The RNA-binding protein Tristetraprolin regulates RALDH2 expression by intestinal dendritic cells and controls local Treg homeostasis

Abstract

AU-rich element (ARE)-mediated mRNA decay represents a key mechanism to avoid excessive production of inflammatory cytokines. Tristetraprolin (TTP, encoded by Zfp36) is a major ARE-binding protein, since Zfp36^−/− mice develop a complex multi-organ inflammatory syndrome that shares many features with spondyloarthritis. The role of TTP in intestinal homeostasis is not known. Herein, we show that Zfp36^−/− mice do not develop any histological signs of gut pathology. However, they display a clear increase in intestinal inflammatory markers and discrete alterations in microbiota composition. Importantly, oral antibiotic treatment reduced both local and systemic joint and skin inflammation. We further show that absence of overt intestinal pathology is associated with local expansion of regulatory T cells. We demonstrate that this is related to increased vitamin A metabolism by gut dendritic cells, and identify RALDH2 as a direct target of TTP. In conclusion, these data bring insights into the interplay between microbiota-dependent gut and systemic inflammation during immune-mediated disorders such as spondyloarthritis.

Introduction

Epithelial barriers such as gut and skin are continuously exposed to environmental stimuli, including commensal and pathogenic microbes. In this context, it is essential for the host to control inflammatory processes in order to prevent local and systemic pathologies. Expression of key pro-inflammatory genes is regulated at the post-transcriptional level by the degradation of their mRNA¹. Presence of cis-acting AU-rich elements (AREs) in their 3’-untranslated regions (3’UTR) allows for the recruitment of RNA-binding proteins that control their turnover and subcellular localization. ARE-mediated mRNA decay is critical to maintain immune homeostasis, since deletion of these regulatory elements in the Tnf gene is sufficient to trigger a spontaneous, microbiota-dependent transmural inflammation in the distal ileum, as well as joint pathology^2–4. The major trans-acting regulatory factor in ARE-mediated Tnf mRNA decay is Tristetraprolin (TTP), encoded by the Zfp36 gene. TTP targets ARE-containing mRNAs for degradation by recruitment of deadenylase complexes^5,6. In addition to TNF, TTP controls the expression of multiple inflammatory mediators such as IL-23, IL-6, CXCL2 or GM-CSF^7–9. As a result, Zfp36^−/− mice develop a TNF- and IL-23-dependent syndrome characterized by sero-negative arthritis, psoriatic-like skin lesions and systemic inflammation^10,11. Many of these features are reminiscent of spondyloarthritis (SpA), a group of interrelated immune-mediated inflammatory rheumatic conditions comprising ankylosing spondylitis (AS), psoriatic arthritis, arthritis/spondylitis with inflammatory bowel disease and reactive arthritis^12,13. Multiple epidemiological studies have established a link between epithelial inflammation and SpA^12,14. Along these lines, we have previously shown that conditional deletion of TTP in epidermal cells was sufficient to trigger most of the cardinal features of the “TTP-deficiency” syndrome, including joint pathology¹⁵. These observations suggest that TTP plays a major role in the maintenance of immune homeostasis at epithelial barriers. One could therefore expect that absence of TTP would lead to dysregulated production of inflammatory cytokines in response to gut microbiota. Herein, we show that, in contrast to Tnf^ΔARE mice, deletion of Zfp36 is not accompanied by histological signs of gut pathology. Nevertheless, Zfp36^−/− mice display a microbiota-dependent increase in intestinal inflammatory markers that corresponds with the onset of systemic, joint and skin inflammation. We further show that lack of overt intestinal pathology is associated with local expansion of regulatory T cells (Tregs). We identified Aldh1a2 (encoding RALDH2) as a direct target of TTP in gut dendritic cells, linking increased retinoic acid metabolism and the local development and function of Tregs. Taken together, these data provide valuable insights into the role of intestinal mucosal homeostasis during immune-mediated inflammatory disorders such as spondyloarthritis.

Results

Zfp36^−/− mice do not exhibit intestinal pathology despite increased inflammatory markers

Since ARE-mediated control of Tnf mRNA stability plays a major role in promoting gut immune homeostasis², we reasoned that this could be driven by TTP. For this purpose, we evaluated the histological features of the ileum and colon of 9 month-old Zfp36^−/− mice. At this age, all mice display clinical signs of systemic inflammation (reduced weight in comparison to their Zfp36^+/+ or Zfp36^+/− littermates) and dactylitis. However, we did not observe major cellular infiltration, distortion of the crypts, or atrophy or flattening of the villi in histology sections of the terminal ileum (Fig 1A). Instead, crypt depth and villus length were even slightly greater. Levels of mRNA expression of a tight junction protein (Ocln, coding for Occludin), and markers of Paneth and goblet cells (Lyz1 and Muc2, respectively), were similar between both groups, suggesting that the ileal barrier architecture as well as epithelial subtypes are maintained in Zfp36^−/− mice (Fig 1B). These results indicate that Zfp36^−/− mice do not display overt signs of pathology in the small intestine. We reached similar conclusions for the large intestine (Fig 1C). In comparison, Tnf^ΔARE mice display clear signs of inflammation and villus destruction as early as 7 weeks of age². Despite this observation, fecal lipocalin 2 (LCN2), a classical marker of gut inflammation, was clearly increased in Zfp36^−/− mice as compared to their littermates (Fig 1D). Furthermore, FACS analysis revealed increased numbers of neutrophils in their lamina propria (SI LP) and mesenteric lymph nodes (MLN) (Fig 1E). We also observed increased mRNA levels for known direct targets of TTP such as Tnf, Cxcl2, Il1b and Il23a, along with Lcn2 and antimicrobial peptides, in the small intestine and to a lesser extent in the colon (Fig 1F–G). Taken together, these results indicate that Zfp36^−/− mice exhibit clear stigmata of local gut inflammation but do not develop overt ileitis nor colitis.

Figure 1. Subclinical gut inflammation in Zfp36^−/− mice.

(A) Villus length and crypt depth in ilea from mice of 6 to 9 months of age (n=5–8 mice/group, pooled from 2 independent experiments) with representative pictures of hematoxylin-eosin (HE) stained ilea. (B) Gene expression of a tight junction protein (Ocln, coding for Occludin), and markers of Paneth and goblet cells (Lyz1 and Muc2, respectively) by RTqPCR (n=8 mice/group, pooled from 2 independent experiments). (C) Representative pictures of HE stained colons. (D) Fecal lipocalin 2 (LCN2) by ELISA (n=14–16 mice/group, pooled from 2 independent experiments). (E) Neutrophil infiltration of the mesenteric lymph nodes (MLN) and the small intestinal lamina propria (SI LP) by flow cytometry (n=18 mice/group, pooled from 6 independent experiments). (F-G) Cytokine and antimicrobial peptide gene expression in total ileum (F) and colon (G) by RTqPCR (n=8 mice/group, pooled from 2 independent experiments). Results in (F-G) are expressed in arbitrary units (normalized against Actb mRNA levels, and each result expressed relative to its WT control group). Results are given as median ± interquartile range, and each dot represents a single mouse. Statistical significance (ns: non-significant, *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001) was assessed by the Mann-Whitney test (A, B, D, F, G) or the Kruskal-Wallis test (E).

Microbiota-dependent subclinical gut inflammation contributes to the TTP-deficiency syndrome.

To determine whether this subclinical gut inflammation was associated with alteration of the microbiome composition, we performed 16S rDNA sequencing of fecal extracts (collected from 5 month-old sex-matched Zfp36^+/+ and Zfp36^−/− littermates separated after weaning). There was no difference in total fecal bacteria levels in the two groups (Fig 2A). We did not observe significant differences in alpha diversity between the two groups based on Shannon and Simpson indexes. With principal coordinates analyses of the Morisita-Horn and Bray-Curtis beta diversity metrics, genotype only explained 13% and 23%, respectively, of the dataset variance (Fig 2B–C). Detailed analysis of the 16S rDNA sequencing data revealed the presence of 56 taxa. Out of these, the relative abundance of 11 taxa were found to be significantly altered in Zfp36^−/− mice (corrected p-value < 0.1, Fig 2D). In particular, Rikenellaceae and Bacteroidaceae families that were found to be enriched in the terminal ileum of AS patients¹⁶ were significantly increased in Zfp36^−/− mice; however, Prevotellaceae, depleted in these patients, were also decreased in these mice (Fig 2D). Several of the amplicon sequencing variants (ASVs) that were increased in Zfp36^−/− mice were identified as Porphyromonadaceae members, a family that was also found to be increased in AS patients (Supplementary Table S1). Thus, TTP-deficient mice display a discrete microbial signature that shares several features with the ones previously described in AS patients.

Figure 2. Contribution of gut microbiome to local and systemic inflammation in Zfp36^−/− mice.

(A-D) Feces were collected from WT or Zfp36^−/− mice at 5 months of age for bacterial DNA sequencing. (A) Total fecal bacteria levels by absolute quantitative RTqPCR (n=4–10 mice/group). (B) Shannon and Simpson alpha-diversity indexes. (C) Principal coordinates analyses of the Morisita-Horn and Bray-Curtis beta-diversity metrics. In these analyses, genotype explains 13% and 23%, respectively, of the dataset variance (ANOSIM analyses, 1000 permutations, p < 0.05). (D) Families and genera significantly affected in Zfp36^−/− mice (q-value < 0.1). (E) Modulation of gut microbiota by broad spectrum antiobiotic therapy (ABT) in Zfp36^−/− mice and WT controls, from 6 weeks of age until the day of sacrifice at 20 weeks of age, followed by flow cytometry, ELISA and RTqPCR experiments. (F) Total fecal bacteria levels by absolute quantitative RTqPCR in feces collected 2 months after the start of ABT (n=8–12 mice/group, pooled from 2 independent experiments). (G) Lipocalin-2 levels by ELISA in feces collected before (n=11–13 mice/group, pooled from 2 independent experiments) and 2 months after the start of ABT (n=5–7 mice/group). (H) Neutrophil infiltration of the small intestinal lamina propria (SI LP) and mesenteric lymph nodes (MLN) by flow cytometry (n=12–17 mice/group, pooled from 3 independent experiments). (I) Cytokine and antimicrobial peptide gene expression in total ileum by RTqPCR, expressed in arbitrary units normalized against Actb mRNA levels and relative to the WT control group (n=10–13 mice/group, pooled from 2 independent experiments). (J) Weight and (K) arthritis severity follow-up (arrow = start of antibiotic therapy) (n=12–17 mice/group, pooled from 3 independent experiments). (L) Serum levels of lipocalin-2 by ELISA (n=8–13, pooled from 2 independent experiments). (M) Neutrophil infiltration of the spleen (n=12–17 mice/group, pooled from 3 independent experiments) and (N) the skin (n=9–12 mice/group, pooled from 2 independent experiments) by flow cytometry. (O) IL-17 producing cells of the skin by flow cytometry (n=12–15 mice/group, pooled from 3 independent experiments). Results are given as median ± interquartile range and each dot represents a single mouse. Statistical significance (ns: non-significant, *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001) was assessed by 2-way ANOVA test (F-I, L-O).

Next, we investigated the potential impact of the gut microbiome on the local and systemic inflammation developed by Zfp36^−/− mice. For this purpose, we modulated the microbiota by broad-spectrum antibiotics in the drinking water, beginning at 6 weeks of age (Vancomycin 0.5g/l, Ampicillin 1g/l, Streptomycin 1g/l and Neomycin 1g/l). We maintained this treatment until analysis at 20 weeks of age (Fig 2E). As expected, antibiotic treatment clearly decreased the number of total fecal bacteria in both groups (Fig 2F). Local gut inflammation was reduced, with normalization of fecal LCN2 levels and decreased mRNA expression of ileal and colonic inflammatory markers in treated vs control Zfp36^−/− mice (Fig 2 and Supplementary Fig S1). There was also a trend towards reduced neutrophil infiltration in the SI LP and MLN from Zfp36^−/− mice (Fig 2G–I). In order to further define the contribution of microbial recognition, signaling through IL-1 receptor family and inflammasome pathways to gut inflammation, we deleted Zfp36⁻ in mice of Myd88^−/− and Casp1/11^−/− backgrounds. Both MyD88 and Caspase 1 have been found to be critical for intestinal inflammation in various models^4,17. We assessed mRNA expression of cytokines and inflammatory markers in ileum samples collected between 5 to 7 months of age (Supplementary Fig S2). We observed that Tnf upregulation in the absence of TTP was MyD88- and Caspase 1/11-dependent. This was not the case for Il1b and Cxcl2. Induction of Lcn2, S100a8 and S100a9 was strongly MyD88- and to a lesser extent Caspase 1/11-dependent. These results support the role of microbiota-derived signals and possibly of IL-1 family-related signaling pathways as a trigger for local gut inflammation in Zfp36^−/− mice.

In order to evaluate whether the altered gut microbiota from Zfp36^−/− mice had an effect on intestinal inflammation when transferred to WT mice, we analyzed gut inflammatory mediators in littermates directly separated after weaning or co-housed for several months with Zfp36^−/− mice. WT mice that were co-housed displayed slight but significant increase in the expression of Il1b, Lcn2, Reg3b and Reg3g in comparison to separated littermates (Supplementary Fig S3). This result suggests that altered gut microbiota from Zfp36^−/− mice contributes to gut inflammation, although this effect did not appear to be dominant.

We then evaluated the effect of oral antibiotic treatment on systemic, joint and skin features of the TTP-deficiency syndrome. Strikingly, Zfp36^−/− mice on antibiotics showed improved weight gain over time, and a prominent delay in the development of the joint pathology, as compared to untreated Zfp36^−/− mice (Fig 2J–K). Likewise, markers of systemic inflammation such as serum LCN2 levels and accumulation of neutrophils in the spleen were significantly decreased (Fig 2L–M). We previously demonstrated that TTP-deficient mice developed an IL-17-dependent skin inflammation¹¹. Upon oral antibiotics treatment, we observed reduced neutrophil recruitment to the skin, as well as decreased frequency of IL-17 production by lymphoid cells, including αβT cells, ILCs and γδT cells (Fig 2N–O). These results strongly suggest that microbiota-dependent intestinal inflammation contributes to the systemic, joint and skin pathologies developed by Zfp36^−/− mice.

Expansion of Tregs in the small intestine of Zfp36^−/− mice

Although Zfp36^−/− mice display clear microbiota-dependent disruption of intestinal immune homeostasis that contributes to systemic inflammation, this process does not lead to overt gut pathology. We explored the mechanisms that could account for this apparent paradox. The TTP-deficiency syndrome is associated with dysregulated Th17 responses¹¹. We evaluated the frequency of Th17 cells in the SI LP of Zfp36^−/− mice and littermates by flow cytometry. While the proportion and absolute number of RORγt⁺ cells among conventional Th cells was comparable in both groups (Fig 3A), their capacity to produce IL-17 (Fig3B) along with steady-state Il17a mRNA levels in ileal samples (Fig 3C) was significantly lower in Zfp36^−/− mice, suggesting that Th17 function rather than differentiation and proliferation was impaired. We therefore hypothesized that local regulatory mechanisms must be at play. We observed 1.5- and 2-fold increases in the proportion of Foxp3⁺ Tregs in CD4⁺ T cells from the mesenteric lymph nodes and SI LP of Zfp36^−/− mice and in the absolute number for SI LP Tregs, respectively (Fig 3D–E). Of note, this was not the case in the spleen and in the LP isolated from the colon, suggesting that gut micro-environmental factors associated to the small intestine are responsible for this observation. We noted that expansion of SI LP Tregs in Zfp36^−/− mice was dependent on microbiota-derived factors as we observed decreased frequency in AB-treated animals (Supplementary Fig 4). Importantly, SI LP Tregs from Zfp36^−/− mice displayed an increased capacity to produce IL-10, and we also noted increased Il10 mRNA levels in ileal samples under steady-state conditions (Fig 3F–G). It is possible to discriminate several subsets of Tregs based on their expression of transcription factors¹⁸. We observed a relative decrease in the proportions of thymic-derived GATA3⁺ and microbiota-dependent RORγt ⁺ subsets, as the expansion of Foxp3⁺ Tregs in Zfp36^−/− mice was restricted to GATA3⁻RORγt⁻ cells (Fig 3H). Taken together, these results indicate that both the frequency and the functionality of Tregs are enhanced in the small intestine of Zfp36^−/− mice. Furthermore, the pattern of transcription factors they express suggests that they are induced in the periphery in response to food-borne antigens¹⁸.

Figure 3. Increased proportion of Treg in the small intestinal GALT of Zfp36^−/− mice.

Flow cytometry and RTqPCR results in WT or Zfp36^−/− mice. (A) Proportion (left) of Th17 cells (RORγt⁺ cells) among Foxp3⁻ CD4 T cells in SI LP (n=12 mice/group, pooled from 3 independent experiments) and absolute number of Th17 in SI LP (right) (n=6–7 mice/group, pooled from 2 independent experiments). (B) Proportion of IL-17 producers among total CD4 T cells (n=4 mice/group). (C) Il17a gene expression in total ileum (n=10–12 mice/group, pooled from 3 independent experiments). (D) Proportion of Treg (Foxp3⁺ cells) among CD4 T cells in spleen, MLN, SI LP (n=16 mice/group, pooled from 5 independent experiments) and colon lamina propria (C LP) (n=7–9 mice/group, pooled from 3 independent experiments). (E) Absolute number of Treg (Foxp3⁺ CD4⁺ T cells) in SI LP (n=8–9 mice/group, pooled from 3 independent experiments). (F) Proportion of IL-10 producing Foxp3⁺ Treg in spleen, MLN and SI LP (n=4–6 mice/group, pooled from 2 independent experiments). (G) Il10 gene expression in total ileum (n=23 mice/group, pooled from 5 independent experiments). (H) Proportion of Foxp3⁺ subsets (GATA3⁺, RORγt⁺ and GATA3⁻RORγt⁻ cells) among CD4 T cells (n=7–8 mice/group, pooled from 2 independent experiments). RTqPCR results are expressed in arbitrary units, normalized against Actb mRNA levels and relative to the WT group. Results are given as median ± interquartile range and each dot represents a single mouse. Statistical significance (ns: non-significant, *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001) was assessed by Mann-Whitney test (A, B, C, E, G) or 2way ANOVA test (D, F, H).

Dendritic cells are the main drivers of this enhanced Treg development and function in the SI LP of Zfp36^−/− mice.

In order to decipher the mechanisms involved in this expansion of Tregs, we evaluated the expression of TTP in the different cellular compartments of the small intestine under steady-state conditions. As most anti-TTP antibodies cannot be reliably used for FACS staining, we used a newly developed Zfp36-V5 epitope tagged knock-in mouse generated by CRISPR/Cas9-mediated genome editing¹⁹. We observed consistent TTP-V5 staining in SI LP myeloid cells, in particular conventional dendritic cells (cDC) and monocytes, and to a lesser extent in macrophages and neutrophils (Fig 4A). We also noted expression of TTP-V5 in up to 30% of Epcam⁺ epithelial cells. In contrast, expression of TTP in most SI LP and intra-epithelial lymphoid subsets, including Tregs, was low or undetectable. We observed higher expression of TTP-V5 in cDC1 isolated from the SI LP in comparison to their counterparts from the spleen or the lung (Fig 4B). We reached similar conclusions for Ly6C⁺monocytes. These results indicate that DCs and monocytic cells from the gut express high levels of TTP under physiological conditions. To define whether the expression of TTP in myeloid cells isolated form the MLN and SI LP was dependent on the presence of gut microbiota, Zfp36-V5 mice were treated with broad-spectrum antibiotics in the drinking water during 4 weeks (Supplementary Fig S5). Antibiotic treatment reduced the frequency of monocytes and neutrophils in the MLN but not in the SI LP. The percentage and intensity of TTP-V5 expression among CD103⁺ and double-positive DCs subsets from the MLN but not the SI LP were decreased in these conditions. These observations suggest that TTP expression in DCs that migrate from the gut to the MLN is partially dependent on microbiota-derived signals. Based on these results, we decided to evaluate the role of TTP within CD11c-expressing cells (CD11cCre-Zfp36^fl/fl; Zfp36^ΔDC, targeting dendritic cells but also macrophage subsets) and myeloid cells (LysMCre-Zfp36^fl/fl; Zfp36^ΔM mice, targeting macrophages, neutrophils and monocytes). As previously described, neither strain developed signs of systemic inflammation^7,15,20. Furthermore, mRNA expression of inflammatory cytokines and antimicrobial peptides was not increased in ileal samples (Fig S6). In comparison to Zfp36^fl/fl controls, we observed an expansion of SI LP Tregs in Zfp36^ΔDC but not in Zfp36^ΔM mice, suggesting that TTP deficiency in DCs rather than in other myeloid subsets affects Treg homeostasis (Fig 4C). However, we cannot exclude that this effect was also mediated by CD11c⁺ intestinal macrophages. As in Zfp36^−/− mice, the proportion of IL-10-producing cells among Tregs was also increased in Zfp36^ΔDC mice as compared to Zfp36^fl/fl mice (Fig 4D). One of the key roles of GATA3⁻RORγt⁻ intestinal Tregs is the maintenance of oral tolerance to food-borne antigens¹⁸. We therefore asked whether deficiency of TTP in DCs affects de novo generation of intestinal Tregs upon oral administration of ovalbumin (OVA). Naive Foxp3⁻ CD4⁺ T cells from transgenic OTII mice were adoptively transferred to Zfp36^fl/fl or Zfp36^ΔDC mice. Mice were subsequently fed with OVA. We observed that conversion and expansion of OVA-specific Tregs in the small intestine and the MLNs was enhanced in Zfp36^ΔDC mice compared to Zfp36^fl/fl mice (Fig 4E).

Figure 4. Enhanced SI LP Treg development and function upon Zfp36 deletion in dendritic cells.

(A-B) Flow cytometry results in WT or Zfp36 V5 epitope tag knock-in mice, using an anti-V5 antibody. (A) Proportion of V5⁺ cells in different immune cell populations (n=3–7 mice/group, pooled from 3 independent experiments) and epithelial cells (n=10 mice/group, pooled from 3 independent experiments) in the small intestine at steady state. (B) Mean fluorescence intensity of V5 in conventional dendritic cells type 1 (cDC1: MHCII⁺CD11c⁺CD103⁺CD11b⁻ cells) and monocytes (CD11b⁺Ly6C⁺Ly6G⁻) from lymphoid organs (spleen, mesenteric lymph nodes) or barrier organs (SI LP, lung) (n=6–7 mice/group from 2 independent experiments). In MLN, cDCs were further differentiated into migratory (mig) or residential (res) subtypes. (C-H) Flow cytometry results in conditional Zfp36 knock-out mice strains (CD11c-Cre Zfp36^fl/fl: Zfp36^ΔDC and LysM-Cre Zfp36^fl/fl: Zfp36^ΔM) and their control Zfp36^flox/flox (fl/fl) mice. (C) Proportion of Treg (Foxp3⁺ cells) among CD4 T cells in SI LP (n=18–37 mice/group, pooled from 3–5 independent experiments). (D) Proportion of IL-10 producing Foxp3⁺ Treg cells in SI LP (n=7 mice/group, pooled from 3 independent experiments). (E) Increased peripheral Treg differentiation in SI LP from Zfp36^ΔDC mice in an oral tolerance model: Zfp36^fl/fl or Zfp36^ΔDC mice were fed for 6 days with ovalbumin, after transfer of naive CD4⁺OTII⁺ cells sorted from OTII mice (n=5–7 mice/group, pooled from 2 independent experiments). (F) Zfp36^fl/fl and Zfp36^ΔDC mice were infected with 10 cysts of T. gondii and sacrificed 7 days later to proceed to flow cytometry and RTqPCR experiments (G-H) (n=5–7 control mice or 16–18 infected mice, from 3 independent experiments). (G) Proportion of Treg (Foxp3⁺ cells) among CD4 T cells in spleen, MLN and SI LP. (H) Proportion of Tbet⁺ cells among Foxp3⁻ CD4 T cells in spleen, MLN and SI LP. Results are given as median ± interquartile range and each dot represents a single mouse. Statistical significance (*P<0.05, **P<0.01, ***P<0.001, ****P < 0,0001) was assessed by 2-way ANOVA (A: WT vs. Zfp36-V5 tagged mice, B: MFI in SI LP from Zfp36-V5 tagged mice compared to MFI in other organs, G, H), by Kruskal-Wallis (C) or by Mann-Whitney test (D, E).

To further define whether Tregs from Zfp36^ΔDC mice were efficient in the suppression of inflammation in vivo, we infected these mice orally with cysts of T. gondii (Fig 4F). In this model, Tregs are required to suppress excessive mucosal Th1 responses²¹. At the peak of the infection, the proportion of Foxp3⁺ Treg cells among CD4⁺ T cells of the SI LP (and MLNs) remained increased in infected Zfp36^ΔDC mice compared to Zfp36^fl/fl infected mice, to the same extent as in non-infected mice (Fig 4G). This increase in Treg cells was associated with a reduction of Tbet-expressing Th cells both locally and in the spleen (Fig 4H). Taken together, our results strongly support the notion that enhanced Treg development and function in the small intestine occurs as a consequence of Zfp36 deletion in gut-associated DCs, and potentially controls the intestinal inflammation observed upon deletion of Zfp36.

RALDH2 is a direct target of TTP

Small intestine CD103⁺ DCs display strong tolerogenic properties and promote the conversion of naive CD4 T cells into Tregs through the production of retinoic acid (RA)²². We therefore evaluated the role of TTP in DC homeostasis. There were no significant differences between Zfp36^ΔDC and Zfp36^fl/fl mice in terms of either total SI LP DC frequency or in the proportions of CD103⁺, CD11b⁺ and double-positive subsets (Fig 5A–B). However, we observed a strong increase in retinaldehyde dehydrogenase (RALDH) activity in SI LP DCs of Zfp36^ΔDC mice using Aldefluor (a RALDH fluorometric substrate) (Fig 5C). RALDH produces RA from retinaldehyde, a derivative of Vitamin A. In order to test if the increased capacity to generate RA by DCs was responsible for the expansion of Tregs in the gut of Zfp36^ΔDC mice, we subjected these mice to a vitamin A-deficient (VAD) or control diet from embryonic day 14 up to 8 weeks of life (Fig 5D). As expected, Aldefluor assays on VAD mice showed a reduction of RALDH⁺ DCs compared to both Zfp36^fl/fl and Zfp36^ΔDC mice on control diets (Fig 5E). The frequency of splenic Tregs was only marginally affected by the VAD diet. In contrast, we observed a decrease in the proportion of SI LP Tregs in VAD Zfp36^ΔDC mice. Importantly, in these conditions, the frequency of Foxp3⁺ cells among CD4⁺ T cells was comparable in Zfp36^fl/fl and Zfp36^ΔDC mice, indicating that increased RALDH activity in TTP-deficient intestinal DCs controls local Treg expansion (Fig 5F).

Figure 5. Aldh1a2 mRNA is a direct target of TTP.

(A-B) Flow cytometry results of DC subsets in SI LP of Zfp36^fl/fl and Zfp36^ΔDC mice. (A) Proportion of conventional dendritic cells (MHCII⁺CD11c⁺) among live CD45⁺ cells (n=15–17 mice/group, pooled from 3 independent experiments). (B) Proportion of cDC subsets of the SI LP among cDCs (n=15–16 mice/group, pooled from 3 independent experiments). (C) Aldefluor assay assessing RALDH activity in MHCII⁺CD11c⁺ cells of the SI LP by flow cytometry. Representative histogram (left) and ΔMFI (MFI sample – (MFI sample+DEAB), see methods) (right) in the SI LP of Zfp36^fl/fl and Zfp36^ΔDC mice (n=12 mice/group, pooled from 3 independent experiments). (D) Pregnant Zfp36^fl/fl or Zfp36^ΔDC mice were fed with a vitamin A-deficient diet (VAD) that was maintained in the pups after weaning until sacrifice at 8 weeks of age, followed by flow cytometry experiments (E-F). Control mice were fed with a normal diet. (E) Representative Aldefluor in MHCII⁺CD11c⁺ cells in SI LP. (F) Proportion of Treg cells in SI LP and spleen (n=9–13 mice/group, pooled from 3 independent experiments). (G) RTqPCR results on sorted SI LP DCs (MHCII⁺CD11c⁺) and non DCs (CD11c⁻) from Zfp36^fl/fl and Zfp36^ΔDC mice. Results are expressed in arbitrary units, normalized against Rpl32 and Ubiquitin and relative to the WT DCs group. Representative of 3 independent experiments in which each dot originates from a pool of 6 mice. Statistical significance (*P < 0.05, **P < 0.01) was assessed by 2-way ANOVA test. (H) Half-life (in min) of Aldh1a2 mRNA in BMDCs from Zfp36^fl/fl or Zfp36^ΔDC mice treated with LPS for 2h prior to Actinomycin D (10μg/μl) and SB203580 (1μM) treatment for the indicated time. mRNA was quantified by RTqPCR and normalized by mRNA levels at t₀ (n=3–4). (I) Schematic representation of the dual reporter plasmids generated to assess the role of the ARE elements: 3’UTR of Aldh1a2 mRNA with (AU+) or without (AU-) ARE was inserted behind the sequence coding for Renilla luciferase (R-Luc). Plasmids containing the WT Globin 3’UTR (AU0 = no ARE) or Globin 3’UTR with AU rich insertion (AUUU)₈ (AU8 = canonical ARE sequence) were used as control. The plasmids also contain the sequence coding for the Firefly luciferase (F-Luc) as a transfection efficiency control. Both luciferases are under the control of a bidirectional CMV promoter. (J) Fold change between the Rluc/Fluc ratio of AU+ and AU- luciferase signals from dual reporter plasmids co-transfected with TTP-Flag (or BOIP-Flag as control) in HEK293T cells (n=7 experiments). (K) Electrophoretic mobility supershift assay. Extracts from HEK293T cells transiently transfected with TTP-Flag or BOIP-flag were incubated with a ³²P-labeled probe of Aldh1a2 ARE of a CTRL probe and either no antibody, an α-V5 or an α-Flag antibody before migration. Representative of 3 independent experiments. Statistical significance (ns: non-significant, *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001) was assessed by Mann-Whitney (A, C) or by 2-way ANOVA test (B, F, G, J).

There are 3 different retinaldehyde dehydrogenases in mice (RALDH1, RALDH2 and RALDH3 encoded by Aldh1a1, Aldh1a2 and Aldh1a3, respectively). We therefore quantified their mRNA expression in FACS-sorted SI LP DCs from Zfp36^ΔDC mice. As a control, we sorted all CD11c⁻ cells. As compared to Zfp36^fl/fl, Zfp36 mRNA expression was strongly decreased in CD11c⁺ cells, validating our strategy for conditional deletion of this gene in DCs. Consistent with previous results¹¹, absence of TTP promoted expression of Il23a in these cells. Aldh1a1 expression was comparable in both groups, while Aldh1a3 was undetectable. In contrast, we observed increased Aldh1a2 expression in Zfp36-deficient SI LP DCs (Fig 5G). We identified several putative ARE sequences in the 3’UTR of the mRNA coding for RALDH2, suggesting that it could be directly regulated by TTP. To define whether Aldh1a2 is a bona fide TTP target, we performed in vitro experiments with bone marrow (BM) derived-DCs. After a short incubation with LPS to induce TTP expression, we evaluated Aldh1a2 mRNA half-life in WT and Zfp36-deficient cells by treating the cells with actinomycin D and SB203580, the latter used to abrogate the inhibitory action of p38 MAPK on TTP activation⁷. As shown in Fig 5H, Aldh1a2 mRNA stability was strongly increased in Zfp36-deficient BMDCs. Next, to assess if the putative ARE sequences found in the 3’UTR of the mRNA coding for RALDH2 were sufficient to promote mRNA destabilization by TTP, we used a bidirectional reporter system²³ in which we inserted either the full-length 3’UTR sequence of Aldh1a2, or a truncated version lacking ARE motifs (Fig 5I). Plasmids containing a synthetic (AUUU)8 ARE motif (AU8) or no ARE (AU0) were used as controls. We evaluated the effect of TTP in co-transfection experiments in HEK293T. As expected, we observed a clear effect of TTP on the ratio between AU8 and AU0 reporter activities as compared with a control plasmid (expressing BOIP, encoded by Ccdc89) (Fig 5J). Similar conclusions were reached for the construct containing Aldh1a2 3’UTR, with and without ARE motifs. Finally, to test the capacity of TTP to directly bind Aldh1a2 ARE, we performed an electrophoretic mobility supershift assay. We incubated the Aldh1a2 ARE probe with extracts from TTP-Flag expressing-HEK293T cells. Addition of α-Flag antibody resulted in a supershift that was not observed with BOIP-Flag or with an anti-V5 antibody, demonstrating that TTP physically interacts with the Aldh1a2 mRNA ARE in this in vitro setting (Fig 5K). Taken together, these results indicate that TTP directly regulates Aldh1a2 mRNA stability in intestinal DCs, and that this mechanism controls local Treg homeostasis.

Discussion

Intestinal homeostasis relies on a coordinated set of innate and adaptive responses that calibrates responses against commensals and food-borne antigens²⁴. Activation of innate immune cells by the constant sensing of microbiota-derived signals has to be tightly regulated to avoid local inflammatory pathology. TTP represents a major mediator of dynamic posttranscriptional regulation²⁵. In immune and stromal cells, TTP regulates the expression of multiple inflammatory cytokines²⁶. It is therefore not surprising that Zfp36^−/− mice, which spontaneously develop a multi-organ inflammatory syndrome, also exhibit high levels of expression of inflammatory markers in both small and large intestines. We observed constitutive expression of TTP in intestinal DC and macrophage subpopulations. Intriguingly, TTP deficiency in CD11c⁺ cells or in myeloid cells was not sufficient to elicit increased expression of inflammatory markers in the gut. We reached similar conclusions in preliminary experiments when we targeted intestinal epithelial cells using Villin^CreZfp36^fl/fl mice. These results suggest that TTP deficiency in multiple cellular compartments could be required to drive dysregulated cytokine production. However, heightened expression of inflammatory cytokines in Zfp36^−/− mice was not associated with any signs of intestinal pathology.

Tregs play a dominant role in the control of intestinal homeostasis, and the gut-associated lymphoid tissue represents a privileged site for the peripheral induction of Treg cells in response to oral antigens^27,28. This is related to the capacity of CD103⁺ DCs to produce retinoic acid^28,29. Although we didn’t demonstrate the heightened capacity of isolated intestinal TTP-deficient DCs to induce Treg development in in vitro assays, we show that TTP directly regulates RALDH2 in these cells, thereby influencing local Treg homeostasis. Of note, retinoic acid supplementation attenuates ileitis in Tnf^ΔARE mice by restoring the balance between Th17 and Treg cells³⁰. Hence, we propose that in the absence of TTP, dysregulated production of inflammatory cytokines in the gut is balanced by enhanced Treg development and function. However, we did not formally demonstrate the role of the RALDH2-Treg axis in promoting intestinal homeostasis in TTP-deficient animals. Other regulatory pathways could contribute to the local control of inflammation in TTP-deficient animals. For instance, TTP was shown to target IL-10, a cytokine that plays a major role in the maintenance of intestinal homeostasis³¹. IL-10 expression is inhibited in BM-derived macrophages from mice with the Zfp36^aa gain-of-function mutation. However, the in vivo effects of this TTP mutation were uniformly anti-inflammatory, suggesting that this mechanism is not dominant³². As TTP directly controls IL-22 production, this could also promote local mucosal integrity and homeostasis³³. Furthermore, a recent report indicated that deletion of Zfp36 in intestinal epithelial cells decreased the severity of dextran-sodium sulfate-induced colitis, possibly by targeting Nos2³⁴. Altogether, these observations could account for the clear phenotypic differences between Zfp36^−/− and Tnf^ΔARE mice.

Microbiota-derived signals set the immunological tone of tissues, both locally and systemically²⁴. Hence, the contribution of altered gut immune homeostasis to the pathogenesis of rheumatic diseases is an area of intense research³⁵. In particular, there is a clear connection between joint symptoms and gut inflammation in spondyloarthritis. Up to 10% of ankylosing spondylitis patients also develop Crohn’s disease during the course of the disease, and it is considered that nearly 50% of patients have subclinical gut inflammation³⁶. Of note, in line with the present observations in Zfp36^−/− mice, ileal inflammation of ankylosing spondylitis patients is characterized by normal IL-17 levels despite increase in IL-23 expression³⁷. Strikingly, this immunological signature was also associated with expansion of IL-10-producing Tregs³⁸. Several studies reported alterations of the gut microbiota composition in patients with ankylosing spondylitis or psoriatic arthritis^16,39–41. Although Zfp36^−/− mice did not display clear dysbiosis, we observed discrete alterations that are compatible to the ones described in ankylosing spondylitis patients¹⁶. It is not clear, however, whether these perturbations are the consequence or the cause of inflammation. Nevertheless, we show that modulation of the gut microbiota with broad spectrum antibiotics consistently reduced intestinal inflammatory markers as well as systemic, joint and skin inflammation. We cannot exclude that long-term oral antibiotic treatment also affects skin microbiota. Indeed, we previously demonstrated that invalidation of Zfp36 in epidermal cells was sufficient to trigger skin and also joint inflammation¹⁵.

Our results are consistent with the observation that raising HLA-B27 transgenic rats in a germ-free environment prevents intestinal inflammation and reduces axial arthritis⁴². Together with the observation that Zfp36^−/− mice bred in conventional housing develop more severe inflammation¹⁵, these experiments support the notion that microbial-derived signals have an exacerbating effect on spondyloarthritis pathological features. The cellular and molecular mechanisms at play remain ill-defined, but the present study indicates that Zfp36^−/− mice might represent an interesting pre-clinical model to address these mechanistic questions. It also provides a good rationale to evaluate the regulation of ARE-mediated RNA degradation in these patients.

In conclusion, our results begin to decipher the complex role of TTP in intestinal immune homeostasis, and provide a framework to study the interconnections between gut, skin and joint inflammation in the context of spondyloarthritis.

Materials and methods

Mice

Zfp36-deficient mice (Zfp36^−/−), LoxP-flanked Zfp36 mice (Zfp36 ^flox/flox) and Zfp36-V5 knock-in mice on a C57BL/6 background were previously described^10,19,20. CD11c-Cre (B6.Cg-Tg(Itgax-Cre)1–1Reiz), LysM-Cre (B6.129P2-Lyz2^tm1(cre)Ifo) mice on a C57BL/6 background were purchased from The Jackson Laboratory. OTII Foxp3-GFP mice were generated by crossing OT-II mice (B6.Cg-Tg(TcraTcrb)425Cbn/J mice, purchased from The Jackson Laboratory) with Foxp3-DTR-GFP mice (B6.129(Cg)-Foxp3^{tm3(DTR/GFP)Ayr} mice), which were kindly provided by Adrian Liston (KU Leuven) with permission from Alexander Rudensky (Howard Hughes Medical Institute and Immunology Program, Sloan-Kettering Institute, New York, NY). For antibiotic treatment, mice were treated for the indicated times with Vancomycin 500mg/l, Ampicillin 1g/l, Neomycin 1g/l (Alfa Aesar) and Streptomycin 1g/l (PanReac Applichem) ad libitum in their drinking water. When indicated, pregnant females were fed vitamin A-deficient (SAFE) or sufficient diets from day 14 of gestation and then kept on the same diet until weaning of the litter⁴³. Upon weaning, mothers were all returned to classical diets, while the weaned younglings were kept on the same special diet until 8 weeks of age, when the experiments were performed. All experiments were performed using littermates as controls. Male and female mice were used for the experiments. In each experiment, mice were age- and sex-matched. All mice were bred and maintained in a specific pathogen-free animal facility. All animal studies were approved by the institutional Animal Care and local committee for animal welfare of the BIOPOLE ULB CHARLEROI.

Flow cytometry analysis.

Cells from spleen, MLN, LP and skin were obtained as described in^15,43. Cells were resuspended in RPMI-1640 medium containing 10% FBS. Single-cell suspensions were incubated with anti-FcγIII/II receptor antibody and stained with fluorochrome-conjugated mAb for phenotype analysis. LIVE/DEAD^™ Fixable Near-IR dye (ThermoFischer) was used to exclude dead cells. Antibodies to TCRβ-PerCP-Cy5.5/FITC (H57–597), CD4-PE-Cy7/BV605 (RM4–5), CD8α-AF700 (53–6.7), Foxp3-APC/FITC (FJK-16s), RORγt-AF647 (Q31–378), GATA3-PE (L50–823), Tbet-PE (eBio4B10), IL-10-BV421 (JES5–16E3), IL-17A-APC/PE (TC11–18H10), MHCII (I-A/E)-PE-Cy7/V500/FITC (M5/114.15.2), CD11c-PerCP-Cy5.5/PE-Cy7 (HL3), CD11b-PE/AF700 (M1/70), CD103-APC/BV421 (2E7/M290), Vα2 TCR-PE (B20.1), Vβ5 TCR-biotin (MR9–4), V5 Tag-AF647, CD326-PE-Cy7 (G8.8), CD45-FITC/AF700/BV510 (30-F11), Ly6G-PerCP-Cy5.5 (1A8), TCRγδ-PerCP-eFluor710 (eBio-GL3), CD127-PE-Cy7 (SB/199), CD3- (145–2C11), CD19- (1D3), B220- (RA3–6B2), Gr1-APC-Cy7 (RB6–8C5), CD90.2-AF700 (30-H12), Ly6C-BV421/BV605 (AL-21), CD3-BV421/BV711 (B3B4/145–2C11), Streptavidin-PE-Cy7/PerCP-Cy5.5 were purchased from Becton Dickinson, Biolegend, eBioscience and Invitrogen (see references in Table S2). For cytokine staining, cells were incubated with PMA (25ng/ml), ionomycin (500ng/ml) and Brefeldin A (20μg/ml) for 2h (skin) or with PMA (50ng/ml), ionomycin (1μg/ml) and Brefeldin A (20μg/ml) for 3h (SI LP) before staining. Intracellular staining was performed using Foxp3/Transcription Factor Staining Buffer kit or Intracellular Fixation & Permeabilization kit (eBioscience) for transcription factor and cytokine analysis, respectively. Concomitant staining for Fopx3 and IL-10 was performed using Foxp3/Transcription Factor Staining Buffer kit (eBioscience). Data were collected on a BD LSRII Fortessa or FACSCanto^™ II (BD Bioscience) and analyzed with FlowJo v10 software.

Arthritis scoring

The paws were monitored for clinical symptoms every week from 6 weeks until 3 months of age, then every two weeks until sacrifice. Disease severity scoring system: 0 = normal; 1 = swelling of one toe; 2 = swelling of more than one toe in the same paw; 3 = swelling of the entire paw or several toes in different paws, 4 = ankylosed paw.

Oral tolerance model

Single-cell suspension of spleen and lymph nodes were obtained from OTII Foxp3-GFP mice and enriched in CD4⁺ T cells by positive selection on a MACS column (Miltenyi). Cells were then labelled with fluorescent dye conjugated monoclonal anti-TCRβ and anti-CD4 antibodies and sorted by flow cytometry using FACSAria III (BD Bioscience) as TCRβ⁺ CD4⁺ GFP⁻ cells up to >95% purity. Purified naïve T cells from OTII Foxp3-GFP mice were adoptively transferred into Zfp36 ^flox/flox or Zfp36^ΔDC mice. Each mouse was injected i.v. with at least 1,5×10⁶ naïve T cells in PBS. Mice were then fed a 1.5% OVA solution (Sigma-Aldrich) in drinking water for 6 days. On day 6, cell suspensions from spleen, MLN and SI LP were prepared for flow cytometry analysis.

Toxoplasma gondii infection

ME-49 type II strain of Toxoplasma gondii was kindly provided by Dr. De Craye (ISP, Belgium) and was used for the production of tissue cysts in C57BL/6 mice. For acute T.gondii infection, cysts collected from the brains of chronically infected mice were counted, and mice were fed 10 T.gondii cysts in Phosphate-Buffered Saline (PBS) solution per oral gavage (200μl/mouse). Mice were sacrificed at day 7 post-infection for analysis.

Histology

Mouse tissues were fixed in 4% paraformaldehyde and directly paraffin embedded. Sections (6μm) were stained with hematoxylin-eosin. Villus length and crypt depth were measured using the Leica Image Viewer Software or NanoZoomer Digital Pathology software; a mean of 10 measures was calculated for each parameter per sample.

Aldefluor assay

An ALDEFLUOR^™ assay kit (STEMCELL Technologies) was used according to manufacturer’s instruction. Briefly, cells were stained with LIVE/DEAD^™ Fixable Near-IR dye (APC-Cy7), MHCII (I-A/E)-PE Cy7 (M5/114.15.2) and CD11c-PerCP-Cy5.5 (HL3), then incubated with ALDEFLUOR^™ activated reagent with or without DEAB inhibitor for 30 min at 37°C in the dark. Cells were washed and suspended in ALDEFLUOR^™ buffer for rapid flow cytometry analysis. Gating strategies used were as follow: for each sample, the gate for Aldefluor⁺ cells was placed according to the same sample + DEAB, so that the Aldefluor⁺ cells in the sample + DEAB represented <1% of parent population. (R)ALDH activity was assessed by the MFI of Aldefluor in DCs. To correct for background fluorescence, the MFI for Aldefluor in the cell sample incubated with DEAB was subtracted from the MFI for Aldefluor in the same sample incubated without inhibitor. Results are presented as ΔMFI = MFI_(sample) – MFI_{(sample + DEAB)}.

Sorting of LP dendritic cells

LP cells were isolated and stained as described above, then sorted on BD FACSAria III (SI LP DCs = live MHCII⁺CD11c^hi and non DCs = live CD11c⁻). The collected cells were resuspended in TRIzol reagent (ThermoFisher) to proceed to RNA extraction.

Gene expression

Total RNA was extracted with NucleoSpin RNA (Macherey) and reverse-transcribed with the RevertAid RT Reverse Transcription kit (ThermoFisher). cDNA was amplified using SYBR green or TaqMan probes. Primer sequences are available in Table S3. Half-life of mRNAs was determined using nonlinear regression (one phase decay) in GraphPad Prism.

Cytokines production

The concentration of LCN2 was determined in serum or feces by ELISA (R&D system). Feces were first homogenized in PBS using MagNA Lyser green beads (Roche). Supernatants were then collected to proceed to ELISA experiments.

Gut microbiota analysis

Genomic DNA was extracted from feces of littermate mice separately housed according to their genotype after weaning using a QIAamp DNA Stool Mini Kit (Qiagen), including a bead-beating step with zirconia beads (BioSpec). Absolute quantification of the total bacteria levels was performed using qPCR (primers presented in Table S4). The samples were PCR-enriched for the V5–V6 region of the 16S rRNA gene and then underwent a library tailing PCR (primers in Table S4). The amplicons were purified, quantified and sequenced using an Illumina MiSeq to produce 2 × 300 bp sequencing products. Initial quality-filtering of the reads was conducted with the Illumina Software, yielding an average of 97 296 pass-filter reads per sample. Quality scores were visualized, and reads were trimmed to 220 bp (R1) and 200 bp (R2). The reads were merged with the merge-Illumina pairs application⁴⁴. All samples were subsampled to 33 000 sequences per sample using Mothur 1.32.1⁴⁵. The UPARSE pipeline implemented in USEARCH was used to further process the sequences. Amplicon sequencing variants (ASVs) were identified using UNOISE3⁴⁶. Taxonomic prediction was performed using the nbc_tax function, an implementation of the RDP Naive Bayesian Classifier algorithm⁴⁷. The phylotypes were computed as percent proportions based on the total number of sequences in each sample. Beta-diversity indexes were calculated using QIIME⁴⁸. PCoA plot of the beta-diversity indexes were obtained using EMPeror⁴⁹.

Cell culture

HEK293T cells (ATCC) were maintained in DMEM medium (Gibco) containing 10% FBS, 50 U/ml Penicillin, 50μg/ml Streptamycin (Pen Strep, Gibco) and 1mM sodium pyruvate (Gibco). Transfections were performed using the calcium phosphate method of transfection. Plasmid DNA was added to a CaCl₂ solution, then added dropwise into a Hepes-buffered phosphate solution. This DNA solution is then incubated at room temperature for 30 min before being spotted on the cell culture. Cells’ medium was changed prior to adding the DNA and after incubation of the cells overnight with the DNA mixture. Cells were harvested on day 3 for experiments. BMDC were differentiated from fresh bone marrow using GM-CSF as follows: Cells obtained from the bone marrow of Zfp36^+/+ and Zfp36^−/− mice were plated at 0,5×10⁶ cells/ml in Petri dishes (10 ml/dish) in RPMI-1640 medium (Lonza) containing 10% HyClone FBS (LPS-free), 50 U/ml Penicillin, 50μg/ml Streptamycin (Pen Strep, Gibco), 1mM sodium pyruvate (Gibco) and non-essential amino acids (Gibco, dilution 100) and supplemented with GM-CSF (20 ng/ml, RnD Systems). Cells were fed at day 3 and 6 of culture with fresh medium containing GM-CSF (20ng/ml). At day 7, cells were seeded (1,5 to 2 × 10⁶ cell/ml) for experiments the next day. For half-life experiments, at day 8, BMDCs were stimulated with LPS (100ng/ml, Sigma) for 2h before treatment with Actinomycin D (10 μg/ml, Sigma) and the P38 inhibitor SB203580 (2 μM, Calbiochem) for the indicated times.

Dual reporter plasmids

The WT 3’UTR of Aldh1a2 gene was amplified by RT-PCR using the primers 5’-agctagagcggccgcggatccgaggccgaggctgaagagca-3’ and 5’-gctcgaagcggccgcgctagcacgaggtttctttttgaa-3’, then cloned into the NotI site of the AU₀ plasmid from Barreau et al.²³, using a ligation independent cloning (LIC) method. For RALDH2ΔARE, a DNA fragment containing the 3’UTR of RALDH without the putative ARE motifs was synthesized (Integrated DNA Technologies), then amplified by PCR before cloning in the AU₀ plasmid.

Electrophoretic mobility supershift assay

Analysis of possible Aldh1a2 mRNA-TTP interactions by EMSA supershift assay was performed as described in⁵⁰ with the following modifications. Cell extracts from HEK293T cells transfected with TTP-Flag of BOIP-Flag were incubated with ³²P-labelled RNA probe, followed by the addition of anti-V5 or anti-Flag antibody prior to sample loading on 5% polyacrylamide non-denaturing gels containing 6% glycerol and 0.5X TBE at 7.5 mA for 16 hours at 4°C.

Statistical analysis

Results are expressed as median ± interquartile range. The statistical significance was assessed using the two-tail Mann-Whitney test, the Kruskal-Wallis test with Dunn correction or the two-way ANOVA test with Sidak correction (GraphPad Prism 6.0). Results were considered significant for P < 0.05 indicated by “*”, P < 0.01 indicated by “**”, P < 0.001 indicated by “***” and P < 0.0001 indicated by “****”. For the gut microbiota sequencing results, p-values were corrected using the Benjamini-Hochberg correction for multiple testing (q-value) and the significance threshold was placed at q-value < 0.1.

Data Availability

Raw data generated during the analysis of the gut microbiota composition can be accessed on SRA (SRA accession: PRJNA558549).