Patent Application
20230408514
The use of DNA to barcode physical compartments and tag intracellular and cell-surface molecules has enabled the use of sequencing to efficiently profile the molecular properties of thousands of cells simultaneously. While initially applied to measuring the abundances of RNA1,2 and identifying regions of accessible DNA3, recent developments in DNA-tagged antibodies have created new opportunities to use sequencing to measure the abundances of cell surface proteins4,5 and intracellular proteins6.
Sequencing DNA-tagged antibodies is particularly useful for profiling cells whose identity and function have long been determined by cell surface proteins (e.g. immune cells) and has several advantages over flow and mass cytometry. First, the number of cell surface proteins that can be measured by DNA-tagged antibodies is exponential to the number of bases in the tag. In theory, all cell surface proteins with available antibodies can be targeted and in practice, panels targeting hundreds of proteins are now commercially available4,7. This contrasts with cytometry where the number of proteins targeted is limited by the overlap in the emission spectrums of fluorophores (flow: 4-48) or the number of unique masses of metal isotopes that can be chelated by commercial polymers (CYTOF: ˜50)8,9. Second, sequencing-based proteomics can readily read out all antibody tagging sequences with one reaction instead of subsequent rounds of signal separation and detection, significantly reducing the time and sample input for profiling large panels and obviates the need for fixation. Third, additional molecules can be profiled within the same cell enabling multimodal profiling of cell surface proteins along with the immune repertoire, transcriptome4, and potentially the epigenome. Finally, sequencing is amenable to encoding orthogonal experimental information using additional DNA barcodes (either inline or distributed) creating opportunities for large-scale multiplexed screens that barcode cells using natural variation10, synthetic sequences11,12, or sgRNAs13,14.
In one aspect provided is an assay method comprising tagging cell surface molecules of cells with DNA-barcoded antibodies and using droplet-based single cell sequencing to determine protein expression profiles of the cells wherein at least 30% of droplets comprise multiple cells and the protein expression profiles for multiple cells simultaneously encapsulated in a single drops are resolved by the combinatorial index of barcodes.
In one aspect provided is an assay method comprising (a) providing a plurality of vessels, each vessel comprising i-a) a plurality of cells from a population, each cell comprising a plurality of cell surface proteins, and ii-a) a panel of staining constructs, wherein each staining construct comprises a handle-tagged antibody and a pool oligonucleotide, wherein each handle-tagged antibody comprises iii-a) an antibody specific for a cell surface protein in (i-a), and iv-a) a handle oligonucleotide attached to the antibody, wherein the handle oligonucleotide comprises a handle sequence that identifies the specificity of the antibody to which it is attached; and each pool oligonucleotide comprises at least the following nucleotide segments: v-a) a handle complement segment complementary to, and annealed to, the handle oligonucleotide, vi-a) a capture complement segment, vii-a) an antibody barcode complement segment having a sequence that identifies the binding specificity of the antibody in (iii-a) and thereby identifies the handle oligonucleotide in (iv-a), and viii-a) a pool barcode complement segment, wherein (vii-a) and (viii-a) are positioned between (v-a) and (vi-a), wherein in each vessel, the staining constructs in the vessel have the same pool barcode complement segments, wherein in at least some vessels at least one staining construct is to a cell surface protein in (i-a); (b) optionally combining the contents of all or some of said plurality of vessels, (c) loading individual stained cells or combinations of individual stained cells into compartments, wherein each stained cell comprises one or more staining constructs bound to a cell surface protein of the cell wherein at least some compartments comprise one or more stained cells and a plurality of droplet oligonucleotides wherein each droplet oligonucleotide comprises a droplet bar code and a capture segment wherein the droplet oligonucleotides in a compartment have the same droplet barcode and droplet oligonucleotides in different compartments have different barcodes wherein the capture segment is complementary to and anneals to the capture complement segment of the pool oligonucleotide; (d) producing sequence fragment structures corresponding to the capture constructs, each sequence fragment structure comprising a droplet barcode, a pool barcode and an antibody barcode whereby a plurality of sequence fragment structures are produced (e) sequencing at least some of the plurality of sequence fragment structures to determine the sequences of the droplet barcode, the pool barcode and the antibody barcode of individual sequence fragment structures; (f) determining from the sequencing in (e) distribution of cell surface proteins on individual cells. The pool barcode and antibody barcode are a compound barcode.
In an approach in step (c) at least some of the compartments have two or more cells loaded therein, and cell surface protein expression profiles of said two or more cells are determined. In some cases at least 30% of the compartments containing cells comprise two or more cells. In some cases the cells in the plurality of vessels in (a) comprise a cell population and a composition or expression of cell surface proteins in the population is determined. In some cases the compartments are droplets or wells. In some cases droplet oligonucleotides (capture oligonucleotides) are attached to beads.
In an aspect provided is a nucleic acid capture complex comprising a handle oligonucleotide, a pool oligonucleotide, and a droplet oligonucleotide. In an aspect provided is a kit comprising two or more of (i) a plurality of handle-tagged antibodies comprising different handle sequences and antibodies with different binding specificities, wherein there is a correlation between each handle sequence and each antibody specificity; (ii) a plurality of pool oligonucleotides with different handle complement sequences, wherein said handle complement sequences are complementary to and can anneal to the handle sequences in (i); and (iii) a plurality of droplet oligonucleotides configured to combine with pool oligonucleotides.
As used herein, “antibody” means an immunoglobulin molecule of any useful isotype (e.g., IgM, IgG, IgG1, IgG2, IgG3 and IgG4); chimeric, humanized and human antibodies, antibody fragments and engineered variants, including, without limitation Fab, Fab′, F(abe)2, F(ab1)2 scFv, dsFv, ds-scFv, dimers, single chain antibodies (scAb), minibodies (engineered antibody constructs comprised of the variable heavy (VH) and variable light (VL) chain domains of a native antibody fused to the hinge region and to the CH3 domain of the immunoglobulin molecule); nanobodies, diabodies (comprising two Fv domains connected by short peptide linkers), and multimers thereof; heteroconjugate antibodies (e.g., bispecific antibodies and bispecific antibody fragments), and other forms that specifically bind to a target polypeptide. “Antibodies” are a type of “affinity reagent” that also includes aptamers, affimers, knottins and the like.
As used herein, the term “monoclonal antibody” has its normal meaning in the art and is an antibody from a population of identical antibodies, including a clonal population produced by cells or a population produced by other means.
As used herein, the term “complementary” refers to Watson-Crick base pairing between nucleotides units of two single stranded nucleic acid molecules or two portions of the same nucleic acid molecule. Complementary sequences or segments can be “exactly complementary” (two nucleic acid segments with 100% complementarity, e.g., the sequence of one segment is the reverse complement of the sequence of the other segment) or “substantially complementary” (two nucleic acid segments with less than 100% complementarity and at least about 80%, at least about 85%, at least about 90%, or at least about 95% complementary). Percent complementarity refers to the percentage of bases of a first nucleic acid segment that can form base pairs with a second nucleic acid segment. Polynucleotides or segments with substantially complementary sequences can anneal to each other under assay conditions to form a double stranded segment. It will be appreciated that a first sequence that can anneal to a second sequence to generate a double-stranded molecule can be referred to as a sequence that is the complement of the second sequence, or, equivalently, the “reverse complement.”
As used herein, two nucleic acid segments that are complementary to each other, or have sequences complementary to each other, or have the relationship in which a first segment has a sequence that is “the complement of” a sequence of a second segment.
As used herein, the terms “anneal” and “hybridize” are used interchangeably to refer to two complementary single stranded nucleic acid segments that base-pair to form a double-stranded segment
As used herein, the term “construct” refers to two or more nucleic acid molecules that are associated by base pairing between a subsequence or segment of a first nucleic acid molecule and a complementary subsequence or segment of a second nucleic acid molecule. Reference to a “Construct” does not include a single, fully double stranded, polynucleotide.
As used herein the term “segment” used in reference to a polynucleotide refers to a defined portion or subsequence of the polynucleotide comprising a plurality of contiguous nucleotides. Typically a segment has 5 to 100 contiguous bases.
As used herein, the terms “oligonucleotide” and “oligo” are used interchangeably and, unless otherwise indicated or clear from context, refer to a single stranded nucleic acid less than 500 bases in length. In some cases, as will be apparent from context, a segment is referred to as an “oligonucleotide” sequence (e.g., “the capture complement is an oligonucleotide sequence contained in a Pool Oligonucleotide”).
As used herein, the terms “nucleic acid” and “polynucleotide” are used interchangeably and usually refer to a single or double-stranded DNA polymer. However, methods and compounds described herein may be carried out using oligonucleotides and Constructs that comprise RNA, DNA/RNA chimeras, and synthetic analogs of DNA or RNA containing non-naturally occurring nucleobase analogs, or analogs of (deoxy)ribose or phosphate or, in the case of DNA, contain uracil in place of thymidine, which are also referred to as nucleic acids or polynucleotides.
As used herein, the term “barcode” or “BC” refers to a short (typically less than 50 bases, often less than 30 bases) nucleic acid sequence that identifies a property of a polynucleotide. For example, in some cases polynucleotides with the same barcode have a common origin, e.g., are from the same vessel or compartment. In various places in this disclosure there is reference, for clarity, to a barcode sequence and a barcode sequence complement. It will be recognized that in a double-stranded polynucleotide the sequence in both strands is informative and can serve as a barcode.
As used herein, the term “vessel” refers to a container in which a solution containing cells, oligonucleotides, and/or constructs can be pooled (combined). Antibody binding and nucleic acid hybridization may occur in a vessel. The term “vessel” does not imply a particular structure or material. Examples of vessels include tubes, wells, and microfluidic chambers.
As used herein, the term “compartment” refers to a structure that can contain one or more cells and one or more nucleic acid Constructs. Examples of compartments include droplets, capsules, wells, microwells, microfluidic chambers, and other containers.
As used herein, “bead” may refer to (but is not limited to) beads of the type used in droplet-based single cell sequencing technologies (inDrop, Drop-seq, and 10X Genomics) which carry or are attached to polynucleotides. Bead technology is well known in the art. Wang et al., 2020, “Dissolvable Polyacrylamide Beads for High-Throughput Droplet DNA Barcoding” Advanced Science 7:8, and references cited therein; Klein et al. Cell 2015, 161, 1187; Macosko et al., Cell 2015, 161, 1202; Lan et al Nat. Biotechnol. 2017, 35, 640; Lareau et al. Nat. Biotechnol. 2019, 37, 916; Stoeckius et al. Nat. Methods 2017, 14, 865; Peterson et al. Nat. Biotechnol. 2017, 35, 936; Zheng et al., Nat. Commun. 2017, 8, 14049.
As used herein, a compartment is “occupied” if it contains at least one cell (i.e., is not empty).
Abbreviations: BC—bar code; CSP—cell surface protein; Ab—antibody; mAb—monoclonal antibody; HTA—Handle-Tagged antibody; HCL—high-concentration loading; UMI—unique molecular identifier.
A major limitation in sequencing-based single-cell proteomics4,7 is the high cost associated with profiling each cell, thus precluding its use across population cohorts or large-scale screens where millions of cells would need to be profiled. Like other single-cell sequencing assays, total cost per cell for proteomic sequencing is divided between cost associated with library construction and the cost for sequencing the library. Because the number of protein molecules per cell is 2-6 orders of magnitude higher than RNA15 and the use of targeting antibodies limits the number of features measured per cell, methods that use tagged antibodies for single cell protein analysis likely yield more information content per read per cell than RNA. However, the costs associated with standard microfluidics based single-cell library construction16 and conjugation of modified DNA sequences to antibodies4 are high. Thus, for single-cell proteomic sequencing to be a compelling strategy for high dimensional phenotyping of millions of cells, there is a major need to develop a workflow that minimizes library and antibody preparation costs.
We describe a simple two round SCI experimental workflow, SCITO-seq, which combinatorically indexes single cells using DNA-tagged antibodies4 and microfluidic droplets to enable cost-effective profiling of cell-surface proteins scalable to 105-106 cells (
Our approach is based, in part, on the discovery that the large number of droplets produced by microfluidic workflows (˜105 for 10X Genomics16) can be used as a second round of physical compartments for single-cell combinatorial indexing (SCI)17-20 resulting in a simple and cost-effective two-step procedure for library construction.
Disclosed herein is a strategy using universal conjugation followed by pooled hybridization to generate large panels of DNA tagged antibodies referred to as “Handle-Tagged antibodies” or “HTA”. Handle-Tagged antibodies are then used to stain cells in individual pools prior to high-concentration loading using commercially available microfluidics devices and methods. Using the current invention, an Antibody Barcode or Handle can be used to identify a cell-surface protein displayed on a cell. Protein expression profiles for multiple (two or more) cells simultaneously encapsulated in a single drop is resolved by the combinatorial index of pool and droplet barcodes. The high concentration loading of stained cells and targeted sequencing reduce the library construction and sequencing costs per cell respectively compared to other single cell sequencing workflows. We demonstrate the feasibility and scalability of SCITO-seq in mixed species and mixed individual experiments profiling 105 cells per microfluidic reaction, a 4-fold increase in throughput compared to standard workflows at the same collision rates. We further illustrate an application of SCITO-seq by profiling 5×104-105 peripheral blood mononuclear cells using a panel of 28 antibodies in one microfluidic reaction from two healthy donors and benchmark the results with mass cytometry (CyTOF). Finally, we demonstrate that targeted sequencing using SCITO-seq can recover the same cell clusters at lower sequencing depths per cell. SCITO-seq can be integrated with existing workflows for multimodal profiling of transcripts22 and accessible chromatin21 and can be a compelling platform for obtaining rich phenotyping data from high-throughput screens of genetic and extracellular perturbations.
Antibodies (or other affinity reagents) used in the invention are attached or conjugated to an oligonucleotide referred to as a “Handle” or “Handle sequence.” The antibody and attached Handle are referred to herein as a “Handle-Tagged Antibody” or “HTA.” Other terms that may be used to describe the antibody-handle complex include “tagged-antibody,” “barcoded antibody,” and “DNA-tagged antibody.” In one approach, each different Handle corresponds to a specific monoclonal antibody or binding specificity.
Handle
The Handle is long enough to form a stable complex with the Handle Complement, described below, under assay conditions. Generally, the Handle is at least 10 bases in length, more often 15 bases in length and often 20 bases in length or longer. For example and not limitation, the length of the Handle can be 10-100 bases, 15-50 bases, or 15 to 25 bases.
Antibodies
The antibody portion of the Handle-Tagged Antibody is typically a monoclonal antibody such as a monoclonal antibody specific for a cell-surface protein (“CSP”). In some embodiments, an antibody specific for a cell-surface protein binds an epitope on the extracellular portion of a cell-surface transmembrane protein. In some embodiments, an antibody specific for a cell-surface protein binds an epitope on a peripheral membrane protein.
It will be recognized that there are a large number of different cell surface proteins. A CSP is generally a naturally occurring protein expressed by a defined, or definable, cell type or types. That is, knowledge of the CSPs expressed by a cell provide information about the cell properties, including type, species, developmental or metabolic state and the like. Any sort of cell can be characterized using the methods of the invention, including cells from an animal, such as a primate (e.g., such as a human), plant, or fungus, and microorganisms.
In certain embodiments the CSP is expressed by and displayed on an immune system cell, such as a lymphocyte, neutrophil, eosinophil, basophil or monocyte. Useful CSPs displayed on immune cells include proteins referred to by cluster of differentiation (CD) designations assigned by HLDA (Human Leukocyte Differentiation Antigens) Workshops. See for example, Beare et al., 2008, “The CD system of leukocyte surface molecules: Monoclonal antibodies to human cell-surface antigens.” Curr. Protoc. Immunol. 80:A.4A.1-A.4A.73, incorporated herein by reference. Exemplary CD proteins are listed in TABLE 1 along with exemplary monoclonal antibodies.
In certain embodiments the CSP is expressed by and displayed on a cell other than an immune system cell. See for example, Bausch-Fluck et al., 2015, “A Mass Spectrometric-Derived Cell Surface Protein Atlas. PLoS ONE 10(4): e0121314. Bausch-Fluck et al., 2015, “The in silico human surfaceome” Proceedings of the National Academy of Sciences November 2018, 115 (46) E10988-E10997; Fonseca et al., 2016, “Bioinformatics Analysis of the Human Surfaceome Reveals New Targets for a Variety of Tumor Types,” International Journal of Genomics Volume 2016, Article ID 8346198. Suitable monoclonal antibodies are described in public databases (e.g., Genbank, NCBI, EMBL, AbMiner, Antibody Central, European Collection of Cell Cultures, The Hybridoma Databank, Monoclonal Antibody Index). New monoclonal antibodies against any specific antigen can be prepared by art-known methods.
In some embodiments the invention is used to detect or quantitate proteins other than cell surface proteins (e.g., cytoplasmic proteins).
Association of Handle and Antibody.
Generally each different antibody is associated with a unique Handle sequence so that determining a Handle sequence identifies properties of the antibody. In general each antibody used in an assay has a different CSP specificity (e.g., anti-CD2, anti-CD17) which is identified by the Handle sequence. In some embodiments two different antibodies recognize the same CSP but, for example, bind to different epitopes and/or have different isotypes. In some embodiments two different antibodies linked to different Handle sequences recognize the same CSP but in different configurations (e.g., distinguishing dimers from monomers). In some embodiments two antibodies with different specificities are tagged with the same Handle sequence, if there is no need to distinguish the corresponding CSPs.
Attachment of the Handle to the Antibody to Form the Handle-Tagged Antibody.
Methods for attaching the Handle oligonucleotide and the antibody to produce the Handle-Tagged Antibody are known in the art. See, e.g., Stoeckius et al., 2018, Genome Biol. 19:224; Peterson et al., 2017, Multiplexed quantification of proteins and transcripts in single cells Nature Biotechnology 35:936-939. In one approach, the Handle oligonucleotide is an amine modified oligonucleotide conjugated to the antibody or a polypeptide constituent thereof. The Handle can be attached to the antibody at its 5-prime end or its 3′ end depending on downstream steps.
The Pool-Oligonucleotide, also referred to as “Pool Oligo,” “Splint Oligo,” “Secondary Oligo,”.and “Ab-Pool Oligo” has the structure and elements listed below. Particular embodiments of the Pool Oligo are shown in
A “Handle Complement” (H′), an oligonucleotide sequence complementary to the Handle sequence. In one approach, the Handle Complement is at the 5′ end of the Pool Oligo. In one approach, the Handle Complement is at the 3′ end of the Pool Oligo. The Handle sequence (or its complement) sometimes has a length of about 20 bp, and usually has a length of 10 to 100 bp, and often 15 to 50 bp.
Elements for connecting the pool oligonucleotide to the droplet olionucleotide. In a hybridization-based approach a “Capture Complement” (C′) which is an oligonucleotide sequence complementary to the capture sequence of the Droplet Oligonucleotide (discussed below). In one approach, the Capture Complement is positioned at the 3′ end of the Pool Oligo is used. The Capture Complement (or Capture sequence) sometimes has a length of about 22 bp, and usually has a length of 10 to 100 bp, and often 15 to 50 bp. In a ligation-based approach the Pool Oligo has a ligatable (e.g., phosphorylated) 5′ terminus that can be ligated to the 3′-terminus of the Droplet Oligonucleotide. Advantageously ligation is facilitated by a Bridge Oligonucleotide (discussed below).
A “Pool Barcode Complement” (PBC′) or “Pool Barcode” is a barcode sequence that identifies the individual pool in which Handle-Tagged Antibodies are combined with Pool Oligos (i.e., Ab-Pool Oligos). For example, the Handle-Tagged Antibodies may be combined with Pool Oligo associated with the Handle-Tagged Antibody.
An “Antibody Barcode Complement” (ABC′) is a sequence that (like the Handle) corresponds to (identifies) the antibody portion of the Handle-Tagged Antibodies.
The “Pool Barcode” and “Antibody Barcode” may be independent barcodes including, for example, barcodes separated by an intervening non-barcode sequence. Alternatively the “Pool Barcode” and “Antibody Barcode” may be a unitary or compound barcode (e.g., a single barcode of contiguous bases that identifies both the pool and antibody. Pool barcodes can also serve as sample barcodes to enable multiplexed SCITO-seq. The choice of separate or compound Pool and Antibody Barcodes will depend on the preferences of the operator. A compound Ab+Pool barcode of a given length (e.g., 10 bp) can encode a larger number of bar code species than separate Pool and Antibody Barcodes with the same total length (e.g., 5 bp each). A compound Ab+Pool barcode often has a length of about 10 bp, such as 5 to 25 bp. The compound Antibody+Pool barcode can be referred to as an “Ab+Pool BC” or complement thereof. However, unless otherwise clear from content, any reference to the Pool Barcode and Antibody Barcode should be understood to refer equally to the compound barcode.
The Pool Oligo may optionally include other sequence features, including an amplification primer binding site or a sequencing primer binding site (which may be the same or different) shown in
The “Droplet oligonucleotide” has the structure and elements listed below. Certain features of the Droplet oligonucleotide vary based on the sequencing platform used. For example, in droplet-based approaches such as 10X Genomics Chromium, inDrop and Drop-seq (see Zhang et al., 2019, Comparative Analysis of Droplet-Based Ultra-High-Throughput Single-Cell RNA-Seq Systems, Molecular Cell 73:130-142.e5, incorporated herein by reference), multiple copies of a Droplet oligonucleotide (generally having the same, unique, sequence) are attached to a bead or similar solid substrate compatible with droplet-based analyses (shown as a circle in
Specific embodiments of the Droplet Oligonucleotide are shown in
A “Capture Sequence” region (C) for association with the Pool Oligonucleotide. Typically the capture sequence is at the 3′ end of the Droplet oligonucleotide. In a hybridization-based approach, the Capture Sequence may be complementary to the Capture Complement of the Pool Oligo. Alternatively, in a ligation-based approach the 3′ terminus of the Droplet Oligo is joined to a ligatable end of the Pool Oligonucleotide (e.g., the 3-prime end of the Droplet Oligonucleotide may be ligated to a phosphorylated 5′ end of the pool oligonucleotide.)
A “Droplet barcode” (DBC) sequence, which is typically 5′ to the Capture Sequence. The DBC is configured so that there is one DBC sequence per compartment (discussed below). In bead-based systems each bead is associated with a unique DBC (represented as many copies in or on the bead). In well-based systems each well contains multiple copies of a well-specific BC. The term “Droplet barcode” does not require that the compartment be a droplet.
The Droplet oligonucleotide may contain additional barcodes, such as a unique molecular identifier or UMI.
The Droplet oligonucleotide typically include other features, such as amplification primer binding sites or sequencing primer binding sites (which may be the same or different) shown in
The SCITO assay is used to characterize the distribution of multiple CSPs in a cell population, and therefore uses a panel of multiple Handle-Tagged Antibodies. In various embodiments the number of different CSPs for which there are Handle-Tagged Antibodies in an assay is at least 3, at least 5, at least 10, at least 12, at least 15, at least 10, or at least 25 such as, for example, from 3 to 100, from 5 to 50, from 10 to 50, from 15 to 50, or from 25 to 50.
Exemplary panels for human immune cells include:
As noted above, any type(s) of cells may be used in the assay. Generally a sample contains is a heterogeneous mixture of multiple cells types (e.g., peripheral blood cells) or a heterogeneous mixture of similar cells exposed to different conditions, having different developmental histories, or the like. Cells used in the assay may be prepared by known means (e.g., washing, optional fixation).
A panel of Handle-Tagged Antibodies representing the CSPs being assayed is selected and the Handle-Tagged Antibodies are pooled into a single mixture (“panel pool”). Generally the panel pool contains equal amounts of each represented antibody. However, the relative proportions of individual Handle-tag antibodies can vary and can be selected by the practitioner based on the cell population, the affinity of different antibodies for the corresponding antigen, etc.
The number of different Handle-Tagged Antibodies, exclusive of controls, may be equal to the number of surface proteins being assayed for.
As illustrated in
As illustrated in
The Handle complement sequences of the Pool Oligos and Handle sequences of the Handle-Tagged Antibodies are allowed to anneal in the vessel to form the “Staining Construct.” As a result, each pool or compartment contains Pool Oligos that have a common Pool Barcode (which identifies the pool), and contains Antibody Barcodes, Handle sequences, and Handle Complement sequences all of which identify the antibody specificity of the Handle-Tagged Antibody. In one approach, the Handle is attached at its 3′ terminus to the antibody (see, e.g.,
Table 2 and
It will be recognized that when a unitary or compound Pool Barcode-Antibody Barcode (Ab+PBC) is used, each pool or compartment contains Pool Oligos containing compound Pool Barcode-Antibody Barcode in which all identify the Pool and subsets identify the Antibody.
It will be recognized that it is not required that all of the Pool Barcodes (or Pool-identifying portions of the unitary Pool Antibody Barcode) in a vessel are necessarily the same (i.e., identical sequence) so long as the pool is identified by the sequence.
A plurality of cells is added to each well, whereby the cells in each well are stained with (bound by) the Staining Constructs. Thus, each cell displaying a CSP(s) is bound to one or more Staining Constructs containing an antibody-specific Handle and antibody specific barcode (PBC′) and a pool barcode (ABC′).
In one approach, cells are combined with Handle-Tagged antibodys (HTAs) prior to adding Pool Oligos. Pool Oligos may be added after HTAs have bound cells. Alternatively, cells, HTAs and Pool Oligos can be combined at the same time and self assemble to produce stained cells. These approaches may have advantages in certain microfluidic work-flows, but are likely to result in increased background. Generally, as discussed above, HTAs and Splint Oligos are allowed to associate to form a complex prior to being combined with cells.
Following staining, the stained cells may be combined into a mixture prior to distribution into compartments.
The compositions and methods of the invention can be carried out using droplet-based methods, including the InDrop, Drop-seq, 10× Genomics Chromium platforms and non-droplet based methods as discussed in § 5 above. See Zhang et al., 2019, Comparative Analysis of Droplet-Based Ultra-High-Throughput Single-Cell RNA-Seq Systems, Molecular Cell 73:130-142.e5; Mimitou et al., 2019, Multiplexed detection of proteins, transcriptomes, clonotypes and CRISPR perturbations in single cells Nature Methods 16:409-412; Fan et al., 2015, Expression profiling. Combinatorial labeling of single cells for gene expression cytometry Science, 347:1258367; and Han et al., 2018, Mapping the mouse cell atlas by Microwell-seq, Cell, 172:1091-1107.e17, each of which is incorporated herein by reference. In general, reagents and methods described in the literature or materials from manufacturers can be adapted to the present invention.
According to the present invention, the stained cells are pooled and distributed into wells or droplets. Loading cells can be carried out using art known means including using commercially available devices used for droplet-based single cell sequencing. See, e.g., Section 10.
Conventional cell analysis methods generally require that individual cells are contained in separate compartments, typically according to a Poisson distribution. For example, the 10× literature recommends steps to maximize the number of droplets that have a single cell (single cell encapsulation), and minimize the number of droplets that are empty or contain two or more than two cells. See Zheng et al., 2017, Massively parallel digital transcriptional profiling of single cells Nature Communications 8, Article number: 14049 and kb.10xgenomics.com/hc/en-us/articles/218166923-How-often-do-multiple-Gel-Beads-end-up-in-a-partition. For the 10X Genomics platform, Poisson loading at the recommended concentrations of 2×103-2×104 cells result in collision rates of 1-10%. However, greater than 97%-82% of droplets do not contain a cell, leading to wasted reagents. In contrast, according to the present methods, antibody binding to CSPs from two cells, or two or more cells, in the same droplet (multiplets) can be distinguished and resolved based on the information provided by barcodes. In the present methods cells may be loaded at high concentrations where the majority of droplets will contain at least one cell. tunable to a targeted collision rate. For example, for a commercially available microfluidic platform where ˜105 droplets are formed, a loading concentration of 1.82×105 cells results in 84% of droplets containing at least one cell but only 4.4% of droplets containing greater than four cells. To yield 105 resolved cells at a collision rate of 5% for this loading concentration, 11 antibody pools would be needed. At 160 pools and 5% collision rate, 1×106 cells can be profiled in one microfluidic reaction with an average of 18.9 cells captured per droplet. In some embodiments at least 25% of compartments occupied by at least one cell (i.e., not empty) contain two cells, sometimes at least 30%, at least 40%, at least 50%, or at least 60%. In some embodiments at least 25% of occupied compartments contain more than one cell (i.e., two or more cells), sometimes at least 30%, at least 40%, at least 50%, or at least 60%. It will be apparent that, in relation to the number of cells in a compartment or droplet, there is an upper limit beyond which benefits diminish. This in some embodiments the multiplicities of encapsulation (MOE) or number of cells per occupied compartment range from 1 to 10 cells per droplet, e.g., up to 10, up to 9, up to 8, up to 7, up to 6, up to 5, or up to 4
As illustrated in
In another approachAs illustrated in
In addition to the DBC, PBC, and ABC (sometimes referred to as “the three barcodes”) the Sequence fragment structure will include elements that allow sequencing of the three barcodes. The three barcodes can be sequenced in a single read, as two paired-end reads (also called mate pair reads), or any other fashion that identifies the combinations of the three barcodes associated on any Sequence Fragment Structure. For example, referring to
It will be within the ability of a person of skill in the art to generate a sequenceable Sequence Fragment Structure using enzymes such as reverse transcriptase, DNA polymers, DNA ligase and art-known strategies such as primer extension, and to prepare a sequencing library. Sequencing may be carried out using any suitable massively parallel sequencing platform, including, for example, Illumina's cluster based sequencing by synthesis platforms and MGI's DNBSeq platforms.
Using the present invention, data from each individual cell includes three identifiers (barcodes): Handle-Tagged Antibody, Pool Oligonucleotide, Droplet Oligonucleotide, and optionally UMI data. As discussed below, using this approach the surface protein expression profiles of multiple encapsulated cells (multiplets) within a droplet can be resolved by the combinatorial index of Antibody Barcode, Pool Barcode (e.g., Ab+PBC) and Droplet Barcode.
As cell loading is governed by a Poisson distribution, the major limitation of standard droplet-based single cell sequencing (dsc-seq) workflows is ensuring encapsulation of single cells to reduce the number of collisions. This results in suboptimal cell recovery, reagent usage, and inflated library construction costs. For the 10X Genomics single-cell sequencing platform, Poisson loading at the recommended concentrations of 2×103-2×104 cells result in cell recovery rates (CRR) of 50-60%16,22 and collision rates of 1-10%. However, at these concentrations, 97%-82% of droplets do not contain a cell, leading to wasted reagents. One approach to decrease the library preparation cost and increase the sample and cell throughput of dsc-seq is to “barcode” samples using either natural genetic variants10,23,24 or synthetic DNA molecules11,12,25 prior to pooled loading at 5×104-8×104 cells, reducing the proportion of droplets without a cell to ˜65%-45%. Because simultaneous encapsulation of cells within a droplet can be detected by the co-occurrence of different sample barcodes (e.g., genetic variant or synthetic DNA tags) with the same droplet barcode (DBC), sample multiplexing increases the number of singlets recovered per microfluidic reaction while maintaining a low effective collision rate tunable by the number of sample barcodes. However, since collision events can only be detected but not resolved into usable single-cell data, the maximum loading concentration that minimizes total cost is ultimately limited by the overhead cost incurred for sequencing collided droplets.
Single-cell combinatorial indexing (SCI) is an alternative, scalable approach to control the collision rate of single-cell sequencing by labeling subsequent rounds of physical compartmentalization with DNA barcodes. While standard SCI approaches require more than two rounds of combinatorial indexing to sequence 105-106 cells17-20, recent advances utilizing droplet-based microfluidics for combinatorial indexing have enabled simplified two-round workflows to achieve the same throughput21,22. For applications where only a set of targeted markers are needed such as high-throughput screens and clinical biomarker profiling, current SCI workflows profiling the entire epigenome or transcriptome per cell is not optimized for sensitivity and would likely result in prohibitively high sequencing costs.
An element of SCITO-seq arises from the recognition that Poisson loading naturally limits the number of cells within a droplet even at very high loading concentrations. Thus, indexing cells using a small number of antibody pools will ensure that the combinatorial index (Ab+PBC and DBC) will identify a cell at low collision rates even at high loading concentrations. Theoretically, given P pools, C cells loaded, D droplets formed, the collision rate is given as
while rate of empty droplets is given by
(see § 23, Methods). Our derivation of the collision rate differs from previously reported estimates derived from the classical birthday problem22, which did not account for higher order collision events of more than two cells with the same barcode. These closed form derivations of the collision and empty droplet rates are nearly identical to those obtained based on simulations. For example, when 6×105 droplets are formed, a loading concentration of 1.82×105 cells (target recovery of 105 cells) results in 84% of droplets containing at least one cell but only 4.4% of droplets containing greater than four cells. To yield 105 resolved cells at a collision rate of 5% for this loading concentration, only 10 antibody pools would be needed to achieve a total cost of 3.1¢/cell. Note that as the library preparation cost quickly diminishes for SCITO-seq with increasing number of pools, the total cost per cell is dominated by antibody costs. Therefore, while 384 pools achieves the maximal 12-fold reduction in cost compared to standard single-cell proteomic sequencing (2.2 vs 26 cents), 10 antibody pools can already achieve a 8-fold reduction in cost (3.1 vs 26 cents) while minimizing experimental complexity (
To demonstrate the feasibility and scalability of SCITO-seq, we performed a mixed species experiment by pooling human (HeLa) and mouse (4T1) cells, splitting into five aliquots, and staining each pool with anti-human CD29 (hCD29) and anti-mouse CD29 (mCD29) antibodies labeled with pool-specific barcodes (
We next sought to further assess the scalability of SCITO-seq and its applicability to resolve quantitative differences in cellular composition based on surface protein expression. We isolated and mixed primary CD4+ T and CD20+ B cells from two donors at a ratio of 5:1 (T:B) for donor 1 and 1:3 (T:B) donor 2. The mixed cells were aliquoted into five pools and each stained with pool-barcoded anti-CD4 and anti-CD20 antibodies (
Merging the ADT data across the five pools, anti-CD4 and anti-CD20 antibodies stained the expected cell types defined by the transcriptome. Based on the ADTs, we estimated 40% of CCDs to be between cell-type multiplets, which is consistent with estimates from the transcriptomic analysis (49.6%,
We next assessed if SCITO-seq can capture unequal distributions of B and T cells from the two donors, especially from CCDs that encapsulated multiple cells. For this analysis, we focused only on 45,240 CCDs (donor 1: 25,630, donor 2: 19,610) predicted to contain cells from only one donor based on genetic demultiplexing. Within CCDs with only one antibody pool barcode detected, analysis of the proportions of T and B cells (T:B200K:5.0:1 for donor 1 and 1:2.8 for donor 2) mirrored the expected proportions for each of the two donors and was consistent with estimates obtained from the transcriptomic data. Encouragingly, approximately the same proportions were estimated in CCDs with multiple pool barcodes (multiplets) (T:B200K 4.0:1 for donor 1 and 1:2.9 for donor 2).
Because pool-specific effects appear to be minimal in SCITO-seq, the pool-specific antibody barcodes could be used to directly label samples, obviating the need for orthogonal sample barcoding. To demonstrate this application, we performed another experiment where we stained one donor per pool and each pool contained different barcoded antibodies (e.g., pool 1 contains CD4-BC1 while pool 2 contains CD4-BC2, etc.). For loading concentrations of 2×104 and 5×104 cells, we obtained 17,730 and 34,549 post-processing CCD, sequenced to a per CCD depth of 964 and 1,540 reads for the ADT and 20,951 and 14,332 reads for the RNA. We observed the expected proportion of T and B cells per donor based on the distribution of the expression of CD4 and CD20 respectively. After resolution, we recovered 18,680 and 41,059 cells at collision rates of 7.4% and 18.6% respectively. Estimates of co-occurrence frequencies of different pool and antibody barcodes were highly correlated (r=0.99, p-value<0.001) with observed values.
To demonstrate SCITO-seq's applicability for high-dimensional and high-throughput cellular phenotyping, we profiled peripheral blood mononuclear cells (PBMCs) from two healthy donors using a panel of 28 monoclonal antibodies across 10 pools. After staining, pooling, and processing 2×105 cells in a single 10X channel using 3′V3 chemistry, we sequenced the resulting ADT and RNA libraries and obtained 49,510 post-filtering CCDs (
We separately analyzed the merged ADT and RNA data by normalizing the counts, performing dimensionality reduction, and constructing a k-nearest neighbor graph (see § 23, Methods). Leiden clustering based on either merged ADT or RNA counts (
We further assessed the accuracy of SCITO-seq for quantitative immune phenotyping by comparing the compositional estimates obtained from CCDs with a single detected pool barcode (singlets) versus those with multiple detected pool barcodes (multiplets). We focused the analysis only on CCDs with cells from one donor as estimated using genetic multiplexing. UMAP projections for resolved cells originating from singlets vs multiplets were qualitatively similar (
One advantage of SCITO-seq as a tool for high-dimensional and high-resolution phenotyping is the high information content obtained by profiling protein abundance. This is demonstrated by downsampling of the 2×105 dataset where only ˜25 UMIs/cell corresponding to ˜60 reads/cell (assuming 45% library saturation) were needed to achieve an Adjusted Rand Index (ARI) of >0.8 for assigning cells to the same clusters in the full dataset (
To further demonstrate the flexibility and scalability of SCITO-seq beyond the number of markers detectable by competing flow and mass cytometry methods9,26, we evaluated the performance of SCITO-seq using a 60-plex custom panel and a commercial Totalseq-C(TSC) 165-plex antibody panel. To achieve compatibility with the commercial TSC panel where anti-body oligos are conjugated on the 5′ end versus the 3′ end for SCITO-seq, we designed a set of splint oligos to hybridize to each of the 165 15 bp antibody barcodes in the panel.
For both experiments, we further leveraged the pool barcodes encoded in each set of splint oligos as a sample label to enable multiplexing. We stained the same 10 donors in 10 distinct pools using either panel and loaded 4×105 cells to tune our targeted recovery to 2×105 cells per experiment. In the 60-plex experiment, we recovered 69,733 CCDs and resolved 219,063 cells (
After removal of collided barcodes based on the number of expressed markers (see § 23, Methods), we obtained 175,930 and 175,000 cells in the 60-plex and 165-plex experiments respectively. After normalization, dimension reduction, and k-nearest neighbor graph construction, the cells were clustered into 26 and 19 clusters respectively and visualized in UMAP space (
The increase in throughput of SCITO-seq can be particularly useful for large-scale profiling of multiple samples. This is further facilitated by the pool barcodes in the splint oligo design which can be used to directly label samples obviating the need for orthogonal sample barcoding (
We sought to enable combinatorially indexed multimodal profiling of the transcriptome and surface proteins by combining SCITO-seq with the recently published scifi-RNA-seq22. Scifi-RNA-seq generates combinatorial indices by adding pool-specific barcodes on transcripts through in-situ reverse transcription and ligates the DBC from the 10X single-cell ATAC-seq (scATAC-seq) gelbeads. See Datlinger et al., 2019, Ultra-high throughput single-cell RNA sequencing by combinatorial fluidic indexing, bioRxiv, incorporated herein by reference. To first enable compatibility of SCITO-seq with the scATAC-seq chemistry, we modified the bead hybridization sequence of the splint oligo to be complementary to the ATAC-seq gelbead sequence. After droplet emulsion breakage and subsequent harvest with silane DNA-binding beads, DNA was eluted and amplified to add sequencing adaptors. We applied the modified SCITO-seq workflow to profile PBMCs from one donor in five pools with 12 broad phenotyping surface markers using the 10X scATAC-seq chemistry. As a proof of principle, we loaded 5×104 cells to recover 21,460 cells and identified the expected clusters of T, B, myeloid, and NK cells expressing the canonical surface proteins demonstrating the compatibility of SCITO-seq with scATAC-seq chemistry.
Scifi-RNA-seq utilizes a bridge oligo to facilitate the ligation of DBCs within scATAC-seq gelbeads and requires a number of cycling conditions that is not directly compatible with SCITO-seq. To enable multimodal profiling, we next designed an orthogonal bridge oligo specific to the SCITO-seq design to assist capture and ligation of SCITO-seq ADTs to the 10X scATAC-seq gelbead capture sequence (
After pre-processing, we obtained an average of 310 UMIs per cell for the RNA library (average 146 genes/cell) and an average of 550 UMIs per cell for the ADT library. After normalization of the ADT counts, dimensionality reduction, and k-nearest neighbor graph construction, we identified 5 clusters using Leiden clustering visualized in UMAP space (
To generate compatible secondary oligos with scifi-RNA-seq, we conjugated unique 20 bp 5′ amine modified oligos to each of our six antibodies, varying from our previous 3′ amine conjugation to present a favorable orientation of the secondary oligonucleotide (Splint Oligo) for capture in a similar fashion to transcripts in the scifi-RNA-seq workflow. In addition, we spiked-in an additional orthogonal bridge oligo for the in-emulsion ligation to reduce competition of transcripts and ADT molecules for the bridge oligo. We stained 5 pools of a mixture of 5 cell lines for 30 min prior to washing and executing the scifi-RNA-seq protocol. After the scifi-RNA-seq workflow, we loaded 3×104 into the 10× chromium controller using the lox ATAC-seq kit. After emulsion breakage as in the 10× user guide, we saved 4 μl of the 24 μl silane bead elution for ADT library construction. The ADT sample index PCR reaction was set up with 4 μl of sample, 5 μl of P5 primer (10 μM), 5 μl of i7 index primer (10 μM), 50 μl of KAPA HiFi mastermix, and 36 μl of RNAse-free water. Cycling conditions were as follows: 98° C. for 45 s, followed by 12 cycles of 98° C. for 20 s, 54° C. for 30 s, 72° C. for 20 s, and ending with a final extension of 72° C. for 1 min. We cleaned up and selected the fragments using AMPure XP beads at a ratio of 1.2X, prior to a final elution in 20 μl. To construct the gene expression library, we used a plexWell 96 Library Preparation kit (Seqwell ref PW096-1) to tagment 10 ng of DNA per reaction. This pre-loaded Tn5 was used to ease the number of tagmentations in the scifi-RNA-seq workflow and increase the reproducibility with a commerical product over custom-loaded Tn5s. The final gene expression library sample index PCR was performed as-is in the scifi-RNA-seq workflow. The resulting libraries were sequenced on a Novaseq 6000 Si v1.0 flow cell with the following read configuration: 21:8:16:78 (Read1:i7:i5:Read2).
To process the transcriptomic data, the generated fastqs (R1:21 bp, R2:16 bp, R3:78 bp) were stitched to make a final R1 file containing a droplet barcode (16 bp)+well barcode (11 bp)+UMI (8 bp) per read. We used kallisto version 0.46.1 and specified the cell barcode as 27 bp (16+11; droplet and well barcode bp lengths) and ran bustools to produce count matrices (www.kallistobus.tools/getting_started). To process the ADT fastqs (same read configuration as RNA) were stitched to produce a final R1 file (35 bp), R3 data was trimmed to 10 bp (encoding antibody barcode) for barcode alignment. These reads were then processed using a modified dropseq pipeline (v2.4.0; aligner swapped to bowtie (v2.4.2)) (www.github.com/broadinstitute/Drop-seq/releases). Counts were then normalized as done in the PBMC experiment above for both ADT and RNA. RNA genes were determined based on manual curation after running the Wilcoxon's test for determining highly variable marker genes. For overlap analysis in
We initially designed a secondary oligo compatible with the 10×ATAC-seq kit by changing the hybridizing end of the splint oligo to the reverse complement of the Read 1 Nextera sequence) from the feature barcode capture sequence (10×3′v3). We modified the microfluidic cell and enzyme mixture to the following mastermix; 4 μl of 10 mM dNTP, 16 μl of RT buffer (5×), 4 μl of Maxima H minus, and cells and RNAse free water up to 80 μl. After running the solution through a 10× chip E reaction as in the 10× user guide, the GEMs were thermocycled at 53° C. for 45 min and 85° C. for 5 min. The emulsion was broken as in the 10× user guide and ADT fragments were eluted in 40 μl. We performed an index PCR with the following conditions: 40 μl of sample, 50 μl of 2×KAPA HiFi HotStart ReadyMix, 1 μl each of P5 primer (100 uM) and universal read 2 Nextera primer, and 8 μl of RNAse-free water. The sample was cycled as follows: initial denaturation at 98° C. for 45 s, cycled 12× at 98° C. for 20 s, 54° C. for 30 s, and 72° C. for 20 s, followed by a final extension at 72° C. for 1 min.
To scale SCITO-seq to a commerical platform, we modified our secondary oligo (Splint Oligo) to be compatible with Biolegend's TS-C platform (normally used for the 10×5′ kits) for the 10×3′V3 kit. To do this, we changed the antibody hybridization region in our original 3′v3 design to the reverse complement of antibody specific TS-C barcode (15 bp) sequences. After emulsion breakage, we followed the index PCR protocol as per manufacturer's recommendations (10× Genomics, CG000185 Rev D, page 52).
In additional embodiments, the Handle oligonucleotide is attached to the antibody via a noncovalent link, such as a streptavidin-biotin link, or a cleavable link, such as a disulfide bridge.
In additional embodiments, affinity reagents other than antibodies may be used to recognize CSPs. These include, for example, aptamer, affirmer, and knottins. See, e.g., U.S. Pat. No. 8,481,491; Cochran, Curr. Opin. Chem. Biol. 34:143-150, 2016; Moore et al., Drug Discovery Today: Technologies 9(1):e3-ell, 2012; Moore and Cochran, Meth. Enzymol. 503:223-51, 2012; Jayasena, et al., Clinical Chemistry 45:1628-1650, 1999; Reverdatto et al., 2015, Curr. Top. Med. Chem. 15:1082-1101. This disclosure should therefore be read as if each and every reference to “antibodies” referred equally to other “affinity reagents” not limited to aptamers, affirmers, and knottins.
In certain embodiments, some of all of the antibodies or other affinity agents to which the Handle is attached bind to cell surface proteins (e.g., peripheral membrane proteins or the extracellular portion of transmembrane proteins). In additional embodiments some or all of the antibodies or other affinity reagents used in an assay bind to any of (a) a cell-surface antigen other than a protein (e.g., cell membrane lipid); (b) intracellular proteins (e.g., cytoplasmic proteins).
The approach described herein can be use with 3′ or 5′ conjugation of the Handle to the antibody, as well as with various commercial platforms and devices. In one approach, the Handle oligonucleotide is conjugated at its 3′ end to the antibody protein as illustrated in
It will be recognized that a pool oligonucleotide may associate with a droplet oligonucleotide by hybridization of complementary sequence or, alternatively a pool oligonucleotide may associate with a droplet oligonucleotide by ligation. In one embodiment of the ligation option the orientation of the pool oligonucleotide is reversed and there is a concomenant reversal of the orientation of the antibody handle (handle is associated with antibody at its 5′ end rather then its 3′ end. The various embodiments described in detail in this disclosure are not intended to be limiting in any fashion. The reader will recognize that rearrangements consistent with the practice of the method may be made and are contemplated here. hybridization the droplet [0106] All references to bar codes should be understood to include either the bar code or the complement of the bar code, as will be clear from context, and reference to “bar code” or “bar code complement” should be so understood. Likewise, it will be recognized the references to oligonucleotides and segments therein should be understood to include the complement when it is clear from the description that such complementarity with an element is required for the association of bar codes and other elements as described herein.
Orthogonal assays: The methods described herein can be combined with simultaneous profiling of additional modalities such as transcripts and accessible chromatin or tracking of experimental perturbations such as genome edits or extracellular stimuli. See, for example, Peterson et al., 2017, Multiplexed quantification of proteins and transcripts in single cells Nature Biotechnology 35:936-939; Stoeckius et al., 2017, Simultaneous epitope and transcriptome measurement in single cells. Nature Methods 14: 865-868 and Datlinger et al., 2019, Ultra-high throughput single-cell RNA sequencing by combinatorial fluidic indexing. bioRxiv
In an additional embodiment the sequence of the Handle sequence(s) associated with each stained cell is determined. In some embodiments, the Handle is positioned so that it flanked by primer binding sites in the Sequence Fragment Structure, for example, as shown in
a. Closed Form Derivation of Collision and Empty Droplet Rates
Suppose there are P pools of cells. For pool p, cells arrive according to a Poisson point process with rate λp>0 (abbreviated PPP(λp)), where the unit of time corresponds to the inter-arrival time of droplets. In the most general formulation, we assume that the point processes for different pools are independent. Further, we assume the probabilities of a gel/bead and a cell encapsulated into a droplet as ρpb and ρpc, respectively. Therefore, by Poisson thinning, the arrival of cells follows PPP(ρpcλp).
We are interested in the probability of the event (called collision) that a droplet contains two or more cells from the same pool. Let N, denote the number of cells from pool p successfully loaded into a droplet. Then, N1, N2, . . . , Np where Np˜Poisson (ρpcλp), are independent random variables, and [Collision] can be computed as 1−
[No Droplet Collision]. Here
[No Droplet Collision] represents a probability that every droplet contains≤1 pool barcode. Therefore, we derive:
where the third equality follows from independence.
Next we condition [Droplet Collision] on
[Non-empty Droplet], which is the probability that a droplet contains a cell at a given observation,
[Non-empty Droplet]=1−
[Empty Droplet], where:
If there are D droplets formed and a total of C cells loaded evenly across the P pools (i.e., there are
cells per pool), then
for all pools p=1, 2, . . . , P and that ρpb becomes a nuisance parameter. If we further assume that ρpc=ρc=1 for all p=1, 2, . . . , P, then [Droplet Collision] and
[Empty Droplet] simplify as
And finally, to estimated conditioned probability of barcode collisions:
A second collision rate we can calculate is the cell barcoding (droplet barcode+pool barcode) collision rate which can be computed as the conditional probability that a particular pool p∈{1, 2, . . . , P} has a collision in a given droplet, given that the droplet contains at least one cell from that pool. If we assume that there are D droplets formed and a total of C cells are distributed evenly across P pools, then we obtain:
for all p∈{1, 2, . . . , P}.
The above conditional probability is related to the proportion of the number of pools with a collision in a given droplet, relative to the total number of pools each with at least one cell represented in the droplet. More precisely,
b. Simulation of Collision and Empty Droplet Rate.
For simulating the collision rates and empty droplet rates, we assumed a cell recovery rate of 60% and 105 droplets are formed per microfluidic reaction resulting in D=6*104. For C cells loaded, cell containing droplets are simulated using a Poisson process where λ=C/D. Assuming each simulated droplet i contains γi cells, we then compute the number of pool barcodes not tagging a cell in each droplet as:
the number of pool barcodes tagging exactly one cell as:
and the number of pool barcodes tagging greater than one cell as:
BCN
i
=P−BC0i−BC1i
The conditional collision rate is estimated as:
[Collision in pool p|Droplet contains at least one cell from pool
c. Estimates of Antibody Conjugation, Library Construction, and Sequencing
Cost for library conjugation is estimated to be $4 per antibody per μg using the Thunderlink conjugation kit and assuming averaged costs for input antibodies as purchased for our 60-plex panel. Cost for library preparation is estimated to be $1,500 per well as advertised by 10X Genomics. Cost for sequencing is estimated as $22,484 per 12B reads as advertised by Illumina.
d. Primary Antibody Oligonucleotide Conjugation
For the species mixing experiment, anti-human CD29 and anti-mouse CD29 antibodies were purchased from Biolegend (cat. 303021, 102235) and conjugated per antibody using a ThunderLink kit (Expedeon cat. 425-0000) to distinct 20 bp 3′ amine-modified HPLC-purified oligonucleotides (IDT) to serve as hybridization Handles. Antibodies were conjugated at a ratio of 1 antibody to 3 oligonucleotides (oligos). In parallel, oligos similar to current antibody sequencing tags were directly conjugated at the same ratio for comparison. Sequences for the hybridization oligonucleotides and directly conjugated oligos were designed to be compatible with the 10× feature barcoding system by introducing a reverse complementary sequence to the bead capture sequence, alongside a batch and antibody specific barcode for demultiplexing. Conjugates were quantified using Protein Qubit (Fisher cat. Q33211) for antibody titration and flow validation. Also, we orthogonally quantified using the protein BCA assay. For the human donor mixing experiment, CD4 and CD20 antibodies (Biolegend cat. 300541, 302343) were conjugated as described above.
e. Antibody-Specific Hybridization Design
After conjugation of primary Handle oligos, antibodies were combined and pools of oligos were used to hybridize the primary Handle sequences prior to staining. Of note, only one conjugation was done per antibody with the previously mentioned 20 bp oligonucleotide.
To avoid non-specific transfer of oligonucleotides between the different antibody clones and the same antibody clone from different wells, each clone received a unique 20 bp Handle (Antibody Handle). To sequence with antibody and batch specificity, a 10 bp barcode was added to the Pool Oligo which consisted of a reverse complementary sequence to the antibody specific primary Handle sequence (20 bp), TruSeq Read2 (34 bp), batch barcode (10 bp), and capture sequence (22 bp) (
f. Determination of Non-Specific Transfer of Oligonucleotides Between Antibodies
To determine the optimal concentration of hybridizing oligonucleotides for cell staining, we performed a mixed cell line experiment to determine the level of background staining of free oligonucleotides. A mixture of lymphoblastoid cells and primary monocytes were stained with CD14 and CD20 antibodies and hybridized with oligonucleotides with different fluorophores (FAM and Cy5 respectively) per antibody for 15 minutes at room temperature. Concentrations of hybridizing oligonucleotides with different concentrations (1 uM and 100 uM) were tested. Antibodies directly conjugated to fluorophores served as a positive control antibodies (CD13-BV421, Biolegend cat. 562596) to gate respective populations.
g. Validation of Saturation of Hybridization Oligonucleotides Using Flow Cytometry
To determine the saturation of available primary oligo Handles, 1 ug of conjugated CD3 antibody (Biolegend) was hybridized with a 1 ul of 1 uM of a reverse complementary oligo with a Cy5 modification (IDT modification/5Cy5/). After a 15 minute incubation at room temperature, 1 ul of 1 uM of the same reverse complementary oligo but with a FAM modification (IDT modification/56-FAM/) was added to the reaction and additionally incubated for 15 minutes. The cocktail was then added to 1×106 PBMCs pre-stained with Trustain FcX (Biolegend cat. 422302).
h. 10× Genomics Run for SCITO-Seq
Washed and filtered cells were loaded into 10× Genomics V3 Single-Cell 3′ Feature Barcoding technology for Cell Surface Proteins workflow and processed according to the manufacturer's protocol. Afterindex PCR and final elution, all samples were run on the Agilent TapeStation High Sensitivity DNA chip (D5000, Agilent Technologies) to confirm the desired product size. A Qubit 3.0 dsDNA HS assay (ThermoFisherScientific) was used to quantify final library for sequencing. Libraries were sequenced on a NovaSeq 6000 (Read1 28 cycles, index 8 cycles and Read2 98 cycles). R2 cycle can be reduced further for cost reduction (depending on the number of pool+antibody barcode length).
i. Mixed Species Experiment
HeLa and 4T1 cells were ordered from ATCC (ATCC cat. CCL-2, CRL-2539) and cultured in complete DMEM (Fisher cat. 10566016, 10% FBS (Fisher cat. 10083147) and 1% penicillin-streptomycin (Fisher cat. 15140122)) in a 37° C. incubator with 5% CO2 on 10 cm culture dishes (Corning). Prior to staining, cells were trypsinized at 37° C. for 5 minutes using 1 ml Trypsin-EDTA (Fisher cat. 25200056) and were quenched with 10 ml complete DMEM. Cells were harvested and centrifuged at 300×g for 5 minutes. Cells were resuspended in staining buffer (0.01% Tween-20, 2% BSA in PBS) and counted for concentration and viability using a Countess II (Fisher cat. AMQAX1000). HeLa and 4T1 cells were then mixed at equally and 1×106 cells were aliquoted into two 5 ml FACS tubes (Falcon cat. 352052) and volume normalized to 85 ul. Cells were stained with 5 ul of Trustain FcX for 10 minutes on ice. Cell mixtures were stained with a pool of human and mouse CD29 antibodies, either with the direct or universal design, in a total of 100 ul for 45 minutes on ice. Cells were then washed 3 times with 2 ml staining buffer and centrifuged at 300×g for 5 minutes to aspirate supernatant. Cells were then resuspended in 200 ul of staining buffer and counted for concentration and viability as before. Cells from each stained pooled were mixed and 2×104 or 1×105 cells were loaded into the lox chromium controller using 3′ v3 chemistry.
j. Human Donor Mixing Experiment
PBMCs were collected from anonymized healthy donors and were isolated from apheresis residuals by Ficoll gradient. Cells were frozen in 10% DMSO in FBS and stored in a freezing container at −80° C. for one day before long term storage in liquid nitrogen. Cells from two donors were quickly thawed in a 37° C. water bath before being slowly diluted with complete RPMI1640 (Fisher cat. 61870-036, supplemented with 10% FBS and 1% pen-strep) before centrifugation at 300×g for 5 minutes at room temperature. Cells were resuspended in EasySep Buffer (STEMCELL cat. 20144) at a concentration of 5×107 cells/ml before being subject to CD4 and CD20 negative isolation (STEMCELL cat. 17952, 17954). Isolated cells were counted and mixed at a ratio of 3 CD4:1 CD20 for donor 1 and a ratio of 1 CD4:3 CD20 for donor 2 for a total of 1.2×106 cells per donor. The cells were centrifuged at 300×g for 5 minutes at room temperature and resuspended in 85 ul of staining buffer and incubated with 5 ul of Human TruStain FcX (Biolegend cat: 422301) for 10 minutes on ice in 5 ml FACS tubes. Cells from each donor were either mixed prior or stained with well specific barcode hybridized antibody oligo conjugates for 30 minutes on ice. Staining was quenched with the addition of 2 ml staining buffer and washed as previously mentioned. Cells were resuspended in 0.04% BSA in PBS and cells from each well were counted, pooled equally, and then passed through a 40 um strainer (Scienceware cat. H13680-0040). The final strained pool was counted once more prior to loading into a 10× chip B with 2×104 cells, 5×104 cells, 1×105 cells, and 2×105 cells.
k. Mass Cytometry of Healthy Controls
PBMCs were isolated, cryopreserved, and thawed from the same donors as previously described. Once thawed, the cells were counted, and 2×106 cells from each donor were aliquoted into cluster tubes (Corning cat. CLS4401-960EA), and live/dead stained with cisplatin (Sigma cat. P4394) at a final concentration of 5 uM for 5 minutes at room temperature. The live/dead stain was quenched and washed with autoMACS Running Buffer (Miltenyi Biotec cat. 130-091-221). Cells were then stained with 5 uL of TruStain FcX for 10 minutes on ice before surface staining. Mass cytometry antibodies were previously titrated using biological controls to achieve optimal signal to noise ratios. The antibodies in the panel were pooled into a master cocktail and incubated with cells from the two donors and stained for 30 minutes at 4° C. After washing twice with 1 ml autoMACS Running Buffer, the cells were resuspended and fixed in 1.6% PFA (EMS cat. 15710) in MaxPar PBS (Fluidigm cat. 201058) for 10 minutes at room temperature with gentle agitation on an orbital shaker. Samples were then washed twice in autoMACs Running Buffer, and then three times with 1X MaxPar Barcode Perm Buffer (Fluidigm cat. 201057). Each sample was then stained with a unique combination of three purified Palladium isotopes obtained from Matthew Spitzer and the UCSF Flow Cytometry Core for 20 minutes at room temperature with agitation as previously described28. After three washes with autoMACS Running Buffer, samples were combined into one tube and stained with a dilution of 500 uM Cell-ID Intercalator (Fluidigm cat. 201057), to a final concentration of 300 nM in 1.6% PFA in MaxPar PBS at 4° C. until data collection on the CyTOF three days later. Immediately before running on the CyTOF machine, the sample tube was washed once with each autoMACS Running Buffer, MaxPar PBS, and MilliQ H2O. Once all excess proteins and salts were washed out, the sample was diluted in Four Element EQ Calibration Beads (Fluidigm cat. 201078) and MilliQ H2O to a concentration of 1e6 cells/mL and run on a CyTOF Helios at the UCSF Flow Cytometry Core.
I. Comparing Mass Cytometry (CyTOF) and SCITO-Seq
Data was transferred from the CyTOF computer, normalized and de-barcoded using the premessa package (www.github.com/ParkerICI/premessa). Clean files were uploaded to Cytobank (www.ucsf.cytobank.org/) for gating and manual identification of immune cell subsets. Files containing only singlet events were exported from Cytobank and analyzed with CyTOFKit2 package (github.com/JinmiaoChenLab/cytofkit2). Through CyTOFkit2, events were clustered using Rphenograph with k=150 and visualized via UMAP for proportion determination.
m. Pre-Processing and Initial Filtering
Both the species mixing experiments and human donor mixing experiments were processed using Cell Ranger 3.0 Feature Barcoding Analysis using default parameters. For cDNA and ADT alignment, we specified the input library type as ‘Gene Expression’ and ‘Antibody Capture’ respectively as recommended. For ADT alignment, specific barcode sequences (Ab+pool) were specified as a reference. Reads were aligned to the hg19 and mm10 concatenation reference for species mixing experiment. For all human experiments, the reads were aligned to the human reference genome (GRCh38/hg20). We first removed RBC and Platelets and removed cells with more than 15% of mitochondrial gene related reads. We further removed genes with less than 1 counts across all cells.
n. Normalization for Species Mixing and T/B Cell Human Donor Mixing Experiment
For cDNA counts, data was normalized by dividing each UMI counts to the total UMI counts and multiplied by 10,000. Then, the data was log 1p transformed (numpy.log 1p). Finally, the data was scaled to have mean=0 and standard deviation=1. Clustering was done using the Leiden algorithm29 using 10 nearest neighbors and a resolution of 0.2 for mixed species and two-donor experiment with two cell types (T and B cells).
To normalize ADT counts in species mixing experiment, the data was log transformed and standardized to have mean=0 and standard deviation=1. For ADT counts in two human donor mixing experiment with two cell types, after log transformation of the raw data, we used a Gaussian Mixture Model in scikit-learn package in python to normalize the data with the following parameters (convergence threshold 1e-3 and max iteration to 100, number of components 2). The data was normalized by z-score like transformation (log transformed raw value−mean of the posterior means of two components/mean of the posterior standard deviations).
o. Implementation of an Algorithm for Batch Demultiplexing and Multiplet Resolution
Considering all antibodies in each pool, we normalized each value by dividing mean expression value of CD45 counts across all pool (considered as a universal expression marker) for each droplet barcode yielding a p*m matrix (p is the number of pool and m is number of droplet barcodes). Then, the matrix was CLR normalized and demultiplexed using HTODemux from Seurat (v3.0) (www.satijalab.org/seurat/) to classify the droplet barcode to a pool or unassigned (we discretized the value of 0 or 1). Using this binary matrix, we iterated over p times (where discretized value equals 1) to get final resolved matrix of (n*r) where n is the number of antibodies used and r is the resolved number of cells. For each iteration, we selected the columns that were positive for the above-mentioned discretized matrix. An additional round of HTODemux was used to re-classify the ‘Negative’ cells from initial classification because most of the initial classification which deemed the cells negative had a UMAP distributions which were contained in the original clusters.
p. Analysis of PBMC Experiment: Normalization and Resolution of Multiplets
To normalize cDNA data for PBMC experiments, we used the same normalization method as described above. To generate UMAP based on ADT counts for PBMC experiment, we performed batch demultiplexing the multiplet resolution using the algorithm described previously. Then, the resolved matrix (n*r) goes through similar normalization as in the cDNA processing. Raw values are normalized to total counts of 10,000 per cell and log 1p transformed. Then, the values are standardized (mean 0, standard deviation 1) per batch. Using this normalized values, PCA was performed to reduce the dimensionality. Leiden clustering was done with 10 neighbors and 15 PCs from the previous step. Resolution value for 1.0 is used to assign clusters for whole PBMC experiments. Finally, UMAP was run to visualize resolved total cells. To remove collided cells in 60-plex and 165-plex experiment, we computed the average number of UMIs expressed per cell and thresholded cells based on the quantile distribution (>80% in the UMI distribution is filtered out) to remove cells and also manually inspect expression across all leiden clusters to exclude the cluster that expresses multiple markers.
q. Analysis of PBMC Experiment: Demultiplexing Donor Identity
For demultiplexing the donors, a VCF file containing donor genotype information and the bam file output from the Cell Ranger pipeline were used as inputs for demuxlet (Freemuxlet) with default parameters. For donors without genotypic information, we used Freemuxlet (https://github.com/statgen/popscle/) to assign droplet barcodes to the corresponding donor.
r. Analysis of PBMC Experiment: Downsampling Experiment with Adjusted Rand Index Calculations
To evaluate the quality of clustering at a given downsample, Adjusted Rand Index (ARI) was used as the comparison metric. Leiden clustering was performed on the full dataset and resulting cluster labels were taken as ground truth cell type assignments. To determine an optimal Leiden resolution for downsampling, clustering was performed 5 times at a range of resolutions. A resolution that produced consistently high ARI was then used to generate ground truth labels and perform clustering on downsampled data. Data was downsampled to a specified mean UMI/Antibody/cell using scanpy (1.4.5.post3) to downsample total reads. Downsampled data was then clustered and labels compared to full dataset clustering with ARI.
The invention has been described in this disclosure with reference to the specific examples and illustrations. The features of these examples and illustrations do not limit the practice of the claimed invention, unless explicitly stated or otherwise required. Changes can be made and equivalents can be substituted to adapt to a particular context or intended use as a matter of routine development and optimization and within the purview of one of ordinary skill in the art, thereby achieving benefits of the invention without departing from the scope of what is claimed and their equivalents.
For all purposes in the United States of America, each and every publication and patent document referred to in this disclosure is incorporated herein by reference in its entirety to the same extent as if each such publication or document was specifically and individually indicated to be incorporated herein by reference.
The Sequence Listing written in file 103182-1233370-004510WO_SL.txt created on Apr. 30, 2021, 1 KB, machine format IBM-PC, MS-Windows operating system, is hereby incorporated by reference in its entirety for all purposes.
This application is a national phase application of PCT Application No. PCT/US2021/023039, filed Mar. 18, 2021, which claims benefit of U.S. provisional application No. 62/991,529, filed Mar. 18, 2020, the entire content of which is incorporated herein by reference.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/US2021/023039 | 3/18/2021 | WO |
| Number | Date | Country | |
|---|---|---|---|
| 62991529 | Mar 2020 | US |
