nature immunology

Review article https://doi.org/10.1038/s41590-023-01678-9

Decoding the tumor microenvironment with

spatial technologies

Received: 18 August 2023 Logan A. Walsh 1,2

& Daniela F. Quail 1,3,4

Accepted: 10 October 2023

Published online: 27 November 2023 Visualization of the cellular heterogeneity and spatial architecture of the
tumor microenvironment (TME) is becoming increasingly important to
Check for updates
understand mechanisms of disease progression and therapeutic response.
This is particularly relevant in the era of cancer immunotherapy, in which
the contexture of immune cell positioning within the tumor landscape
has been proven to affect efficacy. Although single-cell technologies have
mostly replaced conventional approaches to analyze specific cellular
subsets within tumors, those that integrate a spatial dimension are now
on the rise. In this Review, we assess the strengths and limitations of
emerging spatial technologies with a focus on their applications in tumor
immunology, as well as forthcoming opportunities for artificial intelligence
(AI) and the value of integrating multiomics datasets to achieve a holistic
picture of the TME.

The TME is a complex and dynamic ecosystem composed of a diverse that overcome these challenges would be highly valuable to understand
array of cell types and extracellular components. Understanding the functional heterogeneity within the tumor niche.
cellular heterogeneity and spatial organization of the TME is crucial Accordingly, integration of spatial resolution into single-cell
for unraveling the underlying mechanisms that drive tumor progres- datasets has emerged as a powerful approach to unravel the complex
sion and therapeutic responses. Traditionally, analysis of the TME has architecture of the TME, providing invaluable context for interpreting
focused on bulk sequencing and/or flow-based approaches that rely interactions between tumor cells and the immune system. Notably,
on dissociated tissues, obscuring the intricate cellular relationships recent advances in spatial transcriptomics, spatial proteomics and spa-
and regional heterogeneity that define the tumor landscape. In addi- tial metabolomics, and the combination thereof, have all dramatically
tion, many of these techniques require live samples, creating major shifted our understanding of tissue homeostasis and cancer-associated
challenges in the time required to generate complete patient datasets inflammation, providing new insights into cell–cell interactions, mul-
and minimize batch effects. Although multiplex histology overcomes ticellular units and broader communication networks within the TME
some of these limitations, commonly used protocols enable only a few (Fig. 1). With the parallel emergence of these datasets and a rapidly
canonical markers to be imaged at one time, thus restricting unbiased expanding field in AI, we are at the forefront of an unprecedented
discoveries in favor of targeted hypotheses. As a consequence, the era in biomedical research, and we have an opportunity to shape its
advent of single-cell technologies, including those compatible with trajectory to address challenges that could profoundly affect patients.
fixed tissues, has revolutionized our ability to unravel the cellular archi- Here, we provide an overview of contemporary advances in spa-
tecture of tumors, leading to a more comprehensive understanding of tially resolved single-cell technologies and highlight how spatial ori-
tumor biology. Nevertheless, several important limitations remain. For entation has a pivotal role in understanding cancer progression and
example, cell types that are rare or that cannot withstand dissociation therapeutic responses. We place special emphasis on advances in
protocols, such as neutrophils, are often under-represented. Moreover, tumor immunology and the TME, as the general timeline of spatial
it is difficult to effectively evaluate phenotypes that rely on cellular omic innovation has been reviewed elegantly elsewhere1,2. Within
organization, including tertiary lymphoid structures (TLS). Thus, tools each section, we provide examples of how multiple tiers of spatial

1
Rosalind and Morris Goodman Cancer Institute, McGill University, Montreal, Quebec, Canada. 2Department of Human Genetics, Faculty of Medicine,
McGill University, Montreal, Quebec, Canada. 3Department of Physiology, Faculty of Medicine, McGill University, Montreal, Quebec, Canada.
4
Department of Medicine, Division of Experimental Medicine, McGill University, Montreal, Quebec, Canada. e-mail: [email protected];
[email protected]

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1982

Review article https://doi.org/10.1038/s41590-023-01678-9

Transcriptomics Proteomics Epigenomics Metabolomics

Intensity
m/z
Metabolic IMC/MIBI TOF-SIMS
smFISH Cyclic IHC/IF MALDI OrbiSIMS
seqFISH FISSEQ CODEX Spatial ATAC–RNA-seq DESI metaFISH
MERFISH BaSISS Mass cytometry Spatial CUT&Tag–RNA-seq AP-MALDI TOF-SIMS–IMC
seqFISH+ Slide-TCR-seq Spectral fluorescence MERFISH–CUT&Tag t-MALDI-2 MALDI–IMC

Data integration 3D spatial omics 4D physiology

Time

SpatialDE STARmap CODA ZipSeq intMEMOIR

MISTy ExSeq WildDISCO Live-seq Zombie
GLUE 3D IMC DISCO–MS MEMOIR Circadian tools

Fig. 1 | Using multiomic technologies to decode tumor immune dynamics. Moreover, spatiotemporal assessment of tumor immune responses over time
Single-cell spatial omic tools have been developed to analyze tumor sections at the single-cell level is being integrated into these datasets, either through
and tissues in two dimension, including single-cell spatial transcriptomics, time-controlled barcoding or sampling techniques. In the figure, examples of
proteomics, epigenomics and metabolomics. Many of these can be combined specific technologies under each category are listed (the metabolomic image
to integrate multiple tiers of data within one sample. Emerging technologies was sourced from ref. 152). AP-MALDI, atmospheric pressure MALDI; DESI,
that expand these tools into the third dimension, including entire tissue desorption electrospray ionization; smFISH, single-molecule FISH; t-MALDI-2,
volumes or even whole organisms, are now being developed by rebuilding transmission-mode MALDI-2; 4D, four dimensional.
tissues via serial section alignment or through tissue-clearing techniques.

omics datasets can be integrated to enable comprehensive analysis RNA species were probed. In 2015, multiplex error-robust FISH (MER-
across tumor landscapes and how these can extend into the third or FISH) was developed as a highly multiplexed tool to profile copy number
fourth dimensions. Leveraging the power of spatial technologies and and spatial location of thousands of RNA species within individual
integrating them with AI and multi-tiered datasets offer unprecedented cells, shown first in human fibroblasts6. Addressing the limitations of
insights into the cellular architecture of the TME, paving the way for other sequential hybridization techniques, MERFISH leveraged com-
the development of more effective immunotherapies and personalized binatorial labeling of target sequences (which bind RNA) and readout
cancer treatments. sequences (which bind fluorescent probes) with successive rounds
of hybridization and imaging to decode RNA and apply error-robust
From cell suspensions to spatial transcriptomics encoding schemes. It has since been applied to address several ques-
Multiplex mRNA detection in situ primarily began around 2008 with tions in cancer immunology; for example, MERFISH was used in human
the advent of single-molecule fluorescence in situ hybridization (FISH), glioblastoma to measure 135 transcripts in macrophages and tumor
which showed simultaneous detection of three mRNA species in sin- cells of the mesenchymal subtype, where macrophage accumulation is
gle cells from tissues and cell lines from multiple species3. In 2014, a hallmark feature7. This technique has also been employed to measure
sequential FISH (seqFISH) expanded the multiplexing capacity of 479 transcripts within multicellular immunity ‘hubs’ and investigate
single-molecule FISH by integrating multiple rounds of sequential their correlation with the effectiveness of immune checkpoint blockade
hybridization and imaging4. One of the first major applications for efficacy in human lung cancer8. A benefit of MERFISH is that it provided
seqFISH included profiling 249 genes in over 16,000 cells within the a higher level of multiplexing than other multiplex FISH techniques;
mouse hippocampus, revealing a detailed atlas of cellular heterogeneity however, it still shared the potential for low signal-to-noise ratios due to
at unprecedented resolution5. The main limitation of the first iteration high optical density and crowding of transcripts within cells, especially
of seqFISH was scalability, as it accumulated substantial error as more for low-abundance transcripts. In 2019, a new iteration of seqFISH was

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1983

Review article https://doi.org/10.1038/s41590-023-01678-9

developed, seqFISH+, which achieved subcellular detection of >10,000 data with protein detection have been developed33–36; most have been
genes in single cells with a low error rate and a high subdiffraction limit9. limited to measurements of ~100 proteins or less, with the exception
This technique overcame the challenge of transcript crowding and of spatial cellular indexing of transcriptomes and epitopes (CITE)-seq,
provided superior resolution of multiplex RNA-profiling approaches. which includes 200–300 antibody-derived DNA tags combined with
Complementing FISH-based technologies, in situ sequencing (ISS) whole-transcriptome profiling37. Such multiomic platforms have
involves the detection and quantification of RNA molecules directly uncovered a deeper understanding of tumor–immune interactions
in tissues and combines the advantages of next-generation sequenc- in the context of broader cellular microenvironmental networks and
ing (NGS) with spatial information provided by microscopy. The first physiological states, for example, spatially restricted hepatic mac-
major reports of ISS were between 2013 and 201510–12, in which fluo- rophage populations in the context of obesity36.
rescent in situ RNA sequencing (FISSEQ) established the feasibility of Finally, spatial sequencing efforts have also been used to under-
sequencing complementary DNA from ~8,000 genes within single cells stand the landscape of B and T cell receptor (TCR) clones within tis-
and tissues10. More recently, whole-genome sequencing was used to sues. For example, in 2019, multi-region TCR-seq was performed on
inform base-specific ISS (BaSISS) and interrogate spatially resolved early-stage untreated lung tumors using 220 tumor regions from 72
phenotypes across clonally mosaic multifocal breast tumors using patients from the Tracking Cancer Evolution through Therapy (TRAC-
a 153-gene panel13. This technique was particularly valuable in map- ERx) project38. Both regional and ubiquitous TCRs were found within
ping clonal growth patterns of metastases within distinct immune tumors; these were spatially heterogeneous between patients and cor-
microenvironments of the lymph node, thus providing a platform to related with spatial mutational heterogeneity across tumor regions38,
study the interplay between cancer genomics and the tumor immune supporting the notion that the intratumoral TCR repertoire is shaped
microenvironment, both pivotal factors affecting clinical outcomes. by the neoantigen landscape. However, this approach involved sam-
As an alternative to fluorescent probes used for ISS, NGS-based tech- pling different tumor regions rather than true spatial sequencing
niques leverage spatially barcoded capture of RNA molecules from in situ. Shortly thereafter, in 2022, slide-TCR-seq was developed to
tissue sections, followed by high-throughput transcriptomic sequenc- sequence whole transcriptomes and TCRs across entire tumor sec-
ing14–18. These approaches have even been combined with spatial tions at a resolution of 10 μm39. Slide-TCR-seq enabled analysis of T cell
epigenetic profiling during brain development via spatial assay for location, clonotype and gene expression and was applied to investigate
transposase-accessible chromatin (ATAC)–RNA-seq or spatial cleav- immune TME interactions in renal cell carcinoma and melanoma. T cell
age under targets and tagmentation (CUT&Tag)–RNA-seq19 (of note, clones were spatially variable in their gene expression profiles, and
MERFISH can also be combined with CUT&Tag approaches for spatial cellular neighbors of specific T cell clones, including monocytes and
single-cell epigenomic profiling20). tumor cells, exhibited a unique repertoire of differentially expressed
By the end of 2019, two prominent commercial platforms emerged genes39. Moreover, greater clonal expansion was detected for T cells
at the forefront of NGS-based spatial profiling in cancer research, within tumor regions than for those within TLS, and these were more
Visium Spatial Gene Expression (10x Genomics)14,15 and GeoMx Digi- likely to be CD8+ rather than CD4+. This finding suggests that T cells
tal Spatial Profiler (NanoString)21,22. While Visium is based on spatial within tumor-associated TLS can do more than simply support the
transcriptomic arrays and GeoMx relies on microdissection, both offer expansion of tumor antigen-specific T cells; for example, they may
high gene expression quantification compared to other imaging-based support the development of B cells. As a potentially more accessible
methods (for example, MERFISH). In recent years, these NGS-based approach, the Visium platform can be modified to enable spatial mRNA
platforms have gained substantial interest from the cancer research and TCR sequencing in parallel40. In 2022, this approach was used to
community, particularly owing to user-friendly technical and analytic characterize infiltrating T cells in human brain metastases, where it
pipelines that are accessible even to laboratories without specialized was discovered that distinct T cell clones occupy unique niches within
expertise. Although these platforms do not capture true single-cell the TME41. Together, these technologies begin to paint a picture of TCR
resolution, they provide an opportunity to learn about disease biology clonal diversity across the tumor landscape and provide insight into
at scale across large patient cohorts and have now been broadly applied tumor antigen-specific T cell responses.
to understand tumor immunology in human breast23, kidney24, lung25,
brain26, skin27, pancreas28, ovarian29 and many other cancers, and have The rise of spatial proteomics in histology
even been adapted to understand the intratumoral microbiome30. Reso- One of the central limitations of transcriptomic approaches is a reli-
lution is indeed substantially lower than in situ hybridization-based ance on freshly frozen tissues, thus creating challenges for translation
approaches, often including clusters of ~10–100 cells compared to to large-scale human studies for which FFPE tissues are more readily
subcellular resolution, thus creating challenges for analyzing small available. In recent years, the use of FFPE for transcriptomics has seen
immune populations or clusters, like TLS. As a result, major stakehold- substantial advancements. Yet, fresh sections inevitably have less deg-
ers are shifting their focus toward probe-based protocols, which has radation of RNA transcripts. Moreover, antibody-based analysis pro-
resulted in the development of next-generation platforms such as the vides many advantages over mRNA detection, including widespread
Xenium Analyzer (10x Genomics; targeted ISS based)11,12 and CosMx use in pathology. To this end, spatial proteomics has gained substantial
Spatial Molecular Imager (NanoString; untargeted in situ hybridization interest as a tool to visualize details within the TME beyond typical
based)31,32 for RNA and protein co-detection compatibility. For example, histopathology methods. Enthusiasm surrounding these technologies
CosMx was first used in non-small cell lung cancer and breast cancer to was exemplified in three back-to-back articles mapping cellular atlases
quantify ~1,000 RNA species and >100 proteins at subcellular resolu- of the human kidney42, the small intestine43 and the maternal–fetal
tion in formalin-fixed paraffin-embedded (FFPE) tumor tissue31. From interface44 using multiomics datasets. These advances have been pro-
these data, the dynamics of 100 ligand–receptor pairs over time and pelled by pioneering efforts in the TME field, accelerating progress not
space were analyzed based on gene expression changes in 18 different only within oncology but also across various other research domains.
cell types, including tumor cells, epithelial cells, fibroblasts, endothe- Mass-based imaging uses metal-tagged antibodies to identify
lial cells and 14 functionally distinct immune subsets. Interactions specific protein targets within tissues. These are subsequently quan-
between programmed cell death ligand 1 (PD-L1) and programmed tified using time-of-flight (TOF) mass spectrometry imaging (MSI),
cell death protein 1 (PD-1) were low in one of five lung tumors at the capitalizing on the distinct mass differences between metal tags. The
tumor–T cell interface, coinciding with unique macrophage-enriched two main competing platforms are multiplexed ion beam imaging
neighborhoods and transcriptionally diverse regional macrophage sub- (MIBI; Ionpath) and Hyperion imaging mass cytometry (IMC; Stand-
sets31. Additional spatial technologies for integrating transcriptomic ard BioTools), which share similar multiplexing capacity but differ

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1984

Review article https://doi.org/10.1038/s41590-023-01678-9

in several important ways (for example, IMC is destructive to tissue laboratories. To elevate more traditional immunostaining proce-
whereas MIBI is not). The first published uses of MIBI and IMC in cancer dures, cyclic staining protocols for multiplex histology, such as IBEX63,
were by two studies in 2014, where they were was applied to human FFPE SABER64, RareCyte65 and others66–68, have been developed both for
breast cancer tissue as proof of concept that they could provide mean- immunohistochemistry (IHC) and immunofluorescence, some of
ingful detail about tumor biology45,46. These technologies were later which have been commercialized (for example, the Lunaphore ‘COMET’
expanded to include more markers (>35) and larger patient cohorts47,48, platform). Cyclic histology involves the sequential staining of multiple
including ~500 patients from the METABRIC cohort for comprehensive targets using chromogen-conjugated or fluorophore-conjugated anti-
‘phenogenomic’ analyses49. Echoing molecular subtyping discoveries bodies, followed by antibody stripping and repeated cycles of staining.
from large-scale sequencing efforts by The Cancer Genome Atlas, Cyclic IHC was used with 29 multiplexed antibodies to characterize
these studies revealed previously unknown histological subtypes of the immune landscape of human pancreatic ductal adenocarcinoma,
breast cancer and spatially resolved traits, including tumor and TME revealing histological subtypes based on the pattern and abundance
features that were associated with patient outcomes48 and genomic of immune infiltration (myeloid enriched, lymphoid enriched and
drivers of disease49,50. For example, among patient tumors with an hypo-inflamed)69. Leukocyte density was heterogeneous across tumor
immune-enriched stroma, those with localized T cells were associated regions; for example, lymphocytes, including CD8+ T cells, were
with longer survival than those with a T cell-inflamed organization48, enriched within the tumor-adjacent stroma compared to the invasive
thus providing more granular detail to previous observations that the tumor epithelium, whereas myeloid densities between these regions
T cell architecture in breast cancer affects survival outcomes51. A surge were similar. Moreover, the CD8+ T cell/CD68+ macrophage ratio was
of IMC studies were subsequently published, expanding its applica- higher within the tumor and adjacent stroma regions in long-term sur-
tion to examine TME heterogeneity and immunotherapy response vivors than in short-term survivors (median overall survival of 58 versus
in primary lung tumors52, melanoma53, breast cancer54, glioblastoma 9 months), with region-specific enrichment in CD8+ T cell functional
and brain metastasis55, and to assess the effect of host physiology (for subsets. In patients treated with cytotoxic therapies before surgery,
example, obesity) on tumor immunology56. MIBI and IMC have also lymphoid-enriched tumors were associated with prolonged overall sur-
been applied to understand microenvironmental mechanisms of inva- vival, as were those with high CD8+ T cell density or a high CD8+ T cell/
sive breast cancer progression57 and therapeutic responses in patients. CD68+ macrophage ratio69. These findings align with prior work with
For example, IMC was used to dissect the cellular basis of anti-HER2 eight-plex cyclic IHC, in which CD8+ T cell spatial proximity to cancer
(trastuzumab) efficacy, revealing that the proximity of CD8+ T cells to cells was associated with improved survival in patients with pancreatic
the extracellular domain of HER2 (the antibody target) is important ductal adenocarcinoma (PDAC)70. Similar results were found with cyclic
for therapeutic responses58. Similarly, IMC identified mechanisms of IF in lung squamous cell carcinoma, where subsets of immunologically
immune checkpoint blockade efficacy in melanoma53 and lung cancer59 hot, warm and cold tumors were described71. When overlaid with five
patient samples, highlighting important roles for specific functional molecular subtypes identified through unsupervised clustering of
subsets of T cells and monocytes. The main advantage that these tools multiomic datasets, hot tumors aligned with an ‘inflamed secretory’
provide is their compatibility with antibodies and FFPE tissues with subtype, characterized by enrichment in immune-related pathways and
limited channel spillover (unlike multiplex fluorescence, for example). the lung squamous cell carcinoma secretory subtype72,73, thus drawing
However, they are impractical for high-throughput imaging due to parallels between genomic and immunological signatures.
slow acquisition speed (1 mm2 can take 1–2 h), limited metal isotopes For both cyclic IHC and immunofluorescence, the primary chal-
in nature and, in some cases, its destructive acquisition process, which lenges are the progressive degradation of tissue quality through each
negates the possibility of sequential rounds of staining. As a result, cycle and the buildup of background signal, particularly with fluores-
there is limited ability to maximize multiplexing capacity. cence. Spectral imaging systems, such as the PhenoImager platform
Alternatively, non-ablative DNA-barcoding techniques, such as (Akoya) or Vectra Polaris (PerkinElmer), use spectral unmixing algo-
co-detection by indexing (CODEX; PhenoCycler, Akoya), have substan- rithms to separate the overlapping emission spectra of fluorophores,
tially expanded the number of antibodies used in multiplex compared enabling enhanced multiplexing capabilities (7–9+ antibodies per
to IMC. Rather than metals, antibodies are conjugated to unique DNA cycle). This has resulted in important discoveries for understanding
oligonucleotides, followed by a multicycle reaction that includes itera- immune mechanisms of cancer progression, including examination
tive fluorescent imaging of up to three antibodies at a time60. In 2020, of macrophage phenotypes74,75, TLS76,77 and immunotherapy efficacy
CODEX was used to characterize the tumor immune landscape of in diverse cancers such as prostate78, renal79 and skin cancers80,81. A pri-
colorectal cancer using 56 multiplexed antibodies on tumor samples mary advantage of cyclic IHC or immunofluorescence is preservation of
from 35 patients, representing 140 total regions of interest61. Investiga- standard resolution for clinical pathology without requiring alterations
tion of the tumor margin revealed that spatially organized TLS at the to antibody chemistry. However, the optimization of these protocols
invasive front were associated with prolonged overall survival com- remains substantial. The Opal multiplex staining system addresses
pared to tumor boundaries with a more diffuse immune architecture61. some of these challenges, offering a kit-based method to multiplex up
Neighborhood analysis further revealed nine cellular neighborhoods to seven different antibodies, and is compatible with the PhenoImager
that were conserved across images and described how those neighbor- platform. Within the domain of flow cytometry, spectral fluorescence
hoods related to survival. For example, in neighborhoods enriched for (for example, Aurora, Cytek Biosciences) is also gaining popularity
granulocytes, co-enrichment with PD-1+CD4+ T cells was associated over other multiplexing methods such as CyTOF mass cytometry (for
with a survival benefit61, raising interest in the functional relationship example, the Helios system, Standard BioTools). This is because tran-
between these two populations in cancer. Functional differences in the sitioning between fluorescent antibody panels is more cost effective,
spatial topography of PD-1+CD4+ T cells was also shown in cutaneous consistent and time efficient than converting from fluorescence to
T cell lymphomas, where their relative proximity to tumor cells and metal isotopes. In addition, the potential for multiplexing is greater
regulatory T cells (that is, ‘SpatialScore’) was associated with anti-PD-1 with spectral fluorescence, in principle. Thus, similar concepts have
response in clinical trials62. As both CODEX and IMC technologies have been translated to tissue imaging.
been commercialized, they now provide prevalidated antibody subpan-
els for immune cell characterization in human tissues, thus reducing Spatial analysis of tumor metabolism
the time required to expand panels to their full capacity. With the advent of spatial single-cell technologies, there is emerging
Both MSI and CODEX spatial proteomics technologies have lim- interest in developing equivalent capabilities for metabolomics. How-
itations, including being costly and not widely accessible to most ever, mapping the spatial distribution of metabolites within tissues

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1985

Review article https://doi.org/10.1038/s41590-023-01678-9

at the single-cell level poses unique challenges compared to that of lipid distribution within tumor-associated TLS would be an exciting
RNA or protein, in part due to the incompatibility of metabolites with avenue of investigation. In cancer, MALDI and IMC have also been
most amplification and labeling methods. To avoid this, proteomic integrated with spatial transcriptomics in glioblastoma26. From these
multiplex imaging tools can be adapted to focus on metabolic enzymes analyses, three major metabolic modules in tumors were uncovered:
that can be used as a proxy for metabolite changes. In colorectal can- enrichment in the pentose phosphate pathway, phosphoadenylate
cer, MIBI was used to study the ‘metabolic regulome’ of CD8+ T cells metabolism, or glycolysis and amino sugar metabolism, the latter of
and understand metabolic competition within the TME82. A subset of which mapped to ‘reactive-hypoxic’ regions enriched for macrophages
CD8+ T cells expressing CD39 and PD-1 were found to be in a metaboli- and T cells and co-occurred with distinct copy number alterations26.
cally repressed state when positioned at a distance from the tumor– These findings suggest that distinct metabolic niches might shape
immune interface, compared to those in close proximity, suggesting immune clustering and create localized pockets of genomic instabil-
niche-driven regulation of immune cell functionality. Aligning with ity. Finally, building on metabolic analysis even further are tools like
this notion, metabolism-focused cyclic immunofluorescence was also metaFISH, which combines 16S rRNA-targeted FISH with atmospheric
applied to characterize nutrient partitioning between tumor cells and pressure MALDI on the same tissue section to study host–microbe
T cells in colorectal cancer in the context of obesity83. This approach metabolic interactions at a resolution of 3 μm95. First used to map the
demonstrated that a high-fat diet is associated with regional enhance- metabolome of deep-sea mussels, metaFISH revealed coordinated
ment of fatty acid uptake by tumor cells, thus depleting them from metabolic adaptation between host epithelial cells and associated
CD8+ T cells and causing T cell exhaustion and insufficient surveillance symbiotic bacteria of specific taxonomic identity. In sum, the inte-
against cancer. Extending these analyses to nutrient partitioning with gration of metabolic MSI with cell type-identification approaches
additional immune cells of the TME in situ84, as well as competition opens new opportunities to study tumor–immune nutrient parti-
between tissues systemically, will be the next challenge. tioning or even host–microbe metabolic dependencies and might be
Advances in metabolite profiling at cellular resolution have largely used to explore how cancer-associated microbial communities shape
focused on lipidomics. Various MSI techniques, such as matrix-assisted anti-tumor immunity.
laser desorption–ionization (MALDI)85 or desorption electrospray
ionization86, enable label-free targeting of lipid species based on their Visualizing tumor heterogeneity in the third
mass-to-charge ratio and characteristic fragmentation patterns. While dimension
these technologies can reach resolutions of ~20 µm, even greater reso- As many spatial capabilities improve to accommodate integration of
lution can be achieved with atmospheric pressure MALDI (resolution multiomic platforms, a natural progression is advancing these tech-
of ~1.4 µm)87 or transmission-mode MALDI-2 (resolution of 600 nm)88. nologies into 3D. In 2018, spatially resolved transcript amplicon readout
TOF secondary ion mass spectrometry (TOF-SIMS) provides further mapping (STARmap) was developed to map up to ~1,000 genes simul-
insight into the spatial distribution of lipids within tissues with sub- taneously in thick tissue blocks from the mouse brain at single-cell
micron resolution by employing a focused ion beam to desorb and resolution, and identified both neuronal and non-neuronal lineages,
ionize molecules from the sample surface, generating secondary ions including microglia, based on a 1,020-gene set96. Later in 2021, expan-
that are subsequently analyzed by mass spectrometry; however, it has sion sequencing using a polymer-based and hydrogel-based method
a lower mass-to-charge ratio, thus making it challenging to distinguish (ExSeq) was developed to generate RNA maps at nanoscale in tissues
metabolites. Specialized instruments like the Orbitrap (OrbiSIMS) or and revealed region-dependent transcriptional states in both tumor
nanoscale (NanoSIMS) mass analyzers have further improved the reso- and immune compartments in patient liver metastasis biopsy mate-
lution of select metabolites, lipids and sugars, albeit with more limited rial97. Transcriptional changes upon physical contact between tumor
spectral range, such that fewer species can be analyzed simultaneously. and immune cells were examined; for example, S100A8 was upregu-
For example, OrbiSIMS has been used to characterize metabolic intra- lated in B cells when in close proximity to EGFR+ tumor cells, which can
tumoral heterogeneity in adult glioblastoma and discriminate between engage the innate immune system. In the proteomic space, IMC has also
normal, tumor and necrotic regions89, where immune cell-enrichment moved to 3D capability, where images of serial sections are aligned and
patterns are known. Some of these MSI approaches have even been stacked98 to visualize individual cell coordinates in the third dimension.
applied successfully to three-dimensional (3D) metabolic imaging90,91; This approach has the unique advantage of understanding cellular biol-
for example, 3D OrbiSIMS has been applied to pediatric brain tumors ogy in the context of larger structures that are poorly represented in
including medulloblastoma to explore spatial metabolic heterogeneity thin two-dimensional (2D) slices, like blood vessels. The first report of
across specific molecular subtypes92. 3D IMC was in 2022, where it was applied to a HER2+ ductal breast tumor
Of course, a limitation of all these MSI approaches is the lack of cell over 152 consecutive sections99. Cells within the microenvironment
identification, highlighting a need for strategies to integrate protein showed clear patterns of 3D localization; for example, T cells tended
analysis so that canonical lineage markers can be overlaid with meta- to cluster around and along the endothelium, whereas macrophage
bolic tissue maps. This is particularly critical to discriminate metabo- clusters showed distinct distribution patterns (for example, uniform
lite signals from the immune compartment within tumors, where versus clustered distributions). Importantly, 2D IMC does not capture
individual cells can be highly infiltrative. To achieve this, a single-cell cells positioned above or below the section plane, resulting in incom-
spatially resolved metabolic (scSpaMet) pipeline was developed to plete insight into cellular interactions and neighborhood structures. By
profile cells of the TME with untargeted metabolomics combined with contrast, the 3D method offers a more complete perspective on these
multiplex protein analysis. This pipeline integrates metabolite imaging interactions. For example, there was a higher tendency for heterotypic
via TOF-SIMS with IMC to effectively identify cell types and metabolic interactions among endothelial cells and CD8− T cells (putative helper
changes within the same sample, including in 3D via serial section T cells) in 3D that was missed in 2D, whereas the opposite was true of
measurements93. This approach was first developed to measure the specific fibroblast subsets99. Similarly, in 2023, highly multiplexed
spatial distribution of 189 different compounds across immune cell cyclic immunofluorescence was performed on whole-slide serial sec-
populations and their corresponding 3D correlates in tonsil. Immune tions from colorectal cancer, generating a 3D atlas of cell state transi-
cells were found to have spatially unique fatty acid profiles across tions and immune interactions100. This approach revealed new insights
the tissue landscape, particularly when B cells within the germinal into spatial patterning of tumors, sometimes via gradients spanning
centers were compared with T cells surrounding the germinal cent- multiple millimeters of tissue, and highlighted how processes often
ers93, consistent with their dependency on fatty acids as an energy studied in 2D, such as TLS formation or epithelial-to-mesenchymal
source94. Whether such an approach could be useful to understand transition, are organized into larger 3D space. Such findings call for

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1986

Review article https://doi.org/10.1038/s41590-023-01678-9

a re-evaluation of spatial cellular relationships that are impossible tumor and metastatic lung and gut, compared to <10 in the rest of the
to capture with single-plane tissue sections or tissue microarrays, as body. In principle, immune cells can then be targeted via DNA origami
used by the majority of research laboratories for practical reasons. drug mimetics and biodistribution can be visualized at the single-cell
This includes understanding larger multicellular structures that are whole-body level110. A list of validated antibodies is continually updated
known to affect cancer biology, including tumor-associated TLS or for the research community (https://discotechnologies.org/wild-
vascularization processes. However, there are still important limita- DISCO/) and currently includes multiple T cell, B cell and macrophage
tions, as 3D proteomics is limited in its capacity to disentangle cellular markers, all highly relevant for understanding systemic mechanisms of
heterogeneity given its reliance on a restricted number of antibodies, anti-tumor immunity. Finally, DISCO clearing has also been combined
and requires extensive serial sections of precious patient material. with laser capture microdissection and mass spectrometry-based
Mapping tissues using machine learning and hematoxylin and profiling (DISCO–MS), developed first in mouse models of brain inflam-
eosin (H&E) staining, exemplified by the CODA technique101, is an mation and Alzheimer’s disease and a human atherosclerotic heart111.
approach to visualize 3D tissue architecture in large tissue volumes Although DISCO–MS cannot resolve single cells, regions of interest of
(>1 cm3). Like 3D IMC, CODA requires serial sectioning of tissue sam- approximately 60-cell clusters are extracted from whole-body speci-
ples; however, its ability to leverage machine learning to discern spe- mens during imaging, which can be combined with genetic tracing
cific tissue components or cell types and reconstruct the 3D tissue models to extract particular regions of cell enrichment (for example,
architecture only on the basis of H&E represents an exciting advance. fluorescently traced Lysm was used to track myeloid cells)111. Together,
CODA was first published in human PDAC, integrating >4,114 H&E these whole-body clearing technologies represent the next step in 3D
serial sections at ×20 magnification from 13 normal, precancerous and spatial imaging to understand systemic dynamics of immune function
tumor samples, capturing up to 3.5 cm3 in tissue volume and 1.6 billion between and within tissues; their value in understanding mechanisms
cells101. Deep learning was used to identify ten tissue compartments of cancer immunology has yet to be explored in detail.
without reliance on antibodies, including normal ductal epithelium,
pancreatic cancer precursor cells, PDAC, smooth muscle cells, acini, Analytical challenges and multiomics data
adipocytes, collagen, islets of Langerhans, lymph nodes and nerves. integration
A detailed map was generated capturing the multistage process of One of the major limiting factors impeding progress in spatial omics
PDAC tumorigenesis, and 3D structural phenotypes of connectivity is that analysis is often inaccessible to the average research laboratory
of precursor lesions were identified (tubular, lobular or dilated) along and requires specialized bioinformatic expertise. Several analytic chal-
with their microenvironmental properties. CODA was further applied lenges must be addressed (Fig. 2): an initial hurdle is cell segmentation,
to cell type discovery in skin (hair follicles, sweat glands, oil glands, for which accuracy completely dictates the quality of data output.
epidermis, vasculature and collagen), lung (bronchioles, alveoli, vas- Complex tissue architecture and variations in cell morphology make
culature, cancer metastasis and collagen) and liver (bile duct, hepato- it difficult to define precise cell boundaries, particularly for cell types
cytes, vasculature and collagen)101. It has also enabled CODA-guided with cell bodies distant from the nucleus (for example, microglia and
microdissection and multi-region genomic analysis, including tar- neurons). Additionally, the identification of ‘doublets’, where two or
geted and whole-exome sequencing102. These efforts have revealed more cells are mistakenly processed as a singular unit, can lead to cell
the multifocal and genetically heterogeneous nature of pancreatic misclassification and inaccurate transcript or protein assignment.
intra-epithelial neoplasia lesions within surgically resected human User-friendly tools that are widely accessed, such as ilastik112 and Cell-
tissue (~13 pancreatic intra-epithelial neoplasia lesions were found per Profiler113, offer pixel classification and cell-segmentation capabili-
cm3 of tissue)102. The next challenge will be to discern specific immune ties but struggle with accurately identifying complex cell boundaries.
subsets using deep learning and H&E, which seems plausible given that Several pipelines have been developed specifically for multiplex image
trained pathologists can achieve this by eye; such an approach would analysis, including MCMICRO and others114,115. Deep learning-based
be highly valuable not only in 3D analyses but also in 2D spatial analyses algorithms (for example, U-Net), have shown promise in improving
where resolution, tissue size and availability and cost are major limiting cell-segmentation accuracy by leveraging large annotated datasets.
factors of other technologies. Of note, metal-based counterstains like However, these approaches require extensive training and optimiza-
iridium and ruthenium mimic brightfield H&E and are compatible with tion. To address this limitation, several innovative approaches have
IMC on the same tissue section103, although the time and cost of IMC on been developed for cell segmentation. Cell segmentation across scales
the scale of several cubic centimeters is impractically high. (CIRCLE)116 is a comprehensive workflow for nuclear segmentation,
Both CODA and 3D IMC rely on extensive serial sectioning, thus integrating both a convolutional neural network-based approach
requiring manipulation of tissues and retrospective digital recon- and a foreground detection methodology followed by active contour
stitution of 3D images. As an alternative approach, tissue-clearing refinement to segregate overlapping cells. To accommodate cell size
techniques combined with high-resolution imaging have provided variations, a scale-selection process evaluates two independent sets of
major advances in visualizing tissues in 3D while keeping organ systems candidate cell detections at different magnifications, using a salience
completely intact. These methods involve the removal of lipids from the score derived from the structure tensor of the image. Cellpose117 is a
tissue, rendering it optically transparent, and subsequent labeling of versatile model designed to segment various cell types without the
target molecules using fluorescent probes. Tissue-clearing techniques, need for parameter tweaks, additional training data or model retrain-
for example, CLARITY104,105 or PACT106, have been combined with RNA ing. It incorporates two new features including a reversible conversion
probes for volumetric in situ hybridization assessment across intact from training set masks to vector gradients, which can be predicted by
whole tissues or whole bodies. Similarly, tissue-clearing techniques a neural network, and a vast dataset of diverse cell images. Moreover,
have been adapted for use with antibodies. WildDISCO107 combines Cellpose introduces a 3D cell-segmentation method that leverages the
whole-body immunostaining with DISCO-based tissue-clearing meth- existing 2D model without necessitating new 3D-labeled data. Finally,
ods108,109 to fully extract cholesterols from cell membranes and enable Mesmer118 is an automated algorithm capable of nuclear and whole-cell
deep and uniform penetration of antibodies. It has been used to visual- segmentation across various tissue types and imaging methods. The
ize the peripheral nervous system, lymphatic and vascular systems and two primary innovations involve a scalable method for generating vast
the immune system at the single-cell level in wild-type mice. In a 4T1 amounts of pixel-specific training data and a deep learning system that
breast cancer model, wildDISCO was used to image TLS with anti-CD23, uses these data to achieve human-level segmentation proficiency. To
anti-CD3 and anti-CD45 antibodies in the whole body of BALB/c mice, facilitate its widespread use in the research community, Mesmer was
which were found at a frequency between 10 and 25 within the primary incorporated into the open-source DeepCell platform119, complete with

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1987

Review article https://doi.org/10.1038/s41590-023-01678-9

a web interface and plugins tailored for ImageJ and QuPath. Consensus
on automated and user-friendly tools that can accurately segment
1 Cell segmentation cells across diverse tissue types and address the challenges posed by
complex cellular morphologies are urgently needed as the field of
• CIRCLE • SPATA
• Mesmer • NucleoAlzer spatial biology rapidly expands.
• Cellpose • ilastik Once cells have been segmented, the next hurdle is cell lineage
• Squidpy • DeepCell
• Tangram • Stardist
identification, which becomes increasingly difficult with fewer mark-
ers. Furthermore, with the increase in single-cell data, it has become
apparent that the distinction between cell type and activation state
is more blurred than ever before120. In the best-case scenario, lineage
assignment based on single-cell transcriptomics considers complex
gene expression patterns for each cell. However, even then, numerous
cell-identification methods have been developed over the years, and
2 Lineage assignment there is no consensus regarding which tools or marker combinations
• Seurat • Cell-ID are best. Spatial clustering further delineates cellular architecture by
• scran • SPOTlight categorizing cells based on their inherent attributes, like protein or
• SingIR • Tangram
mRNA levels, and the characteristics of adjacent cells, thus integrat-
• SpatialCPie • CODA
• RCTD • EcoTyper ing consideration of cell niches. The predominant methods for spatial
clustering derive from two foundational strategies: hidden Markov
random field (HMRF)121 and graph neural networks (GNNs)122. HMRF is
a statistical model that blends the concepts of hidden Markov models
and Markov random fields. In the context of spatial cellular architec-
ture, HMRF can be used to consider both the inherent features of a cell
and the influences of its neighbors when determining its state or clus-
3 Spatial analysis
ter. GNNs iteratively aggregate information from neighboring nodes
• Squidpy • NICHES (cells) to update the state or features of a node. This mechanism helps
• Giotto • SpaOTsc in capturing the local spatial context around each cell. By leveraging
• SpatialScore • ImaCytE
• CellCharter • STRISH the iterative information aggregation mechanism, GNNs can effectively
• CytoMAP • Spark analyze local neighborhoods in spatial data by capturing and highlight-
• histoCAT • Visinity
• stLearn ing patterns or anomalies in local spatial structures. However, with the
rapid evolution of more complex deep learning methods, AI-driven cell
phenotyping is poised to become a predominant strategy in navigating
this intricate landscape.
Indeed, deep learning-based approaches to cell identification are
4 3D organization advancing, such as real-time cell sorting based on cell morphology
(COSMOS) or single-cell RNA-seq cell typing (scGPT). Notably, scGPT
• Fiji Virtual Stack Slice has been applied to over 800 integrated datasets, learning from >45
• HyperStackReg
• ElasticStackAlignment million cells across >680 cell types123. The model can be effortlessly
• 3D watershed adjusted to attain top-tier performance for diverse downstream tasks,
• CODA
• Agave such as multi-batch integration, multiomic synthesis, cell type identifi-
cation, genetic perturbation forecasting and gene network deducing.
These signatures have yet to be applied to spatial datasets, although
this is undoubtedly on the horizon. Additional approaches include
EcoTyper124, a machine learning model designed for the extensive
detection and validation of cell states and multicellular communities
Maturation and
reprogramming 5 Spatiotemporal kinetics using bulk, single-cell and spatial gene expression data. In its applica-
tion to 12 primary cell lineages from 16 human cancers, EcoTyper pin-
• Pseudotime
• Synthetic recorders pointed 69 transcriptional cell states. The majority of these states were
• Zipseq unique to tumor tissues, consistently present across different tumor
types, and held substantial prognostic value. ImmuNet125 is another
machine learning-based phenotyping tool that is optimized specifically
Activation for immune cells within multiplex imaging data, in which highly dense
tissue can be challenging to resolve. Beyond cell type assignment, sev-
eral innovative approaches can capture cell lineage relationships and
Fig. 2 | Stages and challenges associated with single-cell spatial omics.
infer developmental trajectories. Trajectory-inference algorithms (for
A major remaining challenge in the spatial omic field is analysis among example, Monocle126, Slingshot127 and RNA velocity128) reconstruct the
many competing tools and lack of consensus among experts. Hurdles chronological sequence of cell states, providing insight into cellular
that need to be overcome in sequence include (1) accurate cell differentiation processes and cell state transitions. How such analytic
segmentation, (2) choosing between unsupervised or supervised approaches will inform our understanding of the spatial contexture
cell lineage assignment, (3) identifying tools for 2D spatial analysis and pseudotime of anti-tumor immune responses will be an area of
(for example, interactions, neighborhoods, communities) and, for interest going forward.
specialized technologies, the added challenge of (4) 3D and/or The third hurdle is spatial analysis. Beyond individual cells, spatial
(5) 4D analyses in which new analytical tools are still being developed. patterns and interactions of larger cellular neighborhoods or commu-
For each challenge listed (1–5), a non-exhaustive list of commonly used nities play pivotal roles in determining cellular behavior, cooperation
tools is displayed. and overall tissue dynamics. Decoding the intricate web of cell–cell

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1988

Review article https://doi.org/10.1038/s41590-023-01678-9

communications, be it through direct physical contact or the release recording systems have been developed to spatially map single-cell
and reception of signaling molecules, offers insights into the collec- lineage histories, such as MEMOIR145, intMEMOIR146 and Zombie147.
tive behavior of cellular communities. Much of the roadmap is acces- Although temporal resolution is often inaccessible using archived
sible through major landmark studies that have developed various tumor samples from patients, which are often fixed for biobanking,
tools for studying spatial phenotyping, interactions, neighborhoods it can be integrated over longer time scales. For example, mathemati-
and broader communities, including but not limited to histoCAT129, cal modeling has been applied to IMC data to fit circadian-controlled
Squidpy130, CellCharter131, Giotto132, BayesSpace133 and various pack- neutrophil dynamics in patient lung tumors based on time of surgery148.
ages within the Bioconductor project134. Alternatively, there are several This approach was then used to predict circadian-optimized treatment
open-source and commercially available tools that offer complete schedules to target neutrophils in vivo, while at the same time avoiding
end-to-end analysis, such as AstroPath80 or Visiopharm Phenoplex neutropenia148. Such analytic approaches might inform future appli-
(although high subscription costs limit accessibility for many academic cations for chronopharmacology in patients with cancer and reveal
institutions). Complementary approaches have been developed to mechanisms of time-specific efficacy of immune-targeted therapies149.
solve specific challenges associated with spatial datasets, such as Thus, developing computational frameworks that combine spatial,
understanding spatially variable expression patterns (for example, temporal and multiomics data and employing machine learning algo-
SpatialDE135, SPARK136, trendsceek137, MISTy138, scHOT139). SpatialDE, rithms for time series analysis offer opportunities to uncover dynamic
rooted in Gaussian process regression, clusters gene expression vari- cellular interactions and pathways. Advances in integrative analysis
ability into spatial and nonspatial components, identifying spatially tools are needed to fully achieve temporal resolution and unravel the
variable genes by assessing the variance ratio of these components. intricacies of time-dependent cancer biology in spatial multiomic
By contrast, SPARK is founded on a generalized linear spatial model, studies as they evolve in other fields150.
incorporating a range of spatial kernels, making it scalable and adept
at pinpointing spatially expressed genes across extensive spatial loca- Addressing clinical challenges through AI
tions. MISTy adopts a holistic approach, evaluating the entire expres- An emerging question is whether and how spatial omics datasets will
sion profile across different spatial or functional contexts, requiring no revolutionize clinical management of patients with cancer. Spatial
external information. It uniquely constructs a non-linear model based proteomic techniques, such as IMC, have the potential to elevate
on the expression of each marker as influenced by other markers. Often, standard pathology practice; by correlating proteomic profiles with
these tools are applied either to transcriptomic data or to proteomic clinical outcomes, machine learning models can identify spatial pat-
data but rarely to both; indeed, the development of analytic solutions terns and molecular signatures associated with disease progression.
that integrate spatial proteomics with transcriptomics has been a These models can then be applied prospectively to tackle challenging
leading focus in this field, paving the way for enhanced understanding clinical scenarios and help guide treatment decisions. For instance,
and exploration of complex biological systems. For example, GLUE140 patients with early-stage lung cancer will undergo surgery as standard
models regulatory interactions across omic platforms and can be useful of care, and because this is curative in many cases, there is usually no
for spatial analysis, as it can integrate omic layers with shared features follow-up treatment. However, up to one-third of patients face a recur-
(for example, spatial transcriptomics and single-cell RNA-seq). rence and might have benefited from further cytotoxic intervention.
Beyond 2D multiomics, emerging tools such as CODA or 3D Deep learning techniques applied to IMC datasets from patients with
multiplex imaging aim to extend analytic capabilities into the third early-stage lung cancer can predict with >95% accuracy which indi-
dimension. While adding another dimension presents new challenges, viduals are at a high risk of recurrence52, enabling the development of
the foundational principles for analysis are somewhat analogous to more tailored treatment strategies moving forward. Combining deep
those in 2D contexts. By contrast, an emerging hurdle will be effective learning with other spatial omics datasets, such as radiographs, offers
integration of the fourth dimension of time, particularly for analysis enhanced accuracy in cancer detection by identifying subtle patterns
of immune responses that are highly sensitive to circadian rhythms often missed by humans. This integration promises improved early
and rapid developmental or activation trajectories. Although 3D spa- diagnosis, better prognosis predictions and streamlined radiological
tiotemporal analysis of immune cell behaviors has been achieved workflows, leading to more efficient and effective cancer care. This
elegantly with intravital microscopy141, such approaches are limited approach has demonstrated promise in predicting overall survival
by the number of cell types that can be traced at one time. ZipSeq (a for patients with advanced-stage colorectal cancer151. However, suc-
play on ‘zip codes’) uses photocaged DNA oligonucleotides to imprint cessful implementation of machine learning approaches relies on the
light-sensitive barcodes onto live cells within intact tissues on demand availability of large, well-annotated clinical datasets at a scale that
during microscopy142. This enables single-cell RNA-seq data to be over- will require cooperation across research communities to consolidate
laid onto imaging data, creating a time-scaled map of gene expression. resources. Ongoing research and collaborations between experts in
ZipSeq was applied to a PyMT breast tumor model to understand spatial proteomics, machine learning and clinical oncology are thus
mechanisms of anti-tumor immunity as peripherally recruited cells essential to harness the full potential of spatial omics datasets for pre-
invaded the tumor142. Combining regional localization with pseudotime dictive modeling and clinical management of patients with cancer.
trajectory analysis enabled myeloid and T cell differentiation states
to be mapped to the infiltration of these cells into the tumor interior. Conclusions and future perspectives
Within the myeloid compartment, the spatiotemporal pattern of dif- The spatial omic technologies discussed here serve as powerful dis-
ferentiation aligned with the notion that recruited monocytes differ- covery tools for unraveling the complexities of the TME and have the
entiate into macrophages and are reprogrammed locally within tumors potential to transform clinical management of patients with cancer.
into a tumor-associated macrophage (TAM)-like state, coinciding with a However, a crucial question that remains is how to practically apply
shift in T cell exhaustion status. Unlike tools that use genetic photoacti- and scale these technologies for routine clinical use. Discoveries made
vatable fluorescent reporters143, ZipSeq has potential for application in using comprehensive techniques such as IMC must be refined and
live patient biopsies, where the activity of immune cells within tumors translated into more practical antibody combinations that can be read-
might affect eligibility for certain immunotherapies. Similarly, Live-seq ily employed in the clinical setting. For example, machine learning
can track gene expression changes in live cells over time; for exam- algorithms that predict recurrence of early lung cancer were initially
ple, it was used as a transcriptomic recorder in macrophages before developed with a 35-plex antibody panel; however, prediction accuracy
and after stimulation with lipopolysaccharide, and in adipose tissue was maintained when scaled down to the six most informative mark-
stromal cells before and after their differentiation144. Other synthetic ers52. This downsizing will be vital to address challenges related to the

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1989

Review article https://doi.org/10.1038/s41590-023-01678-9

slower and more expensive nature of some spatial omic techniques. 8. Chen, J. H. et al. Spatial analysis of human lung cancer reveals
To harness the clinical utility of spatial omics, it will be essential to organized immune hubs enriched for stem-like CD8 T cells and
identify the most informative spatial features (for example, through associated with immunotherapy response. Preprint at bioRxiv
attention mapping of deep learning predictions) and streamline the https://doi.org/10.1101/2023.04.04.535379 (2023).
workflows to make them feasible, cost effective and time efficient for 9. Eng, C. L. et al. Transcriptome-scale super-resolved imaging in
the widest possible user base. tissues by RNA seqFISH. Nature 568, 235–239 (2019).
Alternatively, there is an exciting scenario in which the need for 10. Lee, J. H. et al. Highly multiplexed subcellular RNA sequencing
scaling down spatial discoveries for clinical translation might be allevi- in situ. Science 343, 1360–1363 (2014).
ated, owing to the rapid evolution of both spatial technologies and AI. 11. Ke, R. et al. In situ sequencing for RNA analysis in preserved tissue
Drawing parallels with the early days of the genomics era, when sequenc- and cells. Nat. Methods 10, 857–860 (2013).
ing a whole genome cost millions of dollars, the trajectory of progress 12. Lee, J. H. et al. Fluorescent in situ sequencing (FISSEQ) of RNA for
indicates that technological advances can lead to cost reduction and gene expression profiling in intact cells and tissues. Nat. Protoc.
enhanced accessibility. If history has taught us anything, it is that technol- 10, 442–458 (2015).
ogy evolves rapidly, prices come down swiftly and hence we must con- 13. Lomakin, A. et al. Spatial genomics maps the structure, nature
tinue to guide the direction of progress. The same trajectory of advances and evolution of cancer clones. Nature 611, 594–602 (2022).
and cost reductions might apply to spatial omic technologies, enabling 14. Vickovic, S. et al. High-definition spatial transcriptomics for in situ
them to be seamlessly integrated into clinical practice without the need tissue profiling. Nat. Methods 16, 987–990 (2019).
for extensive downsizing. If these technologies can reliably inform clini- 15. Stahl, P. L. et al. Visualization and analysis of gene expression
cal outcomes, they will find their way into the clinic and be further fueled in tissue sections by spatial transcriptomics. Science 353,
by the integration of AI algorithms to decode spatial features that provide 78–82 (2016).
valuable insights for disease management. As a result, the clinical appli- 16. Rodriques, S. G. et al. Slide-seq: a scalable technology for
cation of spatial omic technologies holds immense promise and has the measuring genome-wide expression at high spatial resolution.
potential to reshape the landscape of precision medicine. Science 363, 1463–1467 (2019).
A final consideration is accessibility. Many of the technologies 17. Stickels, R. R. et al. Highly sensitive spatial transcriptomics at
discussed here are not yet commercialized and are thus only accessible near-cellular resolution with Slide-seqV2. Nat. Biotechnol. 39,
to a limited number of groups. Making these technologies available to 313–319 (2021).
the research community more broadly and/or sharing the extensive 18. Cho, C. S. et al. Microscopic examination of spatial transcriptome
datasets they generate through open-access platforms will be essential using Seq-Scope. Cell 184, 3559–3572 (2021).
for advancing treatments and fostering new discoveries. Moreover, 19. Zhang, D. et al. Spatial epigenome–transcriptome co-profiling of
data accessibility is often hindered by disparities in data standards and mammalian tissues. Nature 616, 113–122 (2023).
varying storage and collection methods across institutions. Concerns 20. Lu, T., Ang, C. E. & Zhuang, X. Spatially resolved epigenomic
about patient privacy, intellectual property rights and the lack of uni- profiling of single cells in complex tissues. Cell 185,
fied platforms further complicate the ability to share and integrate 4448–4464 (2022).
data effectively. Overcoming these obstacles requires standardized 21. Zollinger, D. R., Lingle, S. E., Sorg, K., Beechem, J. M. & Merritt, C.
protocols and robust governance that balance scientific goals with R. GeoMx RNA assay: high multiplex, digital, spatial analysis of
individual rights and safety. RNA in FFPE tissue. Methods Mol. Biol. 2148, 331–345 (2020).
In conclusion, while spatial omic technologies currently serve 22. Merritt, C. R. et al. Multiplex digital spatial profiling of proteins
as powerful discovery tools to understand disease processes, their and RNA in fixed tissue. Nat. Biotechnol. 38, 586–599 (2020).
ultimate value lies in their successful translation and integration into 23. Wu, S. Z. et al. A single-cell and spatially resolved atlas of human
routine clinical practice. The challenge ahead is to streamline and breast cancers. Nat. Genet. 53, 1334–1347 (2021).
optimize these technologies for clinical use. By harnessing the power 24. Li, R. et al. Mapping single-cell transcriptomes in the intra-tumoral
of AI, spatial omic technologies have the potential to transform clini- and associated territories of kidney cancer. Cancer Cell 40,
cal decision making, enhance patient outcomes and pave the way for 1583–1599 (2022).
personalized approaches to disease management in the future. 25. Hao, D. et al. The single-cell immunogenomic landscape of B and
plasma cells in early-stage lung adenocarcinoma. Cancer Discov.
References 12, 2626–2645 (2022).
1. Lewis, S. M. et al. Spatial omics and multiplexed imaging to 26. Ravi, V. M. et al. Spatially resolved multi-omics deciphers
explore cancer biology. Nat. Methods 18, 997–1012 (2021). bidirectional tumor–host interdependence in glioblastoma.
2. Bressan, D., Battistoni, G. & Hannon, G. J. The dawn of spatial Cancer Cell 40, 639–655 (2022).
omics. Science 381, eabq4964 (2023). 27. Ji, A. L. et al. Multimodal analysis of composition and spatial
3. Raj, A., van den Bogaard, P., Rifkin, S. A., van Oudenaarden, A. architecture in human squamous cell carcinoma. Cell 182,
& Tyagi, S. Imaging individual mRNA molecules using multiple 497–514 (2020).
singly labeled probes. Nat. Methods 5, 877–879 (2008). 28. Moncada, R. et al. Integrating microarray-based spatial
4. Lubeck, E., Coskun, A. F., Zhiyentayev, T., Ahmad, M. & Cai, L. transcriptomics and single-cell RNA-seq reveals tissue
Single-cell in situ RNA profiling by sequential hybridization. architecture in pancreatic ductal adenocarcinomas.
Nat. Methods 11, 360–361 (2014). Nat. Biotechnol. 38, 333–342 (2020).
5. Shah, S., Lubeck, E., Zhou, W. & Cai, L. In situ transcription 29. Anadon, C. M. et al. Ovarian cancer immunogenicity is governed
profiling of single cells reveals spatial organization of cells in the by a narrow subset of progenitor tissue-resident memory T cells.
mouse hippocampus. Neuron 92, 342–357 (2016). Cancer Cell 40, 545–557 (2022).
6. Chen, K. H., Boettiger, A. N., Moffitt, J. R., Wang, S. & Zhuang, X. 30. Galeano Nino, J. L. et al. Effect of the intratumoral microbiota
RNA imaging. Spatially resolved, highly multiplexed RNA profiling on spatial and cellular heterogeneity in cancer. Nature 611,
in single cells. Science 348, aaa6090 (2015). 810–817 (2022).
7. Hara, T. et al. Interactions between cancer cells and immune cells 31. He, S. et al. High-plex imaging of RNA and proteins at
drive transitions to mesenchymal-like states in glioblastoma. subcellular resolution in fixed tissue by spatial molecular
Cancer Cell 39, 779–792 (2021). imaging. Nat. Biotechnol. 40, 1794–1806 (2022).

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1990

Review article https://doi.org/10.1038/s41590-023-01678-9

32. He, S. et al. High-plex multiomic analysis in FFPE at subcellular 57. Risom, T. et al. Transition to invasive breast cancer is associated
level by spatial molecular imaging. Preprint at bioRxiv https://doi. with progressive changes in the structure and composition of
org/10.1101/2021.11.03.467020 (2022). tumor stroma. Cell 185, 299–310 (2022).
33. Liu, Y. et al. High-spatial-resolution multi-omics sequencing via 58. Carvajal-Hausdorf, D. E. et al. Multiplexed (18-plex) measurement
deterministic barcoding in tissue. Cell 183, 1665–1681 (2020). of signaling targets and cytotoxic T cells in trastuzumab-treated
34. Vickovic, S. et al. SM-Omics is an automated platform for patients using imaging mass cytometry. Clin. Cancer Res. 25,
high-throughput spatial multi-omics. Nat. Commun. 13, 3054–3062 (2019).
795 (2022). 59. Sorin, M. et al. Single-cell spatial landscape of immunotherapy
35. Ben-Chetrit, N. et al. Integration of whole transcriptome spatial response reveals mechanisms of CXCL13 enhanced antitumor
profiling with protein markers. Nat. Biotechnol. 41, 788–793 (2023). immunity. J. Immunother. Cancer 11, e005545 (2023).
36. Guilliams, M. et al. Spatial proteogenomics reveals distinct and 60. Black, S. et al. CODEX multiplexed tissue imaging with
evolutionarily conserved hepatic macrophage niches. Cell 185, DNA-conjugated antibodies. Nat. Protoc. 16, 3802–3835 (2021).
379–396 (2022). 61. Schurch, C. M. et al. Coordinated cellular neighborhoods
37. Liu, Y. et al. High-plex protein and whole transcriptome orchestrate antitumoral immunity at the colorectal cancer
co-mapping at cellular resolution with spatial CITE-seq. invasive front. Cell 182, 1341–1359 (2020).
Nat. Biotechnol. 41, 1405–1409 (2023). 62. Phillips, D. et al. Immune cell topography predicts response to
38. Joshi, K. et al. Spatial heterogeneity of the T cell receptor PD-1 blockade in cutaneous T cell lymphoma. Nat. Commun. 12,
repertoire reflects the mutational landscape in lung cancer. 6726 (2021).
Nat. Med. 25, 1549–1559 (2019). 63. Radtke, A. J. et al. IBEX: an iterative immunolabeling and chemical
39. Liu, S. et al. Spatial maps of T cell receptors and transcriptomes bleaching method for high-content imaging of diverse tissues.
reveal distinct immune niches and interactions in the adaptive Nat. Protoc. 17, 378–401 (2022).
immune response. Immunity 55, 1940–1952 (2022). 64. Saka, S. K. et al. Immuno-SABER enables highly multiplexed
40. Hudson, W. H. & Sudmeier, L. J. Localization of T cell clonotypes and amplified protein imaging in tissues. Nat. Biotechnol. 37,
using the Visium spatial transcriptomics platform. STAR Protoc. 3, 1080–1090 (2019).
101391 (2022). 65. Lin, J. R. et al. High-plex immunofluorescence imaging and
41. Sudmeier, L. J. et al. Distinct phenotypic states and spatial traditional histology of the same tissue section for discovering
distribution of CD8+ T cell clonotypes in human brain metastases. image-based biomarkers. Nat. Cancer 4, 1036–1052 (2023).
Cell Rep. Med. 3, 100620 (2022). 66. Banik, G. et al. High-dimensional multiplexed
42. Lake, B. B. et al. An atlas of healthy and injured cell states and immunohistochemical characterization of immune contexture in
niches in the human kidney. Nature 619, 585–594 (2023). human cancers. Methods Enzymol. 635, 1–20 (2020).
43. Hickey, J. W. et al. Organization of the human intestine at 67. Lin, J. R., Fallahi-Sichani, M., Chen, J. Y. & Sorger, P. K. Cyclic
single-cell resolution. Nature 619, 572–584 (2023). immunofluorescence (CycIF), a highly multiplexed method for
44. Greenbaum, S. et al. A spatially resolved timeline of the human single-cell imaging. Curr. Protoc. Chem. Biol. 8, 251–264 (2016).
maternal–fetal interface. Nature 619, 595–605 (2023). 68. Lin, J. R. et al. Highly multiplexed immunofluorescence imaging
45. Angelo, M. et al. Multiplexed ion beam imaging of human breast of human tissues and tumors using t-CyCIF and conventional
tumors. Nat. Med. 20, 436–442 (2014). optical microscopes. eLife 7, e31657 (2018).
46. Giesen, C. et al. Highly multiplexed imaging of tumor tissues 69. Liudahl, S. M. et al. Leukocyte heterogeneity in pancreatic ductal
with subcellular resolution by mass cytometry. Nat. Methods 11, adenocarcinoma: phenotypic and spatial features associated with
417–422 (2014). clinical outcome. Cancer Discov. 11, 2014–2031 (2021).
47. Keren, L. et al. A structured tumor–immune microenvironment in 70. Carstens, J. L. et al. Spatial computation of intratumoral T cells
triple negative breast cancer revealed by multiplexed ion beam correlates with survival of patients with pancreatic cancer.
imaging. Cell 174, 1373–1387 (2018). Nat. Commun. 8, 15095 (2017).
48. Jackson, H. W. et al. The single-cell pathology landscape of breast 71. Satpathy, S. et al. A proteogenomic portrait of lung squamous cell
cancer. Nature 578, 615–620 (2020). carcinoma. Cell 184, 4348–4371 (2021).
49. Ali, H. R. et al. Imaging mass cytometry and multiplatform 72. Wilkerson, M. D. et al. Lung squamous cell carcinoma mRNA
genomics define the phenogenomic landscape of breast cancer. expression subtypes are reproducible, clinically important,
Nat. Cancer 1, 163–175 (2020). and correspond to normal cell types. Clin. Cancer Res. 16,
50. Danenberg, E. et al. Breast tumor microenvironment structures 4864–4875 (2010).
are associated with genomic features and clinical outcome. 73. Cancer Genome Atlas Research Network. Comprehensive
Nat. Genet. 54, 660–669 (2022). genomic characterization of squamous cell lung cancers. Nature
51. Gruosso, T. et al. Spatially distinct tumor immune 489, 519–525 (2012).
microenvironments stratify triple-negative breast cancers. 74. Obradovic, A. et al. Single-cell protein activity analysis identifies
J. Clin. Invest. 129, 1785–1800 (2019). recurrence-associated renal tumor macrophages. Cell 184,
52. Sorin, M. et al. Single-cell spatial landscapes of the lung tumour 2988–3005 (2021).
immune microenvironment. Nature 614, 548–554 (2023). 75. Vayrynen, S. A. et al. Composition, spatial characteristics, and
53. Moldoveanu, D. et al. Spatially mapping the immune landscape prognostic significance of myeloid cell infiltration in pancreatic
of melanoma using imaging mass cytometry. Sci. Immunol. 7, cancer. Clin. Cancer Res. 27, 1069–1081 (2021).
eabi5072 (2022). 76. Helmink, B. A. et al. B cells and tertiary lymphoid structures
54. Wang, X. Q. et al. Spatial predictors of immunotherapy response promote immunotherapy response. Nature 577, 549–555 (2020).
in triple-negative breast cancer. Nature 621, 868–876 (2023). 77. Ruffin, A. T. et al. B cell signatures and tertiary lymphoid
55. Karimi, E. et al. Single-cell spatial immune landscapes of primary structures contribute to outcome in head and neck squamous
and metastatic brain tumours. Nature 614, 555–563 (2023). cell carcinoma. Nat. Commun. 12, 3349 (2021).
56. McDowell, S. A. C. et al. Neutrophil oxidative stress mediates 78. Gao, J. et al. VISTA is an inhibitory immune checkpoint that is
obesity-associated vascular dysfunction and metastatic increased after ipilimumab therapy in patients with prostate
transmigration. Nat. Cancer 2, 545–562 (2021). cancer. Nat. Med. 23, 551–555 (2017).

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1991

Review article https://doi.org/10.1038/s41590-023-01678-9

79. Au, L. et al. Determinants of anti-PD-1 response and resistance in 102. Braxton, A. M. et al. Three-dimensional genomic mapping of
clear cell renal cell carcinoma. Cancer Cell 39, 1497–1518 (2021). human pancreatic tissue reveals striking multifocality and genetic
80. Berry, S. et al. Analysis of multispectral imaging with the heterogeneity in precancerous lesions. Preprint at bioRxiv
AstroPath platform informs efficacy of PD-1 blockade. Science https://doi.org/10.1101/2023.01.27.525553 (2023).
372, eaba2609 (2021). 103. Catena, R., Montuenga, L. M. & Bodenmiller, B. Ruthenium
81. Jacquelot, N. et al. Blockade of the co-inhibitory molecule PD-1 counterstaining for imaging mass cytometry. J. Pathol. 244,
unleashes ILC2-dependent antitumor immunity in melanoma. 479–484 (2018).
Nat. Immunol. 22, 851–864 (2021). 104. Chung, K. et al. Structural and molecular interrogation of intact
82. Hartmann, F. J. et al. Single-cell metabolic profiling of human biological systems. Nature 497, 332–337 (2013).
cytotoxic T cells. Nat. Biotechnol. 39, 186–197 (2021). 105. Sylwestrak, E. L., Rajasethupathy, P., Wright, M. A., Jaffe, A. &
83. Ringel, A. E. et al. Obesity shapes metabolism in the tumor Deisseroth, K. Multiplexed intact-tissue transcriptional analysis at
microenvironment to suppress anti-tumor immunity. Cell 183, cellular resolution. Cell 164, 792–804 (2016).
1848–1866 (2020). 106. Yang, B. et al. Single-cell phenotyping within transparent
84. Reinfeld, B. I. et al. Cell-programmed nutrient partitioning in the intact tissue through whole-body clearing. Cell 158,
tumour microenvironment. Nature 593, 282–288 (2021). 945–958 (2014).
85. Soltwisch, J. et al. Mass spectrometry imaging with laser-induced 107. Mai, H. et al. Whole-body cellular mapping in mouse using
postionization. Science 348, 211–215 (2015). standard IgG antibodies. Nat. Biotechnol. https://doi.org/10.1038/
86. Wiseman, J. M., Ifa, D. R., Song, Q. & Cooks, R. G. Tissue imaging s41587-023-01846-0 (2023).
at atmospheric pressure using desorption electrospray ionization 108. Ueda, H. R. et al. Tissue clearing and its applications in
(DESI) mass spectrometry. Angew. Chem. Int. Ed. Engl. 45, neuroscience. Nat. Rev. Neurosci. 21, 61–79 (2020).
7188–7192 (2006). 109. Cai, R. et al. Panoptic imaging of transparent mice reveals
87. Kompauer, M., Heiles, S. & Spengler, B. Atmospheric pressure whole-body neuronal projections and skull–meninges
MALDI mass spectrometry imaging of tissues and cells at 1.4-µm connections. Nat. Neurosci. 22, 317–327 (2019).
lateral resolution. Nat. Methods 14, 90–96 (2017). 110. Molbay, M. et al. Single-cell precision nanotechnology in vivo.
88. Niehaus, M., Soltwisch, J., Belov, M. E. & Dreisewerd, K. Preprint at bioRxiv https://doi.org/10.1101/2023.07.24.550304
Transmission-mode MALDI-2 mass spectrometry imaging (2023).
of cells and tissues at subcellular resolution. Nat. Methods 16, 111. Bhatia, H. S. et al. Spatial proteomics in three-dimensional intact
925–931 (2019). specimens. Cell 185, 5040–5058 (2022).
89. He, W. et al. Untargeted metabolomic characterization of 112. Berg, S. et al. ilastik: interactive machine learning for (bio)image
glioblastoma intra-tumor heterogeneity using OrbiSIMS. Anal. analysis. Nat. Methods 16, 1226–1232 (2019).
Chem. 95, 5994–6001 (2023). 113. McQuin, C. et al. CellProfiler 3.0: next-generation image
90. Passarelli, M. K. et al. The 3D OrbiSIMS-label-free metabolic processing for biology. PLoS Biol. 16, e2005970 (2018).
imaging with subcellular lateral resolution and high 114. Windhager, J., Bodenmiller, B. & Eling, N. An end-to-end workflow
mass-resolving power. Nat. Methods 14, 1175–1183 (2017). for multiplexed image processing and analysis. Preprint at bioRxiv
91. Fletcher, J. S. & Vickerman, J. C. A new SIMS paradigm for 2D and https://doi.org/10.1101/2021.11.12.468357 (2021).
3D molecular imaging of bio-systems. Anal. Bioanal. Chem. 396, 115. Schapiro, D. et al. MCMICRO: a scalable, modular
85–104 (2010). image-processing pipeline for multiplexed tissue imaging.
92. Linke, F. et al. Identifying new biomarkers of aggressive group 3 Nat. Methods 19, 311–315 (2022).
and SHH medulloblastoma using 3D hydrogel models, single cell 116. Karimi, E. et al. Machine learning meets classical computer vision
RNA sequencing and 3D OrbiSIMS imaging. Acta Neuropathol. for accurate cell identification. Preprint at bioRxiv https://doi.org/
Commun. 11, 6 (2023). 10.1101/2022.02.27.482183 (2022).
93. Ganesh, S. et al. Spatially resolved 3D metabolomic profiling in 117. Stringer, C., Wang, T., Michaelos, M. & Pachitariu, M. Cellpose: a
tissues. Sci. Adv. 7, eabd0957 (2021). generalist algorithm for cellular segmentation. Nat. Methods 18,
94. Weisel, F. J. et al. Germinal center B cells selectively oxidize fatty 100–106 (2021).
acids for energy while conducting minimal glycolysis. 118. Greenwald, N. F. et al. Whole-cell segmentation of tissue
Nat. Immunol. 21, 331–342 (2020). images with human-level performance using large-scale
95. Geier, B. et al. Spatial metabolomics of in situ host–microbe data annotation and deep learning. Nat. Biotechnol. 40,
interactions at the micrometre scale. Nat. Microbiol. 5, 555–565 (2022).
498–510 (2020). 119. Bannon, D. et al. DeepCell Kiosk: scaling deep learning-enabled
96. Wang, X. et al. Three-dimensional intact-tissue sequencing of cellular image analysis with Kubernetes. Nat. Methods 18,
single-cell transcriptional states. Science 361, eaat5691 (2018). 43–45 (2021).
97. Alon, S. et al. Expansion sequencing: spatially precise in situ 120. Kobak, D. & Berens, P. The art of using t-SNE for single-cell
transcriptomics in intact biological systems. Science 371, transcriptomics. Nat. Commun. 10, 5416 (2019).
eaax2656 (2021). 121. Chatzis, S. P. & Tsechpenakis, G. The infinite hidden
98. Kartasalo, K. et al. Comparative analysis of tissue reconstruction Markov random field model. IEEE Trans. Neural Netw. 21,
algorithms for 3D histology. Bioinformatics 34, 3013–3021 1004–1014 (2010).
(2018). 122. Scarselli, F., Gori, M., Tsoi, A. C., Hagenbuchner, M. & Monfardini,
99. Kuett, L. et al. Three-dimensional imaging mass cytometry for G. The graph neural network model. IEEE Trans. Neural Netw. 20,
highly multiplexed molecular and cellular mapping of tissues and 61–80 (2009).
the tumor microenvironment. Nat. Cancer 3, 122–133 (2022). 123. Cui, H. et al. scGPT: towards building a foundation model for
100. Lin, J. R. et al. Multiplexed 3D atlas of state transitions single-cell multi-omics using generative AI. Preprint at bioRxiv
and immune interaction in colorectal cancer. Cell 186, https://doi.org/10.1101/2023.04.30.538439 (2023).
363–381 (2023). 124. Luca, B. A. et al. Atlas of clinically distinct cell states
101. Kiemen, A. L. et al. CODA: quantitative 3D reconstruction of large and ecosystems across human solid tumors. Cell 184,
tissues at cellular resolution. Nat. Methods 19, 1490–1499 (2022). 5482–5496 (2021).

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1992

Review article https://doi.org/10.1038/s41590-023-01678-9

125. Sultan, S. et al. ImmuNet: a segmentation-free machine learning 147. Askary, A. et al. In situ readout of DNA barcodes and single base
pipeline for immune landscape phenotyping in tumors by edits facilitated by in vitro transcription. Nat. Biotechnol. 38,
muliplex imaging. Preprint at bioRxiv https://doi.org/10.1101/ 66–75 (2020).
2021.10.22.464548 (2023). 148. Milette, S. et al. Circadian control of neutrophil extracellular
126. Trapnell, C. et al. The dynamics and regulators of cell fate trap formation temporally regulates metastatic lung cancer
decisions are revealed by pseudotemporal ordering of single progression. In Molecular Analysis for Precision Oncology
cells. Nat. Biotechnol. 32, 381–386 (2014). Congress 2023 abstract 76P (2023).
127. Street, K. et al. Slingshot: cell lineage and pseudotime 149. Qian, D. C. et al. Effect of immunotherapy time-of-day infusion
inference for single-cell transcriptomics. BMC Genomics 19, on overall survival among patients with advanced melanoma
477 (2018). in the USA (MEMOIR): a propensity score-matched analysis
128. La Manno, G. et al. RNA velocity of single cells. Nature 560, of a single-centre, longitudinal study. Lancet Oncol. 22,
494–498 (2018). 1777–1786 (2021).
129. Schapiro, D. et al. histoCAT: analysis of cell phenotypes and 150. Nobori, T. et al. Time-resolved single-cell and spatial gene
interactions in multiplex image cytometry data. Nat. Methods 14, regulatory atlas of plants under pathogen attack. Preprint at
873–876 (2017). bioRxiv https://doi.org/10.1101/2023.04.10.536170 (2023).
130. Palla, G. et al. Squidpy: a scalable framework for spatial omics 151. Negreros-Osuna, A. A. et al. Radiomics texture features in
analysis. Nat. Methods 19, 171–178 (2022). advanced colorectal cancer: Correlation with BRAF mutation and
131. Varrone, M., Tavernari, D., Santamaria-Martínez, A. & Ciriello, G. 5-year overall survival. Radiol. Imaging Cancer 2, e190084 (2020).
CellCharter: a scalable framework to chart and compare cell 152. Madonna, M. C. et al. In vivo metabolic imaging identifies lipid
niches across multiple samples and spatial -omics technologies. vulnerability in a preclinical model of Her2+/Neu breast cancer
Preprint at bioRxiv https://doi.org/10.1101/2023.01.10.523386 residual disease and recurrence. NPJ Breast Cancer 8, 111 (2022).
(2023).
132. Dries, R. et al. Giotto: a toolbox for integrative analysis and Acknowledgements
visualization of spatial expression data. Genome Biol. 22, We thank M. Sorin for his valuable feedback on this paper. We
78 (2021). acknowledge funding from the Canadian Institutes of Health Research
133. Zhao, E. et al. Spatial transcriptomics at subspot resolution with (CIHR; PJT-162137, L.A.W.; PJT-159742, PJT-178306, D.F.Q.), TRANSCAN-3
BayesSpace. Nat. Biotechnol. 39, 1375–1384 (2021). (jointly supported by CIHR TRN-184696 and Canadian Cancer
134. Amezquita, R. A. et al. Orchestrating single-cell analysis with Society 707710, D.F.Q.), the Brain Tumor Funders’ Collaborative and
Bioconductor. Nat. Methods 17, 137–145 (2020). Canada Foundation for Innovation. L.A.W. holds a Rosalind and Morris
135. Svensson, V., Teichmann, S. A. & Stegle, O. SpatialDE: Goodman Chair in Lung Cancer Research, and D.F.Q. holds a Tier II
identification of spatially variable genes. Nat. Methods 15, Canada Research Chair in Tumour Microenvironment Research.
343–346 (2018).
136. Sun, S., Zhu, J. & Zhou, X. Statistical analysis of spatial expression Author contributions
patterns for spatially resolved transcriptomic studies. Nat. D.F.Q. and L.A.W. contributed equally to all aspects of the article.
Methods 17, 193–200 (2020).
137. Edsgard, D., Johnsson, P. & Sandberg, R. Identification of spatial Competing interests
expression trends in single-cell gene expression data. Nat. The authors declare no competing interests.
Methods 15, 339–342 (2018).
138. Tanevski, J., Flores, R. O. R., Gabor, A., Schapiro, D. & Additional information
Saez-Rodriguez, J. Explainable multiview framework for Correspondence and requests for materials should be addressed to
dissecting spatial relationships from highly multiplexed data. Logan A. Walsh or Daniela F. Quail.
Genome Biol. 23, 97 (2022).
139. Ghazanfar, S. et al. Investigating higher-order interactions in Peer review information Nature Immunology thanks Linghua Wang
single-cell data with scHOT. Nat. Methods 17, 799–806 (2020). and the other, anonymous, reviewer(s) for their contribution to the
140. Cao, Z. J. & Gao, G. Multi-omics single-cell data integration peer review of this work. Primary Handling Editor: N. Bernard, in
and regulatory inference with graph-linked embedding. Nat. collaboration with the Nature Immunology team.
Biotechnol. 40, 1458–1466 (2022).
141. Crainiciuc, G. et al. Behavioural immune landscapes of Reprints and permissions information is available at
inflammation. Nature 601, 415–421 (2022). www.nature.com/reprints.
142. Hu, K. H. et al. ZipSeq: barcoding for real-time mapping of single
cell transcriptomes. Nat. Methods 17, 833–843 (2020). Publisher’s note Springer Nature remains neutral with regard to
143. Medaglia, C. et al. Spatial reconstruction of immune niches by jurisdictional claims in published maps and institutional affiliations.
combining photoactivatable reporters and scRNA-seq. Science
358, 1622–1626 (2017). Springer Nature or its licensor (e.g. a society or other partner) holds
144. Chen, W. et al. Live-seq enables temporal transcriptomic exclusive rights to this article under a publishing agreement with
recording of single cells. Nature 608, 733–740 (2022). the author(s) or other rightsholder(s); author self-archiving of the
145. Frieda, K. L. et al. Synthetic recording and in situ readout of accepted manuscript version of this article is solely governed by the
lineage information in single cells. Nature 541, 107–111 (2017). terms of such publishing agreement and applicable law.
146. Chow, K. K. et al. Imaging cell lineage with a synthetic digital
recording system. Science 372, eabb3099 (2021). © Springer Nature America, Inc. 2023

Nature Immunology | Volume 24 | December 2023 | 1982–1993 1993