Abstract
Autosomal dominant polycystic kidney disease (ADPKD) is a rare genetic
disorder characterised by numerous renal cysts, the progressive
expansion of which can impact kidney function and lead eventually to
renal failure. Tolvaptan is the only disease-modifying drug approved
for the treatment of ADPKD, however its poor side effect and safety
profile necessitates the need for the development of new therapeutics
in this area. Using a combination of transcriptomic and machine
learning computational drug discovery tools, we predicted that a number
of existing drugs could have utility in the treatment of ADPKD, and
subsequently validated several of these drug predictions in established
models of disease. We determined that the anthelmintic mebendazole was
a potent anti-cystic agent in human cellular and in vivo models of
ADPKD, and is likely acting through the inhibition of microtubule
polymerisation and protein kinase activity. These findings demonstrate
the utility of combining computational approaches to identify and
understand potential new treatments for traditionally underserved rare
diseases.
Keywords: autosomal dominant polycystic kidney disease, rare diseases,
gene expression profiling, machine learning, drug discovery, drug
repositioning, mebendazole, tubulin modulators
Introduction
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 ([44]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
([45]Spithoven et al., 2014a). While ADPKD is classified as a rare
disease with a prevalence of 3-5 per 10,000 ([46]Willey et al., 2016;
[47]Solazzo et al., 2018; [48]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 ([49]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 ([50]Torres et al., 2012;
[51]Torres et al., 2017a; [52]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 Ca^2+ 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 ([53]Bergmann et al., 2018; [54]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 ([55]Song et al.,
2017; [56]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
([57]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 ([58]Müller et al., 2021; [59]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 ([60]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 ([61]Kim and Edelstein, 2012), tyrosine kinase inhibition
([62]Tesar et al., 2017), AMP-activated protein kinase (AMPK)
activation ([63]Perrone et al., 2021; [64]Brosnahan et al., 2022) and
somatostatin receptor agonism ([65]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.
Materials and methods
Gene expression datasets and differential expression analysis
Gene expression files were downloaded from GEO under accession numbers
[66]GSE7869 ([67]Song et al., 2009), [68]GSE24352 ([69]Pandey et al.,
2011) and [70]GSE72554 ([71]Menezes et al., 2016). These represented
the highest quality datasets available to model ADPKD gene expression
at the time this study was conducted, representing both mouse and human
PKD1-driven ADPKD signatures across different stages of kidney cyst
development.
Differential expression analyses were carried out on these datasets
using QIAGEN Omicsoft Suite to create disease stage-specific gene
signatures, a summary of which can be seen in [72]Table 1 below.
Following robust multi-array average normalisation ([73]Irizarry et
al., 2003), differential expression was performed with Limma
([74]Ritchie et al., 2015), and p-values were re-adjusted with the
Benjamini-Hochberg correction for multiple testing ([75]Benjamini and
Hochberg, 1995). Disease vs. normal gene signatures were generated by
comparing differentially expressed genes from cystic kidney tissue to
healthy kidney tissue. We used a human dataset to compare cystic tissue
isolated from ADPKD patients to healthy renal cortical tissue from
nephrectomised kidneys ([76]Song et al., 2009), and used mouse datasets
to compare cystic kidneys isolated from early-onset (MGI:2182840)
([77]Pandey et al., 2011) and late-onset (MGI:3612341) ([78]Menezes et
al., 2016) mouse models of ADPKD to healthy kidneys isolated from
control littermates. In the late-onset mouse model dataset, disease vs.
normal gene expression signatures were further segregated by sex, as
this characteristic was determined to be the largest secondary source
of variation in gene expression by the original authors. ADPKD
progression signatures were generated by comparing differentially
expressed genes from late cystic tissue to early or minimally cystic
tissue. We used the human dataset to compare cystic kidney tissue
isolated from ADPKD patients to minimally cystic kidney tissue from the
same patients, and used the early-onset mouse model dataset to compare
cystic tissue from kidneys isolated late in the disease stage (E17.5)
to cystic tissue from kidneys isolated earlier in the disease stage
(E14.5).
TABLE 1.
Summary of ADPKD gene expression data.
GEO dataset Species Model Disease signature (contrast) description
ID Type Cystic stage Sex
[79]GSE7869 [80]Song et al. (2009) Human Patient-derived cystic kidney
tissue. Cysts of different sizes + minimally cystic samples. Control
normal tissue from nephrectomised kidneys 175 Disease vs. Normal Late +
Early Unknown
176 Progression Late vs. Early Unknown
[81]GSE24352 [82]Pandey et al. (2011) Mouse Pkd1-null ([83]MGI:2182840)
embryonic kidneys. Model rapidly develops progressive kidney cysts from
embryonic day (E) 15.5d 203416 Disease vs. Normal E17.5 Unknown
203343 Progression E17.5 vs. E14.5 Unknown
[84]GSE72554 [85]Menezes et al. (2016) Mouse Pkd1-conditional null
([86]MGI:3612341) adult kidneys. Later-onset model with Pkd1
conditionally deleted at postnatal day (PND) 40 203348 Disease vs.
Normal PND102-210 Female
203349 Disease vs. Normal PND102-210 Male
[87]Open in a new tab
Disease gene expression matching
A variation of connectivity mapping ([88]Lamb et al., 2006) was applied
to match ADPKD disease signatures to potential drug treatments.
Briefly, drug signatures derived from cell line treated expression
profiles were downloaded from clue.io and processed according to the
supplementary methods file from the connectivity mapping study
([89]Lamb et al., 2006). The subsequently generated drug signatures
were matched against the top 500 significantly upregulated and
downregulated genes from Disease vs. Normal signatures and Progression
signatures (c.f.[90]Table 1) using the connectivity score. The
connectivity score is an enrichment metric based on the
Kolmogorov-Smirnov statistic to measure how closely the gene expression
signature of a disease aligns with the gene expression patterns found
in a set of drug signatures. It does so by examining where the up- and
down-regulated genes from the disease are located in a ranked list of
genes representing drug treatment effects on cell gene expression. In
these drug signatures, the most relevant genes are typically found at
the very top or bottom of the ranked list. A higher connectivity score
indicates that the up- and down-regulated genes from the disease are
concentrated in these high-priority areas, suggesting a stronger
alignment between disease and drug, and a possible therapeutic
connection. A caveat of the original connectivity mapping approach is
that it only considers one way to represent the input disease
signature, i.e., systematically considering an arbitrary number of up-
and downregulated genes. However, sometimes, downregulated genes drive
the disease phenotype more than upregulated genes and vice versa. To
that end, we developed disease gene expression matching (DGEM) which
incorporates the original connectivity mapping method, as well as an
additional module that extracts an optimal configuration for the input
disease signature, i.e., the configuration which yields the higher
connectivity score for a set of control drug signatures. This set of
control drug signatures is selected to represent an ideal therapeutic
gene expression profile to be matched against the input disease
signature. Note that the standard connectivity mapping routine is not
modified, only the disease input signature is adjusted to better model
the disease expression according to the target therapeutic control
signature.
In this study, two configurations were investigated: 1. The top 250 up-
and 250 downregulated genes and 2. The top 500 differentially expressed
genes regardless of direction. Then for each disease contrast (c.f.
[91]Table 1), the optimal configuration was obtained by running
connectivity mapping and assessing which configuration produced the
highest connectivity score for the control drug(s). In this case,
metformin was selected as a sole control treatment from which to
benchmark, as it had at the time demonstrated clinical efficacy against
ADPKD relevant functional endpoints ([92]Pisani et al., 2018), and
importantly, was represented within the CMap library, unlike other
potential clinical controls such as tolvaptan. For each disease
signature, drug predictions were ranked in descending order of the
highest absolute connectivity score. All factors related to a drug
treatment (such as concentration and cell line) were condensed into a
single rank for each drug. This was done by taking the highest
connectivity score from all the different signatures that represent
these experimental conditions for that drug.
Consensus pathway enrichment analysis
To enrich the annotation of drug-disease gene expression signatures
with putative pathway terms, distinct gene sets representing the
therapeutic action of mebendazole and cloperastine in ADPKD disease
(mebendazole-ADPKD, cloperastine-ADPKD respectively) were compiled.
These sets were derived from the principal genes — specifically, the
top 100 overlapping genes irrespective of directional influence —
underpinning the connectivity scores between each drug signature and
its matching disease signature (c.f. [93]Table 1 and [94]Supplementary
Table S1). Furthermore, signatures obtained from multiple
concentrations of each drug were amalgamated to formulate a unified
signature for each drug. Overlapping genes were only considered from
signatures with a connectivity score ≥0.5. Consensus pathway
enrichments (WikiPathways 2019) were obtained using the Enrichr
Consensus Terms Appyter
([95]https://appyters.maayanlab.cloud/#/Enrichr_Consensus_Terms)
([96]Wang et al., 2013; [97]Wang et al., 2016). WikiPathways was chosen
as the main pathway database because it has comprehensive information
on rare disease pathways. Its strong curation process, which includes
automatic quality checks and a clear review system, ensures that only
verified and approved content is included ([98]Martens et al., 2020).
ADPKD-enriched knowledge graph
Treatment predictions were made from an ADPKD enriched knowledge graph
using an algorithm derived from a method published by [99]Himmelstein
et al., 2017. The graph consists of 165 K nodes of 8 different types
(DRUG, DISEASE, GENE, etc.) associated by over 700 M connections
sourced from published data sets and internally curated Healx data. The
algorithm calculates features for the disease using a “degree weighted
path count” metric from over 450B paths. This data set is then used to
train a neural network to recognise features indicative of known
treatments for similar diseases. The trained network was then used to
suggest novel treatments from a set of 20 K drugs connected to the
disease in the graph. When tested using a held out set of known
treatments, the algorithm has produced an AUC score of 0.906 (+/− 0.12)
and F1 score of 0.894 (+/− 0.16).
Multi-scale interactome
The multi-scale interactome (MSI) algorithm ([100]Ruiz et al., 2021)
plays a crucial role in elucidating potential treatments for ADPKD by
comprehensively analysing the intricate network of molecular
interactions within cells as defined in a knowledge graph. By
integrating graph data across various biological scales, from molecular
to cellular levels, the algorithm learns diffusion profiles which
encodes the effects of drugs and diseases propagating through proteins
and biological functions.
The knowledge graph used includes many edges including disease-protein
(72K edges), drug-protein (52K edges), protein-protein (300K edges),
protein-biological functions (44K edges), and biological
function-biological function (63K edges). The MSI algorithm, through
diffusion profiles, was used to encode the propagation effects for
every disease and drug through proteins and biological functions in the
knowledge graph.
The diffusion profiles are encoded through biased random walks that
start at the drug or disease graph node. These profiles depend on
scalar weights
[MATH:
wdrug,wdisease,wprote
in,wbiologic
al func
tion
:MATH]
, and the probability α of continuing the walk. In this study, we used
the default parameters determined by [101]Ruiz et al., 2021:
[MATH:
wdrug=3.21,wd
isease=3.54,wprotein=4.40,wb<
mi>iological function=
6.58 :MATH]
and
[MATH: α=0.859 :MATH]
. By leveraging these encodings, the most relevant proteins and
biological functions in the drug and disease diffusion profiles were
ranked. For the predicted ADPKD treatments, we extracted paths between
the drug and the disease node ‘ADPKD’ in the knowledge graph, filtering
those paths that involve only the top k (here, k = 10) ranked proteins
and biological functions. This process generated subgraphs (depicted in
[102]Figures 4A,B) that facilitate the identification of potential
therapeutic targets and the exploration of treatment strategies
tailored to the unique molecular dysregulations associated with ADPKD.
FIGURE 4.
[103]FIGURE 4
[104]Open in a new tab
A multiscale interactome approach identifies key proteins and
signalling pathways linking the approved treatment, tolvaptan, and the
putative treatment, mebendazole, with ADPKD. Using this approach, a
link is described between ADPKD disease biology and the approved drug
tolvaptan (A), as well as the putative treatment mebendazole (B), in
the first case identifying well established drug targets and processes
associated with tolvaptan, and in the second case putative targets and
processes of mebendazole for further experimental validation.
Cell culture
In vitro studies in patient-derived 3D kidney cyst cultures were
conducted by OcellO B.V. (Leiden, the Netherlands), since acquired by
Crown Bioscience. Patient genotypes were as follows: Donor 1 (PKD1:
c.10594C>T Gln3532*), Donor 2 (PKD1: c.5622G>A p.Trp 1874*) & Donor 3
(PKD1: c.5861dup p. (Asn1954Lysfs*36)). Primary ADPKD patient kidney
cells derived from resected patient kidneys were mixed with
PrimCyst-Gel (Crown Bioscience B.V.) and cultured in 384-well plates
with minimally supplemented kidney base medium (Crown Bioscience B.V.)
for 24 h, after which the cells were treated with or without 2.5 µM of
the cyst swelling stimulus 1-deamino-8-D-arginine vasopressin (ddAVP)
and varying concentrations of tolvaptan or test compound. Treatment
with staurosporine at 250 nM was used as a positive control for cell
death. After 48 h, cultures were fixed, permeabilised and stained with
rhodamine-phalloidin and Hoechst before imaging. Cysts were segmented
using detection of Hoechst-stained nuclei and
Rhodamine-phalloidin-stained cellular f-actin, and cyst area determined
by calculating the area in pixels of each object in every in-focus
plain, which was then averaged per well (Ominer^® image analysis
software, Crown Bioscience B.V.). Cyst area was normalised to
DMSO-treated control wells (0%), and in stimulated conditions
additionally to ddAVP-treated control wells (100%).
Kinase screen
Kinase screening data was generated by the KinaseProfiler service at
Eurofins Discovery (Le Bois l’Evêque, France). Using a radiometric
assay system, mebendazole (10 µM) was screened against a panel of 94
human kinase targets using the Km ATP concentration for each kinase.
Percentage inhibition was reported relative to uninhibited control
conditions.
Animal studies
In vivo work was conducted at InnoSer Belgie N.V (Diepenbeek, Belgium),
according to standard operating procedures and methods described by the
Association for Assessment and Accreditation of Laboratory Animal Care,
and utilised the tamoxifen-inducible iKsp-Pkd1 ^del mouse as previously
described ([105]Leeuwen et al., 2007; [106]Leonhard et al., 2016).
Animals were housed in individually ventilated cages with sterilised
corn cob bedding at 21°C ± 2°C and 40%–70% humidity on a 12/12
dark/light cycle with ad libitum access to food and water.
Kidney-specific disruption of the Pkd1 gene was induced in
KspCad-CreER^T2;Pkd1 ^lox,lox mice by oral tamoxifen administration
(150 mg/kg/day PO) on PND18, 19 & 20, and compound administration began
at PND42 and continued until study termination on PND110. Mebendazole
was solubilised in 10% DMSO: 90% 2-Hydroxypropyl-b-cyclodextrin (20%
w/v in physiological saline), and administered PO at 10, 20 or 30 mg/kg
QD, or 10 or 15 mg/kg BID (BID 5/7 days, QD 2/7 days), and was compared
with a matched QD vehicle treated group. The vasopressin V2 receptor
antagonist tolvaptan was dosed in medicated food at 0.1% w/w (prepared
by SSNIFF, Germany) as a positive control. Tamoxifen untreated mice, in
which the Pkd1 locus remained intact (Pkd1 WT) served as a negative
control. Blood urea was measured weekly from PND74 in 50–100 µL of
blood sampled from the submandibular vein, and once blood urea levels
reached 20 mmol/L, ESRD was said to have developed, and animals were
sacrificed. The study was terminated when at least 50% of animals in
the P18 iKsp-Pkd1 ^del vehicle-treated group developed ESRD, which
occurred in this study at PND109. One hour after final dosing on
PND110, all remaining study animals were sacrificed by exsanguination
via cardiac puncture under isoflurane anaesthesia, followed by cervical
dislocation. Blood was processed for measurement of terminal blood
urea, and both kidneys were removed from the abdominal cavity, weighed,
and processed for histopathology as detailed below. Animals found dead
or euthanised for reasons other than ESRD are detailed in
[107]Supplementary Figure S1, and were excluded from statistical
analysis.
Histopathology
Kidneys were fixed in 10% formalin for 24 h, after which they were cut
in the transverse direction and stored in 70% ethanol. After paraffin
embedding, kidneys were sectioned, stained with hematoxylin and eosin
and digitally scanned for assessment of cystic load. For evaluation of
cystic index, a colour thresholding method was applied using the image
analysis system HALO (Indica Labs, Albuquerque, NM, United States) to
identify total cystic area of each section (sum of all lesions with a
lumen diameter > 9 µm), which was then normalised to total section area
(excluding dilated veins and the pelvic cavity) using the following
calculation: (cystic area/total area) x 100%. For the semi-quantitative
evaluation of cystic grade, kidney sections were scored on an ordinal
scale of 0–5, with 0.5 intervals, based on the following criteria: 0 =
no cysts visible; 1 = from 1 to a few, scattered small cysts; 2 = mild
number of cysts; 3 = moderate number of cysts; 4 = numerous cysts; 5 =
almost all of the parenchyma replaced by cysts.
Ultrasound
Mice were anaesthetised with isoflurane, and hair removed from the
right side of the abdomen prior to collection of a 3D ultrasound scan
of the complete right kidney with a Vevo 3100 imaging system (FUJIFILM
VisualSonics Inc., Toronto, Canada).
Statistical analysis
Statistical analysis and curve fitting was performed using GraphPad
Prism 9.5.1. For in vivo study data, normality was first evaluated
using D’Agostino-Pearson’s omnibus K2 test. If data were normally
distributed, or could be corrected using log transformation, parametric
analysis was performed using one-way ANOVA prior to Dunnet’s multiple
comparisons test. If data were ordinal or not normally distributed,
non-parametric analysis was performed using the Kruskal-Wallis test
prior to Dunn’s multiple comparisons test. The Log-rank (Mantel-Cox)
test was used for kidney survival analysis. Statistical significance
for all testing was assumed when p < 0.05. Data are presented as mean ±
standard deviation for all in vitro and in vivo data.
Results
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 ([108]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 ([109]Song et al., 2009; [110]Pandey et al., 2011; [111]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 ([112]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 [113]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 ([114]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 ([115]Reif et al.,
2011) ([116]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.
FIGURE 1.
[117]FIGURE 1
[118]Open in a new tab
Drugs predicted by computational drug matching inhibit cystic growth in
3D human cellular models of ADPKD. (A) In the absence of ddAVP
stimulation, tolvaptan has no effect on cyst growth, whereas
demeclocycline, cloperastine and mebendazole dose dependently inhibit
cyst growth in primary cell cultures derived from three ADPKD donor
kidneys. (B) Cyst growth is augmented in the presence of ddAVP
stimulation, and is dose-dependently inhibited by tolvaptan,
demeclocycline, cloperastine and mebendazole in primary cell cultures
derived from three ADPKD donor kidneys. Dotted line at 0% represents
unstimulated cyst growth, while the dashed line at 100% represents
ddAVP-stimulated cyst growth. (C) Representative images of 3D-cultured
cysts from Donor 2 treated with control and test compounds, with
Hoechst staining in blue for nuclei and rhodamine-phalloidin staining
in pink for cellular f-actin. Cell cultures were derived from resected
kidney tissue of ADPKD patients, and carry the following PKD1
mutations: c.10594C>T Gln3532* (Donor 1), c.5622G>A p.Trp 1874* (Donor
2) and c.5861dup p. (Asn1954Lysfs*36) (Donor 3). Error bars represent
standard deviation. Scale bars in (C) represent 200 µm.
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 ([119]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
([120]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
([121]van Hastel and Torres, 2017), though until now definitive
evidence of this hypothesis has been lacking.
FIGURE 2.
[122]FIGURE 2
[123]Open in a new tab
Exploration of data underlying predictive modules provides insight into
pathways potentially involved in the therapeutic effect of validated
drug predictions. (A) Deeper interrogation of the ADPKD-enriched
knowledge graph, which gave rise to demeclocycline as a drug
prediction, suggests that vasopressin signalling is the causal link
between demeclocycline and ADPKD. (B, C) Consensus gene enrichment
analysis of the gene expression overlap between ADPKD disease
signatures (as outlined in [124]Table 1) and cloperastine (B) and
mebendazole (C) drug signatures suggests biological pathways
(WikiPathways 2019) which may be relevant in describing the therapeutic
effect in each case. In (B), ADPKD-cloperastine gene expression
enrichment is represented separately for ADPKD disease signatures 175
(blue), 203343 (orange) and 176 (green). In (C), ADPKD-mebendazole gene
expression enrichment is represented separately for ADPKD disease
signatures 203348 (blue) and 203349 (orange).
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
([125]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 ([126]Figure 2B). These pathways have
been linked to varying degrees to the disease biology of ADPKD
([127]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)
([128]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
([129]Wei et al., 2013) and PPAR ([130]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 ([131]Gu et
al., 2012; [132]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 ([133]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 ([134]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
([135]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 ([136]Nygren et al., 2013; [137]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 ([138]Sasaki et al., 2002;
[139]Bai RY. et al., 2015; [140]Choi et al., 2021; [141]Gallia et al.,
2021; [142]Mansoori et al., 2021; [143]Williamson et al., 2021;
[144]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 ([145]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 ([146]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 ([147]Dagorn et al., 2023; [148]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 ([149]Leeuwen et al., 2007; [150]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 ([151]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 ([152]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 ([153]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 ([154]Figures 3B–E).
Twice daily oral administration of mebendazole at 15 mg/kg
significantly alleviated kidney cystic load by 33% ([155]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 ([156]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 ([157]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 ([158]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
([159]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
([160]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 ([161]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 ([162]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) ([163]Figure 3H).
FIGURE 3.
[164]FIGURE 3
[165]Open in a new tab
Mebendazole ameliorates kidney phenotypes in the P18 iKsp-Pkd1 ^del
mouse model of ADPKD. (A) Pkd1 disruption was induced by daily
tamoxifen administration from PND18-20, and mebendazole, tamoxifen or
vehicle were administered from PND42-110. Ultrasound was used to
measure kidney volume at PND41 (baseline), PND75 and PND94, while blood
urea was measured weekly from PND74 and used to determine when animals
reached ESRD (blood urea >20 mmol/L). (B) As demonstrated in
hematoxylin and eosin stained kidney sections from PND110 animals, Pkd1
disruption results in enlarged cystic kidneys, which are attenuated
with twice daily mebendazole treatment (scale bar represents 2.5 mm).
(C) Twice daily treatment with 15 mg/kg mebendazole significantly
reduces cystic index of Pkd1 KO mice; ** = p < 0.01 Dunnet’s multiple
comparisons test. (D) Twice daily treatment with 10 mg/kg or 15 mg/kg
mebendazole significantly reduces cystic grade in Pkd1 KO mice, as
assessed by blinded histopathological scoring; * = p < 0.05 and ** = p
< 0.01 Dunn’s multiple comparisons test. (E) Body weight normalised
kidney weight is increased in Pkd1 KO mice, but significantly
attenuated with twice daily treatment with 10 mg/kg or 15 mg/kg
mebendazole; * = p < 0.05 and *** = p < 0.001 Dunnet’s multiple
comparisons test. (F) Twice daily treatment with 10 mg/kg or 15 mg/kg
mebendazole significantly reduces increased kidney volume in Pkd1 KO
mice at the study dosing mid-point PND75; * = p < 0.05 and *** = p <
0.001 Dunn’s multiple comparisons test. (G) Terminal blood urea is
elevated in Pkd1 KO mice, but attenuated by BID treatment with 10 mg/kg
or 15 mg/kg mebendazole; * = p < 0.05 and ** = p < 0.01 Dunn’s multiple
comparisons test. (H) Kaplan-Meier analysis of kidney survival from
ESRD reveals a significant treatment effect. (C–G) Each symbol
represents one animal; error bars represent standard deviation; n =
14–22 per treatment group (Pkd1 WT group excluded from analysis).
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
([166]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 ([167]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 ([168]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 ([169]Grimm et al., 2006), the
renin-angiotensin-aldosterone system, dysfunction of which is both a
downstream consequence of, and direct contributor to ADPKD ([170]Hian
et al., 2016), and components of the G protein-coupled receptor
signalosome, which regulate ADPKD-associated proteins such as CFTR and
EGFR ([171]Hama and Park, 2016; [172]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 ([173]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
([174]Morris et al., 1998; [175]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) ([176]Dakshanamurthy et al., 2012) and ABL1 (aka
Abl) ([177]Nygren et al., 2013), to phosphorylation processes also
regulated by EGFR. VEGF signalling has previously been linked with
ADPKD pathophysiology ([178]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 ([179]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 ([180]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
([181]Figure 5A) and CBS agents from the wider family ([182]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 ([183]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)
([184]Bello-Reuss et al., 2001; [185]Tao et al., 2007), cyclin
dependent kinase 1 (CDK1) ([186]Zhang et al., 2021) and Met ([187]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 ([188]Lacey, 1988), and includes molecules such as
albendazole with no appreciable kinase activity ([189]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.
FIGURE 5.
[190]FIGURE 5
[191]Open in a new tab
The therapeutic effect of mebendazole in ADPKD is likely driven by an
inhibitory effect on microtubule polymerisation and ADPKD-relevant
kinases. (A) In addition to mebendazole, other benzimidazole
anthelmintics with microtubule depolymerising activity are also active
in primary cell cultures derived from an ADPKD donor kidney. (B)
Members of the wider class of CBS ligands with microtubule
depolymerising activity are also active in primary cell cultures
derived from an ADPKD donor kidney. (C) At 10µM, mebendazole (MBZ)
inhibits a number of protein kinases, including several ADPKD-relevant
kinase targets. Cell cultures in (A, B) were derived from resected
kidney tissue of an ADPKD patient carrying a (C) 5622G>A p.Trp 1874*
mutation in PKD1. Error bars represent standard deviation.
Discussion
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
([192]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 ([193]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 ([194]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
([195]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 ([196]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) ([197]Li and Yang, 2015). Loss of polycystin-1 is
associated with an increase in HDAC6 ([198]Liu et al., 2012) and SIRT2
([199]Zhou et al., 2014) expression, and a decrease in acetylated
α-tubulin ([200]Zhou et al., 2014), while inhibition of either
deacetylase has been shown to augment α-tubulin acetylation levels and
ameliorate ADPKD-relevant phenotypes ([201]Zhou et al., 2014;
[202]Cebotaru et al., 2016; [203]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 ([204]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
([205]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)
([206]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 ([207]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 ([208]Dakshanamurthy et al., 2012; [209]Issa et al.,
2015; [210]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 ([211]Tao et
al., 2007), but paradoxically anti-VEGF treatment augmented cyst growth
and accelerated functional decline in the same model ([212]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
([213]Horie et al., 1994), while further mechanistic studies confirmed
hyperactivation of Met signalling in Pkd1 models, pharmacological
inhibition of which rescued disease phenotypes ([214]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 ([215]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 ([216]Nygren et al., 2013; [217]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 ([218]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 ([219]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, [220]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 ([221]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.
Acknowledgments