Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public Review):
Summary:
In this manuscript, Day et al. present a high-throughput version of expansion microscopy to increase the throughput of this well-established super-resolution imaging technique. Through technical innovations in liquid handling with custom-fabricated tools and modifications to how the expandable hydrogels are polymerized, the authors show robust ~4-fold expansion of cultured cells in 96-well plates. They go on to show that HiExM can be used for applications such as drug screens by testing the effect of doxorubicin on human cardiomyocytes. Interestingly, the effects of this drug on changing DNA organization were only detectable by ExM, demonstrating the utility of HiExM for such studies.
Overall, this is a very well-written manuscript presenting an important technical advance that overcomes a major limitation of ExM - throughput. As a method, HiExM appears extremely useful, and the data generally support the conclusions.
Strengths:
Hi-ExM overcomes a major limitation of ExM by increasing the throughput and reducing the need for manual handling of gels. The authors do an excellent job of explaining each variation introduced to HiExM to make this work and thoroughly characterize the impressive expansion isotropy. The dox experiments are generally well-controlled and the comparison to an alternative stressor (H2O2) significantly strengthens the conclusions.
Weaknesses:
(1) Based on the exceedingly small volume of solution used to form the hydrogel in the well, there may be many unexpanded cells in the well and possibly underneath the expanded hydrogel at the end of this. How would this affect the image acquisition, analysis, and interpretation of HiExM data?
The hydrogel footprint covers approximately 5% of the surface within an individual well and only cells within this area are embedded in the polymerized hydrogel for subsequent processing steps. Cells that are outside of this footprint are not incorporated into the gel because these cells are digested by Proteinase K and washed away by the excess water exchange in the gel swelling step. Note that different cell types may require higher or lower concentrations of Proteinase K to adequately digest cells for expansion while maintaining fluorescence signal. Given the compatibility of HiExM with 96-well plates, this titration can be performed rapidly in a single experiment. Although cells outside of the hydrogel footprint are removed prior to imaging, we do occasionally observe Hoechst signal that appears to be underneath the gels. We believe this signal is likely from excess DNA from digested cells that was not fully washed out in the gel swelling step. This signal is both spatially and morphologically distinct from the nuclear signal of intact cells and it does not affect image acquisition, analysis, or data interpretation.
(2) It is unclear why the expansion factor is so variable between plates (e.g., Figure 2H). This should be discussed in more detail.
The variability in expansion factor across plates can likely be attributed to the small volume of gel solution (~250 nL) required for expansion within 96 well plates. Small variations in gel volume could impact gel polymerization compared to standard ExM gels. For example, gels in HiExM are more sensitive to evaporation because of the ~1000x reduced volume compared to standard expansion gel preparations, resulting in an increased air-liquid-interface. Evaporation in HiExM gels would increase monomer and cross linker concentrations, leading to variation in expansion factor across plates. We note that expansion factor is robust within well plates and that variance is slightly increased between plates. These considerations are discussed in the revised manuscript.
(3) The authors claim that CF dyes are more resistant to bleaching than other dyes. However, in Figure. S3, it appears that half of the CF dyes tested still show bleaching, and no data is shown supporting the claim that Alexa dyes bleach. It would be helpful to include data supporting the claim that Alexa dyes bleach more than CF dyes and the claim that CF dyes in general are resistant to bleaching should be modified to more accurately reflect the data shown.
We did not show data using Alexa dyes because these fluorophores are highly sensitive to photobleaching using Irgacure and thus we could not obtain images. In contrast, some CF dyes are more robust to bleaching in HiExM including CF488A, CF568, and CF633 dyes. We have recently adapted our protocol to PhotoExM chemistry which is compatible with a wider range of fluorophores as described by Günay et al. (2023) and as shown in Fig. S16.
(4) Related to the above point, it appears that Figure S11 may be missing the figure legend. This makes it hard to understand how HiExM can use other photo-inducible polymerization methods and dyes other than CF dyes.
We revised the legend for revised Fig. S11 (now Fig. S16) as follows: Example of a cell expanded in HiExM using Photo-ExM gel chemistry. Photo-ExM does not require an anoxic environment for gel deposition and polymerization, improving ease of use of HiExM. Mitochondria were stained with an Alexa 647 conjugated secondary antibody, demonstrating that HiExM is compatible with additional fluorophores when combined with Photo-ExM.
(5) The use of automated high-content imaging is impressive. However, it is unclear to me how the increased search space across the extended planar area and focal depths in expanded samples is overcome. It would be helpful to explain this automated imaging strategy in more detail.
We imaged plates on the Opera Phenix using the PreciScan Acquisition Software in Harmony. In brief, each well is imaged at 5x magnification in the Hoechst channel to capture the full well at low resolution. Hoechst is used for this step given its signal brightness, ubiquity across established staining protocols, and spectral independence from most fluorophores commonly conjugated to secondary antibodies. Using this information, the microscope detects regions of interest (nuclei) based on criteria including size, brightness, circularity, etc. Finally, the positional information for each region is stored, and the microscope automatically images those regions at 63x magnification. The working distance for the objective used in this study is 600 µm which is sufficient to capture the entirety of expanded cells in the Z direction. This strategy minimizes offtarget imaging and allows robust image acquisition even in cultures with lower seeding density. A detailed description of the automated imaging strategy is included in the methods section of the revised manuscript.
(6) The general method of imaging pre- and post-expansion is not entirely clear to me. For example, on page 5 the authors state that pre-expansion imaging was done at the center of each gel. Is pre-expansion imaging done after the initial gel polymerization? If so, this would assume that the gelation itself has no effect on cell size and shape if these gelled but not yet expanded cells are used as the reference for calculating expansion factor and isotropy.
Pre-expansion imaging is performed after staining is complete, but prior to the application of AcX, which is the first step of the HiExM protocol. Following staining and imaging, plates can be sealed with parafilm and stored at 4˚C for up to a week prior to starting the expansion protocol. We typically image 61 fields of view at the center of the well plate (where the gel will be deposited) to obtain sufficient pre-expansion images as shown in Figure 2b (left). After preexpansion imaging, we perform the HiExM protocol followed by image acquisition. We then tile all the images, as shown in Figure 2b, and compare tiled images from the same well pre- and post-expansion to manually identify the same cells. Comparisons of the pre- and postexpansion images of the same cell are used to calculate expansion factor and isotropy measurements as described. A detailed description of this process is included in the revised manuscript.
(7) In the dox experiments, are only 4 expanded nuclei analyzed? It is unclear in the Figure 3 legend what the replicates are because for the unexpanded cells, it says the number of nuclei but for expanded it only says n=4. If only 4 nuclei are analyzed, this does not play to the strengths of HiExM by having high throughput.
We performed the doxorubicin titration assay across four different well plates (n=4). For each condition, the total number of expanded nuclei measured was 118, 111, 110, 113, and 77 for DMSO, 1nM, 10nM, 100nM, and 1µM, respectively. For SEM calculations, we included the number of independent experiments to avoid underestimating error. We revised the Fig. 3 legend to include these experimental details.
(8) I am not sure if the analysis of dox-treated cells is accurate for the overall phenotype because only a single slice at the midplane is analyzed. It would be helpful to show, at least in one or two example cases, that this trend of changing edge intensity occurs across the whole 3D nucleus.
For this analysis, the result is heavily dependent on the angle at which the edge of the nucleus intersects the image plane in the orthogonal view. For this reason, we opted to only use the optimal image plane for each nucleus. We repeated our analysis on an image using multiple optical sections to demonstrate this point. These new data are included as Fig. S11 of the revised manuscript.
(9) It would be helpful to provide an actual benchmark of imaging speed or throughput to support the claims on page 8 that HiExM can be combined with autonomous imaging to capture thousands of cells a day. What is the highest throughput you have achieved so far?
The parameters that dictate imaging speed in HiExM include exposure time, z-stack height, and number of fluorophore channels. Depending on the signal intensity for a given channel, exposure times vary from 200ms to 1000ms. For z-stack height, we found that imaging 65 sections with 1µm spacing allowed for robust identification of each region of interest in the 5x pre-scan. As an example, collecting images for a full well plate (e.g., 20 images per well with 4 channels) requires approximately 24 hours of autonomous image acquisition using the Opera Phenix. Depending on cell size, this process yields imaging data for 1200 cells (1 cell per field of view) to 6000 cells (5 cells per field of view). Different autonomous imagers as well as improving staining techniques that increase signal:noise can be expected to significantly decrease the exposure time as it will reduce the number of z-stacks needed for each region.
Reviewer #2 (Public Review):
Summary:
In the present work, the authors present an engineering solution to sample preparation in 96well plates for high-throughput super-resolution microscopy via Expansion Microscopy. This is not a trivial problem, as the well cannot be filled with the gel, which would prohibit the expansion of the gel. A device was engineered that can spot a small droplet of hydrogel solution and keep it in place as it polymerizes. It occupies only a small portion of space at the center of each well, the gel can expand into all directions, and imaging and staining can proceed by liquid handling robots and an automated microscope.
Strengths:
In contrast to Reference 8, the authors' system is compatible with standard 96 well imaging plates for high-throughput automated microscopy and automated liquid handling for most parts of the protocol. They thus provide a clear path towards high-throughput ExM and highthroughput super-resolution microscopy, which is a timely and important goal.
Weaknesses:
The assay they chose to demonstrate what high-throughput ExM could be useful for, is not very convincing. But for this reviewer that is not important.
We believe the data provide an example of the utility of HiExM that would benefit experiments that require many samples (e.g., conditions, replicates, timepoints, etc.) by enabling easier sample processing and autonomous acquisition of thousands of nanoscale images in parallel. The ability to generate large data sets also enables quantitative analysis of images with appropriate statistical power. The intention of this work is to provide a proof-of-concept example of the robustness, accessibility, and experimental design flexibility of HiExM.
Reviewer #3 (Public Review):
Summary:
Day et al. introduced high-throughput expansion microscopy (HiExM), a method facilitating the simultaneous adaptation of expansion microscopy for cells cultured in a 96-well plate format. The distinctive features of this method include 1) the use of a specialized device for delivering a minimal amount (~230 nL) of gel solution to each well of a conventional 96-well plate, and 2) the application of the photochemical initiator, Irgacure 2959, to successfully form and expand the toroidal gel within each well.
Strengths:
This configuration eliminates the need for transferring gels to other dishes or wells, thereby enhancing the throughput and reproducibility of parallel expansion microscopy. This methodological uniqueness indicates the applicability of HiExM in detecting subtle cellular changes on a large scale.
Weaknesses:
To demonstrate the potential utility of HiExM in cell phenotyping, drug studies, and toxicology investigations, the authors treated hiPS-derived cardiomyocytes with a low dose of doxycycline (dox) and quantitatively assessed changes in nuclear morphology. However, this reviewer is not fully convinced of the validity of this specific application. Furthermore, some data about the effect of expansion require reconsideration.
The application we chose was intended as a methods proof-of-concept that could enable future deep biological investigations using HiExM. We believe the data provide an example of the utility of HiExM for collecting thousands of nanoscale images that would benefit experiments that require many samples (e.g., conditions, replicates, timepoints, etc.). The ability to generate large data sets also enables quantitative analysis of images with appropriate statistical power. The intention of this experiment was to provide a proof-of-concept example of the robustness, accessibility, and experimental design flexibility of HiExM.
The variability in expansion factor across plates can likely be attributed to the small volume (~250 nL) deposited by the device posts. Small variations in gel volume could impact gel polymerization compared to standard ExM gels. For example, HiExM gels are more sensitive to evaporation due to an increased air-liquid-interface because they are ~1000x smaller than standard expansion gel preparations. Evaporation in HiExM gels likely increases monomer and cross linker concentrations, leading to variation in expansion factor across plates. We note that expansion factor is robust within well plates and that the expansion factor can be more variable between plates, likely due to differences in gel volumes and evaporation. Future iterations of the platform are expected to control for these environmental conditions. These differences are discussed in the revised manuscript.
Recommendations for the authors:.
Reviewer #1 (Recommendations For The Authors):
(1) Please include a scale bar in Figure 3a.
A scale bar has been added to Figure 3a.
(2) Please show the data related to nuclear volume after dox treatment.
We have added a supplementary figure (Fig. S10) showing nuclear volume and sphericity for post-expansion nuclei as well as nuclear area and circularity for pre-expansion nuclei.
(3) I think it would be extremely helpful for the method as a whole if analysis code and files for device fabrication were made publicly available rather than upon request.
The analysis code has been included in the supplementary files as CM_Hoechst_Analysis_for publication.ipynb. Device design files are also available at the supplementary files link as hiExM_device.SLDPRT (96-well plate device) and MultiExM_24_July28_2022.SLDPRT (24-well plate device).
(4) Some details are missing from the methods, such as the concentration of AcX used for HiExM, the concentration of antibodies, etc. Related, how long does the photopolymerization take? Just the 60 seconds that the UVA light is on?
Additional protocol details are included in the methods section of the revised manuscript. The photopolymerization does only take 60 seconds.
Reviewer #2 (Recommendations For The Authors):
(1) The first three references are chosen a little strangely here. I suggest citing STED, SIM, and PALM/STORM from the original manuscripts here. Also, EM is technically not a super-resolution technique as it is within the resolution of electron beams. This reviewer would stay with light microscopy methods when discussing "super-resolution".
We removed the reference to EM and added citations to the original publications for SIM, STED, and STORM.
(2) The sentence after citation 4 is a little off in its meaning.
We have edited the sentence to improve clarity.
(3) It is highly useful and great that the authors include the observations on the effect of photopolymerization with Irgacure 2959 on dyes.
(4) In the discussion, the authors could mention new high NA silicone oil objectives that may further optimise the resolution in their scheme.
We added a sentence in the discussion to reflect this important point.
(5) The files for the manufacture of the HiExM devices must be in the supplementary data rather than available on request.
The Solidworks designs for the 96 and 24 well plate devices are included in the supplementary files as hiExM_device.SLDPRT and MultiExM_24_July28_2022.SLDPRT, respectively.
(6) It would be useful if the authors could discuss their thoughts on the high throughput processing of expansion factors in the data analysis routine.
We added details to the methods section describing how images are processed and analyzed.
Reviewer #3 (Recommendations For The Authors):
Major:
(1) In the experiments depicted in Figure 3, the authors attempted cellular phenotyping using hiPCS-derived cardiomyocytes treated with doxorubicin (dox). They addressed that the relative intensity of Hoechst at the nuclear periphery increased solely in post-expansion images, although this trend is not clearly evidenced in the provided data (e.g., DMSO control vs. 1 nM dox, Figure 3b). Moreover, this observed phenomenon lacks clear biological significance and may not be suitable as a demonstration for proof-of-concept (POC) acquisition. It is crucial to delineate the biological processes linked with the specific enhancement of DNA binding dye signals in the nuclear periphery and how to rule out the possibility of heterogeneous redistribution of nuclear components rather than enhancing resolution. For instance, if this change can be associated with a biological process such as DNA damage, quantitative detection of the accumulated proteins related to DNA repair, or the specific histone marks, may be more suitable and less susceptible to heterogeneous expansion factors. Additionally, the authors noted the absence of significant changes in nuclear volume, yet the corresponding data was not presented. Moreover, the application insufficiently demonstrated the HiExM's scalable feature employing various well plates. If only acquiring images of dozens of nuclei (Figure 3 legend, p15), a single well per condition would suffice. Therefore, it is necessary to elucidate why this application necessitates a 96-well format for demonstration purposes. The potential experimental design should also incorporate the requirement for well-to-well replication and the acquisition of features at the individual well level, rather than at the single-cell level. Also, related to Figure S10, whether outer gradient slope, but not inner gradient slope, is linked to apoptosis (Page 8, Line 2-4) remains unclear in the H2O2-treated cells.
We believe the data provide an example of the utility of HiExM that would benefit experiments that require many samples (e.g., conditions, replicates, timepoints, etc.) by enabling easier sample processing and autonomous acquisition of thousands of nanoscale images in parallel. The ability to generate large data sets also enables quantitative analysis of images with appropriate statistical power. The intention of this work is to provide a proof-of-concept example of the robustness, accessibility, and experimental design flexibility of the HiExM method. As discussed in the manuscript, dox treatment is associated with DNA damage, cellular stress, and apoptosis, and commonly observed at high dox concentrations (>200 nM) in in vitro studies using conventional microscopy. Our data suggest that cardiomyocytes exhibit sensitivity to lower concentrations of dox than previously anticipated. Although direct evidence specifically linking dox to increased DNA condensation at the nuclear periphery is limited, the known proapoptotic effects of dox strongly suggest that our observations correlate with these changes. We have now included the data analysis on nuclear morphology in revised Fig. S10. We agree that deeper biological interpretation of the observed changes in Hoechst signal upon dox treatment (or other cellular stressors such as H2O2) using HiExM and whether these changes are correlated with DNA damage or other cellular alterations remains an exciting future direction to develop a more sensitive platform for assessing drug responses.
For expanded samples, we performed the doxorubicin titration assay across four different well plates (n=4). For each condition, the total number of nuclei measured was 118, 111, 110, 113, and 77 for DMSO, 1nM, 10nM, 100nM, and 1µM, respectively. We apologize for the confusion with respect to the number of replicates and cells analyzed. For SEM calculations, we used the number of independent experiments to avoid underestimating error.
(2) In Figure 2b, do the orange arrows indicate the same cell with a unique shape in both the pre- and post-expansion images? Additionally, in Figure 3b, why do the pre- and post-expansion nuclei exhibit such different global shapes? Considering that the gel may freely rotate within the well during expansion, it raises doubts about whether one can identify cells with consistent shapes in both the pre- and post-expansion images. Furthermore, this reviewer observed a similar issue regarding reproducibility among different well plates, as shown in Figure 2h. The panel illustrates that different plates yielded distinct populations of gel sizes. The expansion factors provided in the figure legend (page 13) ranged from 3.5x to 5.1x across gels, indicating a relatively large variation in expansion size. What is the reason behind these variations, and how can they be minimized? These variations could become critical when considering large-scale screening across multiple plates.
The orange arrow is intended to indicate the same cell with a unique shape in both the pre- and post-expansion images, albeit at a different orientation given that the gel is not fixed within the well. We agree that improved methods to identify the same cells pre- and post-expansion could facilitate error measurements. We have referenced recent methods that could be combined with HiExM to automate and improve error and distortion detection to the discussion of the revised manuscript.
Fig. 2 illustrates the ability of HiExM to achieve reproducible gel formation with minimal error within gels, wells, and across plates, measurements consistent with proExM. While uniform within gels, the expansion factor is somewhat variable between gels and plates. We attribute these differences primarily to the small size of the gels, making them vulnerable to the effects of evaporation between experiments. We note this variability should be taken into consideration for studies where absolute length measurements between plates are important for biological interpretation. Future iterations of the platform that allow precise delivery of gel volumes and that minimizes environmental exposure are expected to improve the expansion factor reproducibility across plates to further enable the use of HiExM as a tool for high-throughput nanoscale imaging.
Minor:
(1) Considering the signal loss due to photobleaching and fluorophore dilution during expansion, protein imaging may occasionally lack the sensitivity required to detect subtle morphological changes in cellular machinery. This potential limitation should be addressed or discussed in the text.
A sentence reflecting this point has been added to the manuscript.
(2) On page 15, the figure legend for panel d states, "Heatmaps of nuclei in b showing..." However, it appears that the panel referred to in this sentence corresponds to panel c.
The typo has been fixed.
(3) The type of glass 96-well plate utilized in this study should be specified, as the quality of the product could impact the expansion results.
The supplier and product number of the well plate used in our study has been added to the methods section.
(4) In Figure S3, the raw pixel values of CF305 dye are exceptionally low. Is there a specific reason for the very low signals observed when using this dye?
CF® 350 (305 was a typo) does not excite well at 405 nm, which is the excitation wavelength for the channel we used.
