A new AI-powered blood test could help identify the biological signals of Parkinson’s years before symptoms appear — raising the possibility of earlier diagnosis and new approaches to treatment.
The question
Could a blood test reveal that Parkinson’s disease is developing years before the first tremor, movement difficulty or other symptoms appear?
Researchers at University College London (UCL) and University Medical Center Goettingen in Germany believe they have taken an important step towards answering that question, developing a machine-learning approach that identified a Parkinson’s-associated pattern in blood samples up to seven years before the onset of motor symptoms.
The research, published in Nature Communications this month, focused on eight proteins in the blood that appear to change in people with Parkinson’s disease.
The problem
Parkinson’s is a progressive neurodegenerative disorder involving the loss or impairment of nerve cells in a region of the brain called the substantia nigra. These cells produce dopamine, an important chemical involved in controlling movement.
By the time symptoms such as tremor, slowness of movement and gait problems become apparent, significant biological changes have already taken place.
That creates a difficult problem for researchers developing treatments intended to slow or stop the disease: how do you identify people who are at risk before the disease has progressed far enough to produce obvious symptoms?
To date, people with Parkinson’s are generally treated after symptoms appear, using therapies including dopamine replacement. Researchers were increasingly interested in finding biological markers that could identify the disease at an earlier stage, when there may be a greater opportunity to protect vulnerable brain cells.
“We need to diagnose patients before they have developed the symptoms. We cannot regrow our brain cells and therefore we need to protect those that we have,” said Professor Kevin Mills, from the UCL Great Ormond Street Institute of Child Health, and senior author of the research.
“At present we are shutting the stable door after the horse has bolted and we need to start experimental treatments before patients develop symptoms.”
The research
The team set out to identify blood-based biomarkers associated with Parkinson’s and then use machine learning to determine whether a combination of those biomarkers could distinguish people with the disease.
The researchers identified a panel of eight proteins whose levels were associated with Parkinson’s-related biological processes.
The machine-learning model was then trained to recognise the pattern created by those proteins.
In an initial analysis, the model was able to distinguish people with Parkinson’s from controls with 100% accuracy within the study sample. The researchers then asked a more difficult question: could the same biological pattern identify people who had not yet developed Parkinson’s but were likely to do so?
How it works
Blood sample
↓
Eight protein biomarkers measured
↓
Machine-learning model identifies the pattern
↓
Pattern compared with Parkinson’s-associated profile
↓
Potential identification of people at increased risk before symptoms
The eight proteins included markers associated with processes such as inflammation, complement activation, protein handling and Wnt signalling. The researchers suggest that these biological changes may provide clues about what is happening during the earlier stages of Parkinson’s.
Testing the prediction
To investigate whether the test could identify Parkinson’s before symptoms appeared, the researchers looked at people with isolated rapid eye movement sleep behaviour disorder (iRBD).
People with iRBD can physically act out their dreams during sleep. The condition is known to be associated with a substantially increased likelihood of later developing a synucleinopathy, including Parkinson’s disease.
The researchers analysed blood samples from people with iRBD and found that the machine-learning model identified a Parkinson’s-associated profile in 79% of them.
The participants were then followed over time.
At the time of the analysis, some had gone on to develop Parkinson’s disease or dementia with Lewy bodies. The model had correctly identified a number of these individuals in advance, with the earliest correct classification occurring 7.3 years before motor symptoms developed.
Why it matters
The significance of the research is not simply that AI can analyse blood.
The larger opportunity is the possibility of moving Parkinson’s research further towards earlier biological detection.
If researchers can reliably identify people who are in the early stages of the disease before significant symptoms develop, those people could potentially become candidates for clinical trials of treatments designed to slow or prevent progression.
Dr Michael Bartl, from University Medical Center Goettingen, said:
“By determining 8 proteins in the blood, we can identify potential Parkinson’s patients several years in advance. This means that drug therapies could potentially be given at an earlier stage, which could possibly slow down disease progression or even prevent it from occurring.”
The research also suggested that some of the biomarkers were linked to biological processes involved in Parkinson’s, potentially giving researchers clues about possible therapeutic targets.
A less invasive route to diagnosis
There is another attraction to the approach: the sample comes from blood.
Professor David Dexter, Director of Research at Parkinson’s UK, described the research as “a major step forward in the search for a definitive and patient friendly diagnostic test for Parkinson’s”.
He also highlighted the fact that biological markers measured in blood could be considerably less invasive than a lumbar puncture, which was increasingly being used in Parkinson’s clinical research at the time.
That matters because a blood test could potentially be easier to repeat and deploy at scale than more invasive forms of investigation.
What the research does — and doesn’t — show
The results are promising, but they do not mean that there was already a general-purpose blood test capable of predicting Parkinson’s in the wider population.
The study was conducted using particular research groups, including people with iRBD who were already considered to be at increased risk.
Nor did the research demonstrate that early identification would prevent Parkinson’s.
What it showed was that a specific panel of blood proteins, analysed using machine learning, could identify a Parkinson’s-associated biological profile and correctly predict subsequent disease in some participants years before motor symptoms appeared. Further validation was needed before the approach could be considered for broader clinical use.
That distinction is important.
Prediction is not prevention.
The next scientific challenge is to establish how accurately the test performs in different populations and whether identifying people at an earlier stage can ultimately lead to better outcomes.
What’s next?
The research team was continuing to test the accuracy of the approach in people considered to be at higher genetic risk of Parkinson’s, including people carrying mutations associated with the disease.
The researchers also hoped to develop a simpler blood-spot test, in which a small drop of blood could be placed on a card and sent to a laboratory for analysis.
Professor Mills said the ambition was to develop a test that could eventually be translated into large NHS laboratories, although this would require further research and funding.
The bigger question
The research points towards a broader change taking place in medicine.
For many diseases, diagnosis begins when symptoms become visible. But advances in molecular biology, proteomics and machine learning are creating new possibilities for identifying biological changes before those symptoms emerge.
Parkinson’s therefore presents a particularly interesting test case.
Can medicine move from treating disease once it becomes visible to identifying its biological signature while there is still time to intervene?
For the researchers behind this study, the answer may ultimately depend on how successfully these early biological signals can be validated, understood and translated into treatments.
The blood test is not yet the solution.
But it offers another way of looking for the disease — potentially several years before the body makes it obvious that something is wrong.
__
Sign up for our newsletter to get the latest innovation news and more sent to your inbox.
Iain is a writer, journalist and lecturer, and former editor of two international business magazines. Iain is now editor of Innovators Magazine, as well as the creative director for OnePoint5Media.
