Objectives SARS-CoV-2 has evolved rapidly into several genetic clusters. However, data on mutations during the course of infection are scarce. This study aims to determine viral genome diversity in serial samples of COVID-19 patients. Methods Targeted deep sequencing of spike gene was performed on serial respiratory specimens from COVID-19 patients using nanopore and Illumina sequencing. Sanger sequencing was then performed to confirm the single nucleotide polymorphisms. Results A total of 28 serial respiratory specimens from 12 patients were successfully sequenced using nanopore and Illumina sequencing. A 75-year-old patient with severe disease had a mutation, G22017T, identified in the second specimen. The frequency of G22017T increased from ≤5% (nanopore: 3.8%; Illumina: 5%) from first respiratory tract specimen (sputum) to ≥60% (nanopore: 67.7%; Illumina: 60.4%) in the second specimen (saliva; collected 2 days after the 1st specimen). The difference in G22017T frequency was also confirmed by Sanger sequencing. G22017T corresponds to W152L amino acid mutation in the spike protein which was only found in <0.03% of the sequences deposited into a public database. Spike amino acid residue 152 is located within the N-terminal domain, which mediates the binding of a neutralizing antibody. Conclusions A spike protein amino acid mutation W152L located within a neutralizing epitope has appeared naturally in a patient. Our study demonstrated that monitoring of serial specimens is important in identifying hotspots of mutations, especially those occurring at neutralizing epitopes which may affect the therapeutic efficacy of monoclonal antibodies.