Sperm epigenetics in the study of male fertility, offspring health, and potential clinical applications

ABSTRACT

The mammalian sperm contains a highly unique and specialized epigenetic landscape that offers a great degree of interesting research opportunities. One key discriminating feature of the mature sperm epigenome is that it, in theory, represents both remnant marks used throughout spermatogenesis to generate sperm cells competent to perform their function, but also marks that appear to be useful beyond fertilization. Key questions must be asked about the utility of these marks and the multiple purposes that may be served. It is this unique epigenetic landscape that has driven some labs to begin to study the links between aberrant sperm epigenetic patterns and various forms of infertility, from idiopathic to alterations in sperm motility, morphology, and viability and fertilization capacity. Because of the unique nature of the sperm epigenome and the patterns found in mature sperm that appear to reflect perturbations in spermatogenesis that may ultimately have effect on pregnancy outcomes, some researchers believe that these marks may provide predictive insight that can be exploited. Indeed, there is emerging data suggesting that the predictive power of DNA methylation and RNA signatures in sperm likely exceeds that which can be found with traditional assessments of male infertility. This review will focus on the utilization of the sperm epigenome as a potential diagnostic tool in the context of male infertility, as well as the potential difficulties associated with such an approach.

KEYWORDS:

• DNA methylation
• epigenetics
• infertility
• male fertility
• RNA

Introduction

Despite the inherent quiescent nature of sperm, multiple important epigenetic marks exist that create a highly specialized and unique landscape such as nuclear protein composition, DNA methylation, and spermatozoal RNAs [Carrell 2012]. These marks are in large part thought to simply be remnants of epigenetic transitions seen throughout spermatogenesis, though some may be influential downstream during fertilization and embryogenesis as well [Hammoud et al. 2009]. The mature sperm has an extremely unique structure that differs greatly from that of somatic cells due to the replacement of 90-95% of histone proteins with protamines [Oliva 2006]. The addition of protamines to the chromatin architecture aids in nuclear compaction by forming very tight, toroid structures that are necessary for sperm motility as well as protection of DNA from oxidation and fragmentation [Oliva 2006; Jenkins and Carrell 2011].This transition occurs via a stepwise mechanism involving the replacement of canonical histones with intermediary transition proteins and then final replacement by protamines 1 and 2 (P1 and P2) which generally occur at approximately a 1:1 ratio in fertile sperm [Oliva 2006; Ward and Coffey 1991]. The tightly condensed, toroid structures found in mature sperm are in part responsible for the transcriptionally inert nature of sperm due to a lack of available locations for the binding of transcription factors [Oliva 2006].

Considering 5-10% of histones are retained following protamination it is thought that these histones may provide a means for further epigenetic regulation. It has been indicated that histone retention may occur at deliberate locations throughout the genome and that histones can undergo chemical modifications that may further regulate post fertilization transcription [Hammoud et al. 2009; Miller et al. 2010]. The histone protein family relating to nucleosomes is composed of H2A, H2B, H3, and H4. These four proteins as well as a histone protein variant – testes-specific histone H2B (tH2B) are capable of influencing gene activation via various chemical modifications to lysine (K) and serine (S) residues present at histone tails. Gene activation can be influenced by acetylation of H3 and H4, ubiquitination of H2B, or methylation of H3K4 and H3K79 (as recent evidence suggests), with each modification creating a more open chromatin structure [Nguyen and Zhang 2011]. Likewise, inactivation can occur following deacetylation of H3 and H4, ubiquitination of H2A, and methylation of H3K9 and H3K27 [Carrell 2012; Jenkins and Carrell 2011; Yamauchi et al. 2011]. Histone retention appears to be most prominent at promoters of gene families that relate to development, which gives credence to the idea that sperm histone modifications may play a role in transcription in the developing embryo [Hammoud et al. 2009; van der Heijden et al. 2008]. Even more recent evidence suggests that sperm histone modifications do impact the embryo and could potentially have impact on the offspring as well [Siklenka et al. 2015].

DNA methylation is an additional epigenetic mark that has been shown to be a powerful regulator of gene expression in somatic cells. Methylation occurs on cytosine residues in cytosine-phosphate-guanine (CpG) dinucleotides, and is controlled by the DNMT family of proteins [Portela and Esteller 2010]. DNMT proteins are responsible for both initiating and maintaining methylation marks. CpGs found in high concentrations at gene promoters are termed CpG islands and can inhibit transcription when hypermethylated. Lack of methylation marks at CpG islands encourages active transcription by creating an access for transcription factors [Jenkins and Carrell 2011]. Many regulatory processes in the developing embryo require gene silencing such as genomic imprinting and X-chromosome inactivation. Both of these processes are heavily influenced by DNA methylation. Paternal DNA is actively demethylated following fertilization, and methylation patterns are reset in the developing embryo [Panning and Jaenisch 1996]. Ten-eleven translocation (TET) enzymes are largely responsible for this active removal of methylation marks. Studies confirming the generally hypomethylated state of CpG islands at gene promoters relating to development are promising for determining a role for sperm methylation in the epigenetic regulation of the developing embryo [Hammoud et al. 2009; Arpanahi et al. 2009].

A relatively controversial, though intriguing, epigenetic aspect of the sperm epigenome is the role of spermatozoal RNAs. Many potential roles have been hypothesized following the determination that sperm does contain RNAs at very low levels despite being transcriptionally inert [Carrell 2008; Krawetz 2005]. One such hypothesis states that spermatozoal RNAs may have a role in protamination and histone retention due to a regional association supported by the discovery that these RNAs lie in the nuclear envelope [Miller et al. 2005]. Additional hypotheses introduce the idea that RNA transcripts are stable in the embryo following fertilization and can contribute to early embryogenesis and may have a role in regulating gene expression in the embryo [Ostermeier et al. 2004]. We will further explore the potential role of RNAs and other epigenetic marks throughout this review and will attempt to identify the potential utility of these marks; even when causative links between various epigenetic profiles and infertility remain unclear.

The nature of the sperm epigenome, difficulty in study and potential clinical utility

The overarching role of epigenetics in normal cell function is well recognized and has been thoroughly studied in many disease states including various cancers [Momparler 2003]. However, its use as an indication, or cause, of cellular or physiological abnormalities, particularly in the field of fertility, is only beginning to be explored thoroughly and is currently not well understood. This is primarily a result of multiple factors in the study of epigenetics such as the largely continuous nature of the variables involved (and the resultant potential for subtle effects), as well as the poor description of normality, as each cell type has unique signatures with some degree of variation that depends on the region of interest, individual being screened, and type of epigenetic mark being assessed. As an example, if a study identified a 10% change in regional DNA methylation at CpGs within 500 base pairs of the transcription start site (TSS) of a particular gene of interest, how do we determine the potential biological impact? In most cases, with our current understanding of the epigenome, this question simply can not be answered using a straightforward scientific approach. In theory, because of the relatively continuous nature of regional DNA methylation, such a change could cause subtle alterations to gene expression. However, an equally likely scenario (and some may argue a more likely scenario) could be that such a change is biologically irrelevant due to noisy data or that, if real, this would likely represent a ‘silent epimutation’ (a true biological signature that does not affect a change in gene activity). To further compound the difficulty in interpretation of such an alteration, the relevance of a subtle epigenetic change may depend on many factors including the region of interest, as the regulatory paradigm at each region is likely quite variable, as well as the individual and tissue being assessed. Simply put, while we have learned a great deal about the nature of the epigenome, far more still must be understood, and the development of new technologies and techniques is undoubtedly requisite in this effort. Some of these technologies and techniques with extremely high resolution are beginning to emerge. One example is the RNA sequencing technique Drop-Seq [Macosko et al. 2015]. This high power method enables single cell resolution of complete transcriptomes by utilizing a novel microfluidics approach to isolate and to prepare barcoded libraries for each cell at very low costs. The resulting data sets provide remarkable insight into epigenetics at the single cell resolution, which offer tremendous possibilities in both research and diagnostic testing development. Many other new technologies are currently in development and will provide novel tools with which innovators in the field can create new and interesting approaches to basic science discovery, diagnosis, and even treatment.

Importantly, despite our current gaps in understanding, which make it difficult to fully interpret data from studies investigating tissues from abnormal or diseased groups, there is reason to believe that epigenetic data may offer excellent opportunities and applications in the clinic in the near future. While we may not be able to fully test the impact of many epigenetic alterations, we have identified many epigenetic signatures in sperm that are very tightly associated with various sperm abnormalities and/or clinical outcomes [Aoki et al. 2006; Jenkins et al. 2016; Liu et al. 2012]. While very few convincing causative relationships have been established between sperm epigenetic patterns and real clinical outcomes, the associations appear to be real and thus offer potential utility in clinical diagnostic testing. This is likely due to the remarkable nature of the mature sperm epigenome and its reflection of perturbations that may have occurred during spermatogenesis [Hammoud et al. 2014] (Figure 1).

Figure 1. Diagram depicting the nature of the mature sperm epigenome. Based on recent evidence, during the process of spermatogenesis DNA methylation remains quite stable while RNA expression and nuclear protein composition is altered significantly. The mature sperm epigenome reflects these marks and contains remnant epigenetic signatures from this process, but also appears to have marks that are influential in fertilization, embryogenesis, and beyond.

Great potential

The efficacy of utilizing sperm epigenetic signatures as potential markers for various fertility related diseases and/or predictors for success in assisted reproductive techniques (ART) is becoming more widely explored [Jenkins et al. 2016; Jodar et al. 2015; Aston et al. 2015]. The potential of this approach is very real and, for a number of reasons, represents an excellent opportunity for further exploration.

The advent of various technologies to quickly, reliably, and relatively inexpensively screen large portions of the epigenome with high resolution has created large amounts of available data that can be mined to improve our current understanding of the relationship between the sperm epigenome and various fertility phenotypes [Hrdlickova et al. 2016]. RNA sequencing is among these powerful tools that has allowed for the assessment of multiple types of RNA, from non-coding RNAs (ncRNA) to mRNAs [Hrdlickova et al. 2016; Jodar et al. 2013]. While total RNA content is very low in sperm, multiple studies have suggested that various forms of RNA in the mature sperm may play a role (or be indicative of normality) not only in sperm development but also in embryogenesis [Jodar et al. 2013; Jodar et al. 2015]. While controversial, there are data that support the idea that members of the miRNA 34C family are important in the embryogenesis process [Yuan et al. 2015; Liu et al. 2012]. Further, recent data has also identified 648 RNA ‘elements’ which are all required to be present to facilitate the best chances at successful IVF outcomes [Jodar et al. 2015]. In effect, this study showed that the absence of even a single member of this group of RNA elements in sperm is associated with decreased success rates in various fertility treatments [Jodar et al. 2015]. Further work from multiple authors have demonstrated the links between sperm RNA transcripts and pregnancy outcomes. One study has shown that sperm gene expression patterns (as measured by TaqMan Array) are predictive of pregnancy outcomes from IUI [Bonache et al. 2012]. Further, work from 2010 also described similar findings, which suggest that the sperm transcriptome is quite distinct between individuals who achieve pregnancy through IUI and those who do not [Garcia-Herrero et al. 2010]. While much of the RNA content in sperm appears to be ‘remnant’ (utilized in the process of spermatogenesis and simply left over in the quiescent mature sperm), there do appear to be important transcripts present for the function of the mature sperm and potentially the embryo and beyond [Liu et al. 2012; Yuan et al. 2015; Jodar et al. 2013; Jodar et al. 2015].

DNA methylation analyses have progressed greatly in recent history and the available techniques are commonly used in the assessment of sperm DNA methylation patterns [Tang et al. 2015]. In humans, multiple options exist for the rapid assessment of DNA methylation profiles at sites that are potentially important for transcriptional regulation. One of the simplest approaches with tremendous coverage is Illumina’s HumanMethylation450 Bead Chip array, which has recently been replaced by the MethylationEPIC array with over 850,000 methylation sites probed [Moran et al. 2016]. This simple format allows for the rapid and reliable screening of the majority of well-annotated gene promoters, CpG islands, multiple enhancers, as well as gene body methylation sites with single base pair resolution. While whole genome bisulfite sequencing still provides the most information about a given sample, multiple variations of reduced representation bisulfite sequencing (RRBS) exist that also provide excellent data with more limited coverage at a lower cost [Fouse et al. 2010]. With the available restriction enzymes that can be used independently (or in concert) during library preparation, RRBS is a versatile tool that can be tailored to cover many regions of interest for each unique project. Each of these techniques potentially yields tremendous amounts of data that may provide informative patterns that are associated with various forms of fertility phenotypes. Additionally, each of these techniques is potentially scalable and could have utility as diagnostic tools. As an example, recent data generated with Illumina’s 450K array has identified signatures in mature sperm methylation patterns that appear to be predictive of the likelihood that an individual will need to go through the IVF process or if less invasive therapeutic interventions may be effective [Aston et al. 2015]. This is in stark contrast to the poor predictive power of many currently utilized tests of male fertility. Clearly, there is great potential in using these techniques in the clinic, but a great deal of further work is required.

A notoriously low bar

It is well established that there are significant limitations to currently available diagnostic tests to ascertain a man’s reproductive potential [WHO 2010]. In fact, many have cited the need to identify improved approaches with better predictive power specifically considering pregnancy outcomes as well as the most likely therapies to improve these outcomes based on diagnostic test results. Although some semen analysis results suggest associations with decreased fecundity, more often these results are insufficient to predict pregnancy outcomes or to guide clinical decision making beyond general recommendations. The variability between ejaculates further complicates this problem making it particularly difficult to draw any conclusions about the actual status of an individual’s ‘fertility’ or their ability to sire offspring [Poland et al. 1985].

While many still find value in the basic semen analysis measures as it can identify some of the most severe fertility disorders in men, it is still unclear what semen analysis data suggest about a population or a single sperm’s real potential to fertilize an egg and generate healthy offspring. In addition, the regular semen analysis provides little information regarding potential underlying problems in spermatogenesis that are the ultimate cause of the disorder. The latter, in particular, has not been fully explored and yet is essential to better understand spermatogenic deficits and develop targeted medical interventions in male infertility. This is not surprising since studies of adult germ line stem cells and other sperm precursors are difficult to perform in human due to the difficulty in obtaining such tissue (testicular biopsies). One key area of potential in assessing the health of spermatogenesis via observation of the mature sperm only (which is ideal as it is easily collected with generally sufficient cell numbers) is assessing sperm DNA methylation patterns. The key benefit of such an approach is that, based on currently available mammalian studies, DNA methylation signatures remain remarkably stable throughout the process of spermatogenesis [Hammoud et al. 2014]. While RNA transcripts and histone placement/modifications change dramatically throughout sperm production/maturation, with the histone to protamine transition being likely the most dramatic example, DNA methylation marks appear to be largely unchanged from the adult germline stem cell to the mature sperm (Figure 1). Thus, studying this mark in the mature sperm potentially offers unique insight into the process of spermatogenesis in normal cells and some concept of aberrant processes in abnormal cells. This represents an excellent first step to understanding the etiologies of commonly seen sperm disorders. Further, despite the difficulty in obtaining testicular biopsies, a better understanding of the impact of altered spermatogenesis on sperm function and potentially on pregnancy outcome is required to further our knowledge of these processes and allow us the opportunity to identify potential areas for innovation and intervention.

The utility of sperm epigenetics

The study of male infertility and associated sperm abnormalities presents many challenges [Carrell and Aston 2011]. Foremost among them is the multifactorial nature of these disorders. In effect, it is likely that the same, or highly similar, altered fertility phenotypes arise from multiple different etiologies, making identification of common causative elements quite difficult. As an example, an excellent review published in 2013 outlines the many potential causes of oligozoospermia alone [McLachlan 2013]. It is not surprising then that the typical diagnostic tests that focus only on identification of a phenotype (that may arise from any number of underlying abnormalities), often fall short in their predictive capacity. While some etiologies (such as oligozoospermia or asthenozoospermia, for example) may be associated with fertilization abnormalities or poor pregnancy outcome, many others are not (thus the not uncommonly seen occurrence of pregnancy in couples whose male partner displays one or more abnormal semen analysis measure). As a result of this inherent shortcoming in diagnostic testing, in most infertility cases the true causative abnormality of the most important measure (pregnancy outcome) is elusive. In consideration of this overarching difficulty in the fertility clinic, it appears that other testing options must be pursued. The search for such screening techniques must emphasize a real potential to enable enhanced pregnancy prediction to have real value in clinical decision-making.

While still not fully vetted, the use and analysis of large epigenetic data sets (primarily RNA seq and DNA methylation data), appears to have the potential to provide more informative and predictive diagnostic value than the traditionally utilized male fertility testing techniques [Aston et al. 2015; Jodar et al. 2015]. This is in large part due to the fact that the analysis of ‘big data’ with the currently available bioinformatics tools and predictive model building is better suited to identifying difficult to discern biological patterns (such as those seen in the analysis of multifactorial diseases). These data analysis methods are being successfully employed in diabetic cohorts to predict disease progression of other complex diseases including retinopathy, neuropathy, nephropathy, and heart disease [Cichosz et al. 2016]. With these tools, subtle signatures that are seen in groups of individuals who share similar phenotypes can be most easily identified, and predictive models can be constructed. Such an analysis is reasonable biologically as the epigenome offers a unique opportunity to study altered cellular processes in general. In effect, the epigenome in mature sperm is very likely affected by (or affects) large scale alterations associated with infertile phenotypes and as such, when analyzed, has the potential to reveal some of these subtle patterns that would not be found by typical approaches that rely on altered methylation signatures at one common loci or abnormal expression of a single RNA molecule in the majority of cases. Instead, analyses of large-scale datasets of RNA expression or DNA methylation have the potential to identify and cluster samples with similar epigenetic profiles that move beyond the patterns at a few select genomic regions. Data from a recent analysis of embryo quality and epigenetic signatures supports this assumption [Aston et al. 2015]. Specifically, while no clear differentially methylated regions were identified (between poor and high quality embryo development measures), bulk analysis of CpGs in sperm that were not individually informative of embryo quality demonstrate clear predictive value increases. Interestingly, the predictive value appears to increase with the inclusion of more data (CpG counts as high as 480,000 from array data), despite the fact that these CpGs are not independently predictive of embryo quality when observed individually. Though there is much that still must to be learned and a great deal of additional testing and refining is required, it appears that there is real value in assessing large epigenetic data sets to develop predictive tools with diagnostic value in the treatment of male infertility.

Limitations

There are inherent difficulties in the study of epigenetics. Though not prohibitive, these difficulties are pronounced in the study of mammalian sperm and specifically in the context of using the sperm epigenome to develop predictive tools to assess an individual’s ability to sire offspring. Particular caution should be paid to the potential of somatic cell contamination. While not catastrophic, it is important to note this as a potential limitation that must be carefully monitored and, where possible, significantly decreased or eliminated with a bench top assay or in silico.

Epigenetic signatures (DNA methylation, RNA, and nuclear protein composition) are quite unique in sperm when compared to any somatic cell type. It is this inherent difference that makes the identification and removal of somatic cells so essential. In effect, even a small degree of differences in somatic cell contamination levels between two study groups (if not properly addressed) will very likely result in significant findings that are not a result of sperm biological differences, but instead an issue of differences in the cell populations being analyzed. It is reasonable to consider that in collected human semen samples many contain different somatic cell types including white blood cells (commonly found), epithelial cells from many origins, and even vaginal epithelial cells (when collected via intercourse with a sterile condom). While somatic cell contamination can cause dramatic issues with epigenetic data analysis (particularly in the case of sperm epigenetic studies), there are ways to avoid such issues. Any study assessing epigenetic alterations between two groups should carefully evaluate the likelihood of contamination, and every effort should be made to remove contaminating cells prior to processing.

Once sample processing has been finalized there are additional steps that can be taken during data analysis steps to remove the issues of somatic cell contamination. When utilizing array-generated data, as is commonly done in the case of DNA methylation in human samples, it is essential to consider these limitations as these datasets can be particularly susceptible to misleading data from contaminates. DNA methylation array data is effectively a composite average of all cells in the population from which DNA was extracted. The ability to identify and remove contamination signals is extremely difficult though some analytical efforts have been made in an attempt to determine the proportion of various cell types in a given analysis. This has been effectively performed in blood, as the composition of cell types is essential background information upon which your data analysis must be based [Houseman et al. 2015]. Because of this, white blood cell contamination can be effectively removed in silico (by utilizing bioinformatic tools that have been constructed using these available data to determine and remove contaminating signals). However, in sperm vs. somatic cell assessments, there are additional nuances to consider, not the least of which is the likely presence of other somatic cell types beyond white blood cells. Some of the most discriminating genomic loci between somatic cells and sperm are imprinted regions that appear to be altered in sperm in some cases of infertility [Kitamura et al. 2015]. Thus, utilizing these marks to discern between sperm and somatic cells may be inherently flawed. Sequencing technologies available today do offer the ability to remove some regions that appear to come from contaminating cell types as single reads and whole nucleic acid sequences can be assessed and removed. While we are able in some cases to remove or account for a degree of this contamination, every effort should still be made to remove these contaminating cells prior to analysis as they can reduce power of epigenetic analyses by either decreasing the number of reads that can be confidently included in an analysis or by increasing the overall noise in the data, making identification of subtle marks far more difficult. Taken together, it is essential in the development of sperm epigenetic testing protocols that robust somatic cell removal techniques are employed (either prior to processing or in silico) to ensure the highest quality, and most predictive, data.

Potentially the most notable of all limitations associated with the design of any new or improved diagnostic test targeting male factor infertility is the fact that this diagnosis is associated with a condition that affects the interaction between two individuals. Inherently, all diagnostic tests of either female or male infertility are limited in this way. In testing the efficacy of a given diagnostic approach it is important that every effort is made to isolate male factor infertility so that results can be properly interpreted. Further, in practice, using such tests that only target male infertility will be difficult if considered independently. In fact, in most cases these types of tests may only be useful diagnostically when there is a lack of strongly significant female factors and other extremely severe abnormalities (severe oligozoospermia or azoospermia) have been ruled out. Despite this, there still remains a great need for innovation and improvement in testing for infertility in general, and in particular for male factor issues.

Conclusions and future directions

The sperm epigenome is remarkably unique and offers interesting opportunities for study. Great efforts have been exercised to understand the nature of this epigenetic landscape and the role of sperm epigenetic patterns in normal sperm development and functionality, as well as in the embryo and the offspring. While clear causative relationships between sperm epigenetic signatures and fertility phenotypes have been elusive, a great deal of data suggests associations between the sperm epigenome and various forms of infertility and poor pregnancy outcomes. Because the sperm epigenome provides a general view of cellular activity, these often-overlooked marks appear to have the potential to predict outcomes more accurately than traditional methods for measuring sperm quality. Further, because of the nature of these marks, subtle patterns can be identified even in cases where no clear causes (genetic or otherwise) can be elucidated. Perhaps in many cases the causative perturbation of the infertile phenotype is not epigenetic in nature and instead, an assessment of the sperm epigenome simply reflects altered cellular activity that is associated with infertility. Further work must be performed and new technologies developed to enable the targeted exploration of epigenetic signatures in hypothesis driven research.

Many of the limitations mentioned herein will begin to be solved as increasing high resolution techniques emerge. Already we are seeing the development of elegant approaches to test epigenetic signatures at the single cell level. As mentioned earlier, there are single cell techniques that are now being implemented in labs throughout the world that not only offer unparalleled resolution, but also drastically reduced costs of single cell level epigeneomics [Macosko et al. 2015]. Excellent reviews are available that address this trend to high resolution epigenetic analysis and these techniques should be rapidly adopted in the study of germ cells for basic science discovery and to determine potential utility in the clinic [Schwartzman and Tanay 2015; Bock et al. 2016].

The implementation of new technologies and careful consideration of study design and potential limitations will enable us to answer some of the most difficult questions in complex diseases. Particularly in male factor infertility, a great deal of work still must be done to understand the nuances of the disorder and the real driving factors behind the condition. Despite our need for more data to fully elucidate the dynamics of male infertility, our current understanding and ability to utilize epigenetic signatures as a snapshot of potential aberrant cellular functionality offers real potential to improve diagnostics in the field of infertility and beyond. Future discoveries will facilitate a great degree of innovation and will provide more data with which diagnostics can be refined or further innovated.

Declaration of interests

TGJ, DTC, and KIA hold IP in sperm epigenetics, aging and infertility. DTC is also an owner of Episona. ERJ has nothing to disclose.

Additional information

Notes on contributors

Timothy G. Jenkins

Overall outline of review: TGJ, KIA, DTC; Writing: TGJ, ERJ; Review: KIA, DTC.

Kenneth I. Aston

Overall outline of review: TGJ, KIA, DTC; Writing: TGJ, ERJ; Review: KIA, DTC.

Emma R. James

Overall outline of review: TGJ, KIA, DTC; Writing: TGJ, ERJ; Review: KIA, DTC.

Douglas T. Carrell

Overall outline of review: TGJ, KIA, DTC; Writing: TGJ, ERJ; Review: KIA, DTC.