Synthetic miR-34a against solid tumours: a predictable failure

Abstract

A prolific scientific literature attributes pro- or anti-oncogenic properties to many human microRNAs (“miRNAs”). While many of these studies are based on unpersuasive analyses, one candidate suppressor tumour miRNA, miR-34a, appeared convincing enough to be administered to human patients in a clinical trial—with disappointing outcomes. Here, we review possible reasons for that failure, and their implications for other miRNAs.

Introduction

Hong et al. [1] describe the results of a clinical trial where, for the first time, a synthetic miRNA has been administered to human patients. miRNAs are small regulatory RNAs with poor biochemical specificity: they recognise their target RNAs by sequence complementarity, and a perfect match to the miRNA “seed” sequence (nucleotides 2–7) is the best-known predictor for target binding [2]. Because the seed is only a hexamer, perfect seed matches are very frequent in the transcriptome. Even when selecting phylogenetically conserved seed matches in 3′ UTRs (which tend to be the most efficient binding sites), more than 60% of human coding genes are predicted miRNA targets [3]. Experimental identification of miRNA targets (e.g., with immunoprecipitation of the effector complex followed by deep-sequencing of the recruited RNAs) also supports this view, with hundreds of targets being typically found for each miRNA and a good agreement between such experiments and computational predictions [4]. It is nonetheless important to note that miRNA-guided repression is very modest, and many of these interactions are unlikely to trigger physiological consequences at the macroscopic scale, even though they are real at the molecular scale [5].

Inevitably, some cancer-related genes (either favouring or inhibiting any aspect of tumour appearance, growth, immune evasion, metastasis,…) appear in the long lists of molecular targets for most human miRNAs. The fact that a given gene (say, an oncogene) bears a seed match for a miRNA in its 3′ UTR, and that a reporter gene fused to that UTR is repressed in cultured cells upon artificially large overexpression of the miRNA, is not proof that the miRNA is a tumour suppressor. miRNA overexpression is notoriously prone to artifacts [6]. Moreover, even if the repression is real in vivo it may not be large enough to translate into a clinically-relevant phenotype at the macroscopic scale.

This is why, despite the impressively large corpus of studies attributing cancer-related functions to many human miRNAs, rigorous in vivo confirmation is missing for most of them. Among this complicated bibliographic landscape, miR-34a was standing like one of the most reliably validated tumour suppressor miRNAs: its gene was reported to be frequently lost in various cancers, and multiple studies in cultured cells and in animals suggested that miR-34a overexpression reduces tumour growth by several mechanisms (inducing apoptosis, inhibiting tumour immune evasion,…). Synthetic miR-34a therefore became the first miRNA-based drug reaching clinical phase I in oncology, with the administration of MRX34 (a synthetic miR-34a/miR-34a* duplex encapsulated in a liposomal vehicle) to patients with various types of solid tumours [1, 7].

Poor clinical efficiency, recurrent adverse events

The initial administration regimen (“BIW”: 4-week cycles with two intravenous infusions per week for 3 weeks followed by 1 week of rest) proved difficult to follow because of various adverse events, causing many patients to discontinue the study [7]. An improved regimen (“QDx5”: 3-week cycles with five intravenous infusions during the first week, followed by 2 weeks of rest, together with dexamethasone pre-medication) could be followed more easily in the clinic, but once again most patients experienced adverse events (including hepatotoxicity and potential chronic inflammation suggesting immune-mediated toxicity) and disease progression [1]. The study even had to be prematurely stopped after five drug-related severe adverse events, resulting in four patient deaths [1]. Clinical efficacy was limited: out of 66 evaluable patients (for 85 recruited patients in total), partial response was observed for only 3 patients. For two of them, the response was observed several weeks after MRX34 treatment discontinuation. While such a delay may suggest immune-mediated action on tumour [1], it could also mean that the response was completely unrelated to MRX34, especially because each of these patients had been heavily treated with other therapies prior to the MRX34 trial.

Overall, the lack of convincing efficiency, and the frequency and severity of adverse events, put an end to the study. While similar examples of disappointing results are common in clinical trials, we think that the failure of MRX34 could have been anticipated.

Questionable initial assumptions

It had been abundantly repeated that the miR-34a gene is frequently mutated, or otherwise inactivated, in human cancers—a behaviour indeed expected from tumour suppressors. But this assumption is based on very little evidence (e.g., deficient statistical analyses, or taking cultured cell lines as an approximation for tumours). It is however possible to explore precisely the genetic status of numerous genes as well as their level of expression in primary tumours and normal adjacent tissues in a wide variety of cancers. Such extensive, carefully controlled analyses, do not confirm the frequent inactivation of miR-34a in any of the cancer types with appropriate data available [8]. In addition, minimal genetic ablation of the miR-34a gene in immortalised cell lines does not recapitulate the assumed endogenous anti-proliferative activity of this miRNA. Instead, accurate miRNA quantification with RT-ddPCR reveals that miR-34a impacts cell proliferation only when strongly overexpressed: for example, in the heavily studied HCT-116 cell line, a 10 nM transfection of synthetic miR-34a significantly represses cell proliferation and results in a > 8000-fold increase in intracellular miR-34a, while a 1 nM transfection over-expresses it by >490-fold without noticeably affecting cell proliferation [8].

Strikingly, the tumour-suppressive function of endogenous miR-34a had already been seriously questioned by an in vivo study in mouse [9]. miR-34a/b/c KO animals do not display defects in p53-dependent proliferation control or increased susceptibility to spontaneous, irradiation-induced, or cMyc-initiated tumorigenesis. Yet that convincing experiment was overwhelmed by a large number of cultured cell overexpression experiments (with unrealistically high miR-34a levels triggering cell growth arrest).

Finally, because the observation of the tumour-suppressive properties of miR-34a relies only on ectopic experiments, the reported therapeutic potentiality of this miRNA is more likely the result of artifactual effects, including (i) titration of the endogenous miRNA machinery components, (ii) activation of the interferon pathway, (iii) increasing endogenous target occupancy, hence achieving higher target repression, or (iv) repression of novel low-affinity targets that are not affected with endogenous miRNA concentrations [6]. Consequently, while the clinical study with MRX34 aims at replenishing the levels of a lacking tumour suppressor miRNA in cancerous cells, it may provide the whole body with supraphysiological amounts of artificial miRNAs. Thus, even if the introduction of synthetic miRNAs were efficient in reducing tumours, it would not be due to getting back to normal conditions but to a cytotoxic agent with unpredictable secondary effects.

miR-34a overexpression in patients

It is difficult to evaluate the overexpression of miR-34a mediated by MRX34 administration to human patients, but several indirect pieces of information can provide a rough estimation. In non-human primates, after a single dose of MRX34 at 1 mg/kg, miR-34a concentration in blood increases from ≈0.04 to ≈300 ng/mL after 24 h [10]. In tissues dissected at the same timepoint, miR-34a concentration was strongly increased, with much variability from tissue to tissue (fold-change ranging from ≈ 8 [brain] to ≈80,000 [liver] with most tissues exhibiting a fold-change close to 100) [11].

In the human clinical trial, miR-34a concentration in blood was measured both with the BIW [7] and the QDx5 [1] schedules, revealing a concentration ranging between ≈1000 and 10,000 ng/mL for the former and between ≈1000 and 20,000 ng/mL for the latter. It is therefore reasonable to hypothesise that the sustained administration of such high doses of MRX34 to human patients resulted in larger overexpression in tissues than what had been observed in non-human primates.

In terms of molecular response, the repressive effect of administered MRX34 was evaluated in white blood cells, quantifying several candidate miR-34a target RNAs before and after treatment—with a reported repression of these genes after treatment [1]. The significance of this effect is not clear though, with post-treatment timepoints being pooled before the analysis, and with statistical analyses missing for the RNA-Seq-based measurement.

What is next?

The MRX34 trial has been the first study administering a synthetic miRNA to human patients, and it could provide useful informations for future attempts. In particular, the observed adverse events seem to be due to immune activation-related toxicity, and it will be important to determine whether such effects are miRNA sequence-dependent or -independent.

As discussed in Hong et al. [1], the same liposomal vehicle had already been used in humans without triggering such toxicity, hence it is unlikely to be responsible. But RNA by itself is immunogenic, raising the possibility that every future miRNA-based drug may trigger similar adverse events. Yet Hong et al. [1] recall that pre-clinical studies with MRX34 did not show such immune activation-related toxicity, even in non-human primates, suggesting that RNA immunogenicity per se may not explain the effect.

In an alternative explanation, MRX34-related adverse events could be sequence-specific (meaning that future, different synthetic miRNAs, may not have the same effect). Sequence-specific miRNA effects are most likely to occur through the repression of endogenous mRNAs exhibiting 3′ UTR seed matches to the artificially abundant miRNA (while the endogenous miR-34a, far less abundant, would not have the same effect). The absence of immune activation-related toxicity in non-human primates would then be easy to rationalise, because 3′ UTR sequences are very different across species.

If this were the case, then future miRNA-based therapies may face a fundamental problem: their secondary effects could only be evaluated in humans themselves, with animal models being almost useless for safety assessment. Classical drugs (e.g., small molecules or antibodies) typically act by interacting with proteins, which tend to be highly conserved (both in terms of sequence and structure) among related species—a feature not shared by mRNA 3′ UTRs.

Future development of miRNA therapeutics will also require safe and efficient delivery vehicles, which pose another challenge for the industry. Biodistribution measurements in non-human primates revealed a highly heterogeneous delivery of MRX34 from organ to organ, implying that some tissues may prove difficult to reach—therefore calling for the development of novel, dedicated delivery methods.

Conclusion

An overview of the scientific literature may give the impression that many human miRNAs are involved in a large range of cancers. While some miRNAs could be real oncogenes or tumour suppressors, it is important to keep a critical eye on this apparent consensus. miRNA overexpression experiments, and enrichment analyses in automatic annotation terms for predicted miRNA targets, do not constitute any proof for an actual, physiological activity of that miRNA in oncogenesis.

Institutional incentives may have biased the orientation of miRNA research efforts towards molecular oncology, resulting in a biased description of miRNA biological action. This possible effect could be spectacularly illustrated by the “word cloud” option offered by the miRBase database [12]: screening for terms frequently associated with miRNA names in open-access articles, the word “cancer” shows up almost systematically as one of the most frequent terms for any given miRNA.

The fact that even the most convincing candidate tumour suppressor miRNA, miR-34a, turned out to be a predictable disappointment, is a cruel warning for other attempts of application of miRNA-based knowledge in cancer treatment. In order to save resources, and patients’ hopes, it is certainly advisable to avoid unrealistic miRNA overexpression experiments, shallow analyses on miRNA gene mutation in cancer, and unsound extrapolations from the automatic annotation of predicted miRNA targets. Because these methods proved deceiving for miR-34a, it is likely that many other popular miRNAs in oncology are similarly delusive.

Author contributions

SM and HS wrote the manuscript.

Funding

None.

Ethics approval and consent to participate

Not applicable.

Consent to publish

Not applicable.

Competing interests

The authors declare no competing interests.