Nature: Stanford research team achieved non-invasive early lung cancer screening

According to a study published today in the journal Nature, researchers at Stanford University have developed a machine learning method that can identify and screen patients with early-stage lung cancer
.

The method is based on the detection of tumor-derived DNA in blood samples (liquid biopsy), which means that screening of people at high risk of lung cancer is early and non-invasive

In traditional lung cancer screening, CT scans are usually recommended for high-risk groups
.

This model has been shown to reduce deaths related to lung cancer

Blood testing based on liquid biopsy technology is a new cancer detection method that is currently popular, but most of the applicable objects of liquid biopsy are often patients with advanced cancer
.

After all, these patients have higher blood levels than early patients

 

In the latest paper published today in Nature, Maximilian Diehn from Stanford University and his colleagues optimized existing sequencing methods for evaluating circulating tumor DNA (ctDNA)
.

 

They improved the extraction of DNA and identified changes that are expected to serve as effective disease markers
.

The researchers used this method to show that although ctDNA is very low in patients with early-stage lung cancer, it is a powerful prognostic indicator

The researchers then used the data to improve a machine learning method that is used to predict the presence of lung cancer-derived DNA in blood samples
.

In the initial samples of 104 early-stage non-small cell lung cancer patients and 56 matched controls, this method can distinguish early-stage lung cancer patients from risk-matched controls; in another independent verification consisting of 46 cases and 48 controls, the study The personnel confirmed the above results

Liquid biopsy 

In recent years, the performance of tumor liquid biopsy has attracted special attention
.

As a branch of in vitro diagnosis, liquid biopsy can reduce the risk of detection through non-invasive sampling, and has the advantages of high efficiency, accuracy and economy

Even without treatment, cancer cells will continue to divide and die under normal circumstances
.

When cancer cells die, they release DNA fragments into the blood

At present, the detection objects of liquid biopsy include circulating tumor cells (CTCs), circulating tumor DNA (ctDNA), circulating RNA (circulating RNA) and exosomes
.

Among them, ctDNA has received more and more attention due to its broad research prospects
.
ctDNA (circulating tumor DNA) is a cell-free DNA (cfDNA) with characteristic markers, which can be qualitatively, quantified and tracked through high-throughput sequencing technology
.
 

The characteristic markers of ctDNA that have been discovered include differences in site mutation, nucleosome occupancy, and methylation modification
.
According to the difference of these indicators, the early diagnosis of tumor, the dynamic monitoring of tumor development and curative effect, drug resistance detection, recurrence risk assessment and prognosis prediction are carried out
.
 

Stanford University professor Maximilian Diehn once said that ctDNA can not only diagnose solid tumors, but also monitor treatment response, detect small residual lesions, and target drug-resistant mutations
.
It may be the preferred non-invasive tumor screening method
.
"One of the exciting events in this field is that circulating tumor DNA can be used in many different clinical situations
.
" 

Combination of molecular technology and machine learning 

In this latest study, the researchers introduced a method for analyzing circulating tumor DNA through deep sequencing (CAPP-Seq) to better achieve early cancer screening and personalized analysis
.
 

The researchers found that although the level of ctDNA in early lung cancer is very low, ctDNA already exists before most patients receive treatment, and its existence has a strong significance for prognosis
.

 

A team led by Maximilian Diehn and Ash Alizadeh conducted the study

 

The researchers also found that the free DNA (cfDNA) somatic mutations in most lung cancer patients reflect clonal hematopoietic mutations (mutations from white blood cells) and are non-recurrent
.
Compared with tumor-derived mutations, clonal hematopoietic mutations occur in longer cfDNA fragments and lack the mutation characteristics associated with smoking
.

 Combining these findings and other molecular characteristics, the researchers developed and prospectively verified a machine learning algorithm called "lung cancer likelihood in plasma, lung-clip" (lung cancer likelihood in plasma, lung-clip).
Plasma, lung clip), can well distinguish early lung cancer patients and risk-matched control group
.

 

Schematic diagram of the possibility of plasma lung cancer (lung-clip)

 

The researchers said that this non-invasive lung cancer screening method combines improved molecular technology with machine learning to detect the presence of lung cancer cell-derived cfDNA in blood samples, and a considerable part of early lung cancer can be detected through plasma
.
 

Unlike previous liquid biopsy studies that attempted to develop pan-cancer screening analysis, this time the researchers focused on non-small cell lung cancer, using the unique characteristics of lung cancer to reduce unidentified confounding factors in test results
.
Impact
.
 

In addition, unlike previous studies that did not perform validation or used control cohort cross-validation, the calcium study adopted independent validation to avoid the possibility of over-fitting the model leading to over-optimistic results
.
 

Researchers believe that one potential application of Lung-CLiP is as an initial screening for high-risk groups
.
Positive patients can be further tested and confirmed
.
This may increase the number of lung cancer screenings each year, thereby saving more lives
.
 

Thesis title: Integrating genomic features for noninvasive early lung cancer detection

Abstract of the paper:

paper:

https://doi.
org/10.
1038/s41586-020-2140-0

 

(Source: Internet, reference only)

---