Trends in

Genetics OPEN ACCESS

Review

Opportunities and tradeoffs in single-cell

transcriptomic technologies
Matilde I. Conte, 1 Azahara Fuentes-Trillo, 1 and Cecilia Domínguez Conde 1,
*

Recent technological and algorithmic advances enable single-cell transcriptomic Highlights

analysis with remarkable depth and breadth. Nonetheless, a persistent challenge Different single-cell sequencing ap-
is the compromise between the ability to profile high numbers of cells and the proaches have unique strengths and
achievement of full-length transcript coverage. Currently, the field is progressing weaknesses, with cellular throughput
and transcript coverage being generally
and developing new and creative solutions that improve cellular throughput, anticorrelated.
gene detection sensitivity and full-length transcript capture. Furthermore, long-
read sequencing approaches for single-cell transcripts are breaking frontiers Characterization of rearranged V(D)J se-
quences in lymphocytes is a valuable
that have previously blocked full transcriptome characterization. We here present complement to single-cell analysis of
a comprehensive overview of available options for single-cell transcriptome adaptive immune responses.
profiling, highlighting the key advantages and disadvantages of each approach.
Single-cell transcriptomics start to go be-
yond gene-level quantification and incor-
porate differential transcription start site
The evolving landscape of single-cell technologies and isoform usage in a cell-state-specific
manner.
Single-cell sequencing technologies have opened new paths to unravel cellular heterogeneity and
dynamic biological processes at unprecedented resolution. At the same time, the analysis of Long-read sequencing technologies are
single-cell full-length transcripts remains limited in cellular throughput. In this review, we discuss the next frontier in single-cell genomics
technical developments that are expanding the molecular readouts that we can obtain from at scale.
single-cell transcriptomics. We primarily focus on the intersection between advances in the exper-
imental components of these technologies, with some considerations on the computational side.

When deciding on a single-cell genomics strategy, a key decision is choosing between through-
put and the richness of transcriptome characterization. Here, we highlight the strengths and
weaknesses of different state-of-the-art single-cell transcriptomics methods. We intend to
provide the reader with a guide on how to make an informed decision on the most appropriate
technology and analytical approach.

The tradeoff between throughput and sequence coverage

The landscape of single-cell RNA-seq (scRNA-seq) methods has evolved rapidly in the past
decade. Technical advances in microfluidics, robotics, and multiplexing strategies have led to var-
ious solutions each with specific pros and cons. When designing a new project, a fundamental
aspect is whether cell throughput or transcript coverage is more relevant for the particular goal
of the project. For example, if we are aiming to generate a comprehensive single-cell atlas of a bi-
ological specimen, higher cell numbers is likely to be the best option to generate a comprehensive
cell census, including rare cell types. By contrast, a study focusing on alternative splicing (AS) 1
Human Technopole, Viale Rita
(see Glossary) will require methods that maximize transcript coverage and allow robust isoform Levi-Montalcini 1, 20157 Milan, Italy
reconstruction.

Plate-based protocols isolate cells by sorting into plates, where they are lysed (Figure 1). In the
popular Smart-seq methods, polyadenylated transcripts are captured via oligo-dT [1,2]. Alterna-
*Correspondence:
tive plate-based approaches, such as MATQ-seq, achieve amplification of non-polyadenylated [Link]@[Link] (C. Domínguez
transcripts with a random priming strategy [3]. Generally, these methods provide higher gene Conde).

Trends in Genetics, January 2024, Vol. 40, No. 1 [Link] 83

© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license ([Link]
 Trends in Genetics
OPEN ACCESS

detection sensitivity than droplet-based methods. In Smart-seq2, tagmentation generates Glossary

fragments covering full-length transcripts [1]. The main drawbacks of this approach are its limited scal- Allele-specific expression (ASE):
ability due to cost and logistics, as well as suboptimal quantification due to PCR amplification bias. quantification of gene expression at
allelic resolution.
Droplet-based protocols isolate cells by encapsulation in aqueous droplets in a water-in-oil emul- Alternative polyadenylation (APA):
molecular process that generates
sion. Cells are co-encapsulated with oligo-dT capture oligos and additional reagents that allow alternative transcripts of a gene by adding
the droplet-specific barcoding of transcripts (Figure 1). The main advantage of these methods polyA tails at different locations on
is their high cellular throughput allowing comprehensive characterization of tissues, including the 3′ end.
Alternative splicing (AS): molecular
rare cell types. Multiplexing strategies using cell hashing via either oligo-tagged antibodies [4] or
process that can generate multiple
lipids [5] have further increased the cellular throughput. Currently, 10X Genomics commercializes mature transcripts from a single gene by
two main modalities that enrich either the 3′ or the 5′ ends of transcripts after the fragmentation joining different exon combinations.
step [6]. Since single-cell methods start from very low amounts of RNA material requiring many Expression quantitative trait loci
(eQTL): genomic regions that have an
amplification cycles, PCR duplicates may confound quantification [7,8]. Unique molecular
impact on the expression levels of nearby
identifiers (UMIs) are incorporated in these methods and allow the deduplication of reads de- (cis-eQTL) or distal (trans-eQTL) genes.
rived from the same original RNA molecule and therefore improve quantification accuracy [9–12]. Haplotype switching: different
individuals presenting with the same
regulatory effect show opposite allelic
One of the main improvements in the new generation of plate-based methods, such as the
imbalances.
Smart-seq3 protocol [2], has been the incorporation of UMIs at the 5′ end of captured transcripts HLA field resolution: level of detail with
(Figure 1). Compared with Smart-seq2, this newer approach has further modifications in the which HLA alleles can be named
chemistry showing an even higher number of genes detected per cell [2]. Despite the low scalabil- following a standardized nomenclature.
The first field corresponds to the allele
ity of plate-based approaches, recent modifications have alleviated some limitations [13,14]. group, the second field to the HLA
protein, the third field to synonymous
A third single-cell sequencing approach that is gaining momentum among the community is variants in the coding region, and the
based on the combinatorial barcoding of cells by in situ retrotranscription and multiple rounds fourth field to variants in the noncoding
region.
of in situ ligation [15] (Figure 1). Combinatorial barcoding has pushed the boundaries of through- Human leukocyte antigen (HLA):
put by allowing profiling of hundreds of samples and millions of cells in an accessible setup. This group of polymorphic genes located on
approach also includes UMIs. In addition, the gene detection sensitivity is comparable with drop- chromosome 6 of the human genome
encoding surface proteins involved in
let-based methods, with an expected improvement in transcript coverage and amelioration of 3′
antigen presentation to adaptive
bias due to the combination of oligo-dT capture oligos with random hexamers [16]. Other advan- immune cells.
tages of this method are that it does not require specialized equipment and cells can be fixed and Killer-cell immunoglobulin-like
stored prior to RNA isolation, therefore offering flexibility compared with other multiplexing strat- receptors (KIRs): group of polymorphic
genes located on chromosome 19 of the
egies based on cell hashing [15]. Last, creative hybrid approaches that combine the above
human genome encoding surface
methods are pushing the boundaries even further. An example is the scifi-RNA-seq protocol, proteins expressed primarily on NK cells
which involves combinatorial barcoding followed by droplet-based encapsulation, exponentially and involved in innate immunity.
increasing the cellular profiling capacity [17]. Template-switch oligo (TSO):
oligonucleotide sequence that binds to a
string of untemplated C nucleotides
In conclusion, plate-based methods provide deeper information at the individual transcript level, added during reverse transcription to
enabling allele-specific analysis or isoform-level expression, as discussed later in further detail. synthesize the second cDNA strand.
While the largest studies with plate-based methods comprise a few thousand cells [18,19], Transcriptional bursting:
transcription resulting from stochastic
droplet-based and combinatorial indexing methods have enabled studies with up to ~2 million pulses with variable frequency and size.
cells [20,21] and ~4 million cells sequenced [22,23], respectively. Unique molecular identifier (UMI):
sequence added to RNA molecules as a
Targeted sequencing of 5′ ends: transcription start sites (TSSs) and V(D)J tag to track duplication of sequences
during PCR amplification.
repertoires V(D)J: germline gene segments in the
Transcription starts from the 5′ end of transcripts. An emerging application of droplet-based variable region of antigen receptor loci
methods enriching 5′ ends is the ability to accurately map TSS. In bulk RNA-seq, 5′ tag protocols that undergo somatic recombination.
X-chromosome inactivation (XCI):
have been shown to detect proximal promoters [24] and even distal cis-regulatory elements [25].
epigenetic silencing of one of the two X
This renders single-cell 5′ capture methods amenable to this type of analysis. A novel tool called chromosomes during the development
SCAFE [26] (Table 1) maps TSSs from scRNA-seq 5′ data sequenced in a specific configuration of mammalian organisms.
[27]. This tool can efficiently reduce the amount of miscalled TSS reads commonly caused by

84 Trends in Genetics, January 2024, Vol. 40, No. 1

Trends in Genetics
OPEN ACCESS

    

Trends in Genetics

Figure 1. Overview of single-cell RNA-seq approaches. Single-cell RNA-seq starts with the fundamental step of isolating single cells. Different methods are shown.
In 10X Genomics (left), single cells are separated by encapsulation in droplets using a water-in-oil emulsion. Smart-seq (center) involves flow-cytometry-based cell sorting
into plates. The combinatorial barcoding (right) strategy uses multiple rounds of indexing to generate unique labels for individual cells. Most common single-cell protocols
start by capturing the polyA tails of mRNA with an oligo-dT followed by retrotranscription. Specific reverse transcriptases add untemplated cytosines at the end of the newly
synthesized cDNA strand enabling the use of template-switch oligos (TSOs), which favors the second strand synthesis only after full reverse transcription (RT) of mRNA
molecules. This makes possible the capture of full-length transcripts in Smart-seq and alleviates 3′ bias. A key difference between the Smart-seq and tagged-end methods,
such as 3′ and 5′ capture protocols from 10X Genomics, lies in the targeted amplification of the tagged ends after fragmentation/tagmentation. In addition, Smart-seq3
incorporates unique molecular identifiers (UMIs) at the 5′ end of transcripts. Both 5′ UMI reads and internal non-UMI reads may be strategically combined to improve
transcript-level analysis. Combinatorial barcoding methods use both oligo-dT and random hexamers for transcript capture. The generation of unique cell barcodes
relies on an initial in situ retrotranscription step followed by several rounds of in situ ligation and posterior RT + template switching.

strand invasion arising from template-switching reactions [27]. This method represents a promis-
ing complementary approach to scATAC-seq to uncover the single-cell landscape of regulatory
regions.

The more extended use of single-cell methods enriching 5′ ends, however, has been the profiling of
the variable regions of antigen receptors. Antigen receptor loci include T cell receptors (TCRs) [alpha-
beta (ab) and gamma-delta (gd)] and B cell receptors (BCRs). All of these loci undergo somatic
recombination of V, D, and J gene segments, located at the 5′ end, and further diversification events
that ultimately lead to a huge sequence diversity that constitutes the antigen receptor repertoire
(Figure 2A). The first studies that performed simultaneous profiling of transcriptomes and antigen
receptors at the single-cell level were based on full-length technologies [28,29]. This was followed
by the development of droplet-based methods with higher cell profiling throughput that relied on
the capture of 5′ transcript ends, where the variable regions of antigen receptor genes lie [30].
These methods perform a targeted amplification of V(D)J loci [31], which can include TCRgd loci [32].

An interesting application of this framework is to use antigen receptor features to inform differentiation
trajectory interpretation. An example of this is provided in an in-depth analysis of B cell repertoires in
human tonsils [33]. Here the authors performed pseudotime trajectory analysis of B cell maturation.
This analysis based only on transcriptome data revealed a novel cell state that they named pre-
germinal-center B cells. BCR features, such as the levels of somatic hypermutation and the isotype
class, were concordant with a pre-germinal-center B cell state and were used as an orthogonal

Trends in Genetics, January 2024, Vol. 40, No. 1 85

 Trends in Genetics
OPEN ACCESS

Table 1. Key resources

Method Application Repository
SCAFE TSS analysis [Link]
scDALI ASE [Link]
Airpart ASE [Link]
DAESC ASE [Link]
SNPsplit ASE [Link]
SCALE ASE [Link]
scHLApers Polymorphic loci (HLA) [Link]
arcasHLA Polymorphic loci (HLA) [Link]
scHLAcount Polymorphic loci (HLA) [Link]
KIRid Polymorphic loci (KIR) [Link]
T1K Polymorphic loci (KIR/HLA) [Link]
SICILIAN Splicing analysis [Link]
SpliZ Splicing analysis [Link]

readout to interpret the identity of these cells. In a similar way, a study of developing T cells in the
human thymus used trajectory analysis to identify dynamically regulated genes [34]. In this case, pro-
ductive and unproductive TCR transcripts generated during T cell maturation were overlaid on the
gene expression trajectory to support its biological relevance [34] (Figure 2B). More recently, this
approach has been extended by the integration of V(D)J features in the trajectory inference step
[35]. This led to improvements in the identification of transcription factors (TFs) driving differentiation
into either CD4 or CD8 T cell lineages. These studies reflect the advantages of using V(D)J information
alongside gene expression for robust pseudotime trajectory analysis.

Mature T and B cells undergo clonal expansions after exposure to environmental or self-antigens.
Clonality analysis can therefore be applied to dissect adaptive immune responses across different
body sites [36,37] as well as under different conditions such as infection, autoimmunity, and
cancer [38–45] (Figure 2C). In a remarkable early example of this application, simultaneous
TCR and transcriptome analysis of breast cancer T cells revealed a spectrum of cellular activation
states within individual expanded clones [43]. Additionally, recent studies have revealed specific
properties of expanded clones that correlate with the tissue of origin, also providing evidence of
clonal sharing across cell states in the steady state [36,37].

The latest innovation in this area has been the integration of the actual antigen specificity of lympho-
cytes. Antigens are processed by antigen-presenting cells and loaded onto MHC molecules
encoded in the human leukocyte antigen (HLA) locus. Antigen-specific clonally expanded T
cells can be identified through DNA-barcoded MHC monomers/multimers coupled with scRNA-
seq (Figure 2D). This approach has enabled the tracking of severe acute respiratory syndrome
coronavirus 2 (SARS-CoV-2)-specific T cells in coronavirus disease 2019 (COVID-19) patients or
vaccinated individuals, offering insights into the in vivo dynamics of T cell activation and the effects
of repeated antigen exposure [38,46,47]. The addition of this layer of information in the single-cell
toolbox promises to open new vistas in our understanding of host–pathogen interactions.

Single-cell allele-specific expression (ASE)

Genes are transcribed either from one of the two parental alleles (monoallelic) or from both alleles
(biallelic) [48,49] (Figure 3A,B). Differential allelic expression can be mediated by epigenetic silenc-
ing [e.g., genomic imprinting, X-chromosome inactivation (XCI)] or by transcriptional

86 Trends in Genetics, January 2024, Vol. 40, No. 1

Trends in Genetics
OPEN ACCESS

(A) (B)

(C) (D)
P
D

F
C

Trends in Genetics

Figure 2. Single-cell profiling of antigen receptor variable sequences. (A) V(D)J recombination of the antigen receptor loci occurs at early stages of B and T
lymphocyte development in primary lymphoid organs (bone marrow and thymus). This process of combinatorial diversity between the V(D)J segments [light chains in
the B cell receptor (BCR) and alpha-gamma chains in the T cell receptor (TCR) do not contain germline D segments], is responsible for the generation of a wide
spectrum of surface lymphocyte receptors for the recognition of antigens. (B) Developmental trajectory inference can be applied to the study of the differentiation and
maturation of T cell lymphocytes by using gene expression patterns and TCR chain productivity. (C) Analysis of BCR/TCR repertoires can give us information about the
clonality of immune repertoires. Clonal relationships can be inferred from BCR sequence data using grouping and hamming distance methods. Clonotype abundance
can be used as a measure of clonal expansion between healthy and infection states and to characterize the expansion of gene expression signatures and clone
overlap among cell types and states. (D) Antigen specificity can be targeted in single-cell studies by using MHC barcoded monomers/multimers recognizing specific
foreign antigens.

modulation linked to cis-acting expression quantitative trait loci (eQTL). The detection of
single-nucleotide variants (SNVs) from RNA-seq data is instrumental for the identification of alleles
of origin in ASE studies. In single cells, methods capturing full-length transcripts (e.g., Smart-seq2/3)
have been preferred to accurately quantify ASE [50].

One of the first single-cell ASE studies combined Smart-seq2 with bulk genomic sequencing to
distinguish expression from maternal and paternal X chromosomes and detect XCI status across
tissues [51]. Through the strategic utilization of genomic phased information and aggregated full-
length scRNA-seq data across 940 single cells (to eliminate potential false negatives arising from
stochastic monoallelic expression), it is feasible to identify genes escaping XCI that are actively
expressed on both alleles, at single-cell resolution [51]. In addition, single-cell ASE could be lever-
aged to investigate the dynamics of XCI during development. For instance, by adopting the Smart-
seq2 approach, Cheng et al. [52] investigated the onset of XCI in 1724 single cells of an early

Trends in Genetics, January 2024, Vol. 40, No. 1 87

 Trends in Genetics
OPEN ACCESS

(A) (B)
M

G G



 

(C) (D)

Trends in Genetics

Figure 3. Single-cell transcriptomics beyond gene expression. (A) Schematic representations of parental chromosomes, intron–exon structure of gene locus, and
allele-specific RNA spliced products are shown. (B) Schematic description of allelic regulation scenarios across single cells: biallelic expression, when both alleles are
equally represented (top); imprinted expression, when one allele is epigenetically silenced (middle); and random monoallelic expression (RME) when monoallelic
expression is dynamic and stochastic across closely related cells (bottom). Schematic gene locus (left), allelic reads (center), and allele ratio distribution in single cells
(right) are represented for each scenario. (C) Expression analysis of human leukocyte antigen (HLA) and killer-cell immunoglobulin-like receptor (KIR) genes. High
sequence similarity among KIR and HLA genes (e.g., Chr6 HLA locus, on the top) and allelic diversity across individuals and cells (bottom) are described. (D) Intron–
exon structure of gene locus and alternative spliced RNA molecules are shown. Schematic representation of differential transcript usage across single cells in a
differentiation trajectory.

murine female embryo. They revealed that XCI exhibits heterogeneous rates of progression across
different germ layers (epiblast, extraembryonic ectoderm, and visceral endoderm) and reactivation
of genes that escape XCI is initiated during the XCI process. ASE analysis also helped in under-
standing the impact of cis-acting genetic variants on the expression levels of the long noncoding
RNA Xist (a master regulator of XCI) and in turn the efficiency of X-chromosome silencing [53].

Surprisingly, bulk RNA-seq studies revealed monoallelic expression for many autosomal genes,
which randomly varies across organs and tissues [54]. However, bulk methods were unable to
highlight the mechanistic nature of this random monoallelic expression (RME) and distinguish
whether it is clonally inherited (e.g., similar to XCI or genomic imprinting) or a transient phenome-
non. Thus, quantification of ASE in scRNA-seq studies could be helpful in disentangling this co-
nundrum. SNPsplit [55] (Table 1) is a tool developed for bulk RNA-seq to assign reads
overlapping known SNVs to the allele of origin, which can be used to quantify ASE in single
cells. Interestingly, ASE analysis in single cells revealed RME of autosomal genes (~17%

88 Trends in Genetics, January 2024, Vol. 40, No. 1

Trends in Genetics
OPEN ACCESS

mRNA, 26% of long noncoding genes) with high variability across closely related cells [56]. Analysis
of transcriptional kinetics in allelic scRNA-seq revealed that cell-type-specific gene expression is
modulated by variation in burst sizes and frequencies [57]. By using Smart-seq3, which demon-
strated improved sensitivity in transcriptional kinetics analysis, the authors demonstrated that
RME is largely due to transcriptional burst frequency, while clonal inheritance of RME is very rare
(~1%) [58]. This phenomenon, observed especially among low-expressed genes, is related to
the stochastic and dynamic expression of one allele. Despite being subject to debate [59],
transcriptional burst kinetics should be considered when approaching allelic imbalance studies in
single cells. For this purpose, specific statistical frameworks, such as SCALE, are required to
retrieve ASE from scRNA-seq, accounting for allele-specific transcriptional bursting [60,61].

ASE can be leveraged for eQTL analysis at the single-cell level. For instance, Cuomo et al. [18]
measured ASE across a pseudotime trajectory in genetically diverse human differentiating
iPSCs, which are heterozygous for eQTLs. Notably, the authors managed to profile 36 000
cells from 125 individuals using the Smart-seq2 technology. They linked eQTL to ASE in cell
types and states using GATK in ASEReadCounter mode [62], a tool originally developed for
bulk data, which calculates count reads per allele using heterozygous SNVs. Using the
same dataset, a statistical tool named scDALI [63] (Table 1) has been developed with the
goal of uncovering subtle differential regulatory effects through ASE in scRNA-seq data,
which was prevented by the use of tools developed for bulk methods. scDALI identifies allelic
imbalance across different cell types or states by using genotype data and pre-identified
eQTLs. Similarly, Airpart [64] (Table 1) works by partitioning data into groups of genes and
cells with similar allelic imbalance patterns. Neither tool is designed to handle the complexity
of studies including multiple individuals. DAESC [65], a recently published framework,
overcomes these limitations by using a mixture model that effectively handles complexities
such as haplotype switching.

Despite the great potential of ASE for eQTL analysis, improvements are required in the sensi-
tivity, transcript coverage, and scalability of the current methods to enable population-scale
studies.

Typing polymorphic loci at single-cell resolution

HLA and the killer-cell immunoglobulin-like receptor (KIR) are transmembrane receptors in-
volved in immune responses. As mentioned earlier, HLA genes encode MHC molecules involved
in antigen presentation. KIR genes modulate the cytotoxic activity of NK cells and T cells. Expres-
sion analysis of HLA and KIR genes across individuals and single cells is challenging but holds
potential to understand predisposition to autoimmunity and infections [66,67]. The main difficulty
arises from their high degree of allelic diversity (Figure 3C). They comprise ~200 and 15 genes,
respectively, and have numerous subtypes, further complicating their classification and study.
Tailored methods have been developed to type and analyze the expression of HLA and KIR
genes, which similar to the approaches for ASE, require accurate and sensitive SNV detection
from single-cell transcriptomics data.

Droplet-based methods enriching 5′ tagged ends can be used to type HLA genes as the 5′ region
of these genes is enriched for variability, and coverage across those sequences is typically high
[68]. The computational tool scHLAcount [69] is designed to evaluate HLA expression and can
be used after HLA typing with tools such as arcasHLA [70], as shown by Lau et al. [71]. However,
typing from scRNA-seq data is challenging and often results in partial success, especially when
trying to achieve more than one HLA field resolution [68]. A new pipeline called scHLApers
[72] improves the quantification of single-cell HLA expression by using a personalized HLA

Trends in Genetics, January 2024, Vol. 40, No. 1 89

 Trends in Genetics
OPEN ACCESS

reference. This approach, inspired by bulk methods, improves on traditional methods by utilizing
matched genotypes to provide more accurate typing of HLA [72].

Typing of KIRs is notoriously complicated as they harbor high sequence similarity. Vento-Tormo
et al. [66] developed KIRid (Table 1), a method that uses donor-specific reference alleles to map
single-cell reads generated with Smart-seq2. The authors quantified KIR expression in individual
cells and defined three major decidual natural killer (NK) cell states at the maternal–fetal interface.
Also, they predicted the specific immunomodulatory functions of those NK cells in the first trimes-
ter of pregnancy, with potential implications in pregnancy outcomes. T1K [73] (Table 1) is a recent
tool developed for KIR genes, also applicable to HLA, which calculates allele abundance by simul-
taneously aligning candidate reads against KIR reference alleles extracted from the Immuno Poly-
morphism Database (IPD). T1K applies a Poisson model for quality scoring to select a maximum
of two alleles per gene. Importantly, T1K also calculates the ASE, detects new SNVs, and sup-
ports Smart-seq3 data.

Overall, scRNA-seq data hold the potential to inform on the complexity of polymorphic genes, al-
beit with certain limitations. Despite the development of these recent tools, achieving deep reso-
lution is still challenging and new creative strategies are further needed in the field.

From gene-level to transcript-level quantification of expression

Differential use of TSSs, AS and alternative polyadenylation (APA) can generate multiple tran-
scripts from the same gene locus. These different transcripts can modulate cell differentiation and
function (Figure 3D). However, most scRNA-seq studies focus on gene-level quantification. This
scarcity of transcript-level analysis is related to the lack of robust techniques for the accurate iden-
tification and quantification of individual transcripts in single cells.

In this section, we discuss methods based on short-read and long-read sequencing. Systematic
assessments of short-read scRNA-seq concluded that robust isoform-level analysis is precluded
by low transcript coverage per cell [74,75]. Nonetheless, methods that do not attempt to recon-
struct isoforms but perform splice junction quantification have opened the possibility of inferring
transcript-specific expression. For example, SICILIAN [76] (Table 1) infers true-positive splice
junctions retrieved by splice-aware aligners. SICILIAN splice junction quantification can then be
used to calculate differential splicing events with the SpliZ score [77], which accounts for the lim-
itations of single-cell methods. The Tabula sapiens study, covering multiple human tissues and
cell lineages, used SICILIAN in combination with SpliZ, revealing thousands of cell-type-specific
splice variants [78]. This was achieved using both 3′-end droplet-based data and Smart-seq2
data; however, the latter yielded an approximately sevenfold greater number of genes with com-
putable SpliZ scores [78].

In comparison with Smart-seq2, the Smart-seq3 protocol has further improved quantification
with the incorporation of UMIs and isoform reconstruction by combining UMI reads and internal
reads to identify transcripts with salmon [2,79]. This makes Smart-seq3 the currently most sen-
sitive short-read sequencing protocol to detect cell-state-specific differential transcript usage.
Even if some degree of transcript-level analysis is possible from droplet-based and plate-based
methods, there remain high levels of dropout and quantification uncertainty that call for further
methodological developments.

Long-read sequencing technologies such as Pacific Biosciences (PacBio) and Oxford Nanopore
Technologies (ONT) represent a paradigm shift with regard to the characterization of nucleic acids
and, in particular, full transcriptomes. One of the unique advantages of long reads is the possibility

90 Trends in Genetics, January 2024, Vol. 40, No. 1

Trends in Genetics
OPEN ACCESS

of direct profiling of full isoforms derived from co-occurring distal regulatory events. To get a sense of Outstanding questions
the leap that these technologies represent, if we compare the number of reconstructed RNA mole- Can we increase the throughput of full-
cules by the previously mentioned Smart-seq3 protocol sequenced with Illumina (short read) versus length methods to obtain high tran-
script coverage at scale?
PacBio, double the number of full-length transcripts were detected with long-read technology [2].
Can we infer the mechanisms underlying
Until recently, the cell throughput of scRNA-seq methods compatible with long-read sequencing allelic regulation at single-cell resolution?
remained limited. However, in the past 5 years we have seen numerous examples and creative so-
Can long-read sequencing improve
lutions combining high-throughput droplet-based cell encapsulation with long-read sequencing.
the detection of alternatively spliced
One of the initial challenges when using ONT datasets has been the low accuracy of individual transcripts?
base calls. Gupta et al. analyzed isoform expression in mouse cerebellum cells by combining
10X barcoded cDNA molecules with short-read (Illumina) and long-read (both ONT and PacBio) Can we achieve accurate detection of
polyA tails at single-cell resolution?
sequencing [80]. Lebrigand et al. used a similar approach combining 10X barcoded cDNAs se-
quenced with both ONT and Illumina and designed an algorithm to use the Illumina data to improve
UMI and cell barcode identification by ONT [81]. Further variations of these methods are being im-
plemented to increase sequencing throughput. The MAS-seq approach based on PacBio, which
generates composite sequencing libraries by concatenating cDNA fragments, has shown un-
matched sensitivity and has become the first commercial solution of this kind [82]. In addition,
recent developments in ONT library construction [83,84] and ONT chemistry have improved
sequence accuracy, leading to algorithms that no longer rely on Illumina data to maximize the
recovery of UMIs and cell barcodes. Another alternative approach called LR-Split-seq used com-
binatorial barcoding instead of droplet-based methods followed by ONT sequencing and may
represent the most cost-effective option [85]. Importantly, all of these methods require rigorous,
artifact-aware transcript reconstruction and annotation tools, such as IsoQuant [86].

Overall, these pioneer studies support the critical role of gene regulation at the level of individual tran-
scripts in development and other dynamic cellular processes. Further improvements in the cost effec-
tiveness and logistics of these methods are needed to unlock their full potential in the future.

Concluding remarks
The landscape of single-cell transcriptomics methods is evolving rapidly. On the one hand, while
droplet-based methods have prevailed in large-scale studies, alternative high-throughput options
such as combinatorial barcoding are coming onto the scene. The next bottlenecks to overcome
in this type of study will be to reduce sequencing costs and design computational frameworks to
handle large datasets. On the other hand, lower-throughput plate-based methods remain the
most robust option enabling the capture of full-length transcripts, with unique applications such
as ASE and transcript reconstruction. As the single-cell analysis toolbox continues to expand, in-
cluding multimodal [87,88] and long-read applications, the well-established techniques are being
adapted for population-level studies to dissect interindividual differences representing a key pros-
pect in this field of research [89,90] (see Outstanding questions).

Acknowledgments
We acknowledge Tatjana Hirschmugl for graphical images.

Declaration of interests
The authors have declared no competing interest.

References
1. Picelli, S. et al. (2013) Smart-seq2 for sensitive full-length tran- 3. Sheng, K. et al. (2017) Effective detection of variation in single-cell
scriptome profiling in single cells. Nat. Methods 10, 1096–1098 transcriptomes using MATQ-seq. Nat. Methods 14, 267–270
2. Hagemann-Jensen, M. et al. (2020) Single-cell RNA counting at 4. Stoeckius, M. et al. (2018) Cell hashing with barcoded antibodies
allele and isoform resolution using Smart-seq3. Nat. Biotechnol. enables multiplexing and doublet detection for single cell geno-
38, 708–714 mics. Genome Biol. 19, 224

Trends in Genetics, January 2024, Vol. 40, No. 1 91

 Trends in Genetics
OPEN ACCESS

5. McGinnis, C.S. et al. (2019) MULTI-seq: sample multiplexing for 33. King, H.W. et al. (2021) Single-cell analysis of human B cell mat-
single-cell RNA sequencing using lipid-tagged indices. Nat. uration predicts how antibody class switching shapes selection
Methods 16, 619–626 dynamics. Sci. Immunol. 6, eabe6291
6. Zheng, G.X.Y. et al. (2017) Massively parallel digital transcrip- 34. Park, J.-E. et al. (2020) A cell atlas of human thymic development
tional profiling of single cells. Nat. Commun. 8, 14049 defines T cell repertoire formation. Science 367, eaay3224
7. Kivioja, T. et al. (2011) Counting absolute numbers of molecules 35. Suo, C. et al. (2023) Dandelion uses the single-cell adaptive im-
using unique molecular identifiers. Nat. Methods 9, 72–74 mune receptor repertoire to explore lymphocyte developmental
8. Fu, G.K. et al. (2011) Counting individual DNA molecules by the origins. Nat. Biotechnol. Published online April 13, 2023.
stochastic attachment of diverse labels. Proc. Natl. Acad. Sci. [Link]
U. S. A. 108, 9026–9031 36. Poon, M.M.L. et al. (2023) Tissue adaptation and clonal segrega-
9. Jaitin, D.A. et al. (2014) Massively parallel single-cell RNA-seq for tion of human memory T cells in barrier sites. Nat. Immunol. 24,
marker-free decomposition of tissues into cell types. Science 309–319
343, 776–779 37. Domínguez Conde, C. et al. (2022) Cross-tissue immune cell
10. Macosko, E.Z. et al. (2015) Highly parallel genome-wide expres- analysis reveals tissue-specific features in humans. Science
sion profiling of individual cells using nanoliter droplets. Cell 161, 376, eabl5197
1202–1214 38. Minervina, A.A. et al. (2022) SARS-CoV-2 antigen exposure his-
11. Hashimshony, T. et al. (2016) CEL-Seq2: sensitive highly- tory shapes phenotypes and specificity of memory CD8 T cells.
multiplexed single-cell RNA-seq. Genome Biol. 17, 77 Nat. Immunol. 23, 781–790
12. Jovic, D. et al. (2022) Single-cell RNA sequencing technologies 39. Yoshida, M. et al. (2022) Local and systemic responses to
and applications: a brief overview. Clin. Transl. Med. 12, e694 SARS-CoV-2 infection in children and adults. Nature 602,
13. Hahaut, V. et al. (2022) Fast and highly sensitive full-length single- 321–327
cell RNA sequencing using FLASH-seq. Nat. Biotechnol. 40, 40. Ramesh, A. et al. (2020) A pathogenic and clonally expanded B
1447–1451 cell transcriptome in active multiple sclerosis. Proc. Natl. Acad.
14. Hagemann-Jensen, M. et al. (2022) Scalable single-cell RNA se- Sci. U. S. A. 117, 22932–22943
quencing from full transcripts with Smart-seq3xpress. Nat. 41. Penkava, F. et al. (2020) Single-cell sequencing reveals clonal ex-
Biotechnol. 40, 1452–1457 pansions of pro-inflammatory synovial CD8 T cells expressing
15. Rosenberg, A.B. et al. (2018) Single-cell profiling of the developing tissue-homing receptors in psoriatic arthritis. Nat. Commun. 11,
mouse brain and spinal cord with split-pool barcoding. Science 4767
360, 176–182 42. Argyriou, A. et al. (2022) Single cell sequencing identifies clonally
16. Tran, V. et al. (2022) High sensitivity single cell RNA sequencing expanded synovial CD4+ TPH cells expressing GPR56 in rheu-
with split pool barcoding. bioRxiv Published online August 27, matoid arthritis. Nat. Commun. 13, 4046
2022. [Link] 43. Azizi, E. et al. (2018) Single-cell map of diverse immune pheno-
17. Datlinger, P. et al. (2021) Ultra-high-throughput single-cell RNA types in the breast tumor microenvironment. Cell 174,
sequencing and perturbation screening with combinatorial fluidic 1293–1308.e36
indexing. Nat. Methods 18, 635–642 44. Guo, X. et al. (2018) Global characterization of T cells in non-
18. Cuomo, A.S.E. et al. (2020) Single-cell RNA-sequencing of differ- small-cell lung cancer by single-cell sequencing. Nat. Med. 24,
entiating iPS cells reveals dynamic genetic effects on gene ex- 978–985
pression. Nat. Commun. 11, 810 45. Gao, S. et al. (2022) Single-cell RNA sequencing coupled to TCR
19. Sade-Feldman, M. et al. (2019) Defining T cell states associated profiling of large granular lymphocyte leukemia T cells. Nat.
with response to checkpoint immunotherapy in melanoma. Cell Commun. 13, 1982
176, 404 46. Fischer, D.S. et al. (2021) Single-cell RNA sequencing reveals
20. Sinha, S. et al. (2022) Dexamethasone modulates immature neu- ex vivo signatures of SARS-CoV-2-reactive T cells through “re-
trophils and interferon programming in severe COVID-19. Nat. verse phenotyping”. Nat. Commun. 12, 4515
Med. 28, 201–211 47. Lindeboom, R.G.H. et al. (2023) Human SARS-CoV-2 challenge
21. Svensson, V. et al. (2020) A curated database reveals trends in resolves local and systemic response dynamics. medRxiv Pub-
single-cell transcriptomics. Database (Oxford) 2020, baaa073 lished online April 25, 2023. [Link]
22. Cao, J. et al. (2019) The single-cell transcriptional landscape of 13.23288227
mammalian organogenesis. Nature 566, 496–502 48. Ferguson-Smith, A.C. (2011) Genomic imprinting: the emer-
23. Cao, J. et al. (2020) A human cell atlas of fetal gene expression. gence of an epigenetic paradigm. Nat. Rev. Genet. 12, 565–575
Science 370, eaba7721 49. Reinius, B. and Sandberg, R. (2015) Random monoallelic ex-
24. Vitezic, M. et al. (2010) Building promoter aware transcriptional pression of autosomal genes: stochastic transcription and
regulatory networks using siRNA perturbation and deepCAGE. allele-level regulation. Nat. Rev. Genet. 16, 653–664
Nucleic Acids Res. 38, 8141–8148 50. Probst, V. et al. (2022) Benchmarking full-length transcript single
25. Hirabayashi, S. et al. (2019) NET-CAGE characterizes the dy- cell mRNA sequencing protocols. BMC Genomics 23, 860
namics and topology of human transcribed cis-regulatory ele- 51. Tukiainen, T. et al. (2017) Landscape of X chromosome inactiva-
ments. Nat. Genet. 51, 1369–1379 tion across human tissues. Nature 550, 244–248
26. Moody, J. et al. (2022) SCAFE: a software suite for analysis of 52. Cheng, S. et al. (2019) Single-cell RNA-seq reveals cellular hetero-
transcribed cis-regulatory elements in single cells. Bioinformatics geneity of pluripotency transition and X chromosome dynamics
38, 5126–5128 during early mouse development. Cell Rep. 26, 2593–2607.e3
27. Tang, D.T.P. et al. (2013) Suppression of artifacts and barcode 53. Pacini, G. et al. (2021) Integrated analysis of Xist upregulation
bias in high-throughput transcriptome analyses utilizing template and X-chromosome inactivation with single-cell and single-allele
switching. Nucleic Acids Res. 41, e44 resolution. Nat. Commun. 12, 3638
28. Redmond, D. et al. (2016) Single-cell TCRseq: paired recovery of 54. Kravitz, S.N. et al. (2023) Random allelic expression in the adult
entire T-cell alpha and beta chain transcripts in T-cell receptors human body. Cell Rep. 42, 111945
from single-cell RNAseq. Genome Med. 8, 80 55. Krueger, F. and Andrews, S.R. (2016) SNPsplit: allele-specific
29. Stubbington, M.J.T. et al. (2016) T cell fate and clonality inference splitting of alignments between genomes with known SNP geno-
from single-cell transcriptomes. Nat. Methods 13, 329–332 types. F1000Res 5, 1479
30. Pai, J.A. and Satpathy, A.T. (2021) High-throughput and single- 56. Deng, Q. et al. (2014) Single-cell RNA-seq reveals dynamic, ran-
cell T cell receptor sequencing technologies. Nat. Methods 18, dom monoallelic gene expression in mammalian cells. Science
881–892 343, 193–196
31. Efremova, M. et al. (2020) Immunology in the era of single-cell 57. Larsson, A.J.M. et al. (2019) Genomic encoding of transcriptional
technologies. Annu. Rev. Immunol. 38, 727–757 burst kinetics. Nature 565, 251–254
32. Mimitou, E.P. et al. (2019) Multiplexed detection of proteins, tran- 58. Larsson, A.J.M. et al. (2021) Transcriptional bursts explain auto-
scriptomes, clonotypes and CRISPR perturbations in single cells. somal random monoallelic expression and affect allelic imbal-
Nat. Methods 16, 409–412 ance. PLoS Comput. Biol. 17, e1008772

92 Trends in Genetics, January 2024, Vol. 40, No. 1

Trends in Genetics
OPEN ACCESS

59. Vigneau, S. et al. (2018) High prevalence of clonal monoallelic ex- 75. Buen Abad Najar, C.F. et al. (2020) Coverage-dependent bias cre-
pression. Nat. Genet. 50, 1198–1199 ates the appearance of binary splicing in single cells. Elife 9, e54603
60. Kim, J.K. and Marioni, J.C. (2013) Inferring the kinetics of sto- 76. Dehghannasiri, R. et al. (2021) Specific splice junction detection
chastic gene expression from single-cell RNA-sequencing data. in single cells with SICILIAN. Genome Biol. 22, 219
Genome Biol. 14, R7 77. Olivieri, J.E. et al. (2022) The SpliZ generalizes “percent spliced
61. Jiang, Y. et al. (2017) SCALE: modeling allele-specific gene ex- in” to reveal regulated splicing at single-cell resolution. Nat.
pression by single-cell RNA sequencing. Genome Biol. 18, 74 Methods 19, 307–310
62. Van der Auwera, G.A. and O’Connor, B.D. (2020) Genomics in 78. Tabula Sapiens Consortium et al. (2022) The Tabula Sapiens: a
the cloud: using Docker, GATK, and WDL in Terra, O’Reilly multiple-organ, single-cell transcriptomic atlas of humans. Sci-
Media ence 376, eabl4896
63. Heinen, T. et al. (2022) scDALI: modeling allelic heterogeneity in 79. Patro, R. et al. (2017) Salmon provides fast and bias-aware quan-
single cells reveals context-specific genetic regulation. Genome tification of transcript expression. Nat. Methods 14, 417–419
Biol. 23, 8 80. Gupta, I. et al. (2018) Single-cell isoform RNA sequencing char-
64. Mu, W. et al. (2022) Airpart: interpretable statistical models for acterizes isoforms in thousands of cerebellar cells. Nat.
analyzing allelic imbalance in single-cell datasets. Bioinformatics Biotechnol. Published online October 15, 2018. [Link]
38, 2773–2780 10.1038/nbt.4259
65. Qi, G. et al. (2023) Single-cell allele-specific expression analysis 81. Lebrigand, K. et al. (2020) High throughput error corrected nanopore
reveals dynamic and cell-type-specific regulatory effects. Nat. single cell transcriptome sequencing. Nat. Commun. 11, 4025
Commun. 14, 6317 82. Al’Khafaji, A.M. et al. (2023) High-throughput RNA isoform se-
66. Vento-Tormo, R. et al. (2018) Single-cell reconstruction of the quencing using programmed cDNA concatenation. Nat.
early maternal–fetal interface in humans. Nature 563, 347–353 Biotechnol. Published online June 8, 2023. [Link]
67. Gutierrez-Arcelus, M. et al. (2020) Allele-specific expression 1038/s41587-023-01815-7
changes dynamically during T cell activation in HLA and other au- 83. Cole, C. et al. (2020) Complete characterization of the human im-
toimmune loci. Nat. Genet. 52, 247–253 mune cell transcriptome using accurate full-length cDNA se-
68. Solomon, B.D. et al. (2023) Prediction of HLA genotypes from quencing. Genome Res. 30, 589–601
single-cell transcriptome data. Front. Immunol. 14, 1146826 84. Volden, R. et al. (2018) Improving nanopore read accuracy with the
69. Darby, C.A. et al. (2020) scHLAcount: allele-specific HLA expres- R2C2 method enables the sequencing of highly multiplexed full-
sion from single-cell gene expression data. Bioinformatics 36, length single-cell cDNA. Proc. Natl. Acad. Sci. U. S. A. 115,
3905–3906 9726–9731
70. Orenbuch, R. et al. (2020) arcasHLA: high-resolution HLA typing 85. Rebboah, E. et al. (2021) Mapping and modeling the genomic
from RNAseq. Bioinformatics 36, 33–40 basis of differential RNA isoform expression at single-cell resolu-
71. Lau, D. et al. (2022) Integration of tumor extrinsic and intrinsic tion with LR-Split-seq. Genome Biol. 22, 286
features associates with immunotherapy response in non-small 86. Prjibelski, A.D. et al. (2023) Accurate isoform discovery with
cell lung cancer. Nat. Commun. 13, 4053 IsoQuant using long reads. Nat. Biotechnol. 41, 915–918
72. Kang, J.B. et al. (2023) Mapping the dynamic genetic regulatory 87. Baysoy, A. et al. (2023) The technological landscape and applications
architecture of genes at single-cell resolution. medRxiv Published of single-cell multi-omics. Nat. Rev. Mol. Cell Biol. 24, 695–713
online March 20, 2023. [Link] 88. Heumos, L. et al. (2023) Best practices for single-cell analysis
23287257 across modalities. Nat. Rev. Genet. 24, 550–572
73. Song, L. et al. (2023) Efficient and accurate KIR and HLA 89. Cuomo, A.S.E. et al. (2023) Single-cell genomics meets human
genotyping with massively parallel sequencing data. Genome genetics. Nat. Rev. Genet. 24, 535–549
Res. 33, 923–931 90. Kang, J.B. et al. (2023) Methods and insights from single-cell ex-
74. Westoby, J. et al. (2020) Obstacles to detecting isoforms using pression quantitative trait loci. Annu. Rev. Genomics Hum.
full-length scRNA-seq data. Genome Biol. 21, 74 Genet. 24, 277–303

Trends in Genetics, January 2024, Vol. 40, No. 1 93