On 10th December 2012, UK Prime Minister David Cameron launched a Report on the Strategy for UK Life Sciences One Year On by announcing that the Government has earmarked £100 million to “sequence 100,000 whole genomes of NHS patients at diagnostic quality over the next three to five years”. This ambitious initiative – which will focus initially on cancer, rare diseases and infectious diseases – aims to train a new generation of genetic scientists, stimulate the UK life sciences industry and “revolutionise” patient care.
There is no doubt that this investment offers a major opportunity for the UK to firmly establish itself as a world-leader in medical genomics. However, deciding how best to use the £100M to maximise patient benefit will be a challenge. There are numerous implementation issues, outlined in the PHG Foundation’s response to the announcement. Not least of these is the urgent need for informatics provision to facilitate storage, processing, annotation, interpretation and secure access to both genomic and phenotypic data. This will involve determining appropriate ethical and operational standards across a broad range of questions.
But there is one particularly crucial question that needs to be answered early on: what is the most appropriate assay to use for clinical implementation? All the literature released by the Government, and quoted extensively by the press, states quite categorically that the money will be used for “sequencing whole genomes”. Surely this can’t really be true? (I certainly hope it’s just coincidence that if you multiply a £1000 genome by 100,000 patients you reach the magic figure of £100 million…) If it is the case, there are several major problems.
It would be a mistake to implement whole genome sequencing in the NHS now.
Firstly, although the Government’s Report asserts that “we will soon be able to sequence a human genome for less than £1000”, we can’t actually do this at the moment. In fact, according to data from the US National Human Genome Research Institute, a genome cost $7,666 (around £5k) about a year ago. Even if this were to become possible in the next year or so, the £1000 price-tag for sequencing excludes the substantial costs of labour, informatics and training. At this early stage in planning, it is still unclear how the money will be divvied up, but I think we could safely assume that given these additional and essential costs, perhaps only half the money might be available for generating the sequence itself. Which will not stretch anywhere near 100,000 whole patient genomes.
Secondly, the interpretation of whole genome sequence data is still in its infancy. While we are starting to get to grips with interpreting the pathogenicity of variants in known genes, variants in the other 98% of the genome remain essentially a mystery. This problem is confounded by the reduced analytical accuracy of current next generation sequencing machines for whole genomes versus targeted multigene or exome sequencing, as the depth of coverage may be substantially lower in coding regions. There is a currently a significant trade-off between the extent of coverage (i.e. whole versus partial genomes) and the depth of coverage (i.e. guaranteed accuracy of particular targeted regions). So opting for genomes instead of gene-targeted approaches may actually reduce the diagnostic yield! This is major problem for rare disease sequencing, as the ability to accurately call novel variants in key pathogenic genes may be sacrificed, thus preventing a diagnosis. It is even more of a problem for cancer genome sequencing due to sample heterogeneity; if just a few percent of tumour cells have the critical mutation, high read depths are essential for cancer diagnosis, prognosis and stratifying treatments. Although low average read depths have been successfully applied in population datasets such as the 1000 Genomes Project, a read depth of at least 30x is needed at every possibly causal site (not as an average across the genome) for diagnostic purposes, and even greater for low prevalence somatic mutations.
Finally, we also need to ask: what is the most appropriate test to use in each population? For diagnosing rare diseases, a parent-offspring trio model using exome sequencing is now well established in the literature. This approach allows many thousands of variants to be filtered down to a handful of possible candidates that can then be evaluated, which typically results in a high diagnostic yield. Currently, a whole parent-offspring trio could have their exomes sequenced for less than the cost of a single whole genome. In cancer, the few hundred genes known to be associated with cancer development, progression and treatment response could be targeted and sequenced for the cost of a single gene test, offering a high diagnostic yield across all cancers. High quality data could be obtained using formalin-fixed paraffin-embedded tumour samples which are already collected in hospitals around the country. Moreover, unlike whole genome sequencing, there would be no need to sequence the individual’s germline genome for comparison as population variation data could be used as a normal control dataset, effectively halving the cost of the test. And for infectious diseases (about which I know very little, so won’t say very much), the only sensible approach at this stage would be microbial sequencing not human genome sequencing.
So why the obsession with whole patient genomes? The genome has become a cultural and political icon, a proxy for the incredible achievements of modern science and a beacon of hope for the future of medicine. And it will undoubtedly become an important tool in years to come. But at this point in time, the difficulties involved in clinical interpretation mean that whole genome sequencing simply does not offer either sufficient value for money or clinical utility for mainstream use.
Whole genomes are currently expensive, inaccurate and hard to interpret. More research is needed before they could be implemented in mainstream medicine and many countries are already engaged in major genomic research endeavours. However,the UK National Health Service almost uniquely has the potential to directly translate this technology and embed genomics into clinical practice – an opportunity which could be wasted if the difficulties with interpreting whole genome data decimate the diagnostic rate. The alternative approaches enabled by next generation sequencing – namely multigene targeted sequencing – would increase not only the number of patients who could be sequenced but also the proportion who receive a diagnosis from this initiative.
NHS money should be spent on intelligent genomic strategies that offer the most benefit to the most patients.