The publication of Khush et al. in this month's issue (1) provides an opportunity to consider some of the challenges in doing genetic association studies in transplantation. While rigid rules for such manuscript submissions are neither warranted nor desirable, there are several guidelines published for gene association study publications that are well worth reviewing (2,3). These guidelines are also directly relevant to gene association studies in transplantation and the Journal does not need to create any new guidelines for this reason. Thus, in the present editorial, I will use this new article in the Journal to consider some science-based guidelines specifically for transplantation.
There is no question that the physiological stability of the donor prior to surgical organ recovery is a significant factor in effectively managing the donor procedure, influences the selection for cardiac donation and impacts the early posttransplant course of the recipients. With that context, Khush et al. examined 2048 consecutive brain dead donors to study almost 1500 with a focus on four known and functional genetic polymorphisms of β-adrenergic receptors and determined their association with left ventricular function during the donor management period. Their results in the discovery cohort demonstrate significant associations with two of these genetic polymorphisms and cardiac function after brain death in the donors. There is also a matching trend in some metrics for posttransplant cardiac allograft dysfunction in the recipients.
Linking genetic polymorphisms to donor cardiac function is instructive at several levels. One point is that these specific genetic markers were chosen because they are already well established to be modifiers of β-adrenergic receptor function and clinical outcomes. These correlations are in conditions characterized by excessive catecholamine release that drives strong adrenergic signaling such as in myocardial infarctions and heart failure (4–8). Demonstrating these specific genetic correlations with cardiac function at the onset of donor brain death supports the current paradigm that brain death and catecholamine overexpression are directly linked and important to the clinical management of all donors as well as selection of donors for heart transplantation. The authors also point out that in certain donors, higher requirements for catecholamine support may simply reflect their genetics for lower receptor sensitivities. In that situation, such support should not always be considered as a negative factor in selecting donors for heart transplants.
One strength of the approach used in this study and that is generalizable to other studies is the selection of a small number of high value genetic polymorphisms with established functional and clinical significance in other populations. Discovery and validation of novel associations is much more difficult and risky. By using this selection strategy of a small number of known polymorphisms, the investigators also decreased the statistical burden of multiple corrections testing, obtaining higher statistical power with fewer subjects. In other words, true discovery association studies require analysis of hundreds of thousands of candidate polymorphisms. However, if this focused strategy is chosen, it must be selectively applied to a transplant study group whose biology and functional outcomes are likely to match the known associations made in the other types of patients used for the original discovery and validations. Moreover, the results in the chosen transplant population must be informative for clear and mechanism-driven reasons. If done correctly, this approach using mechanism-linked, established polymorphisms to enhance transplant studies is also supported by probability mathematics. The established probability of finding the association in the original disease group is now carried forward to the probability of this association being correct in the transplant patients if the latter are chosen in the study design to properly reflect the underlying biology and physiology. Thus, in the present example, brain death is known to be associated with acute catecholamine release and cardiovascular dysfunction, and these events are known to be associated with functional genetic polymorphisms of the β-adrenergic receptors being studied. In turn, catecholamine excess causing cardiac dysfunction leading to myocardial injury is a “perfect storm” waiting for an organ donor. Thus, if the authors established a significant association to the transplant population, then they would basically have a good start on the mechanism as it is now highly reasonable to assume that it is mediated by these same β-adrenergic receptor pathways.
The obvious downsides to this strategy are that there is no guarantee that any association will be found in the transplant population chosen regardless of how good the hypothesis and this approach using already known polymorphisms precludes any novel discovery. It is also important to emphasize here that there is little to no value in genetic association studies that simply prove that a known association taken from another field is also operant in transplant subjects. For example, simply validating polymorphisms in a posttransplant group already linked to risks of diabetes, obesity or cancer in the general population does not reveal anything not already obvious to any clinician and has at least not to date suggested any new approaches to the field for reducing risks or improving outcomes in transplants. The two objectives should be to clearly establish a statistically significant association and to establish a mechanism-based or clinical practice-based value for that association that contributes something new to transplantation science and medicine.
Another strength of Khush et al. is the proper use of a discovery cohort and an independent validation cohort. One clear point for anyone planning an association study is that validation must be an integral part of the design. Validation is classically the confirmation of a candidate genetic association by testing it in a second, independent cohort. But validation can also be done in some cases by compelling studies of a mechanism using some model system though this is not the same as simply showing immunohistochemical or qPCR evidence. In the present case, Khush et al. are to be congratulated for taking a more nuanced approach to their design and data analysis by appreciating that the management of donors in their system changed significantly after 2006. By resisting the temptation to simply balance numbers and randomize their total subject population into two groups regardless of time or historical events, they recognized the critical value of defining each cohort and their outcome metrics by also considering changes in the actual clinical practices they were attempting to study.
In fact, the genetic associations found in the discovery group were not validated in their second cohort and the authors are straightforward in discussing this result. The failure to validate could mean that the associations were not true. I doubt that for three reasons. First, the discovery cohort is comprised of over 1000 donors, the odds ratios observed are high for associations and the results are significant in all the different race/ethnicity combinations studied. These are all signs that the associations claimed are real. Second, the validation cohort is actually too small for sufficient statistical power particularly as it represents a random race/ethnicity mix. This underlines the impact of the author's correct decision in defining their validation cohort, in this case a negative impact. One could argue here that they should have waited another few years before publishing to increase the validation cohort size. I believe they would respond that the work was compelling enough at this point and that they honestly discussed this issue in the Discussion. The Journal agreed. Third, I think that what actually happened after 2006 is that the diligent professionals involved in their donor management learned from experience to optimize practices without knowing the genetic mechanisms involved. The resulting reduced incidence of left ventricular dysfunction undermined the validation but does not mean that the original association discovered was not true. In fact, the authors explain that after 2006 they routinely used higher doses of phenylephrine when required, which is consistent with the receptor polymorphisms they studied that could reduce the sensitivity of the catecholamine receptors. In a similar way, everyone now knows that high doses of calcineurin inhibitors can lead to damaging and progressive nephrotoxicity but we have all learned by experience to manage these drugs correctly and minimize it to the maximal extent possible. Unfortunately, the authors did not go back to test for a correlation with phenylephrine doses.
If it is true that experience-driven clinical decision choices obscured the genetic associations with β-adrenergic receptor polymorphisms after 2006, it highlights another challenge to any strategy for testing genetic associations in retrospective studies using archived samples. Clinical practice does and must change as medicine evolves and bedside experience with real patients is a powerful driver of change. But in genetic association terms, one challenge is to inform these inevitable clinical changes by study strategies with the emphasis on genetic polymorphisms that reveal real biology and mechanisms and so advance transplantation. Discovering what the clinicians have already discovered and managed can sometimes be scientifically valuable, but will rarely transform the field.
Finally, doing any genetic association study, particularly prospective, therapy-controlled genetic studies, requires relatively large subject numbers for adequate statistical power. The present design had a correctly powered, retrospective discovery cohort but failed to study a similarly sized and powered validation cohort that also represented changes in clinical practice. An alternative design would be to do discovery in a retrospective archive cohort and then validate the association in a properly powered prospective study. Another limitation was the choice to mix all race/ethnicities in their analysis, a fact that was clearly acknowledged and statistically managed. But without including genetic tests of race/ethnicity admixture-defining polymorphisms, the authors are ultimately depending on self-reported designations in highly genetically mixed American populations. It is still a workable strategy if the numbers are large enough in either subjects or clinical impacts on the outcome metrics as demonstrated by the discovery cohort results. But this choice worked against the authors in the validation cohort by adding to the subject number challenge.
Successful genetic association studies, retrospective but especially prospective, usually represent multi-institutional agreements to standardize clinical management and sample acquisition. This cooperation is another challenge in the current climate of academic and economic competition and in this particular area of genetics does not serve the transplant field well. It is much easier to pull a thousand samples from a frozen archive but the downside of this strategy is obvious and the fact that the lead in discovery genetics has been consistently taken by other fields such as cancer, cardiovascular disease and diabetes is not only because these are more common diseases (see recent review of genetic associations for cardiovascular disease (9)). Though transplantation has so many advantages for medical research that have significantly multiplied the scientific and clinical impact of our work historically, in the next phase of genetics discovery and validation we will have to work harder and cooperate more.