The RNA binding protein ZFP36L2 displays tissue-selective mRNA targeting in mice Authors: George Stephen Stephenson, Dalia Fleifel, Jeanette Gowen Cook, Alain Laederach, Silvia Beatriz Vieira Ramos Published: 2026 ABSTRACT ABSTRACTZFP36L2 (zinc finger protein 36 like 2, C3H type-ZFP) is an RNA-binding protein targeting transcripts rich in adenine-uridine elements (AREs). Previous transcriptomic analysis suggested that ZFP36L2 displays a distinct transcript preference or ‘specific activity’. However, this analysis was restricted to a few tissues. Here, using experimental data in multiple tissues we detected a remarkable transcript selectivity depending on the tissue. Given that ZFP36L2 accelerates the degradation of specific ARE-transcripts upon binding, we obtained differential expression transcriptomic data on a Zfp36l2 knock-out mouse model to delve into the mechanisms governing this tissue-specific targeting. Transcriptomic analyzes of up-regulated ARE-transcripts in six tissues, lung, liver, bone marrow, spleen, kidney, and ovary of the Zfp36l2-deficient mouse confirmed that there is high tissue preference in ZFP36L2 targets. We observed only one common up-regulated gene, Apol11b, among these six different tissues. However, we do observe common trends, specifically an enrichment in protein coding genes in the up-regulated genes, consistent with these RBP primarily targeting genes on their 3’ UTRs. Interestingly, we observed a significant increase in the proportion of IgV (immunoglobulin) genes being up-regulated. We further performed eCLIP (Enhanced Cross-Linking&ImmunoPreciptation) on a mouse cell line to explore potential binding sites of ZFP36L2. AU-Rich Element score (AREscore) analysis revealed enrichment in both up-regulated genes and eCLIP peaks, although some differences were observed in flanking residue composition. Our findings provide new insights into the intricate regulatory network orchestrated by ZFP36L2, opening avenues for exploring its potential roles in different tissues. Background The Zinc Finger Protein 36 Like 2 (ZFP36L2) belongs to the class of adenine-uridine-rich element (ARE)-RNA binding proteins (ARE-RBPs). ARE-RBPs either stabilize or promote degradation of their target transcripts [1–4]. AREs comprise a variety of distinct adenine- and uridine-rich nucleotide sequences, typically located in the 3’ UTRs of mRNAs that have relatively short half-lives [5,6]. The AUUUA pentamer is considered the minimum core motif, and AREs are subdivided in several classes [7,8] based on pentamer number and arrangement. Although investigators have extensively identified specific binding sequences for RBPs, the consequences of the RNA/protein interactions are more difficult to assess. The tristetraprolin (TTP) family [9], or the zinc finger protein 36 (ZFP36) family, comprises a group of RBPs that interact directly with ARE-containing transcripts. Once the protein/RNA complex is formed, the most evident effect is to accelerate mRNA degradation by recruiting the CCR4/CAF1 deadenylases [10] and PABPC1 [11] in addition to decapping enzyme [12] and exosome. Most studies of the ARE specificity of the ZFP36 family are focused on the prototypical family member, ZFP36 [12–14]; comparatively little is known about the specificity of the other two family members, ZFP36L1 and ZFP36L2. In the study described here, we focused on the specificity of ZFP36L2. The ubiquitous expression of Zfp36l2 mRNA in tissues/organs (GTEx RNA-seq data) suggests a range of yet to be evaluated functions of Zfp36l2. Particularly, the functions of ZFP36L2 in the lungs, liver, and kidney have not been adequately assessed, despite the association of many RBPs with diseases of these organs, such as ischaemic lung injury [15], chronic metabolic and inflammatory fatty liver diseases and cancer [16] and diabetic kidney [17]. Mouse models provided the first evidence for the importance of ZFP36L2. Disruption of the first of the two Zfp36l2 exons resulted in viable animals. However, the females were infertile because of arrest of early embryo development at the 2-cell stage [18]. Interestingly, complete knockout of Zfp36l2 resulted in premature death of homozygous pups by the end of the second week of life [19,20], likely due to anaemia consequent to haematopoietic stem cell defects. These initial functional role for ZFP36L2 in mouse, has also been confirmed in humans. More recently, Zheng et al. reported that expression of biallelic deleterious polymorphisms in the human Zfp36l2 gene also leads to female infertility [21], and Iwanaga et al. found that acute leukaemia was associated with mutations in Zfp36l2 [22]. While other ARE-RBPs, such as TIAR and HuR, have been implicated in the chemotherapy resistance of haematologic malignancies [23]. Zhang et al. and Chousal et al. identified ZFP36L2-regulated mRNAs in erythrocyte lineage cells using an ex vivo knockdown [24] and an oocyte knockout model [25], respectively. Surprisingly, when we compared our splenic up-regulated gene dataset from Zfp36l2-fKO mouse [19] with these two datasets of genes up-regulated in the absence of ZFP36L2 we identified only one up-regulated genes in common to all three tissues. This single up-regulated gene overlap was in marked contrast to the approximate 7,000 expressed/detected genes in common among these three different tissues. Nevertheless, this result was obtained using three different biological models and distinct methodologies, microarray for the erythrocyte lineage, single cell RNA-seq (scRNA-seq) for oocytes and RNA-seq in spleen. These observations led us to hypothesize that ZFP36L2 has preferential mRNA targets depending on the tissue. If that is the case, ZFP36L2 would regulate distinct groups of genes depending on the tissue. To determine whether this apparent ZFP36L2 tissue preference was present in a range of tissues, here we used one biological model, our newly developed Zfp36l2 flox global knockout mouse model (L2-fKO) and the same methodology, RNA-seq, to obtain different tissue data sets. This mouse line, L2-fKO, lacking Zfp36l2 in all tissues, was created by crossing the CMV-Cre mouse (Jackson Labs) with a Zfp36l2-flox/flox mice line, where exon 2 of Zfp36l2 is flanked by flox sequences [19]. Six organs, lungs, liver, bone marrow, spleen, kidney and ovary were collected from experimental and control animal groups. RNA-seq data from all samples were analysed using a poly(A) library selection. The obtained results support our hypothesis that ZFP36L2 has distinct mRNA target preference depending on the tissue; only one up-regulated gene, Apol11b, was common to all tissues. Results Differential gene expression analysis in six mouse tissues To identify potential ZFP36L2 mRNA targets, we performed differential gene expression analysis of RNA-seq data collected from six different mouse tissues from WT and L2-fKO pups, carrying and lacking both Zfp36l2 alleles in all tissues, respectively, ages D6 to D9 post-birth. This developmental age was chosen, because these animals die during their second week of life as we reported previously due to haematopoietic defects [19,20]. We identified differentially expressed genes as having log2|Fold Change| >1 and padj <0.05. In Figure 1, in the left and right upper quadrants delineated by the red dotted lines in A to F, down-regulated genes are coloured in blue and up-regulated genes are in orange. Genes with a log2|Fold Change| <1 or padj >0.05 were classified as unchanged genes (grey). Data sets obtained for all tissues, satisfactorily clustered for the biological replicates in our RNA-seq data, based on the principal component analysis (PCA) and fold-change versus normalized mean counts (MA) plots as shown in Supplementary Figure S1A. To exclude that genes in the different organs are expressed in a sex-biased pattern in our samples, we performed PCA and MA analysis by sex in the liver (an organ where many genes are expressed in a sex-biased pattern) and observed no clustering by sex (Supplementary Figure S1B), suggesting no sex effect at that particular prepubertal age investigated. We classified coding and non-coding genes based on the protein coding annotations in Ensemble mouse (GRCm38). We chose the Ensemble annotations because they provide a coarse level of annotation, especially for non-coding genes. In the volcano plots and pie charts shown in Figure 1, we separated up-, down-, and unchanged genes into two sub-categories, coding (darker colours) and non-coding (lighter colours). Figure 1.Differential gene analysis of Zfp36l2 knockout mouse. We collected RNA sequencing data from six mouse tissues for both wild-type (WT) and Zfp36l2 knockout mice. Down-regulated genes are colored in blue, up-regulated in orange, and unchanged in gray. The inset pie charts divide the gene sets as coding and non-coding. Genes with a log2(Fold change) > |1| and padj < 0.05 were considered differentially regulated in (A) lung, (B) liver, (C) bone marrow, (D) spleen, (E) kidney and (F) ovary.The image shows volcano plots and pie charts for differential gene expression in six mouse tissues: A) lung, B) liver, C) bone marrow, D) spleen, E) kidney and F) ovary. Each plot displays log subscript 2 fold change on the x-axis and negative log subscript 10 p-adjusted on the y-axis. Red dotted lines indicate thresholds for significance. Pie charts categorize genes as up-regulated or down-regulated, coding or non-coding. The central diagram maps tissue locations: lung, liver, bone marrow, spleen, kidney and ovary. Legends explain plot and tissue color coding. Percentages in pie charts show gene distribution: coding and non-coding. The data highlights significant gene expression changes across tissues, with specific counts and percentages for each category.Volcano plots and pie charts illustrate the differential gene expression in six WT and Zfp36l2-KO mouse tissues. Differential gene analysis of Zfp36l2 knockout mouse. We collected RNA sequencing data from six mouse tissues for both wild-type (WT) and Zfp36l2 knockout mice. Down-regulated genes are colored in blue, up-regulated in orange, and unchanged in gray. The inset pie charts divide the gene sets as coding and non-coding. Genes with a log2(Fold change) > |1| and padj < 0.05 were considered differentially regulated in (A) lung, (B) liver, (C) bone marrow, (D) spleen, (E) kidney and (F) ovary. The fraction of coding vs. non-coding genes varied among tissues, as did the number of genes regulated. Ovary had the highest percent of non-coding regulated genes (Figure 1(F), 26.1% and 23.3% for down and up-regulated, respectively), whereas kidney (Figure 1(E), 10.5% and 5.8%) had the lowest. In general, for the down-regulated genes (blue), the percent of coding to non-coding was similar to the percent of coding to non-coding for unchanged genes (grey). In contrast, the percent of coding genes was more variable when we compared up-regulated coding genes (orange) vs unchanged genes. That is, in BM and kidney, the percent of coding genes (light orange) was greater compared with the unchanged genes, whereas the percent of coding genes was lower in the other tissues. Interestingly, the largest number of up-regulated coding genes was observed in bone marrow (1135) and spleen (430), which are tissues where haematopoiesis occurs at this development stage, D6 to D9 post-birth. Possibly, the pancytopenia phenotype of the knockout mouse is a result of these up-regulated mRNAs that are normally targets for ZFP36L2. Interestingly, among all six tissues we analysed, bone marrow and spleen had the highest expression levels of Zfp36l2, 7000 and 6000 TPMs in the WT mice, respectively (Supplementary Figure S2A). To test if the increased or decreased number of targets in each of these tissues was likely due to the absence of Zfp36l2 and not a potential compensatory effect of the other family members, we calculated the expression levels of Zfp36 and Zfp36l1 in all tissues analysed. In all the tissues, the other family members do not seem to compensate for the lack of Zfp36l2, except for the ovary where both Zfp36l1 and Zfp36 were slightly increased (Supplementary Figure S2B and C). Gene annotation of differentially regulated genes To identify trends common among the different tissues, we plot all genes (unchanged, down- and up-regulated) in a single volcano plot, where each gene was counted only once, in Figure 2(A). We observe that only in the up-regulated genes (light orange) there is a ~ 6.4% enrichment in protein coding genes relative to unchanged. Whereas we observed only a slight increase (1.5%) in coding genes that were down-regulated (dark blue). Thus, the absence of ZFP36L2 led to preferential upregulation of coding genes, suggesting that ZFP36L2 mainly downregulates coding sequence genes in the tissues studied. Figure 2.Differential gene analysis of Zfp36l2 knockout mouse across all tissues and cell cycle progression in splenocytes. We combined our analysis across all six mouse tissues for both wild-type (WT) and Zfp36l2 knockout animals. Down-regulated genes are colored in blue, up-regulated in orange, and unchanged in gray. The inset pie charts divide the genes by coding and non-coding. Genes with a log2(Fold change) > |1| and padj < 0.05 were considered differentially regulated. (A) Volcano plot of differential gene expression showing a general increase in the fraction of coding mRNAs in the up-regulated genes relative to non-coding and down-regulated gene sets, consistent with preferential targeting of 3’ UTRs by ZFP36L2. (B) Percent change in gene type annotations (as per RefSeq gene annotations) for up- and down-regulated genes in all six tissues. There was a 6.4% increase in the fraction of coding genes in the up-regulated set (relative to unchanged), whereas in the down-regulated set there was a 1.5% relative increase. Similarly, IgV genes (immunoglobulin variable regions) were overrepresented in the up-regulated gene set. Conversely, long non-coding RNAs were overrepresented in down-regulated genes but underrepresented in up-regulated genes. (C) RefSeq annotations include additional “other” categories, none of which are significantly over- or under-represented in the differential gene expression analysis. (D) WT and (E) KO representative samples of isolated splenocytes were assessed for EdU incorporation and DNA content by flow cytometry, after a 60 min exposure to EdU ex vivo. Both animals were of the same age, D8 post birth, and from the same litter. (F) Percent of cells in each cell cycle phase, G1, S and G2-M are illustrated. Samples were analyzed using FlowJo software. An ANOVA test was used to determine significance for multiple samples (*p = 0.01). (G) Comparison of cells from WT (n = 5) and L2-fKO (n = 3) animals in S-phase using a t test for comparing two samples (***p = 0.0006).The image consists of multiple panels analyzing gene expression and cell cycle in WT and L2-fKO splenocytes. A) A volcano plot shows differential gene expression with coding and non-coding genes categorized as down-regulated, unchanged, or up-regulated, when all tissues were combined. B) A bar graph displays percentage differences in gene types between up-regulated and down-regulated genes. C) A chart lists percentage differences in various gene categories. D) Flow cytometry plot for WT splenocytes shows DNA content and EdU incorporation with cell cycle phases G1, S and G2-M. E) Similar plot for L2-fKO splenocytes. F) Bar graph compares cell cycle phases between WT and L2-fKO, highlighting significant differences. G) A plot compares the percentage of cells in S-phase between WT and L2-fKO, with statistical significance indicated.Combined differential gene expression from all tissues and cell cycle analysis in WT and L2-fKO splenocytes. Differential gene analysis of Zfp36l2 knockout mouse across all tissues and cell cycle progression in splenocytes. We combined our analysis across all six mouse tissues for both wild-type (WT) and Zfp36l2 knockout animals. Down-regulated genes are colored in blue, up-regulated in orange, and unchanged in gray. The inset pie charts divide the genes by coding and non-coding. Genes with a log2(Fold change) > |1| and padj < 0.05 were considered differentially regulated. (A) Volcano plot of differential gene expression showing a general increase in the fraction of coding mRNAs in the up-regulated genes relative to non-coding and down-regulated gene sets, consistent with preferential targeting of 3’ UTRs by ZFP36L2. (B) Percent change in gene type annotations (as per RefSeq gene annotations) for up- and down-regulated genes in all six tissues. There was a 6.4% increase in the fraction of coding genes in the up-regulated set (relative to unchanged), whereas in the down-regulated set there was a 1.5% relative increase. Similarly, IgV genes (immunoglobulin variable regions) were overrepresented in the up-regulated gene set. Conversely, long non-coding RNAs were overrepresented in down-regulated genes but underrepresented in up-regulated genes. (C) RefSeq annotations include additional “other” categories, none of which are significantly over- or under-represented in the differential gene expression analysis. (D) WT and (E) KO representative samples of isolated splenocytes were assessed for EdU incorporation and DNA content by flow cytometry, after a 60 min exposure to EdU ex vivo. Both animals were of the same age, D8 post birth, and from the same litter. (F) Percent of cells in each cell cycle phase, G1, S and G2-M are illustrated. Samples were analyzed using FlowJo software. An ANOVA test was used to determine significance for multiple samples (*p = 0.01). (G) Comparison of cells from WT (n = 5) and L2-fKO (n = 3) animals in S-phase using a t test for comparing two samples (***p = 0.0006). To better understand whether there was a major change in any gene category of up- or down-regulated, we calculated the percent difference between up-regulated vs. unchanged genes (Figure 2(B), orange) and down-regulated vs unchanged genes (Figure 2(B), blue). As mentioned earlier, the Ensemble-specific annotation of mouse genes was used for this categorization. When we subdivided the genes, it became clear that most up-regulated genes were protein coding genes, i.e. mRNAs. Additionally, among the protein coding genes, the immunoglobulin variable (IgV) genes comprised a relatively high percent of up-regulated genes, 2.40%; thus, the IgV genes are represented in their own category (Figure 2(B)). The immunoglobulin variable genes encode the variable portion of the immunoglobulin chains, and they generate diversity and specificity of the immunoglobulins. Long non-coding RNAs (lncRNAs, containing or not containing introns) and processed pseudogenes lacking introns that are thought to arise from retrotransposition were underrepresented in the up-regulated category. ‘To be Experimentally Confirmed’ genes, i.e. genes with poly (A) features but which are not experimentally validated, were slightly underrepresented in the up-regulated group and overrepresented in the down-regulated group. All other gene categories had minor variations (Figure 2(C)). Because IgV genes are only expressed in B cells, and the spleen is the primary site for maturation of early B cell populations during D5-D9 post-birth, we decided to isolate splenocytes from WT and L2-fKO and assess their proliferation by measuring EdU incorporation. EdU is a thymidine analog that is incorporated into DNA by cells undergoing active DNA replication. In the spleen, the B cells are activated and undergo rapid proliferation along with differentiation into plasma cells and memory cells, that then exit the spleen. Note that B cells are one component of the heterogeneous cell types that compose the spleen. However, because we used polysucrose density gradient centrifugation to first remove most red cells, debris and other cell types, our ‘splenocytes’ were enriched for immune cells including monocytes and lymphocytes, in particular B cells. Surprisingly, the L2-fKO ‘splenocytes’ had nearly 5-fold reduction in S phase population (Figure 2(E and G)), with a correlating increase in G1/G0 (Figure 2(F)). The G1 phase is a critical stage for B cell differentiation because during this period, B cells receive signals that trigger their activation and subsequent differentiation into either memory B cells or plasma cells. Overlap of up-regulated genes and gene ontology analysis As mentioned, ZFP36L2 binds AU-rich elements in 3’ UTRs, and this interaction results in RNA degradation. Thus, the up-regulated genes in the knockout mice are of particular importance to our analysis. Our previous transcriptomic analysis raised the idea that ZFP36L2 display a distinct tissue targeting specificity [19]. As mentioned earlier, when we performed a similar analysis using previously published data [24,25] in early burst-forming unit-erythroid and oocyte, respectively, and spleen in our knockout, we found a single common gene that did not have an ARE [19]. However, this low overlap could be due to different biological models and use distinct methodologies, thus justifying the present investigation using the same biological model and methods. Here we perform an overlap analysis on our new experimentally obtained data sets to identify genes up-regulated in multiple tissues and test our idea that ZFP36L2 has preferential mRNA targets depending on the tissue. This analysis is illustrated in Figure 3(A) as an upset plot. Strikingly, only a single gene, apolipoprotein L 11b (Apol11b), was up-regulated in all six tissues (S1 in Figure 3(A and E)). To answer if the up-regulated genes were equally detected among all tissues, we analysed the expression of all up-regulated genes in each specific tissue by obtaining the mean TPM values from WT animals (Supplementary Table 1). The up-regulated genes are detected in each tissue at similarly high percentage rates, varying from 74% to 83% (Supplementary Table S2). This suggests that up-regulated genes in each tissue are not uniquely expressed in each tissue, but rather up-regulated in specific tissues. Additionally, only few genes that were uniquely up-regulated (Figure 3(A) on the right) are present solely in that tissue (Suplementary Table S3), except for ovary. Thus, most genes that were uniquely up-regulated in one tissue are expressed in all tissues but only up-regulated in that one. Figure 3.Overlap and Gene Ontology analysis of up-regulated genes in Zfp36l2 knockout tissues. (A) Upset plot of up-regulated genes in six tissues. Only one gene, Apol11b, was up-regulated in all six tissues (subset S1). The genes simultaneously up-regulated in four or more tissues are labeled S1-S8, and their respective names are shown in E. Three up-regulated genes (subsets S2, S3) were common to five tissues, and 13 genes were upregulated in common in four tissues (subsets S4–S8). However, most of the up-regulated genes were upregulated in only one tissue, indicating an extensive tissue-selectivity targeting of ZFP36L2. (B) Gene Ontology analysis of up-regulated genes in three or more tissues. (C) Gene Ontology analysis of up-regulated genes in two tissues. (D) Gene Ontology analysis of up-regulated genes in bone marrow and spleen. The overrepresented terms in B, C, and D are mostly related to immune response. (E) ribbon diagrams of GO terms for up-regulated genes in four or more tissues, organized by common sets. (F) upset plot of unchanged genes in six tissues and their tissue distribution, confirming that most of the genes were detectable in the RNA-seq of all six tissues.The infographic illustrates gene expression analysis across various tissues, focusing on up-regulated genes. It includes several sections: A) An upset plot showing up-regulated genes in six tissues, highlighting subsets S1 to S8. Only one gene is consistently upregulated across all tissues. B) Gene ontology analysis for genes up-regulated in three or more tissues reveled genes related to immune response. C) Gene ontology analysis for genes up-regulated in two tissues, also resulted in immune-related terms. D) Analysis for bone marrow and spleen, share genes involved in immune response. E) Ribbon diagrams display GO terms for genes up-regulated in four or more tissues, organized by common sets (S1 to S8). F) An upset plot of unchanged genes across six tissues, illustrating their common detectability in all six tissues. The analysis indicates tissue-selectivity targeting of ZFP36L2, with Apol11b being the sole gene up-regulated in all tissues. The Gene Ontology terms are predominantly relate to immune response, highlighting that immune-related processes are common to multiple tissues.Infographic showing gene expression analysis in tissues with upset plots and gene ontology insights. Overlap and Gene Ontology analysis of up-regulated genes in Zfp36l2 knockout tissues. (A) Upset plot of up-regulated genes in six tissues. Only one gene, Apol11b, was up-regulated in all six tissues (subset S1). The genes simultaneously up-regulated in four or more tissues are labeled S1-S8, and their respective names are shown in E. Three up-regulated genes (subsets S2, S3) were common to five tissues, and 13 genes were upregulated in common in four tissues (subsets S4–S8). However, most of the up-regulated genes were upregulated in only one tissue, indicating an extensive tissue-selectivity targeting of ZFP36L2. (B) Gene Ontology analysis of up-regulated genes in three or more tissues. (C) Gene Ontology analysis of up-regulated genes in two tissues. (D) Gene Ontology analysis of up-regulated genes in bone marrow and spleen. The overrepresented terms in B, C, and D are mostly related to immune response. (E) ribbon diagrams of GO terms for up-regulated genes in four or more tissues, organized by common sets. (F) upset plot of unchanged genes in six tissues and their tissue distribution, confirming that most of the genes were detectable in the RNA-seq of all six tissues. Only 17 genes (listed in Figure 3(E)) were up-regulated in four or more tissues (subsets S1-S8 in Figure 3(A)) of a total of 10,887 genes present in common, less than 0.1%. Notably, the three tissues with the highest number of up-regulated genes in common are liver, spleen and bone marrow; tissues where ZFP36L2 has a known role in vivo, in immuno-haematopoiesis [19,20,24,26,27]. The largest number of shared up-regulated genes, 184, was in spleen and bone marrow. Again, this finding likely reflected the fact that both tissues have haematopoietic functions at this developmental age, and that the KO phenotype, pancytopenia, revels potential ZFP36L2 targets in those organs. To identify potential biological processes, cellular components, or molecular functions significantly overrepresented, we performed a Gene Ontology enrichment analysis of up-regulated genes present in more than three tissues (Figure 3(B)), two tissues, (Figure 3(C)), and only common to spleen and bone marrow (Figure 3(D)). This analysis identified some common terms, such as ‘response to Type II interferon’ and ‘cell killing’ (Figure 3(B)) and ‘lymphocyte mediated immunity’ (Figures 3(C and D)). These terms likely reflected the haematologic defect of pancytopenia observed in the L2-fKO, as well as cellular stress responses resultant from the phenotype. In addition, ‘immune system process’ and cellular response, such as ‘signaling receptor activity’ and ‘response to stimulus’ were overrepresented (darker blue) in most of the 17 genes up-regulated in four or more tissues (Figure 3(E)). Similarly, for genes up-regulated in two tissues (Figure 3(C)), the enriched GO terms were also related to immune functions, so that they provided little clarification of specific pathways or molecular processes modulated by Zfp36l2. To provide a more robust way to compare independently derived gene expression data sets, we also performed GeneOverlap [28] analysis (Supplementary Figure S4A) followed by a Gene Set Enrichment Analysis (GSEA) to identify potential common pathways modulated by the lack of Zfp36l2. The GSEA for each tissue in a rank-ordered list is summarized in an upset plot (Supplementary Figure S4B) and listed as a heatmap of hallmark pathways (Supplementary Figure S4C). No single pathway was significantly enriched in all six tissues. However, inflammatory pathways such as TNF-α and IL6 were significantly enriched in 5 out of 6 tissues (Supplementary Figure S4B and C, coloured in green), although in bone marrow they were negatively enriched. Additionally, five other inflammatory pathways (inflammatory response, IL2-Stat5 signalling, allograft rejection, IFN-α and IFN-γ responses, Supplementary Figure S4B and S4C, coloured in purple) were significantly enriched in liver, kidney, lung and ovary, but not in spleen and bone marrow. Overall, liver had the highest number of significantly enriched pathways (20, with Z-score >1), followed by kidney [16], lung [15], ovary [12], bone marrow [9], and spleen [1]. Likely, the predominance of inflammatory pathways observed in the L2-fKO mouse is a consequence of the decreased number of all white blood cells. These findings are consistent with our GO analysis where ‘immune system process’ and cellular response, such as ‘signaling receptor activity’ and ‘response to stimulus’ were overrepresented (Figure 3(E)). Finally, other significantly enriched pathways were observed only in some specific tissues, for example ‘epithelial mesenchymal transition’ in liver and bone marrow, as well as ‘mTORC1 signaling’ strongly represented in spleen. It is important to compare the up-regulated upset plot in Figure 3(A) with the same analysis for the unchanged genes shown in Figure 3(F). Approximately 70% of all unchanged genes (10,887/~15,000genes Figure 3(F)) were expressed in all six tissues. In contrast 0.04% (1/2,583) of up-regulated genes were found in all tissues. Thus, we would expect far greater overlap in the differentially regulated genes than what was observed in Figure 3(A). We also performed a similar analysis of down-regulated genes (Supplementary Figure S3). Zfp36l2 was found among the down-regulated genes in all six tissues, confirming the lack of this gene in the L2-fKO mouse model as expected. The contrasting overlap analysis of up-regulated genes (Figure 3(A)) versus unchanged genes (Figure 3(F)) suggested that the genes uniquely up-regulated in each tissue would be genes most likely to reveal specific organ functions regulated by ZFP36L2. Thus, we performed GO enrichment analysis of up-regulated genes in each tissue (Figure 4). The GO analysis in each tissue revealed functions and biological processes related to the physiology of the tissue itself. This result was not necessarily surprising because each gene set was uniquely up-regulated in an individual tissue. For example, for the liver (Figure 4(B)), after the inflammatory genes common to all tissues, we observed that there was a category of genes affected by a lack of Zfp36L2 which were related to the catabolic and metabolic function of this organ, such as ‘NADH regeneration’ and ‘glucose catabolic process to pyruvate’. In fact, nearly the entire glycolytic pathway from glucose-6-phosphate to pyruvate was de-repressed in the knockout (Gpi1, Pfkl, Pfkp, Aldoa, Tpi1, Gapdh, Pgk1, Pgam1, Eno1 and Pkm; log2 fold change +1.1 to +2.8, all padj <0.05), accompanied by the lactate exporter Slc16a3 (MCT4) and the hypoxia regulators Egln3 and Ddit4, suggesting that ZFP36L2 normally restrains a glycolytic, lactate-exporting program in the liver. Whereas in the bone marrow (Figure 4(C), Table 1), the up-regulated genes were related to the bone marrow environmental niche and the extracellular matrix that are required for proper differentiation of stem cells into each differential blood cell type. Figure 4.Gene Ontology analysis of up-regulated genes in each tissue. We performed Gene Ontology enrichment analysis on genes uniquely up-regulated in each tissue to identify enriched terms specific to each tissue. We selected terms with padj < 0.05 and gene enrichment ratio > 0.01. The GO terms identified for A) lung, B) liver, C) bone marrow, D) spleen, E) kidney and F) ovary.The image presents Gene Ontology analysis of genes up-regulated in only one mouse tissue, featuring a central mouse diagram labeled with organs: lung, liver, bone marrow, spleen, kidney and ovary. Each organ connects to a graph showing gene enrichment data. A) Lung: highlights response to type II interferon, tolerance induction and cholesterol homeostasis. B) Liver: includes NADH regeneration, canonical glycolysis and glucose catabolism to pyruvate. C) Bone Marrow: lists collagen-containing extracellular matrix, Wnt signaling pathway and UV-A response. D) Spleen: features antigen processing, MHC protein complex and oxidoreductase activity. E) Kidney: emphasizes positive regulation of GTPase activity and triglyceride lipoprotein particles. F) Ovary: shows cytokine-mediated signaling, neutrophil chemotaxis and chemokine activity. Each graph details gene ratio, adjusted p-value and count.Gene Ontology analysis of genes up-regulated in only one mouse tissue: lung, liver, bone marrow, spleen, kidney, ovary.Table 1.Gene ontology enriched terms, once immune terms common to multiple tissues are excluded.TissueDistinctive (non-shared) gene program — representative terms and genesLiverGlycolysis / NADH regeneration / glucose-to-pyruvate (Gpi1, Pfkl, Pfkp, Aldoa, Tpi1, Gapdh, Pgk1, Pgam1, Eno1, Pkm)Bone marrowBone, cartilage and connective-tissue development; biomineralization (Col1a1, Col2a1, Bmp7, Sp7, Mmp13, Fgfr2)KidneyXenobiotic glucuronidation: glucuronosyl-/UDP-glycosyltransferase activity (Ugt1a9, Ugt1a10, Ugt3a1, Ugt3a2)LungTriglyceride / lipid metabolism, lipase inhibition (Apoc3, Angptl3, Apoa4, Sik1)OvaryMeiosis / germ cell: synaptonemal complex, germ-cell nucleus (Sycp3, Fkbp6, Morc1, Tbpl2, Lhx8)SpleenVascular (VEGF-response) and ion-channel regulation (Pik3cb, Emilin1, Kcnq1, Kcne3/Kcne4) Gene Ontology analysis of up-regulated genes in each tissue. We performed Gene Ontology enrichment analysis on genes uniquely up-regulated in each tissue to identify enriched terms specific to each tissue. We selected terms with padj < 0.05 and gene enrichment ratio > 0.01. The GO terms identified for A) lung, B) liver, C) bone marrow, D) spleen, E) kidney and F) ovary. Table 1.Gene ontology enriched terms, once immune terms common to multiple tissues are excluded.TissueDistinctive (non-shared) gene program — representative terms and genesLiverGlycolysis / NADH regeneration / glucose-to-pyruvate (Gpi1, Pfkl, Pfkp, Aldoa, Tpi1, Gapdh, Pgk1, Pgam1, Eno1, Pkm)Bone marrowBone, cartilage and connective-tissue development; biomineralization (Col1a1, Col2a1, Bmp7, Sp7, Mmp13, Fgfr2)KidneyXenobiotic glucuronidation: glucuronosyl-/UDP-glycosyltransferase activity (Ugt1a9, Ugt1a10, Ugt3a1, Ugt3a2)LungTriglyceride / lipid metabolism, lipase inhibition (Apoc3, Angptl3, Apoa4, Sik1)OvaryMeiosis / germ cell: synaptonemal complex, germ-cell nucleus (Sycp3, Fkbp6, Morc1, Tbpl2, Lhx8)SpleenVascular (VEGF-response) and ion-channel regulation (Pik3cb, Emilin1, Kcnq1, Kcne3/Kcne4) Gene ontology enriched terms, once immune terms common to multiple tissues are excluded. Overall, some trends emerged from the GO analysis. Most of the common terms reflected genes related to immune system function, such as response to type II interferon (lungs), regulation of T cell activation (liver), collagen-containing extracellular matrix (bone marrow), immunoglobulin-mediated immune response (spleen), antigen processing and presentation of exogenous peptide antigen (kidney), and cytokine-mediated signalling (ovary). To distinguish the organ-specific signal from the immune terms common to these largely immune-related tissues, we separated GO terms that recurred across tissues from those enriched in a single tissue (Table 1). The shared terms were dominated by generic immune and chemotaxis processes, whereas the tissue-specific terms reflected the characteristic physiology of each organ, including glycolysis in liver, skeletal and connective-tissue development in bone marrow, xenobiotic glucuronidation in kidney, triglyceride and lipid handling in lung, and meiotic germ-cell programs in ovary. Therefore, the overall analysis suggested that ZFP36L2 likely regulates not only different immune pathways in general but also genes related to the physiology of each organ. The only gene up-regulated in all six tissues was Apol11b, thus we further test if there was a direct interaction of ZFP36L2 protein and this transcript. This gene is predicted to enable chloride channel action and lipid binding activity, but the specific function remains to be determined. In all six L2-fKO tissues, Apol11b had higher RNA-seq read depth in the knockout (Figure 5(A)). In most WT tissues, Apol11b was expressed at low levels, except for spleen and bone marrow (Figure 5(B)). Apol11b has only one ARE, a 7-mer type (UAUUUAU), in its 3’ UTR, 73 nucleotides downstream of the stop codon. To determine whether ZFP36L2 directly binds to Apol11b mRNA, we synthesized an RNA probe that included this ARE and performed a gel shift assay. ZFP36L2 clearly bound to Apol11b (Figure 5(C)). Another family member, ZFP36, apparently did not bind, whereas ZFP36L1 bound weakly (Figure 5(D)). Immunoblotting showed that all ZFP36 family members were present at comparable levels in the protein extracts used for the gel shift assay (Figure 5(E)). These results suggested that Apol11b is a specific mRNA target of ZFP36L2, although ZFP36L1 had a lower affinity for this transcript. In the present in vivo model, Zfp36l1 expression is not increased in any KO tissue, except for a modest increase in the ovary (Supplementary Figure S2B), thus ZFP36L1 is unable to compensate for the effect of ZFP36L2 in controlling Apol11b in vivo. Interestingly, an RNAfold analysis of the 29-nt Apol11b probe predicted that the ARE was within a loop (Figure 5(F)) resembling the structure we obtained for the functional ARE in the LHR transcript using SHAPE-Map [29]. Figure 5.Apol11b, the only gene up-regulated in all six tested tissues. (A) Upper panel shows the four transcript variants of Apol11b. Lower panel shows coverage plots for Apol11b from RNA-seq results. RNA-seq coverage for WT is shown in blue, and KO is shown in red; black dashed bars indicate the location of the 7-mer ARE in the 3’ UTR. The pattern of read depth coverage is similar in all tissues, suggesting no change in relative transcript isoform abundance, but instead an overall change in gene expression. (B) Expression of Apol11b across WT tissues based on transcript per million (TPM) obtained by RNA-seq. (C) RNA electrophoretic mobility shift assays were performed by incubating 10 μg of protein extract from HEK 293 cells transfected with ZFP36L2-C176S (lane 2) or ZFP36L2 (lanes 3) constructs with Apol11b probe. Lane 2 corresponds to a mutation (mC176S) in the first mZFD that abrogates binding. (D) EMSA performed by incubating 10 μg of protein extract from HEK 293 cells transfected with GFP (lane 2) or with vectors expressing the GFP-ZFP36 proteins with Apol11b probe. The sequence of the Apol11b mouse probe, 29 nt, is shown above the EMSA. (E) Immunoblotting of protein extracts used in the EMSA in panel D were probed with GFP-antibody. Five µg of protein extracts were loaded per lane. (F) Secondary structure prediction for the Apol11b probe using RNAfold web server. The color-code scale shown at the bottom is for the prediction confidence for each base pair in the structural model.Panel A shows Apol11b transcript variants derived from chromosome 15 with RNA-seq coverage for WT and KO tissues. The coverage plots indicate expression differences. Panel B presents a box plot of Apol11b expression in tissues, with higher TPM in spleen and bone marrow. Panel C displays RNA electrophoretic mobility shift assays with ZFP36L2-C176S and ZFP36L2, showing binding only in lane 3, but not in 2. Panel D shows EMSA with GFP and GFP-ZFP36 proteins, indicating binding in lanes 3 and 4, corresponding to ZFP26L1 and ZFP36L2, respectively. Panel E provides immunoblotting results of protein extracts used in EMSA, probed with GFP-antibody. Panel F showns predicted secondary structure of the Apol11b probe using RNAfold, with a confidence scale for base pair prediction. The probe sequence used is 5’AAUGUCAAAAUUUUAUUUUCUUAGAAG3’. Panels are arranged top-to-bottom, left-to-right, highlighting Apol11b expression and ZFP36L2 interaction.Infographic on Apol11b expression, binding to ZFP36L2 using EMSA and structure prediction. Apol11b, the only gene up-regulated in all six tested tissues. (A) Upper panel shows the four transcript variants of Apol11b. Lower panel shows coverage plots for Apol11b from RNA-seq results. RNA-seq coverage for WT is shown in blue, and KO is shown in red; black dashed bars indicate the location of the 7-mer ARE in the 3’ UTR. The pattern of read depth coverage is similar in all tissues, suggesting no change in relative transcript isoform abundance, but instead an overall change in gene expression. (B) Expression of Apol11b across WT tissues based on transcript per million (TPM) obtained by RNA-seq. (C) RNA electrophoretic mobility shift assays were performed by incubating 10 μg of protein extract from HEK 293 cells transfected with ZFP36L2-C176S (lane 2) or ZFP36L2 (lanes 3) constructs with Apol11b probe. Lane 2 corresponds to a mutation (mC176S) in the first mZFD that abrogates binding. (D) EMSA performed by incubating 10 μg of protein extract from HEK 293 cells transfected with GFP (lane 2) or with vectors expressing the GFP-ZFP36 proteins with Apol11b probe. The sequence of the Apol11b mouse probe, 29 nt, is shown above the EMSA. (E) Immunoblotting of protein extracts used in the EMSA in panel D were probed with GFP-antibody. Five µg of protein extracts were loaded per lane. (F) Secondary structure prediction for the Apol11b probe using RNAfold web server. The color-code scale shown at the bottom is for the prediction confidence for each base pair in the structural model. AU-rich element analysis of regulated genes Because ZFP36L2 binds to adenine-uridine-rich elements (ARE) within the 3 ’UTRs of mRNAs, we used the AREScore algorithm to analyse the sequences of the 3’ UTRs of up- and down-regulated genes. This algorithm is based on quantifying three typical features of AREs: the number of ‘AUUUA’ pentamer motifs, the proximity between the pentamers, and the presence of AU content surrounding the pentamer [30]. A high AREScore indicates a high likelihood that the 3’ UTR is a target of the ZFP36 family of proteins. When we compared the AREScores from the up-regulated genes (Figure 6(A), orange, mean 3.75) with the down-regulated genes (Figure 6(A), blue, mean 2.82) from all six tissues combined, the mean score was clearly greater for the up-regulated set (****p = 1.48 × 10−12). Thus, the 3’ UTRs of up-regulated genes preferentially contain ARE sequences likely to bind ZFP36L2. Additionally, when we performed a similar analysis for each tissue individually (Figure 6(B)), the AREScores for up-regulated genes were also significantly greater for each tissue, except for lung and ovary. Note that the largest difference in AREScore was for bone marrow (p = 1.55 × 10−5) and liver (p = 3.58 × 10−5). AREScore did not correlate with the number of tissues in which a gene was up-regulated (Spearman ρ = −0.03, p = 0.17), consistent with the predominantly tissue-restricted response in which 82% of up-regulated genes were uniquely up-regulated in a single tissue. Notably, tissue-specific up-regulated genes carried AREScores comparable to broadly-regulated genes and significantly higher than downregulated genes (mean 3.97 vs 2.96, p = 3 × 10− 9), indicating that the AREScore enrichment is a pervasive feature of the up-regulated population rather than a property of a few consistently regulated genes. In contrast, the AREScores in up-regulated non-coding genes were not increased in comparison to the down-regulated genes (Supplementary Figure S5). In fact, the mean of AREscores in non-coding up-regulated genes was significantly lower than AREscores in down-regulated non-coding genes. Most of the long non-coding genes are nuclear [31], whereas ZFP36L2 is mainly cytosolic with shuttling properties [32]. These findings, support our results presented in Figure 2(A), suggesting that ZFP36L2 does not seem to modulate non-coding genes in our mouse model. Figure 6.ARE analysis of 3’ UTRs in differentially regulated genes. (A) AREScore analysis of all coding gene 3’ UTRs for up- (orange) and down- (blue) regulated genes. The mean AREScore of up-regulated genes was greater compared with down-regulated genes (3.75 vs 2.82, p = 1.48 × 10−12, using a traditional two-tailed t-test). (B) A greater AREScore was also observed in up-regulated genes in most individual tissue, consistent with the fact that ZFP36L2 preferentially binds AREs. (C) Motif analysis of the sequences flanking the AREs in the 3’ UTRs of up-regulated genes. In each individual tissue, there was a common flanking U predominance.Image A displays a violin and box plot of ARE scores for down-regulated and up-regulated coding genes, with means of 2.82 and 3.75, respectively. The y-axis ranges from 0 to 50. A bracket indicates p = 1.48 x 10^-12 with four asterisks. Image B shows paired violin and box plots by tissue: kidney, BM, lung, spleen, liver, ovary, each with down- and up-regulated genes. p-values are: kidney p = 1.78 x 10^-2*, BM p = 1.55 x 10^-5****, lung p = 4.38 x 10^-1, spleen p = 9.93 x 10^-4**, liver p = 3.58 x 10^-5****, Ovary p = 1.11 x 10^-1. Image C features sequence logo plots for BM, kidney, liver, lung, ovary, spleen. The y-axis isinformation (bits) from 0.000 to 0.200, x-axis positions 0 to 24. A highlighted band spans positions 10 to 14 with AUUUA above it. Letter stacks show U, A, G, C heights, with U predominance near the highlighted region. Small tissue icons accompany each logo.Plots compare ARE score in down- and up-regulated genes, blue and orange, respectively and show flanking sequence motifs across tissues. ARE analysis of 3’ UTRs in differentially regulated genes. (A) AREScore analysis of all coding gene 3’ UTRs for up- (orange) and down- (blue) regulated genes. The mean AREScore of up-regulated genes was greater compared with down-regulated genes (3.75 vs 2.82, p = 1.48 × 10−12, using a traditional two-tailed t-test). (B) A greater AREScore was also observed in up-regulated genes in most individual tissue, consistent with the fact that ZFP36L2 preferentially binds AREs. (C) Motif analysis of the sequences flanking the AREs in the 3’ UTRs of up-regulated genes. In each individual tissue, there was a common flanking U predominance. We performed position weight matrix (PWM) analysis of the flanking sequences of 5-mer AREs in the 3’ UTRs of up-regulated genes for each tissue individually (Figure 6(C)). Uridine was the flanking nucleotide with the highest probability adjacent to the core 5-mer AUUUA motif in all tissues. This result was consistent with our previous work that showed the minimum binding requirement of ZFP36L2 to be a 7-mer UAUUUAU motif [19]. Although some differences in ARE flanking PWMs were observed across tissues, the differences were small. Interaction of ZFP36L2 with the transcriptome The transcriptome includes all RNA transcripts expressed in each cell type at a specific time or in a particular condition. Based on the results in Figure 2(B), we concluded that ZFP36L2 tends to target mRNAs that have protein coding sequences, i.e. open reading frames. To determine whether ZFP36L2 tends to regulate mRNAs high or low expressed in each tissue, we obtained the expression values in transcripts per million (TPMs) for all the genes captured by our RNA-seq. The up-regulated genes had a 6-fold decrease in expression levels in comparison to down-regulated genes (TPM ~100 vs TPM ~600, respectively, ****p < 0.00005), Supplementary Figure S6A, suggesting that ZFP36L2 preferentially targets mRNAs that are expressed at low levels in mouse tissues. To explore mRNA sequences directly bound to ZFP36L2, we performed duplicate enhanced UV crosslinking followed by immunoprecipitation experiments with an anti-ZFP36L2 antibody (eCLIP); a rabbit IgG was used as negative control. Other investigators have performed eCLIP experiments with ZFP36L2, however those studies were conducted only in human cell lines [27] and never using murine lines. Because the 3’ UTRs are not well conserved, in contrast to 5’ UTRs that tend to be conserved in evolution, we deemed it necessary to collect murine-specific binding data to test how they overlapped with our up-regulated data set. Initially, we attempted these experiments in primary bone marrow cells from WT pups at D7 to D9 post-birth because that period precedes the initial phenotypic signs observed in the L2-fKO animals. However, these samples did not pass IP quality controls, and we had to resort to an established mouse culture cell line. To explore that we used MLTC-1 cells because, previously, we detected endogenous ZFP36L2 by immunoblotting in these cells [29,33]. MLTC-1 is a Leydig cell line derived from the mouse testis, thus raising the question if this would be a limitation of performing eCLIP in these cells to compare with our RNA-seq data. To address that we checked how many genes are expressed in common between our RNA-seq from all six tissues and a recent RNA-seq from MLTC-1 cells [34]. Most of the genes, 68%, captured in our RNA-seq from mouse tissue overlap with genes expressed in MLTC-1 cells (Supplementary Figure S7), whereas only 13% are expressed exclusively in MLTC-1 cells. We identified 2,143 eCLIP sequences with log2(Fold Change) >1 and log10(p) <1. These peaks mapped to 597 different genes. To explore how these findings could be interpreted in the context of our mouse model, we intersected up-regulated genes in each tissue with these 597 eCLIP genes (Figure 7(A)). We noticed that the highest intersection occurred with bone marrow, i.e. 34 genes. Liver was second with 10 genes, and spleen was third with 7 genes. Notably, these three organs are important sites of haematopoiesis during the early development stage of D6 to D9 post-birth, the period when we collected samples for RNA-seq. Of note, overall, eCLIP peaks captured by the specific antibody were derived from genes expressed at a higher TPM (>3,000) (Supplementary Figure S6B). Figure 7.Intersection of up-regulated genes and eCLIP peaks. (A) Upset plot of genes in which at least one ZFP36L2 eCLIP peak present in both eCLIP duplicate samples from MLTC-1 cells was present in the 3’ UTR and in the set of up-regulated genes from each tissue. (B) Genomic track of 3’ UTR of Serpinh1, a gene up-regulated in bone marrow. RNA-seq results show higher coverage plot in the KO (red) versus the WT condition (blue). The lower panel shows raw ZFP36L2 IP in green compared with the IgG control in purple. The reproducible eCLIP peaks (across two conditions) are shown as yellow rectangles. One 7-mer ARE is present at the 5’ end of the first eCLIP peak.Image A shows an upset plot with 13 gene intersection columns and ′Intersection size′ from 0 to 550. Bar values are 529, 34, 10, 7, 6, 3, 2, 1, 1, 1, 1, 1, 1. Sets include up-regulated genes in kidney, ovary, lung, liver, spleen, bone marrow tissues and simultaneouly captures by eCLIP peaks. Image B depicts genomic coverage and immunoprecipitation tracks over Serpinh1 3′ UTR, with Chromosome 7 coordinates from 99,345,400 to 99,346,200. ′KO coverage′ and ′WT coverage′ are expressed in the y-axis labeled ′Read Depth′ from 0 to 3.0 x 10^4. A gene model track shows Serpinh1 3′ UTR. Below, a sequence logo reads ′UUC AUA UUU AUG GCC.captured by ZFP36L2 IP′ but not by ′IgG IP′ eCLIP track.‘The upset plot shows gene set intersections of up-regulated genes and eCLIP peaks and Serpinh1 3’ untranslated region coverage and eCLIP peaks. Intersection of up-regulated genes and eCLIP peaks. (A) Upset plot of genes in which at least one ZFP36L2 eCLIP peak present in both eCLIP duplicate samples from MLTC-1 cells was present in the 3’ UTR and in the set of up-regulated genes from each tissue. (B) Genomic track of 3’ UTR of Serpinh1, a gene up-regulated in bone marrow. RNA-seq results show higher coverage plot in the KO (red) versus the WT condition (blue). The lower panel shows raw ZFP36L2 IP in green compared with the IgG control in purple. The reproducible eCLIP peaks (across two conditions) are shown as yellow rectangles. One 7-mer ARE is present at the 5’ end of the first eCLIP peak. To exemplify how these intersections of up-regulated and eCLIP sequences looked in our raw data, we chose one representative gene, Serpinh1, up-regulated in bone marrow (Figure 7(B)). The upper panel of Figure 7(B) illustrates the 3’ UTR coverage obtained by RNA-seq for the L2-fKO condition (red) compared with WT (blue). There was significantly greater read depth of the 3’ UTR in the KO, consequently Serpinh1 was classified as an up-regulated gene in the bone marrow. In the lower panel, we plotted the window of one reproducible eCLIP peak (present in both IP data sets, yellow) found in Serpinh1 (green). Interestingly, the black dashed bar corresponds one UAUUUAU motif. The lower panel of Figure 7(B) shows evidence for strong enrichment of the ZFP36L2 IP track (green) when we compared the data with the IgG IP control (purple), resulting in the peak call. In Supplementary Figure S8 we show coverage tracks for four additional representative bone marrow genes, Col4a1, Fstl1, Lox, and Lpl. We chose these four genes because they had at least one ARE of the 5-mer type (AUUUA) in 3’ UTR. A flanking motif analysis around the AREs in eCLIP peaks showed an enrichment for uridines surrounding the AREs (Supplementary Figure S8E). Whereas a de novo motif analysis using HOMER found three of the six first ranked motifs, as being U enriched motifs. This U enrichment is particularly visible in motif #5 (Supplementary Figure S9). Expression of the mRNA-degradation machinery in the Zfp36l2 knockout Because ZFP36L2 acts as a sequence-specific adaptor that recruits the general mRNA-decay machinery to ARE-containing transcripts rather than degrading RNA itself, the up-regulation of target mRNAs in the knockout could in principle arise from changes in the levels of the degradation enzymes. To test this, we examined the differential expression of 115 genes spanning the mRNA-degradation machinery across all six tissues (Supplementary Figure S10): the catalytic and structural core of decay (Supplementary Figure S10A), its activators and adaptors (Supplementary Figure S10B), and the sequence-specific RNA-binding proteins that recruit the machinery to specific mRNAs (Supplementary Figure S10C). The catalytic machinery was essentially unchanged in the knockout: across the 71 detected core genes the changes were small (median per-gene maximal |log2FC| ≈ 0.34), and only Noct and Patl2 reached significance with |log2FC| ≥1, each in a single tissue and in opposite directions, with no coordinated regulation of any decay module. By contrast, and as expected, Zfp36l2 was the most strongly and consistently down-regulated gene of the entire set (log2FC −1.85 to −4.09; all padj <0.05), confirming the knockout, and the few additional changes were largely restricted to the recruiter layer (for example, the paralog Zfp36 in bone marrow). These results indicate that de-repression of ARE-containing targets in the Zfp36l2 knockout reflects loss of ZFP36L2-mediated recruitment of an intact degradation machinery rather than a change in the abundance of the degradation enzymes. Discussion RNA-binding proteins (RBPs) are associated with liver, kidney, and lung diseases. In the liver, 92 RBPs are up-regulated in hepatocellular carcinoma [35]. Members of a prominent class of RBPs, to which ZFP36L2 belongs, the adenine-uridine-rich element (ARE)-binding proteins (ARE-RBPs), are involved in chronic metabolic and inflammatory fatty liver diseases and cancer [16,36]. Other ARE-RBPs are associated with hypoxia-response signalling in kidney epithelial cells [37], diabetic kidney disease [17], and protection from ischaemic lung injury [15]. Despite the associations of ARE-RBPs with liver, kidney, and lung disorders, the mechanism(s) by which these RBPs contribute to disease are unclear. To determine whether ZFP36L2 has mRNA tissue preference and uncover the function of this protein in organs not previously investigated, we analysed potential mRNA targets of ZFP36L2 in different tissues, using our newly developed Zfp36l2 flox global knockout mouse model (L2-fKO). The goal of our investigation was to define the mechanism and ‘rules’ that favour tissue-specific function of ZFP36L2, an ARE-RBP. Knowing that upon binding, ZFP36L2 accelerates the degradation of specific ARE-containing transcripts, we expect that target genes would be up-regulated in the L2-fKO animals. To identify these transcripts, we obtained differential expression transcriptomic data from lung, liver, bone marrow, spleen, kidney, and ovary. We found multiple transcripts up-regulated in each of these tissues as expected. However, there was a notable transcript selectivity of ZFP36L2 according to tissue even though most genes are expressed in all six tissues (Figure 3 and Supplementary Table 2). Only one up-regulated gene, Apol11b, was commonly up-regulated in the six different tissues investigated, suggesting that ZFP36L2 selectively targets different mRNAs in different tissues. Other alternative explanations would include (a) tissue-specific expression of target mRNAs (i.e. the gene is not expressed in some tissues, so cannot be regulated), (b) differential ARE accessibility due to secondary structure or competing RBPs, and (c) functional compensation by ZFP36L1 or ZFP36. ARE accessibility would require obtaining RNA secondary structure data for example using SHAPE-map and direct competition of ZFP36L2 with other RBPs, which are beyond the scope of our current investigation using our mouse model, thus cannot be excluded. However, the data presented here does not support possibilities (a) and (c). In Supplementary Table S1 (under GSE283043) we showed that the up-regulated genes are detected in each tissue at similarly high percentage rates, varying from 74% to 83%, tissues. Except for ovary, only few genes that were uniquely up-regulated are present solely in that tissue (Suplementary Table S3). These results suggest that up-regulated genes are expressed in all tissues but only up-regulated in that one, talking against possibility (a). To investigate functional compensation, possibility (c), we measure measured Zfp36 and Zfp36l1 expression in all six tissues (Supplementary Figure S2) and found no major compensation except in ovary. Another compensatory mechanism would be differences in mRNA-degradation machinery, which does not seem to be the case, given that the catalytic machinery was essentially unchanged in the knockout (Supplementary Figure S10). Despite this observed tissue selectivity of ZFP36L2 in choosing its mRNA targets in a particular tissue, we observed common trends. First, we clearly noticed that most of the up-regulated genes were coding genes, confirming at the transcriptomic level that ZFP36L2 preferentially modulates protein coding mRNAs. Second, AREscores were significantly increased only in up-regulated genes but not in down-regulated genes. This increase in AREscores was particularly evident in bone marrow, spleen and liver, which are haematopoietic organs at the early developmental age investigated here. Similarly, when ZFP36L2 was knocked down in a primary erythroid cell lineage [24] and knocked out in oocytes [], AREScores were increased in the up-regulated genes. The AREScore algorithm was developed and optimized for ZFP36 [30]. The wide use of AREScore confirmed its usefulness for detecting potential targets of other ZFP36 family members. Although some differences in ARE flanking PWMs were observed across tissues, the differences were small, suggesting that the tissue specificity of up-regulated genes is modulated by factors other than specific sequence motifs recognized by the tandem zinc finger domain of ZFP36L2. Third, when identifying common pathways in multiple tissues in up-regulated genes using Gene Ontology and GSEA we found terms related to immune system processes, response to stimulus and inflammatory pathways. Whereas genes uniquely up-regulated in a single tissue tended to be related to functions and biological processes particular to the physiology of that tissue. Surprisingly, among the coding genes modulated by ZFP36L2 the IgV genes were overrepresented (Figure 2). The IgV genes are unique to B cells, they encode the variable regions of antibodies, the domains that determine the uniqueness of an antibody. Notably, IgV genes undergo an extensive V(D)J recombination in B cells. V(D)J recombination is a process of somatic gene rearrangement unique to lymphocytes where variable (V), diversity (D), and joining (J) gene segments are recombined by the activity of the recombination activating genes (RAG). This recombination leads to an increased somatic hypermutation (SHM) rate with significant gene instability in germinal centre B cells. Thus, we hypothesize that ZFP36L2 can influence SHM in B cells, what would require investigation of DNA mutation or hypermutation in these cells. Future gene instability assays and DNA sequencing data in L2-fKO B cells will be necessary to investigate the mechanism by which ZFP36L2 contributes to genome stability or other DNA damage response (DDR). Other ARE-RBPs, such as AUF1, HuR, KHSRP, TIA-1, TIAR, including ZFP36L2, have been implicated in preserving genome stability by regulating DNA damage response [38]. In addition to this connection between ZFP36L2 and gene instability, we observed a cell cycle effect in B cells from the L2-fKO animals. These cells were arrested in G1 (or G0) of the cell cycle and very few cells entered in S phase when ZFP36L2 was absent. This finding supports our hypothesis that lack of ZFP36L2 affects DNA damage in splenic L2-fKO B cells. Note that Somatic hypermutation (SHM) is intrinsically linked to the B cell cycle, with high-rate mutagenesis restricted primarily to the G0/G1 phases. Interestingly, Galloway et al. observed that in pro-B cells, ZFP36L1 and ZFP36L2 are critical for cell cycle before B cells are mature and express Ig receptors in the cell surface. They proposed that both RBPs suppress mRNAs whose protein products cooperatively promote transition to the S phase of the cell cycle [39]. Recently [40], reported that the dKO of ZFP36L1 and ZFP36L2 results in G2/M arrest due to accumulation of DNA damage in S phase [40], as opposed to our observed G1 arrest as a result of L2-fKO. In our study, we used enriched B cells from an in vivo system, whereas in [40], they used an in vitro GC B cell culture system. It is possible that dKO of both ZFP36L1 and L2 in an in vitro system will result in different cell cycle effects compared to the in vivo single ZFP36L2-KO performed in our study. While they observed cell cycle-related genes such as Cdkn1a, Ccnd2, Ccne2, Cep76, Mdm2, Csnk1e, Nup88, and Ska2 increased in the dKO, none of these eight genes were captured in our eCLIP experiments. However, 11 other cell cycle-related genes were captured in our list of 2,143 eCLIP reproducible peaks, but none of them contained the minimal binding sequence for ZFP36L2, an ARE of the 7-mer type (UAUUUAU) suggesting an indirect control of ZFP36L2 on the cell cycle genes rather a direct binding. Only future studies may help clarify the underlying reasons for these variations. Additionally, in other biological models, ZFP36L2 has been implicated in directly modulating genes involved in terminal differentiation of other haematopoietic cell types, such as erythroid [24] and myeloid [27] cells, pointing to a progenitor defect in haematopoietic stem cells (HSC). The present work shows that ZFP36L2 affects B cell proliferation, which we suspect is not a cell-autonomous effect given that the up-regulation of IgV genes is only correlative. The analysis of overlap among up-regulated genes in multiple tissues revealed only one common gene, Apol11b. We confirmed that ZFP36L2 interacts with the 3’ UTR of Apol11b using gel-shift assays, but surprisingly this gene was not captured in our eCLIP data set. When we performed our eCLIP experiments we anticipated that even though MLTC-1 cells were unrelated to most tissues studied here, most genes would be commonly expressed. We confirmed that, given that approximately 70% of the genes expressed in these cells were simultaneously detected in the RNA-seq from all six mouse tissues (Supplementary Figure S7). However, there was still a possibility that Apol11b was not expressed in MLTC-1 cells. In fact, this is the case, Apol11b was not detected on a recent MTLC-1 RNA-seq [34], which we now confirmed by qPCR (not shown). This illustrates one of the main restrictions of eCLIP, it captures only genes expressed in the cell line it was performed, which may not reflect the same genes expressed in primary tissues. Therefore, some other hits captured in our RNA-seq can still be good candidates for direct ZFP36L2 target. Potentially, eCLIP experiments performed in a primary cell line, such as splenocytes, would overcome this limitation. Potentially, using splenocytes which are enriched in B cells would also clarify if ZFP36L2 directly interact with IgV genes. Additionally, we would like to pinpoint that eCLIP itself has its own limitations. One of them being that eCLIP does not capture all transcripts containing a consensus recognition for a given protein. Other groups have observed that this is the case for Unkempt, an RBP containing a cluster of CCCH-type zinc finger domains [41,42]. The reasons why, can be attributed to fragments of RNA captured that do not correspond to the short sequence involved in the direct binding. These could be considered false positive signals captured by the crosslinking between other domains of the protein that are in proximity with the RNA but are not directly mediating the binding. Aiming to increase the quality of eCLIP generated data, novel advances have been recently introduced, such as Size-Matched Input (SIM) improved library preparations and better bioinformatics pipelines [43,44]. Undoubtedly, the incorporation of footprinting SHAPE with eCLIP to identify nucleobases that are precisely hydrogen bound to the protein is one of the most promising advances to directly detect potential RNA consensus recognition sequence and RBPs [45]. Also extremely promising is the combination of non-isotopic ligation induced by UV crosslink and IP followed by mass spec [46]. Because of the intrinsic eCLIP limitations mentioned above, electrophoretic mobility shift assay (EMSA) was the alternative we employed to test the interaction of ZFP36L2 with Apol11b transcript. Given that Apol11b was significantly increased in multiple L2-fKO tissues and our EMSA showed a robust binding of ZFP36L2 to Apol11b, collectively we concluded that Apol11b is an in vivo target of ZFP36L2. When we performed a flanking motif analysis around the AREs captured by eCLIP peaks, an enrichment for uridines surrounding the AREs (Supplementary Figure S8E) was noticed, which was similar to the results when we used the up-regulated genes combined (Figure 6(C)). Consistent with that, when we applied a de novo motif analysis using HOMER to find specific nucleotide sequences recognized by the tandem zinc finger domain of ZFP36L2, we found three of the six first ranked motifs are U-enriched motifs, particularly in motif #5, UUUAUGC, (Supplementary Figure S9). Intriguingly, certain eCLIP peaks in our data did not contain a canonical AUUUA motif. Our observation was consistent with other ZFP36L2 eCLIP datasets using human cell lines [27], raising the possibility that ZFP36L2 binding is mediated by factors other than solely the presence of an ARE sequence motif. In summary, most eCLIP peaks tended to be uridine-rich, consistent with the analysis of sequences flanking ARE in the 3’ UTRs of up-regulated genes (Supplemental Figures S8E and S9) but our eCLIP data did not captured the classic ARE 7-mer consensus recognition sequence (UAUUUAU). Maybe the presence of other RNA-binding proteins or miRNAs interacting with sequences in the 3’ UTRs obscured our eCLIP detection. For example, the RBP ELAVL2 shares some ARE-RNA targets with ZFP36L2 [19] thus, potentially they compete for the binding to the same ARE sequence. However, we did not design this study to exclude these possibilities, but only to investigate potential ZFP36L2 targets. Around 10% of our eCLIP peaks simultaneously overlap with up-regulated genes indicating some agreement between these two orthogonal approaches, however one would expect greater overlap. We suspect that the smaller than expected overlap was related to the inability of eCLIP, in this particular case, to detect genes expressed at low levels. Most of the up-regulated transcripts had TPMs below 200, which is considered a low level of expression. Whereas most eCLIP peaks captured in our experiments had TPMs above 1000. This corresponds to one order of magnitude difference in the expression levels using these different experimental approaches (Supplementary Figure S6), potentially explaining the smaller than expected overlap between eCLIP and up-regulated genes. Because of that, only highly expressed ZFP36L2 targets were likely to be detected by eCLIP. Thus, in our case, the combination of RNA-seq and eCLIP is still useful but perhaps too stringent to broadly uncover transcripts targeted by ZFP36L2, particularly because they tend to be expressed at low levels. These unexpected results demonstrate an underappreciated complication of integrating differential gene analysis with eCLIP data. However, the 57 genes from eCLIP that overlap with up-regulated genes in tissues relevant for haematopoiesis identified deserve further careful investigation. Once binding to ZFP36L2 is confirmed and consequent RNA degradation, these genes may shed light in understanding myelodysplastic syndrome in children, a rare disease that happens in four out of one million births and has been limitedly investigated. Conclusions Among all RNA binding proteins that recognize adenine-uridine rich elements, ZFP36L2 plays a unique and vital role in RNA metabolism, as it is required for normal haematopoiesis. Here, by investigating potential ZFP36L2-mRNA targets in multiple tissues, we provide a new understanding into the intricate regulatory network orchestrated by ZFP36L2, opening avenues for testing distinct RNA decay functions of ZFP36L2 according to its tissue environment. Additionally, our results demonstrate an underappreciated complication of integrating differential gene analysis with eCLIP data. Material and methods Mouse model and sample collection We followed the ARRIVE 2.0 guidelines to report our animal experiments. Conditional Zfp36l2 knockout mice carrying loxP sites flanking exon 2 of Zfp36l2 (Zfp36l2fl/fl) were crossed with B6.CTg(CMV-Cre)1Cgn/J (stock#006054) mice from Jackson Laboratory to generate animals with Zfp36l2 alleles floxed as described [19]. These animals are referred to as L2-fKO or KO; they lacked Zfp36l2 in all tissues from the early embryo stage onward. In both groups, WT and L2-fKO, were distinguished based on genotyping as previously described in [19]. Animals carrying both wild-type copies of Zfp36l2 (WT) were used as controls. To minimize any potential confounders associated with development stage and sex, the collected tissues were derived from age-matched animals (D6 to D9 post-birth) of both sexes. In most collections, the WT and KO animal pairs were derived from the same litter aiming to minimize environmental or gestational effects. Thus, it was decided a priori that the animals to be included in the study had to be age-matched and derived from the same litter. Based on these criteria, even if a WT and a KO were from the same matched age but were born from a different litter, they were excluded and we did not collect samples. No randomization was used given that all collected samples were included. Only the person who genotyped the animas was aware of the group allocation, further conducted experiments, outcome and data analysis were performed blindingly. The primary outcome measured was genes differentially regulated and cell cycle analysis performed in both groups. Traditional two-tailed t-test was used for analysis. Carbon dioxide (CO2) inhalation was the method of euthanasia used for mice, followed by cervical dislocation. Forty-six samples were collected from 23 pairs of animals, three for lung, five for liver, five for bone marrow, four for spleen, three for kidney, and three for ovary. Immediately after death, tissues were rapidly collected, in less than 10–15 min and total RNA was extracted. For each tissue collected, the independent replicates clustered in a PCA and the variability for both groups is summarized in Supplementary Figure S1. The tissues were collected from animals on days 6 to 9, i.e. at a short window post-birth and reflecting a similar developmental stage (except for one ovarian sample collected at D13). The samples were collected at that age because the KO animals die before they are 2 weeks old, likely due to a severe pancytopenia of the blood cells. Animals were housed in the Dental School vivarium of the UNC School of Medicine, Chapel Hill. The Institutional Animal Care and Use Committee of the University of North Carolina approved all animal experimentation (IACUC ID# 24–004). RNA sequencing and data analysis Total cellular RNA was freshly extracted from dissected mouse tissues using RNeasy plus (Qiagen) and quantified using a NanoDrop spectrophotometer (ThermoFisher). Three micrograms of total RNA from each sample treated with DNase (TurboDNA-free kit, Invitrogen) for 60 min. The total RNA was re-measured, and 2000 ng of DNase-treated RNA was submitted to Azenta Life Science for RNASeq. All 46 samples passed QC with RNA Integrity Numbers (RIN) >6 (average 9.2 ± 1.02) and RNA fragments >200 of 70% or higher. The DNase-treated RNA samples were then used to prepare a library using a poly(A) selection. RNA sequencing was performed using an Illumina HiSeq platform by Azenta Life Science; the total yield was 388,685,940 reads. Each sample, representing a specific tissue, yielded an average of 35–60 million reads. Initial quality control involved using Trimmomatic v.0.36 to trim potential adapter sequences and low-quality nucleotides. For alignment, STAR aligner v.2.5.2b was used, and reads were mapped to the Mus musculus ENSEMBL GRCm38 reference genome. Specifically, we used a two-pass mapping strategy in STAR. In the first pass, STAR was used to generate a database of splice junctions from the unannotated data. This database was then incorporated into the second pass, resulting in improved alignment accuracy for the entire read sequences. Additionally, parameters were optimized for sensitive detection of splice junctions, including setting – outSAMstrandField intronMotif and – alignIntronMin 20—alignIntronMax 500,000. More than 97.5% of the trimmed reads mapped to the reference genome. Unique gene hit counts were subsequently determined using featureCounts from the Subread package v.1.5.2. Because of the strand-specific nature of the library, only unique reads mapping within exon regions were counted, maintaining strand specificity. Differential gene expression analysis was conducted using DESeq2. A direct comparison between wild-type and L2-fKO samples was performed. The Wald test was used to calculate p-values and log2 fold changes. Genes were considered differentially expressed if they exhibited an adjusted p-value below 0.05 and an absolute log2 fold change exceeding 1. We downloaded 3’ UTR sequences from the NCBI Biomart [47] and analysed the longest sequence for each gene. The AREScore program [30] was used to compute the AREscores for all 3’ UTRs and then analysed for the up-regulated and down-regulated genes [30]. We compared mean AREScores using a traditional two-tailed t-test. Splenocyte isolation The entire spleen was dissected from five WT and three KO Day 8 pups. Studied animals were from the same litter. Any adipose tissue was trimmed away from the organ, and the spleen was placed on a 40-μ Nylon mesh filter in a 35-mm petri dish. Once the spleen was finely pressed through the filter mesh, the round haematopoietic cells moved through easily, whereas the stromal vascular fraction remained on the filter. Three mL of filtered cells was collected from the petri dish and delicately layered on top of 2 mL of Histopaque-1077 (Sigma, density: 1.077) in a 15 mL conical tube and centrifuged in a swinging bucket for 35 min at 416 g at 32°C. A Polysucrose™ density gradient centrifugation was used to remove small red cells and debris. After centrifugation, the cloud of cells in the interface was collected, adjusted to a final volume of 4 mL, and washed twice with PBS. These freshly collected splenocytes were counted and plated in 35-mm petri dishes at a concentration of 20–30 × 106 cells/mL. The medium was RPMI, enriched in GM-CSF, and containing 10% FCS, 2 mM glutamine, 15 mM HEPES pH7.4, 0.2% NaHCO3, penicillin (100 units/mL), and streptomycin (100 μg/mL). Ethynyl-2’-deoxyuridine staining using click chemistry for cell cycle analysis Once in the incubator at 4% CO2 and 37°C, the splenocytes were treated with 10 μM EdU (Santa Cruz Biotechnology) for 60 min. Cells were then collected, washed with 1% BSA-PBS, and fixed in 4% paraformaldehyde in PBS for 15 min at ambient temperature. Fixed cells were permeabilized with 0.5% Triton X-100 in 1% BSA-PBS at ambient temperature for 15 min, centrifuged, then washed once with 1% BSA-PBS, and centrifuged again before labelling. For EdU detection, samples were incubated in PBS with 1 mM CuSO4, 1 μM Alexa 647-azide (Life Technologies), and 100 mM freshly made ascorbic acid for 30 min at ambient temperature in the dark [48]. Cells were washed and resuspended in 200 μL of PBS, 1% BSA, 0.5% Triton X-100 containing 100 μg/mL RNAse (Promega) and 1 μg/mL DAPI (Life Technologies) overnight at 4°C in the dark. On the following day, the cells were washed, resuspended in 500 μL of PBS, 1% BSA, 0.5% Triton X-100. Data were collected with an Attune NxT Flow cytometer and analysed with FlowJo software. For control samples we did not add EdU ex vivo and used this as the threshold of detection for EdU [49]. Cell culture, transfections, and protein extracts HEK 293 cells (American Type Culture Collection CRL-3216) were maintained in DMEM medium supplemented with 10% foetal bovine serum, penicillin (100 units/mL), and streptomycin (100 μg/mL). Transient transfection of 2 × 106 cells seeded in a 100-mm plate with different mouse Zfp36l2 plasmids was performed using Lipofectamine 2000 (Life Technologies) according to the manufacturer’s protocol [33]. The transfection mixture containing 1 ug of DNA for each construct was incubated with the cells for 12–18 h and then replaced with fresh medium for a further 24-h incubation, after which the cells were lysed for protein extraction. Immunoblot analysis Protein samples were quantitated by Bradford assay (Bio-Rad). Five μg of protein per lane was loaded on a 10% Tris-Glycine polyacrylamide gel and electrophoresed under denaturing conditions. The proteins were transferred from the gel to a nitrocellulose membrane using the Trans-Blot® TurboTM System (Bio-Rad). The membrane was blocked for 1 h with 5% non-fat milk (w/v) in a Tris-buffered saline solution containing 0.1% TWEEN (TBS-T). The following primary antibodies were used for protein detection: mouse monoclonal anti-GFP (Abcam #AB290), dilution 1/1000. The HRP-conjugated secondary antibodies were anti-mouse (Gena-N931V), dilution 1/5000. The HRP signal was developed using the Bio-Rad Western Clarity TM chemiluminescent substrate. Preparation of RNA probe for RNA electrophoretic mobility assay An RNA probe was synthesized with the Riboprobe System-T7 (Promega) using DNA primers having sequences immediately downstream from a T7 promoter, as described [50]. The RNA was body-labelled during the transcription process, which was performed in the presence of [α-32P] UTP (3000 Ci/mmol; PerkinElmer). The probe was designed to be ~30 nucleotides when we considered the location of the ARE motif. The synthesized RNA probe was separated from the free nucleotides using Sephadex G50 columns (GE Healthcare Life Sciences) and electrophoresed on a 16% polyacrylamide urea gel. The probe was purified from excised gel fragments after detection by autoradiography, as described [29]. The amount of RNA probe used in each EMSA sample was calculated to be ∼10 femtomoles. The sequence of the RNA probe contained a single ARE of the 7-mer type (AUAAUAU) from the 3’ UTR of Apol11b. The corresponding sequence was Apol11b: 5’- AAUGUCAGAAAUAUUUAUUUUCUUGAAGA − 3’. RNA electrophoretic mobility assay HEK 293T cells were transiently transfected with wild type mZfp36l2 (WT-L2), mouse Zfp36l2-C175S (disruption of the first zinc finger domain), ZFP36-family members fused to GFP [32], or GFP vector plasmids. Then, protein extracts were prepared as described [29]. The protein extracts were incubated for 15 min at ambient temperature with 0.2 × 105 cpm of 32P-labelled RNA probe in a final volume of 20 μL containing 10 mM HEPES (pH 7.6), 40 mM KCl, 3 mM MgCl2, 0.5 μg/μL heparin, and 60 ng/uL yeast tRNA, as described [29]. The resultant mixtures were applied to 6% nondenaturing acrylamide (37.5:1) gels and subjected to electrophoresis at 150 V for 15 min followed by electrophoresis at 200 V for 90 min in 0.4× Tris-borate/EDTA running buffer. The gels were dried, exposed to film (Carestream BIOMAX MR Film), and developed after 12–20 h of exposure. Each gel shift assay was repeated three times, using different protein extracts to ensure rigour and reproducibility. Cell culture, UV crosslinking, and eClip MLTC-1 cells (American Type Culture Collection, CRL 2065) were cultured in RPMI medium with 10% foetal bovine serum, penicillin (100 units/ml), and streptomycin (100 μg/ml), at 37°C in an incubator with 5% CO2 in atmospheric air. MLTC-1 cells are derived from murine Leydig tumour cells, in which we have previously shown high expression of ZFP36L2 in Westen blots [33]. MLTC-1 cells at 70–80% confluency from a 100-mm petri dish were trypsinized, washed in cold PBS twice, counted and adjusted to 20 × 106 cells/ml. A total of 80 × 106 cells were resuspended in 4 mL of PBS and placed in a clean 100-mm petri dish, on top of an ice plate, inside the UV crosslinker (Hoefer UVC 500 crosslinker). The distance from the 254-nm UV bulbs to the cells was 11 cm. Then, the cells were exposed to 254-nm light with an energy setting of 400 mJoules/cm2. After UV crosslinking, cells were washed in PBS, counted again, and pelleted (20 × 106 cells) in an Eppendorf tube and snap frozen. Enhanced Crosslinking and Immunoprecipitation (eCLIP) was performed by EclipseBio BioInnovations Inc (San Diego; https://www.eclipsebio.com/) according to [51] protocol. In short, total RNA was extracted from the frozen pellets of MLTC-1 cells after UV-crosslinking, quantified, and analysed for quality using an Agilent 4200TapeStation. Twenty ug of RNA with a RIN of 8.35 was used for immunoprecipitation with the validated Abcam ab70775 antibody. As a negative control, we used a rabbit IgG (ab313801 from Abcam). Following the immunoprecipitation, the samples were treated with proteinase K. The IP-RNA was first fragmented and then used to create a library using an RNA adapter on the 3’ end of the target RNA. The RNA was reverse transcribed resulting in cDNAs ligated to single-stranded DNA adapters on the 3’ end to serve as identifiers against PCR duplicates. Finally, the paired-end cDNA fragments were amplified to obtain sufficient material for high-throughput sequencing. eCLIP analysis The CLIPper algorithm [52] was used to identify RBP-binding peaks. This method calculates significant read clusters relative to gene-level read distribution and length, then filters false positives by requiring a substantial IP/input library fold-change enrichment (we used log2 ≥3). Two thousand one hundred and forty-three eCLIP sequences were identified with log2(Fold Change) >1 and log10(p) <1. These peaks mapped to 597 different genes. For the flanking motif analysis around the AREs we started with all 2,143 reproducible eCLIP peaks, then filtered to those with positive ARE scores, resulting in 320 eCLIP peaks. From these 320 peaks, we extracted 800 WUUUW motif instances (with 10nt flanks) for the Logomaker visualization. A standard motif analysis for RBPs in our eCLIP peaks was initially performed by EclipseBio and did not identify any known motifs commonly attributed to RBPs. Thus, we performed a more advanced de novo motif analysis using HOMER v.5.1 (Hypergeometric Optimization of Motif EnRichment). For our HOMER analysis we used 2,143 reproducible (peaks present in both experimental replicate) and filtered for the peaks that were located only at the 3’ UTRs, which resulted in 258 peaks. Using these 258 peaks we run HOMER to identify potential novel motifs bound to ZFP36L2. As a background control we used other 7,308 non-reproducible eCLIP peaks (i.e. 3’UTR peaks that did not overlap with the reproducible peaks). Statistical calculations Traditional two-tailed t-test was used in most analysis, except for Supplementary Figure S4, when the Welch’s two-sample t-test was used to compare mean expression levels between WT and KO samples, because it assumes the expression levels are approximately normally distributed but allow for unequal variances in different tissues, which was the case. p values are *<0.05, **<0.005, ***<0.0005 and ****<0.00005, respectively. Supplementary Material Supplementary Table 1_page 1.docx ARRIVE guidelines 2_RNA Biol 2026.pdf Supplementary Figures S1-S10.docx Supplementary Table 3_FINAL.docx Supplementary Table 2_FINAL.docx