Roadmap for spatial transcriptomic profiling of HIV reservoir in tissues

Abstract

Purpose of review:

Mechanisms of HIV persistence in tissues are distinct from that in the blood. Spatial transcriptomic profiling examines HIV-infected cells, surrounding neighborhoods, and tissue microenvironment in unprecedented resolution. Spatial profiling captures cytokine gradients, distances between HIV-infected cells and immune effectors (and their function versus exhaustion), and cell-cell interactions. We present an overview of spatial transcriptomic platforms and a workflow of quality controls, sanity check, and bioinformatic analysis.

Recent findings:

The selection of spatial profiling methods should base on the research question, resolution, breadth of coverage, the expresssion level of RNA of interest, tissue quality, and tissue size. Advanced spatial transcriptomic profiling can capture RNA molecules at high resolution (<1 μm) and thus enable near-single cell profiling at genomewide (~20,000 genes) breadth. Specifically, poly-A-based mRNA capture can identify previously unknown targets, while targeted RNA capture increases sensitivity in low-quality tissues. In targeted capture, however, the increase in target numbers frequently decreases sensitivity. Coupling ATAC-seq, protein capture, and T cell receptor sequencing to spatial platforms is ongoing.

Summary:

Spatial transcriptomic profiling uncovers mechanisms of HIV persistence in tissues and informs therapeutic strategies. Investigators should ensure the rigor of analysis, validate findings, and avoid reporting signatures with unknown biological significance.

Introduction

The life-long persistence of HIV reservoir is the barrier to cure. While HIV-infected cells in blood have been extensively studied [1–7], HIV reservoir within tissues are not fully understood. The lymph node and the gut account for the majority of the reservoir [8], with cell types in tissues distinct from those in the blood. HIV infects cells expressing CD4 receptor and CCR5 or CXCR4 coreceptors, primarily CD4+ T cells [such as memory CD4+ T cells, tissue resident memory T cells (TRM), and T follicular helper cells (TFH)], macrophages (exhibiting a spectrum of classically activated M1-like to alternatively activate M2-like macrophages, such as microglia in the brain, Kupfer cells in the liver, and lung macrophages), and dendritic cells [9] (DCs, including conventional dendritic cells (cDCs) and plasmacytoid dendritic cells (pDCs). Although follicular dendritic cells (FDCs, a subset of fibroblastic reticular cells, not DCs) do not express CD4, they retain infectious HIV viral particles in the B cell follicle in the lymph node as an HIV reservoir [10]. Furthermore, the mechanism of HIV persistence in tissues is different from that in blood. HIV-infected cells persist in immune sanctuary sites (where immune effectors cannot reach due to the lack of homing signals, such as exclusion of CD8+ T cells from the B cell follicles in the lymph node due to the lack of CXCR5 homing receptor [11]), anatomical sanctuary sites [where immune effectors cannot reach due to anatomical barriers, such as blood brain barrier (BBB) blocking the entry into the central nervous system (CNS)], tissues providing survival benefit for HIV-infected cells (such as BACH2-driven long-lived memory programs in the gut) [12], and tissues where the drug level is low [13]. Through spatial profiling, recent studies have provided insights into reduced type I interferon responses [14] and decreased cytolytic effector function [15] in the germinal centers where HIV+ cells reside. Distinct tissue microenvironment has been captured in the gut [16], cervix [17], tuberculosis granuloma in the lung [18], lung cancer [19], and atherosclerotic plaques [20]. Investigating HIV persistence in tissues is critical for HIV eradication, in order to target HIV-infected cells with distinct mechanisms of persistence, reinvigorate immune effector function in tissue microenvironment, and enhance accessibility of immune effectors to HIV-infected cells within immune sanctuary and anatomical sanctuary sites.

Spatial transcriptomic profiling uncovers mechanisms of HIV persistence and immune evasion

The difference of HIV tissue profiling between traditional single-cell RNA-seq in cell suspensions versus spatial profiling (Figure 1A) is the ability to identify (a) not only HIV-infected cells, but also immune cells around them in physical proximity (distance), (b) not only highly expressed genes in HIV-infected cells, but also a gradient of genes in the cells surrounding HIV-infected cells (the neighborhood), (c) not only bioinformatically computed cell-cell interactions, but also in situ visualization of ligand-receptor interactions in sender-recipient cells. These tools will discover (a) mechanisms of HIV persistence in the tissue compartment, such as expressing genes that promote the survival of HIV-infected cells or through ligand-receptor interactions with neighboring cells, (b) how HIV-infected cells escape immune clearance, depending on the immune effectors (such as CD8+ T cells) surrounding HIV+ neighborhoods and their respective function (such as cytotoxic, exhaustion, and proliferation gene expression in these immune effectors), and (c) different tissue microenvironment at different stages of HIV infection, such as acute infection, viral suppression, viral rebound, such as gradients of inflammatory cytokines and differences in immune cells recruited to the HIV+ cells. Overall, spatial profiling may nominate mechanisms of HIV persistence, immune evasion, markers for viral rebound, and therapeutic targets for HIV elimination.

Figure 1. Spatial transcripomic profiling at near-single cell resolution and genomewide breadth identifies mechanisms of HIV persistence in tissues.

(A) Discoveries made possible by spatial transcriptomic profiling that single-cell RNA-seq from suspensions of cells cannot achieve: (a) cytokine gradients surrounding HIV-infected cells and neighborhoods, (b) distance between HIV-infected cells (and their survival and immune escape programs) and immune effectors (and their function versus exhaustion phenotypes), and (c) cell-cell interactions, as examined by ligand expression from sender cells and receptor expression on adjacent recipient cells. (B) Caveat of spatial transcriptomic profiling: cell boundary definition (cell segmentation) is key to accuracy. Users should examine cell boundary annotation in respective programs and do not solely rely on default settings. Until cell segmentation incorporates 3D location of RNA transcripts, current cell segmentation methods suffer from mixture of RNA transcripts from cells underneath the annotated cell of interest. (C) Resolution of spatial transcriptomic platforms. (D) The number of genes captured and the size of the capture area. (E) Recommended workflow for the selection and analysis of spatial transcriptomic profiling.

Why spatial transcriptomics – genomewide capture of immune responses preserving tissue organization and cell-cell interactions

Fluorescent in situ hybridization (FISH) [11, 21–23], flow cytometry [24–27], mass cytometry, and single-cell RNA-seq-based [28, 29] methods profile HIV-infected cells and immune effectors in tissues. However, FISH-based (up to 12 targets) and mass cytometry-based (up to 50 targets) methods can only capture a limited number of selected RNA or protein of interest and may not be sufficient to interrogate multiple cell types and immune pathways at the same time. Flow cytometry, CyTOF, and single-cell RNA-seq based methods examine cell suspensions, and information of tissue organization, compartmentalization, and cell-cell interactions (the distance between and respective ligand-receptor interactions between cells, such as an HIV-infected cell versus an immune effector) are lost.

Spatial transcriptomic profiling, named method of the year in 2020 by Nature Methods [30], represents an essential advancement in elucidating the complexities and heterogeneity immune cell interactions in tissues. Several important review articles nicely summarized and compared recent advances in spatial transcriptomics [31–38]. First, the number of targets substantially increased to hundreds or thousands of targets in targeted capture [39–41] and unbiased genomewide (~20,000) profiling in poly-A-based capture [42–49]. Second, the versatility of these platforms extending from RNA capture to other modalities, such as transcription factor accessibility (by spatial ATAC-seq) [50, 51], protein expression [42], and even T cell receptor (TCR) profiling [52, 53]. These advancements in spatial profiling have discovered mechanisms of cancer persistence in the tumor microenvironment [54], such as evolution of cancer clones [55], exhaustion of tumor infiltrating lymphocytes (TILs) [56], and immune-suppressive tumor associated macrophages (TAM) [57]. We hope that the advancement in spatial profiling can similarly accelerate our understanding of HIV persistence in tissues by identifying mechanisms of HIV-infected cell survival and immune effector dysfunction.

Caveats of spatial transcriptomics – rigorous negative controls and cell type calling are required to avoid false discoveries

While these spatial technologies seem innovative and promising, several caveats need to be considered with caution. First, spatial technologies should be used to answer a biologically important question, not to find signatures with no biological meanings. For example, enrichment of HIV in B cell follicles has been rigorously demonstrated by low-throughput but orthogonal methods such as RNA FISH [11]. The use of new technologies needs to be designed to discover new mechanisms, not to confirm findings that can be done by orthogonal methods. Second, if spatial technologies do not capture the biological processes of interest, these resource-demanding platforms should not be used. For example, biological processes involving protein processing (such as degradation through ubiquitination, kinase activation through phosphorylation, caspase cleavage, and nuclear translocation for transcription factor function) cannot be answered by RNA profiling. Third, negative control of HIV+ cells and cautious cell type annotation need to be rigorously examined to avoid false positive detection of HIV in bizarre cell types. Biological sanity check and detailed dissection of individual cell types are required. Using default settings of commercial spatial technologies for cell type calling will likely lead to wrong assignment of cells and thus false discoveries [58]. Fourth, regardless of cell segmentation method used, spatial technologies inevitably assign RNA from different cells (aligned from top to bottom at the 5–10 μm cross section) as single cell, leading to T cells expressing B cell markers (Figure 1B). Finally, findings from omics-based technologies, whenever possible, should be validated by orthogonal wet-lab approaches to confirm rigorous and generalizable biological insights.

Choosing between platforms – resolution, number of target genes, and sensitivity

The choice of spatial profiling platforms should depend on the biological question of interest. There are five technical determinants of decisions – resolution, number of target genes, sensitivity of target detection, tissue size, and multi-omic coupling (Figure 1C–1D).

Resolution

Advanced spatial transcriptomic platforms can reach near-single cell resolution (such as Visium HD, Stereo-Seq, Xenium, CosMx, Seq-Scope, Open-ST, and Nova-ST) and reduce pooling mixtures of cells (such as Visium, DBiT-seq, Slide-Seq, and GeoMx). The first iterations of spatial transcriptomic methods, such as 10x Genomics Visium, profiles regions of 55 μm in diameter (Figure 1C). Given that CD4+ T cell is around 10 μm in diameter, 55 μm resolution captures a mixture of gene expression from many different cells. While there are bioinformatic programs (such as SPOTlight [59]) which attempt to estimate the different cellular composition within the 55 μm region, it is challenging to tease out the cellular program of HIV+ cells when they are mixed with uninfected cells. Advancement in single-cell platforms improves resolution to 10 μm (GeoMx [60], DBiT-seq [42], Slide-seq [61]), 2 μm (Visium HD) [62], 0.6 μm (based on the cluster distance in the three Novaseq flow cell-based spatial transcriptomic methods Seq-Scope [44], Open-ST [46, 47], and Nova-ST [48]), 0.5 μm (StereoSeq [49]), and 0.1 μm [based on the resolution of the microscope in MERSFISH [40], CosMx [46], Xenium (380-plex), and Xenium Prime 5K] [63]. While advanced methods claim single-cell or subcellular resolution, as mentioned above, spatial technologies inevitably assign RNA from different cells (aligned from top to bottom at the 5–10 μm cross section) as single cell, leading to T cells expressing B cell markers. Therefore, the so-called single cell profile in spatial transcriptomics needs to be interpreted with caution. Importantly, commercial default cell segmentation methods (calling cell boundaries and thus define transcripts within a so-called single cell) may not be accurate. Biological sanity checks are required to capture single-cell transcriptome closer to ground truth.

Number of target genes versus sensitivity of detection

Genomewide oligo-dT-based capture makes discoveries but may not capture lowly expressed genes.

The human genome contains approximately 20,000 protein-coding genes. Oligo-dT-based sequencing methods captures the polyadenylated mRNA and can presumably capture all protein coding mRNAs (and some non-coding RNA that are polyadenylated), such as Visium, Slide-Seq, DBiT-Seq, Slide-seq, Stereo-seq, Seq-Scope, Open-ST, and Nova-ST (Figure 1D). Therefore, oligo-dT-based sequencing platforms are genome-wide, allowing discovering genes not previously known to be important in biological process of interest in tissues and potentially as novel therapeutic targets. The strength in genome-wide capture comes with the inevitable weakness of sensitivity of target detection. Typically, if one cell is sequenced at 20,000 reads to 40,000 reads per cell, the most abundant RNA reads come from housekeeping genes, which does not distinguish different cell types. The next abundant RNA reads may be cell specific genes, such as CD3E and CD8A for CD8+ T cells and CD19 and MS4A1 (CD20) for B cells. Genes that can further distinguish cell subsets at a more granular level may not be expressed at high levels. For example, CD4 RNA expression is known to be low even in purified CD4+ T cells having high CD4 surface protein expression. Transcription factors (low expression level) and cytokines (short half-lives) are also known to be low in abundance in RNA-seq and thus may not be captured if sequencing depth is insufficient. Genes that are expressed in even lower levels, such as HIV RNA, may require much deeper sequencing depth to capture. Unfortunately, some of the lowly expressed genes may be degraded if the tissue quality is suboptimal. Therefore, oligo-dT-based genomewide RNA-capture can identify previously unknown targets but requires deeper sequencing and good tissue quality. If top differentially expressed genes between cell subsets are mainly housekeeping genes (such as transcription and translation machinery) or mitochondrial genes (indicating tissue degradation), such result indicates the lack of capturing key immune programs (such as CD4 and CD8 T cell effector genes), suggests poor tissue quality or processing, and should not be over-interpretated. Of note, although most platforms describe high quality RNA capture and excellent cell-type specific gene detection, some tissues used for benchmarking (such as mouse embryo, brain, and tumors) have much higher gene expression than human lymphocytes (such as CD4 and CD8). Thus, human lymph node spatial profiling results may be harder than other profiling tissues having higher transcription levels, such as tumors or the brain.

Targeted capture enriches lowly expressed genes but may miss novel targets of importance.

Targeted capture of gene of interest using hybridization-based probes and amplification avoids capturing the highly abundant (but not informative) housekeeping genes. increases the sensitivity of detecting lowly expressed genes of interest, and may be used for degraded tissues. Such platforms include RNAscope (~12 targets), GeoMx DSP (>18,000 targets), MERFISH (~10,000 targets), Xenium (~5,100 targets), CosMx SMI (~18,000 targets), seqFISH+ (~10,000 targets), and Visium HD (~20,000 targets). If the gene list is designed carefully and is sufficient to capture genes of interest, targeted capture may better capture immune phenotypes (focusing on lowly expressed signature transcription factors and cytokines) and detect lowly expressed genes (such as HIV). However, given that these targeted capture methods rely on high-resolution microscopy, the optical hindrance (that different highly expressed genes in the same cellular location may interfere with detection of the others), nonspecific staining (background noise and false positive detection), and the requirement of specific instruments. The key caveat is that if the target probe set only confirms existing knowledge and does not include novel genes to probe previously unknown biological process, the chance of making new discoveries may be lower than genome-wide unbiased profiling. Overall, microscopic imaging of targeted probes (such as Visium HD, Xenium, MERFISH, GeoMx, and CosMx) may suffer from optic hindrance and decreased sensitivity when target number increases, while Nova-Seq sequencing chip-based platforms (such as Seq-Scope, Open-ST, and Nova-ST) allows genomewide poly-A-based capture but relies on tissue quality and sequencing depth.

Tissue RNA quality.

Given that targeted capture may identify lowly expressed genes from degraded tissues, the tissue RNA quality should be taken into consideration. For example, for good quality tissue (DV200 ≥ 50%, i.e. the proportion of RNA fragments >200 nucleotides, is ≥50%), genome-wide profiling methods can be used. However, for degraded tissue (such as DV200 <50% or formaldehyde fixed tissues), targeted capture should be considered [64].

Tissue size versus the capture field size

Most spatial transcriptomic profiling methods only have a small capture area – from 1 mm² to 2 cm² (Figure 1D), which is much smaller than a typical histology slide (25 mm x 75 mm). This is because profiling a full slide requires substantial sequencing cost (of billions of reads to cover the whole slide), probe reagent cost (to maintain optimal concentration across the whole slide), or lengthy time for high-resolution microscopic imaging. While these small field sizes can be helpful for small tissues having dense cell types of interest (such as mouse embryo, brain, and tumor), a much larger capturing field will be needed to capture the rare HIV-infected cells among tissues, such as the gut and the brain.

Multi-omic coupling – expanding spatial RNA-seq to coupled ATAC-seq and protein capture

Extending from RNA-seq-based profiling, new advances enable spatial ATAC-seq to capture transcription factor accessibility and protein profiling, and even coupling these technologies with RNA-seq-based profiling. While ATAC-seq allows capturing transcription factor accessibility, current platforms cannot reach single-cell resolution yet. These gradients of transcription factor activity, unlike single-cell ATAC-seq profiling, may be helpful for embryogenesis (following caudal-rostral gradients of transcription factor activities) but may not yet be ready for single-cell immune profiling.

Spatial protein profiling, such as CODEX [65, 66] and PANINI [67], can be coupled with spatial transcriptomics and overcomes the caveat of low RNA expression of genes of interest. While the number of protein available is still limited, the ongoing advancement in technology is currently increasing the number of protein capture and therefore the breadth and robustness of coupled RNA-protein multiomic profiling.

Workflow of spatial profiling bioinformatic analysis to avoid false positive discoveries

Here, we present a recommended workflow of spatial profiling (Figure 1E). The accuracy of biological assays relies on removing low quality noise, correct annotation of cell types, and identification of differences (such as differential gene expression or different composition of cell types) between the experiment and the control. For example, in flow cytometry, most scientists follow such standard procedures, to gate away low-quality cells (using forward and side scatter), doublets (using area-versus-height or width-versus-height in forward and side scatter), and dead cells (by viability staining). Then, cell type annotation (such as CD4+ versus CD4– cells) and differential gene expression relies on gating strategies based on robust negative controls (such as unstained or isotype control) and positive controls. Of note, such streamlined standard procedures are not yet readily available in commercialized platforms. In this review, we outline these important steps for scientists to consider in their own analyses. Using results from direct output from commercial platforms, without investigator-driven quality control and sanity check, may lead to false discovery of HIV in a wrong cell type or nomination of cellular pathways or marker genes that are far from ground truth. Of note, these procedures are currently evolving in the spatial transcriptomics field. Readers are encouraged to also refer to benchmarking studies [68–71] and review articles [38, 72]. Overall, the goal of sophisticated bioinformatic analyses is not to generate colorful plots of signatures of no biological meanings – the goal is to use rigorous bioinformatic analysis and biological sanity checks to reach as close as possible to ground truth for pathways, cell types, and markers that can be validated by orthogonal wet-lab methods.

Quality control

The first question after obtaining spatial transcriptomics data would be – is the quality good for biologically informative interpretation? In a typical flow cytometry, scientists gate away from cell debris and dead cells to avoid false nomination of biological process that are simply reflecting dead cells (in which only mitochondrial, cellular degradation, and housekeeping genes can be detected and do not reflect the actual cellular phenotype before death).

Number of genes and RNA count per cell.

In the context of spatial transcriptomics, the first quality control parameter is the number of genes per cell, or the number of RNA molecules per cell [which can be counted as unique molecular identifiers (UMIs) or RNA count] [33]. While single-cell RNA-seq in suspension can typically capture ~2,000 genes per cell, spatial transcriptomic platforms typically capture much fewer genes, as low as 200–1,000 genes per cell. While there has not been a consensus on the threshold for low-quality of cells and doublets (like those thresholds used in single-cell RNA-seq of suspension cells), a 250 UMI cutoff for 10 μm squared grids in SeqScope [73], a 500–45,000 UMI cutoff for 55 μm Visium spots [74], and a 10 transcripts for cell cutoff in CosMx [63] have been reported.

We recommend three immediate parameters as biological sanity check, which are not included in commercial platforms – cell segmentation, highly expressed genes, and HIV detection in negative controls.

Cell segmentation.

First, perform biological sanity check on cell segmentation methods - does the cell boundary calling by the spatial platform make biological sense? For example, a lymphocyte typically has a 9:1 nucleus: cytoplasm (N:C) ratio and ~10 μm in size. Users should examine cell boundary calling by eye and ask – are there (a) a presumably round lymphocyte, ~10 μm in diameter, that was drawn in to much larger areas that contain transcripts from other cells, (b) multiple cells being merged together and called one cell (wrong segmentation), (c) areas without good cell staining being called as cells (false positive), and (d) areas of good cellular staining that was not annotated as cells (false negative, missing cells)? Then, users should optimize parameters based on expected cell size, nuclei size, cell-cell distance and minimize these artifacts, and try different cell segmentation programs. Relying solely on default settings may generate false positive discoveries.

Cell segmentation significantly influences the accuracy of spatial transcriptomic analyses. Although each spatial profiling platform has its own optimized segmentation method, current methods are not yet absolutely accurate. Cell segmentation methods are broadly classified into image-based versus transcript-based approaches. Image-based segmentation (such as Xenium default multi-modal cell segmentation, CellPose [75] and QuPath [76]) relies heavily on tissue imaging quality based on hematoxylin and eosin (H&E), DAPI (nuclear staining), and cell membrane staining for cell boundary calling. Transcript-based segmentation refinement methods (such as Baysor [77] and Segger [78]) leverages spatial transcriptomic data combined with reference single-cell RNA sequencing data to refine image-based segmentation and thus improves accuracy by correcting potential artifacts or errors.

Highly expressed genes.

Second, examine the most highly expressed genes in the dataset. If lymphoid tissues are profiled, are typical T lymphocyte genes (such as CD3E, CD8A) and B lymphocyte genes [such as MS4A1 (CD20), CD19] among the most highly expressed genes? If the most highly expressed genes are mitochondrial genes, ribosomal proteins, and transcription machinery, it is likely that tissue or RNA degradation prevents the discovery of true markers that shows immune cell composition in the tissue.

Set HIV detection threshold using negative controls.

Third, examine HIV positive reads in an uninfected sample. Such negative control samples should always be included in spatial transcriptomic profiling to set rigorous threshold for HIV detection and to avoid false positive calling of HIV because of nonspecific staining.

Cell cluster visualization and annotation

Visualization of cell type clusters.

Gene expression profiles should be normalized by SCTransform [79] or stSME [80] after spatial domain identification depending on the library size variation. While there are robust batch effect correction (integration) methods in single-cell RNA-seq of cell suspension, batch effect correction methods in spatial transcriptomics remains limited [81, 82]. Using bioinformatic pipelines [83, 84] to nominate different cell type clusters by gene expression profiles and their spatial location, different cell clusters can be visualized on a dimension reduction plot, such as UMAP (uniform manifold approximation and projection) [85, 86] or on spatial locations by spatial clustering [81]. The number of clusters should not be based on a default setting or an arbitrary number. Rather, the users should use knowledge from tissue atlas (such as lymphoid tissue atlas [87, 88] or gut atlas [89]) to build a biological insight on the number of cell types expected in the tissue.

Cell type annotation.

Then, the users can subcluster each cell clusters, examine highly expressed genes on a heatmap, examine cell type-defining genes on a dot plot (of the percent and level of key genes of interest in each cluster), and use biological knowledge and tissue atlas references to annotate each cell type (rather than giving cluster numbers without biological meanings). Of note, as mentioned above, the limit of spatial profiling is that regardless of cell segmentation method used, a so-called well-defined single cell may still contain transcripts from cells underneath, i.e. a T cell may also have B cell or FDC gene expression. Interpretation should be made with caution.

Cell type annotation at the spatial level.

The power of spatial transcriptomics is the ability to visualize gene expression in original tissue organization. When naming cell clusters, users should examine the distribution of each cluster and name them accordingly, i.e. taking tissue location into consideration for more accurate cell type calling. As a sanity check, users should examine whether cell type annotations make biological sense, such as visualizing T cell clusters in the T cell zone and B cell clusters in B cell follicles in lymphoid tissues.

Cell type deconvolution.

One critical caveat of spatial profiling is mixed capture of multiple cells types in the same spot which can conceal the genuine cellular profile. In platforms having low resolution and units containing multiple cell types, such as Visium, Slide-seq, DBiT-seq, and GeoMx, further analysis is recommended to dissect the proportion of cell type mixtures in the captured spots, so-called cell type deconvolution. Based on a benchmarking study [68], CARD [90], Cell2location [91], Tangram [92], RCTD [93], and SPOTLight [59] can be considered for cell type deconvolution.

Making discoveries

Differential gene expression identifies cellular markers of pathways of HIV+ cells or neighborhoods.

One strength of spatial profiling is the ability to define immune cells adjacent to versus away from HIV-infected cells [14, 23]. In addition to identification of HIV-infected cells, the neighborhoods of viral infection (surrounding HIV-infected cells) [67] can be delineated. Cellular composition analysis, differential gene expression analysis, and RNA expression distance analysis to identify differences between HIV neighborhoods versus those away from HIV-infected cells within the same tissue, or between different biological conditions (such as acute versus chronic viral infection). These analyses may nominate immune pathways surrounding HIV-infected cells and nominate markers as therapeutic targets.

Cell-cell interaction analysis identifies bona fide ligand (from sender cells) – receptor (on receiver cells) interactions in adjacent cells.

Spatial cell-cell communication refers to the interactions between neighboring cells within their tissue microenvironment, mediated through ligand-receptor signaling pathways, cytokines, growth factors, or direct physical interactions [94]. Evaluation of cell-cell communication relies on three key information: cell location, ligand-receptor database, and the RNA expression in each cell. Advanced bioinformatic tools (such as CellChat2 [95] and COMMOT [96]) integrate cell-cell distance, gene expression of cytokine ligand from sender cells and cytokine receptor expression in receiver cells, and visualize such interactions with not only ranked probability of interaction but also in the spatial context. For example, CellChat2 focuses on a database of ligand-receptor interactions and COMMOT uses a computational approach to infer communication in space. Overall, spatial cell-cell interaction analysis advances bioinformatic predictions into in situ visualization of ligand-receptor interactions between cells in physical contact.

Make discoveries that can be validated, not signatures of no biological meanings.

We would like to reiterate that the output of single-cell multiomic analysis and spatial profiling should not be colorful plots or signatures without biological meanings. Mechanisms of HIV persistence, pathways facilitating immune evasion, and markers of HIV-infected cells should be validated by orthogonal wet-lab approaches, such as FISH, immunofluorescent staining, or immunohistochemistry (IHC).

Conclusions

Tissue samples from people living with HIV requires altruistic donation and invasive procedures. The use of spatial profiling maximizes the capacity of discovery from a handful of markers of interest to genome-wide probing, sub-cellular resolution, with spatial context of tissue compartments. The current spatial profiling platforms and bioinformatic analyses are still under development, with encouraging ongoing progress in (a) coupling protein detection with RNA profiling (such as DBiT-seq, MERFISH, Xenium, and CosMx), (b) increasing the capture area, genome-wide breadth, with lower cost (such as Seq-Scope, Open-ST, and Nova-ST), (c) targeted probe design to allow capturing lowly expressed genes, and (d) improvement of cell segmentation and cell-cell interaction bioinformatic methods. As single-cell multiomics in methods of suspension cells becoming one of the essential tools in immunology and cancer studies, spatial profiling in tissues has advanced cancer and immunology discoveries and will accelerate discoveries in HIV eradication and microbial-host interactions.

Key points.

1. Spatial profiling can identify cytokine gradients in HIV+ neighborhoods, measure distance between HIV-infected cells and immune effectors (and capture respective immune effector function), and examine cell-cell interactions through ligand-receptor expression from adjacent sender-recipient cells.

2. Advanced spatial transcriptomic platforms can reach near-single cell resolution (such as Visium HD, Stereo-Seq, Xenium, CosMx, Seq-Scope, Open-ST, and Nova-ST) and reduce pooling mixtures of cells (such as Visium, DBiT-seq, Slide-Seq, and GeoMx).

3. Microscopic imaging of targeted probes (such as Visium HD, Xenium, MERFISH, GeoMx, and CosMx) may suffer from optic hindrance and decreased sensitivity when target number increases, while Nova-Seq sequencing chip-based platforms (such as Seq-Scope, Open-ST, and Nova-ST) allows genomewide poly-A-based capture but relies on tissue quality and sequencing depth.

4. The best practice of of spatial profiling should include rigorous cell boundary definition (cell segmentation), quality control (number of genes and RNA captured per cell), santity check (detection of genes that capture different cell types instead of housekeeping genes and ribosomal proteins), stringent threshold for HIV detection based on negative controls, insightful annotation of cell clusters (not arbitrary cluster numbers), and findings that can be validated and close to ground truth (not signatures without biological meanings).