Large-scale evolutionary surveillance of the 2009 H1N1 influenza A virus using resequencing arrays

Lee, Charlie Wah Heng; Koh, Chee Wee; Chan, Yang Sun; Aw, Pauline Poh Kim; Loh, Kuan Hon; Han, Bing Ling; Thien, Pei Ling; Nai, Geraldine Yi Wen; Hibberd, Martin L.; Wong, Christopher W.; Sung, Wing-Kin

In order to enable large-scale identification of variations of H1N1(2009) viruses from multiple patient samples, it is necessary to develop a low-cost method for rapidly wholegenome sequencing the H1N1 samples. Historically, sequencing of viral genomes is performed using standard dye termination technologies. These conventional sequencing technologies produce accurate data but are too slow, costly and labour-intensive to be practical for large-scale epidemiologic or evolutionary investigations in viral outbreaks. Oligonucleotide resequencing microarrays that are capable of identifying nucleotide sequence variants may offer an alternative solution (2, 3) and in recent years, have been used for detecting and subtyping influenza viruses (4, 5) . By analysing sequences generated from tiling probes across targeted regions of various strains of the influenza virus [e.g. partial fragments of the haemagglutinin (HA) and neuraminidase (NA) genes], important information such as viral subtypes, lineages and sequence variants can be determined. Apart from influenza, resequencing microarrays have also been used to obtain whole-genome primary sequences for orthopoxviruses (6) , biothreat viruses (7) and SARS (8) . The reported studies mainly use platform accompanying software that employs probabilistic base-calling algorithms such as ABACUS (3) and Nimblescan PBC (8) . Although statistically sound, these methods are susceptible to hybridization noise caused by factors such as poor probe quality, poor amplification or mutations. This results in numerous ambiguous and false positive base calls that may affect the accuracy of downstream evolutionary analysis. Efforts have been made to improve the call rates and accuracies of existing probabilistic base-calling algorithms. For example, Model-P uses probe and sequence features to build intensity-prediction models that compute maximum likelihood scores for base-calling (9) . Another approach filters low-confidence base calls from problematic regions (e.g. regions with high mutation rates or repeats), thereby reducing the number of false-positive base calls (10) . Depending on the stringencies of the filters used, call rates may suffer as a result.