Noninvasive Recognition and Biomarkers of Early Allergic Asthma in Cats Using Multivariate Statistical Analysis of NMR Spectra of Exhaled Breath Condensate

Asthma is prevalent in children and cats, and needs means of noninvasive diagnosis. We sought to distinguish noninvasively the differences in 53 cats before and soon after induction of allergic asthma, using NMR spectra of exhaled breath condensate (EBC). Statistical pattern recognition was improved considerably by preprocessing the spectra with probabilistic quotient normalization and glog transformation. Classification of the 106 preprocessed spectra by principal component analysis and partial least squares with discriminant analysis (PLS-DA) appears to be impaired by variances unrelated to eosinophilic asthma. By filtering out confounding variances, orthogonal signal correction (OSC) PLS-DA greatly improved the separation of the healthy and early asthmatic states, attaining 94% specificity and 94% sensitivity in predictions. OSC enhancement of multi-level PLS-DA boosted the specificity of the prediction to 100%. OSC-PLS-DA of the normalized spectra suggest the most promising biomarkers of allergic asthma in cats to include increased acetone, metabolite(s) with overlapped NMR peaks near 5.8 ppm, and a hydroxyphenyl-containing metabolite, as well as decreased phthalate. Acetone is elevated in the EBC of 74% of the cats with early asthma. The noninvasive detection of early experimental asthma, biomarkers in EBC, and metabolic perturbation invite further investigation of the diagnostic potential in humans.

哮喘在儿童与猫群中均具有较高患病率，亟需无创诊断手段。本研究旨在通过呼出气冷凝液（exhaled breath condensate, EBC）的核磁共振波谱（nuclear magnetic resonance, NMR），无创区分53只猫在诱导过敏性哮喘发作前后的代谢差异。采用概率商归一化（probabilistic quotient normalization）与广义对数变换（glog transformation）对波谱进行预处理后，统计模式识别的性能得到显著提升。通过主成分分析（principal component analysis）与偏最小二乘判别分析（partial least squares with discriminant analysis, PLS-DA）对106条预处理后的波谱进行分类时，与嗜酸性哮喘无关的方差会显著削弱分类效果。通过滤除混杂方差，正交信号校正（orthogonal signal correction, OSC）结合PLS-DA可大幅改善健康状态与早期哮喘状态的分离效果，其预测特异性与灵敏度均可达94%。将OSC增强应用于多层PLS-DA模型后，预测特异性进一步提升至100%。对归一化波谱进行OSC-PLS-DA分析显示，猫过敏性哮喘最具潜力的生物标志物包括：丙酮水平升高、核磁共振峰位于5.8 ppm附近的重叠峰代谢物、含羟苯基的代谢物，以及邻苯二甲酸酯水平降低。在74%的早期哮喘猫的EBC样本中，丙酮水平均有所升高。本研究实现了实验性早期哮喘的无创检测，鉴定出EBC中的生物标志物并揭示了代谢扰动，可为人类哮喘诊断潜力的后续研究提供参考。