Prenatal counseling and the detection of copy-number variants

To the Editor: Penetrance is defined as the proportion (or percentage) of individuals with a genotype known to cause a disease who have signs or symptoms of the disease. Although genotype has classically referred to individual genes or mutations, the term has also been applied in the context of copy-number variations (CNVs) detected by microarrays. Vassos et al.¹ proposed calculating penetrance for CNVs associated with schizophrenia using a Baysian approach that considered the overall population probability of an individual having the disorder, P(D); the probability of a specific CNV in individuals with the disorder ; and the probability of finding the CNV in subjects who do not have the disorder . Rosenfeld et al.,² in their article titled “Estimates of Penetrance for Recurrent Pathogenic Copy-Number Variations,” have extended the approach to calculate the probability of any abnormal pediatric phenotypic outcome given the detection of some of the more commonly encountered CNVs. However, the calculations of Rosenfeld et al.² are based on some questionable assumptions that materially affect the results.

To calculate the posterior risk following the detection of a CNV, Rosenfeld et al.² consider the overall population incidence of a disorder (prior risk) to be equal to the incidence of any genetic disorder that might be seen in a child or young adult (excluding gross chromosome abnormalities). This disease estimate of P(D) = 0.05 is based on data in the British Columbia Health Surveillance Registry collected during 1952–1983.³ For the probability of each CNV in an affected population, , Rosenfeld et al.² used the data from a heterogeneous group of referrals for microarray testing, heavily weighted toward individuals with developmental delay/intellectual disability, epilepsy, and autism spectrum disorders. Presumably, this also preferentially included referrals for which there was a high suspicion for a syndrome caused by a microdeletion or for which there were multiple anomalies. It is not clear precisely what proportion of cases had disorders that were potentially caused by a CNV, but it seems clear that this referral population did not include the full spectrum of disorders that are in the British Columbia Registry. This is because there would be no indication for microarray analyses on most cases with dominant disorders (which could be established by family history and/or phenotype), many recessive diseases (e.g., identified by phenotype, newborn screening, prenatal carrier screening), or many other conditions (e.g., strabismus, clubfoot, and hip dislocation) for which there has been little or no data suggesting microarray may be informative. For multifactorial disorders, the environmental factors that determine phenotype may also substantially differ between the Registry cases and contemporary test referral cases. The probability of a CNV in the unaffected population,

, should not include any patients with phenotypes associated with the disease, but it is not entirely clear that control groups described by Rosenfeld et al.² were all disease free.

The following example, for illustrative purposes only, shows how the calculations by Rosenfeld et al.² might provide incorrect estimates of disease risk. Suppose that the 0.05 disease incidence consists of two types of disorders: type A, comprising 0.04 abnormal cases, are those with phenotypes that are unlikely to have a genetic disorder detectable by microarray and would not typically be referred for the test; and type B, which are the remaining 0.01 that are appropriate for microarray testing and match the referrals included by Rosenfeld et al.² Incidences of the CNVs in the affected and unaffected populations appropriate for testing are assumed to be as reported by Rosenfeld et al.² Results are presented for the distal 16p11.2 deletion and 15q11.2 deletion, which Rosenfeld et al.² report to have the highest and lowest risk for abnormality of the CNVs they evaluated. As can be seen from Table 1 , the net risk associated with the presence of the CNV is substantially lower under the A or B model. In fact, overall risk for abnormality if the 16p11.2 deletion is present is 28% or about five to six times the background (5%) risk, whereas overall risk if the 15q11.2 deletion is present is 6%, which is very close to the background (5%) risk. Both of these risks are substantially lower than those calculated by Rosenfeld et al.² The term “penetrance” is not appropriate because the probability calculated is for all abnormalities, not just those associated with the CNV.

Table 1 Probability of an abnormal outcome when there is a 4% risk for a disorder unrelated to those detectable by microarray (type A) and a 1% birth incidence of a disorder that could be associated with a CNV (type B)

Although the proposed Baysian approach to calculating the significance of a CNV for a broader set of disorders seems to be theoretically possible, I suggest that there is currently insufficient reliable data to generate the posterior risks. This is because the phenotype associated with each CNV is poorly defined and therefore the prior risk and prevalence of CNVs in appropriate populations are uncertain. Moreover, the diagnosis of disorders such as autism and schizophrenia is often imprecise and there may also be ascertainment bias in diagnosis, e.g., a patient who is initially considered to have an “uncertain” diagnosis is revised to “affected” following the detection of a CNV. Alternatively, a CNV is identified in a patient with an entirely unrelated disorder and it is assumed that the CNV is causal. This latter bias may explain why very high rates of CNVs associated with autism and neurocognitive alterations were reported in prenatal microarray studies performed because of fetal structural abnormalities identified by ultrasound; the ultrasound information was used to evaluate whether the CNV was likely to be clinically significant.⁴

Invasive prenatal tests are now mostly used to confirm a specific chromosome abnormality detected through cell-free fetal DNA tests and/or ultrasound and to provide reassurance following other screening. For many women, the presentation of a finding of a CNV of uncertain clinical significance may be very unhelpful. The challenge posed by using microarray testing needs to be met through enhanced professional education about the strengths and limitations of the testing, individualized counseling of women considering the test, and guidance on test utilization and interpretation from professional groups such as the American College of Medical Genetics and Genomics.

Disclosure

The author declares no conflict of interest.