Woollam Mark, Angarita-Rivera Paula, Thakur Sanskar, Daneshkhah Ali, Siegel Amanda P, Hardin Dana S, Agarwal Mangilal
Integrated Nanosystems Development Institute, Indiana University Indianapolis, Indianapolis, 46202, USA.
Department of Chemistry and Chemical Biology, Indiana University Indianapolis, Indianapolis, 46202, USA.
Sci Rep. 2025 May 25;15(1):18257. doi: 10.1038/s41598-025-00284-z.
Persons with type 1 diabetes (T1D) must track/control their blood glucose (BG) levels to avoid hypoglycemic events (BG < 70 mg/dL), which in the most severe cases can lead to seizures or even death. Canines may lead the way toward innovative testing solutions, as they can be trained to identify hypoglycemia simply and noninvasively by smelling exhaled volatile organic compounds (VOCs). To identify breath-based biomarkers of hypoglycemia, samples were collected during two consecutive summers at a diabetes camp (Cohort 1 and Cohort 2), and VOCs were analyzed by gas chromatography-mass spectrometry. Conserved VOCs between the two cohorts were identified, but individual VOCs alone had low accuracies for detection. Therefore, supervised multivariate statistical analysis was undertaken to identify a biosignature in the training data set (Cohort 1) that could detect hypoglycemia with higher accuracy (sensitivity = 94.8%/specificity = 95.0%). When this model was blindly tested on Cohort 2, hypoglycemia was classified with sensitivity = 90.0%/specificity = 89.9%. Ultimately, this study makes strides toward clinical validation through verifying biomarkers of hypoglycemia in hundreds of breath samples. These results may be translated to design a sensor array that could be integrated into a portable breathalyzer to increase glycemic control in persons with T1D.
1型糖尿病（T1D）患者必须跟踪/控制其血糖（BG）水平，以避免低血糖事件（BG < 70 mg/dL），在最严重的情况下，低血糖会导致癫痫发作甚至死亡。犬类可能引领创新的检测解决方案，因为它们可以通过嗅闻呼出的挥发性有机化合物（VOC），经过训练以简单且非侵入性的方式识别低血糖。为了识别基于呼吸的低血糖生物标志物，在连续两个夏天于一个糖尿病营地（队列1和队列2）采集了样本，并通过气相色谱 - 质谱法分析VOC。确定了两个队列之间保守的VOC，但单独的个体VOC检测准确率较低。因此，进行了监督多元统计分析，以在训练数据集（队列1）中识别一种生物特征，该生物特征能够以更高的准确率检测低血糖（灵敏度 = 94.8%/特异性 = 95.0%）。当该模型在队列2上进行盲测时，对低血糖的分类灵敏度 = 90.0%/特异性 = 89.9%。最终，本研究通过在数百个呼吸样本中验证低血糖生物标志物，朝着临床验证迈出了步伐。这些结果可能转化为设计一种传感器阵列，该阵列可集成到便携式呼气酒精含量测定仪中，以改善1型糖尿病患者的血糖控制。
