A member of the wider group of ciliopathic disorders, autosomal dominant polycystic kidney disease (ADPKD) is caused primarily by mutations in the PKD1 or PKD2 genes, encoding the ciliary proteins polycystin 1 (PC1) and polycystin 2 (PC2), respectively (Bergmann et al., 2018). Pathophysiologically, the disease is characterised by the formation and expansion of fluid-filled cysts throughout the parenchyma of the kidney, and in some cases extra-renal organs such as the liver and pancreas. The variable but progressive growth of kidney cysts leads to structural remodelling of the kidney and eventually a significant impairment in renal function in most cases, with a median age of progression to end-stage renal disease (ESRD) of 58 years (Spithoven et al., 2014a). While ADPKD is classified as a rare disease with a prevalence of 3-5 per 10,000 (Willey et al., 2016; Solazzo et al., 2018; Willey et al., 2019), it is the most common genetic cause of renal failure, accounting for approximately 10% of all patients requiring renal replacement therapy (Spithoven et al., 2014b), and remains a significant public health burden. Tolvaptan is the only approved therapy targeting the underlying pathophysiology of ADPKD, and has demonstrated moderate efficacy in slowing renal cyst growth and preserving kidney function (Torres et al., 2012; Torres et al., 2017a; Torres et al., 2017b). However, the common tolvaptan side effect of polyuria and the rare but serious complication of idiosyncratic hepatic injury mean there remains an unmet need for safer and more effective therapies for ADPKD patients.

PC1 and PC2 are transmembrane proteins which form a multimeric complex in a number of cellular membranes, including the primary cilia, where they are thought to transduce extracellular signals through the regulation of a plethora of intracellular pathways. Disease-causing mutations in either gene product result in dysregulation of these intracellular signalling cascades, most notably in kidney epithelial cells, with a reduction in intracellular Ca²⁺ and elevation of cAMP leading to aberrant modulation of downstream pathways converging on defects in cell polarity, and increases in cell proliferation and extracellular fluid secretion (Bergmann et al., 2018; Gall et al., 2019). Beyond primary processes driving cyst initiation and expansion, secondary interstitial inflammation and fibrosis have been implicated in later stage ADPKD pathology and ESRD (Song et al., 2017; Zhang et al., 2020). As the mechanistic understanding of key pathogenic pathways has evolved, so too have opportunities for therapeutic intervention. Tolvaptan is an antagonist of the vasopressin type-2 receptor (V2R), and negatively impacts cyst growth through the repression of aberrantly elevated cAMP levels in collecting duct cells (Reif et al., 2011). In line with this primary mechanism of action however, tolvaptan also induces significant clinical aquaresis, which can present a barrier to initiating or maintaining treatment, particularly in younger patients with preserved estimated glomerular filtration rate who would most benefit from early and prolonged intervention (Müller et al., 2021; Calvaruso et al., 2023). While the mechanism is not well understood, there also remains a small but significant risk of tolvaptan-induced hepatic injury, which necessitates frequent and regular monitoring of liver function for the duration of treatment and further increases the burden on patients and healthcare providers (Müller et al., 2021). Beyond V2R antagonism, a number of diverse mechanisms for therapeutic intervention have been evaluated in randomised controlled clinical trials, including mTOR inhibition (Kim and Edelstein, 2012), tyrosine kinase inhibition (Tesar et al., 2017), AMP-activated protein kinase (AMPK) activation (Perrone et al., 2021; Brosnahan et al., 2022) and somatostatin receptor agonism (Griffiths et al., 2020). Unfortunately these efforts have yet to yield any further regulatory approvals, and thus there remains a significant need to identify additional tractable targets and mechanisms for the treatment of ADPKD.

Drug discovery for rare diseases suffers from a unique set of challenges, not least of which is an often incomplete understanding of disease pathogenesis from which to identify new therapeutic targets. Despite advances in unravelling the molecular aetiology of ADPKD, the complex processes involved give rise to a plethora of putative targets for prioritisation, validation and development. In order to address this problem using a target agnostic approach, we applied computational drug discovery approaches to interrogate publicly available gene expression datasets and an ADPKD-augmented rare disease knowledge graph to identify and prioritise clinical-stage candidate drugs for the treatment of ADPKD. We screened a focused library of drug predictions through preclinical disease model systems, and validated the antiparasitic drug mebendazole as an effective treatment in human cellular models and a genetic animal model of ADPKD. Using in silico and experimental approaches, we further probed the putative mechanisms of action of mebendazole in ADPKD. This work highlights the utility of combining computational and experimental methods in the discovery of new treatments and therapeutic targets for rare diseases such as ADPKD, which often suffer from complex or incompletely understood pathophysiological processes.

In order to uncover new therapeutic avenues for ADPKD, we made use of two independent but complementary drug prediction paradigms. The first, disease gene expression mapping (DGEM), is based on connectivity mapping (Lamb et al., 2006), and predicated on the concept that disease-induced perturbations in gene expression can be used to query libraries of drug-induced gene expression signatures in order to identify drugs which might induce therapeutic transcriptional changes in a given disease state. In order to implement this approach for ADPKD, we created disease signatures from publicly available datasets which represented early and late stage disease states, as well as disease progression, in both mouse models of disease and human patient tissue (Song et al., 2009; Pandey et al., 2011; Menezes et al., 2016). We then used these signatures to query the CMap drug database in order to identify drug signatures with the highest connectivity to disease states. While variations of this approach have proven successful in identifying new therapeutic candidates across numerous diseases (Musa et al., 2017), there are a number of limitations, including the limited size of the CMap library, and the exclusive use of human cancer cell lines for deriving its drug perturbation signatures. For these reasons, we also made use of an ADPKD-augmented knowledge graph, from which novel drug-ADPKD links were derived by a neural network trained to recognise graph characteristics of known treatments for similar diseases, via an algorithm termed degree-weighted path count (DWPC). We combined the output from the DGEM and DWPC predictive modules, and prioritised thirteen drugs for preclinical evaluation based on the strength of each prediction across computational prediction sets, novelty, as well as clinical and regulatory considerations which might enable rapid clinical translation. A full list of these prioritised predictions can be seen in Supplementary Table S1. In order to evaluate these drug predictions, we made use of a patient-derived in vitro model system which recapitulates three dimensional (3D) cyst growth in the correct genetic context, and has previously been used to interrogate novel therapeutic strategies (Dagorn et al., 2023). We evaluated drug predictions alongside tolvaptan in the presence and absence of the cyst swelling stimulus desmopressin (ddAVP), and determined that three predicted drugs—demeclocycline, cloperastine and mebendazole—demonstrated dose-dependent inhibition of cyst growth in both ddAVP-stimulated and unstimulated conditions, while tolvaptan only exhibited dose-dependent effects on cyst growth in the presence of ddAVP, as expected based on its mechanism of action (Reif et al., 2011) (Figure 1). We next used information and data generated during the prediction process to gain insights into the potential therapeutic mechanisms of these active molecules.

Demeclocycline is a tetracycline antibiotic which was first described over half a century ago. It was predicted as a potential treatment for ADPKD via interrogation of an ADPKD-enriched knowledge graph, and upon further investigation, it appeared this prediction was driven primarily via links to vasopressin signalling (Figure 2A). Indeed, aside from its use in the treatment of susceptible bacterial infections, demeclocycline has been employed for several decades in the treatment of inappropriate antidiuretic hormone syndrome; a feature shared with the standard of care in ADPKD, tolvaptan. In this regard, demeclocycline acts as a physiological antagonist of the V2R, reducing the abundance of adenylate cyclases downstream of V2R, and subsequently leading to a reduction in cAMP-dependent aquaporin 2 transcription (Kortenoeven et al., 2013). The intersection of this mechanism with the direct effects of tolvaptan on the V2R, and the known involvement of this pathway in ADPKD pathogenesis have led some to speculate that demeclocycline may indeed be an efficacious treatment (van Hastel and Torres, 2017), though until now definitive evidence of this hypothesis has been lacking.

The antitussive agent cloperastine has several known targets, including histamine, as well as the sigma receptors σ1 and σ2 which are thought to be responsible for its primary antitussive effect (Gregori-Puigjané et al., 2012). In the current study, cloperastine was predicted using transcriptomic drug matching, thus we used gene enrichment analysis to interrogate the gene expression overlap between drug signatures and disease signatures with the highest connectivity scores, in order to explore pathways which might be relevant for its treatment effect in ADPKD. The highest ranked consensus pathways for the cloperastine-ADPKD gene expression overlap were associated with amino acid metabolism, peroxisome proliferator-activator receptor (PPAR) signalling, nuclear factor κB (NF-κB) and nuclear receptors (Figure 2B). These pathways have been linked to varying degrees to the disease biology of ADPKD (Harris and Torres, 2014), however the link to the primary pharmacology of cloperastine was initially less obvious. Interestingly, a molecular target recently associated with cloperastine is a methyltransferase, protein arginine methyltransferase 5 (PRMT5) (Prabhu et al., 2023). PRMT5 regulates gene expression via the dimethylation of arginine residues in histone and non-histone protein targets, and has been associated with the direct control of both NF-κB (Wei et al., 2013) and PPAR (Huang et al., 2018) transcriptional regulation, which align with the top consensus pathways identified here in the cloperastine-ADPKD gene expression overlaps. Furthermore, PRMT5 has been associated with the regulation of cell cycle and immune cell invasion in the context of cancer (Gu et al., 2012; Abe et al., 2023), which intersect with the lower ranked cloperastine-ADPKD pathways described here. Whether PRMT5 represents a new therapeutic target for ADPKD remains to be determined, however there are a number of clinical candidates evaluating the potential of this target for oncology indications (Zheng et al., 2023). Unfortunately, in addition to many potentially therapeutic targets, cloperastine is also a potent inhibitor of human ether-a-go-go (hERG) and induces QT prolongation in vivo, thus demonstrating undesirable proarrhythmic potential (Takahara et al., 2012). While there are obvious limitations in the repurposing of antibiotics or proarrhythmic drugs for a chronic indication such as ADPKD, the ability of our methods to predict demeclocycline and cloperastine as active agents in line with known pharmacology in the case of demeclocycline, and potentially new pharmacology in the case of cloperastine validates the potential of our in silico methodology to identify known and novel treatment approaches.

Mebendazole is a broad spectrum anthelmintic agent that has been in clinical and veterinary use for over 50 years. It is believed to mediate its therapeutic effect via binding to parasitic β-tubulin and subsequent inhibition of microtubule polymerisation, thus disrupting key cellular processes leading to paralysis and death of the parasite (Lacey, 1988). In addition to its affinity for parasitic β-tubulin, mebendazole also inhibits microtubule polymerisation in mammalian cells, and moreover demonstrates affinity for a number of kinase targets (Nygren et al., 2013; Issa et al., 2015). These properties, combined with an extensive history of safe human use, have led to significant interest and advancements in the repurposing of mebendazole for oncological indications (Sasaki et al., 2002; Bai RY. et al., 2015; Choi et al., 2021; Gallia et al., 2021; Mansoori et al., 2021; Williamson et al., 2021; Hegazy et al., 2022). Like cloperastine, mebendazole was also predicted as a putative treatment for ADPKD using transcriptomic drug matching methods, and so we again performed gene enrichment analysis of the drug-disease gene expression overlap using the most highly connected disease contrasts in order to gain insight into the potential mechanisms underlying the therapeutic effect in this case (Figure 2C). Key pathways identified from this gene expression overlap included those related to cell cycle, cancer, inflammation and metabolism. These pathways are all highly relevant for both ADPKD and cancer, which are known to share many molecular similarities (Seeger-Nukpezah et al., 2015), and suggest that mebendazole may have a similar mode of action in the treatment of both disease entities. Due to the potent effects of mebendazole on human cellular models of ADPKD, and its previously demonstrated potential as a clinical repurposing candidate in oncology, we decided to focus further efforts on in vivo validation and mechanism-based studies to understand if mebendazole and/or its targets might also represent a therapeutic strategy for ADPKD.

In order to assess the efficacy of mebendazole against ADPKD-relevant phenotypes in vivo, we utilised the P18 iKsp-Pkd1 ^del mouse model, which employs a tamoxifen-inducible cadherin promoter to selectively disrupt Pkd1 expression in the kidney epithelium at postnatal day (PND) 18, and has previously been employed to evaluate new therapeutic strategies (Dagorn et al., 2023; Song et al., 2023). While perinatal inactivation of Pkd1 in this genetic background produces a rapid and severe model of kidney cystogenesis and functional decline over a compressed timeline, inactivation of Pkd1 at PND18 induces an adult-onset, slowly progressing model of cystic kidney disease which leads to renal failure over the course of several months, and offers the opportunity to evaluate treatment strategies over a chronic time window (Leeuwen et al., 2007; Leonhard et al., 2016). After disruption of the Pkd1 locus at PND18, mice were treated from PND42 with once daily (QD; 10, 20 or 30 mg/kg) or twice daily (BID; 10 or 15 mg/kg) mebendazole via oral gavage, or tolvaptan dosed in medicated food (0.1% w/w) until study termination (Figure 3A). The study was terminated at PND110, 1 day after >50% of animals in the untreated control group had been euthanised due to end-stage renal disease (ESRD), defined as a blood urea level >20 mmol/L. Weekly body weight measurements indicated that mebendazole doses of 10 mg/kg QD, 20 mg/kg QD, 10 mg/kg BID and 15 mg/kg BID were well tolerated in these mice, however some animals treated with 30 mg/kg QD experienced weight loss, which necessitated a dosing holiday and reduced dose of 25 mg/kg QD in this group from PND59 (Supplementary Figures S1A, B). Despite this intervention, only 14/22 animals made it to the study endpoint or termination in this treatment group, compared with almost all animals from the remaining treatment groups, suggesting the maximum tolerated exposure of mebendazole was exceeded in this group (Supplementary Figure S1C). Over the course of this study, P18 iKsp-Pkd1 ^del mice developed highly cystic kidneys which were on average 7-fold the normalised weight of Pkd1 wildtype mouse kidneys (Figures 3B–E). Twice daily oral administration of mebendazole at 15 mg/kg significantly alleviated kidney cystic load by 33% (Figure 3C), while twice daily administration of mebendazole at 10 mg/kg and 15 mg/kg, significantly reduced kidney cystic grade as assessed by histopathological scoring (Figure 3D). In addition to effects on cystic load, twice daily administration of mebendazole at 10 mg/kg and 15 mg/kg also significantly ameliorated increased kidney weight by 32% and 42%, respectively (Figure 3E). As a further measure of structural disease progression, we took advantage of the chronic nature of this model system to evaluate kidney volume longitudinally using a non-invasive ultrasound imaging approach. On PND75, kidney volume in Pkd1 wildtype mice was increased by an average of 47% compared with pre-dosing values taken on PND41, whereas in the same timeframe, kidney volume in vehicle treated Pkd1 KO mice increased by an average of 291%, consistent with the development of enlarged cystic kidneys by the midpoint in this study (Figure 3F). Consistent with ameliorative effects on terminal kidney structure, twice daily administration of 10 mg/kg and 15 mg/kg mebendazole significantly reduced kidney volume expansion at this point in the study compared with vehicle treated Pkd1 KO mice, from an average of 291% to 132% and 177%, respectively (Figure 3F). By PND94, only twice daily administration of 15 mg/kg had a significant effect on reducing kidney volume in Pkd1 KO mice, however it should be noted that by this late stage of the study, animals in the comparator vehicle-treated Pkd1 KO group had begun to enter ESRD and drop out of the study, potentially reducing statistical power to detect differences in kidney volume between treatment groups (Supplementary Figure S2). In order to evaluate the effects of mebendazole on preserving kidney function, we compared blood urea levels in each animal from terminal samples taken when animals reached ESRD, or at the study terminus, whichever came earlier. Terminal blood urea levels were an average of 6.7 mmol/L in Pkd1 WT mice, and 18.7 mmo/L in vehicle treated Pkd1 KO mice, indicating substantial functional decline across the course of the study (Figure 3G). Twice daily treatment with 10 mg/kg and 15 mg/kg mebendazole significantly reduced blood urea levels by 48% and 41% to 9.8 mmol/L and 11.0 mmol/L, respectively, compared with vehicle treated Pkd1 KO mice (Figure 3G), suggesting a significant rescue of renal functional decline in mebendazole-treated Pkd1 KO mice, while kidney survival analysis further confirmed a significant treatment effect (Mantel-Cox test p < 0.01) (Figure 3H).

These findings demonstrate that mebendazole treatment ameliorates ADPKD phenotypes in a chronic, slowly-progressing mouse model of disease, and further suggest that increasing the frequency of mebendazole administration to deliver the same dose over two administrations in order to maintain constant levels of exposure improves both efficacy and tolerability. In contrast to mebendazole, the positive control and current standard of care for ADPKD patients, tolvaptan, surprisingly failed to produce significant effects on any measured endpoints. It should be noted that constant exposure to therapeutic levels of tolvaptan is required, which can be difficult to control by using the standard approach of tolvaptan dosing via medicated food. Based on the significant effects of mebendazole on kidney cyst growth in human cellular models and an in vivo model of ADPKD, we next focused on determining the likely mechanism behind this therapeutic effect.

Given the multifactorial nature of ADPKD pathophysiology, and the polypharmacological nature of mebendazole, we made use of a recently described computational framework called the multiscale interactome to understand how mebendazole may treat ADPKD at the molecular level (Ruiz et al., 2021). This framework is predicated on the concept that drugs often do not treat diseases via direct modification of disease-associated proteins, but via a complex propagation of signals through protein-protein interactions and biological functions and processes. When these complex drug-disease relationships are represented as diffusion profiles, they can be used to understand how a given drug might impact a given disease, particularly in cases where a direct molecular link is not immediately obvious.

To understand how this framework performs in describing drug-disease relationships for ADPKD, we initially focused on the well understood link between tolvaptan and ADPKD (Figure 4A). In line with known drug pharmacology and disease biology, the tolvaptan-ADPKD diffusion profile identifies V2R as the sole drug target for tolvaptan, and also identifies key proteins known to be perturbed in the ADPKD disease state, including V2R itself, as well as polycystin-2, angiotensinogen, cystic fibrosis transmembrane conductance regulator (CFTR) and epidermal growth factor (EGFR). While V2R, the molecular target of tolvaptan, is directly linked to PKD pathophysiology (Wang et al., 2008), the diffusion profile also describes the interaction of V2R with other first order ADPKD-associated proteins via additional entities or biological functions, adding more nuance to the drug-treatment picture. There are interactions described in the diffusion profile between V2R and the process of cell proliferation, which is itself dysregulated in ADPKD by the polycystins (Grimm et al., 2006), the renin-angiotensin-aldosterone system, dysfunction of which is both a downstream consequence of, and direct contributor to ADPKD (Hian et al., 2016), and components of the G protein-coupled receptor signalosome, which regulate ADPKD-associated proteins such as CFTR and EGFR (Hama and Park, 2016; Sussman et al., 2020). These findings demonstrate the utility of the multiscale interactome approach in describing complex drug-disease relationships between two well described entities.

We next utilised the same approach to gain insight into the unknown drug-disease relationship between mebendazole and ADPKD (Figure 4B). As anticipated, disease-associated proteins in the mebendazole-ADPKD diffusion profile share significant overlap with those identified in the tolvaptan-ADPKD diffusion profile described above, however the drug targets of mebendazole and their downstream associations are markedly different to those of tolvaptan. Mebendazole is a well known microtubule depolymerising agent, and in line with this mode of action, two tubulin proteins are herein described as drug targets, Tubulin beta 4B, and Tubulin alpha 1A. The diffusion profile makes a direct connection between these microtubule proteins and the ADPKD-associated proteins EGFR and CFTR, trafficking of which has indeed previously been linked to the state of the microtubule network (Morris et al., 1998; Liu et al., 2012), and may explain in part the downstream effects of mebendazole-induced microtubule destabilisation on ADPKD-relevant networks. Aside from its microtubule disrupting action, mebendazole is also an inhibitor of a number of kinases, and accordingly is linked in the diffusion profile via two of its known protein kinase targets, vascular endothelial growth factor receptor 2 (VEGFR2) (Dakshanamurthy et al., 2012) and ABL1 (aka Abl) (Nygren et al., 2013), to phosphorylation processes also regulated by EGFR. VEGF signalling has previously been linked with ADPKD pathophysiology (Tao et al., 2007), and herein the diffusion profile suggests this could be mediated in part via the regulation of apoptosis, which is itself a key process known to be dysregulated in the disease state (Zhou and Li, 2015). On the other hand, Abl has not been explicitly linked to ADPKD previously, although intriguingly it is linked both here in the diffusion profile and experimentally elsewhere to direct regulation of EGFR (Tanos and Pendergast, 2006). Using a multiscale interactome approach we have identified a number of potential molecular targets and functions that may drive the therapeutic efficacy of mebendazole in ADPKD, which we next sought to explore using experimental approaches.

Mebendazole is part of the benzimidazole class of anthelmintics, which itself is a member of a wider family of structurally diverse compounds which bind to the colchicine-binding site (CBS) of β tubulin, inhibiting the polymerisation of microtubules. Given its known pharmacology as a microtubule inhibitor, and its predicted link to ADPKD pathophysiology via tubulin protein targets uncovered by the multiscale interactome approach, we evaluated the potency and efficacy of a range of compounds with affinity for the CBS in the 3D phenotypic assay, to establish if the therapeutic effect of mebendazole in ADPKD is indeed being driven primarily via this mechanism. We determined that in addition to mebendazole, a number of benzimidazole anthelmintics (Figure 5A) and CBS agents from the wider family (Figure 5B) demonstrate dose-dependent inhibition of kidney cyst growth. Given the additional link of mebendazole to ADPKD via two protein kinase targets predicted by the multiscale interactome approach, we assessed the ability of mebendazole to inhibit a diverse panel of kinases, which was enriched with additional kinase targets previously known or predicted to be inhibited by mebendazole, or previously implicated in ADPKD. Utilising a cell-free assay platform, we observed that mebendazole inhibited 15 out of 94 tested kinases by >50% at 10 µM (Figure 5C). Both of the kinase targets predicted in the mebendazole-ADPKD diffusion profile, VEGFR2 (KDR) and Abl, were inhibited by mebendazole at this concentration, while overall four targets with known links to ADPKD pathophysiology were represented: VEGFR1 (Flt1) and VEGFR2 (KDR) (Bello-Reuss et al., 2001; Tao et al., 2007), cyclin dependent kinase 1 (CDK1) (Zhang et al., 2021) and Met (Qin et al., 2010). While the consistent activity of a wide range of microtubule polymerisation inhibitors on cyst growth in vitro, which in the case of the benzimidazole anthelmintics broadly correlates with their biochemical potency in inhibiting mammalian tubulin polymerisation (Lacey, 1988), and includes molecules such as albendazole with no appreciable kinase activity (Nygren et al., 2013), suggests that mebendazole acts primarily via microtubule polymerisation inhibition to inhibit cyst growth in ADPKD, there remains an intriguing possibility that inhibition of several protein kinase targets also contributes to the therapeutic effect of mebendazole observed in preclinical ADPKD models. In summary, in silico modelling and pharmacological and biochemical data support the notion that the primary effects of mebendazole in ADPKD are likely driven via binding to the CBS of β tubulin, inhibition of microtubule polymerisation, and disruption of downstream microtubule dynamics, with a possible contribution via the inhibition of select protein kinase targets.

Using transcriptomic and machine-learning drug discovery approaches, we predicted a number of existing drugs which may have therapeutic potential in ADPKD, and subsequently validated several of these predictions in relevant disease model systems. Among these predictions, the anthelmintic drug and anti-cancer repurposing candidate mebendazole ameliorated cyst growth in multiple human cellular models of ADPKD, and disease-relevant phenotypes in a slowly progressing mouse model of ADPKD, inhibiting both kidney cyst growth and kidney size expansion, while rescuing declining kidney function. In silico insight and mechanistic studies revealed that the anti-cystic effect of mebendazole in ADPKD is likely driven primarily by its inhibitory effect on microtubule polymerisation, with a potential contribution from the inhibition of protein kinase targets known or predicted to be involved in disease pathophysiology.

In addition to mebendazole, we determined that a number of well known microtubule depolymerising agents attenuated cystic growth in a human cellular model of ADPKD. Aside from a fundamental role in cell division, microtubules also form the core axoneme structure of the cilia, and facilitate essential functions such as cilia assembly/disassembly, and intraflagellar transport of ciliary proteins (Conkar and Firat-Karalar, 2021). Previous research has highlighted the importance of the microtubule cytoskeleton in kidney cyst formation and progression, and the impact of targeting microtubule dynamics and function on disease pathophysiology. As early as 1994, researchers using paclitaxel observed that hyperstabilisation of microtubules was sufficient to blunt kidney cyst expansion in vitro and in vivo (Woo et al., 1994), and more recently it was shown that the microtubule stabilising compound 1-Indanone was able to rescue ADPKD phenotypes by correcting abnormal cilia length and aberrant cilia signalling pathways (Li et al., 2023). A recent phenotypic screening effort identified a number of microtubule stabilisers as well as destabilisers as anti-cystic agents in murine and human cellular models of ADPKD, suggesting that depolymerisation as well as hyperstabilisation might both be viable therapeutic strategies (Asawa et al., 2020). The diverse functions of the microtubule network are facilitated in part through extensive posttranslational modifications of its α and β tubulin subunits (Wloga et al., 2017), and recent findings have highlighted the potential of targeting these modifications in polycystic kidney disease. Acetylation of α-tubulin is associated with stable microtubule polymers, and is primarily regulated by the opposing actions of the acetylase α-tubulin acetyltransferase 1, and the deacetylases histone deacetylase 6 (HDAC6) and sirtuin 2 (SIRT2) (Li and Yang, 2015). Loss of polycystin-1 is associated with an increase in HDAC6 (Liu et al., 2012) and SIRT2 (Zhou et al., 2014) expression, and a decrease in acetylated α-tubulin (Zhou et al., 2014), while inhibition of either deacetylase has been shown to augment α-tubulin acetylation levels and ameliorate ADPKD-relevant phenotypes (Zhou et al., 2014; Cebotaru et al., 2016; Yanda et al., 2017). While non-selective targeting of microtubule dynamics has the potential to induce undesirable toxicity due to the role of microtubules in essential cellular processes (Dumontet and Jordan, 2010), selective targeting of dysfunctional elements of the network such as this might open up the possibility of more tractable treatment opportunities for ADPKD. Indeed, while targeting HDAC6 has shown promise in preclinical models of ADPKD as described above, HDAC6-null mice with hyperacetylated α-tubulin are viable and develop normally (Zhang et al., 2008), suggesting that disease-selective targeting of the microtubule network might be possible to achieve while limiting undesirable toxicity.

Numerous growth factors and their respective protein kinase-mediated signalling pathways have been implicated in ADPKD, including among others epidermal growth factor, insulin growth factor, platelet-derived growth factor and vascular endothelial growth factor (VEGF) (Formica and Peters, 2020). While interventions targeting a number of these pathways have demonstrated promise in preclinical models of polycystic kidney disease, only the Src/Bcr-Abl inhibitor bosutinib and the multi-kinase inhibitor tesevatinib have thus far progressed into randomised clinical trials (Zhou and Torres, 2023). Using the multiscale interactome framework, we predicted that in addition to tubulin, mebendazole may also be acting via kinase inhibition to elicit a treatment response in ADPKD. To explore this prediction experimentally, we screened mebendazole against a kinase panel enriched with targets predicted or known to be inhibited by mebendazole, or important in ADPKD, and determined that in addition to the predictions VEGFR2 and Abl, mebendazole also inhibited VEGFR1, Met and CDK1, which have all been implicated in ADPKD. These findings corroborate existing literature - which is drawn upon by the multiscale interactome method to make inferences - by replicating direct inhibition of VEGF2 and Abl by mebendazole (Dakshanamurthy et al., 2012; Issa et al., 2015; Ariey-Bonnet et al., 2020), and also extend these studies by implicating mebendazole in the inhibition of additional ADPKD-relevant kinase targets it has not previously been demonstrated to interact with. A number of studies have linked pro-angiogenic VEGF signalling to ADPKD progression, although the picture is admittedly complex: suppression of either VEGFR1 or VEGFR2 expression suppressed cyst growth in the Han:SPRD model of polycystic kidney disease (Tao et al., 2007), but paradoxically anti-VEGF treatment augmented cyst growth and accelerated functional decline in the same model (Raina et al., 2011). Almost 30 years ago, it was observed that renal-cyst lining cells of PKD patients aberrantly expressed both the mitogenic hepatocyte growth factor and its receptor tyrosine kinase Met (Horie et al., 1994), while further mechanistic studies confirmed hyperactivation of Met signalling in Pkd1 models, pharmacological inhibition of which rescued disease phenotypes (Qin et al., 2010). Finally, a recent study identified the cell cycle regulator CDK1 as a key promoter of early cystogenesis in ADPKD, and crucially demonstrated that genetic ablation of this gene inhibited cell cycle progression and ameliorated kidney disease phenotypes in vivo (Zhang et al., 2021). Whether interactions with any or all of these targets and signalling pathways in addition to the microtubule network is essential for the activity of mebendazole in ADPKD requires further experimental validation, however a synergistic effect of mebendazole on tubulin and protein kinases has been hypothesised to underlie its potent anticancer activity (Nygren et al., 2013; Bai R.-Y. et al., 2015), and may make biological sense in the current context given the similarities in signalling pathways between ADPKD and cancer (Seeger-Nukpezah et al., 2015), and the representation of cancer-related pathways in the mebendazole-ADPKD gene expression overlap. A polypharmacological mode of action may indeed be desirable in a multifaceted disease such as ADPKD.

This study builds on previous work using computational methods to identify new drugs and therapeutic candidates for ADPKD. Malas and colleagues utilised a combined transcriptomic and cheminformatic approach to prioritise and ultimately validate several novel compounds in a 3D cystic assay (Malas et al., 2020). Their approach involved the creation of a disease-stage specific transcriptomic signature, from which candidate genes were identified and linked to molecules through publicly available drug databases. Targets and molecules were ultimately filtered for validation based on biological and chemical insights, and the potential for clinical translation. Earlier this year, Wilk et al., 2023 applied a similar transcriptomic approach to us, in that case making use of publicly available transcriptomic datasets to create Pkd2-specific ADPKD disease signatures, from which signature reversion was sought from the Library of Integrated Network-based Cellular Signatures (LINCs) drug signature database in order to identify drug repurposing candidates. While one group has previously made use of a knowledge graph-based approach to prioritise preclinically active compounds with the highest chance of clinical translation (Malas et al., 2019), to our knowledge, the current study provides the first combined application of transcriptomic and machine-learning approaches to identify and prioritise putative treatments for ADPKD, and further deconvolute potential mechanisms of action for experimental validation.

In summary we report, using computational, in vitro and in vivo approaches, that the anthelmintic drug mebendazole ameliorates disease-relevant phenotypes in cellular and animal models of ADPKD. We further show that this effect is likely primarily due to the inhibitory effect of mebendazole on the polymerisation of microtubules, which underlie cellular processes important in ADPKD, including cell proliferation, transport, and cilia signalling, and extends previous work linking the importance of the microtubule network to ADPKD pathophysiology. We also describe the inhibitory profile of mebendazole on known and novel protein kinase targets, some of which have previously been implicated in ADPKD, suggesting mebendazole may be acting via polypharmacology to impact disease mechanisms. We acknowledge that further experimental efforts will be required to confirm the actions of mebendazole on these putative targets in relevant disease model systems. It would be particularly informative to investigate these mechanisms in dedicated in vivo studies, where the effects of mebendazole on a wider range of ADPKD-relevant cell types and phenotypes could be evaluated. Notwithstanding these limitations, this work supports the combined role of in silico and experimental approaches in the discovery of new treatments and therapeutic pathways for rare diseases with complex or poorly understood pathophysiology.