As with the advent of many new diagnostic technologies in medicine, the use of microarrays has provided unprecedented power to detect biochemical, morphological, and physiological differences, sometimes of unknown or unconfirmed clinical significance. This power also creates challenges that have important ethical implications that should be considered. Many of the ethical issues raised by microarrays are not new. For example, clinical geneticists and genetic counselors have extensive experience thinking and communicating about ethical issues with their patients, and many of the issues in microarrays parallel issues involved with other genetic diagnostic techniques. Furthermore, many of the ethical issues involved with the application of microarray technologies exist broadly for many new diagnostic technologies.

The use of microarrays raises at least four issues that exist broadly for many new diagnostic technologies. Although the ethical issues in microarray application to complex traits may not all be novel, we suggest that these issues merit consideration and reexamination in this context. First, do we understand normal variation enough to know what is “abnormal” or “pathogenic”? If not, use of the tests could cause more harm than good to patients and their families because of the probability that they provide inaccurate information that may lead to unnecessary anxiety, clinical procedures, or even termination of pregnancy.¹ Second, because new technologies like microarrays allow inferences to be made about traits that are not necessarily associated with morbidity or mortality, or traits that are nonmedical (such as personality or hair color), there is concern that the availability of the technology may increase pressure on prospective parents and clinicians to screen for “imperfections” to eliminate them.² Such screening raises the specter of eugenics³ and raises questions about the appropriate role of the clinician in providing information that is not about health and disease.

Third, new powerful technologies, like the comprehensive and detailed analysis of the genome afforded by microarray analysis, creates a greater potential for incidental findings. Incidental findings are those that are identified by the application of a new technology that were not part of the original scope of what was being examined. This is a common theme of new diagnostic technologies: their ever-increasing ability to observe phenomena before the clinical significance is known. For example, radiologists are increasingly encountering unidentified bright objects, or UBOs, in magnetic resonance images (MRI) of the brain. A study of MRI scans of healthy children found 13% with “abnormalities.” However, on follow-up examination and after re-scanning 24 months later, none of these UBOs was found to be of clinical significance.⁴ In addition, the rates of incidental findings in neuroimaging vary widely, including reports of 20% of scans with incidental findings among asymptomatic individuals⁵ or even 40% among research participants.⁶ However, much lower rates of findings with clinical significance suggest the possibility of high false-positive rates.⁷ Results like these suggest that applying these new diagnostic tools in a clinical context prematurely may be unethical because it increases the potential for harm to patients, either because of over-treatment, under-treatment, unwarranted labeling and stigmatization, or a false sense of security.

Fourth, despite our current lack of a comprehensive understanding of the implications of microarray analysis, the large number of loci analyzed will create an unprecedented volume of information for patients, counselors, and health care providers. This makes it very challenging to envision what adequate informed consent means and how to achieve it.

However, we would add to this set of criteria by suggesting that the issues raised by this kind of research extend beyond the question of whether the design and conduct of the research itself is ethical, to the presentation and interpretation of the research results. This is especially relevant in the context of complex traits. Array CGH provides new opportunities to look for the genetic components of diseases and traits for which other techniques have been less successful. Complex traits, with multiple environmental and genetic risk factors, have not successfully yielded to linkage or candidate gene approaches, and only recently are genome-wide scans being applied in the attempt to characterize the genes involved. This failure to identify causal genes is particularly evident for traits that are involved in neurodevelopment and cognition, such as autism and mental retardation. Many of these traits do not have clear patterns of heritability in families; yet there is a strong desire on the part of parents and clinicians to understand the etiology, identify a clear diagnostic with clear predictive value, and avoid having children born with the condition.

In the context of this failure to convincingly identify genes, new ways of understanding genomic variation and new technological tools are exciting because of their potential promise for yielding answers where few have previously come. It is for this reason that many research groups are moving quickly to apply these new approaches to samples from complex traits.

To make specific recommendations about the ethical implications of this kind of research, we analyzed publications in PubMed of original studies of autism using array CGH (using keywords “autism” and “array” or “copy number”) published April 2005 through April 2007. This search yielded 14 publications that fell into three main categories.

The first category contained studies explicitly focusing on the association between copy number variants and autism, either idiopathic or syndromic.^9–12 These studies ranged in population/sample size, from the largest at 6709 samples,¹⁰ an intermediate population with 389 samples,⁹ and two smaller populations of 29 samples¹¹ and six samples.¹²

A second category focused on genes or genomic regions either with some evidence of association with autism themselves or with syndromes known to have autistic phenotypes.^13–19 These included regions in chromosome 15q, including those involved in Prader-Willi and Angelman syndromes,^14–17 Williams-Beuren syndrome,¹⁹ patients with corpus callosum anomalies,¹⁸ and patients with autism itself.¹³

The third category included studies identifying novel deletions or deletion syndromes in single individuals or very small groups of individuals that included autistic phenotypes.^20–22 These studies resulted from applications of array CGH in larger studies of autism,¹⁹ idiopathic intellectual disability,²¹ and genetics clinical practice.²² They included deletions on chromosomes 2, 21 17, 20 and the X chromosome.²²

Without more data on the frequency of these kinds of changes, de novo or inherited in the population, it is difficult to make conclusive statements about causation. For sequence variations and microscopic changes, about which much more is known, variants can be more easily distinguished among those recognized or likely to be a cause of a disorder, those that are not likely to be causal and those of unknown clinical significance.²³ However, at this stage of research, the finding of a CNV not previously identified does not in itself constitute evidence for pathogenicity.

The articles also vary in the language and terms they use to describe causal relationships and clinical relevance. Although the use of language may seem like an issue of mere semantics, it has powerful societal implications for diagnosis and identity for people who have autism, a disease for which no genes or other causes have been consistently identified. The recently published article by Sebat et al. in Science represents the extreme of confidence about causation, with the title: “Strong association of de novo copy number mutations and autism.” The authors also state, “the strong association of de novo CNVs with ASD is consistent with such mutations being a primary cause in most cases rather than merely contributory.”⁹ These authors base their evidence of causation on the difference in CNVs identified in cases and controls, stating, “Since the difference in rate between autism and control is so marked, we can make a fair presumption that many of the lesions we observe are contributory to the disorder.” However, this difference was only 10% (n = 12) in cases compared with 2%⁵ in controls, which was statistically significant using their analytic approach, but it is arguable whether “marked” or “significant” translates to a broader contribution to the disorder.⁹

A study by Jacquemont et al. published in 2006¹¹ applied array CGH to 29 patients presenting with syndromic autism spectrum disorders (ASD), or ASD associated with other clinical features such as facial dysmorphism, limb or visceral malformations, and growth abnormalities. They referred to the identified CNVs as “clinically relevant rearrangements” and “pathogenic” because they were not previously identified in normal variants in the general population and were either de novo, or not present in either unaffected parent, or occurred only on a maternal X chromosome that was then inherited by an affected son. None of the variants were recurrent or overlapped in the eight patients.

Other articles are more cautious in the use of language around causation. They state that their goal is to identify potential candidate genes for autism, which would then need to be examined in different, larger, and more generalized populations.¹⁹ Others cite the need to exclude known rare causes of ASD to reduce the noise in linkage studies, candidate genes studies, and whole-genome studies and allow for the easier identification of candidate genes.¹⁰

The clinical and phenotypic definition of the cases is also variable among studies. In the case of autism, it is important to consider what diagnostic criteria were used for recruitment and whether the study is focused on idiopathic or syndromic autism. How representative are the cases of the majority of autism in the population? Are the cases included in the study from multiplex or simplex families? What kinds of conclusions are made about the phenotypic differences and causality? Additionally, what controls, if any, are examined? Are controls unaffected family members or unrelated individuals? For unaffected siblings, were they evaluated for the presence of any autism-related phenotypes? All of these questions affect the interpretability and generalizability of the results.

Early decisions in study design and data collection may also affect other aspects of array CGH analysis. Genomic DNA may not be available for any samples, particularly when DNA banks are being used as sources; therefore, researchers may not be able to confirm that the CNVs are not just an artifact of immortalization of cell lines. Cell lines from affected patients may have been treated differently than those of control samples, including more cell line passages, perhaps leading to an increase in chromosomal changes that are artifacts of immortalization. Furthermore, researchers may not have collected genomic DNA from parents or unaffected siblings; therefore, it may not be possible to determine whether CNVs are inherited or occur de novo.

Second, most array CGH findings require validation on genomic DNA samples, yet many studies do not have genomic DNA available and only have lymphoblastoid cell lines available for analysis. Investigators must choose either not to validate findings or to recontact participants to obtain genomic DNA. Such recontacting is not ethical if the informed consent says that participants will not be re-contacted or if the investigators are using samples from a databank and are not the original investigators and therefore should not have access to identifiable information. Even if consent forms and investigator relationships made recontact ethically acceptable, it would be challenging to recontact participants without alarming them about the reason.

Third, it is important to consider whether the use of array CGH might ever require investigators to recontact subjects under any circumstances because of the pathogenic importance of any finding, even an incidental finding. Such findings may include, for example, CNVs that are known to be associated with severe conditions for which some preventive therapy might be appropriate, or for which cautious reproductive decisions might be warranted (for example, large changes on the X chromosome in the mother). If CNVs known to be associated with clinically significant conditions are included in arrays used for research, it is important for researchers to consider beforehand whether and how to convey this information to subjects and how subjects should consider the possibility of receiving such information during the consent process. To the extent that array CGH is used to evaluate chromosomal integrity generally (as opposed to targeted analysis of specific loci), it may generate findings of known or likely clinical significance, such as a large deletion, even if they were not specifically sought as part of the research question, or even if the chromosomal abnormality is not known to be the cause of any particular phenotype of research interest. In these cases, the research team should have plans for results to be validated and provided to the subject by a CLIA-certified lab. However, we realize that some of these findings may be ambiguous in terms of clinical validity and that a CLIA-certified lab that can perform an appropriate test to validate a research finding may not exist. In these cases, the nature of the finding is likely to be too preliminary to offer to research subjects.