In this issue of Brain, Bushra Wali et al. (2014) describe the efficacy of progesterone in an animal model of ischaemic stroke and, in so doing, provide data of importance for clinical translation, while at the same time raising the bar for high quality reporting of in vivo research (Wali et al., 2014).
The development of new treatments for stroke is a tricky business (O'Collins et al., 2006), and so finding an established drug that might have therapeutic efficacy is a tantalizing prospect. Abundant preclinical data support a neuroprotective effect of progesterone in experimental stroke and traumatic brain injury, and a phase III trial is under way. Given that the safety profile of progesterone is already known, such reprovisioning would accelerate drug development. A potential confounding influence is sex- and age-related differences in circulating progesterone. Women who experience stroke are usually post-menopausal, and modelling the menopause in animals is generally achieved by ovariectomy. In a meta-analysis of the effects of progesterone in animal models of focal cerebral ischaemia, Wong et al. (2013) found that progesterone increased mortality in young female hormonally-intact animals, and had no effect on lesion volume in adult ovariectomized females (Wong et al., 2013). In human studies, patients taking hormone replacement therapy (either oestrogen alone or oestrogen plus progesterone) at the time of their stroke seemed to have a worse outcome (Bath and Gray, 2005); however, the effect of progesterone on its own is unknown.
Wali et al. (2014) avoid the confounding effect of circulating progesterone and effects of ovariectomy by restricting attention to aged (12-month-old) male rats. They systematically explore the limits to efficacy in ways that will help inform clinical trial design. For delay to treatment, the authors establish that efficacy is retained at 6-h delays but lost at 24-h delays, thus validating the use of a 6-h window in clinical trials. For drug dose, they show maximum efficacy at the lowest dose tested [16 mg/kg on Day 1 (in two doses 5–6 h apart), 8 mg/kg for the next 6 days], with 32 mg/kg (Day 1)/16 mg/kg thereafter marginally less effective, and 64 mg/kg (Day 1)/32 mg/kg thereafter clearly less effective. These doses are at the higher end of those identified in the review of Wong et al. (2013) who found no dose-response relationship. These dosage regimes are, however, loosely analogous to those that have been used in some clinical trials in traumatic brain injury, where the total daily dose has ranged from 2 mg/kg to 12 mg/kg.
The study by Wali et al. (2014) is also important in that it explores efficacy in aged animals with outcomes measured at extended times after the stroke has occurred. One difficulty in assessing the efficacy of drugs used to treat stroke is that residual infarct volume may be a poor guide to functional improvement, and in many animal models (Howells et al., 2010) the neurobehavioural consequences of focal cerebral ischaemia resolve rapidly, meaning that long-term efficacy is difficult to measure. Using a combination of behavioural endpoints, Wali et al. (2014) demonstrate firstly that there is a sustained behavioural deficit 3 weeks after the injury, and secondly that the efficacy of progesterone was retained at these later time points. This response appeared to be independent of the delay to treatment—that is, the beneficial effect of treatment initiated 6 h after the induction of focal ischaemia was retained whether outcome was measured 3 days, 9 days or 21 days later. Conversely, the lack of benefit of treatment initiated 24 h after ischaemia was largely consistent across times of outcome measurement, although it is a simplification to say there was no efficacy with late treatment; for locomotion (distance travelled in the open field) and for measures of gait quality (stride length, paw print area, limb swing speed) the data are consistent with a biologically important effect, which the experiment was not powered to detect. As the authors point out, if efficacy is sustained beyond 6 h this would substantially broaden the pool of patients potentially eligible for progesterone treatment. This efficacy at longer delays to treatment suggests effects on regeneration and repair as well as neuroprotection, consistent with reports that progesterone promotes post-ischaemia synaptogenesis (Zhao et al., 2011), increases circulating endothelial progenitor cells (Li et al., 2012), and promotes neurogenesis (Barha et al., 2011) and neural regeneration (Li et al., 2012).
The most important thing about this study, however, is the effort made by the authors to minimize the risk that their findings are confounded by bias (Kilkenny et al., 2010; Landis et al., 2012). By randomly allocating animals to the various experimental groups, they ensure that the only systematic differences between the groups were those designed into the experiment. Of course, there will be differences between laboratory animals, but the investigators performed a power calculation to ensure that their experiment was large enough to account for these chance differences and still see any treatment effects of a size considered important. The authors take care to report the number of animals recruited to each experimental cohort, and the number surviving ischaemia surgery and to the time that outcome is measured. This allows the reader to judge whether improved outcome may be because of a healthy survivor effect (O'Collins et al., 2011), and is particularly important where a drug has previously been reported to increase mortality. Finally, by blinding the assessment of outcome the authors give us greater confidence that the behavioural differences they observed are true differences between groups, and not a product of how the animals were handled before the test or of the optimism of the observer.
In terms of scientific rigour and the quality of reporting, this then is an excellent study. There are areas where further improvements could be made. The method of randomization might be described in greater detail for instance; or the reader might be given access to a date-stamped study protocol in which a primary outcome measure had been defined, allowing us to be confident that there was no selective outcome reporting bias (Tsilidis et al., 2013). Nonetheless, this study serves as an example to the field.