Spatial genomics enables in situ gene-expression profiling of complex tissues with high spatial resolution, allowing for the identification and spatial mapping of molecularly defined cell types in intact tissues (Bressan et al., 2023; Larsson et al., 2021; Zhuang, 2021; Close et al., 2021). Spatially resolved gene-expression profiling has been achieved through two categories of approach: (1) imaging-based approaches that allow simultaneous visualization and quantification of hundreds to thousands of genes in individual cells; (2) next-generation sequencing-based methods, which use nucleic acid barcodes to encode positional information onto transcripts, followed by sequencing to determine both the genetic identity and spatial location of the transcripts (Bressan et al., 2023; Larsson et al., 2021; Zhuang, 2021).

These approaches have been mainly applied to thin-tissue sections of 10–20 µm thickness. However, many functional and anatomical studies of biological systems require volumetric measurement of thick-tissue blocks. In theory, three-dimensional (3D) volumes can be reconstructed using serial thin sections. In practice, it is challenging to align these serial-section images due to the tissue deformation that occurs during sectioning, which in turn makes it more challenging to accurately determine the 3D morphology and organization of cells within tissues. Additionally, multimodal characterizations of the molecular, morphological, and functional properties of cells in tissues often require registration of transcriptomic images with images that capture other cellular properties through thick-tissue or live-animal imaging, such as neuronal activity imaging. Alignment of multiple thin-section transcriptomic images with such thick-tissue or live-animal volumetric images also poses significant challenges. Hence, the ability to perform spatially resolved single-cell gene-expression profiling in thick tissues will be highly beneficial for these and other purposes.

Recent studies (Nobori et al., 2023; Wang et al., 2018; Wang et al., 2021) have demonstrated the 3D gene-expression profiling of dozens of genes in intact tissues with thicknesses ranging from 100 to 300 µm. Despite these advancements, the vast majority of today’s spatial transcriptomic measurements are still derived from tissues with thicknesses of only 10–20 µm, largely due to the scarcity of methods capable of thick-tissue 3D transcriptomic measurements.

In this work, we report an extension of multiplexed error-robust fluorescence in situ hybridization (MERFISH) to thick-tissue 3D transcriptomic imaging. MERFISH is a single-cell genome-scale imaging method that allows simultaneously imaging and qualification of RNAs from thousands of genes (Chen et al., 2015; Xia et al., 2019; Fang et al., 2022). MERFISH has also been extended to allow spatially resolved 3D-genome imaging and epigenomic profiling of individual cells (Lu et al., 2022; Su et al., 2020). So far, MERFISH measurements have only been conducted on thin-tissue sections of ~10 µm in thickness. Thick-tissue imaging presents several challenges. First, the fluorescence background caused by out-of-focus signals reduces image quality. Second, spherical aberration introduced by refractive-index mismatches leads to degradation in image resolution and quality, particularly in the deeper regions of thick-tissue samples. Third, the gel-based tissue clearing protocol, often used in tissue MERFISH imaging, was originally optimized for thin-tissue imaging and may not be adequate for imaging thick samples. Here, we address the above-described challenges by using spinning disk confocal microscopy to eliminate out-of-focus signals, adopting deep learning to speed up confocal imaging, and optimizing the MERFISH protocol for thick-tissue imaging.

First, we used spinning disk confocal microscopy to achieve optical sectioning and eliminate the out-of-focus fluorescence signals, which significantly improved the signal-to-noise ratio (SNR) in the MERFISH images of thick-tissue sections (Figure 1—figure supplement 1). However, the spinning-disk confocal detection geometry also cuts a large amount of in-focus fluorescence signals, and hence a substantially longer exposure time or higher illumination light intensity is required to achieve a high SNR for imaging individual RNA molecules in the sample. This leads to a drastic reduction in imaging speed and/or a substantial photobleaching of out-of-focus fluorophores before they are imaged. We reasoned that deep learning, which has been used to improve the quality of fluorescence microscopy images in various applications (Laine et al., 2021; Ouyang et al., 2018; Weigert et al., 2018), could potentially improve the SNR of confocal MERFISH images acquired at high speed or with low illumination intensity. To test this approach, we performed MERFISH imaging to measure 242 genes (Supplementary file 1–3) in tissue sections of the mouse cerebral cortex, imaging the same fields of view (FOVs) with both a long (1 s) and a short (0.1 s) exposure time per frame to obtain high and low SNR image pairs, respectively. As expected, the low-SNR MERFISH images acquired with 0.1 s exposure time led to a substantially lower (4× lower) detection efficiency compared to the high-SNR measurement acquired at the 1 s exposure time (Figure 1a and b). To test whether deep learning can enhance the quality of the 0.1 s images, we trained a neural network based on a subset of the short- and long-exposure-time image pairs and subsequently used this model to improve the quality of the remaining images taken with the short-exposure time (see ‘Materials and methods’). This approach indeed improved the SNR of the 0.1 s images substantially (Figure 1c). As a result, the detection accuracy and efficiency of MERFISH images acquired with 0.1 s exposure time were increased and became nearly identical to that measured with 1 s exposure time (Figure 1d). This allowed us to acquire high-quality confocal MERFISH images at high speed under low illumination intensity.

Figure 1 with 2 supplements see all

Download asset Open asset

Figure 1—source data 1

This source data file contains source data for Figure 1b and d.

https://cdn.elifesciences.org/articles/90029/elife-90029-fig1-data1-v1.xlsx

Download elife-90029-fig1-data1-v1.xlsx

In thick-tissue imaging, it is important to avoid aberration caused by refractive-index mismatch. We have previously used high numerical-aperture (NA = 1.4) oil-immersion objectives for thin-tissue MERFISH imaging. While these objectives have a high detection efficiency of fluorescence signals from thin samples close to the glass substrate, they present significant challenges when imaging thick samples. The refractive-index mismatch between the immersion oil and tissue samples leads to substantial aberration, which reduces the sensitivity and accuracy of RNA detection in the deeper regions of thick samples. To overcome this problem, we switched to water-immersion objectives, which have a better refractive-index match with the samples. Despite having a lower NA (NA = 1.15 or 1.2), these water objectives allowed for the detection of individual RNA molecules without substantial decay of detection sensitivity and efficiency across the entire depth of 100-µm- and 200-µm-thick-tissue sections (Figure 1—figure supplement 2).

Finally, we often use gel-based tissue clearing for MERFISH imaging of tissue samples. Our gel-based MERFISH protocol developed using thin-tissue samples also need to be optimized for imaging thick-tissue samples. In MERFISH, we label cellular RNAs with a library of encoding oligonucleotide probes that contains barcode-determining readout sequences and then detect the barcodes bit-by-bit with dye-labeled readout probes complementary to the readout sequences (Chen et al., 2015). The encoding and readout probe labeling conditions previously developed for thin-tissue imaging may not be optimal for thick-tissue imaging. Moreover, additional challenges may arise in thick-tissue imaging. To optimize the MERFISH protocol for thick-tissue imaging, we imaged the 242 genes in 100-µm-thick mouse brain sections with 1 µm step size for each z-plane. Upon optimization of probe labeling conditions (Figure 2—figure supplement 1a–d), we observed bright and consistent signals from individual RNA molecules across the entire depth of 100-µm-thick-tissue samples in each bit (Figure 2—figure supplement 1e). However, unexpectedly, despite consistent detection of RNA molecules in individual bits, the copy number of RNA molecules identified for individual genes decreased substantially with the tissue depth and exhibited a poor correlation with bulk RNA-seq data (Figure 2—figure supplement 2a–c). We reasoned that this may be due to the displacement of RNA molecules between imaging rounds in the thick-tissue sample that made these molecules difficult to decode and identify from their multibit images. To test this, we imaged fiducial beads embedded in the polyacrylamide gel, which is often used for sample clearing in MERFISH imaging of tissue samples (Moffitt et al., 2016a). Indeed, we observed a significant displacement in the positions of beads in all three dimensions (x, y, and z) between imaging rounds, especially in the deeper part of the thick-tissue samples (Figure 2—figure supplement 2d).

We identified three factors contributing to the displacement of RNA molecule positions between imaging rounds: (1) the piezo-actuator used for z-scanning of the objective did not consistently place the focal plane at defined z-positions of the sample during each imaging round; (2) the MERFISH protocol includes multiple buffer-exchange steps, and the expansion or shrinkage of the polyacrylamide gel upon buffer changes contributed to the displacement of RNA molecules across imaging rounds; and (3) two-color imaging was used to measure two bits in each hybridization round, and the axial chromatic aberration between the two colors resulted in misalignment of RNA molecules between bits measured in different color channels.

To solve the first problem, we conducted tests using various piezo actuators and chose the Queensgate OP400 piezo, which demonstrated the highest precision and reproducibility. This specific piezo provided negligible z-position error for sample thicknesses under 200 µm. To minimize the gel expansion effect, we used a low concentration of gel initiator and low polymerization temperature (4°C) to generate polyacrylamide gel with a lower degree of buffer-dependent expansion (Boyde, 1976) and chose MERFISH buffers that further minimize the gel-expansion effect (Figure 2—figure supplement 3a and b). In addition, to ensure that the gel recovered to the same size before imaging, the gel was allowed to relax for an extra 10 min following buffer exchange to the imaging buffer in order to completely remove any residual gel expansion induced by other buffers (Figure 2—figure supplement 3c). To eliminate the effect of chromatic aberration, we calibrated the axial chromatic aberration by imaging fiducial beads in the two-color channels, which allowed precise alignment of images between these channels.

To validate our optimized thick-tissue MERFISH protocol, we first imaged the expression of the 242 genes in 100-µm-thick sections of the mouse cortex. In addition to MERFISH imaging of RNAs, we also imaged the nuclear DAPI stain and total polyadenylated mRNA signals for 3D cell segmentation (Figure 2a). Individual RNA molecules were clearly identified in the thick-tissue MERFISH images (Figure 2b and c, Figure 2—figure supplement 4a). The average RNA copy numbers for individual genes were highly correlated with the RNA abundance measured by bulk RNA sequencing (Figure 2d). The detected RNA number and the correlation with bulk RNA-seq did not show any decay across the entire tissue depth (Figure 2e and f). We then compared the RNA copy number of individual genes detected per unit area per z-plane with the results from 10-µm-thin-tissue MERFISH measurements performed previously using an epifluorescence setup (Zhang et al., 2021). The detection efficiency of confocal thick-tissue measurements per z-plane was ~20% higher than the epi thin-tissue measurements (Figure 2g), which may be due to the background reduction by confocal optical sectioning.

Figure 2 with 4 supplements see all

Download asset Open asset

Figure 2—source data 1

This source data file contains source data for Figure 2d-h.

https://cdn.elifesciences.org/articles/90029/elife-90029-fig2-data1-v1.xlsx

Download elife-90029-fig2-data1-v1.xlsx

We then performed 3D cell segmentation using DAPI and total polyadenylated mRNA signals (Figure 2—figure supplement 4b), which allowed us to determine the expression profiles of the 242 genes in each individual cell. When we compared the RNA copy number per cell detected in the 100-µm-thick samples with the results obtained from individual 10-µm-thick sections of the same sample (obtained by dividing the 100 µm z-range into 10 equal-thickness sections), we found that the former was twofold higher than the latter (Figure 2h). This is presumably because most of the cell bodies were completely captured within the 100-µm-thick tissues, whereas many cells were only partially captured in the 10-µm-thick-tissue sections. Although normalization of the RNA copy number by cell volume could partially mitigate this problem in thin-tissue imaging, detecting too few RNA molecules per cell could nonetheless introduce significant noise and imposing a cell volume threshold to reduce noise would in turn reduce the number of cells measured. Moreover, nonuniform intracellular distribution of RNAs could further reduce transcriptome profiling accuracy when only part of a cell is imaged. Thus, thick-tissue MERFISH imaging should allow more accurate gene-expression profiling of individual cells.

Next, we used single-cell expression profiles derived from our 3D thick-tissue MERFISH measurement to identify transcriptionally distinct cell populations in the mouse cortex. We observed all previously known subclasses of excitatory neurons (layer 2/3 [L2/3] intratelencephalic-projection neurons [L2/3 IT], L4/5 IT, L5 IT, L6 IT, L6 Car3 neurons, L5 extratelencephalic projection neurons [L5 ET], L5/6 near projection neurons [L5/6 NP], L6 cortical-thalamic projection neurons [L6 CT], and L6b neurons), inhibitory neurons (marked by Lamp5, Sst, Vip, Pvalb, and Sncg, respectively), and non-neuronal cells (oligodendrocytes, oligodendrocyte progenitor cells, astrocytes, microglia, endothelial cells, pericytes, vascular leptomeningeal cells, smooth muscle cells) (Figure 3a), as well as transcriptionally distinct cell clusters within these subclasses (Figure 3—figure supplement 1). The 3D MERFISH images further showed the expected layered organization of transcriptionally distinct neuronal cell populations in the cortex (Figure 3b).

Figure 3 with 4 supplements see all

Download asset Open asset

Figure 3—source data 1

This source data file contains source data for Figure 3e-h.

https://cdn.elifesciences.org/articles/90029/elife-90029-fig3-data1-v1.xlsx

Download elife-90029-fig3-data1-v1.xlsx

To test if our thick-tissue 3D MERFISH approach can image beyond 100 µm thickness, we next measured 156 genes (Supplementary files 4 and 5) in a 200-µm-thick section of the mouse anterior hypothalamus. The RNA copy numbers of individual genes measured per cell per z-plane at different tissue depths showed excellent correlation with each other with only a slight reduction of the RNA copy number across the entire tissue depth (Figure 3—figure supplement 2). From the MERFISH-derived single-cell expression profiles, we identified 21 excitatory neuronal clusters, 26 inhibitory neuronal clusters, and 7 non-neuronal cell subclasses in this region (Figure 3c and d, Figure 3—figure supplement 3a). Most of the transcriptionally distinct neuronal clusters showed distinct and localized spatial distributions, some of which were predominantly located in a single hypothalamic nucleus such as inhibitory cluster I1 in the bed nucleus of the stria terminalis, inhibitory cluster I5 in the suprachiasmatic nucleus, and excitatory cluster E20 in the nucleus of reuniens thalamic (Figure 3d, Figure 3—figure supplement 3b).

It has been shown previously that the same type of interneurons (Lamp5, Vip, Sst, or Pvalb-positive inhibitory neurons) tend to form juxtaposed pairs in the mouse visual cortex (Wang et al., 2018) and human middle and superior temporal gyrus (Fang et al., 2022). It has been hypothesized that such tightly juxtaposed structures might be mediated by electric gap junctions, which is important for synchronized firing patterns (Amitai et al., 2002; Gibson et al., 1999; Wang et al., 2018). To test whether we can observe such tightly juxtaposed interneuron pairs in the thick-tissue 3D MERFISH images of the mouse cortex obtained in this work, we performed the following two types of analyses. First, we determined the nearest-neighbor neurons for each interneuron and classified it as either having another interneuron of the same type as the nearest neighbor or having a different type of neuron (excitatory or inhibitory) as the nearest neighbor. We found that the distances between the former category of nearest-neighbor pairs were significantly smaller than the distances between the latter category of nearest-neighbor pairs, even though the density of interneurons in the cortex is substantially lower than that of the excitatory neurons (Figure 3e). In the second analysis, we determined for each interneuron its nearest-neighbor interneurons, regardless of the subtype identity, and determined the distribution of the nearest-neighbor distances among all inhibitory neurons. We observed a bimodal distribution: while the major peak was centered at about 30 μm, a significant fraction of nearest-neighbor distances fell in a peak that was <10 μm (marked by arrow in Figure 3f). Visualization of these latter interneurons in the MERFISH images suggests that they were not only juxtaposed but also formed junctions so tight that led to morphological deformations of the cell bodies (Figure 3—figure supplement 4a). As a result, the median distance between those interneurons was less than 10 µm even though the average diameter of a single interneuron was approximately 15 µm (Figure 3—figure supplement 4b). These results are consistent with previous observation that interneurons tend to form tightly juxtaposed pairs in the mouse and human cortex (Fang et al., 2022; Wang et al., 2018).

We next performed the same two types of nearest-neighbor analysis for interneurons in the hypothalamus. In the first analysis, we found that the distances between the nearest-neighboring pairs formed by the same type of interneurons were not significantly smaller than those between the nearest-neighboring pairs formed by an interneuron and a neuron of a different type (Figure 3g). Unlike in the cortex, where the interneurons formed a scattered distribution, the interneuron subtypes formed tight spatial clusters in the hypothalamus (Figure 3—figure supplement 3b). Those interneurons being classified as having a different neuronal cell types as the nearest neighbor tended to be at the edge of their own clusters, hence being close to other cell types. Therefore, these nearest-neighbor distances were not necessarily greater than the distance between interneuron pairs of the same types. In the second analysis, we found that the distribution of the nearest-neighbor distances among all interneurons formed a single-modal distribution centered at a small distance (Figure 3h), consistent with the observation of tight spatial clusters of interneuron subtypes. While we were also able to find tightly juxtaposed interneuron pairs with deformed cell bodies in the hypothalamus (Figure 3—figure supplement 4c), such examples were much rarer than those in the cortex. Hence, it appears that interneurons potentially have a lower tendency to form such juxtaposed interneuron pairs with tight junctions in the hypothalamus than in the cortex, but it is also worth noting that the spatially clustered distributions of interneuron subtypes could make it more difficult to detect such tightly juxtaposed interneuron pairs in the mouse hypothalamus quantitatively.

In summary, we developed a method that allowed high-quality 3D MERFISH imaging in thick-tissue samples. One of the major challenges of thick-tissue RNA imaging is the low signal-to-background ratio of individual RNA molecules. To address this challenge, signal amplification has been previously used for thick-tissue imaging. For example, in STARmap (Wang et al., 2018) and PHYTOMap (Nobori et al., 2023), enzymatic process of rolling circle amplification (Krzywkowski and Nilsson, 2017; Banér et al., 1998) is used to generate DNA amplicons, thereby enhancing the signals. However, the requirement of enzymes in this process may potentially restrict or slow reagent penetration in thick tissues. Alternatively, EASI-FISH (Wang et al., 2021) employs hybridization chain reaction (HCR) (Dirks and Pierce, 2004) for signal amplification. Since HCR oligos are short (50–100 nt), they can potentially penetrate thick tissue more rapidly. This latter approach was originally used to measure the expression of dozens of genes sequentially and the number of genes that can be measured is limited by the number of available HCR amplifiers. During the revision of our manuscript, a bioRxiv preprint (Gandin et al., 2024) reported cycleHCR – a technology that leverages multiplexed barcoding and split HCR, which has enabled the measurement of hundreds of genes within a 300-μm-thick mouse embryo. Additionally, another bioRxiv preprint (Sui et al., 2024) reported approaches that enable 3D in situ quantification of thousands of genes and their corresponding translation activities in 200-μm-thick-tissue blocks using STARmap and RiboMap. In this work, we overcome the challenge of low signal-to-background ratio in thick-tissue imaging by combining optical sectioning with deep learning, without involving signal amplification. The oligonucleotide probes used in our thick-tissue 3D MERFISH approach are only 20-base long, which allows for efficient penetration into thick tissues. We demonstrated 3D MERFISH of tissue blocks up to 200 µm in this work, and the consistently high performance along the entire tissue depth suggests that our approach can be applied to image thicker tissue samples. Together, these advances expand the toolkits for transcriptome imaging of thick tissues and will benefit many important applications. For instance, as we have shown here, imaging thick tissues allows us to capture the full cell-body volume of nearly all imaged cells, and hence increases the gene-expression profiling accuracy. Second, confocal imaging removes out-of-focus fluorescence background, which can potentially allow us to image more genes or the same number of genes with fewer imaging rounds. Third, we also anticipate confocal imaging to facilitate cell segmentation, especially in regions with high cell density. Fourth, because the shape of cells can be better determined by volumetric imaging, thick-tissue imaging should facilitate integrated measurements of gene-expression profiles and morphology of individual cells to investigate the relationship between these two properties. Fifth, thick-tissue single-cell transcriptome imaging will also facilitate more efficient detection of interactions between molecularly defined cell types. Finally, we also envision this approach to facilitate combined measurements of neuronal activity, for example, by calcium and voltage indicator imaging or electrophysiology measurements, with transcriptomic profiling of individual cells to reveal the functional roles of molecularly defined cell types. Because our thick-tissue 3D MERFISH method uses commercially available spinning disk confocal microscopes and we have made our image analysis code publicly available, we anticipate that this method can be readily adopted by many laboratories.

The MERFISH image acquisition software is available at Zenodo (Babcock et al., 2019, doi:10.5281/ZENODO.3264857). The 3D MERFISH decoding analysis software is available at Zenodo (Fang et al., 2024, doi:10.5281/ZENODO.13356943). The data and analysis notebooks are available at Dryad https://doi.org/10.5061/dryad.w0vt4b922.

1. 1. Fang R
   2. Halpern A
   3. Rahman M
   4. Huang Z
   5. Lei Z
   6. Hell S
   7. Dulac C
   8. Zhuang X

   (2024) Dryad Digital Repository

   Data from: Three-dimensional single-cell transcriptome imaging of thick tissues.

   https://doi.org/10.5061/dryad.w0vt4b922

1. 1. Amitai Y
   2. Gibson JR
   3. Beierlein M
   4. Patrick SL
   5. Ho AM
   6. Connors BW
   7. Golomb D

   (2002) The spatial dimensions of electrically coupled networks of interneurons in the neocortex

   The Journal of Neuroscience 22:4142–4152.

   https://doi.org/10.1523/JNEUROSCI.22-10-04142.2002

   • PubMed
   • Google Scholar
2. Software

   1. Babcock H
   2. Moffitt J
   3. Boettiger A
   4. Emanuel G
   5. Sepulveda L
   6. Dempsey GT
   7. Kayikcioglu T

   (2019) ZhuangLab storm-control, version v2019.06.28

   Zenodo.

   https://doi.org/10.5281/zenodo.3264856
3. 1. Banér J
   2. Nilsson M
   3. Mendel-Hartvig M
   4. Landegren U

   (1998) Signal amplification of padlock probes by rolling circle replication

   Nucleic Acids Research 26:5073–5078.

   https://doi.org/10.1093/nar/26.22.5073

   • PubMed
   • Google Scholar
4. 1. Blondel VD
   2. Guillaume JL
   3. Lambiotte R
   4. Lefebvre E

   (2008) Fast unfolding of communities in large networks

   Journal of Statistical Mechanics 2008:e10008.

   https://doi.org/10.1088/1742-5468/2008/10/P10008

   • Google Scholar
5. 1. Boyde TRC

   (1976) Swelling and contraction of polyacrylamide gel slabs in aqueous solutions

   Journal of Chromatography A 124:219–230.

   https://doi.org/10.1016/S0021-9673(00)89737-X

   • Google Scholar
6. 1. Bressan D
   2. Battistoni G
   3. Hannon GJ

   (2023) The dawn of spatial omics

   Science 381:eabq4964.

   https://doi.org/10.1126/science.abq4964

   • PubMed
   • Google Scholar
7. 1. Chen KH
   2. Boettiger AN
   3. Moffitt JR
   4. Wang S
   5. Zhuang X

   (2015) RNA imaging:spatially resolved, highly multiplexed RNA profiling in single cells

   Science 348:aaa6090.

   https://doi.org/10.1126/science.aaa6090

   • PubMed
   • Google Scholar
8. 1. Close JL
   2. Long BR
   3. Zeng H

   (2021) Spatially resolved transcriptomics in neuroscience

   Nature Methods 18:23–25.

   https://doi.org/10.1038/s41592-020-01040-z

   • PubMed
   • Google Scholar
9. 1. Dirks RM
   2. Pierce NA

   (2004) Triggered amplification by hybridization chain reaction

   PNAS 101:15275–15278.

   https://doi.org/10.1073/pnas.0407024101

   • PubMed
   • Google Scholar
10. 1. Fang R
    2. Xia C
    3. Close JL
    4. Zhang M
    5. He J
    6. Huang Z
    7. Halpern AR
    8. Long B
    9. Miller JA
    10. Lein ES
    11. Zhuang X

    (2022) Conservation and divergence of cortical cell organization in human and mouse revealed by MERFISH

    Science 377:56–62.

    https://doi.org/10.1126/science.abm1741

    • PubMed
    • Google Scholar
11. Software

    1. Fang R
    2. Emanuel G
    3. Babcock H

    (2024) rx3fang/MERlin, version v3.0.0-elife

    Zenodo.

    https://doi.org/10.5281/zenodo.13356943
12. Preprint

    1. Gandin V
    2. Kim J
    3. Yang LZ
    4. Lian Y
    5. Kawase T
    6. Hu A
    7. Rokicki K
    8. Fleishman G
    9. Tillberg P
    10. Castrejon AA
    11. Stringer C
    12. Preibisch S
    13. Liu ZJ

    (2024) Deep-tissue spatial omics: imaging whole-embryo transcriptomics and subcellular structures at high spatial resolution

    bioRxiv.

    https://doi.org/10.1101/2024.05.17.594641

    • Google Scholar
13. 1. Gibson JR
    2. Beierlein M
    3. Connors BW

    (1999) Two networks of electrically coupled inhibitory neurons in neocortex

    Nature 402:75–79.

    https://doi.org/10.1038/47035

    • PubMed
    • Google Scholar
14. 1. Hörl D
    2. Rojas Rusak F
    3. Preusser F
    4. Tillberg P
    5. Randel N
    6. Chhetri RK
    7. Cardona A
    8. Keller PJ
    9. Harz H
    10. Leonhardt H
    11. Treier M
    12. Preibisch S

    (2019) BigStitcher: reconstructing high-resolution image datasets of cleared and expanded samples

    Nature Methods 16:870–874.

    https://doi.org/10.1038/s41592-019-0501-0

    • PubMed
    • Google Scholar
15. 1. Krzywkowski T
    2. Nilsson M

    (2017) Fidelity of RNA templated end-joining by chlorella virus DNA ligase and a novel iLock assay with improved direct RNA detection accuracy

    Nucleic Acids Research 45:e161.

    https://doi.org/10.1093/nar/gkx708

    • PubMed
    • Google Scholar
16. 1. Laine RF
    2. Jacquemet G
    3. Krull A

    (2021) Imaging in focus: an introduction to denoising bioimages in the era of deep learning

    The International Journal of Biochemistry & Cell Biology 140:106077.

    https://doi.org/10.1016/j.biocel.2021.106077

    • PubMed
    • Google Scholar
17. 1. Larsson L
    2. Frisén J
    3. Lundeberg J

    (2021) Spatially resolved transcriptomics adds a new dimension to genomics

    Nature Methods 18:15–18.

    https://doi.org/10.1038/s41592-020-01038-7

    • PubMed
    • Google Scholar
18. 1. Lu T
    2. Ang CE
    3. Zhuang X

    (2022) Spatially resolved epigenomic profiling of single cells in complex tissues

    Cell 185:4448–4464.

    https://doi.org/10.1016/j.cell.2022.09.035

    • PubMed
    • Google Scholar
19. 1. McGinnis CS
    2. Murrow LM
    3. Gartner ZJ

    (2019) DoubletFinder: doublet detection in single-cell rna sequencing data using artificial nearest neighbors

    Cell Systems 8:329–337.

    https://doi.org/10.1016/j.cels.2019.03.003

    • PubMed
    • Google Scholar
20. 1. Moffitt JR
    2. Hao J
    3. Bambah-Mukku D
    4. Lu T
    5. Dulac C
    6. Zhuang X

    (2016a) High-performance multiplexed fluorescence in situ hybridization in culture and tissue with matrix imprinting and clearing

    PNAS 113:14456–14461.

    https://doi.org/10.1073/pnas.1617699113

    • PubMed
    • Google Scholar
21. 1. Moffitt JR
    2. Hao J
    3. Wang G
    4. Chen KH
    5. Babcock HP
    6. Zhuang X

    (2016b) High-throughput single-cell gene-expression profiling with multiplexed error-robust fluorescence in situ hybridization

    PNAS 113:11046–11051.

    https://doi.org/10.1073/pnas.1612826113

    • PubMed
    • Google Scholar
22. 1. Moffitt JR
    2. Bambah-Mukku D
    3. Eichhorn SW
    4. Vaughn E
    5. Shekhar K
    6. Perez JD
    7. Rubinstein ND
    8. Hao J
    9. Regev A
    10. Dulac C
    11. Zhuang X

    (2018) Molecular, spatial, and functional single-cell profiling of the hypothalamic preoptic region

    Science 362:eaau5324.

    https://doi.org/10.1126/science.aau5324

    • Google Scholar
23. 1. Nobori T
    2. Oliva M
    3. Lister R
    4. Ecker JR

    (2023) Multiplexed single-cell 3D spatial gene expression analysis in plant tissue using PHYTOMap

    Nature Plants 9:1026–1033.

    https://doi.org/10.1038/s41477-023-01439-4

    • PubMed
    • Google Scholar
24. 1. Ouyang W
    2. Aristov A
    3. Lelek M
    4. Hao X
    5. Zimmer C

    (2018) Deep learning massively accelerates super-resolution localization microscopy

    Nature Biotechnology 36:460–468.

    https://doi.org/10.1038/nbt.4106

    • PubMed
    • Google Scholar
25. 1. Pachitariu M
    2. Stringer C

    (2022) Cellpose 2.0: how to train your own model

    Nature Methods 19:1634–1641.

    https://doi.org/10.1038/s41592-022-01663-4

    • PubMed
    • Google Scholar
26. 1. Stuart T
    2. Butler A
    3. Hoffman P
    4. Hafemeister C
    5. Papalexi E
    6. Mauck WM
    7. Hao Y
    8. Stoeckius M
    9. Smibert P
    10. Satija R

    (2019) Comprehensive integration of single-cell data

    Cell 177:1888–1902.

    https://doi.org/10.1016/j.cell.2019.05.031

    • PubMed
    • Google Scholar
27. 1. Su JH
    2. Zheng P
    3. Kinrot SS
    4. Bintu B
    5. Zhuang X

    (2020) Genome-scale imaging of the 3D organization and transcriptional activity of chromatin

    Cell 182:1641–1659.

    https://doi.org/10.1016/j.cell.2020.07.032

    • PubMed
    • Google Scholar
28. Preprint

    1. Sui X
    2. Lo JA
    3. Luo S
    4. He Y
    5. Tang Z
    6. Lin Z
    7. Zhou Y
    8. Wang WX
    9. Liu J
    10. Wang X

    (2024) Scalable spatial single-cell transcriptomics and translatomics in 3D thick tissue blocks

    bioRxiv.

    https://doi.org/10.1101/2024.08.05.606553

    • Google Scholar
29. 1. Wang X
    2. Allen WE
    3. Wright MA
    4. Sylwestrak EL
    5. Samusik N
    6. Vesuna S
    7. Evans K
    8. Liu C
    9. Ramakrishnan C
    10. Liu J
    11. Nolan GP
    12. Bava FA
    13. Deisseroth K

    (2018) Three-dimensional intact-tissue sequencing of single-cell transcriptional states

    Science 361:eaat5691.

    https://doi.org/10.1126/science.aat5691

    • PubMed
    • Google Scholar
30. 1. Wang Y
    2. Eddison M
    3. Fleishman G
    4. Weigert M
    5. Xu S
    6. Wang T
    7. Rokicki K
    8. Goina C
    9. Henry FE
    10. Lemire AL
    11. Schmidt U
    12. Yang H
    13. Svoboda K
    14. Myers EW
    15. Saalfeld S
    16. Korff W
    17. Sternson SM
    18. Tillberg PW

    (2021) EASI-FISH for thick tissue defines lateral hypothalamus spatio-molecular organization

    Cell 184:6361–6377.

    https://doi.org/10.1016/j.cell.2021.11.024

    • PubMed
    • Google Scholar
31. 1. Weigert M
    2. Schmidt U
    3. Boothe T
    4. Müller A
    5. Dibrov A
    6. Jain A
    7. Wilhelm B
    8. Schmidt D
    9. Broaddus C
    10. Culley S
    11. Rocha-Martins M
    12. Segovia-Miranda F
    13. Norden C
    14. Henriques R
    15. Zerial M
    16. Solimena M
    17. Rink J
    18. Tomancak P
    19. Royer L
    20. Jug F
    21. Myers EW

    (2018) Content-aware image restoration: pushing the limits of fluorescence microscopy

    Nature Methods 15:1090–1097.

    https://doi.org/10.1038/s41592-018-0216-7

    • PubMed
    • Google Scholar
32. 1. Xia C
    2. Fan J
    3. Emanuel G
    4. Hao J
    5. Zhuang X

    (2019) Spatial transcriptome profiling by MERFISH reveals subcellular RNA compartmentalization and cell cycle-dependent gene expression

    PNAS 116:19490–19499.

    https://doi.org/10.1073/pnas.1912459116

    • PubMed
    • Google Scholar
33. 1. Zhang M
    2. Eichhorn SW
    3. Zingg B
    4. Yao Z
    5. Cotter K
    6. Zeng H
    7. Dong H
    8. Zhuang X

    (2021) Spatially resolved cell atlas of the mouse primary motor cortex by MERFISH

    Nature 598:137–143.

    https://doi.org/10.1038/s41586-021-03705-x

    • PubMed
    • Google Scholar
34. 1. Zhuang X

    (2021) Spatially resolved single-cell genomics and transcriptomics by imaging

    Nature Methods 18:18–22.

    https://doi.org/10.1038/s41592-020-01037-8

    • PubMed
    • Google Scholar
35. Software

    1. ZhuangLab

    (2023) Storm-control, version swh:1:rev:07689317aed5e7a001a7694655389e1f2e2e2d5e

    Software Heritage.

    https://archive.softwareheritage.org/swh:1:dir:e71452362132a3e8af9e6b64349aea123e331f8c;origin=https://github.com/ZhuangLab/storm-control;visit=swh:1:snp:384a489a81cce76b8c60a4a245c94bd42c655796;anchor=swh:1:rev:07689317aed5e7a001a7694655389e1f2e2e2d5e

Author details

1. Rongxin Fang

   Howard Hughes Medical Institute, Department of Chemistry and Chemical Biology, Department of Physics, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing

   Contributed equally with

   Aaron Halpern

   Competing interests

   inventor on patents applied for by Harvard University related to thick-tissue 3D MERFISH. (Serial number: 63/506,283)

   "This ORCID iD identifies the author of this article:" 0000-0003-0107-7504

2. Aaron Halpern

   Howard Hughes Medical Institute, Department of Chemistry and Chemical Biology, Department of Physics, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Conceptualization, Software, Formal analysis, Validation, Investigation, Methodology, Writing – review and editing

   Contributed equally with

   Rongxin Fang

   Competing interests

   inventor on patents applied for by Harvard University related to thick-tissue 3D MERFISH. (Serial number: 63/506,283)

3. Mohammed Mostafizur Rahman

   Howard Hughes Medical Institute, Department of Molecular and Cellular Biology, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Investigation, Writing – review and editing

   Competing interests

   No competing interests declared

4. Zhengkai Huang

   Howard Hughes Medical Institute, Department of Chemistry and Chemical Biology, Department of Physics, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Investigation, Writing – review and editing

   Competing interests

   No competing interests declared

5. Zhiyun Lei

   Howard Hughes Medical Institute, Department of Chemistry and Chemical Biology, Department of Physics, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Formal analysis, Writing – review and editing

   Competing interests

   No competing interests declared

6. Sebastian J Hell

   Howard Hughes Medical Institute, Department of Chemistry and Chemical Biology, Department of Physics, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Formal analysis, Writing – review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0009-0006-6291-9639

7. Catherine Dulac

   Howard Hughes Medical Institute, Department of Molecular and Cellular Biology, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Resources, Supervision, Funding acquisition, Writing – review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-5024-5418

8. Xiaowei Zhuang

   Howard Hughes Medical Institute, Department of Chemistry and Chemical Biology, Department of Physics, Center for Brain Science, Harvard University, Cambridge, United States

   Contribution

   Conceptualization, Resources, Supervision, Funding acquisition, Investigation, Methodology, Writing – original draft, Project administration, Writing – review and editing

   For correspondence

   zhuang@chemistry.harvard.edu

   Competing interests

   cofounder and consultant of Vizgen.inventor on patents applied for by Harvard University related to thick-tissue 3D MERFISH. (Serial number: 63/506,283)

   "This ORCID iD identifies the author of this article:" 0000-0002-6034-7853

• 2,980

  views

• 186

  downloads

• 2

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.