Login / Signup

Microsatellite instability profiles of gastrointestinal cancers: comparison between non-colorectal and colorectal origin.

Aya Shinozaki-UshikuAkiko KunitaAkiko IwasakiMoe KatoSho YamazawaHiroyuki AbeTetsuo Ushiku
Published in: Histopathology (2022)
Microsatellite instability (MSI) is a major carcinogenic pathway with prognostic and predictive implications. The validity of polymerase chain reaction (PCR)-based MSI testing is well established in colorectal cancer; however, the data are limited in non-colorectal gastrointestinal cancers. The aim of this study is to clarify the detailed MSI profiles of non-colorectal gastrointestinal cancers and to investigate the differences from those of colorectal cancers. MSI testing was performed using paired tumour/normal tissues of 123 mismatch repair-deficient cancers detected by immunohistochemistry including 80 non-colorectal cancers (eight oesophagogastric junction (EGJ), 57 gastric and 15 small intestine) and 43 colorectal cancers. Fragment size analysis revealed that the mean nucleotide shifts of five markers (Promega panel) were the highest in the stomach (6.4), followed by colorectum (5.7), small intestine (5.0) and EGJ cancers (mean = 4.0; P = 0.015, versus stomach). All cases showed ≥ 1 nucleotide shift in ≥ 2 markers and were considered as MSI-high. However, when the cut-off was set to ≥ 3 nucleotide shifts in ≥ 2 markers, three EGJ (37.5%), two small intestine (13.3%) and two gastric (3.5%) cancers showed false-negative results. In addition, cases with isolated loss of MSH6 or PMS2 showed smaller nucleotide shifts than those in others. MSI testing is applicable to non-colorectal gastrointestinal cancers; however, a subset can yield false-negative results due to subtle nucleotide shift in multiple markers. Analysis of paired tumour/normal tissues and careful interpretation is necessary to avoid false-negative results and ensure appropriate treatment.
Keyphrases
  • gene expression
  • machine learning
  • young adults
  • electronic health record
  • deep learning