The statement that "an important predictor of dyslexia, phonological awareness, can be understood as poor auditory structuring ability applied to language" raises some questions for me.

/1/ If dyslexia correlates with poor auditory structuring ability, it seems very strange that some alleles leading to dyslexia would also lead to musical ability. One would generally assume that musical ability would involve making good (rather than poor) use of auditory information. This leads me to ask whether your research plans to find out what particular aspects of human auditory ability (in people with the music/dyslexia alleles) might enhance musical learning/performance while simultaneously worsening literacy learning/performance.

/2/ Does the possession of phonological awareness indeed predict the possession of poor reading ability? From other research, I had always assumed that possessing phonological awareness went along with good (rather than poor) reading ability.

/3/ Similarly, it seems strange to see phonological awareness (rather than its absence) described as "poor auditory structuring ability." Can you please clarify how the ability to isolate and sequence the successive sounds within a spoken word would constitute "poor" (rather than good) "auditory structuring ability"?

To the Editor: Dear Sir

We read with interest the study by Antoniou et al, [1] in which they compared a number of the described methods for assessing the probability that a BRCA1 or BRCA2 gene mutation is the cause of a family history of breast or ovarian cancer in 1934 families of non-Ashkenazi Jewish origin. In this study a number of methods, particularly the BOADICEA model, demonstrated a high degree of discrimination between families who are mutation positive and negative on current testing. We believe these data fully support the authors conclusion that “More systematic use of these models in clinic… would have the advantage of ensuring equity of access to genetic testing as well as making the management decision making process clearer and more explicit.” In their accompanying commentary, Hopper et al [2] go further and suggest that the accuracy of the best methods, such as BOADICEA, warrant them taking the prime position in the decision-making process when assessing which families should go forward for clinical BRCA1 and BRCA2 mutation testing. When it comes to clinical application, however it appears that the picture revealed by these data is more complex and show that the assessment of families needs to occur within a framework of clinical judgement if the resources available for genetic testing are to be converted into the greatest possible health benefits for these families.

In the study by Antoniou et al, the majority of patients who had been tested through a clinical service had a pre-test probability of detecting a mutation below the 20% threshold suggested in the UK NICE guidelines [3] (as assessed by any of the methods used; BOADICEA, BRCAPRO, IBIS, the Myriad data, and the Manchester Score). Hopper et al., note one reason for this is that the NICE recommendations were only introduced in 2004, but a more fundamental reason is that there remains no accepted “gold standard” for establishing the pre-test risk and indeed the NICE guidelines do not prescribe a specific model for this purpose. In the past the assessment of mutation risk was guided by clinical criteria believed to equate to a high level of risk. However, while these criteria proved sensitive, they suffered from a lack of specificity and over-testing of low risk families was common. More recently, more accurate probabilistic methods [4,5], empirical data [6] and clinical scoring [7] systems have become available that improve case selection, but although all the methods identify a low risk group of families to be excluded from testing, they do not necessarily agree on which families belong in that group. In a cohort of 209 non-Ashkenazi families who underwent BRCA1 and BRCA2 testing in Victoria [8] we found that concordance between the assessment methods examined (which did not include BOADICEA) was low. Overall complete agreement as to which families should be tested (at the 10% threshold used in Victoria) as assessed by BRCAPRO, the Myriad tables, and the Manchester Score (combined score of 15) was only 44%. In low to moderate risk families (BRCAPRO score <30%) the agreement fell to only 12%. We would be interested in the equivalent levels of concordance between the methods in the Antoniou study, but our data suggest that despite the appearance of objectivity, the definition of a low risk family remains highly dependent on exactly which assessment model is used.

In their paper Antoniou et al., report good accuracy for a number of methods in discriminating families across a range of pre-test probabilities – the majority of positive tests occurring in the high risk families as expected. In clinical practice, however, it is rare for the decision to test a high risk family to present a clinical challenge. Instead, further assistance is required with the assessment of low to moderate risk families who represent a sizeable majority of the referrals to familial cancer centres and the bulk of newly diagnosed breast cancer patients. There is evidence in this paper that the methods described are also less effective when assessing low to moderate risk families. For families around the 20% threshold for testing (i.e. 10-30%), the odds ratio that a family with a gene mutation would be selected for testing by any of the methods ranged from 0.28 to 1.35 and in no case was this significantly better than the likelihood that would be expected by chance alone (OR = 1) [Table 1.]. Considered from this point of view, the assessment methods are effective in reducing the overall number of families who would be offered testing in this low-moderate risk group, but this short-term economic benefit would only be converted into improved long-term clinical outcomes and economic benefits (i.e. more women from mutation positive families receiving the right specialist advice and management) if the resources saved could be directed at finding other families who are more likely to be mutation positive to take their place.

We believe that the reported performance in the low-moderate risk groups cannot be interpreted to mean that there is no room for clinical judgment in the decision-making process. Instead these data show that a wider consideration of an individual or families specific circumstances is essential when interpreting the output of these methods. Improved discrimination amongst the large number of families with low-moderate risk based on family history requires further development of the models to incorporate other sources of predictive information such as pathological characteristics and tumour immuno-phenotyping [8,9,10]. Currently this information forms a key component of clinician’s judgement. In addition, where resources are scarce, a clinician assessing a family lying close to the testing threshold must consider more than small differences in pre- test probability, but examine the potential clinical impact of finding a mutation as a whole, for the individual as well as their families, when determining how that resource can be used to achieve the best health outcomes. Although it is possible to imagine an algorithm that was able to account for economic benefits or potential life years saved within a given pedigree based on the outcome of testing, this clearly remains some way off and these considerations also remain within the sphere of clinical judgement.

When we consider the wider context of breast cancer screening and management it is clear that genetic testing represents a very modest cost in comparison to life-long screening programs, risk-reducing interventions or the management of breast cancer. Consequently, it is far from clear that a strategy that merely further restricts access to testing will ultimately deliver the increased equity and efficiency that is hoped for. Uncritical use of the current methods would reduce the chances of identifying mutation carriers who do not have a conventional cancer family history, often due to reasons such as paternal inheritance or a limited pedigree structure [11]. Their modest performance in the low-moderate risk group further suggests that used in isolation they are not well suited to the routine consideration of the underlying genetic aetiology that increasingly forms part of the clinical decision making that takes place with every new breast cancer diagnosis.

Overall we conclude that, despite the encouraging results reported by Antoniou et al, for now, at least in the clinical setting, BOADICEA and other assessment tools should remain an instrument for use by a clinician rather than the other way around.

Paul A James, Marion Harris, Geoffrey J Lindeman and Gillian Mitchell

References:

[1] Antoniou AC, Hardy R, Walker L, Evans DG, Shenton A, Eeles R, Shanley S, Pichert G, Izatt L, Rose S, Douglas F, Eccles D, Morrison PJ, Scott J, Zimmern RL, Easton DF, Pharoah PDP. Predicting the likelihood of carrying a BRCA1 or BRCA2 mutation: validation of BOADICEA, BRCAPRO, IBIS, Myriad and the Manchester scoring system using data from UK genetics clinics J Med Genet 2008; 45: 425 - 431

[2] Hopper JL, Dowty JG, Apicella C, Southey MC, Giles GG, Winship I. Towards more effective and equitable genetic testing for BRCA1 and BRCA2 mutation carriers. J Med Genet 2008 45: 409-410

[3] National Institute for Clinical Excellence. Clinical guideline 14. Familial breast cancer: The classification and care of women at risk of familial breast cancer in primary, secondary and tertiary care. 2004. National Institute for Clinical Excellence.

[4] Parmigiani G, Berry D, Aguilar O. Determining carrier probabilities for breast cancer-susceptibility genes BRCA1 and BRCA2. Am J Hum Genet 1998; 62(1):145-158.

[5] Antoniou AC, Pharoah PP, Smith P, Easton DF. The BOADICEA model of genetic susceptibility to breast and ovarian cancer. Br J Cancer 2004; 91(8):1580-1590.

[6] Frank TS, Deffenbaugh AM, Reid JE, Hulick M, Ward BE, Lingenfelter B, Gumpper KL, Scholl T, Tavtigian SV, Pruss DR, Critchfield GC. Clinical characteristics of individuals with germline mutations in BRCA1 and BRCA2: analysis of 10,000 individuals. J Clin Oncol 2002; 20(6):1480-1490.

[7] Evans DG, Eccles DM, Rahman N, Young K, Bulman M, Amir E, Shenton A, Howell A, Lalloo F. A new scoring system for the chances of identifying a BRCA1/2 mutation outperforms existing models including BRCAPRO. J Med Genet 2004; 41(6):474-480.

[8] James PA, Doherty R, Harris M, Mukesh BN, Milner A, Young M-A, Scott C. Optimal Selection of Individuals for BRCA Mutation Testing: A Comparison of Available Methods. J Clin Oncol 2006 24: 707-715

[9] Lakhani SR, van de Vijver MJ, Jacquemier J, Anderson TJ, Osin PP, McGuffog L, Easton DF. The pathology of familial breast cancer: predictive value of immunohistochemical markers estrogen receptor, progesterone receptor, HER-2, and p53 in patients with mutations in BRCA1 and BRCA2. J Clin Oncol 2002; 20(9):2310-2318.

[10] Lakhani SR, Reis-Filho JS, Fulford L, Penault-Llorca F, van d, V, Parry S, Bishop T, Benitez J, Rivas C, Bignon YJ, Chang-Claude J, Hamann U, Cornelisse CJ, Devilee P, Beckmann MW, Nestle-Kramling C, Daly PA, Haites N, Varley J, Lalloo F, Evans G, Maugard C, Meijers-Heijboer H, Klijn JG, Olah E, Gusterson BA, Pilotti S, Radice P, Scherneck S, Sobol H, Jacquemier J, Wagner T, Peto J, Stratton MR, McGuffog L, Easton DF. Prediction of BRCA1 status in patients with breast cancer using estrogen receptor and basal phenotype. Clin Cancer Res 2005; 11(14):5175-5180.

[11] Weitzel JN, Lagos VI, Cullinane CA, Gambol PJ, Culver JO, Blazer KR, Palomares MR, Lowstuter KJ, MacDonald DJ. Limited Family Structure and BRCA Gene Mutation Status in Single Cases of Breast Cancer. JAMA 2007;297(23):2587-2595.

Dear Editor

We thank the Human Gene Mutation Database (HGMD) team for critically analysing our results, and highlighting some potential problems with our analysis.

Many of the criticisms raised by Stenson et al. relate to mutations that were outside the terms of reference of the study. For instance, a major criticism raised by Stenson et al. was that the review was not comprehensive viz we neglected to compare all categories of mutations. We limited our comparison to single-base-mutations, as it was found too difficult to reliably text-mine other mutation types from OMIM. This could be viewed as indirect discrimination against HGMD but these mutations do represent the majority of characterized mutations in the database and thus represent a reasonable variable to quantify.

We concluded that both OMIM and HGMD were the most comprehensive, but differences between them, including missing genes, highlighted the importance of using both resources when searching for information on a gene.

Another criticism was our failure to distinguish between somatic and heritable mutations but nor did we claim to do so. Both are associated with disease. Also, it is not clear why HGMD does not include mitochondrial mutations, as these are inherited.

Claim 1 - 143 genes are present in OMIM but have no corresponding HGMD entry.

Of the 143 genes identified as not having single-base mutations in HGMD although they are present in OMIM:

• 30% of these genes were missing from HGMD because the database was not up-to-date.
• Another 45% were excluded from HGMD because they did not fit HGMD's criteria for inclusion (mitochondrial, somatic, low quality).
• Although CD2AP is listed in HGMD it does not have any single base mutations and therefore was fairly counted. Presumably this is the case for the other 9% listed in this category.

The remaining 14% deserves further scrutiny. A genuine error did arise with the 12 genes mentioned that included FUZ. This error arose because many genes in OMIM and HGMD do not follow the HGNC official gene names. In our attempt to ensure that we compared the same gene from OMIM and HGMD, we converted all genes to their HGNC gene name, where available. Unfortunately, for a few cases, genes were assigned the wrong HGNC name. An example is the FUZ gene highlighted by Stenson et al. as having no mutations. This was a problem originating from OMIM where the gene described was not found to be a standard name, but a non-unique alias - FY, which is used for two different genes. FY was converted to the wrong HGNC name - FUZ, rather than the correct gene - DARC. This highlights a further problem in the general mutation databases and the importance that datasets adhere to the HGNC to prevent confusion.

The remaining eight genes, which included GPT, are listed as normal polymorphic variants. These are legitimately excluded from HGMD which specifically addresses disease but included in OMIM which has wider terms of reference "a catalog of human genes and genetic disorders". In the context of the purpose of our study which focused on disease, it is fair to call their inclusion into question.

Claim 2 - 226 genes in OMIM contain more mutation entries than HGMD.

Of the 83 genes listed by HGMD:

• 34% are indeed missing or were added to HGMD after our analysis and were fairly counted.
• 28% contained additional somatic allelic variants - we did not distinguish between somatic and heritable mutations in our study.
• 21% have an equal number or fewer mutations than HGMD - this is only true if you include all mutation types rather than just single-base mutations.
• This leaves 18% that contained additional polymorphic variants, haplotypes or non-disease associated variants - we agree that it was probably not fair to include these.

Claim 3 - Many mutations are missing from HGMD that were published in the journal Human Mutation.

Stenson et al. claim that the data presented in Supplementary Table 2 is "highly misleading" and that "HGMD is not missing any of the mutations that the authors claim." We have (yet again) performed searches in the public version of HGMD for the variations listed in Supplementary Table 2 from issues 25(5) - 27(12) of Human Mutation and stand by our original assertion that many mutations are missing from the public version of HGMD.

While the mutations under examination may very well have been entered into HGMD Professional within "1-2 of their publication in the Human Mutation issue cited in the table" as Stenson et al. claim, due to the delayed release of HGMD to non-subscribers, these mutations are not available in the public version, the specific database release our study sought to examine.

Stenson et al. go on to claim that "Several mutations had already been described in the literature prior to their publication as 'novel' in Human Mutation (e.g. PAX6 1410delC)." We note that the source of the PAX6 1410delC mutation (Sale et al. 2004) claims it as a novel mutation, and our inclusion of it, rather than counting against our study, could be seen as an indicator of the inaccuracies of variation databases, as non-novel mutations are being, not only mislabelled as novel, but published as such.

Finally, Stenson et al. note that we included "many neutral and somatic variants." As we outlined above, we never claimed to distinguish between somatic and heritable mutations.

As we noted in our paper, it is not always clear if a particular mutation is included in HGMD due to gene name changes between publications and HGMD, and HGMD’s use of non-standard nomenclature. In these cases, we have endeavoured to give HGMD the benefit of the doubt.

Claim 4 - R158Q in PAH is in error.

We thank Stenson et al. for pointing out the error in the electronic version of our manuscript regarding R158Q in PAH, we have now updated the main manuscript. The Dworniczak et al. (1989) article reports a G>A base change gives rise to R158E, but it gives rise to R158Q. Both OMIM and HGMD had rectified this error in their databases, but there is no annotation to indicate the correction. The original source, which lacks a published correction is cited as the definitive reference. This error has been corrected in our manuscript but it does highlight the need for detailed annotation of mutations eg conflicting reports.

Claim 5 - HGMD is missing two specific genes (COL9A1 and PTCH2) & Claim 6 - Patchy coverage of gene and mutation data in HGMD.

Again, the use of single-base mutations only is a fair assessment as they do make up the majority of HGMD. The purpose of our study was to gauge what data is currently available and HGMD actually did well in comparison to the other databases.

Claim 7 - The authors claim no competing interests.

Competing interests usually refers to commercial interests. The interest of "several of the authors" is not to set up "from scratch a new and all embracing human variation/mutation databases". The aim is a complete collection of variation, and their phenotypes so that they can be curated expertly in locus specific databases that are public and which can be harvested/sent to comprehensive databases such as HGMD and those at NCBI, EBI and UCSC. We are pleased to say that a consortium of members of HGVS (www.hgvs.org) has been responsible for hundreds of LSDBs being created, being publicly available and made use of by HGMD and this effort will continue also under the banner of the Human Variome Project (www.humanvariomeproject.org).

In conclusion, we congratulate the curators of HGMD and OMIM for providing two such crucial resources for inherited disease diagnostics and research. Our study raised a number of explicable problems outlined by Stenson et al., many of which highlight the problems of the mutation databasing field viz: Non use of standard nomenclature, non coverage of all types of variation, lack of annotation of corrections and the need for public and private versions of HGMD due to funding strictures.