Two articles out from a large multicenter team (led out of the Janelia campus of HHMI) demonstrate fearsome advances in the imaging technique called focused ion beam scanning electron microscopy (FIB-SEM). Until fairly recently, that was mainly used in the semiconductor industry to look over the fine details of circuit design and production, but it's been increasingly applied to biological samples. Briefly, what you do is produce a high-quality beam of gallium ions and aim it at the sample. This hits the surface and produces showers of secondary electrons and ions (some neutral species, too, for that matter). You can use those secondary electrons that you detect to do straight FIB imaging, or you can alternate between the ion beam and classic scanning electron microscopy (SEM, which hits the sample directly with a beam of electrons). That works because at higher ion currents, you can actually abrade the surface of the sample with the gallium beam, and you could in theory just work your way down, milling off layer by fine layer, and imaging the new surfaces by SEM all the way.
Getting that to work well in practice has taken some refinements. You could use it to work down and generate a three-dimensional picture of the sample structure, and that's been done across small surface areas and depths in cellular samples. But these latest papers report improvements in several directions - the FIB milling has been made much more precise, for one, and the SEM-style electron detection has also been enhanced greatly. Both these needed to happen at once - the previous backscattered electron detection was pretty slow, and needed an extended period to build up a high-res image. You could always rev up the ion beam to produce more electrons, but that also gives a lot more sputtering and noise, and the imaging couldn't keep up with the milling effect. As it is, the new system produces high resolution images across much larger sample sizes (one hundredfold greater as compared to the previous work) at five times lower radiation levels, which reduces damage to the sample and improves the data in that way as well. Fans of electron diffraction and cryo-EM will recognize the sorts of signal-to-noise and resolution improvements in electron detection that have led to those techniques advancing over the years.
In this paper, the authors report machine-learning training to deal with the (huge) mass of data that these experiments produce, automatically labeling major cell structures and producing a three-dimensional picture of the cellular environment, and in this one they announce "OpenOrganelle2", an open-access library of such data across a variety of cell types and tissues. The level of detail and the picture it provides of the cell is startling. At right is a cube cut out of the middle of a HeLa cell. The purple stuff at the top is chromatin, and then you have the dark green nuclear membrane separating that from the cytoplasm. Those yellow blobs are lysosomes, and the green is endoplasmic reticulum. The layered light blue stuff is Golgi, as you might figure, and the dark blue are endosomes. The small grey tubes are, in fact, microtubules, and the small red structures are vesicles associated with the Golgi apparatus. As the authors point out, seeing these organelles in their natural state really does give you a different picture. Based on electron microscopy in slices, for example, the Golgi structure is often drawn as a compact stack of folded membranes, but you can see that in reality it takes up more room than the usual picture, and is far more complex - those stacks are present, but they're connected by winding tubes and passageways.
The inside of a cell is crowded. It's a thick mass of tangled stuff, and these images are only the beginning of the complication. I would be interested to see if FIB-SEM can pick up on the larger biomolecular condensates, for example, temporary membraneless droplets that phase-separate from the cytoplasm. Actually, looking at these sorts of pictures, you wonder what we mean by "cytoplasm" at all - if you're picturing a volume of jelly-like stuff with occasional organelles floating around in it, that doesn't seem very accurate compared to the yarn bag we have in reality. And of course there are finer levels of detail that this technique can't reach - consider the complicated structure of the nuclear pores, the way that the ER and Golgi (and those vesicles) are in fact full of proteins in various stages of processing, and the huge numbers of enzymes and transporters that decorate those membrane surfaces. And then add in the fact that all of these structures are in constant motion - jostling microtubles being pushed by expanding and contracting endosomes, the nuclear membrane bulging as some chunk of chromatin is unwound for transcription, vesicles of proteins squeezing and shoving through the mass delivering their cargos to the inside of the cell membrane itself, and so much more. If you just run with the colors in the picture shown and imagine all of these things to be rather dense gelatin (lime, lemon, cherry, "blue raspberry" and all), that's probably about right. It's a lot to think about. . .