Genetically proxied inhibition of Phosphodiesterase-5 and cancer risks: A drug-target Mendelian randomization analysis.

Introduction
Phosphodiesterase 5 (PDE5) inhibitors including sildenafil, tadalafil, vardenafil, and avanafil are commonly used in the treatment of erectile dysfunction (ED) and pulmonary arterial hypertension (PAH)1. PDE5 is a metallophosphohydrolase that degrade cyclic guanosine monophosphate (cGMP) and thus play a key role in cGMP-related signaling pathway2. Given the safety profile of approved drugs, there is considerable interest in exploring PDE5 inhibitors as novel treatment approach to diseases involved cGMP-related signaling pathway3–5.
cGMP regulates cancer cell proliferation, migration, adhesion, and the tumor microenvironment6,7, such as blood flow, angiogenesis, inflammation, and immune response. Accumulative evidence with in vitro and in vivo studies found that PDE5 was overexpressed in multiple cancer types (e.g., prostate, colon, lung, oral squamous and breast cancer)8–12, and restoration of intracellular cGMP signaling with PDE5 inhibitors might have an anticancer activity. However, there is also some preclinical evidence suggested that reduced PDE5 expression was associated with increased invasiveness, metastasis, and melanin synthesis of melanoma cells, which may facilitate melanoma development13,14.
Epidemiological studies have variably reported both carcinogenic and antineoplastic effects of PDE5 inhibitors medication. For example, some observational studies have reported that PDE5 inhibitors users have lower risk of several cancers (i.e. colorectal cancer [CRC], head and neck squamous cell carcinoma, chronic lymphocytic leukaemia)15–17, while PDE5 inhibitors use has been associated with an increased risk of melanoma and other skin cancers18–20. In addition, long-term PDE5 inhibitor use have been linked to an elevated risk of biochemical relapse of prostate cancer21. However, recent studies found no evidence of a positive association between exposure to PDE5 inhibitors and risk of cancer22–25. Repurposing of PDE5 inhibitors in clinical use for malignant conditions remain challenging26–29.
The causal relationship between PDE5 inhibitor use and effect on cancer has not been determined, and epidemiological studies are vulnerable to unmeasured confounding, reverse causation and various forms of bias (e.g. selection bias, immortal time), which may preclude a causal interpretation30. Although a number of clinical trials are investigating the anti-cancer uses of PDE5 inhibitors31, such studies typically have limited follow-up periods, and may not able to adequately capture outcomes with long induction and latent period, such as cancer.
Using germline variants within the genes encoding drug targets as instrumental variables (proxies) that mimic the effect of pharmacological agents, drug target Mendelian randomization (MR) is an efficient way to evaluate the causality for lifelong pharmacological exposures on disease outcomes. Given the widespread use of PDE5 inhibitors owing to patents expiration, and long induction period of cancers, as well as continuing uncertainty over a causal link of PDE5 inhibitors with cancer risk, we used a two-sample MR design to investigate the associations of genetically proxied PDE5 inhibition with risk of 22 different types of cancer.

Methods

Study overview
We aimed to evaluate the causal role of PDE5 inhibition on common cancer risk using two-sample drug-target MR study design. We focused on cis-acting protein quantitative trait loci (pQTLs) in the vicinity of the PDE5A gene to mimic the action of the protein target since most drug targets are proteins. An outline of the current study is shown in Fig. 1. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology-Mendelian Randomisation reporting guidelines (Table S1). Participant consent and ethical approval was not required for this study because all summary-level genetic data were obtained from sources available to the public.

Data sources
To select single-nucleotide polymorphisms (SNPs) to proxy PDE5 inhibition, we used summary statistic from the UK Biobank Pharma Proteomics Project (UKB-PPP), which quantified 2940 plasma proteins levels involving 54,219 participants32. To validate the SNPs as instruments of PDE5 inhibition, we estimated their associations with ED and PAH from independent GWAS data in positive control Mendelian randomisation analyses33. To ensure that the associations of ED and PAH were not identified by chance, we also examined their effect on diastolic blood pressure (DBP) using summary data from the largest GWAS of blood pressure involving 757,601 individuals of European-ancestry34.
As for variant-outcome associations, summary genetic association data were obtained from 22 cancer-specific genome-wide association study in the discovery stage. Detailed information on included datasets in discovery stage is summarized in Table 1. We then used the data from the FinnGen study for replication analysis (87,505 cancer cases and 314,193 controls)35. These analyses were restricted to participants of European ancestry. Further information on diagnostic criteria for cancer, genotyping, imputation, and quality control measures for these studies is available in the original publications. Sample overlap is likely to be nominal in this study because the exposure and outcome GWAS were conducted with largely independent samples.

Instrument construction
Given that most drug targets are proteins, common (minor allele frequency > 0.01) SNPs associated with plasma PDE5A concentration at genome-wide significance (P < 5 × 10−8) that were in or within ± 100 kb from the PDE5A gene (GRCh37/hg19, chr4: 120,415,550 − 120,550,146) were considered as instrumental variables (IVs) for PDE5 inhibition. These IVs were further clumped using LD threshold of r2 < 0.001 with a clumping window of 150 kb on the basis of the 1000 Genomes Project European data. To explore potential horizontal pleiotropic effects, we searched AstraZeneca PheWAS Portal36, a curated database employing 17,361 dichotomous and 1,419 quantitative phenotypes, for traits associated with the genetic instruments to mimic PDE5 inhibition (p < 1 × 10−8). The strength of genetic instruments measured by calculating the F statistic and an F-statistics above 10 suggests minimal weak instrument bias. To validate the genetically proxied PDE5 inhibition instruments, positive control analyses were performed with ED, PAH and DBP as the outcome.

Statistical analysis
Causal estimates were generated using random-effects inverse-variance weighted (IVW) approach37. If an SNP was missing from the outcome GWAS, we replaced it with a proxy SNP in high LD (r2 > 0.80). The heterogeneity between SNPs were evaluated by Q statistic. The MR Steiger test was also performed to validate direction of association38. To further explore the robustness of significant results, colocalisation analysis was also performed when a statistically significant Mendelian randomisation association was detected. It assessed the probability for five hypotheses: H0, no causal variant for either trait; H1, a causal variant for trait 1 but not trait 2; H2, a causal variant for trait 2 but not trait 1; H3, association with both traits but at separate causal variants; and H4, association with both traits at a shared causal variant39. A posterior probability of H4 greater than 0.80 was considered to be strongly colocalised.
Odds ratios (ORs) with 95% confidence intervals (CIs) of cancers are one standard deviation (SD) decrease in protein level through genetically proxied PDE5 inhibition. The pooled effects from discovery and replication datasets were obtained using meta approach with a fixed-effects model weighted. To account for multiple testing across analyses, a Bonferroni correction was used to establish a p value threshold of < 0.0023 (0.05/22 cancer types) for the pooled effects. All statistical analyses were two-sided and conducted using the ‘TwoSampleMR’, ‘meta’, and ‘coloc’ packages in R (version 4.3.0).

Results

Instrumental variables
Characteristics of the 2 SNPs used as instruments were presented in Table 2, with mean F statistic of 347. Genetically proxied PDE5 inhibition was associated with decreased erectile dysfunction (P = 0.023), pulmonary arterial hypertension (P = 0.044) and diastolic blood pressure (P = 0), suggesting the validity of genetic instruments (Table S2). AstraZeneca PheWAS also confirmed that none of the selected genetic variant was deemed to be at risk of horizontal pleiotropy.

Main results
The MR estimates for the effect of genetically proxied PDE5 inhibition on risk of different cancers from two independent datasets and pooled analysis were shown in Table S3. Genetically proxied PDE5 inhibition was strongly associated with decreased risk of colorectal cancer in both discovery (P = 1.56 × 10−6) and replication datasets (P = 4.25 × 10−8). The combined OR of colorectal cancer was 0.80 (95% CI: 0.75–0.86, P = 6.15 × 10−11) for genetically predicted one-standard deviation decrease in plasma PDE5A level through PDE5 inhibition (Fig. 2). In addition, there was strong evidence of an protective effect of genetically proxied PDE5 inhibition on gastric cancer in the combined analysis (OR = 0.48, 95%CI: 0.34–0.68; P = 4.70 × 10−5), although we failed to detect significant association in the discovery stage. Results from Cochran’s Q test (all P > 0.05) implied no apparent evidence of heterogeneity (Table S4). In addition, the correct causal direction was confirmed using the MR-Steiger directionality test (P for Steiger test < 0.05, Table S5). However, colocalisation analysis showed limited evidence of shared causal variants in PDE5A for colorectal cancer (PP.H4 = 57.5%) or gastric cancer (PP.H4 = 51.6%, Table S6).

We did not find evidence of an association between genetically proxied PDE5 inhibition and other 20 cancers in the combined analysis after Bonferroni correction (Fig. 2).

Discussion
To the best of our knowledge, this is the first MR study systematically evaluated the causality between PDE5 inhibition and cancer risk. We found robust inverse association of PDE5 inhibition with risk of colorectal cancer, which corroborates with finding from a recent meta-analysis of 3,518 cases40. Besides, we also found strong evidence for an protective effect of PDE5 inhibition on gastric cancer. Of note, we did not observe a causal effect of PDE5 inhibition on other 20 types of cancer. This included the absence of a carcinogenic effect of genetically-proxied inhibition of PDE5, implying previous observational studies investigating the relationship between sildenafil and melanoma and skin cancer risk may be subject to residual confounding or bias.
The mechanisms that links PDE5 inhibitors and lower CRC risk are not completely known and several underlying mechanisms have been proposed. It has been reported that PDE5 inhibitors can suppress proliferation and apoptosis in the mouse colonic epithelium by activating the cGMP/protein kinase G II pathway41–43. A range of in vivo studies showed that the sildenafil reduced tumor inflammation in polyps by suppressing the secretion of several pro-inflammation cytokines and immune cell infiltration10,44. Similar effects were also reported that sildenafil and tadalafil delayed tumor outgrowth, even by 50-70% in immune-competent mice45. Recently, the ability to reduce redox stress of PDE5 inhibitors in the colon mucosa have also been suggested to be associated with its anticancer effect46. Genetic evidence from The Cancer Genome Atlas further supported the observed antitumor effects in which CRC patients (including male and female) without PDE5 mutation suffered poorer survival rates than those with mutation of PDE547. In addition, sildenafil and other PDE5 inhibitors were show to enhance the lethality of multi-kinase inhibitors48, and associated with increased efficacy of some commonly used cancer drugs (5-fluorouracil and anti-PD-1 antibody)49. A recent study suggested that sildenafil abolished the effects of chronic stress on CRC by activating the cGMP pathway through metabolite produced by gut microbiota50. Given the guanylate cyclase/cGMP/PDEs signaling pathway in maintaining the homeostasis of intestinal epithelial cells51, these observations highlight the therapeutic potential of PDE5 inhibitors for CRC.
In consideration of the established records on drug efficacy, safety, and adverse effects, our MR findings provide vital evidence to explore the potential of repurposing PDE5 inhibitors in prevention or treatment of CRC. Given that PDE5 inhibitors are commonly prescribed for male patients, the protective effect on CRC observed in previous observational studies was exclusively based among male population, and the anticancer effect of PDE5 inhibitors for female remain unknown. In the present study, the associations for CRC was detected in the largest GWAS of colorectal cancer and replicated in FinnGen, which both included high proportion of female participants, suggesting a similar protective effect on CRC for both male and female. However, further clinical trials are warranted to validate the preventive or therapeutic effect of PDE5 inhibitors for CRC, especially in female population.
The protective effect of PDE5 inhibition and GC has been first revealed in this study. It is worth noting that this protective effect was observed from studies with relative small sample size (1,568 cases), while associations were directionally consistent in discovery and replication datasets. Thus, further observational study is needed to check whether the PDE5-GC association could be replicated even though the incidence rate of GC in European population is relative low. The molecular mechanism by which PDE5 inhibition influences the risk of GC remain unknown. However, a recent study demonstrated that sildenafil mediates the inhibition of downstream IL-6/JAK/STAT3 signaling pathway activation, and thus suppressing tumor growth of GC52. Findings from our MR analysis and mechanism study raises the possibility of PDE5A inhibitors as a potential anticancer drug for GC.
Some limitations should be acknowledged when interpreting the results of this study. First, our MR analysis estimated the effect of PDE5 inhibition based on on-target reductions in PDE5A protein level rather than the short-term effects of PDE5 inhibitors use in practice. Second, there was little evidence from co-localisation analyses to support the association between genetically proxied PDE5 inhibition and CRC or GC outcome, which may reflect insufficient statistical power. Thus, our study is more helpful in determining the direction of associations rather than quantifying the magnitude of causal effect. Third, gene-environment and/or gene-gene interactions, as well as potential off-target effects cannot be estimated from summary-level statistic data, although accumulated evidence suggested a complex interplay of these factors influence cancer risk. Fourth, we cannot definitively rule out the possibility that our MR results might be biased by horizontal pleiotropy, although we leveraged cis-acting variants from previous MR study and showed the validity of these genetic instruments. However, results from different datasets remain consistent. Fifth, we can’t completely rule out the potential bias due to sample overlapping between exposure and outcome using summary data. However, the overlapping probably involves < 1% of all samples and it is unlikely to affect precision of our estimates. Additionally, sensitivity analyses (e.g., the weighted Median and MR-Egger analyses) can not be performed due limited number of genetic instruments available. Lastly, our findings was primarily based on European populations, thus may have limited generalizability to other ethnic groups.
In summary, the present study provides MR evidence that genetically proxied PDE5 inhibitors use was associated with lower risk of CRC and GC, highlighting its potential for drug repurposing. More importantly, we did not identify harmful causal effect of genetically proxied PDE5 inhibition on other 20 types of cancer, which may help to eliminate concerns regarding PDE5 inhibitors medication in carcinogenesis promotion. Future studies should focus on elucidating the mechanism underlying the inverse association between PDE5 inhibition and CRC and GC.

Electronic supplementary material
Below is the link to the electronic supplementary material.