Author response:
The following is the authors’ response to the original reviews.
eLife Assessment
This interesting study adapts machine learning tools to analyze movements of a chromatin locus in living cells in response to serum starvation. The machine learning approach developed is useful, the experiments are well controlled, and the data are solid. The study would be greatly strengthened by testing key predictions made using perturbation experiments. This work will be of interest to those studying chromosome biology and gene expression patterns.
We thank eLife for this nice assessment. We indeed believe that the presented machine learning approach will be useful for many types of research questions, and this was the main aim of this manuscript.
Public Reviews:
Reviewer #1 (Public review):
Summary:
Redchuk et al. explore the dynamic properties of chromatin upon serum starvation using machine learning approaches. They use CRISPR-tagging to visualize a region on chromosome 1 in human cells and show that in their system, chromosome 1, but not the previously reported chromosomes 10, 13, and X, undergo a change in radial position upon serum starvation. Live cell imaging showed a position change towards the periphery after serum starvation. They then apply a machine learning algorithm for the analysis of the imaging data, which reveals changes in nuclear area during serum starvation and longer displacements of the chromosome 1 locus near the nuclear periphery. Differential behavior of homologues is also reported.
Strengths:
(1) The study of chromatin dynamics is an interesting and important area of research.
(2) The use of machine learning approaches to analyze live cell imaging data is timely.
(3) With serum starvation, the authors use a simple, well-controllable model system.
Weaknesses:
(1) This study only provides limited new insight into chromatin dynamics.
We respectfully disagree with this conclusion. To the best of our knowledge, our study is the first to provide any insights into chromatin dynamics upon serum starvation. Previous studies are solely based on studies in fixed cells, and the dynamics have remained unexplored. Moreover, for example the notion that homologous chromosomes show differential dynamic behavior is novel and will likely have implications and relevance to many chromatin-based processes beyond the example studied here.
(2) It was not immediately evident what the use of machine learning approaches added to this study. It appears that the main conclusions could have been reached by conventional analysis.
First, we would like to point out that the other reviewer found our machine learning analysis pipeline a major strength of our manuscript. Indeed, analyzing single features and assessing their impact on the studied phenomenon could have been achieved relatively easily by conventional analysis. However, this analysis would have ignored the interactions (some of which were not intuitively obvious) between different features and thereby limited the knowledge gain from the experiment.
Unbiased analysis of the interactions between the different features would have been already very difficult and time-consuming with conventional approaches. We believe that our analysis pipeline, especially with the Shapley values, addresses the key issue of combinatorial explosion prominent to multiparametric data, such as imaging data, and helps the researcher to navigate complex datasets.
(3) There are several specific technical points:
(a) It was not clear what the CRISRP-Sirius probes actually labelled. The chromosome 1 sgRNA sequence is provided, but I could not find information as to which region(s) of the chromosome are actually labelled (size, location, etc.).
We have added a schematic as Supplementary Figure 1A to show the region of the chromosome that is labelled. In addition, the target sequence, together with the relevant references can be found in the Materials and methods (page 16). Please see also below Reviewer #1 (Recommendations for the authors) point 4a.
(b) The authors visualize a relatively small region of chromosome 1 but make conclusions regarding the entire chromosome. Additional probes on the same chromosome should be used.
Related to this point, the discussion of why the authors are unable to reproduce the prior findings of relocation of chromosomes 10, 13, and X is not satisfying. It would be worth comparing the FISH-based painting of entire chromosomes, which generated the results suggesting relocation of these chromosomes, with the point-labelling method used here.
We agree that our approach to labeling chromosome 1 is very different than the FISH-based probes utilized before. However, we also feel that we discuss this aspect, and the difference between our and previous results, which may also stem from the used cell model, in quite a detail in the first paragraph of the results (page 4). Also, we are very careful throughout the manuscript to indicate that here we study the dynamics of a specific chromosome loci, not the entire chromosome, and have further amended the text to emphasize this. In the future, it would be very interesting to study the dynamics of also other loci of chromosome 1. As indicated also below in response to reviewer 2, we have failed to identify further gRNAs that would reliably and reproducibly label further chromosome 1 loci, suggesting that we would need to change the labeling system entirely. Unfortunately, this is not in the scope of this manuscript. Please see also below Reviewer #1 (Recommendations for the authors) point 1.
(c) The study lacks controls. Since in their hands chromosomes 10, 13, and X do not change position, they should be used as a negative control in all experiments demonstrating a shift in the location of chromosome 1.
We disagree that our study lacks controls, since we use telomeres as controls throughout the manuscript. Please see also below Reviewer #1 (Recommendations for the authors) point 2,3.
(d) I did not find information about the spatial or temporal resolution of the imaging modality. This is important to assess whether the observed changes in position, relative to time, are meaningful.
To estimate the spatial resolution, we have added new data using fixed cells (Supplementary figure 1E; corresponding text in results on page 5); temporal resolution is indicated in Materials and methods (page 17). Please see also below Reviewer #1 (Recommendations for the authors) point 4d.
(e) The authors analyze surprisingly early timepoints (up to 40 minutes) of serum starvation. Would these results look different if longer serum starvation timepoints of several hours were analyzed?
We chose to analyze early time points of serum starvation based on the previous literature reporting the chromosome relocation within the first 15 minutes of starvation. Indeed, the results might look very different later during serum starvation, since we already observe differences between 0-20 min vs 20-40 min into starvation (see for example Figure 5A-D). Analyzing further time points is not in the scope of this manuscript.
(f) The authors can do a better job of explaining what the biological meaning of the various parameters (DistR, TDist, etc.) they measure is.
We have amended Table 1 to describe the measured features more clearly. Please see also below Reviewer #1 (Recommendations for the authors) point 4e.
(g) I did not understand the reasoning for the authors' conclusion of differential behavior of homologues. Please explain this better, or idealy use more direct labeling methods that identify the individual homologues.
The differential behavior of homologues is best demonstrated in Figure 6H, which shows that in serum-containing media, the peripheral homolog has equal probability of being faster or slower compared to its homolog. However, the distribution changes upon starvation, with the peripheral loci being more frequently the faster homolog. We completely agree that further studies are needed to understand this phenomenon better, but changing the labeling method is not in the scope of this manuscript.
(h) In many figures, statistical analysis of the data is missing, including, but not limited to, Figures 1B, C, G, Figures 4, 5, 6.
We have added a Supplementary table to include inferential statistics. See also below Reviewer #1 (Recommendations for the authors) point 4b.
(i) No information is provided throughout the manuscript as to how many cells were analyzed in each experiment. This should be indicated in every figure legend.
The number of analyzed loci or nucleus is indicated in every figure. See also below Reviewer #1 (Recommendations for the authors) point 4c.
Reviewer #2 (Public review):
Summary:
The study demonstrates that CRISPR-Sirius provides a powerful approach to investigating chromosome dynamics in living cells during environmental stress. By focusing on serum starvation, the authors show that this process induces global nuclear changes, including a reduction in nuclear area and increased morphological dynamism, while at the same time driving specific reorganization of chromosome 1. Chromosome 1 relocates toward the nuclear periphery and displays distinctive patterns of motion, maintaining overall motility but punctuated by occasional long-distance displacements, particularly near the nuclear envelope. Importantly, the analysis reveals that homologous copies of chromosome 1 do not behave uniformly: peripheral loci become more mobile and responsive to starvation, whereas central homologs remain comparatively stable, often associated with nucleolar subcompartments. By integrating live imaging with machine learning and explainable AI analysis, the study highlights the complexity of nuclear organization and provides valuable insights into how chromosome-specific and locus-specific responses to stress are orchestrated within the three-dimensional nuclear landscape.
Strengths:
The study uses live-cell imaging to investigate the dynamics of loci during starvation. Livecell tracking and data interpretation are carried out using machine learning and AI models, which is a major strength.
Weaknesses:
The manuscript is at times difficult to follow, partly because the methodological descriptions are highly specialized, especially for non-expert biologists. In addition, the observations are not tested for a mechanistic basis. Experiments that could provide deeper insights are missing, for example, why chromosome 1 moves, why the peripheral homologue dislocates, or why a "long jump" is observed at the periphery even though the speed of the loci does not change. It is also unclear whether a displacement of 0.5 μm is functionally meaningful.
We appreciate the comment about the readability of our manuscript, and have seriously evaluated this point. We also completely agree that it would be interesting and important to understand the mechanistic and functional basis of the observed changes in chromatin dynamics take place upon serum starvation. However, we feel that it is not in the scope of the present manuscript. See also below Reviewer #2 (Recommendations for the authors) points 3,7-11.
Recommendations for the authors:
Reviewing Editor Comments:
I would like to first offer my congratulations on a very interesting study; second, I would like to encourage you to test a few key predictions using a perturbation experiment. Two reviewers with deep expertise in this area were supportive of the work, and both noted that such an addition would greatly increase the impact and visibility of this work in the field. I welcome a revision that addresses this seminal point. Thank you for sending your work to eLife!
We thank eLife for the positive assessment. We have aimed to address all of the reviewers comments and suggestions. However, we feel that some of the suggestions are not in the scope of this particular manuscript, since they would require setting up a different chromatin labeling system.
Reviewer #1 (Recommendations for the authors):
The following experiments would strengthen the study:
(1) Please label additional regions on chromosome 1 so as not to rely on a single point to represent the behavior of the entire chromosome.
This is an excellent suggestion, but unfortunately, despite our extensive efforts, we have failed to identify further gRNAs that would reliably label chromosome loci with the CRISPR-Sirius system. Changing the labeling system is not in the scope of the presented manuscript.
(2) Please use chromosomes 10, 13, or X as a negative control since these chromosomes do not change position in the authors' hands.
(3) Please compare the behavior of the homologues to that of either random loci or control loci on 10, 13, or X to assess whether the differential behavior observed for chromosome 10 is a specific effect.
Related to points 2 and 3, we opted to use telomeres as controls in this study. Throughout the manuscript, the behavior of chromosome 1 loci is compared to telomeres, demonstrating the specific effect of serum starvation on chr 1. For example, Figure 5A and 5B show that when analyzing mean locus displacement, chr1 and telomeres show the opposite behavior.
(4) In addition:
(a) Please provide detailed information on the sequence and location of the probes used.
We have added a schematic showing the location of the probes as Supplementary Figure S1A. In addition, the sequences are indicated in Materials and methods (page 16).
(b) Please provide a statistical analysis in all graphs.
To make statistical analysis more comprehensive, we have added supplementary table 1, showing the results of inferential statistics, namely, two-sided Mann-Whitney (MW) U-test. Descriptive statistics data are shown on figures as kernel density estimation, confidence intervals and bootstrapped changes distributions. See also below Reviewer #2 (Recommendations for the authors) point 5.
(c) Please provide throughout the manuscript in each figure legend information as to how many cells were analyzed in each experiment.
The number of analyzed loci (or nucleus) is indicated in each graph.
(d) Please provide information on the spatial and temporal resolution of the imaging modality.
The imaging settings are indicated in Materials and methods, including the temporal resolution of 0.25 frames per second (page 17). To estimate spatial resolution, and especially its relationship with the observed repositioning of the chromosome loci, we performed experiments in fixed cells, using an optically identical set-up as utilized for live imaging. Unfortunately, the microscope utilized for live imaging was taken out of use by the core facility after submission of the original draft of this manuscript, but we used a microscope with essentially a similar set-up. The data from fixed cells is now presented as Supplementary figure 1E and discussed in results on page 5. This analysis indicates that the change in minimal distance to the nuclear edge, reported in our study under serum starvation in live samples (0.32 and 0.5 micron), is more than one order of magnitude above the static error.
(e) Please better explain what the various measured parameters mean in biological terms.
We have amended Table 1 to provide better explanation of the measured parameters.
(f) Please add a scale bar to Figure 6I.’
Scale bar has been added to figure 6I.
Reviewer #2 (Recommendations for the authors):
Major points:
(1) SHAP analysis identified nuclear area (MA) and its change (sA) as the most predictive features of starvation state, while motility features (MD, MaxD, TD) showed strong interactions with nuclear morphology. Discrete features, such as displacement outliers and homolog subclassification by speed/proximity, influenced classification, particularly in MLP models. Could the authors clarify why morphological and motility features act in combinatorial and context-dependent ways? A biological interpretation of this interdependence would strengthen the study.
Unfortunately, we do not have a good biological interpretation for this. The fact that some interactions are context-dependent indicates that there could be subpopulations of cells/analyzed loci. For example, we found that the predictive value of nuclear area was high in a subgroup of low motility loci (Figure 4F and Supplementary figure 4D-F). We do not believe that adding more speculation would strengthen the study.
(2) The manuscript shows that chromosome 1 moves toward the periphery within the first 20-40 minutes of serum withdrawal. However, it remains unclear whether the locus eventually "touches" the periphery and whether it subsequently stabilizes or retracts. It would be valuable to compute the time point of minimal nuclear distance and examine whether this is transient or sustained.
With the experimental set-up utilized here, we imaged the loci for only two minutes at random time point within the first 40 minutes of the starvation. Hence extracting the time point of minimal nuclear distance is not meaningful from this dataset. As we discuss in the manuscript, following the dynamics of the same locus for longer periods of this would be very interesting in the future. However, this is not in the scope of the present manuscript.
(3) The manuscript is at times difficult to follow, partly because methodological descriptions are highly detailed in the main text. Consider moving more of the methodological content into Supplementary Methods and emphasizing the main results and interpretations in the main text for clarity.
We have carefully evaluated this point. Most methodological descriptions in the manuscript relate to the machine learning models and their explanation with SHAP. As we feel that this combination is an essential part of the manuscript, and likely the aspect that can have widest impact beyond chromatin dynamics studies, we feel that the background and our reasoning related to the chosen methods are important.
(4) The distinction between the first 20 minutes and the latter 40-minute window is intriguing. Could these different time scales be paralleled with early versus delayed gene expression responses to serum starvation? A discussion of this temporal connection would add biological depth.
This is an intriguing idea, and we have added a short note on this in the discussion (page 13). However, as we do not know how the U2OS cells utilized here respond transcriptionally to serum starvation, we are hesitant to speculate too much.
(5) If the observed interpretations are robust, could this be demonstrated more explicitly through statistical principles or reproducibility tests across independent datasets?
To provide further evidence of the robustness of our findings, we have 1) added new data to estimate the spatial resolution (Supplementary figure 1E) and 2) expand the statistical analysis as supplementary table 1. Regarding the spatial resolution (see also the response to reviewer 1), our experiments on fixed cells demonstrate that the change in minimal distance to the nuclear edge, reported in our study under serum starvation in live samples (0.32 and 0.5 micron), is more than one order of magnitude above the static error. Descriptive statistics data are shown on figures as kernel density estimation, confidence intervals and bootstrapped changes distributions. To make statistical analysis more comprehensive, we added a supplementary table, showing the results of inferential statistics, namely, two-sided Mann-Whitney (MW) U-test. MW test was used as a non-parametric statistic, with null hypothesis assuming the samples are coming from the same distribution. Null hypothesis was rejected at the p-value below 0.05. In most cases (bold font in table) MW test results were in accordance with the descriptive statistics confirming the conclusions in the study. In case of exceptions (MD, TD for telomeres and TDist), the results were reported, for example, as an “appearing trend” to reflect descriptive statistics while highlighting certainty levels.
(6) Figure labeling is difficult to follow. Please include abbreviation explanations directly in the figure panels or legends for clarity.
Abbreviations have been added to figure legends. Adding them to figures themselves would have made the figures too busy.
(7) The manuscript reports higher displacement at the nuclear periphery. Can the authors explain why displacement amplitudes increase near the periphery and how this relates to nuclear architecture?
We speculate in the manuscript (results, page 11; discussion, page 14) that actually the lower displacement observed with the central locus may, at least partially, result from anchoring this locus to the nucleolus (Fig 6I). Nevertheless, alternative explanations, such as differences in transcriptional and/or chromatin states may exist (see also the response to point 11), and this is now mentioned in the discussion (page 14).
(8) How is the movement of chromosome 1 directed specifically toward the periphery, rather than being random fluctuations? This point requires clarification.
This is an important question, but unfortunately our data does not provide an answer to this, and suggesting any mechanism would be pure speculation. Nevertheless, our results agree with previous studies utilizing fixed cells that also demonstrated movement of chromosome 1 towards nuclear periphery (Mehta et al., 2010), arguing against random fluctuation.
(9) Only chromosome 1, and not the other tested chromosomes, undergoes this relocalization. Could the authors elaborate on why some chromosomes but not others display this behavior?
Previous studies (Mehta et al., 2010) utilizing chromosome paints in fixed cells actually show the relocalization of several chromosomes upon serum starvation. The fact that we observed the relocalization of only chr1 loci is likely due to the labeling method and/or the cell model utilized in this study. This is quite explicitly discussed in the first paragraph of results (page 4).
(10) The magnitude of these movements appears relatively small (0.5 micron). Can the authors discuss whether such small but reproducible displacements are likely to be biologically meaningful in terms of nuclear function or gene regulation?
At the moment, our experimental set up allows us to analyze the dynamics of only a small portion of chr1, which indeed shows an average 0.5 micron displacement towards the nuclear periphery. Based on the chromosome painting data from fixed cells, the displacement at the level of whole chromosome is significantly larger. As mentioned in the discussion (page 13), the functional implications of radial repositioning of chromosomes upon serum starvation is not known. Therefore further discussion on the relevance of the magnitude reported here would be pure speculation.
(11) Peripheral homologs of chromosome 1 became faster and more dynamic under starvation. Why might these loci be more prone to movement? Could this be linked to differences in transcriptional activity or chromatin state between central and peripheral homologs?
At the moment we favour the idea that the central homolog is constrained by its anchorage to the nucleolus (Figure 6I). However, transcriptional activity and/or chromatin state may also play a role, and this possibility is now mentioned in the discussion on page 14.
Minor points:
(1) Figure legends use inconsistent capitalization and panel labels. These should be standardized across all figures for better readability.
We apologize for these inconsistencies, and have aimed to standardize all labeling.
References
Mehta, I.S., Amira, M., Harvey, A.J., and Bridger, J.M. (2010). Rapid chromosome territory relocation by nuclear motor activity in response to serum removal in primary human fibroblasts. Genome Biol 11, R5.
