Damian Toczydlowski

Damian ToczydlowskiSep 11, 2026

𝟒𝐃 𝐬𝐩𝐚𝐭𝐢𝐨𝐭𝐞𝐦𝐩𝐨𝐫𝐚𝐥 𝐥𝐚𝐧𝐝𝐬𝐜𝐚𝐩𝐞 𝐨𝐟 𝐦𝐢𝐭𝐨𝐜𝐡𝐨𝐧𝐝𝐫𝐢𝐚𝐥 𝐩𝐡𝐞𝐧𝐨𝐭𝐲𝐩𝐞𝐬 𝐚𝐜𝐫𝐨𝐬𝐬 𝐜𝐞𝐥𝐥𝐮𝐥𝐚𝐫 𝐬𝐭𝐚𝐭𝐞𝐬 𝐮𝐧𝐥𝐨𝐜𝐤𝐞𝐝 𝐭𝐡𝐫𝐨𝐮𝐠𝐡 𝐫𝐞𝐩𝐫𝐞𝐬𝐞𝐧𝐭𝐚𝐭𝐢𝐨𝐧 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 I have to say, this still feels a bit surreal. We have only just started getting used to all the new 3D methods in biomedical research.. Now we are already looking at cellular worlds in four dimensions (4D)! Maybe one day we will even reach 5D. Who can say? I remember when 3D and 4D glasses first showed up in cinemas. We wore them to see films in a new way. Actually, it looks like biology is having a similar moment. But this time, the glasses aren't for a movie. Scientists dream of using them to see our cells (and their organelles) from different angles. Even more intriguing... They could let us watch how the parts inside cells change over time. In a new paper in Cell, Agarwal and colleagues focus exclusively on mitochondria, which are far from static, to be honest. These organelles move, divide, fuse, change shape, and reorganise as the cell's context changes. The authors used 𝐥𝐚𝐭𝐭𝐢𝐜𝐞 𝐥𝐢𝐠𝐡𝐭-𝐬𝐡𝐞𝐞𝐭 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐲 𝐭𝐨𝐠𝐞𝐭𝐡𝐞𝐫 𝐰𝐢𝐭𝐡 𝐬𝐞𝐥𝐟-𝐬𝐮𝐩𝐞𝐫𝐯𝐢𝐬𝐞𝐝 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 to create 𝐌𝐢𝐭𝐨𝐒𝐩𝐚𝐜𝐞, a model trained on 4D images of mitochondria under different conditions. What I find especially unique is that the model wasn't given predefined categories of mitochondrial shapes to learn. Researchers provided the model with sequences of 3D microscopy images showing how mitochondria change shape over time in cells exposed to different drugs. During training, it learned to recognise patterns in that diversity on its own. I think that range holds much more biological information than a single snapshot ever could. The model's learned features could even tell apart different conditions and adapt to new biological situations. Most impressively, just the shape and movement of mitochondria could predict membrane potential with an 𝐑² 𝐨𝐟 𝟎.𝟗𝟏. MitoSpace could one day be used for the nucleus, cytoskeleton, endoplasmic reticulum, and other cell structures. I don't know, this might even lead us toward a true 𝟒𝐃 𝐟𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐦𝐨𝐝𝐞𝐥 𝐨𝐟 𝐭𝐡𝐞 𝐜𝐞𝐥𝐥. I'm impressed, this indeed looks quite promising. Check it out!

Damian ToczydlowskiSep 13, 2026

𝐖𝐞 𝐜𝐚𝐥𝐥 𝐞𝐮𝐜𝐡𝐫𝐨𝐦𝐚𝐭𝐢𝐧 "𝐨𝐩𝐞𝐧". 𝐈𝐬 𝐢𝐭 𝐫𝐞𝐚𝐥𝐥𝐲? For years, scientists have used a simple model to describe chromatin: ∙ 𝘏𝘦𝘵𝘦𝘳𝘰𝘤𝘩𝘳𝘰𝘮𝘢𝘵𝘪𝘯 𝘸𝘢𝘴 𝘵𝘩𝘰𝘶𝘨𝘩𝘵 𝘵𝘰 𝘣𝘦 𝘵𝘪𝘨𝘩𝘵𝘭𝘺 𝘱𝘢𝘤𝘬𝘦𝘥 𝘢𝘯𝘥 𝘪𝘯𝘢𝘤𝘵𝘪𝘷𝘦 𝘸𝘩𝘪𝘭𝘦 𝘦𝘶𝘤𝘩𝘳𝘰𝘮𝘢𝘵𝘪𝘯 𝘸𝘢𝘴 𝘤𝘰𝘯𝘴𝘪𝘥𝘦𝘳𝘦𝘥 𝘰𝘱𝘦𝘯 𝘢𝘯𝘥 𝘢𝘤𝘵𝘪𝘷𝘦. However, the nucleus does not always fit into these two categories. A recent study published in 𝘕𝘢𝘵𝘶𝘳𝘦 𝘎𝘦𝘯𝘦𝘵𝘪𝘤𝘴 used 𝐬𝐢𝐧𝐠𝐥𝐞-𝐧𝐮𝐜𝐥𝐞𝐨𝐬𝐨𝐦𝐞 𝐭𝐫𝐚𝐜𝐤𝐢𝐧𝐠 and advanced imaging techniques (like 𝟑𝐃-𝐒𝐈𝐌 and 𝐒𝐓𝐎𝐑𝐌) to study 𝐞𝐮𝐜𝐡𝐫𝐨𝐦𝐚𝐭𝐢𝐧, the generally more accessible and transcriptionally active form of chromatin, in human cells. Their findings might surprise everyone in the epigenetics field, as they indeed make the usual picture of "open euchromatin" seem far too simple. So, do we need to update the textbooks again? It sure seems like it. Rather than forming a loose, extended fiber, active euchromatin was observed as condensed, irregular domains spanning several hundred nanometres. RNA polymerase II, along with active promoters and enhancers, were mainly found at the surfaces and edges of these domains. When the researchers removed cohesin, they observed something even more intriguing. The domains did not just open up. Their overall structure stayed about the same, but their movement changed (See the 3D reconstruction in the video). Actually, nucleosomes moved more freely, the chromatin became more fluid, neighboring domains mixed more, and transcriptional insulation became weaker. This makes the idea of "open chromatin" seem less straightforward than it first appears. Euchromatic chromatin can remain physically condensed while still supporting transcription. What matters here is not just how tightly chromatin is packed, but also how nucleosomes move, how domains are organised, and where the transcription machinery can reach the DNA. So, accessibility, compaction, and mobility may not simply be different points on the same open-to-closed scale. The authors offer an explanation for this phenomenon. Regulatory sequences located deeper within condensed domains can be more difficult to access. However, changes in histone modifications or the removal of nearby nucleosomes might make these regions easier to reach. This finding is rather unexpected for those of us who have always imagined "open chromatin" as less compact, more accessible, and easier to read. There is no doubt that this study challenges that idea. If euchromatin that is being actively transcribed can still be condensed, what does it actually mean to call chromatin "open" now? ___ [The video displays a 3D-SIM volume reconstruction of a fixed HCT116 nucleus, where DNA is labeled with DAPI and euchromatin is marked with H3.3-Halo-TMR]

Damian ToczydlowskiSep 10, 2026

𝐑𝐞𝐜𝐨𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐧𝐠 𝐬𝐢𝐠𝐧𝐚𝐥𝐢𝐧𝐠 𝐡𝐢𝐬𝐭𝐨𝐫𝐢𝐞𝐬 𝐨𝐟 𝐬𝐢𝐧𝐠𝐥𝐞 𝐜𝐞𝐥𝐥𝐬 𝐯𝐢𝐚 𝐩𝐞𝐫𝐭𝐮𝐫𝐛𝐚𝐭𝐢𝐨𝐧 𝐬𝐜𝐫𝐞𝐞𝐧𝐬 𝐚𝐧𝐝 𝐭𝐫𝐚𝐧𝐬𝐟𝐞𝐫 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 Many scientists are now more concerned with understanding how cells choose their next move. We have already spent quite enough time staring at cells under the microscope and, to be fair, it has taught us a lot. By now, we know a great deal about: - what they look like, - how they behave, - which phenotypes they adopt, - how those features can change from one state to another. Over the past few years, we have learnt even more by combining single-cell omics, genome perturbations, imaging readouts and sophisticated computational models. But why are we doing all this? Hidden somewhere in that complexity is something research institutes, biotech (and pharmaceutical) companies would pay a fortune to understand: ∙ 𝘈𝘵 𝘪𝘵𝘴 𝘤𝘰𝘳𝘦, 𝘵𝘩𝘪𝘴 𝘪𝘴 𝘢𝘣𝘰𝘶𝘵 𝘩𝘰𝘸 𝘤𝘦𝘭𝘭𝘴 𝘮𝘢𝘬𝘦 𝘥𝘦𝘤𝘪𝘴𝘪𝘰𝘯𝘴. If we knew precisely how a cell chooses one fate over another, it would be a little like having the instruction manual rather than reverse-engineering “the cellular machine” while it is running. A cell receives signals, weighs them, responds, changes state and, if all goes well, the tissue carries on as if nothing remarkable had happened. Far more consequential is what happens when this logic is disturbed. Disease changes the signals, the cellular communication, and sometimes the very rules under which those decisions are made. Hence, a programme that once maintained homeostasis may suddenly become useless. Or worse, it may begin to sustain disease. If we want to redirect a pathological cell state, knowing where the cell ended up is not enough. We also need to know what brought it there. We knew that most signalling inference from single-cell data relies on ligand-receptor expression or a handful of canonical response genes. That can tell us that signalling is possible, but not necessarily whether a pathway was actually active. Hutchins and colleagues approached the problem in a rather elegant way, publishing an experimental-computational framework in Nature Methods called Intracellular Response to Infer Signaling State, or IRIS. How does it work? IRIS learns transcriptome-wide signatures from combinatorial signalling perturbation screens and transfers them to previously unseen single cells. This allows the model to infer pathway activity and reconstruct signalling histories along developmental trajectories. Scientists predicted an early role for WNT in respiratory mesenchymal specification, which they then validated experimentally and used to improve a stem-cell differentiation protocol. This is a nice example of perturbation, inference, and experiment feeding back into one another! If reconstructing the signalling history of a single cell sounds intriguing to you, check this out!