Joint-constraint Model Based Livecell Super-resolution Microscopy

Optical far-field microscopy enables non-invasive imaging of living samples to reveal the laws of life activities,but the limited spatial resolution restricts its potential applications in biomedicine.Since 2014,the Nobel Prize in Chemistry was awarded for the development of super-resolution fluorescence microscopy,and the approaches to improve spatial resolution have undergone tremendously developing and advancing based on the corresponding physical and chemical theories.Many modalities have been successfully implemented and become routine observation instruments in biomedical laboratories.However,in exchange for spatial resolution,the traditional super-resolution methodology usually introduces the sacrifice to temporal resolution or the increase of phototoxicity.On the other hand,live-cell imaging is a highly dynamic and sensitive process,and the observing process may also affect the state of the organism itself.This thesis intends to optimize the reconstruction model of the existing superresolution microscopy to expand the information extraction ability from the perspective of mathematical calculation.In other words,the spatial resolution is directly improved based on the optimization of the reconstruction model,under the current optical configurations without sacrificing temporal resolution or increasing phototoxicity.Furthermore,a quantitative evaluation method for reconstructed images at the corresponding super-resolution scale is established.The researches can promote the field of live-cell super-resolution microscopy and obtain better spatial resolution and image interpretability under existing hardware settings.The main contents and results are as follows:First,the traditional deconvolution technology is extremely sensitive to noise and has limited ability to expand high-frequency information.To meet these limits,based on the forward imaging model of fluorescence microscopy,a general sparsity constraint is proposed,namely "the improvement of the spatial resolution in fluorescence microscope is equivalent to the increase in the relative sparsity of the image".On the basis of the Nyquist sampling theorem,a general continuity constraint that "the intensity within the scale of the system point spread function should be continuous" is proposed.Combining two prior knowledge,a joint-constraint deconvolution reconstruction model based on the sparsity and the continuity is proposed.The sparsity constraint expands the high-frequency information of the signal,while the continuity constraint suppresses noise amplification.First,simulations were conducted using samples of different structures,and it is in principle proved that the joint-constraint deconvolution model can stably improve the spatial resolution by nearly 2 times.Then,the experimental imaging data from different biological samples which were collected by different modalities,including wide-field and super-resolution structured illumination microscopy(SIM),confocal and stimulated emission depletion microscopy(STED),was used as cross-modal verifications.These data verified that,under real experimental conditions,the method still can stably improve the spatial resolution by around 2 times.Second,the resolution improvement of linear super-resolution SIM is still limited by the diffraction limit,in which the spatial resolution can only be improved by around2 times.To solve this challenge,the linear SIM and the developed joint-constraint deconvolution model are combined(Sparse-SIM)to further increase the spatial resolution under the optical settings,without sacrificing the temporal resolution.Using specially designed DNA origami samples and commercial standard resolution slides,the increase of spatial resolution was validated and calibrated,i.e.,increased from ～120 nm to ～60 nm.The experiments under different signal-to-noise ratio(SNR)conditions were used to test the noise robustness of the system,in which the SparseSIM can still reach 60 nm resolution under 1/4 of the common imaging SNR.Third,the imaging quality and resolution of computational super-resolution techniques are highly related to the specific experimental setup and corresponding samples to be imaged,thus the field is in dire need of an in-situ super-resolution evaluation method that requires no reference structures or images.To address this need,a quantification method based on rolling Fourier ring correlation(r FRC)is proposed.It can directly characterize the reconstruction uncertainty of superresolution images,and quantitatively evaluate the reconstruction quality at the corresponding super-resolution scale.By combining with the resolution scaling error map(RSM),the final quantification results are obtained.The imaging quality of the deconvolution,SIM,and single-molecule localization microscopy results were quantitatively analyzed at the corresponding super-resolution scales,and the accuracy maps were obtained.Based on the obtained error maps,the computational superresolution technologies,namely deconvolution and single-molecule localization methods are further optimized for imaging results that are closer to the ground truth.Finally,the ～60 nm spatial resolution,564 Hz temporal resolution,and enhanced axial contrast of the developed Sparse-SIM enable biologists to resolve intricate structural intermediates and their fast dynamic processes.Several representative livecell experiments were selected to verify the effectiveness and superiority of SparseSIM for live-cell super-resolution imaging.In specific,the Sparse-SIM was used to dissect the ultrafast secretion process of insulin cell vesicles,and the new observation summarizes that there exists a two-stage secreting process.In addition,the observation also includes annular nuclear pores formed by different nucleoporins,network-like actin,circle-like caveolin,lysosome,and lipid droplet,vesicle-like fusion pores,and the relative movements of inner and outer mitochondrial membranes in living cells,facilitating the high-precision characterization of cell metabolism and secretion processes.