Three-dimensional imaging system and method for identifying nucleic acids in thick tissues

Abstract

This disclosure generally relates to microscopy techniques, including confocal microscopy. In some cases, techniques such as MERFISH can be used to identify nucleic acids in a sample in images acquired using a confocal microscope. In some cases, nucleic acids can be identified in three dimensions. In some cases, relatively thick samples, e.g., samples with a thickness of at least 100 micrometers, can be identified. In some embodiments, deep learning or other machine learning techniques can be used to improve image quality and / or accelerate the confocal imaging process.

Description

[Technical Field]

[0001] Related applications This application claims the interests of U.S. Provisional Patent Application No. 63 / 506,283, filed on 5 June 2023, entitled “Three-Dimensional Single-Cell Transcriptome Imaging of Thick Tissues,” which is incorporated herein by reference in its entirety.

[0002] Reference to electronic sequence lists The contents of the electronic sequence listing (H049870798WO00-SEQ-TC.xml; size: 9,274 bytes; and creation date: June 4, 2024) are incorporated herein by reference in their entirety.

[0003] field This disclosure generally relates to microscopy, including confocal microscopy, and spatial genomics. [Background technology]

[0004] Multiplexed error-robust fluorescence in-situ hybridization (MERFISH) enables genome-scale imaging of nucleic acids such as RNA and DNA, and epigenetic elements, in individual cells within intact tissue. To date, MERFISH has been applied to imaging relatively thin tissue samples.

[0005] Recent advances in genome-scale imaging techniques have enabled in-situ gene expression profiling, 3D genome imaging, and epigenome profiling of individual cells, as well as the identification and spatial mapping of molecularly defined cell types in intact tissues. Among these methods, multiplexed error-robust fluorescence in-situ hybridization (MERFISH) allows for the simultaneous imaging of thousands of genes by employing combinatorial labeling, which assigns unique barcodes to individual genes, sequential round imaging to read barcodes from individual RNA molecules, and an error-robust barcoding scheme that ensures high detection accuracy. MERFISH has also been extended to enable spatially degraded 3D genome (3D-gemone) imaging and epigenome profiling of individual cells. See, for example, U.S. Patent No. 11,098,303, incorporated herein by reference. [Overview of the project] [Problems that the invention aims to solve]

[0006] MERFISH measurements have traditionally been performed on thin tissue sections approximately 10 micrometers thick. However, tissue deformation during sectioning makes it difficult to align images of consecutive thin sections and identify the 3D molecular and cellular structure of the tissue. Furthermore, it is difficult to reconcile MERFISH images of thin sections with images of other cellular properties, such as cell morphology and neuronal activity, obtained from imaging thick tissues or living animals, hindering multimodal characterization of the molecular, morphological, and functional properties of cells. Therefore, there is a strong desire to extend MERFISH to thick tissue imaging. However, thick tissue MERFISH imaging presents several challenges. Firstly, fluorescence background caused by out-of-focus signals degrades image quality. Secondly, spherical aberration resulting from refractive index mismatch leads to reduced image resolution and quality in the deeper parts of thick samples. Thirdly, the MERFISH protocol, which has been optimized for imaging thin specimens, is not suitable for imaging thick specimens. Therefore, further improvements are needed. [Means for solving the problem]

[0007] This disclosure generally relates to microscopy techniques, including confocal microscopy. The subject matter of this disclosure may, in some cases, involve related products, alternative solutions to specific problems, and / or multiple different uses of one or more systems and / or articles.

[0008] One embodiment generally involves acquiring images of a sample using a confocal microscope and identifying nucleic acids within the sample using MERFISH.

[0009] Another aspect generally involves acquiring images of a sample using a confocal microscope and using those images to three-dimensionally identify nucleic acids within the sample.

[0010] Another aspect generally involves exposing a sample to multiple nucleic acid probes, identifying the binding of each nucleic acid probe to the sample by acquiring an image of the sample using a confocal microscope, constructing a codeword based on the binding of the nucleic acid probes, and matching at least a portion of the codewords to a valid codeword, and if no match is found, applying error correction to the codeword to form a valid codeword.

[0011] Another aspect generally involves acquiring images of a sample using a confocal microscope, improving the images using machine learning, and identifying nucleic acids within the sample.

[0012] Another aspect generally involves acquiring images of a sample using a confocal microscope, improving the images using machine learning, and identifying nucleic acids within the sample using MERFISH.

[0013] Another aspect generally involves using a confocal microscope to acquire images of a sample and identifying nucleic acids within the sample by exposing the sample to multiple nucleic acid probes and identifying the binding of the multiple nucleic acid probes to the sample.

[0014] Another aspect generally involves using a confocal microscope to acquire images of the sample, using machine learning to improve the images, and identifying nucleic acids in the sample by exposing the sample to multiple nucleic acid probes and identifying the binding of the multiple nucleic acid probes to the sample.

[0015] In another aspect, this disclosure includes methods for producing one or more embodiments described herein. In yet another aspect, this disclosure includes methods for using one or more embodiments described herein.

[0016] Other advantages and novel features of the present disclosure will become apparent from the following detailed description of various non-limiting embodiments of the present disclosure when considered in conjunction with the accompanying drawings.

[0017] Non-limiting embodiments of the present disclosure are illustrated by reference to the accompanying drawings, which are schematic and not intended to be drawn to scale. In the figures, each identical or nearly identical component that is illustrated is typically represented by a single reference numeral. For clarity, not all components are shown in all figures, nor are all components of each embodiment of the present disclosure shown, where illustration is not necessary for one of ordinary skill in the art to understand the present disclosure. In the figures:

Brief Description of the Drawings

[0018] [Figure 1] Figures 1A - 1D show improved performance of confocal MERFISH imaging by deep learning according to one embodiment. [Figure 2] Figures 2A - 2G show imaging of thick brain tissue sections in another embodiment. [Figure 3] Figures 3A - 3F show the spatial organization of cell types in tissue sections in yet another embodiment. [Figure 4] Figure 4 shows a comparison of epifluorescece images and confocal images in yet another embodiment. [Figure 5] Figures 5A - 5D show MERFISH images of RNA molecules in tissue in yet another embodiment. [Figure 6] Figures 6A - 6E show MERFISH encoding and readout in yet another embodiment. [Figure 7] Figures 7A - 7D show displacement of RNA molecules in yet another embodiment. [Figure 8] Figures 8A - 8C show gel swelling in one embodiment. [Figure 9]Figures 9A-9C show 3D MERFISH imaging in mouse tissue in another embodiment. [Figure 10] Figure 10 shows cell type identification in mouse tissue in one embodiment. [Figure 11] Figures 11A-11B show MERFISH imaging in mouse tissue in another embodiment. [Figure 12] Figures 12A-12B show cell type identification in mouse tissue in yet another embodiment. [Figure 13] Figures 13A-13B show multiple read sequences distributed among different populations of nucleic acid probes according to another embodiment. **DETAILED DESCRIPTION**

[0019] The present disclosure generally relates to microscopy techniques, including confocal microscopy. In some cases, techniques such as MERFISH can be used to identify nucleic acids within a sample in an image obtained using confocal microscopy. In some cases, nucleic acids can be identified three-dimensionally. In some cases, relatively thick samples, e.g., samples having a thickness greater than 10 micrometers or greater than 100 micrometers, can be identified. In some embodiments, deep learning or other machine learning techniques can be used to improve image quality and / or speed up the confocal imaging process.

[0020] One aspect generally relates to systems and methods for imaging or identifying nucleic acids within cells or other samples. The nucleic acids can be identified two-dimensionally or three-dimensionally. In some embodiments, the transcriptome of a cell can be identified. Certain embodiments are directed to identifying nucleic acids such as mRNA within a cell at a relatively high resolution. In some embodiments, a plurality of nucleic acid probes are applied to a sample, and their binding within the sample can be identified using, e.g., confocal microscopy, to identify the positions of the nucleic acid probes within the sample.

[0021] The sample may include cell cultures, cell suspensions, biological tissues, biopsies, and other organisms. The sample may also contain nucleic acids even if it does not contain cells. If the sample contains cells, these cells may be human cells or any other suitable cells, such as mammalian cells, fish cells, insect cells, or plant cells. More than one cell type may be present in some cases.

[0022] In certain embodiments, confocal microscopy can be used to analyze relatively thick samples, such as samples with thicknesses of at least 10 micrometers, at least 20 micrometers, at least 30 micrometers, at least 40 micrometers, at least 50 micrometers, at least 60 micrometers, at least 70 micrometers, at least 80 micrometers, at least 90 micrometers, at least 100 micrometers, at least 110 micrometers, at least 125 micrometers, at least 150 micrometers, at least 175 micrometers, at least 200 micrometers, at least 225 micrometers, at least 250 micrometers, at least 300 micrometers, at least 350 micrometers, at least 400 micrometers, at least 450 micrometers, at least 500 micrometers, etc.

[0023] In some cases, confocal microscopy uses spatial pinholes, rotating disks, or other techniques to block out-of-focus light in image formation. In some embodiments, it may be possible to reconstruct a three-dimensional image of a sample by capturing images at different depths of focus of the sample. For example, different images of a sample at different depths of focus can be obtained. In some cases, the images may be spaced apart (e.g., in the z-axis direction) at intervals of at least 1 micrometer, at least 2 micrometers, at least 3 micrometers, at least 4 micrometers, at least 5 micrometers, at least 6 micrometers, at least 7 micrometers, at least 8 micrometers, at least 9 micrometers, at least 10 micrometers, at least 12 micrometers, at least 15 micrometers, at least 20 micrometers, etc.

[0024] Various confocal microscopes can be used according to various embodiments. For example, confocal microscopy methods may include laser scanning confocal microscopes, spinning disk confocal microscopes, dual spinning disk confocal microscopes, and programmable array microscopes (e.g., those that can generate a set of moving pinholes using an electronically controlled spatial light modulator).

[0025] In addition, in some embodiments, one or more acquired images may be improved using machine learning or artificial intelligence techniques such as deep learning, graph convolutional networks, reinforcement learning, neural networks, and recurrent neural networks. In some cases, the machine learning technique may involve, for example, training a machine learning model on a training set of acquired images as considered herein, and then using the trained machine learning model to improve the images. In some cases, the model may be trained using a dataset having appropriate properties (e.g., accuracy, image sharpness, edges in the image, color, shape, etc.). In some cases, the training data may be labeled, for example, to enable supervised learning, unsupervised learning, reinforcement techniques, etc. For example, one or more images may be improved by reducing blur, increasing edge sharpness, improving boundary sharpness, and improving contrast. In some embodiments, the output image may have improved properties or image segments having improved properties. Examples, though not limited to, include improvements in the signal-to-noise ratio or signal-to-background ratio. In some cases, the trained machine learning model may be updated, for example, based on an evaluation of the improved image. In some embodiments, the output of a machine learning model (e.g., one or more improved images) can be evaluated against a target (e.g., nucleic acid localization). In some embodiments, the trained machine learning model can be updated through reinforcement learning or other techniques. Furthermore, in some cases, a pre-trained model can be used that can take input and generate improved image outputs.

[0026] The various methods or processes outlined herein can be coded as software executable on one or more processors employing any one of various operating systems or platforms, and in some cases can be used to improve the underlying computer system on which the software is implemented. Furthermore, such software can be written using any of a number of suitable programming languages ​​and / or programming tools or scripting tools, and can also be compiled as executable machine code or intermediate code that runs on a virtual machine or a suitable framework.

[0027] In this regard, the various inventive concepts described herein may be embodied as at least one non-transient computer-readable storage medium (e.g., computer memory, one or more floppy disks, compact disks, optical disks, magnetic tapes, flash memory, field-programmable gate arrays, or circuit configurations in other semiconductor devices) encoded with one or more programs that implement various embodiments when executed on one or more computers or other processors. The one or more non-transient computer-readable media may be transportable so that the one or more programs stored thereon can be loaded onto any computer resource to implement the various embodiments discussed above.

[0028] In this specification, the terms “program,” “software,” and / or “application” are used in a general sense to refer to any type of computer code or set of computer executable instructions that can be employed to program a computer or other processor to implement various aspects of the embodiments discussed above. Furthermore, it should be understood that, according to one aspect, one or more computer programs that, when executed, perform in the manner discussed herein do not need to reside on a single computer or processor, but can be modularly distributed across different computers or processors to implement various embodiments.

[0029] Computer executable instructions can take many forms, such as program modules, which are executed by one or more computers or other devices. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. Typically, the functions of a program module can be combined or distributed as desired in various embodiments.

[0030] Furthermore, data structures can be stored in any suitable form on a non-transient, computer-readable storage medium. A data structure may have fields that relate to each other via locations within the data structure. Such relationships can be achieved by allocating storage for fields that also have locations in a non-transient, computer-readable medium that convey relationships between fields. However, any suitable mechanism can be used to establish relationships between information within the fields of a data structure, including the use of pointers, tags, or other mechanisms for establishing relationships between data elements.

[0031] As mentioned above, samples can be partially or completely immobilized or embedded in polymers or gels. In some cases, samples can be embedded in relatively large polymers or gels, which can then be sectioned or sliced, for example, using various microtome techniques commonly available to those skilled in the art, to produce smaller portions for analysis. For example, tissues or organs can be immobilized in suitable polymers or gels.

[0032] In some embodiments, a variety of polymers can be used. In some cases, a polymer that is relatively optically transparent can be selected. The polymer may also not deform significantly during the polymerization process, although in some cases the polymer may exhibit some deformation. In some cases, the amount of deformation can be specified as a relative change in size of less than 5, less than 4, less than 3, less than 2, less than 1.5, less than 1.3, or less than 1.2 (i.e., a change in size of 2 means that the sample doubles in linear dimensions), or the reciprocal of these (i.e., a change in size of the reciprocal of 2 means that the sample halves in linear dimensions).

[0033] In some embodiments, the gel may be selected to exhibit relatively low expansion under various conditions, for example. For instance, the gel may exhibit linear expansion of 10% or less, 9% or less, 8% or less, 7% or less, 6% or less, 5% or less, 4% or less, 3% or less, 2% or less, or 1% or less, for example, when various buffers are applied.

[0034] For example, in some cases, the gel can be prepared at relatively low temperatures, such as below 25°C, below 20°C, below 15°C, below 10°C, below 8°C, below 6°C, below 4°C, below 2°C, or below 0°C. Furthermore, in some cases, the gel can be prepared using relatively low concentrations of initiators, such as less than 1% vol / vol, less than 0.5% vol / vol, less than 0.2% vol / vol, or less than 0.1% vol / vol.

[0035] Non-limiting examples of suitable polymers include polyacrylamide and agarose. In some cases, the polymer is a gel or hydrogel. Various polymers can be used in various embodiments involving chemical crosslinking between gel subunits, including but not limited to acrylic acid, acrylamide, ethylene glycol diacrylate, ethylene glycol dimethacrylate, and poly(ethylene glycol dimethacrylate); as well as hydrophobic or hydrogen bonding interactions such as poly(N-isopropylacrylamide), methylcellulose, (ethylene oxide)-(propylene oxide)-(ethylene oxide) terpolymer, sodium alginate, poly(vinyl alcohol), alginate, chitosan, gum arabic, gelatin, and agarose.

[0036] In one embodiment, an anchor probe can be used during the polymerization process. The anchor probe may include a moiety that can polymerize with the polymer during the polymerization process and, for example, can immobilize a target chemically and / or physically. For example, in the case of polyacrylamide, the anchor probe may include an acrydite moiety that can polymerize and be incorporated into the polymer.

[0037] An anchor probe may also include a portion that can interact with and bind to a nucleic acid molecule, or other molecules to be immobilized, such as proteins or lipids, or other desired targets. Immobilization can be covalent or non-covalent. For example, to immobilize a target nucleic acid, the anchor probe may include a nucleic acid containing an acridite portion (e.g., 5' end, 3' end, internal bases, etc.) and a nucleic acid sequence substantially complementary to at least a portion of the target nucleic acid. For example, the nucleic acid may be complementary to at least 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20 or more nucleotides of the nucleic acid. In some cases, the complementarity may be exact (Watson-Crick complementarity), or there may be one, two or more mismatches. In some cases, the anchor probe may be configured to immobilize mRNA, for example, in the case of transcriptome analysis. For example, in one embodiment, the anchor probe may contain multiple thymine nucleotides, for example, sequentially, to bind to the poly-A tail of mRNA. Therefore, for example, the anchor probe may have at least 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20 or more consecutive thymine nucleotides (SEQ ID NO: 2) (e.g., poly-dT moiety). In some cases, at least a portion of the thymine nucleotides may be "locked" thymine nucleotides. These may constitute at least 20%, at least 30%, at least 40%, at least 50%, at least 60%, at least 70%, or at least 80% of these thymine nucleotides. In certain embodiments, locked and unlocked nucleotides may be present alternately. Such locked thymine nucleotides may be useful, for example, to stabilize the hybridization of the poly-A tail of mRNA with the anchor probe.

[0038] Other methods can be used to anchor nucleic acids, or other molecules to be immobilized. In one set of embodiments, nucleic acids such as DNA or RNA can be immobilized by covalent bonding. For example, in one set of embodiments, an alkylating agent can be used that contains a second chemical moiety that can covalently bond to RNA or DNA and be incorporated into polyacrylamide during polymerization. In yet another set of embodiments, the terminal ribose of an RNA molecule can be oxidized with sodium periodate (or another oxidizing agent) to produce an aldehyde, which can then be crosslinked to acrylamide or other polymers or gels. In other embodiments, chemical agents that can modify bases, such as aldehydes, such as paraformaldehyde or glutaraldehyde, alkylating agents, or succinimidyl-containing groups; chemical agents that modify terminal phosphates, such as carboimide, such as EDC(1-ethyl-3-(3-dimethylaminopropyl)carbodiimide); chemical agents that modify internal sugars, such as p-maleimide-phenylisocyanate; or chemical agents that modify terminal sugars, such as sodium periodate can be used. In some cases, these chemical agents can then support a second chemical moiety that can be directly crosslinked to a gel or polymer, and / or this second chemical moiety can be further modified with a compound that can be directly crosslinked to a gel or polymer.

[0039] In yet another embodiment, nucleic acids can be immobilized using an anchor probe having a portion substantially complementary to the DNA or RNA. Between the anchor probe and the nucleic acid, there may be 5, 6, 7, 8, 9, 10, 12, 13, 14, 15, 20, 25, 30, 35, 40, 45, 50 or more complementary nucleotides. In yet another set of embodiments, nucleic acids may become physically entangled in the polymer or gel, for example, due to their length, and therefore may not be able to diffuse from their original positions in the gel.

[0040] In other embodiments, similar anchor probes can be used to immobilize other components onto the polymer or gel. For example, in one embodiment, an antibody that can specifically bind to a suitable target (e.g., another protein, lipid, carbohydrate, virus, etc.) can be modified to include an acridite moiety that allows for its incorporation into the polymer or gel.

[0041] Furthermore, it should be understood that the embedding of the sample within the matrix and the immobilization of nucleic acids (or other desired targets) can be carried out in any suitable order in various embodiments. For example, immobilization may be performed before, during, or after the embedding of the sample. In some cases, the target may be chemically modified or reacted to crosslink with the gel or polymer before or during the formation of the gel or polymer.

[0042] After immobilizing nucleic acids or other suitable molecules onto a polymer or gel, other components in the sample can be "cleared." Such clearance involves the removal of components and / or the degradation of non-target components (e.g., into smaller components, non-fluorescent components). In some cases, at least 50%, at least 60%, at least 70%, at least 80%, or at least 90% of undesirable components in the sample may be cleared. In certain embodiments, for example, multiple clearance steps may be performed to remove various undesirable components. As will be considered, the removal of such components can reduce the background during analysis (e.g., by reducing background and / or off-target binding), while the desired components (such as nucleic acids) are immobilized and therefore not removed.

[0043] For example, proteins can be clarified from a sample using enzymes, denaturants, chelating agents, and chemical agents, which can break down proteins into smaller components and / or amino acids. These smaller components are easily removed physically and / or may be small enough or inactive to have little impact on the background. Similarly, lipids can be clarified from a sample using surfactants and the like. In some cases, one or more of these may be used, for example, simultaneously or sequentially. Non-limiting examples of suitable enzymes include proteinases such as proteinase K, proteases or peptidases, or digestive enzymes such as trypsin, pepsin, or chymotrypsin. Non-limiting examples of suitable denaturants include guanidine hydrochloride, acetone, acetic acid, urea, or lithium perchlorate. Non-limiting examples of chemical agents that can denature proteins include solvents such as phenol, chloroform, guanidinium isocyanate, urea, and formamide. Non-limiting examples of surfactants include Triton X-100 (polyethylene glycol p-(1,1,3,3-tetramethylbutyl)-phenyl ether), SDS (sodium dodecyl sulfate), Igepal CA-630, or poloxamer. Non-limiting examples of chelating agents include ethylenediaminetetraacetic acid (EDTA), citrate, or polyaspartic acid. In some embodiments, such compounds can be applied to a sample to clear proteins, lipids, and / or other components. For example, a buffer (e.g., Tris or tris(hydroxymethyl)aminomethane) can be applied to the sample and then removed.

[0044] Non-limiting examples of DNA enzymes that can be used to remove DNA include DNase I, dsDNase, and various restriction enzymes. Non-limiting examples of techniques for clearing RNA include RNA enzymes such as RNase A, RNase T, or RNase H, or chemical agents, such as alkaline hydrolysis (e.g., by raising the pH above 10). Non-limiting examples of systems for removing sugars or extracellular matrix include enzymes such as chitinase, heparinase, or other glycosylases. Non-limiting examples of systems for removing lipids include enzymes such as lipidase, chemical agents such as alcohol (e.g., methanol or ethanol), or surfactants such as Triton X-100 or sodium dodecyl sulfate. Many of these are readily available commercially. In this way, the background of the sample can be removed, thereby facilitating the analysis of nucleic acid probes or other desired targets using, for example, fluorescence microscopy or other techniques discussed herein. As described above, in various embodiments, various targets (e.g., nucleic acids, certain proteins, lipids, viruses, etc.) can be immobilized, while other non-targets can be cleared using appropriate drugs or enzymes. In a non-limiting example, when proteins (such as antibodies) are immobilized, systems for removing RNA enzymes, DNA enzymes, lipids, sugars, etc., can be used.

[0045] Using these techniques, in some embodiments, nucleic acids in a sample can be identified. The identified nucleic acids may be, for example, DNA, RNA, epigenetic elements, or other nucleic acids present in cells (or other samples). The nucleic acids may be endogenous in cells or added to cells. For example, the nucleic acids may be viral or artificially produced. In some cases, the identified nucleic acids may be expressed by cells. In some embodiments, the nucleic acid is RNA. RNA may be coding RNA and / or non-coding RNA. Non-limiting examples of RNA that can be studied in cells include mRNA, siRNA, rRNA, miRNA, tRNA, lncRNA, snoRNA, snRNA, exRNA, or piRNA.

[0046] In some cases, it is possible to study the majority of nucleic acids within a cell. For example, in some cases, it is possible to identify sufficient RNA present in a cell and construct a partial or complete transcriptome of the cell. In some cases, at least four types of mRNA are identified within the cell, and in some cases, at least three, at least four, at least seven, at least eight, at least twelve, at least four, at least fifteen, at least sixteen, at least two two, at least three ten, at least three one, at least three two, at least five ten, at least six three, at least sixteen, at least seven two, at least seven fifteen, at least one hundred, at least one two hundred, at least one two hundred, at least one fourteen, at least two fifteen, at least two fifteen, at least two fifteen, and at least two fifteen, and at least one 500, at least 1,000, at least 1,500, at least 2,000, at least 2,500, at least 3,000, at least 4,000, at least 5,000, at least 7,500, at least 10,000, at least 12,000, at least 15,000, at least 20,000, at least 25,000, at least 30,000, at least 40,000, at least 50,000, at least 75,000, or at least 100,000 types of mRNA can be identified.

[0047] In some cases, the cellular transcriptome can be identified. It should be understood that the transcriptome generally encompasses not only mRNA but all RNA molecules produced within the cell. Therefore, for example, the transcriptome may also include rRNA, tRNA, siRNA, etc. In some embodiments, at least 5%, at least 10%, at least 15%, at least 20%, at least 25%, at least 30%, at least 40%, at least 50%, at least 60%, at least 70%, at least 80%, at least 90%, or 100% of the cellular transcriptome can be identified.

[0048] The identification of one or more nucleic acids within a cell or other sample can be qualitative and / or quantitative. Furthermore, the identification can also be spatial; for example, the location of nucleic acids within a cell or other sample can be identified in two or three dimensions. In some embodiments, the location, number, and / or concentration of nucleic acids within a cell (or other sample) can be identified.

[0049] In some cases, a large portion of a cell's genome may be identified. Identified genomic segments may be contiguous or scattered across the genome. For example, in some cases, at least four genomic segments may be identified within a cell, and in some cases, at least three, at least four, at least seven, at least eight, at least twelve, at least four, at least fifteen, at least sixteen, at least two hundred, at least three hundred, at least three hundred, at least fifty, at least sixty, at least sixty, at least seventy, at least seventy, at least seventy, at least one hundred It is possible to identify at least 500, at least 1,000, at least 1,500, at least 2,000, at least 2,500, at least 3,000, at least 4,000, at least 5,000, at least 7,500, at least 10,000, at least 12,000, at least 15,000, at least 20,000, at least 25,000, at least 30,000, at least 40,000, at least 50,000, at least 75,000, or at least 100,000 genome segments.

[0050] In some cases, the entire genome of a cell may be identified. It should be understood that the genome generally includes not only chromosomal DNA but also all DNA molecules produced within the cell. Therefore, for example, the genome may also include mitochondrial DNA, chloroplast DNA, plasmid DNA, etc. In some embodiments, at least about 5%, at least about 10%, at least about 15%, at least about 20%, at least about 25%, at least about 30%, at least about 40%, at least about 50%, at least about 60%, at least about 70%, at least about 80%, at least about 90%, or 100% of the cell's genome may be identified.

[0051] In some cases, the cell's epigenome may be identified. The epigenome may include chromatin having chemically modified DNA or chemically modified histones, or chromatin having other DNA-binding proteins. In some embodiments, at least about 5%, at least about 10%, at least about 15%, at least about 20%, at least about 25%, at least about 30%, at least about 40%, at least about 50%, at least about 60%, at least about 70%, at least about 80%, at least about 90%, or 100% of the cell's epigenome may be identified.

[0052] Furthermore, certain embodiments relate to systems and methods that enable the identification of the copy number and spatial localization of thousands of RNA species, genomic segments, and / or epigenetic elements within a single cell. Some of these techniques are known to those skilled in the art, for example, as multiplexed error-robust fluorescence in situ hybridization or "MERFISH." See, for example, U.S. Patent No. 11,098,303, which is incorporated herein by reference in its entirety. In some cases, error correction may also be used. For example, in some embodiments, the codeword may be based on the binding of multiple nucleic acid probes, and in some cases, the codeword may define an error correction code to reduce or prevent misidentification of nucleic acids. In certain cases, a relatively large number of different targets can be identified with a relatively small number of labels, for example, by employing various combinatorial approaches. Error-robust encoding schemes may enable imaging of hundreds to thousands of RNA species, genomic segments, and / or epigenetic elements in a sample.

[0053] For example, as discussed herein, various nucleic acid probes can be used to identify one or more nucleic acids in cells or other samples. Probes may include nucleic acids (or entities that can hybridize to nucleic acids, for example, specifically), such as DNA, RNA, LNA (locked nucleic acid), PNA (peptide nucleic acid), or combinations thereof. In some cases, additional components may be present in the nucleic acid probe, as discussed below. Nucleic acid probes can be introduced into cells using any suitable method.

[0054] For example, in some embodiments, cells are immobilized, for instance, to preserve the position of the nucleic acid within the cell, before introducing the nucleic acid probe. Techniques for immobilizing cells are known to those skilled in the art. In non-limiting examples, cells can be immobilized using chemicals such as formaldehyde, paraformaldehyde, glutaraldehyde, ethanol, methanol, acetone, and acetic acid. In one embodiment, cells can be immobilized using Hepes-glutamate buffer-mediated organic solvent (HOPE).

[0055] Nucleic acid probes can be introduced into cells (or other samples) using any suitable method. In some cases, cells can be sufficiently permeabilized by flowing a fluid containing the nucleic acid probe around them so that the nucleic acid probe can be introduced into the cells. In some cases, cells can be sufficiently permeabilized as part of a fixation process; in other embodiments, cells can be permeabilized by exposure to certain chemicals such as ethanol, methanol, or Triton. Furthermore, in some embodiments, techniques such as electroporation or microinjection can be used to introduce nucleic acid probes into cells or other samples.

[0056] Certain embodiments generally relate to nucleic acid probes introduced into cells (or other samples). Depending on the application, the probe may include any of a variety of entities that can hybridize to nucleic acids, typically by Watson-Crick base pairing, such as DNA, RNA, LNA, and PNA. The nucleic acid probe typically contains a target sequence that can optionally bind specifically to at least a portion of the target nucleic acid. When introduced into cells or other systems, the target system can bind to a specific target nucleic acid (e.g., mRNA, or other nucleic acids considered herein). In some cases, the nucleic acid probe can be identified using a signal ring entity (e.g., considered below) and / or using a secondary nucleic acid probe that can bind to the nucleic acid probe (i.e., the primary nucleic acid probe). Identification of such nucleic acid probes is discussed in detail below.

[0057] In some cases, more than one type of (primary) nucleic acid probe may be applied to the sample, for example, simultaneously. For example, there may be at least 2, at least 5, at least 10, at least 25, at least 50, at least 75, at least 100, at least 300, at least 1,000, at least 3,000, at least 10,000, or at least 30,000 distinguishable nucleic acid probes applied to the sample, for example, simultaneously or sequentially.

[0058] The target sequence can be placed anywhere within the nucleic acid probe (or primary nucleic acid probe or encoding nucleic acid probe). The target sequence may include a region that is substantially complementary to a portion of the target nucleic acid. In some cases, that portion may be at least 50%, at least 60%, at least 70%, at least 75%, at least 80%, at least 85%, at least 90%, at least 92%, at least 94%, at least 95%, at least 96%, at least 97%, at least 98%, at least 99%, or 100% complementary. In some cases, the target sequence may have a length of at least 5, at least 10, at least 15, at least 20, at least 25, at least 30, at least 35, at least 40, at least 50, at least 60, at least 65, at least 75, at least 100, at least 125, at least 150, at least 175, at least 200, at least 250, at least 300, at least 350, at least 400, or at least 450 nucleotides. In some cases, the target sequence may consist of nucleotides with lengths of 500 or less, 450 or less, 400 or less, 350 or less, 300 or less, 250 or less, 200 or less, 175 or less, 150 or less, 125 or less, or 100 or less, or 75 or less, 60 or less, 65 or less, 60 or less, 55 or less, 50 or less, 45 or less, 40 or less, 35 or less, 30 or less, 20 or less, or 10 or less. Any combination of these is also possible, and for example, the target sequence may have lengths such as 10-30 nucleotides, 20-40 nucleotides, 5-50 nucleotides, 10-200 nucleotides, or 25-35 nucleotides, 10-300 nucleotides. Typically, complementarity is determined based on Watson-Crick nucleotide base pairs.

[0059] The target sequences of (primary) nucleic acid probes can be identified by referring to target nucleic acids suspected to be present in cells or other samples. For example, the target nucleic acid of a protein can be identified using the protein sequence by identifying the nucleic acid expressed to form the protein. In some cases, only a portion of the nucleic acid encoding the protein is used, having a length such as those discussed above. Furthermore, in some cases, more than one target sequence can be used to identify a particular target. For example, multiple probes that can bind to or hybridize to different regions of the same target can be used sequentially and / or simultaneously. Hybridization typically refers to the annealing process in which complementary single-stranded nucleic acids associate via Watson-Crick nucleotide base pairs (e.g., hydrogen bonds, guanine-cytosine, and adenine-thymine) to form double-stranded nucleic acids.

[0060] In some embodiments, nucleic acid probes, such as primary nucleic acid probes, may also include one or more “read” sequences. However, it should be understood that read sequences are not required in all cases. In some embodiments, a nucleic acid probe may include 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 or more, 20 or more, 32 or more, 40 or more, 50 or more, 64 or more, 75 or more, 100 or more, 128 or more read sequences. Read sequences can be placed anywhere within the nucleic acid probe. If more than one read sequence is present, the read sequences may be placed adjacent to each other and / or interposed with other sequences.

[0061] If a read sequence exists, its length can be arbitrary. If more than one read sequence is used, the read sequences can be independently the same length or have different lengths. For example, a read sequence could have a length of at least 5, at least 10, at least 15, at least 20, at least 25, at least 30, at least 35, at least 40, at least 50, at least 60, at least 65, at least 75, at least 100, at least 125, at least 150, at least 175, at least 200, at least 250, at least 300, at least 350, at least 400, or at least 450 nucleotides. In some cases, the read sequence may be a nucleotide with a length of 500 or less, 450 or less, 400 or less, 350 or less, 300 or less, 250 or less, 200 or less, 175 or less, 150 or less, 125 or less, or 100 or less, or a nucleotide with a length of 75 or less, 60 or less, 65 or less, 60 or less, 55 or less, 50 or less, 45 or less, 40 or less, 35 or less, 30 or less, 20 or less, or 10 or less. Any combination of these is also possible, and for example, the read sequence may have lengths such as 10-30 nucleotides, 20-40 nucleotides, 5-50 nucleotides, 10-200 nucleotides, or 25-35 nucleotides, 10-300 nucleotides.

[0062] In some embodiments, the read sequence may be arbitrary or random. In certain cases, the read sequence is selected, for example, so as not to bind or hybridize with other nucleic acids suspected to be present in the cell or other sample, in order to reduce or minimize homology with other components of the cell or other sample. In some cases, homology may be less than 10%, less than 8%, less than 7%, less than 6%, less than 5%, less than 4%, less than 3%, less than 2%, or less than 1%. In some cases, homology may be less than 20 base pairs, less than 18 base pairs, less than 15 base pairs, less than 14 base pairs, less than 13 base pairs, less than 12 base pairs, less than 11 base pairs, or less than 10 base pairs. In some cases, the base pairs are continuous.

[0063] In one embodiment, the set of nucleic acid probes may contain a certain number of read sequences, and in some cases may be fewer than the number of targets for the nucleic acid probes. Those skilled in the art will know that with one signaling entity and n read sequences, generally 2 n - Recognize that one different nucleic acid target may be uniquely identified. However, it is not necessary to use all possible combinations. For example, a population of nucleic acid probes may target 12 different nucleic acid sequences but contain 8 or fewer read sequences. As another example, a population of nucleic acids may target 140 different nucleic acids but contain 16 or fewer read sequences. Different nucleic acid sequence targets can be identified separately by using different combinations of read sequences within each probe. For example, each probe may contain read sequences 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, etc. or more. In some cases, each population of nucleic acid probes may contain the same number of read sequences, but in other cases, the number of read sequences present on different probes may differ.

[0064] As a non-limiting example, a first nucleic acid probe may include a first target sequence, a first read sequence, and a second read sequence, while a second different nucleic acid probe may include a second target sequence, the same first read sequence, but with a third read sequence instead of the second read sequence. Such probes can be distinguished by identifying the various read sequences present or associated with a given probe or position, as discussed herein.

[0065] Furthermore, nucleic acid probes (and their corresponding complementary sites on encoding probes) can be constructed using only two or three of the four bases, such as omitting all "G"s or all "C"s in the probe, in certain embodiments. Sequences lacking either "G" or "C" may, in certain embodiments, hardly form secondary structures, potentially contributing to more uniform and faster hybridization.

[0066] In some embodiments, nucleic acid probes may include signaling entities. However, it should be understood that signaling entities are not necessary in all cases; for example, in some embodiments, nucleic acid probes may be identified using secondary nucleic acid probes, as will be further discussed in detail below.

[0067] Other components may also be present in the nucleic acid probe. For example, in one embodiment, one or more primer sequences may be present to enable enzymatic amplification of the probe. Those skilled in the art will recognize primer sequences suitable for applications such as amplification (e.g., using PCR or other suitable techniques). Many such primer sequences are commercially available. Other examples of sequences that may be present in a primary nucleic acid probe include, but are not limited to, promoter sequences, operons, identification sequences, and nonsense sequences.

[0068] Typically, a primer is a single-stranded or partially double-stranded nucleic acid (e.g., DNA) that serves as a starting point for nucleic acid synthesis, allowing a polymerase enzyme, such as nucleic acid polymerase, to extend the primer and replicate the complementary strand. The primer is complementary to the target nucleic acid and hybridizes (e.g., is designed to do so). In some embodiments, the primer is a synthetic primer. In some embodiments, the primer is a primer that does not exist in nature. Primers typically have a length of 10 to 50 nucleotides. For example, primers may have lengths of 10 to 40, 10 to 30, 10 to 20, 25 to 50, 15 to 40, 15 to 30, 20 to 50, 20 to 40, or 20 to 30 nucleotides. In some embodiments, the primer has a length of 18 to 24 nucleotides.

[0069] Furthermore, the components of the nucleic acid probe can be arranged in any suitable order. For example, in one embodiment, the components may be arranged in the nucleic acid probe as primer-reading sequence-targeting sequence-reading sequence-reverse primer. In this structure, each "reading sequence" can contain any number of reading sequences (including 0), as long as at least one reading sequence is present in the probe. Non-limiting structural examples include primer-targeting sequence-reading sequence-reverse primer, primer-reading sequence-targeting sequence-reverse primer, targeting sequence-primer-targeting sequence-reading sequence-reverse primer, targeting sequence-primer-reading sequence-targeting sequence-reverse primer, targeting sequence-reading sequence-primer, reading sequence-targeting sequence-primer, reading sequence-primer-targeting sequence-reverse primer, and so on. Furthermore, the reverse primer is optional in some embodiments, including all of the examples described above.

[0070] After introducing a nucleic acid probe into cells or other samples, the nucleic acid probe can be directly identified by identifying a signaling entity (if any) according to a particular embodiment, and / or by using one or more secondary nucleic acid probes. As previously mentioned, in some cases, identification may be spatial, e.g., two-dimensional or three-dimensional. Furthermore, in some cases, identification may be quantitative, e.g., by identifying the amount or concentration of the primary nucleic acid probe (and target nucleic acid). Furthermore, the secondary probe may include any of the various entities that can hybridize to nucleic acids, e.g., DNA, RNA, LNA, and / or PNA, depending on the application.

[0071] Secondary nucleic acid probes may include recognition sequences that can bind to or hybridize with the reading sequences of primary nucleic acid probes. In some cases, the binding may be specific, or the binding may preferentially bind to or hybridize to only one of the reading sequences in which the recognition sequence exists. Secondary nucleic acid probes may also include one or more signaling entities. If more than one secondary nucleic acid probe is used, the signaling entities may be the same or different.

[0072] The recognition sequence can be of any length, and multiple recognition sequences can be the same length or different. If more than one recognition sequence is used, the recognition sequences can independently have the same length or different lengths. For example, a recognition sequence may have a length of at least 5, at least 10, at least 15, at least 20, at least 25, at least 30, at least 35, at least 40, or at least 50 nucleotides. In some cases, a recognition sequence may have a length of 75 nucleotides or less, 60 nucleotides or less, 65 nucleotides or less, 60 nucleotides or less, 55 nucleotides or less, 50 nucleotides or less, 45 nucleotides or less, 40 nucleotides or less, 35 nucleotides or less, 30 nucleotides or less, 20 nucleotides or less, or 10 nucleotides or less. Any combination of these is also possible, and for example, a recognition sequence may have lengths such as 10-30, 20-40, or 25-35 nucleotides. In one embodiment, the recognition sequence is the same length as the reading sequence. Furthermore, in some cases, the recognition sequence may be at least 50%, at least 60%, at least 70%, at least 75%, at least 80%, at least 85%, at least 90%, at least 92%, at least 94%, at least 95%, at least 96%, at least 97%, at least 98%, at least 99%, or at least 100% complementary to the reading sequence of the primary nucleic acid probe.

[0073] As will be considered, in certain embodiments, nucleic acid probes containing various “reading sequences” are used. For example, a population of primary nucleic acid probes may contain certain “reading sequences” that can bind to certain secondary nucleic acid probes, and the location of the primary nucleic acid probe is identified in the sample using, for example, a secondary nucleic acid probe containing a signaling entity. As mentioned above, in some cases, different nucleic acid probes can be prepared by combining populations of reading sequences in various combinations, for example, a relatively large number of different nucleic acid probes can be prepared using a relatively small number of reading sequences.

[0074] Therefore, in some cases, a group of primary nucleic acid probes (or other nucleic acid probes) may each contain a certain number of read sequences, some of which are shared among different primary nucleic acid probes so that the total group of primary nucleic acid probes contains a certain number of read sequences. A group of nucleic acid probes can have any suitable number of read sequences. For example, a group of primary nucleic acid probes may have 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, etc. In some embodiments, more than 20 may be possible. Furthermore, in some cases, a population of nucleic acid probes may contain a total of 1 or more, 2 or more, 3 or more, 4 or more, 5 or more, 6 or more, 7 or more, 8 or more, 9 or more, 10 or more, 11 or more, 12 or more, 13 or more, 14 or more, 15 or more, 16 or more, 20 or more, 24 or more, 32 or more, 40 or more, 50 or more, 60 or more, 64 or more, 100 or more, 128 or more possible read sequences, although some or all of the probes may each contain more than one read sequence, as discussed herein. In addition, in some embodiments, the nucleic acid probe population may contain 100 or fewer, 80 or fewer, 64 or fewer, 60 or fewer, 50 or fewer, 40 or fewer, 32 or fewer, 24 or fewer, 20 or fewer, 16 or fewer, 15 or fewer, 14 or fewer, 13 or fewer, 12 or fewer, 11 or fewer, 10 or fewer, 9 or fewer, 8 or fewer, 7 or fewer, 6 or fewer, 5 or fewer, 4 or fewer, 3 or fewer, or 2 or fewer read sequences. Any combination of these is also possible, for example, the nucleic acid probe population may contain a total of 10 to 15 read sequences.

[0075] As a non-limiting example of an approach to combinatorially producing a relatively large number of nucleic acid probes from a relatively small number of read sequences, in a population of six different types of nucleic acid probes, each containing one or more read sequences, the total number of read sequences in the population may be four or less. In this example, four read sequences are used for ease of explanation, but it should be understood that in other embodiments, a larger number of nucleic acid probes can be realized using, for example, five, eight, ten, sixteen, thirty-two, or more read sequences, or any other suitable number of read sequences described herein depending on the application. Referring here to Figure 13A, if each primary nucleic acid probe contains two different read sequences, then by using four such read sequences (A, B, C, and D), up to six probes can be identified separately. It should be noted that in this example, the order of the read sequences on the nucleic acid probe is not mandatory, i.e., "AB" and "BA" can be treated as synonymous (however, in other embodiments, the order of the read sequences is mandatory, and "AB" and "BA" do not necessarily have to be synonymous). Similarly, when using five reading sequences (A, B, C, D, and E) in a population of primary nucleic acid probes, up to 10 probes can be identified separately, as shown in Figure 13B. For example, as a person skilled in the art would know, if there are k reading sequences in the population and n reading sequences on each probe, assuming that ordering of the reading sequences is not essential, up to

[0076]

number

[0077] In some embodiments, the read sequences and / or binding patterns of nucleic acid probes in a sample can be used to define error detection and / or error correction codes, for example, to reduce or prevent nucleic acid misidentification or errors. For example, if binding is indicated (e.g., identified using a signaling entity), the location may be identified with a "1," and conversely, if binding is not indicated, the location may be identified with a "0" (and possibly vice versa). Then, for example, multiple rounds of binding identification can be performed using different nucleic acid probes to create, for example, "codewords" for their spatial locations. In some embodiments, codewords can be used for error detection and / or correction. For example, if no match is found against a given set of read sequences or binding patterns of nucleic acid probes, the match may be identified as an error, and the codeword can be structured so that error correction can be applied to the sequence to identify the correct target of the nucleic acid probe, as appropriate. In some cases, a codeword may have fewer "letters" or locations than the total number of nucleic acids encoded by the codeword, for example, if each codeword encodes a different nucleic acid.

[0078] Such error detection and / or error correction codes can take various forms. These various codes, such as Golay codes and Hamming codes, have been previously developed in other fields, such as the telecommunications industry. In one embodiment, the reading sequence or binding pattern of a nucleic acid probe is assigned such that not all possible combinations are assigned.

[0079] For example, if four read sequences are possible and a primary nucleic acid probe contains two read sequences, then up to six primary nucleic acid probes can be identified, but the number of primary nucleic acid probes used may be less than six. Similarly, for k read sequences in a population where each primary nucleic acid probe has n read sequences,

[0080]

number

[0081]

number

[0082] As another example, when using multiple rounds of nucleic acid probes, the number of rounds can be arbitrarily chosen. If each target can give two possible outcomes in each round, such as being detected or not detected, then for n rounds of probes, up to 2 n Although different targets are possible, the number of nucleic acid targets actually used is 2 n It can be any number less than 2. For example, if each round can give more than 2 possible results, such as each target being detected in a different color channel, then for n rounds of probes, 2n (For example, 3) n , 4 n ...) Different targets may be possible. In some cases, the number of nucleic acid targets actually used may be any number less than this. Furthermore, these can be assigned randomly or in a specific way to enhance the ability to detect and / or correct errors.

[0083] For example, in one embodiment, codewords or nucleic acid probes can be assigned to each other in coding space by a Hamming distance, where this Hamming distance represents the number of incorrect "readings" in a given pattern that cause the nucleic acid probes to be mistaken for different valid nucleic acid probes. In a particular case, the Hamming distance can be at least 2, at least 3, at least 4, at least 5, at least 6, and so on. Furthermore, in one embodiment, the assignments can be formed as Hamming codes, e.g., Hamming(7,4) code, Hamming(15,11) code, Hamming(31,26) code, Hamming(63,57) code, Hamming(127,120) code, and so on. In another set of embodiments, the assignments can form SECDED codes, such as SECDED(8,4) code, SECDED(16,4) code, SCEDED(16,11) code, SCEDED(22,16) code, SCEDED(39,32) code, SCEDED(72,64) code, and so on. In yet another set of embodiments, the assignments can form extended binary Golay codes, full binary Golay codes, or ternary Golay codes. In yet another set of embodiments, the assignments can represent a subset of possible values ​​that any of the above codes may take.

[0084] For example, a code having the same error correction characteristics as a SECDED code can be formed by using only binary words containing a fixed number of "1" bits, such as 4, to code a target. In another set of embodiments, the assignment can represent a subset of possible values ​​that the above-described code can take, for the purpose of addressing asymmetric read errors. For example, a code in which the number of "1" bits can be fixed for all binary words used, in which case the proportion of "0" bits measured as "1" or the proportion of "1" bits measured as "0" may differ, can eliminate biased measurements of words with different numbers of "1".

[0085] Therefore, in some embodiments, once a codeword is identified (as considered herein, for example), it can be compared to known nucleic acid codewords. If a match is found, the nucleic acid target can be identified or specified. If no match is found, it can be identified that there is an error in the codeword reading. In some cases, error correction can be applied to identify the correct codeword, thus resulting in the correct identification of the nucleic acid target. In some cases, assuming that only one error exists, the codeword can be selected such that only one possible correct codeword is available, and therefore only one correct identity of the nucleic acid target is possible. In some cases, this can also be generalized to larger codeword intervals or Hamming distances. For example, if two, three, or four errors exist (or possibly more), the codeword can be selected such that only one possible correct codeword is available, and therefore only one correct identity of the nucleic acid target is possible.

[0086] The error correction code may be a binary error correction code, or it may be based on another numbering system, such as a ternary or quaternary error correction code. For example, in one embodiment, more than one type of signaling entity may be used and assigned to different numbers in the error correction code. Thus, in an unrestricted example, a first signaling entity (or possibly more than one signaling entity) may be assigned as "1", a second signaling entity (or possibly more than one signaling entity) may be assigned as "2" ("0" indicates that no signaling entity exists), and the codewords are distributed to define a ternary error correction code. Similarly, a third signaling entity may be additionally assigned as "3" to constitute a quaternary error correction code, and so on.

[0087] As described above, in certain embodiments, signaling entities are identified, for example, to identify nucleic acid probes and / or to create codewords. In some cases, signaling entities within a sample may be spatially identified using, for example, various techniques. In some embodiments, signaling entities may be fluorescent, and techniques for identifying fluorescence within a sample, such as fluorescence microscopy or confocal microscopy, can be used to spatially identify the location of signaling entities within cells. In some cases, the location of entities within a sample may be identified in two or even three dimensions. In addition, in some embodiments, more than one signaling entity (e.g., signaling entities with different colors or emission) can be identified simultaneously and / or sequentially.

[0088] In some embodiments, the spatial location of an entity (and therefore a nucleic acid probe to which the entity may be associated) can be determined with relatively high resolution. For example, the location can be determined with a spatial resolution of, for example, better than about 100 micrometers, better than about 30 micrometers, better than about 10 micrometers, better than about 3 micrometers, better than about 1 micrometer, better than about 800 nm, better than about 600 nm, better than about 500 nm, better than about 400 nm, better than about 300 nm, better than about 200 nm, better than about 100 nm, better than about 90 nm, better than about 80 nm, better than about 70 nm, better than about 60 nm, better than about 50 nm, better than about 40 nm, better than about 30 nm, better than about 20 nm, or better than about 10 nm.

[0089] There are various techniques that allow the spatial location of an object to be optically identified or imaged, for example, using a fluorescence microscope. In some cases, the spatial location can be identified with superresolution or with a resolution better than the wavelength or diffraction limit of light. Non-limiting examples include STORM (stochastic optical reconstruction microscopy), STED (stimulated emission depletion microscopy), NSOM (near-field scanning optical microscopy), 4Pi microscopy, SIM (structured illumination microscopy), SMI (spatial modulation illumination) microscopy, RESOLFT (reversible saturated optical linear fluorescence transition microscopy), GSD (ground state depletion microscopy), SSIM (saturated structured illumination microscopy), SPDM (spectral precision distance microscopy), photo-activated localization microscopy (PALM), fluorescence-activated localization microscopy (FPALM), LIMON (3D optical nanosizing microscopy), and super-resolution optical fluctuation imaging (SOFI).

[0090] The following documents are incorporated herein by reference: U.S. Patent Nos. 10,240,146 and 11,098,303; U.S. Patent Applications Nos. 2017 / 0212986, 2017 / 0220733, 2019 / 0233812, 2019 / 0264270, 2019 / 0276881, 2022 / 0025442, 2022 / 0064697, and Publication No. 2022 / 0205983; and International Publication Nos. WO2016 / 018960, WO2016 / 018963, WO2018 / 089438, WO2018 / 089445, WO2018 / 218150, WO2020 / 123742, WO2020 / 214885, WO2021 / 102122, and WO2021 / 138078.

[0091] Furthermore, U.S. Provisional Patent Application No. 63 / 506,283, filed on June 5, 2023, by Zhuang et al., entitled "Three-Dimensional Single-Cell Transcriptome Imaging of Thick Tissues," is incorporated herein by reference in its entirety.

[0092] The following embodiments are intended to illustrate certain embodiments of the present disclosure, but not to illustrate the entire scope of the present disclosure. [Examples]

[0093] [Example 1] This embodiment presents a method for enabling three-dimensional (3D) imaging of thick tissue specimens by combining confocal microscopy for optical sectioning with deep learning to improve imaging speed and quality. This embodiment also presents a method for enabling three-dimensional (3D) single-cell genome-scale imaging of nucleic acids such as RNA, DNA, and / or epigenetic elements in thick tissue specimens by integrating MERFISH with confocal microscopy and deep learning. This embodiment demonstrates 3D MERFISH with high detection efficiency and accuracy in mouse brain tissue sections up to 200 micrometers thick. 3D MERFISH imaging of thick tissues, and general 3D thick tissue imaging, are expected to facilitate a wide range of biological applications.

[0094] The approach of this embodiment addresses the aforementioned challenges by using spinning disk confocal microscopy to remove out-of-focus fluorescence background, exploring deep learning to accelerate the confocal imaging process, utilizing refractive index-matched objective lenses to eliminate depth-induced spherical aberration, and optimizing the MERFISH protocol for thick tissue imaging. While this embodiment presents a demonstration for identifying RNA species in thick tissue samples, this disclosure is not so limited and should be understood as being generalizable to measure proteins and / or other nucleic acids such as DNA and epigenetic elements, as discussed herein.

[0095] This embodiment significantly improved the signal-to-noise ratio (SNR) in MERFISH images of thick tissue sections by achieving optical sectioning using spinning disk confocal microscopy and eliminating out-of-focus fluorescence background (Figure 4). However, the spinning disk confocal detection geometry also cuts off a large amount of intrafocal fluorescence signal, so to achieve a high SNR for imaging individual RNA molecules in the sample, the exposure time per imaging frame must be significantly longer or the illumination light intensity must be increased. This results in a significant decrease in imaging speed or substantial photobleaching of out-of-focus fluorescent dyes before they are imaged.

[0096] Deep learning has been used to improve the quality of fluorescence microscopy images in various applications and can potentially improve the signal-to-noise ratio (SNR) of confocal MERFISH images acquired at high speeds or low light. To validate this approach, we performed MERFISH imaging of 242 genes in mouse cerebral cortex sections, imaging the same field of view (FOV) at slow (1 sec) and fast (0.1 sec) frame rates, obtaining pairs of high-SNR and low-SNR images, respectively. As expected, the low-SNR MERFISH images acquired at a frame rate of 0.1 sec showed significantly reduced detection efficiency (down to one-quarter) compared to the high-SNR measurements acquired with a 1 sec exposure time (Figures 1A and 1B).

[0097] To verify whether deep learning could improve the quality of 0.1-second images, a neural network was trained on a subset of short-exposure and long-exposure image pairs, and then this model was used to improve the quality of the remaining short-exposure images. This deep learning approach significantly improved the signal-to-noise ratio (SNR) of the 0.1-second images (Figure 1C). As a result, the detection accuracy of MERFISH images acquired at a 0.1-second frame rate improved to nearly the same level as that measured at a 1-second frame rate (Figure 1D). This advancement makes it possible to acquire high-quality confocal MERFISH images at high speed under low light conditions.

[0098] In imaging thick tissues, it is sometimes crucial to avoid aberrations caused by refractive index mismatch. High numerical aperture (NA=1.4) oil immersion objectives have long been used for MERFISH imaging of thin tissues. While these objectives are highly efficient at detecting 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 the tissue sample results in large aberrations, reducing both the sensitivity and accuracy of RNA detection in deep regions of thick samples. To overcome this problem, water immersion objectives with a good refractive index match to the sample have been used. These water objectives, despite having a lower NA (NA=1.15), enabled the detection of individual RNA molecules across the entire depth of tissue sections with thicknesses of 100 and 200 micrometers without a significant attenuation of detection sensitivity and efficiency (Figures 5A-D).

[0099] To optimize thick tissue imaging, 242 genes were imaged in 100-micrometer thick mouse brain sections at a step size of 1 micrometer per z-plane. In MERFISH, cellular RNA was labeled with a library of encoding oligonucleotide probes containing readout sequences that identified barcodes, and then the barcodes were detected bit by bit with readout oligonucleotide probes conjugated with fluorescent dyes. Optimizing probe concentration and incubation time (Figures 6A–6E) resulted in bright, consistent signals from individual RNA molecules across the entire depth of the 100-micrometer thick tissue sample for each bit. Unexpectedly, despite consistent detection of RNA molecules in individual bits, the RNA copy number detected for individual genes per z-plane decreased significantly with tissue depth and showed poor correlation with bulk RNA-seq data (Figures 7A–7C). This is likely due to displacement of RNA molecules between imaging rounds of the thick tissue sample, making it difficult to decode and identify these molecules from multi-bit images. To verify this, fiducial beads embedded in the polyacrylamide gel used for embedding MERFISH samples were imaged. Significant displacements of the bead positions were observed in all three dimensions (x, y, and z) between imaging rounds, particularly in deeper parts of the sample (Figure 7D).

[0100] The displacement of RNA molecules between imaging rounds was thought to be due to one or more factors. For example, the piezo actuator used for z-scanning may not have consistently positioned the sample at the predefined z-position during each imaging round. The polyacrylamide gel is prone to expansion or contraction when buffer conditions change, and its size is not constant, which may have contributed to the displacement of RNA molecules between imaging rounds. Additionally, although two-color imaging was used to measure two bits in each hybridization round, the on-axial chromatic aberration between the two colors increases with increasing imaging depth, which may have caused misalignment of RNA molecules between bits.

[0101] To address the initial problem, various piezo actuators were tested, and the one with the highest accuracy and reproducibility (Queensgate OP400) was selected for this experiment. This allowed the z-position error to be negligible when the sample thickness was less than 200 micrometers. To minimize the gel expansion effect, a polyacrylamide gel with a low degree of buffer-dependent expansion was prepared using a low concentration of gel initiator and a low polymerization temperature (4°C), and a MERFISH buffer was selected to further minimize the gel expansion effect (Figures 8A-8B). Furthermore, to ensure that the gel recovered to the same size before imaging, the gel was relaxed for 10 minutes after two rounds of washing with imaging buffer to completely remove any remaining cutting or probe incubation buffer-induced gel expansion (Figure 8C). To eliminate the effects of chromatic aberration, axial chromatic aberration was calibrated by imaging fiducial beads in two color channels. This allowed for precise image alignment between these channels.

[0102] Next, the optimized thick tissue MERFISH protocol was validated by first imaging the expression of 242 genes in 100-micrometer thick mouse cerebral cortex sections. In addition to RNA MERFISH imaging, DAPI staining for 3D cell segmentation and imaging of total polyadenylated mRNA signals were performed (Figure 2A). Individual RNA molecules were identified in the thick tissue MERFISH images (Figures 2B-2C; Figure 9A). The mean RNA copy number of individual genes correlated highly with the amount of RNA measured by bulk RNA sequencing (Pearson correlation coefficient r = 0.82) (Figure 2D). The correlation between the number of detected transcripts and bulk RNA-seq did not show attenuation across all tissue depths (Figures 2e, f). The RNA copy number of individual genes detected per unit area in each z-plane was compared with the results of previously performed 10-micrometer thin tissue MERFISH measurements using an epifluorescence setup. The detection efficiency of the thick tissue measurements was approximately 20% higher than that of the thin tissue measurements (Figure 2G). This may be due to the background reduction achieved by confocal optical sectioning. In confocal imaging, the depth per z-section is reduced compared to epifluorescence imaging, which could potentially further improve the detection efficiency of thick tissue measurements by 3D MERFISH.

[0103] 3D cell segmentation was performed using DAPI and total polyadenylated mRNA signaling (Figure 9B), allowing for the identification of expression profiles for 242 genes in individual cells. The RNA copy number per cell detected in a 100-micrometer thick sample was found to be twice as high in individual 10-micrometer thick sections of the same sample (obtained by dividing the 100-micrometer z-range into 10 equal-thickness sections) (Figure 9C). This is likely because many cells were only partially captured in the 10-micrometer thick tissue sections, while most cell bodies were completely captured in the 100-micrometer thick tissue. While normalizing RNA copy number by cell volume can partially mitigate this problem in thin tissue imaging, a low number of RNA molecules detected per cell can still introduce significant noise. Imposing a threshold for cell volume to reduce noise, however, reduces the number of cells measured. Furthermore, the non-uniform intracellular distribution of RNA can further degrade the accuracy of transcriptome profiling when imaging only a portion of the cells. Therefore, thick tissue MERFISH imaging should enable more accurate gene expression profiling of entire cells.

[0104] Next, using single-cell expression profiles obtained from 3D MERFISH measurements, we identified transcriptionally distinct cell populations in the mouse cerebral cortex. All previously known subclasses of excitatory neurons (L2 / 3 IT, L4 / 5 IT, L5 IT, L6 IT, L5 ET, L5 / 6 NP, L6 CT, and L6b), inhibitory neurons (marked by Lamp5, Sst, Vip, Pvalb, and Sncg, respectively), and non-neuronal cells (oligodendrocytes, oligodendrocyte progenitor cells, astrocytes, microglia, endothelial cells, pericytes, VLMCs, and smooth muscle cells) were observed (Figure 3A), as well as transcriptionally distinct cell clusters within these subclasses (Figure 10). Furthermore, 3D MERFISH images showed the expected layered structure of transcriptionally distinct neuronal populations in the cerebral cortex (Figure 3B).

[0105] Next, to investigate whether the thick tissue 3D-MERFISH approach could image tissues thicker than 100 micrometers, 156 genes were measured in a 200-micrometer thick anterior hypothalamic section of a mouse. The RNA copy numbers of individual genes, measured per cell in each z-plane at different tissue depths, showed excellent correlation with each other, with only a slight decrease in RNA copy number across the entire tissue depth (Figures 11A, 11B). From single-cell expression profiles derived from MERFISH, 21 excitatory neuronal clusters, 26 inhibitory neuronal clusters, and 7 non-neuronal subclasses were identified in this region (Figures 3C, 3D; Figure 12A). Most transcriptionally distinct neuronal clusters showed a clear and localized spatial distribution, some of which were primarily located in a single hypothalamic nucleus; for example, cluster I1 was located in the bed nucleus of the stria terminalis (BNST), cluster I5 in the suprachiasmatic nucleus (SCH), and cluster E20 in the synodic nucleus (RE) of the thalamus (Figure 3D; Figure 12B).

[0106] It has been previously shown that interneurons in the mouse and human cerebral cortex (Lamp5, Vip, Sst, and Pvalb-positive inhibitory neurons) tend to form juxtaposition structures with other cells of the same type, as reflected by the observation that the nearest-neighbor distance between interneurons of the same type is significantly smaller than the distance between inhibitory neurons of this type and other cell types, despite the fact that the density of interneurons in the cerebral cortex is significantly lower than that of excitatory neurons. This was confirmed by analysis of thick tissue MERFISH data (Figure 3E, left) as well as analysis of whole-brain MERFISH data covering multiple cerebral cortical regions in the mouse brain (Figure 3E, right). It is hypothesized that such juxtaposition structures are mediated by electrical gap junctions, which are important for the synchronization of firing patterns. We investigated whether such juxtaposed pairs of inhibitory neurons are also formed in the hypothalamus using the same nearest-neighbor analysis. Interestingly, unlike inhibitory neurons in the cerebral cortex, inhibitory neurons in the anterior hypothalamus did not show preferential juxtaposition with inhibitory neurons of the same type (Figure 3F).

[0107] In summary, this embodiment demonstrates a method enabling high-performance 3D MERFISH imaging of thick tissue samples. This method is useful for many important applications. Firstly, as shown here, imaging thick tissue allows for capturing the total volume of almost all cells, thus increasing the accuracy of transcriptome profiling. Secondly, confocal imaging eliminates out-of-focus fluorescence background, potentially allowing for imaging of more genes, or the same number of genes, in fewer imaging rounds. Thirdly, confocal imaging can be used to facilitate cell segmentation, especially in areas with high cell density. Fourthly, thick tissue MERFISH imaging should facilitate the integrated measurement of gene expression profiles and morphology of individual cells and the investigation of the relationship between these two characteristics. Finally, this approach also facilitates the combination, for example, the measurement of neuronal activity by calcium imaging with the measurement of transcriptome profiles of individual cells, enabling the elucidation of the functional roles of molecularly defined cell types.

[0108] Figure 1. Deep learning improves the performance of confocal MERFISH imaging. (Figure 1A) Single-bit 242-gene high-pass filtered MERFISH confocal image of a brain tissue section taken with an exposure time of 0.1 seconds (left), and a magnified view of a single cell indicated by a white frame in the left image for further examination (right). (Figure 1B) Correlation between the copy number of individual genes detected per field of view (FOV) using a frame rate of 0.1 seconds and the copy number obtained using a frame rate of 1 second. The median ratio of copy numbers and the Pearson correlation coefficient r are shown. (Figure 1C) The same image as (Figure 1A), but with improved signal-to-noise ratio (SNR) using a deep learning algorithm. (Figure 1D) The same image as (Figure 1B), but with improved SNR of the 0.1-second image using deep learning.

[0109] Figure 2. 3D-MERFISH imaging of thick brain tissue sections. (Figure 2A) 3D images of DAPI and total polyA mRNA from a single FOV in a 100-micrometer-thick mouse brain tissue slice (upper panel). (Figure 2A) 3D images of DAPI and total polyA mRNA from a single FOV in a 100-micrometer-thick mouse brain tissue slice (upper panel), and a single z-plane at a tissue depth of 50 micrometers marked by a box in the upper image (lower panel). (Figure 2B) Maximum projection of 10 consecutive 1-micrometer z-planes of high-pass filtered MERFISH bit images obtained for the cells marked by the box in the lower panel of Figure 2A. (Figure 2C) RNA molecules identified in the same region as (Figure 2B), with the RNA molecules shaded by genetic identity. (Figure 2D) RNA copy number per individual gene per unit area (100 2 micrometers 2 ) vs. FPKM obtained from bulk RNA-seq. r represents the Pearson correlation coefficient. (Figure 2E) Pearson correlation between the RNA copy number per individual gene per z-plane detected by MERFISH and FPKM from bulk RNA-seq at different tissue depths. (Figure 2F) Number of detected RNA molecules per FOV at different tissue depths. (Figure 2G) RNA copy number per individual gene per unit area (100 2 micrometers 2 ) vs. RNA copy number detected by MERFISH measurement of 10-μm-thick tissue by epifluorescence setup. r represents the Pearson correlation coefficient.

[0110] Figure 3. Spatial organization of cell types in the mouse cerebral cortex and hypothalamus by 3D thick tissue MERFISH. (Figure 3A) UMAP visualization of cell subclasses identified in a 100-micrometer thick section of mouse cerebral cortex. Cells are shaded according to subclass identification. (Figure 3B) 3D spatial map of subclasses of excitatory neurons (left), inhibitory neurons (center), and non-neuronal cells (right) identified in a 100-micrometer thick section of mouse cerebral cortex. (Figure 3C) UMAP visualization of major cell types identified in a 200-micrometer thick section of the anterior mouse hypothalamus. Cells are shaded according to cell type identification. (Figure 3D) 3D spatial map of excitatory neurons (left), inhibitory neurons (center), and non-neuronal cells (right) identified in a 200-micrometer thick section of mouse hypothalamus. (Figure 3E) Distribution of nearest nearest distances from cells of individual inhibitory neuron subclasses to cells of allogeneic subclasses ("distance to allogeneic") or heterogeneic subclasses ("distance to heterogeneic") in the mouse cerebral cortex, as measured by MERFISH of the mouse cerebral cortex. (Figure 3F) Distribution of nearest nearest distances in the anterior mouse hypothalamus as described in (Figure 3E). *FDR < 0.01 in (Figure 3E) and (Figure 3F) was identified by the Wilcoxon rank-sum one-sided test and adjusted for FDR by the BH method. Only inhibitory neuron clusters with at least 20 "self-self" interaction pairs were examined and plotted in (Figure 3E-3F).

[0111] Figure 4. Comparison of epifluorescence and confocal MERFISH images in a thick tissue sample. Images of the nucleus, total poly(A) mRNA, and two MERFISH bits were obtained using epifluorescence microscopy and confocal microscopy, respectively. Both confocal and epifluorescence images were acquired with an exposure time of 1 second. The MERFISH bit images were filtered with a high-pass filter to remove cellular background.

[0112] Figure 5. MERFISH images of RNA molecules at different tissue depths in brain tissue sections of 100 micrometers and 200 micrometers thickness. (Figure 5A) Number of RNA molecules detected per FOV at tissue depths of 10 micrometers and 90 micrometers in MERFISH measurements of 242 genes in 100 micrometer thick sections of mouse cerebral cortex. (Figure 5B) Logarithmic distribution of the number of integrated photons of individual RNA molecules at tissue depths of 10 micrometers and 90 micrometers identified in (Figure 5A). (Figure 5C) Number of RNA molecules detected per FOV at tissue depths of 10 micrometers and 190 micrometers in MERFISH measurements of 156 genes in 200 micrometer thick sections of mouse hypothalamus. (Figure 5D) Logarithmic distribution of the number of integrated photons of individual RNA molecules at tissue depths of 10 micrometers and 190 micrometers identified in (Figure 5C).

[0113] Figure 6. Optimization of MERFISH encoding and readout probe labeling conditions. (Figure 6A) Exemplary bit-1 high-pass filtered MERFISH images of MERFISH measurements of 242 genes in 100-micrometer thick sections of mouse cerebral cortex stained with different concentrations of encoding probes. Concentration values ​​refer to the concentration of individual encoding probes. (Figure 6B) Distribution of integrated photon counts for individual RNA molecules identified at different encoding probe concentrations. The signal from individual RNA molecules increased with the encoding probe concentration and reached saturation at 1.0 nM per probe. Therefore, an encoding probe concentration of 1 nM was used to stain thick tissue samples. (Figure 6C) 100-micrometer thick mouse brain slices were stained with MERFISH encoding probes for 242 genes, followed by sequential hybridization with readout probes corresponding to bits 1, 2, 3, and 4 of the barcode, using different readout probe concentrations for each bit. The signal increased with the readout probe concentration, but the background also increased when the probe concentration exceeded 5 nM. Therefore, a readout probe concentration of 5 nM was used for imaging thick tissues. In addition to the probe concentration, the incubation time of the readout probe was also optimized. (Figure 6D) Number of RNA molecules per FOV per z-plane and normalized intensity of individual molecules at different tissue depths. The encoding probe concentration was 1 nM per encoding probe, the readout probe concentration was 5 nM, and the incubation time of the readout probe was 25 minutes.

[0114] Figure 7. Displacement of RNA molecules between different imaging rounds reduces detection accuracy and efficiency. (Figure 7A) Comparison of RNA copy numbers of individual genes per FOV per z-plane detected by MERFISH measurement of 242 genes in 100-micrometer thick mouse cerebral cortex tissue sections and FPKM values ​​obtained by bulk RNA-seq. Pearson correlation coefficient (r) is shown. (Figure 7B) Pearson correlation between RNA copy numbers of individual genes per FOV per z-plane detected at different tissue depths in 100-micrometer thick sections and FPKM measured by bulk RNA-seq. (Figure 7C) Total RNA copy numbers detected per FOV per z-plane at different tissue depths. (Figure 7D) Exemplary images of gel-embedded fiducial beads acquired in two rounds of imaging. Buffer exchange was performed between imaging rounds, mimicking the MERFISH protocol. Due to differences in gel expansion between rounds, the position of the beads changed in x, y, and z directions from round to round. Circles indicate identical beads identified across two imaging rounds, while arrows indicate beads captured in only one imaging round.

[0115] Figure 8. Quantification of gel expansion effect with MERFISH buffer. (Figure 8A) Quantification of gel expansion rate with various buffers used in the MERFISH protocol. The initial gel size was the same as the coverslip, and the expansion rate after buffer exchange was identified as the ratio of the gel size after buffer exchange to the coverslip size. (Figure 8B) In each round of MERFISH imaging, the sample was subjected to processing with a readout probe in a wash buffer (either 10% ethylene carbonate EC or 10% formamide) and hybridized for 15 minutes. The sample was then washed with the wash buffer to remove excess readout probe, followed by processing with an imaging buffer (either glucose-based or bacterial protocatechuate 3,4-dioxygenase rPCO-based imaging buffer). After the imaging process, the sample was treated with Tris(2-carboxyethyl)phosphine (TCEP) cleavage buffer to remove the fluorescence signal, and finally washed with 2× saline-sodium citrate (SSC) solution to remove the cleavage buffer. The gel expansion rates of various buffers used in MERFISH imaging were quantified and are shown here. Reagents marked with * were selected for final use in the 3D MERFISH experiment. The dashed line highlights the expansion rate of 2×SSC, which is the base for all other buffers; i.e., all other buffers contain 2×SSC. (Figure 8C) XZ projection image of fiducial beads embedded in the gel after buffer changes at specified times. Significant distortion occurs in the gel with a wash buffer containing 15% EC in 2×SSC, but recovers after 15 minutes of treatment with 2×SSC.

[0116] Figure 9. 3D MERFISH imaging of 242 genes in 100-micrometer thick sections of mouse cerebral cortex. (Figure 9A) Exemplary images of decoded RNA molecules at different tissue depths. Each image shows the decoded barcode in a z-range of 10 micrometers thickness, as indicated. The lower panel is a magnified and more detailed examination of the area indicated by the white frame in the upper panel. Identified RNA molecules are shaded by their genetic identity. (Figure 9B) Images of DAPI (left) and poly(A) mRNA (center) in an example of a field of view (FOV) used for cell segmentation. Cell boundary segmentation identified using a deep learning-based segmentation algorithm (Cellpose 2.0) is shown in the right panel. (Figure 9C) RNA copy number of individual genes per cell detected in a 100-micrometer thick tissue section versus individual RNA copy number in a z-range of 10 micrometer thickness in the same sample. A 100-micrometer thick section was evenly divided into 10 z-ranges, and the latter was identified.

[0117] Figure 10. Cell types identified in 100-micrometer thick sections of mouse cerebral cortex. UMAP visualization of excitatory (left) and inhibitory (right) neuronal clusters identified in mouse cerebral cortex, color-coded by cluster identity.

[0118] Figure 11. 3D MERFISH imaging of 156 genes in a 200-micrometer thick section of mouse hypothalamus. (Figure 11A) Median RNA copy number per cell along tissue depth in a 200-micrometer thick section of mouse hypothalamus. The first and last 10 micrometers were excluded from analysis due to incomplete cell coverage. (Figure 11B) Pearson correlation coefficient of RNA copy number of individual genes along tissue depth to RNA copy number in the first 1 micrometer of the 200-micrometer thick section.

[0119] Figure 12. Cell types and their spatial configuration identified in 200-micrometer thick sections of the mouse hypothalamus. (Figure 12A) UMAP visualization of excitatory and inhibitory neuronal clusters identified in 200-micrometer thick sections of the anterior mouse hypothalamus, color-coded by cluster identity. (Figure 12B) 2D spatial visualization of individual excitatory and inhibitory neuronal clusters. The hypothalamic nucleus to which each cluster is localized and the top 1 or 2 notable genes for each cluster are listed for each cluster. Three clusters, specifically E20, I1, and I5, are shown by dashed lines along with the corresponding nuclei to which they are localized.

[0120] The various methods and materials used in the above examples are shown below.

[0121] Animals. Adult C57BL / 6J male mice aged 7-9 weeks were used in this study. The mice were maintained in a 12-hour light / 12-hour dark cycle (12:00 noon to 12:00 noon), at a temperature of 22±1°C and humidity of 30-70%, with free access to food and water. Animal care and experiments were conducted in accordance with NIH guidelines and approved by the Harvard University Animal Care Committee (IACUC).

[0122] Tissue preparation for 3D MERFISH. 7-9 week old mice were deeply anesthetized with isoflurane. Transcardiac perfusion with phosphate-buffered saline (PBS) followed by perfusion with 4% paraformaldehyde (PFA) solution. Next, brain tissue was carefully dissected, post-fixed in 4% PFA solution, and treated overnight at 4°C. Afterward, the brain tissue was thoroughly washed with PBS. Next, the brain tissue was embedded in 4% low-melting-point agarose (Thermo Fisher Scientific, 16520-050) to prepare sections 100 or 200 micrometers thick. Sections were obtained using a vibratome (Leica). Finally, these sections were collected in 1×PBS and stored in 70% ethanol. The samples were stored at 4°C, and the sections were allowed to stand overnight and then permeabilized with 70% ethanol.

[0123] The samples were removed from 70% ethanol, washed three times with 2× saline sodium citrate (SSC), and then equilibrated at 47°C for 30 minutes in encoding probe washing buffer (30% formamide in 2× SSC). The washing buffer was aspirated from the samples and carefully transferred to a 2 mL DNA low-binding centrifuge tube containing 50 microliters of encoding probe mixture. The encoding probe mixture contained approximately 1 nM of each encoding probe, 1 micromol of poly(A) anchor probe (IDT), 0.1% wt / v yeast tRNA (15401-011, Life Technologies), and 10% v / v dextran sulfate (D8906, Sigma) in the encoding probe washing buffer. The samples were incubated at 37°C for 24–48 hours. The poly-A anchor probe sequence ( / 5Acryd / TTGAGTGGATGGAGTGTAATT+TT+TT+TT+TT+TT+TT+TT+TT+TT+T(SEQ ID NO: 1)) contains a mixture of DNA and LNA nucleotides, where T+ is locked nucleic acid and / 5Acryd / is a 5' acridite modification. The poly-A anchor allows the polyadenylated mRNA to be immobilized on the polyacrylamide gel during the hydrogel embedding process described later. After hybridization, the sample was washed three times with encoding probe washing buffer at 47°C for 20 minutes each to wash away excess probe, and then washed three times with 2×SSC at room temperature.

[0124] Next, the sample was embedded in a hydrogel to clear the tissue background and remove off-target probe binding. First, the sample was incubated at room temperature for 30 minutes in a monomer solution containing 2M NaCl, 4% (vol / vol) 19:1 acrylamide / bisacrylamide, 60 mM Tris-HCl pH 8, and 0.2% (vol / vol) TEMED. Next, 100 microliters of ice-cold monomer solution containing 0.2% (vol / vol) 488 nm fiducial beads (Invitrogen) were placed on a 40 mm silane-coated coverslip. The silane modification procedure ensured that the hydrogel covalently bonded to the coverslip surface as described above. Then, the tissue was gently transferred to the coverslip using a brush, flattened, and excess monomer solution was carefully aspirated. 100 microliters of ice-cold monomer solution containing 0.1% (wt / vol) ammonium persulfate was dropped onto a hydrophobic glass plate treated with GelSlick (Lonza). Next, the coverslip with the flattened sample was inverted onto the droplet to form a uniform layer of monomer solution. A 50g weight was placed on top of the coverslip to confirm that the tissue remained flat and completely adhered to the coverslip. The sample was left at room temperature for at least one hour to fully polymerize, ensuring that the movable sample slice was completely attached to the coverslip.

[0125] Next, the coverslip on which the polymerized sample was placed was peeled off the glass plate using a thin razor blade. Then, the sample was incubated at 37°C for 24 hours in a digestion buffer containing 2% (wt / vol) sodium dodecyl sulfate (SDS) (ThermoFisher), 0.5% (vol / vol) Triton X-100 (ThermoFisher), and 1% (vol / vol) proteinase K (New England Biolabs) in 2× SSC. After digestion, the sample was washed at room temperature for 1 hour in 2× SSC buffer supplemented with 0.2% (vol / vol) proteinase K. The buffer was changed every 30 minutes to ensure thorough washing.

[0126] MERFISH encoding and readout probes. This study used two sets of MERFISH encoding probes: a previously designed set targeting 242 genes in the mouse primary motor cortex, and another set targeting 156 genes in the preoptic area of ​​the hypothalamus of the mouse brain. In previous studies of the preoptic area, 135 of the 156 genes were imaged using combinatorial MERFISH imaging, and the remaining 20 genes were measured in a series of multicolor smFISH rounds. In this study, these 20 genes were incorporated into combinatorial imaging rounds. Fluorescent readout probes conjugated via disulfide linkage to either Cy5, Cy3B, or Alexa488 dye molecules were purchased from Bio-Synthesis Inc.

[0127] 3D MERFISH Imaging Platform. In this study, two 3D MERFISH imaging platforms were constructed. One (Setup 1) consisted of a Nikon Ti-U microscope body equipped with either a Nikon 40× 1.15 NA immersion lens (Nikon, MRD77410) or a 60× 1.2 NA immersion lens (Olympus, UPLSAPO60XW) and a spinning disk confocal unit (Andor Dragonfly, ACC-CR-DFLY-202-40). Solid-state lasers with wavelengths of 647nm (MBP Communications, 2RU-VFL-P-1500-647-B1R), 561nm (MBP Communications, 2RUVFL-P-1000-560-B1R), 488nm (Coherent, Genesis MX488-1000STM), and 405nm (Coherent, Obis 405-200C) were used for illumination. The 647nm, 561nm, and 488nm outputs were controlled by acousto-optic tunable filters (Crystal Technologies, AODS 20160-8 and PCAOM Vis), while the 405nm output was controlled by direct modulation. The coaxially aligned beams were coupled to the input fiber of a beam homogenizer (Andor, Borealis BCU-120) to provide uniform illumination to a rotating disk. Fluorescence emission was separated in the imaging path using a pentabandpass dichroic (Andor, CR-DFLY-DMPN-06I) and an emission filter (Andor, TR-DFLY-P45568-600). Sample position was controlled by an electric XY stage (Prior, Proscan H117E1N5 / F), and the z-scan was controlled by a piezo objective nanopositioner (Queensgate, OP400 or Mad City Labs, F200S). Prior to image scanning, initial focus was acquired using a custom autofocus system that monitored the position of an IR laser (Thorlabs, LP980-SF15) reflected from the coverslip surface with a CMOS camera (Thorlabs, DCC1545M).The control signals for the laser and piezo were all generated by a DAQ card (National Instruments, PCIe-6353) and synchronized with the imaging start signal of the sCMOS camera (Hamamatsu, Orca-Flash4.0).

[0128] A similar imaging platform (Setup 2) was constructed based on an Olympus IX71 microscope body, a spinning disk confocal unit (Andor, CSU W1), a beam homogenizer (Andor, Borealis BCU 100), a piezo objective lens nanopositioner (Mad City Labs, Nano-F200S), and an XY stage (Marzhauser, Scan IM 112×74). Illumination was supplied by solid-state lasers of 647nm (MBP Communications, 2RU-VFL-P-2000-647-B1R), 561nm (MPB Communications, 2RU-VFL-P-1000-560-B1R), 488nm (MPB, 2RU-VFL-P-500-488-B1R), and 405nm (Coherent, Cube 405), and controlled by a mechanical shutter (Uniblitz, LS6T2).

[0129] Table 1 shows details about which imaging platform was used to acquire the specific data.

[0130] Fluid system and sample chamber. MERFISH samples were imaged on a 40 mm circular coverslip (Bioptech) and mounted in a flow chamber (Bioptech, FCS2) using a 0.5 mm gasket (Bioptech, 1907-1422-500). The fluid system included a peristaltic pump (Gilson, Minipuls 3), four 8-way valves (Hamilton, MVP 36798 with 8-5 distribution valves) assembled to supply up to 24 readout bit solutions, and four additional buffers (2×SSC, wash, cut, and image). Image acquisition and fluid control were fully automated using custom software.

[0131] 3D MERFISH imaging. To prepare the sample for imaging, it was first stained at a concentration of 25 nM per probe using a readout hybridization mixture containing a readout probe conjugated with a complementary probe to a poly(A) anchored probe and then conjugated with the dye Alexa488 via a disulfide bond. The readout hybridization mixture consisted of the readout probe added to a readout probe wash buffer containing 2×SSC, 10% v / v ethylene carbonate (E26258, Sigma), and 0.1% v / v Triton X-100. The sample was incubated in this buffer mixture at room temperature for 30 minutes, then washed for 30 minutes with a readout probe wash buffer containing 1 microgram / ml of DAPI to stain the nuclei in the sample. Next, the sample was washed with 2×SSC for 15 minutes and loaded into the flow chamber. An imaging buffer containing 5 mM 3,4-dihydroxybenzoic acid (P5630, Sigma), 2 mM Trolox (238813, Sigma), 50 micromoles of Trolox skinone, 1:500 recombinant protocatechlate 3,4-dioxygenase (rPCO, OYC Americas), 1:500 mouse RNase inhibitor, and 5 mM NaOH (adjusted to pH 7.0) in 2× SSCs was introduced into the chamber. The imaging buffer was allowed to penetrate deep into the tissue for at least 15 minutes. Next, the sample was imaged with illumination light at a wavelength of 405 nm using a low-magnification 10× air objective lens to create a tiled image of the sample. Then, the region of interest (ROI) of each slice was positioned using this image, and a grid of field of view (FOV) positions covering the ROI was created. After identifying these positions, each FOV position was imaged using a high numerical aperture objective lens. In the first round of imaging, images were acquired using 488 nm and 405 nm channels, imaging 488 nm fiducial beads, total poly(A) mRNA stained with poly(A)-anchor probe, and nuclei stained with DAPI. These two channels were later used for cell segmentation.To compensate for slight differences in stage position, one image of the fiducial beads on the coverslip surface was taken for each imaging round as a spatial reference. A 1-micrometer thick z-stack was collected for all channels in each field of view.

[0132] After the initial imaging round, to remove the fluorescent dye, 2 mL of cleavage buffer containing 2×SSC and 50 mM Tris(2-carboxyethyl)phosphine (TCEP; 646547, Sigma) was flushed through the sample and incubated in a flow chamber for 15 minutes to cleave the disulfide bond linking the dye and the readout probe. Next, the sample was washed with 4 mL of 2×SSC to completely remove any remaining cleavage buffer.

[0133] To perform imaging for subsequent rounds, 3 mL of a read-probe mixture containing the appropriate read-probe was flowed and hybridized for 25 minutes (15 minutes flow, followed by 10 minutes rest). Two read-probes were hybridized in each round, one labeled with Cy5 and the other with Cy3b, and a read-probe mixture containing 5 nM of the appropriate read-probe was used in each round. Next, the sample was washed with 2 mL of read-probe washing buffer, followed by two washes of 2 mL of imaging buffer for 15 minutes each (5 minutes flow, followed by 10 minutes rest). In each round, all FOVs were imaged on the surface with the two read-probes and a 488 nm fiducial bead in 650 nm and 560 nm channels. The hybridization, washing, imaging, and cutting steps were repeated for all rounds to complete MERFISH imaging.

[0134] 3D MERFISH imaging analysis. MERFISH image analysis was performed using a customized version of MERlin, a Python-based MERFISH analysis pipeline.

[0135] Firstly, tiling artifacts were minimized with approximately 10% overlap of adjacent FOVs by stitching the nuclear channels using BigStitcher. Next, the transformation was applied to each FOV to adjust the relative position of each FOV. This allowed for seamless transitions between adjacent FOVs in the final dataset. Secondly, the content-aware deep learning-based image restoration algorithm CSBdeep was used to improve the quality of MERFISH images captured with short exposure times. Specifically, a model was trained individually for each MERFISH bit color channel (560 and 650). To achieve this, for each color channel, 50 image pairs were randomly selected, each containing low and high signal-to-noise ratio (SNR) images. From these imaging pairs, overlapping 128×128 image patches were generated, and these patches were further divided into a training set (80%) and an evaluation set (20%). Using the training dataset, the neural network was trained using CSBdeep with the following parameters: kernel size = 3, training batch size = 10, and training steps per epoch = 50. The mean of the absolute values ​​of the differences was used as the training loss function. Next, the model performance was evaluated using a reserved 20% of the dataset to identify the optimal model parameters. Thirdly, the images taken during each imaging round were aligned based on the fiducial bead image to account for the XY drift of the stage position relative to the imaging in the first round. Fourthly, a high-pass filter was applied to the improved images of each FOV to remove the cellular background. The filtered images were then deconvolved using 20 rounds of Lucy-Richardson deconvolution to sharpen RNA spots, and a low-pass filter was applied to account for slight shifts in the apparent centroid of RNA between imaging rounds. Individual RNA molecules were identified by a pixel-based decoding algorithm.In short, a barcode was assigned to each pixel individually, then adjacent pixels with the same barcode were aggregated into putative RNA molecules, and the list of putative RNA molecules was filtered to enrich the correctly identified transcripts. More specifically, to assign each pixel to one of the barcodes, the intensity vector measured for each pixel was compared to the vector corresponding to a valid barcode. To aid in comparison, the intensity of each image was normalized within the bit by the median intensity across all FOVs in the bit to eliminate hybridization and intensity variations between color channels. After intensity normalization, the intensity variations between pixels were further normalized by dividing the intensity vector of each pixel by its L2 norm. Similarly, each of the pre-designed barcodes was normalized by its L2 norm. To assign a barcode to each pixel, the normalized barcode vector closest to the normalized intensity vector of that pixel was identified. Pixels whose distance from any valid barcode was greater than 0.65 in the first step were excluded, and any pixels with an intensity less than 10 were ignored as they could be out-of-target binding probes or noise-induced artifacts amplified by the deep learning algorithm.

[0136] During model training, two sets of confocal images from the same location were used as input. One set was taken with a short exposure time (100 ms), and the other with a long exposure time (1 second). The short-exposure image was expected to contain a lot of noise. The long-exposure image is of high quality and therefore has a low noise level. These two images were taken from the same location and should contain the same information, although the noise levels are different. Next, the deep learning model was trained to learn how to transform the low-quality, noisy short-exposure image into a high-quality long-exposure image. The goal of the deep learning model was to improve the image quality. The deep learning model was a convolutional neural network.

[0137] After independently assigning a barcode to each pixel, adjacent pixels with the same barcode were aggregated into a single putative RNA molecule. The list of putative RNA molecules was then filtered to enrich correctly identified transcripts, resulting in an overall barcode misidentification rate of 5%. Putative RNAs containing only a single pixel were also removed, as they tend to create background spurious barcodes generated by random fluorescence fluctuations, leading to a significantly higher misidentification rate than those containing two or more pixels. Finally, because 3D MERFISH imaging involves sampling at 1-micrometer intervals, it is possible to capture the same molecule in multiple z-sections. To avoid counting duplicate molecules, identical molecules present in adjacent z-sections were removed before assigning barcodes to individual cells.

[0138] 3D cell segmentation. Cell segmentation was performed using the deep learning-based cell segmentation algorithm, Cellpose 2.0, with co-staining of DAPI and total mRNA. The segmentation model was refined using a user-in-the-loop approach in Cellpose 2.0, starting with the "CP" model and using randomly selected z-slice containing DAPI and poly(A) mRNA channels. After manual training, the segmentation model was applied to a 3D Z-stack for each FOV, and segmentation masks were generated in 3D using the 3D mode of Cellpose 2.0. Cell boundaries were extracted for each cell and exported as polygons.

[0139] Due to approximately 10% overlap in field of view (FOV), a single cell may be captured in two adjacent FOVs, resulting in cell duplication. To address this issue, a method was developed to remove duplicate cells. First, for each cell, the 10 nearest neighbor cells were identified, and their overlapping volume was calculated for each neighboring cell. Next, the overlapping volume was divided by the minimum volume of two adjacent cells to determine the overlap rate between the two cells. The overlap rate ranged from 0% (for cells that do not overlap with any other cells) to 100% (for cells that completely overlap with another cell). If their overlap rate was greater than 40%, the two cells were considered duplicates, and in this case, the cell with the smaller volume was removed. This method allowed for the identification and removal of duplicate cells appearing in multiple adjacent FOVs. By implementing this method, the final dataset contained only unique cells, avoiding any inaccuracies in downstream analyses. Detected RNA molecules were assigned intracellularly if their molecular location was within the cell boundary, resulting in a cell-specific gene matrix.

[0140] Unsupervised clustering analysis of 3D MERFISH data. After obtaining the gene matrix for each cell as described above, the matrix was preprocessed using the following steps. First, cells potentially containing artifacts due to segmentation errors were removed. Specifically, cells with small volume (<300 micrometers) were removed. 2 Cells with low RNA counts (<30), or those captured fewer than 5 times or more than 40 times on a 1-micrometer z-section were excluded. These criteria were chosen to exclude low-quality or insufficiently informative cells. Next, approximately 10% of cells suspected to be doublets, identified using doubletFinder, were removed.

[0141] Following the preprocessing steps described above, single-cell data were analyzed using Seurat as described below. The gene vector was normalized for each cell by dividing each cell by the sum of its total RNA counts, and then multiplying the resulting value by a constant of 10,000 to ensure that all cells contained the same total RNA count. After this normalization, the cell-specific gene matrix was logarithmically transformed. The normalized single-cell expression profiles were z-scored, followed by dimensionality reduction by principal component analysis. Using the first 30 principal components, graph-based Leuven community detection was performed in the 30-principal component space with nearest neighbor size k=15 and resolution r=0.8 to identify clusters.

[0142] From the first round of clustering, excitatory, inhibitory, and non-neuronal cell types were identified based on the expression of canonical marker genes. To further refine the cell classification, the dimensionality reduction and clustering procedures were repeated separately for excitatory and inhibitory neurons as described above. For neuronal cluster identification, the following parameters k=15 and r=5 were used for inhibitory neurons, and k=15 and r=3 were used for excitatory neurons.

[0143] [Table 1]

[0144] While several embodiments of the Disclosure have been described and illustrated herein, those skilled in the art can readily imagine a variety of other means and / or structures to perform the functions described herein and / or to obtain one or more of the results and / or benefits, and each of such variations and / or modifications shall be considered within the scope of the Disclosure. More generally, those skilled in the art will readily understand that all parameters, dimensions, materials and configurations described herein are illustrative, and that actual parameters, dimensions, materials and / or configurations will depend on the particular use or use in which the teachings of the Disclosure are used. Those skilled in the art will recognize or can verify many equivalents to the specific embodiments of the Disclosure described herein without going beyond routine experimentation. Therefore, it should be understood that the embodiments described herein are presented only as examples, and within the scope of the appended claims and their equivalents, the Disclosure may be performed in ways other than those specifically described and claimed. The Disclosure covers the individual features, systems, articles, materials, kits and / or methods described herein. In addition, any combination of two or more such features, systems, articles, materials, kits, and / or methods is included within the scope of this disclosure, provided that such features, systems, articles, materials, kits, and / or methods are not inconsistent with each other.

[0145] In the event that this Specified Version and any document incorporated by reference contain conflicting and / or contradictory disclosures, this Specified Version shall prevail. If two or more documents incorporated by reference contain conflicting and / or contradictory disclosures with respect to each other, the document with the later effective date shall prevail.

[0146] All definitions defined and used herein should be understood to take precedence over dictionary definitions, definitions incorporated by reference in documents, and / or the ordinary meanings of the defined terms.

[0147] As used herein and in the claims, the indefinite articles "a" and "an" should be understood to mean "at least one" unless the opposite is explicitly stated.

[0148] As used herein and in the claims, the phrase “and / or” should be understood to mean “one or both” of the elements thus connected, that is, elements that are sometimes connected and sometimes disjunctive. Multiple elements listed with “and / or” should be interpreted similarly, meaning “one or more” of the elements thus connected. Other elements may exist, as appropriate, in addition to those specifically named by the “and / or” clause, whether or not they are related to the elements specifically named. Thus, as a non-restrictive example, a reference to “A and / or B” when used in combination with open-ended language such as “including” may refer to A only in one embodiment (including elements other than B as appropriate); B only in another embodiment (including elements other than A as appropriate); and both A and B in yet another embodiment (including other elements as appropriate), and so on.

[0149] Where used herein and in the claims, “or” should be understood to have the same meaning as “and / or” as defined above. For example, when separating items in a list, “or” or “and / or” should be interpreted as inclusive, that is, including at least one of the number of elements or the list, but including more than one, and, where appropriate, including further unlisted items. Terms that express a clear opposite meaning, such as “one of” or “exactly one of” or, where used in the claims, “consisting of,” refer to including exactly one element of the number of elements or the list. In general, where used herein, the term “or” should be interpreted only as indicating an exclusive choice (i.e., “one or the other, but not both”) when preceded by an exclusive term such as “either,” “one of,” “one of,” or “exactly one of.”

[0150] As used herein and in the claims, the phrase “at least one” referring to a list of one or more elements should be understood to mean at least one element selected from any one or more elements in the list of elements, but not necessarily including at least one of each and all elements specifically enumerated in the list of elements, nor excluding any combination of elements in the list of elements. This definition also allows for the presence of elements other than those specifically listed in the list of elements to which the phrase “at least one” refers, whether or not they are related to the specifically listed elements. Therefore, as a non-restrictive example, “at least one of A and B” (or synonymously “at least one of A or B” or synonymously “at least one of A and / or B”) could mean, in one embodiment, at least one, optionally more than one, of A (and optionally other elements) in the absence of B; in another embodiment, at least one, optionally more than one, of B (and optionally other elements) in the absence of A; and in yet another embodiment, at least one, optionally more than one, of A, and at least one, optionally more than one, of B (and optionally other elements), and so on.

[0151] Where the term “approximately” is used in reference to a number in this specification, further embodiments of the present disclosure should be understood to include numbers that are not modified by the presence of the term “approximately.”

[0152] Furthermore, unless otherwise explicitly stated, in any method claimed herein that includes more than one step or act, the order of the steps or acts of the method is not necessarily limited to the order in which the steps or acts of the method are described.

[0153] In the claims, as in the specification above, all transitional phrases, such as “comprising,” “including,” “carrying,” “having,” “containing,” “involving,” “holding,” and “composed of,” should be understood to be open-ended, meaning they include without limitation. As stated in the U.S. Patent and Trademark Office's Manual of Patent Examination Procedure (Section 2111.03), only the transitional phrases “consisting of” and “consisting essentially of” are closed or semi-closed transitional phrases, respectively.

Claims

1. Acquiring images of the sample using a confocal microscope, and Identifying nucleic acids in a sample using MERFISH. A method that includes this.

2. The method according to claim 1, further comprising acquiring images at more than one focal plane.

3. The method according to claim 2, comprising acquiring images of a plurality of focal planes separated by at least 1 micrometer in the Z-axis direction.

4. The method according to any of the prior claims, wherein the confocal microscope is a spinning disk confocal microscope.

5. The method according to any of the prior claims, further comprising using machine learning to improve an image of a sample.

6. The method according to any of the prior claims, further comprising running a machine learning model trained on confocal images to receive an input image and output an improved image based at least on the signal-to-noise ratio.

7. The method according to any of the prior claims, further comprising running a machine learning model trained on confocal images to receive an input image and output an improved image based at least on the signal-to-background ratio.

8. The method according to any of the prior claims, further comprising running a machine learning model trained on confocal image characteristics to receive an input image and output an improved image based at least on the signal-to-noise ratio.

9. The method according to any of the prior claims, further comprising running a machine learning model trained on confocal image characteristics, receiving an input image, and outputting an improved image based on at least the signal-to-background ratio.

10. The method according to any of the preceding claims, wherein the sample has a thickness of at least 100 micrometers.

11. Acquiring images of the sample using a confocal microscope, and Using images to identify nucleic acids in a sample in three dimensions. A method that includes this.

12. The method according to claim 11, further comprising using machine learning to improve the image of the sample.

13. The method according to claim 11 or 12, wherein the confocal microscope is a spinning disk confocal microscope.

14. Exposing the sample to multiple nucleic acid probes, For each nucleic acid probe, the binding of the nucleic acid probe within the sample is identified by acquiring images of the sample using a confocal microscope. To generate codewords based on the binding of nucleic acid probes, and For at least a portion of the codewords, match the codewords with valid codewords, and if no match is found, apply error correction to the codewords to form valid codewords. A method that includes this.

15. The method according to claim 14, wherein the confocal microscope is a spinning disk confocal microscope.

16. Acquiring images of a sample using a confocal microscope, Improving images using machine learning, and Identifying nucleic acids in a sample using MERFISH. A method that includes this.

17. The method according to claim 16, wherein the confocal microscope is a spinning disk confocal microscope.

18. Acquiring images of a sample using a confocal microscope, Improving images using machine learning, and Identifying nucleic acids in a sample A method that includes this.

19. The method according to claim 18, wherein the confocal microscope is a spinning disk confocal microscope.

20. Acquiring images of the sample using a confocal microscope, and By exposing a sample to multiple nucleic acid probes and identifying the binding of multiple nucleic acid probes to the sample, nucleic acids within the sample can be identified. A method that includes this.

21. The method according to claim 20, wherein the confocal microscope is a spinning disk confocal microscope.

22. Acquiring images of a sample using a confocal microscope, Improving images using machine learning, and By exposing a sample to multiple nucleic acid probes and identifying the binding of multiple nucleic acid probes to the sample, nucleic acids within the sample can be identified. A method that includes this.

23. The method according to claim 22, wherein the confocal microscope is a spinning disk confocal microscope.

24. The process involves acquiring multiple images of a sample contained in a gel using a spinning disk confocal microscope, wherein the gel exhibits a linear expansion of 5% or less between image acquisitions, and Identifying nucleic acids in a sample using MERFISH. A method that includes this.

25. The method according to claim 24, wherein the gel exhibits linear expansion of 2% or less.

26. The process involves acquiring multiple images of a sample contained in a gel using a spinning disk confocal microscope, wherein the gel exhibits a linear expansion of 5% or less between image acquisitions. Identifying nucleic acids in the sample, and Improving sample images using machine learning A method that includes this.

27. The method according to claim 26, wherein the gel exhibits linear expansion of 2% or less.

Images