Department of Pathology, University Medical Center Utrecht, Utrecht University, Utrecht, the Netherlands.
Department of Endocrinology and Internal Medicine, Amsterdam University Medical Centers, Amsterdam, the Netherlands.
Clin Cancer Res. 2021 Mar 1;27(5):1341-1350. doi: 10.1158/1078-0432.CCR-20-3281. Epub 2020 Dec 22.
The primary origin of neuroendocrine tumor metastases can be difficult to determine by histopathology alone, but is critical for therapeutic decision making. DNA methylation-based profiling is now routinely used in the diagnostic workup of brain tumors. This has been enabled by the availability of cost-efficient array-based platforms. We have extended these efforts to augment histopathologic diagnosis in neuroendocrine tumors.
Methylation data was compiled for 69 small intestinal, pulmonary, and pancreatic neuroendocrine tumors. These data were used to build a ridge regression calibrated random forest classification algorithm (neuroendocrine neoplasm identifier, NEN-ID). The model was validated during 3 × 3 nested cross-validation and tested in a local and an external cohort ( = 198 cases).
NEN-ID predicted the origin of tumor samples with high accuracy (>95%). In addition, the diagnostic approach was determined to be robust across a range of possible confounding experimental parameters, such as tumor purity and array quality. A software infrastructure and online user interface were built to make the model available to the scientific community.
This DNA methylation-based prediction model can be used in the workup for patients with neuroendocrine tumors of unknown primary. To facilitate validation and clinical implementation, we provide a user-friendly, publicly available web-based version of NEN-ID.
仅通过组织病理学确定神经内分泌肿瘤转移的主要起源可能具有一定难度，但这对于治疗决策至关重要。基于 DNA 甲基化的分析已广泛应用于脑肿瘤的诊断工作。这得益于成本效益高的阵列平台的可用性。我们已经扩展了这些努力，以增强神经内分泌肿瘤的组织病理学诊断。
编译了 69 例小肠、肺和胰腺神经内分泌肿瘤的甲基化数据。这些数据用于构建岭回归校准随机森林分类算法（神经内分泌肿瘤标识符，NEN-ID）。该模型在 3×3 嵌套交叉验证中进行了验证，并在本地和外部队列中进行了测试（=198 例）。
NEN-ID 能够非常准确地（>95%）预测肿瘤样本的起源。此外，该诊断方法在一系列可能的混杂实验参数（如肿瘤纯度和阵列质量）下具有稳健性。我们构建了一个软件基础设施和在线用户界面，将该模型提供给科学界使用。
该基于 DNA 甲基化的预测模型可用于未知原发灶的神经内分泌肿瘤患者的检查。为了促进验证和临床应用，我们提供了一个用户友好的、公开可用的基于网络的 NEN-ID 版本。
