards the generation of a predictive tool for the selection of ATRA-sensitive gastric-cancer patients, we defined the constitutive gene-expression profiles of our panel of 27 cell-lines, employing the CCLE/RNA-seq data. The computational approach used identifies a limited number of transcripts whose basal expression levels correlate directly or inversely with the experimentally determined ATRA-score values in a quantitative manner (Fig. 3A). Indeed, 26 and 16 protein-coding mRNAs correlate with ATRA-sensitivity directly (high-Basal-Expression-Levels/high-ATRA-scores) and inversely (low-Basal-Expression-Levels/high-ATRA-scores). According to an analysis performed with the STRING database [23], the products of these 42 genes converge into 4 distinct networks of interacting proteins (Fig. 3B). The largest network (28 elements) contains proteins involved in tissue-development (PITX2/PAX9/ALX3/MEOX1/SIX6/TLE3) and the WNT pathway (WNT2/TLE3/EGF/ERBB3). The second network (6 elements) encloses proteins playing a role in myogenesis and collagen-homeostasis (TPM1/FLNC/NRAP/COL6A1/LOXL1/TLL1). The third and fourth networks consist of factors controlling pre-mRNA-processing (UPF3A/SRRM1/FIP1L1/NUP98) and metabolic-/mitochondrial-homeostasis (CPS1/BHMT/SIRT3/ING1).

Fig. 3

Transcriptomic model based on genes whose basal expression is associated with gastric-cancer cell-lines ATRA-scores. A The panel shows a heat-map illustrating the levels of the 42 genes whose constitutive expression is quantitatively associated with the ATRA-score values of the gastric-cancer cell-lines profiled for their sensitivity to ATRA. The mRNAs directly (high-Basal-Expression-Levels/high-ATRA-scores) and inversely (low-Basal-Expression-Levels/high-ATRA-scores) correlated with ATRA-sensitivity are marked in red and blue respectively. The expression values of the 42 mRNAs in gastric-cancer cell-lines are calculated with the use of the RNA-seq results available in the CCLE (Cancer Cell Line Encyclopedia) database. The values are expressed as Log 2 [CPM (Counts Per Million)]. The G-DIFF and G-INT cell-lines are marked in red and light-blue, respectively. B The panel illustrates the results of a STRING (Search-Tool-for-the-Retrieval-of-Interacting-Genes/Proteins) analysis performed on the 42 gene-products directly or inversely associated with ATRA-sensitivity. The genes directly and inversely associated with ATRA-sensitivity are marked by red and blue dots, respectively. C The panel shows the level of ATRA-sensitivity predicted in the 375 gastric-cancer samples of the TCGA dataset for which RNA-seq results are available. The predictions rest on the 42-gene model shown in panels (A) and (B) and they were generated with the use of a quantitative Similarity-score applied to the RNA-seq data. The ATRA-score threshold value of 0.55 is used to separate the cases predicted to be characterized by high ATRA-sensitivity (≥  0.55) and low ATRA-sensitivity (< 0.55). The G-DIFF and G-INT cases are marked in red and blue, respectively. The number and percentage of G-DIFF and G-INT cases observed in the high ATRA-sensitivity and low ATRA-sensitivity groups are indicated

We used the combined expression levels of the 42 transcripts to calculate Similarity-score values from the RNA-seq of the 375 gastric-cancer cases available in the TCGA-dataset. These Similarity-scores were utilized to predict the corresponding ATRA-score values (Fig. 3C). Indeed, we predict an ATRA-score ≥ 0.55 in 165 cases (high ATRA-sensitivity) and an ATRA-score < 0.55 in the remaining 210 cases (low ATRA-sensitivity). Thus, consistent with the proportion of gastric-cancer cell-lines experimentally responsive to ATRA (52%; Fig. 1C), 44% of gastric-cancers are predicted to be characterized by high ATRA-sensitivity. As expected from the cell-lines data, the G-INT phenotype is prevailing in patients characterized by predicted low ATRA-sensitivity (67%; Fig. 3C).

Effects of ATRA on gene-expression in gastric-cancer cell-lines

To obtain insights into the gene-networks mediating the action of ATRA in gastric-cancer, we performed RNA-seq studies in 13 of our cell-lines (Table S4). Seven and 6 cell-lines are characterized by high (ATRA-score > 0.55) and low (ATRA-score < 0.55) sensitivity to the retinoid, respectively (Fig. 4A). To perform these RNA-seq studies, we exposed each cell-line to vehicle or ATRA (1.0 µM) for 48 h, a time-interval preceding any sign of growth-inhibition. The number of transcripts up- and down-regulated by ATRA in each cell-line (Fig. 4A) shows a direct correlation with the ATRA-scores, as indicated by the linear-regression r-values (Fig. 4B). Indeed, the higher is the ATRA-score determined in each cell-line, the higher is the number of mRNAs up-/down-regulated by ATRA.

Fig. 4

Effects of ATRA on the gene-expression profiles of gastric cancer cells: RNA-seq pathway analysis. Exponentially growing triplicate cultures of the indicated cell lines were exposed to ATRA (1.0 µM) for 48 hours. At the end of the treatment cells were subjected to RNA-seq analysis (processed data in Table S4). A The panel shows the ATRA-sensitivity scores (ATRA-scores) of the 13 gastric cell-lines considered along with the number of genes significantly (FDR<0.1) up-regulated (UP) and down-regulated (DOWN) by ATRA (1.0µM) following 48 hours of exposure. B The diagram illustrates the correlations between the ATRA-score values and the number of genes up-regulated by the retinoid in our experimental conditions. The r-correlation values are indicated. C The RNA-seq data were subjected to pathway analysis using the HALLMARK data set. The Figure illustrates a Dot-Plot of the most significant HALLMARK pathways which are up-regulated (red dots) and down-regulated (blue dots) by ATRA in the indicated cell-lines. The size of the dots is inversely proportional to the FDR (False-Discovery-Rate) values calculated. When the FDR values are <0.1, they are considered to be statistically significant. The dots shown in dark color are statistically significant, while those shown in light color lack significance. Only the most relevant up- (red) or downregulated (blue) pathways are illustrated. The full results of the analysis are available in Fig. S4

To gather further information regarding the pathways regulated by ATRA in the responsive cell-lines, we performed GSEA on the RNA-seq data, using the HALLMARK platform (Figs. 4C and S4). The ATRA-dependent anti-proliferative effects are associated with a decrease in the expression of genes involved in the control of the G2M checkpoint and genes regulated by the E2F transcription-factor. In the retinoid-sensitive cell-lines, the growth-inhibitory action of ATRA seems to involve a specific down-regulation of c-myc target-genes (Figs. 4C and S4). In the majority of the cell-lines, regardless of their sensitivity to the retinoid, ATRA increases the expression of genes controlled by interferon-α (IFNα) and interferon-γ (IFNγ) (Figs. 4C and S4). These data indicate that the retinoid activates a series of IFN-dependent immune-responses, which may be necessary but insufficient for the anti-proliferative effects of ATRA in gastric-cancer cells. As for the potential metabolic pathways involved in the growth-inhibitory action of ATRA, the KEGG-Metabolism platform indicates that ATRA tends to up-regulate genes controlling glycerophospholids and retinol metabolism in all cell-lines (Fig. S5).

ATRA-dependent up-regulation of the genes involved in IFN-dependent immune-responses is accompanied by an increase in the HALLMARK “Allograft-Rejection” gene-set (Figs. 4C and S4). This HALLMARK network contains genes involved in antigen-presentation and T-cell dependent suppression of tumor-growth/metastatic-spread, which suggests that ATRA increases the immunogenicity of gastric-cancer cells. Thus, we took into consideration the REACTOME “Folding-Assembly-and-Peptide-Loading-of-Class-I-MHC” gene-set, which consists of 24 genes playing key roles in antigen-presentation. ATRA up-regulates this gene-network in the majority of our gastric-cancer cell-lines, regardless of retinoid-sensitivity and G-DIFF/G-INT phenotype (Fig. 5A). In particular, ATRA stimulates the expression of the HLA-A/B/C and the B2M genes whose products are components of the Major-Histocompatibility-Complex (MHC). To evaluate whether these ATRA-dependent transcriptomic effects translate into an increase of the HLA-A/B/C surface-antigens, we performed FACS (Fluorescence-Activated-Cell-Sorter) analyses in the gastric-cancer HGC-27, LMSU, KATO-III, AGS and the breast-cancer SKBR3 (positive control) cell-lines, using an anti-HLA-A/B/C antibody (Fig. 5B and C). Consistent with the general up-regulation of the REACTOME gene-network, HGC-27, LMSU and KATO-III cells show an ATRA-dependent increase in HLA-A/B/C surface-expression. Significantly, ATRA does not alter HLA-A/B/C surface-expression in AGS, the only cell-line showing a slight down-regulation of the REACTOME gene-network.

Fig. 5

Effects of ATRA on the process of antigen-presentation in gastric-cancer cell-lines. The indicated cell-lines were exposed to vehicle (DMSO) or ATRA (1µM) for 48 hours. At the end of the treatment, each cell-line was subjected to RNA-seq analysis. A The panel shows a heat-map illustrating the effects of ATRA on the expression levels of the 24 genes constituting the “Folding-Assembly-and-Peptide-Loading-of-Class-I-MHC” REACTOME gene-network in the indicated gastric-cancer cell-lines. The results are expressed as the ATRA-vehicle ratio [Log2FC (Fold-Change)]. The G-INT cell-lines are marked in blue and the G-DIFF cell-lines are marked in red. The cell-lines ATRA-score values are shown below the heat-map, as indicated. B and C The indicated gastric-cancer and the control SKBR3 breast cancer cell-lines were exposed to vehicle (DMSO) or ATRA (1µM) for 48 hours. At the end of the treatment, the cell-lines were subjected to FACS (Fluorescence-Activated-Cell-Sorter) analysis with an anti-HLA/B/C antibody. Panel B shows representative FACS graphs obtained with the indicated gastric-cancer cell-lines. Panel C shows the calculated FACS quantitative data. The data are expressed as the Mean+SD (N=3) of the AUC (Area Under the Curve) values determined from the FACS graphs. The surface expression of HLA/B/C was compared in each of the indicated vehicle-treated and ATRA-treated cell-lines. In case of significance (two-tailed Student’s t-test), the p-values of the comparisons are shown

Genes commonly modulated by ATRA in both G-INT and G-DIFF cell-lines

To identify genes potentially involved in the anti-tumor action of ATRA in G-INT tumors, we performed RNA-seq studies with 8 G-INT cell-lines exposed to ATRA for 48 h. In a first set of experiments, we focused on the G-INT cell-lines, GSU, KATO-III and IM95 (ATRA-score > 0.55) characterized by high ATRA-sensitivity. In these cell-lines, ATRA modulates the expression of a few thousands transcripts (Fig. S6, left). However, only 143 and 154 common mRNAs are up-regulated and down-regulated by ATRA, respectively, in all these cell-lines (Fig. 6A).

Fig. 6

Gene-networks modulated by ATRA in retinoid-sensitive G-INT and G-DIFF gastric-cancer cell-lines. The G-INT/ATRA-sensitive GSU/KATO-III/IM95 and the G-DIFF/ATRA-sensitive HGC-27/GCIY/RERF-GC-1B/LMSU cell-lines were exposed to vehicle (DMSO) or ATRA (1µM) for 48 hours and subjected to RNA-seq analysis. A Upper: heat-maps illustrating the effects of ATRA on the expression of the 297 genes commonly up-regulated (143 genes; UP) and down-regulated (154 genes; DOWN) in the 3 G-INT cell-lines (FDR < 0.1). The results (Mean of 3 independent vehicle-treated and ATRA-treated cultures) are expressed as the ATRA-vehicle ratio [Log2FC (Fold Change)]. Up-regulated/genes=red; Down-regulated/genes=blue; Non-protein-coding/genes=black. Lower: STRING (Search-Tool-for-the-Retrieval-of-Interacting-Genes/Proteins) analysis of the 143 up-regulated gene-products (left diagram; red dots) and the 154 down-regulated gene-products (right diagram; blue dots). The proteins in black squares are encoded by genes up-regulated and down-regulated in the retinoid-resistant G-DIFF cell-lines, AGS, NCI-N87, HuG1-N, MKN45, OCUM-1. B Upper: 2 heat-maps illustrating the effects of ATRA on the expression levels of the 43 genes commonly (FDR < 0.1) up-regulated (33 genes; UP) and down-regulated (10 genes; DOWN) in the in the 4 G-DIFF gastric-cancer cell-lines. The results are expressed as the ATRA-vehicle ratio [Log2FC (Fold Change)] and they represent the mean of 3 independent vehicle-treated and ATRA-treated cultures. Lower: STRING-analysis of the 33 up-regulated gene-products (red-dots) and the 10 down-regulated gene products (blue-dots). The proteins in black squares are encoded by genes up-regulated in the retinoid-resistant G-DIFF cell-line, GSS. C Upper: heat-map illustrating the effects of ATRA on the expression of the genes commonly up-regulated (6 genes; UP) and down-regulated (1 gene; DOWN) in ATRA-sensitive G-INT/G-DIFF cell-lines (FDR < 0.1). The results (mean of 3 vehicle-treated and ATRA-treated cultures) are expressed as the ATRA-vehicle ratio [Log2FC (Fold Change)]. Lower: STRING-analysis of the 6 up-regulated gene products (upper diagram; red dots) and the single down-regulated gene product (lower diagram; blue dots) which are common to the 7 ATRA-sensitive G-DIFF and G-INT cell-lines shown in panels A and B. The type of protein interactions available in the literature are shown in the rectangular boxes

As for the commonly up-regulated mRNAs, a STRING analysis of the data demonstrates that some of the corresponding translated products are part of 12 separate interactomes (Fig. 6A, lower-left). The largest interactome (37 elements) contains proteins involved in lipid-metabolism (HADHB/HADH/ECH1/ECI2/ALDH1A3/ADH1C/DHRS3/CYP26B1/CYP2B6/UGT1A8/SRD5A3/GALC/BGALT5), with particular reference to the retinol metabolic pathway (ALDH1A3/ADH1C/DHRS3/CYP26B1/CYP2B6/UGT1A8). This is consistent with the GSEA results obtained with the KEGG-Metabolism platform (Fig. S5). The second-largest interactome (8 elements) is centered on IRF1 (Interferon-Responsive-Factor-1) and it contains proteins involved in inflammation and antigen-presentation (IRF1/SOCS1/SAMD9/TLR1/PSMB10/CTSS). In this case too, up-regulation of the network is in line with the GSEA results (Fig. S4). The third-largest interactome (7 elements) contains proteins controlling ion-channels (KCNE3/KCNN4) and epithelial-polarization (CYTH3). The fourth-largest network includes the tumor-suppressor VWA5A [33] and 5 other proteins (SAMD12/PLPP1/AGPAT3/CDS2/CPD) regulating membrane phospholipids. Finally, one of the commonly up-regulated genes is DMBT1, a tumor-suppressive demethylase [34], which interacts with PGR.

As for the commonly down-regulated genes, the corresponding products aggregate into 14 interactomes (Fig. 6A, lower-right). The largest interactome (28 elements) contains factors involved in lipid metabolism (MOGAT3/DGAT2/PCK1/NR1H3/SCARB1/PCSK9/PON2/CES2/CYP2C19/CYP4F12/CYP2C18/CYP2W1/SLC16A1). Interestingly, other proteins controlling lipid metabolism (SPTLC3/DEGS2/UGT8/GAL3ST1) cluster inside a smaller network. This supports the idea that the anti-proliferative effects of ATRA in the G-INT neoplastic cells are associated with lipid-homeostasis modulation. The second-largest interactome of down-regulated gene-products (CDH17/ITGA6/COL17A1/DSG3/TRIM29/SERPINB5/PLOD3/LAD1/P3H2/TMEM154/S100A14/S100A16/S100A4) contains proteins involved in cell-adhesion and motility. Down-regulation of this last group of gene-products may relate to a potential increase in the process of epithelial-polarization mentioned in the case of the up-regulated CYTH3 gene.

To evaluate possible associations with ATRA-sensitivity, we conducted similar studies in 5 G-INT cell-lines (MKN45, AGS, NCI-N87, HuG1-N and OCUM-1) characterized by low ATRA-sensitivity (ATRA-score < 0.55). The 5 cell-lines respond to ATRA with the up-/down-regulation of several hundred genes (Fig. 7A). Nevertheless, the number of commonly up-/down-regulated genes is limited to 24 and 3, respectively (Fig. 7A and B). Interestingly, 8 genes (BGALT5/SRI/EPB41L1/TINAGL1/LARGE1/SQSTM1/STK39/DHRS3), are up-regulated in both G-INT cells showing low and high ATRA-sensitivity (Figs. 6A and 7B, black-squares). A similar situation is observed in the case of the down-regulated ID2 gene.

Fig. 7

Gene-networks modulated by ATRA in retinoid-resistant G-INT and G-DIFF gastric cancer cell-lines. The G-INT/retinoid-resistant MKN45/NCI-N87/AGS/OCUM-1/HuG1-N cell-lines and the G-DIFF/retinoid-resistant GSS cell-line were exposed to vehicle (DMSO) or ATRA (1.0 µM) for 48 hours and subjected to RNA-seq analysis. A The panel illustrates the number of genes selectively up-regulated (red) or down-regulated (blue) in each G-INT cell-line (squares) and commonly up-regulated (red) or down-regulated (blue) in the 5 cell-lines (circle). B The left side of the panel shows a heat-map illustrating the effects of ATRA on the expression levels of the 27 genes commonly and significantly (FDR < 0.1) up-regulated (24 genes; UP) and down-regulated (3 genes; DOWN) in the 5 retinoid-resistant G-INT gastric-cancer cell-lines. The results are expressed as the ATRA-vehicle ratio [Log2FC (Fold-Change)] and they represent the mean of 3 independent vehicle-treated and ATRA-treated cultures. The right side of the panel illustrates the results of a STRING (Search-Tool-for-the-Retrieval-of-Interacting-Genes/Proteins) analysis performed on the 24 up-regulated gene-products (red dots) and the 3 down-regulated gene products (blue dots). The proteins included in black squares are encoded by genes which are commonly up-regulated or down-regulated also in the 3 retinoid-sensitive G-INT cell-lines, GSU, KATO-III and IM95. C The panel shows two heat-maps illustrating the effects of ATRA on the expression levels of the 225 genes significantly (FDR < 0.1) up-regulated (137 genes; UP) and down-regulated (88 genes; DOWN) in the retinoid-resistant G-DIFF cell-line, GSS. In addition, the panel illustrates the results of a STRING analysis performed on the 7 gene products which are commonly up-regulated by ATRA in the GSS and the retinoid-sensitive G-DIFF cell-lines, HGC-27, LMSU, GCIY and RERF-GC-1B (black squares) or the retinoid-resistant G-INT cell-lines, MKN45, NCI-N87, AGS, OCUM-1 and HuG1-N (green dots). The results are expressed as the ATRA-vehicle ratio [Log2FC (Fold-Change)] and they represent the mean of 3 independent vehicle-treated and ATRA-treated cultures.

We extended our analyses to the G-DIFF context, performing studies in the G-DIFF cell-lines, HGC-27, RERF-GC-1B, LMSU and GCIY (ATRA-score > 0.55), showing high ATRA-sensitivity, as well as GSS cells, the only G-DIFF cell-line characterized by low ATRA-sensitivity (ATRA-score = 0.50). In HGC-27 and RERF-GC-1B cells, ATRA modulates the expression of 6,063 and 5,557 genes, respectively, while a lower number of genes is up- and down-regulated in LMSU (1,886 genes) and GCIY (914 genes) cells (Fig. S6, right). In these retinoid-sensitive cell-lines, ATRA up- and down-regulates 33 and 10 common genes, respectively (Fig. 6B). The STRING analysis performed on the 33 up-regulated genes demonstrates that the corresponding proteins cluster into 3 networks. The largest network (13 elements) centers on IRF1 and it contains factors regulating inflammation (BIRC3/TLR3/USF1/CCL2) and antigen-presentation (PSME1/PSMB10/TAPBPL2/ERAP1/ERAP2/CTSS). The second-largest network consists of 7 proteins (CYP26B1/DHRS3/NRIP1/RARB/GPRC5A/RARRES3/APOL3) controlling retinoid-metabolism and epithelial-differentiation. As for the 10 down-regulated genes, they do not code for any interacting protein. In retinoid-resistant GSS cells, ATRA causes a significant up- and down-regulation of 137 and 89 genes, respectively (Fig. 7C). Noticeably, 4 of these genes (CYP26B1/RARB/DHRS3/TINAGL1) are equally up-regulated in the retinoid-sensitive G-DIFF cell-lines, GCIY, HGC-27, RERF-GC-1B and LMSU (Fig. 7C, black-squares).

In a last set of studies, we looked for genes up- or down-regulated by ATRA in G-INT and G-DIFF cell-lines characterized by high ATRA-sensitivity (Fig. 6C). In all the G-DIFF and G-INT cell-lines ATRA increases the expression of 6 common genes, i.e. IRF1, DHRS3, CTSS, PSMB10, CYP26B1 and TINAGL1. In contrast, only AHNAK2 is down-regulated by ATRA in the entire set of G-INT/G-DIFF cell-lines. As already observed, the CTSS and PSMB10 gene-products interact with IRF1 and are part of the interactome centered on IRF1 itself (Fig. 6A and B). Furthermore, the CYP26B1 gene-product binds DHRS3 and the 2 proteins are part of a network involved in retinoid metabolism and epithelial-differentiation (Fig. 6A and B).

Role of IRF1 and DHRS3 in the growth inhibitory action of ATRA

Given the potential role played by IRF1 in inflammatory/immunological/IFN-dependent responses [35, 36] and by DHRS3 (Short-Chain-Dehydrogenase/Reductase-Family-16 C-Member-1 or Retinol-Dehydrogenase-17) in retinoid metabolism, we conducted functional studies on the 2 factors in HGC-27 cells.

In a first set of experiments, we transiently transfected HGC-27 cells with 2 IRF1-targeting siRNAs (siIRF1a and siIRF1b) and a control siRNA (siCTRL). We exposed the transfected cells to vehicle or ATRA for 48 h. In accordance with the up-regulation of the corresponding mRNA, ATRA increases the basal levels of the IRF1 protein in mock transfected (no siRNA) and siCTRL transfected HGC-27 cells, as indicated by the Western-blot analyses performed (Fig. 8A). By converse, the IRF1 protein is undetectable in siIRF1a and siIRF1b transfected cells regardless of vehicle or ATRA treatment. As ATRA co-regulates IRF1 and DHRS3 mRNAs, we determined the levels of the DHRS3 protein as well. In native and siCTRL-transfected cells, ATRA induces the DHRS3 protein (Fig. 8A). The two siRNAs, siIRF1a and siIRF1b reduce the basal levels of the DHRS3 protein. More importantly, the retinoid-dependent DHRS3 induction is abolish