Periodic Reporting for period 1 - MitoQuant (Development of Deep-UV Quantitative Microscopy for the Study of Mitochondrial Dysfunction)
Período documentado: 2019-07-01 hasta 2021-06-30
Mitochondria are potentially the most well-known organelle; they are a fundamental part of the eukaryotic cellular machinery and their dysfunction has been linked to many diseases, ranging from diabetes to neurodegeneration. To gain information on these and other sub-cellular compartments, which contain features well below 200 nm, as well as their interactions, recent research focuses more and more on superresolution microscopy to study sub-cellular and even molecular dynamics. Superresolution microscopy, also called nanoscopy, can resolve details below 100 nm and is on its way to become an indispensable tool in biomedical research. However, almost all optical nanoscopy methods require the sample to be fluorescently labelled to break the diffraction barrier of optical imaging. This is often seen as a serious caveat as photobleaching limits the duration of time-lapse recordings, certain labels require the cell membrane to be damaged, and engineered fusion-proteins can alter the native proteins’ function. A further worry is overexpression, i.e. the cell ends up with far too many proteins of a certain type, which limits the meaningfulness of conclusions from those types of experiments. Therefore, label-free techniques with excellent resolution and specificity are of high interest to alleviate these drawbacks.
This is where the MitoQuant project comes into play. Seeing the need for label-free nanoscopy, this project aims to develop a microscopy solution capable of delivering quantitative sub-cellular resolution time-lapse imaging with adequate label-free specificity and contrast that is suitable for organelle imaging. Importantly, and in difference to electron microscopy, this microscope shall be applicable to largely “unprepared” samples and even living cells. Plenty of applications for such a microscope exist. One such example application area, which gives the project its name, is the label-free study of mitochondria and their interaction with other cellular organelles or potentially even individual proteins. The main route to achieving this feat is the exploration of deep ultra-violet (DUV) wavelengths for imaging in combination with quantitative phase microscopy and machine learning.
Therefore, the development of a novel instrument to provide adequate resolution and contrast, approaching label-based live-cell superresolution techniques like structured illumination microscopy, is the first main goal of this project. To counter low signals and yet offer highest resolution and specificity, MitoQuant’s second objective is to recover weak signal from mitochondrial by enhancing image quality through novel image processing based on machine learning.
This is where the MitoQuant project comes into play. Seeing the need for label-free nanoscopy, this project aims to develop a microscopy solution capable of delivering quantitative sub-cellular resolution time-lapse imaging with adequate label-free specificity and contrast that is suitable for organelle imaging. Importantly, and in difference to electron microscopy, this microscope shall be applicable to largely “unprepared” samples and even living cells. Plenty of applications for such a microscope exist. One such example application area, which gives the project its name, is the label-free study of mitochondria and their interaction with other cellular organelles or potentially even individual proteins. The main route to achieving this feat is the exploration of deep ultra-violet (DUV) wavelengths for imaging in combination with quantitative phase microscopy and machine learning.
Therefore, the development of a novel instrument to provide adequate resolution and contrast, approaching label-based live-cell superresolution techniques like structured illumination microscopy, is the first main goal of this project. To counter low signals and yet offer highest resolution and specificity, MitoQuant’s second objective is to recover weak signal from mitochondrial by enhancing image quality through novel image processing based on machine learning.
The MitoQuant project started with the evaluation of suitable avenues to obtain high resolution and contrast in unlabelled biological samples. Hereby, the obtained contrast should be quantitative, which means that the image value for e.g. a mitochondrion in one image has that same value in a different image, but also in a completely different sample or cell type. Such quantitative information is obtained in a label-free context by quantitative phase microscopy (QPM), which sets it apart from conventional label-free high-contrast techniques like differential interference contrast microscopy. Through MitoQuant, the use of DUV light was explored for the first time in this method. DUV wavelengths are much shorter than visible light and thus inherently offer higher resolution. Additionally, this type of illumination can excite intrinsically fluorescent amino acids and certain bio-molecules. The three most important ones in this respect are tryptophan, tyrosine, and NADH share an excitation peak, but emit at slightly shifted bands. Thus, imaging the fluorescent response of a sample using a range of emission filters enlarges the set of orthogonal quantitative measures greatly, which aids feature recovery through machine learning. Using appropriate calibration of the illumination, DUV QPM also has the potential to provide more information than visible light QPM as the excitation coefficients of the sample can be estimated in addition.
During the first year of the project period, such a microscope was design and building started. Due to heavy delays caused by the pandemic and the associated start of the lockdown in month nine of the project, the microscope could only be finished half a year behind schedule. During the lockdown, focus was therefore shifted to software-related aspects of the project ahead of time. This led to a novel approach for illumination calibration in quantitative phase microscopy using deep learning. The developed neural network operates on the Fourier spectra of images, which makes it much more transferable than other networks. We managed to outperform the gold standard method in illumination calibration by a factor of 2 using our approach. Furthermore, progress on computational superresolution microscopy was achieved using the fluctuation imaging method MUSICAL. Eventually, these approaches will be used to enhance autofluorescence image data. In the last six months of the project, lab work was possible again and the microscope could be finished. Its image quality was evaluated on a range of samples and the microscope was found to provide far greater contrast than visible light brightfield microscopy. We investigated this finding and hypothesised that it is due to the strong absorbance of DUV wavelengths in biological samples. Research in this direction is ongoing.
Living up to the spirits of the Marie Curie program, the project leader has learned the local language Norwegian, successfully acquired prestigious young investigator funding from the Norwegian Research Council and initiated the patenting process for a new type of microscopy technique. Personal development was supported by the host university through the Aurora Outstanding program that involved individual mentoring as well as courses in research communication, innovation, and leadership.
Further, a full video series on optical alignment was produced: https://www.youtube.com/watch?v=L6dGgYd4DYE
During the first year of the project period, such a microscope was design and building started. Due to heavy delays caused by the pandemic and the associated start of the lockdown in month nine of the project, the microscope could only be finished half a year behind schedule. During the lockdown, focus was therefore shifted to software-related aspects of the project ahead of time. This led to a novel approach for illumination calibration in quantitative phase microscopy using deep learning. The developed neural network operates on the Fourier spectra of images, which makes it much more transferable than other networks. We managed to outperform the gold standard method in illumination calibration by a factor of 2 using our approach. Furthermore, progress on computational superresolution microscopy was achieved using the fluctuation imaging method MUSICAL. Eventually, these approaches will be used to enhance autofluorescence image data. In the last six months of the project, lab work was possible again and the microscope could be finished. Its image quality was evaluated on a range of samples and the microscope was found to provide far greater contrast than visible light brightfield microscopy. We investigated this finding and hypothesised that it is due to the strong absorbance of DUV wavelengths in biological samples. Research in this direction is ongoing.
Living up to the spirits of the Marie Curie program, the project leader has learned the local language Norwegian, successfully acquired prestigious young investigator funding from the Norwegian Research Council and initiated the patenting process for a new type of microscopy technique. Personal development was supported by the host university through the Aurora Outstanding program that involved individual mentoring as well as courses in research communication, innovation, and leadership.
Further, a full video series on optical alignment was produced: https://www.youtube.com/watch?v=L6dGgYd4DYE
Our initial image experiments in the DUV show a leap in image quality and contrast compared to previously reported attempts in this wavelength range. Further, the ability to image complex phase information and structurally associated autofluorescence virtually aberration-free in this regime is a technical success that goes beyond the state of the art. In a first run of applications, we successfully imaged various standard cell lines like HeLa and U2OS cells. All the measured samples showed a high image quality, contrast well beyond the standard in the visible regime, diffraction-limited resolution of 250µm, and specific autofluorescence. Data analysis is ongoing. The setup and its capabilities has generated huge interest from collaborators in vascular biology, pathology, cardiovascular biology, fishery sciences, and bacteriology. We intend therefore to apply the project to medical and biological fields using funding from future national and EU project calls.
The developed deep learning method delivers a two-fold improvement in accuracy over the current state of the art. The approach and software are published and openly available. Similarly, software was developed for computational nanoscopy, with both software and datasets openly available.
In total, four papers have been published linked to the research conducted in this project.
The developed deep learning method delivers a two-fold improvement in accuracy over the current state of the art. The approach and software are published and openly available. Similarly, software was developed for computational nanoscopy, with both software and datasets openly available.
In total, four papers have been published linked to the research conducted in this project.