MNK2 deficiency potentiates β-cell regeneration via translational regulation

Regenerating pancreatic β-cells is a potential curative approach for diabetes. We previously identified the small molecule CID661578 as a potent inducer of β-cell regeneration, but its target and mechanism of action have remained unknown. We now screened 257 million yeast clones and determined that CID661578 targets MAP kinase-interacting serine/threonine kinase 2 (MNK2), an interaction we genetically validated in vivo. CID661578 increased β-cell neogenesis from ductal cells in zebrafish, neonatal pig islet aggregates and human pancreatic ductal organoids. Mechanistically, we found that CID661578 boosts protein synthesis and regeneration by blocking MNK2 from binding eIF4G in the translation initiation complex at the mRNA cap. Unexpectedly, this blocking activity augmented eIF4E phosphorylation depending on MNK1 and bolstered the interaction between eIF4E and eIF4G, which is necessary for both hypertranslation and β-cell regeneration. Taken together, our findings demonstrate a targetable role of MNK2-controlled translation in β-cell regeneration, a role that warrants further investigation in diabetes.

B oth type 1 and type 2 diabetes manifest with elevated circulating glucose levels caused by the deregulation of insulin signaling and/or the loss of functional insulin-producing β-cells 1 . Although daily insulin injection, lifestyle interventions and various drug treatments can manage the disease, there is currently no available cure. Therefore, stimulating endogenous β-cell regeneration is an attractive curative approach for diabetes; however, current efforts have failed to translate into a clinically approved drug.
Drug screening in vivo for chemicals stimulating β-cell regeneration has the potential to accelerate the drug discovery process as it is performed in a physiological, whole-organism setting. The zebrafish model has emerged as a powerful tool for performing unbiased, large-scale chemical/genetic screens that are directly coupled to phenotypic analyses. These types of screens have identified molecules that have already entered clinical development 2,3 , showcasing the translational potential of the approach. Metabolism, and in particular diabetes research, is an area in which chemical screens using zebrafish have grown in popularity. Such screens have already identified compounds that can stimulate β-cell proliferation [4][5][6] and neogenesis from duct-residing pancreatic progenitors [7][8][9] , two of the main known mechanisms of endogenous β-cell regeneration. Moreover, zebrafish chemical screens for regulators of glucose metabolism have identified compounds with potential use as antidiabetic treatments [10][11][12] . Collectively, these studies have demonstrated the power of the zebrafish model for performing chemical screens to identify compounds that could be repurposed as antidiabetic drugs. However, many hits from phenotypic screens have unknown targets and mechanisms of action that create a bottleneck for further development but, if the targets are defined, can open up new research areas.
In this work, we aimed to identify the molecular mechanism of action of CID661578 (1), the most striking hit from a zebrafish chemical screen for stimulators of β-cell regeneration 4 . By performing a modified yeast two-hybrid screen suited for drug target deconvolution, we identified MAP kinase-interacting serine/ threonine kinase 2 (MNK2) as the molecular target of CID661578 and validated this interaction in vivo. MNK2 participates in initiation of mRNA translation and has been postulated to modulate the process in a transcript-selective fashion 13 . Here, we show that kinase-independent blocking of MNK2 leads to bolstered protein synthesis in the pancreatic duct and that the effects are conserved across zebrafish, pig and organoid cultures of human pancreatic ducts. Overall, our results demonstrate a conserved pathway to stimulate β-cell neogenesis by boosting protein synthesis through targeting MNK2.

Results
Yeast chemical hybrid screen identifies MNK2 as the molecular target of CID661578. In a previous large-scale chemical screen, we identified five small molecules that potently drove β-cell regeneration in zebrafish larvae 4 . Four of the five hit compounds converged on the adenosine pathway and stimulated β-cell proliferation, while the fifth hit compound, named CID661578, had no known molecular target or cellular mechanism of action ( Fig. 1a). To identify the molecular target of CID661578, we used yeast chemical hybrid (YChemH) screening technology, which takes advantage of the  Fig. 1 | yChemH screen identifies MNK2 as the molecular target of CiD661578. a, Schema for the screening of compounds increasing β-cell regeneration using a transgenic zebrafish model for β-cell ablation and approximately 10,000 compounds. The hits included four compounds affecting adenosine signaling and CID661578 with an unknown mechanism. b, Schematic showing the structures of CID661578 and the analog CID661578.6 along with the screening strategy (YChemH). The red circles highlight the structures that were altered in CID661578. Survival of yeast on selective histidine-free medium was the output of the screen for clones expressing interactors of the CID661578.6 bait; TMP, trimethoprim; AD, activation domain. c, Table summarizing the top hits of the YChemH screen from the two cDNA libraries. The A-classified hits (drl and acin1b from the zebrafish embryo library and MKNK2 from the human islet library) have a higher probability of being true targets of CID661578.6 than B-and C-classified hits. d, Validation of the MNK2-CID661578.6 interaction with different concentrations of CID661578.6 bait and an MNK2-expressing yeast clone. DMSO demonstrates the sensitivity to the selective medium, and yeast clones did not survive in the selective histidine-free medium. The interaction between MNK2 and CID661578.6 promoted yeast survival, as illustrated by the multiple colonies at the four spots of inoculation (decreasing levels of inoculation from the top to the bottom). Each condition was tested in two replicates. e, Validation of the zebrafish Mnk2b-CID661578. 6 interaction with different concentrations of CID661578.6 bait and two different DHFR hook vectors. Experiments using the original hook vector, N-LexA-DHFR-C, are listed as 1, 2 and 3. Experiments using the modified vector with the reverse order, N-DHFR-LexA-C, are listed as 4, 5 and 6. Both full-length zebrafish Mnk2b (3 and 6) and a fragment (2 and 4) corresponding to the original fragment of the human MNK2 identified in the screen were used. Human MNK2 was used as a positive control (1 and 4), and zebrafish Mnk2b only mediated binding when expressed by the hook vector with the reverse order (5 and 6) to the one used in the original screen (explaining why zebrafish Mnk2b did not show up as a hit in the original screen). classical yeast two-hybrid system for protein-protein interactions but enables screening for protein targets of small molecules. First, we generated a series of structural analogs of CID661578 to simplify the structure and identified one (termed CID661578.6 (2)) that exerted similar effects on β-cell regeneration as CID661578 (Extended Data Fig. 1). We used yeast clones with a construct that expressed a DNA-binding domain (LexA) coupled to the enzyme dihydrofolate reductase (DHFR) along with a GAL4 activation domain fused to each cDNA in the libraries. We screened two different cDNA libraries, one derived from human islets and another derived from zebrafish embryos, with a total of 135 million and 122 million clones, respectively. The rationale of this approach is that when yeast are incubated with the CID661578.6-derived bait (3), the attached trimethoprim will interact with DHFR on the DNA-binding site of LexA. Interaction between CID661578.6 and a protein fragment (that is, the prey) generated from the cDNA library will lead to yeast survival on the selective histidine-free medium (Fig. 1b). The most likely targets of CID661578 were classified as A hits, and less likely targets were classified as B hits, C hits and so forth, depending on the confidence in the interaction (Fig. 1c).
Next, we followed up on hits that had previously been shown to affect metabolism and could validate the interaction in an MNK2-expressing yeast clone with different concentrations of the CID661578.6-derived bait in the YChemH system (Fig. 1d). Subsequently, we assessed whether the zebrafish homolog of MNK2 (Mnk2b) could also bind CID661578.6 because it did not appear as a hit in the zebrafish cDNA library. To this end, we cloned the cDNA sequence of mknk2b (the genes encoding Mnk kinases are called mknk) in frame with the GAL4 activation domain and used two different DHFR hook vectors of the YChemH system. The YChemH assay results revealed that zebrafish Mnk2b only bound to CID661578.6 when the new, modified DHFR hook vector was used, while human MNK2 bound to CID661578.6 regardless of the vector conformation (Fig. 1e), explaining why Mnk2b did not show up as a hit in the original screen of the zebrafish cDNA library. In summary, through a series of in vitro experiments, we identified MNK2 as the likely molecular target of CID661578.
CID661578.6 promotes β-cell regeneration of ductal origin. To identify the cellular source of the newly formed β-cells, we first assessed their proliferation status. We used the β-cell ablation zebrafish model, Tg(ins:flag-NTR), in which the enzyme nitroreductase (NTR) is expressed under the control of the β-cell-specific insulin promoter. When the prodrug metronidazole (MTZ) is administered, NTR converts it to a toxic byproduct, resulting in the specific ablation of β-cells 14,15 . We treated the zebrafish β-cell ablation model with CID661578 in the presence of EdU to label dividing cells, and quantification of EdU + β-cells showed no alteration in proliferation status between the control and CID661578-treated groups (Extended Data Fig. 2a-c). Next, we performed lineage tracing of the pancreatic ductal cell population using the inducible Tg(tp1:creER T2 ) line (tp1 is a Notch-responsive element characterizing the duct population) and the responder Tg(ubi:switch) line. We induced Cre-mediated recombination (5-6 days postfertilization (d.p.f.)) and let the larvae grow to the juvenile stage, followed by β-cell ablation and treatment with CID661578.6 ( Fig. 2a). Quantification of β-cells along the tail of the pancreas (often referred to as secondary islets) revealed that treatment with CID661578.6 increased the number of regenerating β-cells derived from the ductal cell population (Fig. 2b-e). Moreover, we used a complementary lineage tracing approach to validate these findings, where double-transgenic Tg(tp1:H2BmCherry); Tg(ins:GFP) zebrafish in which tp1 drives the expression of the stable fluorescent protein H2BmCherry as a ductal cell tracer were used, and we confirmed ductal-derived β-cell regeneration (Extended Data Fig. 3a-e). Furthermore, we excluded the possibility that the chemical treatment altered the number and proliferation of the Notch-responsive ductal cells by assaying for EdU incorporation into Tg(tp1:GFP) zebrafish (Extended Data Fig. 2d-g). Thus, by using two different lineage tracing strategies, we identified ductal cells as the cellular source of the newly formed β-cells.
We subsequently assessed glucose levels after ablation of β-cells and treatment of adult fish with CID661578.6 for 3 d. As expected, β-cell ablation caused an increase in blood glucose at 3 d after ablation in the control fish, whereas fish treated with CID661578.6 had significantly lower blood glucose than controls (Fig. 2f). Glucose levels were also reduced in zebrafish larvae following CID661578 or cercosporamide (4) (a previously described selective inhibitor of MNK2 (ref. 16 )) treatment (Extended Data Fig. 2h). The newly formed β-cells also appeared mature in terms of mnx1 expression (Extended Data Fig. 4a-d) as well as being devoid of glucagon and insulin coexpression (Extended Data Fig. 4e-g). Taken together, these results show that compounds interfering with Mnk2 lower glucose levels in both larval and older zebrafish.
Next, we explored mknk2b expression in the adult zebrafish pancreas using a recently published single-cell RNA-sequencing (RNA-seq) dataset 17 that contains all the major pancreatic cell types ( Fig. 2g and Extended Data Fig. 5). The expression of mknk2b was not ductal specific, yet its highest expression was in a cluster of ductal cells (Fig. 2h). Taken together, these results showed that CID661578.6 increased the regeneration of functional β-cells in both larval and older zebrafish by promoting β-cell neogenesis from a ductal origin.
CID661578 targets Mnk2b in vivo to drive β-cell regeneration. Subsequently, we wanted to determine whether the in vivo engagement of Mnk2b is responsible for the observed phenotypes of CID661578 treatment. We treated zebrafish larvae with CID661578, cercosporamide or their combination from 4 to 6 d.p.f. during β-cell regeneration. We observed a dramatic increase in regenerating β-cells in the primary islets of zebrafish larvae following treatment with either chemical, but no additive effect was observed when they were combined (Fig. 3a-e). This result suggested that cercosporamide and CID661578 have similar effects on β-cell regeneration, and the absence of additive/synergistic effects indicated that they converge on a common molecular target/pathway. Further, we found that neither CID661578 nor cercosporamide treatment affected the development of any of the endocrine cell populations in zebrafish larvae, suggesting that the effects of this pathway are restricted to the regenerative state (Extended Data Fig. 6a-i).
We also tested if the effects of the chemicals targeting Mnk2 could be reproduced using genetic approaches. The mknk2 gene is duplicated in the zebrafish genome, and an investigation of the expression of the two paralogs mknk2a and mknk2b in published RNA-seq data revealed that mknk2b is the predominantly expressed paralog in ductal cells 17 . Therefore, we generated a mknk2b full-body knockout zebrafish using CRISPR-Cas9 mutagenesis to target the N-terminal part of the protein. First, we observed physiological β-cell development in the homozygous mutants ( Fig. 3f-i). However, β-cell regeneration was enhanced in the homozygous mknk2b mutants following β-cell ablation ( Fig. 3j-m), suggesting that the absence/inhibition of Mnk2b is responsible for this phenotype. We reproduced the mknk2b-knockout phenotype using a splice-blocking morpholino to knockdown mknk2b (validated by quantitative PCR with reverse transcription (RT-qPCR) (Extended Data Fig. 7k)). Knockdown of mknk2b increased the number of β-cells, while no additive effect was observed from simultaneous mknk2b knockdown and CID661578.6 treatment (Extended Data Fig. 7a-e). In addition, morpholino knockdown of the other two A hits from the zebrafish library, acin1b and drl, did not increase β-cell regeneration, suggesting that these two genes are not responsible for    An unpaired two-tailed Student's t-test was used to assess significance for d (*P = 0.0393), and a two-tailed Mann-Whitney test was used for e (**P = 0.0087). Data are presented as mean values ± s.e.m. The experiment shown in b and c was repeated twice with similar results. f, Blood glucose was measured 3 d post-β-cell ablation (d.p.a.) in 4-month-old fish treated with DMSO or CID661578.6. Blood glucose levels in zebrafish without β-cell ablation were included as a basal-state reference; n = 7 (control), n = 10 (control, 3 d.p.a.), n = 10 (CID661578.6, 3 d.p.a.). A one-way ANOVA followed by Šidák's multiple comparisons test was used to assess significance for f (**P = 0.0078). Data are presented as mean values ± s.e.m. g,h, UMAP plots showing the different cell types present in the adult zebrafish pancreas after reanalysis of published single-cell RNA-seq data (g) and expression of mknk2b (h) at various levels in the different clusters. the observed phenotypes (Extended Data Fig. 7f-m). These results further supported that β-cell regeneration increased in the absence of mknk2b.
Finally, we reasoned that because the knockout and knockdown of mknk2b had similar effects as CID661578.6 treatment, overexpression of the protein would sequester CID661578.6 and reduce β-cell regeneration. To this end, we cloned zebrafish mknk2b and human MKNK2 and overexpressed them under the control of the tp1 promoter in the β-cell ablation model, which was subsequently treated with CID661578.6. Overexpression of either mknk2b or MKNK2 significantly decreased the effect of CID661578.6 on β-cell regeneration ( Fig. 3n-r). These experiments using mosaic overexpression were also confirmed in a stable line overexpressing mknk2b, which showed an even stronger reversal of the CID661578.6 effect on β-cell regeneration (Extended Data Fig. 7n-r). Collectively, these data support Mnk2b as the molecular target of CID661578.6 in vivo and that Mnk2b can restrict β-cell neogenesis from a ductal origin.
CID661578 boosts translation to increase β-cell regeneration. To better understand the molecular mechanism induced by CID661578, we treated zebrafish larvae with CID661578 for 24 h before global metabolomics characterization 18 . After creating a metabolite profile of CID661578-treated zebrafish, we identified differentially regulated metabolites (Fig. 4a). An interesting observation was that the levels of many amino acids were altered following CID661578 treatment. MNK2 interacts with a complex of eukaryotic translation initiation factors and thereby plays a role in protein synthesis 19 . Thus, the metabolomics data indicated possible changes in protein synthesis as a key effect of CID661578, consistent with the known role of MNK2. Moreover, α-d-glucose was significantly downregulated, an observation that we replicated using glucose measurements with an in vitro assay for both CID661578 and cercosporamide ( Fig. 4a and Extended Data Fig. 6j). Metabolite set enrichment analysis of zebrafish-specific pathways showed that downregulated metabolites following CID661578 treatment were related to non-essential amino acid metabolism (Fig. 4b), while the pathways related to upregulated metabolites were less impacted and limited to changes in pyrimidine metabolism (Extended Data Fig. 8a). Further, enrichment analysis of the single-cell RNA-seq data used to examine the expression of mknk2b showed that genes enriched in the ductal cells have a role in mRNA translation (Extended Data Fig. 8b). Taken together, these results further strengthen the hypothesis that CID661578 affects global changes in protein synthesis and glucose metabolism, in agreement with protein synthesis being a highly energy-consuming process that should reduce nutrient levels, including glucose.
To further investigate the changes in protein synthesis in vivo, we measured the incorporation of O-propargyl-puromycin (OPP), a modified amino acid that is incorporated in proteins during translation and can be visualized with a Click-iT reaction. Following β-cell ablation, larvae were incubated with OPP for 20 h concomitant with treatments of CID661578.6, 4EGI-1 or their combination. 4EGI-1 inhibits the interaction between the translation initiation factors eIF4E and eIF4G, which together with MNK2 function within the translation initiation complex 20 . The rationale of this assay was to assess the outcome of altering the translation initiation complex composition during CID661578.6 treatment. OPP incorporation was predominantly observed in the ductal cells of the pancreas, indicating a high protein synthesis rate in this population. CID661578.6 treatment drastically increased the incorporation of OPP in the ductal cells (and the intestine), an effect abolished following cotreatment with the inhibitor 4EGI-1 ( Fig. 4c-h). These data suggest that CID661578.6 boosts protein synthesis and that its effect is dependent on the translation initiation complex.
Encouraged by the observation that 4EGI-1 inhibited CID661578.6-induced protein synthesis, we examined whether 4EGI-1 could also inhibit the induced β-cell regeneration. To this end, zebrafish larvae were treated during the regenerative period with CID661578.6, 4EGI-1 or their combination. Interestingly, 4EGI-1 treatment was also sufficient to inhibit CID661578.6-induced β-cell regeneration ( Fig. 4i-m). Thus, through a combination of metabolomics and protein synthesis measurements, we demonstrated that CID661578.6 induces protein synthesis in vivo and that the effect on both protein synthesis and β-cell regeneration is blocked by targeting the translation initiation complex that MNK2 is a part of.
after treatment with CID661578.6 or cercosporamide. We did not observe any drastic changes in kinase activity with CID661578.6 treatment, whereas cercosporamide decreased the kinase activity of numerous kinases (Extended Data Fig. 9a,b). These results suggest that CID661578.6 does not affect protein synthesis by inhibiting the kinase activity of MNK2.
We then reasoned that CID661578.6 binding to MNK2 could alter the composition of the translation initiation complex. Given that inhibiting the interaction between translation initiation factors eIF4E and eIF4G was sufficient to block the effects of CID661578.6 in vivo, we hypothesized that CID661578.6 binding to MNK2 may enhance the interaction between eIF4E and eIF4G at the mRNA cap.  To test our hypothesis, we treated the COLO 320HSR cell line with CID661578.6, 4EGI-1 or their combination. Subsequently, we pulled down the cap-binding protein eIF4E from cell lysates using beads with an immobilized m 7 GTP structure of the mRNA cap. Treatment with CID661578.6 stabilized the eIF4E-eIF4G interaction, an effect that could be reversed following treatment with the inhibitor 4EGI-1 (Fig. 5d). Subsequently, we performed the same m 7 GTP pulldown assay in vitro using rabbit reticulocytes. The advantage of using this in vitro system is that rabbits do not have an ortholog of MKNK2. CID661578.6 did not increase the eIF4E-eIF4G interaction in rabbit reticulocytes, indicating that the increase in the eIF4E-eIF4G interaction after CID661578.6 treatment is dependent on MNK2 (Fig. 5e). Lastly, we asked how the interaction between CID661578 and MNK2 could affect the recruitment of MNK2 to the translation initiation complex. For this experiment, we used the human pancreatic cancer cell line PANC-1, as it was more efficiently transfected than the COLO 320HSR cell line. We began by validating that both CID661578 and its analog CID661578.6 increased the eIF4G-eIF4E interaction in PANC-1 cells by using the m 7 GTP pulldown assay (Fig. 5f). Next, we transfected PANC-1 cells with a FLAG-MNK2 plasmid, added DMSO/CID661578 and performed immunoprecipitation of MNK2 using anti-FLAG. Immunoblotting against eIF4G (which is the protein that directly interacts with MNK2 in the complex) revealed that CID661578 treatment largely abolished the interaction between MNK2 and eIF4G (Fig. 5g). Finally, we assessed whether phosphorylation of eIF4E was affected by CID661578-induced changes in the translation initiation complex composition. Unexpectedly, we observed that phosphorylation of eIF4E was increased after treatment with CID661578 and cercosporamide (that is, selective interference of MNK2) but was nearly abolished with two broader inhibitors blocking both MNK1 and MNK2 (that is, CGP57380 and eFT508; Fig. 5h). To address whether increased eIF4E phosphorylation is important for β-cell regeneration, we cotreated fish with CID661578 and eFT508 and observed that eFT508 could inhibit the increase in β-cell numbers induced by CID661578 (Fig. 5i). Taken together, these data suggest that CID661578 binds to MNK2, preventing it from interacting with eIF4G, which can increase MNK1-dependent phosphorylation of eIF4E and bolsters the interaction between eIF4E, eIF4G and the mRNA, resulting in increased protein synthesis.
As alterations in both eIF4E phosphorylation and increased eIF4F-complex (that is eIF4E, eIF4G and eIF4A) formation affects mRNA translation in a transcript-selective fashion 21,22 , we sought to identify translationally regulated mRNAs following both CID661578 and cercosporamide treatments. To this end we performed polysome profiling of PANC-1 cells after chemical treatments and used RNA sequencing to quantify total mRNA and mRNA associated with more than three ribosomes (Fig. 5j). We then identified mRNA whose translational efficiency was modulated along with mRNAs with changed abundance and translationally buffered (Fig. 5k,l and Methods). We focused our analysis on transcripts with altered translational efficiencies predicted to affect protein levels and found that there was a highly significant overlap of hypo-and hypertranslated mRNAs between CID661578 and cercosporamide treatments (Extended Data Fig. 10a-c). Overall, we identified a total of 270 hypertranslated and 99 hypotranslated mRNAs that were shared between treatments ( Supplementary Data 1-3). This further highlights that cercosporamide and CID661578 target overlapping molecular pathways in PANC-1 cells. Gene ontology (GO) analysis identified several GO terms enriched among proteins encoded by mRNAs that were hypotranslated in response to both compounds, with the most striking being mitochondrial-related processes (Extended Data Fig. 10d). By contrast, there were no significantly enriched pathways among shared hypertranslated mRNAs. Lastly, we examined the 5′ untranslated region (5′ UTR) sequences in search of features of translationally regulated mRNAs. Our analysis demonstrated differences in GC content among translationally regulated mRNAs following chemical treatments (Extended Data Fig. 10e,f), a feature underlying 5′ UTR structures. Therefore, both CID661578 and cercosporamide modulate mRNA translation in a selective fashion where the hypertranslated mRNAs had 5′ UTRs with low GC content and the hypotranslated mRNAs had 5′ UTRs with high GC content, consistent with previous studies on eIF4E-regulated translation 23,24 .

CID661578-induced β-cell neogenesis translates to mammals.
To examine whether our findings were translatable to mammals, we took advantage of an in vitro culture system of neonatal pig islet aggregates. Pancreata from 3-d-old pigs were digested, and the islet aggregates were generated and cultured in vitro for 3 d before a 5-d treatment with CID661578 or cercosporamide. These islet aggregate preparations are highly enriched in intraislet ductal cells, making them an ideal model to study the effect of the assayed chemicals. MNK2 is expressed in the duct as well as in islets of juvenile and adult pigs (Extended Data Fig. 9c-e). Treatment with either CID661578 or cercosporamide increased the number of insulin + cells in the islet aggregates ( Fig. 6a-d). The number of CK7 + ductal cells decreased after treatment with CID661578, while the number of cells coexpressing insulin and CK7 increased following either treatment (Fig. 6e,f). These results indicated that the new β-cells also have a ductal origin in neonatal pig islets and showed that the increase of β-cells in the zebrafish could be translated to a mammalian model.

2-Phenylethanamine
Adrenochrome/N-benzoylglycinate Taurodeoxycholic acid/taurochanodeoxycholate population of cells along the pancreatic ducts. The MNK2-expressing cells that were located just outside the luminal ductal lining most often did not coexpress MNK2 and CK19. However, we also observed MNK2 + CK19 + ductal cells at other positions in the same ducts (Fig. 6h,i). To address whether CID661578 or cercosporamide can stimulate differentiation of human ductal cells toward β-cells, we generated ductal organoid cultures from healthy human donors (Fig. 6j). Encouragingly, treatment of the human organoid cultures with either CID661578 or cercosporamide increased the expression of INS mRNA compared to the DMSO-treated controls (Fig. 6k). In sum, taking into account the potent effect of MNK2-interfering drugs on β-cell differentiation in neonatal pig islets and human ductal organoids as well as the intriguing expression pattern of MNK2 in humans, further investigation of the translational potential of this class of drugs is warranted.

Discussion
In the current study, we identified the molecular target of CID661578 as MNK2. The MNK2-CID661578 interaction potently induced β-cell regeneration from a pancreatic ductal cell origin and was sufficient to improve glucose control in both larval and adult diabetic zebrafish models. CID661578 was the only hit from our previous chemical screen that did not have a known target 4 . An important step in the drug discovery process is the identification of protein interactors and target engagement in vivo. Here, we used the YChemH system and identified MNK2 as one of the molecular targets of CID661578. Additionally, we used a combination of polysome profiling and in vitro biochemical experiments to shed light on the molecular mechanism of action of CID661578 in pancreatic ductal cells.
The neogenesis of β-cells from pancreatic ductal cells has been previously observed in the zebrafish model and has now been accepted as an endogenous path for β-cell regeneration 7,25-27 . However, whether a similar pancreatic progenitor population resides within the ductal cell compartment in adult mammalian models remains unclear. Different lineage tracing methods in mouse models have yielded various results regarding the contribution of ductal cells to β-cell neogenesis 8,[28][29][30][31][32][33][34][35][36][37][38] . This controversy highlights the need to study ductal cells and their progenitor potential in the zebrafish model to identify new effectors and markers that could be translated to mammalian systems. Interestingly, CID661578 treatment also increased the formation of new β-cells from intraislet ductal cells in cultures of neonatal pig islet aggregates and stimulated INS expression in ductal-derived human organoids. These results demonstrated that the pathway could also be exploited to increase the differentiation of β-cells from a ductal source in mammalian systems.
We also observed that the Mnk2b-CID661578 interaction potently lowered glucose levels in zebrafish, suggested to be due to a combination of increased β-cell regeneration and protein synthesis (one of the most energy-consuming processes in the cell). This was supported by the fact that glucose lowering also occurred during homeostasis following chemical treatment, suggesting that increased glucose consumption is possibly the fuel for the increased protein synthesis. Interestingly, a recent report has linked analogs of cercosporamide to a glucose-lowering effect in mice 39 . Furthermore, Mnk1-and Mnk2-knockout mice exhibit beneficial metabolic outcomes when challenged with a high-fat diet 40,41 . These data are consistent with our observations and open new avenues for exploiting MNK2 in metabolic diseases.
During the course of our study, we performed metabolomics to assess the effects of CID661578 treatment in vivo, which highlighted global changes related to glucose metabolism and protein synthesis. We observed that the chemical treatment increased protein synthesis in vivo, an effect that was most profound in the ductal cell population. Hypertranslation as a mechanism that governs stem cell differentiation has recently been demonstrated for cell types belonging to a few different organs [42][43][44][45] . Our study expands this concept to the pancreas and to a regenerative setting in vivo, suggesting that targeting initiation of translation could represent a conserved process stimulating differentiation and regeneration in multiple systems.
MNKs belong to the MAPK interacting protein kinase family and were identified in screens for interactors of the ERK and p38 MAP kinases 46,47 . MNKs primarily phosphorylate eIF4E and thereby have a context-dependent role in protein synthesis 19 . However, although our data indicated that CID661578 treatment increased protein synthesis, it did not affect the kinase activity of MNK2 in vitro. Instead, we observed that CID661578 increased the interaction between mRNAs, the cap-binding protein eIF4E and the scaffold protein eIF4G, which also binds to MNK1/MNK2 (ref. 13 ), together with an increase in phospho-eIF4E levels. Notably, previous studies in, for example, cancer and immune cells using cercosporamide observed reduced phosphorylation of eIF4E 16,48 . This discrepancy could be attributed to differential activities and upstream regulation of the MNKs. MNK2 contributes to a basal level of phospho-eIF4E, while the activity of MNK1 can be potentiated by upstream signaling to increase phospho-eIF4E in the absence of MNK2 (ref. 49 ). Consistently, reduced expression of MNK2 was previously reported to be associated with increased levels of MNK1 (ref. 50 ). Nevertheless, our studies using pan-MNK inhibitors support that the increased phospho-eIF4E following CID661578 depends on hyperactive MNK1. Our interpretation is that β-cell regeneration is increased when MNK2 is inhibited but is unaffected when both MNK1 and MNK2 are inhibited. Therefore, deleting/interfering with the less efficient kinase might open up for the more efficient kinase, leading to a net increase in phospho-eIF4E. Additionally, we observed that and JAK3 (c) kinase activity in vitro; n = 2 for each concentration tested. Data are presented as mean values ± s.e.m. d, Immunoblotting against eIF4G and eIF4E after an m 7 GTP pulldown assay in lysates of COLO 320HSR cells after 6-h treatment with DMSO, CID661578.6, 4EGI-1 or CID661578.6 together with 4EGI-1. For a loading control, 5% of the input was used. e, Immunoblotting against eIF4G and eIF4E after an m 7 GTP pulldown assay in rabbit reticulocytes treated with the indicated concentrations of CID661578.6. f, Immunoblotting against eIF4G and eIF4E after an m 7 GTP pulldown assay in lysates of PANC-1 cells treated with DMSO, CID661578 or CID661578.6 for 6 h. For a loading control, 1% of the input was used. g, Immunoblotting against eIF4G and FLAG-MNK2 after an immunoprecipitation (IP) assay with anti-FLAG in lysates of PANC-1 cells that were treated for 6 h with DMSO or CID661578. For a loading control, 1% of the input was used; IB, immunoblot. h, Immunoblotting against phospho-eIF4E (Ser 209; p-eIF4E), total eIF4E and actin in lysates of PANC-1 cells after 6-h treatment with DMSO, CID661578, cercosporamide, CGP57380 or eFT508. i, Quantification of the number of β-cells in 6 d.p.f. zebrafish larvae following β-cell ablation and treatment for 48 h with DMSO, CID661578, eFT508 or a combination of CID661578 and eFT508; n = 15 (control), n = 14 (CID661578), n = 17 (eFT508) and n = 15 (CID661578.6 + eFT508). A one-way ANOVA followed by Dunnett's multiple comparisons test was used to assess significance for i (**P = 0.0014 (control versus CID661578) and *P = 0.0283 (CID661578 versus CID661579 + eFT508)). Data are presented as mean values ± s.e.m. Experiments in d-h were repeated at least two times. j, Representative polysome tracings from optimized sucrose gradients of PANC-1 cells treated with DMSO, CID661578 or cercosporamide. k,l, Scatter plots showing log 2 fold changes for total mRNA (x axis) and polysome-associated mRNA (y axis) for the comparisons of CID661578 (k) and cercosporamide (l) to DMSO. Color codes indicate significantly affected mRNAs identified by anota2seq analysis.
CID661578 treatment drastically decreases the binding of MNK2 to eIF4G, resulting in increased interaction of eIF4E and eIF4G at the mRNA cap. Further, our polysome profiling analysis uncovered a total of 369 common translationally regulated mRNAs following treatment with CID661578 and cercosporamide. Coupled with the observed increase in phospho-eIF4E, this is one of the most extensive signatures of the effect of the phospho-eIF4E modification on the translatome described to date.
The zebrafish model has emerged as a powerful system for coupling large-scale screens with desired phenotypic outcomes in vivo.
The ability to screen in vivo rather than using in vitro culture systems offers the advantage of performing chemical screening in a setting where all of the tissues are present and can interact. Here, we report the identification of Mnk2 as the molecular target of CID661578, the most striking hit from a chemical screen for drivers of β-cell regeneration. Our results identified a previously unknown role for Mnk2 in β-cell neogenesis from cells residing within the ductal cell compartment of the pancreas, thereby paving the way for an alternative path to stimulate β-cell neogenesis, and hence regeneration, for the management of diabetes.
For CRISPR-Cas9 mutagenesis, a suitable guide RNA sequence targeting the N-terminal part of the mknk2b gene was designed using the CHOPCHOP web tool (https://chopchop.cbu.uib.no), resulting in the sequence 5′-CAATAA CTTACCAGGTCGGGCGG-3′. Then, IDT's Alt-R CRISPR-Cas9 system was used to create the mutation. Briefly, equal volumes of the custom-made Alt-R CRISPR RNA (crRNA) with a sequence of 5′-CAAUAACUUACCAGGUCG GGGUUUUAGAGCUAUGCU-3′ was annealed with the universal Alt-R trans-activating crRNA (tracrRNA) sequence (IDT) by incubating the solution for 3 min at 95 °C and cooling at room temperature for 15 min to a final concentration of tracrRNA:crRNA of 10 µM. Then, the tracrRNA:crRNA was incubated with the same concentration of Cas9 protein (IDT) at 37 °C for 10 min, and 1 nl of the mixture was injected in the one-cell stage to generate mknk2b-mutant zebrafish. Mutagenesis was confirmed in pooled injected embryos after DNA extraction and qPCR followed by melt curve analysis using the following primers: forward primer 5′-AGGATCCCATCTCCTTGAATCT-3′ and reverse primer 5′-CACCCACAGGAAATAGCTTGAT-3′. An identified founder line that had germline transmission of a 4-base pair deletion (5′-CGAC-3′) at the end of exon 3 was used for all experiments using mknk2b mutants. Genotyping of mutants was done by qPCR with melt curve analysis using the QuantStudio V1.2.4 software.
Chemicals were added to the E3 medium for larvae or facility water for adult zebrafish to a final concentration of 2 or 10 µM CID661578 (as specified in the figure legends), 2 or 10 µM CID661578.6 (OnTarget Chemistry), 500 nM cercosporamide (Tocris Bioscience), 800 nM 4EGI-1 (Tocris Bioscience) and 10 µM eFT508 (MedChemExpress). EdU was added to the E3 medium to a final concentration of 5 mM together with HEPES (10 mM) and developed using the Click-iT EdU Alexa Fluor 647 kit (Thermo Fischer Scientific). OPP was added to E3 medium to a final concentration of 100 µM and developed using a Click-iT Plus OPP Alexa Fluor 647 Protein Synthesis Assay kit (Thermo Fischer Scientific). 4-Hydroxytamoxifen was added to the E3 medium to a final concentration of 5 µM for 24 h.
We estimated glycemia in larvae by measuring free glucose that had not been intracellularly phosphorylated by hexokinases using a Glucose Colorimetric/ Fluorometric Assay kit (BioVision) in pools of four larvae for each time point and condition. Blood glucose measurements in adult zebrafish were performed using a standard glucometer (Freestyle, Abbott). Fish were fasted for 4 h, anesthetized in tricaine (Sigma-Aldrich) and decapitated for blood glucose measurements.
Immunofluorescence and confocal analysis. Immunofluorescence staining and confocal analysis of larvae and 1-month-old fish were performed as previously described 27 . Primary antibodies were used against green fluorescent protein (GFP; to amplify the GFP signal, 1:500; Aves Labs, GFP-1020), glucagon (1:200; Sigma, G6254), insulin (1:100; custom made by Cambridge Research Biochemicals) and tdTomato (1:500; MYBioSource, MBS448092). For experiments in juveniles, the insulin:GFP + area was measured on a flattened projection (average intensity). For the OPP intensity measurements, all larvae were imaged using the same parameters on a confocal microscope. Quantification was performed using the Fiji parameter (mean gray value). The mean gray value was measured from eight tp1:GFP + cells around the islet from a single plane for each larva using the same area as a reference for each cell, and the average of the mean gray value of the eight cells was calculated. All images were acquired with LAS X (v3.5.5.19976) software. The contrast was adjusted for visualization purposes in some experiments, in which case the same adjustments were made for all displayed images from the same experiment. The original unmodified pictures were used for analysis.
Chemical synthesis of CID661578 analogs. A detailed report that includes the steps used to synthesize all the analogs described in this study is provided in Supplementary Note 1.
Yeast chemical hybrid screen. The YChemH screen is based on the same principle as the classical yeast two-hybrid system. Synthesis of the chemical probe for the screen was performed by coupling a terminal carboxylic acid and the trimethoprim-PEG5-NH2 building block to an amide group of CID661578.6 (OnTarget Chemistry). The chemical probe was used as the bait and screened against two cDNA libraries in yeast, one from zebrafish embryos (18-20 h after fertilization) and the other from human islets (Hybrigenics). Survival of the yeast on medium lacking histidine was used to identify positive clones in the screen. All histidine + colonies were collected and screened for false positives, leading to 55 and 85 final positive clones in the zebrafish embryo and human islet libraries, respectively. The clones were sequenced to identify the corresponding prey, and a confidence score (A-E) was ascribed to each interaction based on two different levels of analysis. The confidence score was calculated as previously described 51 . Briefly, a local score considers the redundancy and independence of prey fragments together with the distribution of reading frames and stop codons in overlapping fragments. Then, a global score is calculated based on the interactions observed in all previous screens performed using the same library. The global score shows the probability of a hit being false positive. A hits were categorized as the most confident prey proteins, and E hits were the least probable. The full hit list report for both cDNA libraries is provided in Supplementary Note 2.
For the zebrafish Mnk2b binding experiments, full-length or truncated mknk2b (corresponding to the fragment of the human MKNK2 identified in the original screen) was cloned into the prey vector. Two different hook vectors were used to validate the binding of zebrafish Mnk2b to the bait: 1-N-LexA-eDHFR-C (original hook vector) and 2-N-eDHFR-LexA-C (reverse order hook vector).
Metabolomics. Metabolites were extracted from pools of 10 zebrafish larvae at 5 d.p.f. using a methanol-based extraction, and metabolite analysis was performed using liquid chromatography coupled with high-resolution mass spectrometry, as recently described in detail for zebrafish 27 . Xcalibur software was used for mass spectrometer data collection, and Sieve 2.2 was used for the chromatographic peak alignment. Differential regulation was examined based on t-test analysis in the Morpheus tool (https://software.broadinstitute.org/morpheus). Pathway analysis of the differentially regulated metabolites was performed using MetaboAnalyst 4.0 (ref. 52 ).
For the immunoprecipitation experiments, PANC-1 cells were plated in 10-cm dishes and transfected using the DharmaFECT Duo transfection reagent (Active Motif) with 2 µg of the MNK2-FLAG plasmid, obtained from Origene (NM_199054, RC216704). Cells were lysed with Pierce IP lysis buffer (87788), and 1 mg of protein was incubated for the immunoprecipitation with anti-FLAG (Sigma-Aldrich, 6 µg, F3165) overnight. As a negative control for the immunoprecipitation assay, we used the lysis buffer alone incubated overnight with anti-FLAG. The following day, the samples were incubated for 1 h with 30 µl of A/G magnetic beads (Thermo Fisher Scientific, 88803) and washed four times with NET-2 buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM MgCl 2 , 0.5% Nonidet P-40 and one tablet of protease inhibitors), and proteins were eluted from the beads with loading buffer (4× Laemmli sample buffer, Bio-Rad).
For the reticulocyte m 7 GTP pulldown assay, untreated rabbit reticulocyte lysates (Promega) were incubated with the specified concentrations of CID661578.6 for 1 h at 30 °C followed by m 7 GTP pulldown and western blotting, as described above.
Polysome profiling. To isolate fractions of efficiently translated mRNAs, polysome profiling, using an optimized sucrose gradient, was performed using a recently described optimized sucrose gradient 53 . Briefly, 4 × 10 6 PANC-1 cells were seeded in 15-cm plates 24 h before treatment. Cells were treated with cercosporamide (5 μM), CID661578 (40 μM) or DMSO for 6 h and lysed in a hypotonic lysis buffer. Aliquots of cytosolic lysate were collected from each sample for isolation of total cytosolic RNA. The remaining lysates were layered onto optimized sucrose density gradients (5%:34%:55% (wt/vol)), and samples were centrifuged at 4 °C at 35,000 r.p.m. for 2 h followed by fractionation. Fractions containing mRNAs bound to more than three ribosomes were collected in TRIzol reagent (Thermo Fischer Scientific) and pooled, allowing for isolation of efficiently translated polysome-associated mRNA.
RNA extraction was performed using the TRIzol reagent protocol (Thermo Fischer Scientific) followed by additional purification using an RNAeasy MinElute Cleanup kit (Qiagen). RNA quality was assessed using a Bioanalyzer 2100 with an RNA 6000 Nano kit (Agilent). Smartseq2 sequencing libraries were prepared as previously described 54 using 10 ng of mRNA as input. Libraries were prepared for total cytosolic and polysome-associated fractions from four biological replicates of cells treated with cercosporamide, CID661578 or DMSO. Libraries were pooled and sequenced on an Illumina NovaSeq 6000 platform using a 50-base pair paired-end setup.
RNA sequencing read quality was evaluated using MultiQC (1.7). Adapters and reads mapping to ribosomal RNA were removed using BBDuk (36.59) from the BBTools suite (http://jgi.doe.gov/data-and-tools/bb-tools/) before alignment to hg38 using HISAT2 (2.1.0) with default settings. Reads were summarized using RSubread (2.6.4) featureCounts with default settings and RefSeq gene definitions 55 . Genes with zero counts in at least one sample were removed, and the data were trimmed mean of the M values log 2 normalized. Changes in mRNA abundance (that is, congruent modulation in total and polysome-associated mRNA; downstream of altered transcription or mRNA stability), translation efficiency (that is, changes in levels of polysome-associated mRNA not paralleled by corresponding alterations in total mRNA) and buffering (that is, changes in total mRNA level but not polysome association and therefore predicted to result in unchanged protein levels) between cells treated with cercosporamide or CID661578 and DMSO were assessed using the anota2seq (1.14.0) algorithm 56 . The following thresholds were applied within the anota2seqRun function: maxPAdj = 0.15; deltaP = log 2 (1); deltaT = log 2 (1); deltaPT = log 2 (1.2); deltaTP = log 2 (1.2); maxSlopeTranslation = 2; minSlopeTranslation = −1; minSlopeBuffering = −2; maxSlopeBuffering = 1. Replicate was included in the anota2seq model using the 'batchVec' parameter to account for batch effects. Genes were classified according to their mode of regulation (mRNA abundance or translation) using the anota2seqRegModes function.
GO analysis was performed using the ClueGO plug-in (2.5.8) within CytoScape (3.8.2; https://cytoscape.org). Analyses of 5′ UTRs were based on the RefSeq curated sequences. Differences in 5′ UTR length, GC content and length-corrected fold energy between mRNA in each regulatory mode were assessed using a two-sided Mann-Whitney test.
Single-cell RNA-seq analysis. Single-cell RNA-seq data of adult zebrafish used for the analysis were downloaded from the Gene Expression Omnibus (GEO) under accession number GSE106121 and sample number GSM3032164 (ref. 17 ). The unique molecular identifier counts matrix was imported into R and processed using the Seurat R package version 3.5.1 (ref. 57 ). Low-quality cells with detected gene numbers less than 450 or higher than 2,200 along with mitochondrial genes were removed before downstream analysis. Subsequently, we performed principal component analysis and selected the top 21 significant principal components for dimensional reduction. A graph-based clustering method (Louvain) was used to cluster cells with a resolution of 0.5. Finally, we used the uniform manifold approximation and projection (UMAP; part of the Seurat 3.5.1 package algorithm) to display the relationships within and between different clusters. Enrichment analysis of differentially expressed genes was performed and visualized using the clusterProfiler package (3.10.1).
Dose-response assessment and in vitro kinase screen. Dose-response assessments and the in vitro kinase screen were performed by the International Centre for Kinase Profiling. All kinase assays were performed using a Multidrop 384 instrument at room temperature in a total assay volume of 25.5 μl. DMSO controls or acid blanks and 15 μl of the enzyme mix containing enzyme and peptide/protein substrate in buffer were added to plates containing 0.5 μl of compounds. The compounds were preincubated in the presence of the enzyme and peptide/protein substrate for 5 min before initiation of the reaction by addition of 10 μl of ATP (final concentration selected for each kinase at 5, 20 or 50 μM). The reactions were incubated for 30 min at room temperature before termination by the addition of 5 μl of orthophosphoric acid. The assay plates were collected onto P81 Unifilter plates by a PerkinElmer Harvester and air dried. The dry Unifilter plates were then sealed after the addition of MicroScint O and analyzed in PerkinElmer Topcount scintillation counters.
Pig islet aggregates and immunofluorescence experiments. All procedures involving pigs were performed according to the guidelines established by the Canadian Council on Animal Care. Donor pancreases were surgically removed from neonatal piglets of either sex (Swine Research and Technology Center, University of Alberta). Neonatal porcine islets were isolated and maintained in Ham's-F10 tissue culture medium (Sigma-Aldrich), as previously described 58 . For clarification, because neonatal pigs do not contain intact mature islets structures, the term neonatal porcine islets refers to aggregates of endocrine and exocrine tissue generated in culture following digestion of the pancreas. At the third day of cultivation, the incubation medium was switched to Ham's-F10 medium (without IBMX) supplemented with either 1 µM CID661578 (Sigma-Aldrich) or 1 µM cercosporamide (Tocris). The islets were treated for 5 d, and the medium was replaced with identical fresh medium every 48 h. Samples were collected from each condition, and immunohistochemical staining was performed as described previously 59 . Antibodies against the following proteins were used to stain the pig sections: insulin (1:5, DAKO, IR002), CK7 (3:100, DAKO, clone OV/TL 12/30) and MNK2 (1:200, Sigma-Aldrich, SAB2101483). Human pancreatic sections were stained using the same protocol as described for the porcine sections. All images were acquired with NIS-Elements (version 4.30) software. Human tissues were kindly provided by the Alberta Diabetes Institutes Islet Core, and ethical approval for the use of human samples was obtained from the University of Alberta's Human Research Ethics Board, protocol PRO00001416. Informed written consent was provided at the institutions where the organs were collected.
Human pancreatic ductal organoid culture. Human pancreatic exocrine tissue, obtained from the discarded fraction after human islet purifications using the Ricordi method from cadaveric organ donors with informed written consent, was processed to isolate ductal fragments and generate organoid cultures. Ethical approval for processing pancreatic samples from deidentified organ donors was granted by the Clinical Research Ethics Committee of Hospital de Bellvitge (PR030/22). Ductal fragments were embedded in GFR Matrigel and cultured in human organoid expansion medium 60 . After three to four passages, organoid expansion medium was replaced by a basic medium containing Advanced DMEM/F12, ITS-X (1×), heparin (0.1 mg ml -1 ), nicotinamide (0.1 mM), N-acetylcysteine (0.25 mM), FGF10 (0.1 µg ml -1 ), N2 supplement (1×) and B27 (1×), and organoids were treated with DMSO, CID661578 (50 µM) or cercosporamide (100 µM). The medium was renewed every other day, and the organoids were cultured under these conditions for 8 d. Following treatment, RNA was isolated from ductal organoids using the RNeasy minikit (Qiagen) followed by DNase I treatment (Invitrogen). The RNA was reverse transcribed with SuperScript III reverse transcriptase (Roche) and random hexamers, and qPCR was performed on a 7900 real-time PCR system (Applied Biosystems) using Power SYBR green (Applied Biosystems). Primers used for TBP included forward primer 5′-ATCCCTCCCCCATGACTCCCATG-3′ and reverse primer 5′-ATGATTACCGCAGGAAACCGC-3′, and primers used for INS included forward primer 5′-GCAGCCTTTGTGAACCAACA-3′ and reverse primer 5′-TTCCCCGCACACTAGGTAGAGA-3′.