Scholarlink

Background <p>Breast cancer (BC) is a major problem of public health in western countries. The long-term survival improved thanks to therapeutic progresses and mass screening. Mass screening is based on mammography but displays limitations. Efforts are ongoing to develop accurate and minimally invasive tools for early BC detection. Analysis of liquid biopsies is a promising option, among which the ones based on thermal denaturation profiling provide a “thermodynamic signature” of disease through analysis of plasma protein denaturation profiles. We recently developed a major technical breakthrough of differential scanning calorimetry by switching to nanoDSF (Differential Scanning Fluorimetry), more easily transferrable in clinical routine. Here, we applied it for the first time to samples form BC patients.</p> Methods <p>We retrospectively applied nanoDSF to plasma samples from 176 patients collected in two prospective clinical trials and 61 healthy controls (HC). The profiles were analyzed using four artificial intelligence (AI) algorithms. Our primary objective was to test the potential of this approach to distinguish BC <i>versus</i> HC samples. We also assessed its ability to distinguish early <i>versus</i> advanced BC, and major molecular subtypes of disease.</p> Results <p>The four algorithms provided predictive models displaying very good performances for distinguishing patients from HC. For example, the random forest-based model displayed 96.6% accuracy in properly classifying subjects, 99.4% sensitivity, and 88.5% specificity. These performances were not dependent on the clinicopathological characteristics of BC, and compared favorably to those of mammography-based screening. For comparison, the performances of predictive models centered on the secondary objectives (early <i>versus</i> metastatic stage, hormone receptor (HR)-positive <i>versus</i> HR-negative status, and HER2-positive <i>versus</i> HER2-negative status) were good, but inferior, likely because of the stronger unbalance in the number of patients in each group and of more subtle differences in thermograms between patients’ groups than between patients and HC.</p> Conclusions <p>We reveal the potential of nanoDSF and AI applied to plasma samples to discriminate between BC patients and HC. If these results are confirmed, such approach could represent a minimally-invasive, low risk, quick and low-cost technique, which could help to improve the screening of BC.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The plasma nanoDSF denaturation profiles predict the presence of breast cancer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月5日

查看全文
  • Mathilde Guerin,
  • Rémi Eyraud,
  • Philippe Tsvetkov,
  • Pascal Finetti,
  • Alexandre Giraudo,
  • Carole Tarpin,
  • Aymen Hassin,
  • Didier Bechlian,
  • Emilie Mamessier,
  • Jihane Pakradouni,
  • Jean-Marie Boher,
  • Anthony Goncalves,
  • François Devred,
  • François Bertucci

摘要

Background

Breast cancer (BC) is a major problem of public health in western countries. The long-term survival improved thanks to therapeutic progresses and mass screening. Mass screening is based on mammography but displays limitations. Efforts are ongoing to develop accurate and minimally invasive tools for early BC detection. Analysis of liquid biopsies is a promising option, among which the ones based on thermal denaturation profiling provide a “thermodynamic signature” of disease through analysis of plasma protein denaturation profiles. We recently developed a major technical breakthrough of differential scanning calorimetry by switching to nanoDSF (Differential Scanning Fluorimetry), more easily transferrable in clinical routine. Here, we applied it for the first time to samples form BC patients.

Methods

We retrospectively applied nanoDSF to plasma samples from 176 patients collected in two prospective clinical trials and 61 healthy controls (HC). The profiles were analyzed using four artificial intelligence (AI) algorithms. Our primary objective was to test the potential of this approach to distinguish BC versus HC samples. We also assessed its ability to distinguish early versus advanced BC, and major molecular subtypes of disease.

Results

The four algorithms provided predictive models displaying very good performances for distinguishing patients from HC. For example, the random forest-based model displayed 96.6% accuracy in properly classifying subjects, 99.4% sensitivity, and 88.5% specificity. These performances were not dependent on the clinicopathological characteristics of BC, and compared favorably to those of mammography-based screening. For comparison, the performances of predictive models centered on the secondary objectives (early versus metastatic stage, hormone receptor (HR)-positive versus HR-negative status, and HER2-positive versus HER2-negative status) were good, but inferior, likely because of the stronger unbalance in the number of patients in each group and of more subtle differences in thermograms between patients’ groups than between patients and HC.

Conclusions

We reveal the potential of nanoDSF and AI applied to plasma samples to discriminate between BC patients and HC. If these results are confirmed, such approach could represent a minimally-invasive, low risk, quick and low-cost technique, which could help to improve the screening of BC.