Democratising Live-Cell Adaptive Super-Resolution Microscopy based on SRRF

人类的感知能力很大程度上取决于视觉皮层弥补眼睛结构缺陷的能力。从进化的角度来看，眼睛的结构随着时间的推移几乎没有变化，而神经元计算在补偿“硬件”施加的光学限制和优化其在不同条件下的性能方面发挥了重要作用。显微镜也在经历类似的转变。显微镜的基本光学设计在过去的20年里几乎没有变化。然而，我们已经取得了显着的进步，绕过他们的物理限制，找到新的方法来收集和分析图像。通过计算辅助的方法，我们现在能够实现这些成像系统的分辨率的大规模增加（10倍或更多），并大大减少光学像差。十多年前，超分辨率显微镜，一个致力于提高光学显微镜分辨率的新领域，被构想出来。在这些技术中，诸如PALM、STORM和SRRF等方法使用显微镜中的数据馈送的先进时空分析来估计分子的位置，从而实现接近分子尺度本身（<50 nm）的分辨率。这些技术正在彻底改变生物和生物医学研究。它们使我们能够第一次在纳米尺度上观察细胞的动态结构，超出了任何其他方法的能力。然而，超分辨率显微镜受到其复杂性的限制。这些方法的质量和分辨率在很大程度上取决于对所用荧光团的光物理学和被成像样品的特性有相当多的了解。非伪影高分辨率成像取决于正确地利用这些特性。活细胞成像需要进一步的难度，因为还需要考虑并最小化通过现有方法施加到细胞中的有毒光负担。在这里，我们提出了iSRRF，这是超分辨率显微镜的一种范式转变，其中关于成像条件的决定不是由研究人员决定的，而是由研究样本并学习如何最佳优化成像的人工智能引擎决定的。这种计算方法将最大限度地提高图像质量和分辨率，同时限制样品照明，以最大限度地减少活细胞的光毒性。它将用基于数学定义的行动的经验决策取代人类的直觉。在其基础上，这种方法将基于最近开发的超分辨率径向波动（SRRF）方法，利用其在大多数现代光学显微镜中实现活细胞超分辨率成像的能力，即使那些最初不是为超分辨率设计的。SRRF目前正在改变显微镜领域，导致第一个超分辨率相机（iXon SRRF-Stream由Andor Technology开发）的开发。它将被集成到大多数领先成像公司的显微镜中。遵循类似的模式，iSRRF将作为一种开源，易于使用，易于实现的算法提供，与非常流行的ImageJ和Micro-Manager图像分析和采集平台兼容，使公司和个人用户能够尽可能广泛地使用。作为iSRRF的试点项目，我们将研究细胞分裂。细胞分裂通常被用作显微镜中光毒性的基准，因为光损伤会破坏细胞分裂，如果损伤足够高，则完全阻止细胞分裂。到目前为止，由于与传统成像方法相比需要高激光照明强度，因此应用超分辨率显微镜来研究细胞分裂一直是一个难题。然而，我们的初步数据显示，iSRRF将能够在几个小时内跟踪细胞分裂，同时实现分辨率提高2至10倍。这种能力超出了任何其他现有的超分辨率方法。

Human perception depends heavily on the capacity of the visual cortex to compensate for flaws in the structure of the eye. Evolutionarily, the architecture of the eye changed little over time while neuronal computation has taken on a major role in compensating for the optical limitations imposed by the `hardware' and optimising its performance under different conditions. Microscopy is following a similar shift. The basic optical design of microscopes has changed little over the last 20 years. However, we have achieved remarkable advancements in bypassing their physical limitations by finding new ways to collect and analyse images. Through computational-assisted approaches we are now able to achieve a massive increase in the resolution of these imaging systems (by 10-fold or more) and considerably reduce optical aberrations. Over a decade ago Super-Resolution Microscopy, a new field dedicated to increasing resolution in light microscopy, was conceived. In these techniques, methods such as PALM, STORM and SRRF, use advanced spatio-temporal analysis of the data-feed in microscopes to estimate the location of molecules, achieving a resolution near the molecular scale itself (<50nm). These technologies are revolutionizing biological and biomedical research. They allow us to observe for the first time the dynamic structure of cells at the nanoscale, beyond the capacity of any other method. Super-Resolution Microscopy is however limited by its complexity. The quality and resolution of these methods depends heavily on having considerable knowledge of the photophysics of the fluorophores used and of the characteristics of the sample being imaged. Non-artefactual high-resolution imaging depends on correctly exploiting these traits. Live-cell imaging entails a further degree of difficulty, as one needs to take also into account and minimize the toxic light burden imposed into cells by the existing methods. Here we propose iSRRF, a paradigm shift in Super-Resolution Microscopy where the decisions regarding imaging conditions are not determined by the researcher, but by an artificial intelligence engine that studies the sample and learns how to best optimize imaging. This computational approach will maximize the image quality and resolution while limiting the sample illumination to minimise live-cell phototoxicity. It will replace human intuition with empirical decisions based on mathematically well-defined actions. At its base, this approach will be based on the recently developed Super-Resolution Radial Fluctuations (SRRF) method, taking advantage of its capacity to enable live-cell Super-Resolution imaging in most modern light microscopes, even those not initially designed for Super-Resolution. SRRF is currently transforming the microscopy field, leading to the development of the first Super-Resolution cameras (iXon SRRF-Stream by Andor Technology). It is set to be integrated into the microscopes of most of the leading imaging companies. Following a similar model, iSRRF will be provided as an open-source, easy-to-use, easy-to-implement algorithm compatible with the extremely popular ImageJ and Micro-Manager image analysis and acquisition platforms, enabling the widest possible uptake both by companies and individual users. As a pilot project demonstrating iSRRF, we will study cell division. Cell division is often used as the benchmark for phototoxicity in microscopy as photodamage disrupts cell division, preventing it entirely if the damage is sufficiently high. It has, thus far, been a struggle to apply Super-Resolution Microscopy to study cell division due to the requirement for high laser illumination intensities when compared to conventional imaging methods. Our preliminary data shows however that iSRRF will be capable of following cell division over several hours while achieving a 2-to-10 fold increase in resolution. This capacity is beyond any other existing Super-Resolution method.