Case Note Sections:
The filter at the rear of the eye is a mere 1/10th the thickness of a human hair but plays a critical lifetime role in protecting the photoreceptive cells of the retina. It is exposed to a wide range of biochemical agents and as it is never renewed it accumulates and damage inflicted on it.
When the project was started no method existed for measuring the complex web of accumulated damage comprehensively. Individual types of damage could be analysed one by one but required destroying valuable tissue
Schematic of the Human eye
The main requirements were:
The first step was to gather a database of stressors that were commonly known to impact on the structure of collagens in the human body.
Purified compounds were sourced from commercial laboratory suppliers, academic research groups and in-house preparation.
Controlled in-situ formation of modifications to human proteins was performed in the laboratory to provide a step up in complexity
Signals for various stressors
Reference samples were validated by a range of technologies including separation methods to ensure no significant impurities were present.
In-situ modified proteins were analysed using established protocols to quantify key targets
The outcomes were:
The proof of concept demonstrated that the approaches developed were worth further investment, leading to a substantial funding award from the Medical Research Council to develop the research further.
It was clear that many challenges needed to be solved:
The human filter tissue sat above another tissue that gave a confounding signal that was orders of magnitudes stronger.
Native collagen and hemes (protein responsible for red blood cells) varied greatly, dominating degradation signal.
Background signals were large and extremely variable.
The challenges each required unique solutions that were tailored to meet the specific needs of that challenge:
The first rule of data science is to start with the best data you can, therefore I ensured that the data we collected was the most pertinent. Variation due to competing tissues was solved by creating sampling protocols to ensure isolation of the correct filter layer and exclusion of the underlying tissue.
The second key part of data science is to characterise and understand your dataset in order to know what is relevant information and what is unwanted irrelevant noise. I developed a rigorous signal processing protocol that was tailored for the specific needs, accounting for the basic scientific basis of irrelevant variations.
Levels of filter components vs age
Having proved the concept and validation numerous key developments the next stage was to develop the method further.
The ultimate aim is to measure in-vivo, taking a measurement in living humans, to be able to measure in real time what that persons collagen status is. This needs the method to be:
I then set about defining what these required
Distribution of stressors in the filter
Consultation with clinical experts led to the definition of end-user requirements, namely that the analysis had to take place in under 60 seconds, require no restraint or anaesthetic and no post-measurement care. Consultation with laser safety experts led to hard numbers on the laser power and exposure times, which were many times below the levels used in the PoC and research phases.
Systematic experiments on the sources of error and their impact on the final predicted result were performed, leading to a definition of achievable performance. These experiments made it clear significant technological advances would be required to directly measure the filter in-vivo.
An alternative strategy was formulated whereby a tissue exposed to the exterior (the white of the eye) was found to accumulate the protein modifications in proportion to those in the filter itself.
The achievable performance of the current technology and the required performance for measurement of the external surface of the eye overlapped, demonstrating that this could provide a ready route to identifying the patients collagen status and thereby risk of age related blindness.
Maps of a range of protein modifications at different ages
The project led to new insights into how the protein modifications were formed (3 distinct pathways were evident, briefly the formation of acid side chains and subsequent formation of calcified deposits, lipid oxidation induced modifications and a new class mediated by the presence of the heme protein). The body of work has accumulated over a hundred of scientific citations and continues to accrue them, indicating an ongoing relevance to this critical area of research. The results of the project have also informed a keystone review of the eye filter and its role in eye dieases indicating that the results are impacting the understanding of thought leaders in the field.
Work towards an in-vivo solution will require many more years of development and trails. A number of groups continue to explore the analysis of protein modifications in the sclera, indicating a vibrant research interest in this critical issue.
The signal processing methods developed to solve some of the challenges encountered in the project have been exploited, including in commercial products.
Alan Stitt : Dean of Innovation and Impact, Queen’s University Belfast.
“Dr Beattie was involved in the project from the start and brought enormous insight into the project. My department is focused on clinical understanding of vision impairment. In the early stages I was concerned that the challenges associated with the project would marginalise its results with respect to our mission. However, Dr Beattie not only managed to solve each technical hurdle, he also managed to make the outcome of his work relevant to our goals.“
The project established a new avenue for biochemical research, not only in the direct disease area but also a wide spectrum of diseases involving gradual decay of stable structural proteins found throughout the body. Key innovations of the project were:
The new methods and protocols developed have been exploited in a range of fields not only of academia but also commerce. This is due to many of these issues being typical of real-world data: