Abstract
Cellular interactions are essential for tissue organization and functionality. In particular, immune cells rely on direct and usually transient interactions with other immune and non-immune populations to specify and regulate their function. To study these “kiss-and-run” interactions directly in vivo, we previously developed LIPSTIC (Labeling Immune Partnerships by SorTagging Intercellular Contacts), an approach that uses enzymatic transfer of a labeled substrate between the molecular partners CD40L and CD40 to label interacting cells. Reliance on this pathway limited the use of LIPSTIC to measuring interactions between CD4+ helper T cells and antigen presenting cells, however. Here, we report the development of a universal version of LIPSTIC (uLIPSTIC), which can record physical interactions both among immune cells and between immune and non-immune populations irrespective of the receptors and ligands involved. We show that uLIPSTIC can be used, among other things, to monitor the priming of CD8+ T cells by dendritic cells, reveal the cellular partners of regulatory T cells in steady state, and identify germinal center (GC)-resident T follicular helper (Tfh) cells based on their ability to interact cognately with GC B cells. By coupling uLIPSTIC with single-cell transcriptomics, we build a catalog of the immune populations that physically interact with intestinal epithelial cells (IECs) and identify a receptor-ligand pair necessary for the interaction between IECs and intraepithelial lymphocytes (IELs). Thus, uLIPSTIC provides a broadly useful technology for measuring and understanding cell–cell interactions across multiple biological systems.
Introduction
Physical interactions in which cells use membrane-bound molecules to exchange signals with each other are at the core of multiple tissue functions1, 2. In the immune system they feature prominently, from the priming of T cells by dendritic cells (DCs) that initiates the adaptive immune response to the CD4+ T cell that enables B cells to produce high-affinity antibodies3, 4. More recent work has explored the role of similar interactions between immune and non-immune cells, such as those forming the epithelial barrier of the gut and skin, which are thought to drive transcriptional changes in immune cells that in turn enable them to support tissue function5–8. Despite their importance, direct observation of cell-cell interactions has traditionally been done by microscopy and other forms of imaging9, 10, which have the common limitation that interacting cells cannot be retrieved for downstream analysis11. Thus, the impact of the interaction on cell behavior and the cellular features that lead the interaction to occur in the first place cannot be inferred from traditional imaging data alone. More recently, spatial transcriptomics and high-density imaging technologies have allowed for a more in-depth characterization of the states of cells in the same neighborhood12. Although the more widely available technologies involve a tradeoff between high spatial resolution13 and full transcriptome coverage14, custom platforms can achieve gene expression profiling at cellular size resolution15, 16. However, even when capable of high resolution, transcriptomic and imaging techniques still report on proximity between cells rather than on true physical interaction and signal exchange between membranes, requiring additional indirect methods and assumptions to infer functional interactions computationally17–22. High throughput identification of cellular interactors and full deconvolution of the transcriptomic effects of physical interaction on cellular behavior and function are therefore yet to be achieved.
An attractive strategy to overcome these limitations has been proximity-based labeling across cellular membranes23–27. These strategies rely on equipping “donor” cells with enzymes or other signals that act over short distances to identify “acceptor” cells in either close proximity or physical contact with these donors. An early example of such an approach was our development of LIPSTIC, which uses enzymatic labeling across immune synapses to directly record cell-cell interactions in vivo24. In its first iteration, we engineered LIPSTIC to label interactions driven by a specific receptor-ligand pair—CD40L-CD40—a dependency that restricted its utility to interactions involving effector CD4+ T cells, the primary expressors of CD40L. Here, we report the development of a universal (u)LIPSTIC tool, which enables us to record interactions between multiple cell types of immune and non-immune origin. Coupling of uLIPSTIC with standard single-cell mRNA sequencing (scRNA-seq) methods allows for atlas-type characterization of the “cellular interactome” of a population of interest and even for the discovery of surface molecules that drive some of these interactions. Thus, uLIPSTIC enables us to achieve truly quantitative interaction-based transcriptomics without need for computational inference of transcriptomes or interacting molecules.
Results
A receptor/ligand-agnostic approach for recording cell–cell interactions
LIPSTIC uses the Staphylococcus aureus transpeptidase Sortase A (SrtA)28 to covalently transfer a peptide substrate containing the motif LPETG onto an N-terminal pentaglycine (G5) acceptor. In its original version, catalysis by the very low-affinity (∼1.8 mM) interaction between LPXTG-loaded SrtA and its G5 target29, 30 was favored by genetically fusing each component to one of the members of a receptor–ligand pair24, thus raising local concentration of the reactants above the threshold required for substrate transfer (Fig. 1A). We reasoned that a similarly high local concentration of enzyme and target could also be achieved in a receptor–ligand independent manner by driving very high expression of SrtA and G5 on apposing cell membranes without direct fusion to the interacting molecules. Transfer of substrate under these conditions would provide a universal readout for physical interactions between cells of any type, even in the absence of prior knowledge of the surface molecules that drive these interactions (Fig. 1B).
To test this strategy, we generated a donor–acceptor pair consisting of the “PDK” version of SrtA30 targeted to the plasma membrane by fusion to the human PDGFRB transmembrane domain24 (mSrtA) and the G5 peptide fused to the N-terminus of the mouse Thy1.1 protein, a single glycosyl-phosphatidyl-inositol (GPI)-anchored immunoglobulin domain which is widely used as both an allelic marker and a reporter for retroviral transduction. 3D modeling (Fig. 1C) predicted the maximal distance between membranes at which label transfer would occur to be approximately 14 nm (the total distance spanned when the PDGFRB transmembrane domain is fully extended). This distance is comparable to the inter-membrane span required, for example, for the TCR-MHC interaction (∼15 nm), and narrower than the typical distance separating juxtaposed cell membranes in the absence of receptor-ligand interactions, set by glycocalyx repulsion31. Given the negligible affinity (∼1.8 mM) between SrtA-LPETG and G5 30, such a design would in principle allow for label transfer only when cells were functionally interacting at a close intermembrane distance, without driving artificial interactions between its engineered components.
To gauge the functionality of such an approach, we first transfected separate pools of HEK293T cells with high or low concentrations of plasmids expressing either mSrtA or G5-Thy1.1, then carried out uLIPSTIC labeling by addition of biotin-LPETG substrate to co-cultures of these cells as described24 (Fig. 1D). Labeling of G5-Thy1.1-transfected cells was augmented when donor and acceptor populations were forced to physically interact by co-transfection of constructs encoding CD40L and CD40, respectively, and further increased when the uLIPSTIC components were transfected at the highest concentration (Fig. 1E). Thus, high-level expression of SrtA and G5 on the membranes of interacting cells allows for LIPSTIC labeling to proceed without need for fusion to specific receptor–ligand pairs, in principle enabling universal labeling of physical interactions between cells of any type.
Based on these initial findings, we generated a Rosa26uLIPSTIC mouse allele in which high expression of mSrtA (preceded by a FLAG-tag) or G5-Thy1.1 are driven by the strong cytomegalovirus early enhancer/chicken beta-actin/rabbit beta-globin (CAG) promoter introduced into the ubiquitously expressed Rosa26 locus, using a previously published targeting strategy32 (Fig. 1F and Fig. S1). The upstream G5-Thy1.1 is flanked by LoxP sites, so that Cre-mediated recombination leads to expression of a previously silent downstream mSrtA, switching Cre-expressing cells from uLIPSTIC acceptors into uLIPSTIC donors. The specificity and efficiency of this recombination are determined by the Cre driver used (Fig. S2). To test this system in primary immune cells ex vivo, we crossed Rosa26uLIPSTIC mice to the CD4-Cre and OT-II T cell receptor transgenes to generate mSrtA+ uLIPSTIC donor T cells specific for peptide 323-339 of the model antigen chicken ovalbumin (OVA). Co-culture of mSrtA+ donor T cells with G5-Thy1.1+ B cells showed efficient transfer of labeled substrate between T and B cells only in the presence of OVA323-339, whereas background labeling by non-interacting cells in the absence of OVA323-339 was negligible (Fig. 1G). Substrate transfer was completely prevented by addition of a blocking antibody to MHC-II, necessary for the cognate recognition of B cells by T cells, but not by an antibody to CD40L, which is not itself required to drive this interaction (Fig. 1G). We conclude that uLIPSTIC provides an approach for trans-synaptic labeling of intercellular contacts that is functional regardless which receptor(s) and ligand(s) drive these interactions. Importantly, preventing functional B cell–T cell interactions by blocking antigen presentation completely abrogated label transfer between cells, indicating that high expression of the uLIPSTIC components is itself not sufficient to drive unwarranted cell–cell interactions.
uLIPSTIC labeling of cell–cell interactions in vivo
To test the activity of uLIPSTIC in vivo, we used the antigen-specific interaction between DCs and CD4+ T cells. Since this interaction involves the CD40L–CD40 axis, we could benchmark the performance of uLIPSTIC against the original LIPSTIC system24. To this end, we used a classic in vivo T cell priming model9, 33, where G5-Thy1.1+ DCs loaded with OVA323-339 are injected into the footpads of mice followed by adoptive transfer of mSrtA+ OT-II T cells. Lymphatic migration of DCs to the draining popliteal lymph node (pLN) allows DC–T cell interactions to take place at this site (Fig. 2A). Footpad injection of biotin-LPETG substrate 24 h after T cell transfer led to detectable labeling of on average 6.5% of transferred DCs (Fig. 2B). Comparable numbers were obtained when using the original CD40L–CD40 LIPSTIC system24 (Fig. 2C). Treatment with anti-MHC-II prior to substrate injection blocked labeling in both settings (whereas treatment with anti-CD40L blocked transfer only by the original LIPSTIC), indicating that expression of the uLIPSTIC components alone is insufficient to artificially drive interactions between neighboring cells also in vivo. Thus, uLIPSTIC labeling is equivalent to receptor–ligand-specific LIPSTIC for recording the binding patterns of CD4+ T cells and DCs during in vivo priming. Of note, transferring increasing numbers of mSrtA+ donor T cells increased the degree to which interacting DCs were labeled (Fig. S3A). Thus, uLIPSTIC labeling efficiency varies with the availability of donor cells, likely as a consequence of multiple donor T cells being primed by the same acceptor DC.
We next used uLIPSTIC to record interactions between T cells and DCs not accessible to the original LIPSTIC system, either because they do not involve the CD40L/CD40 interaction or because directionality is inversed. In an in vivo priming model, mSrtA+ OT-I CD8+ T cells were found to detectably label on average 14% of DCs pulsed with their cognate peptide (OVA257-264) but only background levels (0.6%) of DCs pulsed with a control peptide from lymphocytic choriomeningitis virus (Fig. 2D). We then inverted the uLIPSTIC reaction so that endogenous mSrtA+ DCs (in Rosa26uLIPSTIC/+.Clec9aCre/+ mice34) were made to interact with adoptively transferred CD4+ (OT-II) T cells from Cre-negative Rosa26uLIPSTIC/+ donors upon immunization with OVA in alum adjuvant (Fig. 2E). This setup led to detectable labeling of roughly 20% of transferred T cells, which was again fully abrogated by prior injection of a blocking antibody to MHC-II (Fig. 2F-G). Thus, uLIPSTIC can label interactions between T cells and DCs bidirectionally.
To test the ability of uLIPSTIC to record immunological interactions other than priming of naïve T cells by DCs, we chose two settings. We first determined the identity of the cellular partners of regulatory T (Treg) cells in the steady state lymph node. To this end, we used the Foxp3CreERT2 allele35 to achieve tamoxifen-dependent recombination of Rosa26uLIPSTIC specifically in Treg cells. We treated Rosa26uLIPSTIC/+.Foxp3CreERT2/Y mice35 with tamoxifen to induce mSrtA expression in Treg cells, and then administered LIPSTIC substrate to footpads to achieve labeling in the draining pLNs (Fig. 3A,B). This revealed pronounced labeling of most DCs with migratory (MHC-IIhiCD11cint) phenotype, whereas labeling of resident (MHC-IIintCD11chi) DCs was markedly lower (Fig. 3B,C). Importantly, labeling of Foxp3− CD4+ T cells was negligible in this setting, confirming that simple colocalization of these cells with donor Treg cells within the same microenvironment is not sufficient to drive label transfer (Fig. 3B, left). Expression of mSrtA− in roughly equivalent numbers of Treg cells or total conventional CD4+ T cells (the latter achieved by low-dose tamoxifen administration to Rosa26uLIPSTIC/+.CD4-CreERT2 mice36) resulted in much less efficient labeling of migratory DCs by conventional T cells (Fig. S3B,C). Thus, interaction with migratory DCs at steady state, although not a unique property of Treg cells, is more pronounced among this subset (Fig. S3D). Treg cell labeling of migratory DCs was decreased but not completely abrogated by administration of a blocking antibody to MHC-II, confirming that the strong interaction between Treg cells and migratory DCs is partly driven by the TCR-MHC-II axis but suggesting that other receptor-ligand pairs may also contribute to this process (Fig. 3B,C).
As a second test case, we used uLIPSTIC to determine the phenotype of the T cells that provide help to B cells in germinal centers (GCs), which are often difficult to identify unambiguously using the classic Tfh markers CXCR5 and PD-137. To this end, we crossed Rosa26uLIPSTIC to the AicdaCreERT2 allele, which allows for tamoxifen-dependent LoxP recombination in GC B cells38. We immunized Rosa26uLIPSTIC/+.AicdaCreERT2/+ mice38 in the footpads with the model antigen 4-hydroxi-3-nitro-phenylacetyl (NP)-OVA to generate GCs, then treated these mice with tamoxifen 7 and 8 days later to induce mSrtA expression in GC B cells (Fig. 3D). Footpad injection of biotin-LPETG 10 days post-immunization led to a high proportion of uLIPSTIC labeling among CXCR5hiPD-1hi T follicular helper (Tfh) cells but not among CXCR5−PD-1− non-Tfh cells in the pLN (Fig. 3E,F). Notably, only a fraction of T cells expressing lower levels of CXCR5 and PD-1 were labeled by GC B cells, indicating that relatively few of the cells within the “Tfhlow” gate are truly engaged in helping GC B cells. Again, blocking of MHC-II led to total loss of Tfh cell labeling, confirming the specificity of the reaction (Fig. 3E,F).
Finally, to test the ability of uLIPSTIC to record interactions between immune cells and non-immune partners, we measured the transfer of substrate from intestinal epithelial cells (IECs) to the large population of intraepithelial T lymphocytes (IELs) that reside within this compartment39. We crossed Rosa26uLIPSTIC mice to villin-1 (Vil1)-CreERT240, a transgene that drives LoxP recombination in IECs upon administration of tamoxifen (Fig. 3G). Intraperitoneal administration of biotin-LPETG led to efficient loading of substrate onto IECs and its abundant transfer onto a large fraction (median 65%) of the CD45+ IEL population (Fig. 3H,I and S4B). Subsetting IELs into different populations showed that uLIPSTIC labeling followed a gradient corresponding to the stage of differentiation (or adaptation to the epithelial tissue) of these cells. Whereas labeling of “natural” TCRψ8+ and CD8αα+/TCRαý+ IELs was uniformly high, uLIPSTIC signal in induced CD4+ IELs41 followed closely their developmental trajectory5, 42, from background levels in the CD4+ CD8αα−CD103− “conventional” subset to intermediate labeling in CD4+ CD8αα− CD103+ pre-IELs and levels comparable to those of natural IELs in fully epithelium-adapted CD4+ CD8+CD103+ population (Fig. 3J,K). Therefore, uLIPSTIC is capable of recording interactions between epithelial and immune cells in the small intestine. Our data show that close physical interaction with IECs is a common feature of most IELs, which is acquired in a stepwise fashion as CD4+ IELs develop.
We conclude that uLIPSTIC can be used to label a wide variety of immune cell interactions in vivo, both in adoptive transfer and in fully endogenous models. In the latter, uLIPSTIC revealed the interaction preferences of steady-state LN Treg cells, identified populations of Tfh cells capable of providing help to B cells in the GC, and showed stepwise acquisition by intraepithelial CD4+ T cells of the ability to physically interact with IECs.
Using uLIPSTIC for interaction-based transcriptomics
A key feature of uLIPSTIC is its ability to identify the full cellular interactome of a given cell type in an unbiased manner. Reading out this interactome is best achieved by single-cell RNA sequencing (scRNA-seq), which is also unbiased in its ability to identify labeled cell populations. Because LIPSTIC labeling has a wide dynamic range24, coupling it to scRNA-seq also has the potential to identify genes and transcriptional programs quantitatively associated with the degree of interaction between two cell types, which can in principle reveal the molecular pathways that drive a given interaction (Fig. 4A). To explore these possibilities, we profiled the immune interactome of IECs in the small intestine. Rosa26uLIPSTIC/+. Vil1-CreERT2 mice were labeled as in Fig. 3G and the intraepithelial immune cell (CD45+) fraction was stained with fluorescent antibodies for cell sorting plus a DNA-barcoded anti-biotin antibody to detect the uLIPSTIC signal. We sorted CD45+ immune cells in a semi-unbiased manner, using FACS gating to enrich for rarer leukocyte populations (Fig. S4C). These cells were then profiled by droplet-based scRNA-seq using the 10X Genomics platform. Immune cell populations were identified by comparing well characterized marker genes and publicly available gene signatures across the identified cell clusters, coupled with unbiased differential expression analysis and TCR sequence reconstruction (Fig. 4B, Figs. S5-7, and Supplementary Spreadsheets 1,2).
uLIPSTIC revealed broad variation in the extent to which different immune populations interact with IECs, which aligned with the data obtained by flow cytometry. Whereas there was a high degree of labeling among natural IEL subsets (ψ8, CD8αα+ and CD8αý+ T cells), labeling was low or negligible among B cell populations (including naïve-like, activated, and plasma cell subsets), and intermediate in plasmacytoid DCs (Fig. 4B). uLIPSTIC also labeled two less clearly defined leukocyte clusters that interacted strongly with IECs. These included a small population of likely myeloid cells and a larger cluster marked by high expression of genes such as Atxn1 and Btbd11, to which we were unable to assign an identity (Fig. 4C and Figs. S5A,G and S6A,B). As seen by flow cytometry (Fig. 3J,K), CD4+ T cells again showed a gradient in their ability to interact with IECs, which became more apparent when these cells were clustered into subpopulations (Fig. 4D, Fig. S7, and Supplementary Spreadsheet 3). Acquisition of the biotin label largely followed a developmental trajectory that began with a highly polyclonal naïve-like population with low uLIPSTIC signal and followed through a pre-IEL intermediate into a fully differentiated, oligoclonal CD4-IEL state5, 42 (Fig. 4D-F) labeled to a similar extent as natural IELs (Fig. 4C).
To search for the molecular drivers of the IEL–IEC interaction, we calculated the correlation between the uLIPSTIC signal within the CD4+ T cell population and the expression of all detected genes in our dataset (Fig. 4G,H). Expectedly, given the association of uLIPSTIC labeling with CD4-IEL development, we found multiple significant correlations with key markers of IEL differentiation. These included negative correlations with naïve T cell markers such as Sell (encoding for L-selectin) and Tcf7 and positive correlations with the CD4-IEL associated genes such as Ccl5, Gzma, Itgae, and Itgb7 (Fig. 4G,H, Fig. S7E,F, Supplementary Spreadsheet 4)42. Flow cytometry confirmed the strong correlation between uLIPSTIC signal and expression of Itgae and Itgb7, which together encode for the two subunits of the integrin αEý7 or CD103 (Fig. S8). This was of particular interest, because CD103 is a receptor for the IEC adhesion molecule E-cadherin, suggesting that uLIPSTIC may be able to identify receptor-ligand pairs that drive the IEL–IEC interaction. Indeed, we observed a tight positive correlation between uLIPSTIC labelling and the expression of Jaml—encoding for the surface protein junction adhesion molecule-like (JAML)—a member of the immunoglobulin superfamily with homology to junctional adhesion proteins43 (Fig. 4G,H and Fig. S7E,F). JAML is a ligand for the coxsackie and adenovirus receptor (CAR), a tight junction-associated protein that is expressed in the intestinal epithelium44–47. In agreement with gene expression data, in vivo staining with an anti-JAML antibody confirmed the stepwise acquisition of this adhesion molecule during CD4-IEL development, as well as its expression in multiple IEL subsets, including CD8αα+ and ψ8 IELs (Fig. 4I). To verify whether uLIPSTIC had indeed identified an additional receptor/ligand pair involved in the IEL–IEC interaction, we treated Rosa26uLIPSTIC/+.Vil1-CreERT2 mice with antibodies to JAML, and carried out uLIPSTIC labeling 12 hours later. Anti-JAML treatment led to substantial loss of uLIPSTIC signal in all IEL populations tested (Fig. 4K), indicating that JAML is not only a marker of CD4-IEL differentiation, but is also required for of the ability of all IELs to interact with the surrounding IECs. We conclude that uLIPSTIC can be used to achieve quantitative interaction-based transcriptomics, enabling us not only to define the “cellular interactomes” of populations of interest, but also to discern the receptor-ligand pairs that functionally drive those interactions.
Discussion
This study describes a generalization of the LIPSTIC method applicable to cells of any type that interact physically with each other. Unlike the original version of our method24, uLIPSTIC does not require cognate interaction between a pre-specified receptor-ligand pair for label transfer, allowing one to probe the full cellular interactome of a population of interest in an unbiased manner. In the absence of such a requirement, the specificity of labeling is ensured by two properties of our system. First, the interacting components span a maximum intermembrane distance of ∼14 nm, typical of many receptor–ligand interactions including that of TCR with MHC31. Second, the extremely low affinity (millimolar Km) between the SrtA-LPETG donor and the G5 acceptor renders the uLIPSTIC components unlikely to themselves initiate unwarranted interactions between neighboring cells. Such specificity is supported by two separate lines of evidence. First, labeling is abrogated when antibodies are used to block known molecular drivers of a cellular interaction, demonstrating that efficient labeling takes place only when cells are allowed to interact functionally, such as through antigen presentation. Second, not all cells that are physically juxtaposed interact to the same extent, as exemplified by the low degree of labeling, if any, of conventional T cell or resident DC acceptors by Treg cell donors. uLIPSTIC thus differs from and complements methods such as synNotch variants, which, although they can be used to drive transcription of downstream reporter genes26, are based on molecular partners that bind each other with much higher (nanomolar) affinity, and thus are in theory themselves capable of driving cellular interactions49; as well as methods based on label transfer by extracellular diffusion25, which can better be conceived of as approaches to mark close cellular neighborhoods rather than functional interactions. uLIPSTIC also has advantages over cell doublet-based methods50, in that uLIPSTIC labeling is continuous rather than binary, can study interactions that do not survive cell extraction and flow cytometry processing, and does not require computational deconvolution of single-cell transcriptional profiles from doublets, although our system has the disadvantage that it requires genetic engineering of its components.
A central feature of uLIPSTIC is that it can be coupled directly to droplet-based scRNA-seq to achieve quantitative interaction-based transcriptomics. This property can be used in both an “atlas” mode, where the objective is to identify which populations of acceptor cells interact with a given donor lineage, and in “mechanistic” mode, where correlations between uLIPSTIC signal intensity and expression of individual genes allow us to establish the molecular basis of an interaction of interest. Using the second approach, we identify the leukocyte adhesion molecule JAML as playing an important role in the interaction between IELs and IECs in the small intestine. The JAML–CAR interaction has been previously implicated in neutrophil migration across epithelial tight junctions and ψ8 T cell activation and wound healing in the skin44, 48, and its expression has been documented on ψ8 IELs in the intestine48. Our observations using anti-JAML antibodies suggest that JAML–CAR interactions are important for the physical juxtaposition of IECs and IELs, possibly relating to the ability of IELs to migrate through tight junctions or to position themselves above the basement membrane separating the lamina propria and epithelium layers6, 51.
In conclusion, uLIPSTIC provides an unbiased platform for measurement of known cell-cell interactions as well as discovery of new ones. When coupled to scRNA-seq, uLIPSTIC interaction-based transcriptomics has the ability to quantify correlations between the intensity of cell-cell interactions and gene expression, allowing for the discovery of the underlying receptor– ligand pairs that drive these interactions. We expect this tool will be broadly useful for studying cellular interactions in immunology and beyond.
Funding
This study was funded by NIH grants DP1AI144248 (Pioneer award) and R01AI173086 to G.D.V., DP2AI171161 to Y.P., R01AR050452 and R01AR27883 to E.F., and Starr Consortium grant I10-0044 to G.D.V. Work in the Victora laboratory is additionally supported by the Robertson Foundation, and work in the Pritykin lab by the Ludwig Institute for Cancer Research. S.N.-H. was supported by a Bulgari Women & Science Fellowship, S.W. by the NIH/NHGRI training grant 5T32HG003284, M.C.C.C. by the Pew Latin-American Fellows Program, A.C. by a Damon Runyon Postdoctoral Fellowship, and S.M.P. by a CRI/Carson Family Postdoctoral Fellowship (CRI4498) . E.F. and D.M. are investigators supported by HHMI. G.D.V. is a Burroughs-Wellcome Investigator in the Pathogenesis of Infectious Disease and Pew-Stewart Scholar.
Author contributions
S. N.-H. carried out all experim1ental work with assistance from M.C.C.C., A.C., J.T.J., and S.M.P. S.W. and Y.P. designed and carried out all computational analyses. K.F. generated the structural model of the SrtA–G5-Thy1.1 interaction. G.P. conceived the uLIPSTIC system along with G.D.V. Y.P., A.M.B., E.F., D.M. and G.D.V. supervised the work. S.N.-H., S.W., Y.P., and G.D.V. wrote the manuscript, with input from all authors.
Competing Interests
G.D.V. is an advisor for Vaccine Company, Inc. E.F. recently served on the SABs of L’Oreal and Arsenal Biosciences and owns stock futures in the latter company.
Data availability
Final scRNA-seq datasets will be deposited to GEO upon publication. Python code and Jupyter notebooks used for data analysis are available at https://github.com/pritykinlab/ulipstic-analysis.
Methods
Plasmids
All constructs were cloned into the pMP71 vector52, which was modified to express a fluorescent reporter (eGFP or tdTomato) followed by the porcine teschovirus-1 self-cleavable 2A peptide53 and the protein of interest. The SrtA sequence, including an N-terminal Flag-tag, was attached by a single Gly-Gly-Gly-Gly-Ser linker54 to the human PDGFRB transmembrane domain to form mSrtA. The pentaglycine (G5) acceptor sequence was fused at the N terminus of the mouse Thy1.1 protein, downstream of the signal peptide. Sequences of all constructs are included in Supplementary Spreadsheet 5.
Mice
CD45.2 (C57BL6/J), CD45.1 (B6.SJL Ptprca), CD4-Cre55, CD4-CreERT255, and Foxp3eGFP-CreERT235 mice were purchased from the Jackson Laboratories (strain numbers 000664, 002014, 022071, 022356, and 016961, respectively). Clec9aCre mice56 were a kind gift from C. Reis e Sousa (Francis Crick Institute, UK), S1pr2-CreERT2 BAC-transgenic mice57 were generated and kindly provided by T. Kurosaki and T. Okada (Osaka University and RIKEN-Yokohama), and AicdaCreERT2 mice38 were a kind gift from Claude-Agnès Reynaud and Jean-Claude Weill (Université Paris-Descartes). OT-II TCR transgenic (Y chromosome)58 mice were bred and maintained in our laboratory. The Rosa26uLIPSTIC mouse strain was generated by the Rockefeller University Gene Targeting and Transgenics facilities, as described below. All genetically modified strains are bred and maintained under specific pathogen-free conditions at the Rockefeller University’s Comparative Biosciences Center in accordance with institutional guidelines and ethical regulations. All protocols were approved by the Rockefeller University Institutional Animal Care and Use Committee. Male and female 5–12-week-old mice were used in all experiments.
Generation of the Rosa26uLIPSTIC allele
Rosa26uLIPSTIC mice were generated by gene targeting in C57BL/6 embryonic stem cells (ESCs). The Rosa26uLIPSTIC targeting vector is a modification of the Ai9 Rosa26 conditional expression vector32 (Addgene plasmid #22799). G5-Thy1.1 cDNA preceded by a mouse CD40 signal peptide was inserted into a NruI enzyme site in Ai9 immediately downstream of the first loxP site, whereas FLAG-mSrtA cDNA was introduced in place of the tdTomato gene using FseI enzyme sites. Expression of the cassette in ESCs was screened by standard Southern blotting analysis after EcoRI digestion and using a 32P probe targeting a sequence near the promoter region, shortly upstream of the left homology arm. Positive ESCs (7.3 kb band) were karyotyped, injected into blastocysts and chimeric founders were backcrossed to the C57BL6 background for at least six generations. The full sequence of the uLIPSTIC targeting vector and the Southern blot probe is reported in Supplementary Spreadsheet 5.
Isolation of splenic dendritic cells (DCs), CD4+ T cells, and B cells
To isolate DCs, spleens were collected, incubated for 30 min at 37 °C in HBSS (Gibco) supplemented with CaCl2, MgCl2, and collagenase D at 400 U ml-1 (Roche). After digestion, tissue was forced 5 times through a 21-gauge needle and filtered through a 70 µm strainer into a 15 ml falcon tube with PBS supplemented with 0.5% BSA and 2 mM EDTA (PBE). Red-blood cells were lysed with ACK buffer (Gibco), and the resulting cell suspensions were filtered through a 70-μm mesh into PBE. DCs were obtained by magnetic cell separation (MACS) using anti-CD11c beads (Miltenyi Biotec), as per the manufacturer’s instructions. To isolate CD4+ T cells, CD8+ T cells, and B cells, spleens were processed as above, without collagenase digestion. CD4+ T cells and CD8+ T cells were isolated by negative selection using a cocktail of biotinylated antibodies targeting Ter119, CD11c, CD11b, CD25, B220, NK1.1, and either CD8 (for CD4+ isolation) or CD4 (for CD8+ isolation), followed by anti-biotin beads (Miltenyi Biotec), as per the manufacturer’s instructions. B cells were obtained by negative selection using anti-CD43 beads (Miltenyi Biotec), as per the manufacturer’s instructions.
Adoptive cell transfers
For DC transfer experiments, splenic DCs were isolated as described above from mice subcutaneously injected with 1 x 106 B16 melanoma cells that constitutively secrete FMS-like tyrosine kinase 3 ligand (Flt3L)59 10 days prior to harvest. Cells were resuspended at 107 cells/ml and incubated with 10 μM OVA323–339, LCMV-GP61–80, OVA257-264, or LCMV276-286 peptides (Anaspec) in RPMI + 10% FBS, for 30 min at 37 °C. For cell labelling, CFSE or CTV (ThermoFisher Scientific) was added to a final concentration of 2 μM during the last 5 or 20 minutes of incubation, respectively. Cells were washed three times in RPMI + 10% FBS and resuspended at 2 × 107 cells/ml in PBS supplemented with 0.4 μg ml−1 LPS (Sigma-Aldrich). DCs were injected (5 × 105 cells in 25 μl) subcutaneously into the hind footpads. For CD4+ T cell and CD8+ T cell transfer experiments, T cells isolated as described above were injected intravenously in 100 μl PBS per mouse.
Immunizations
Mice were immunized by subcutaneous injection into the hind footpad of 10 μg OVA or 10 μg NP-OVA (Biosearch Technologies) adsorbed in alum (Imject Alum, ThermoFisher Scientific) at 2:1 antigen:alum (v:v) ratio in 25 μl volume.
Antibody treatments
For CD40L and MHC Class II blocking experiments in vivo, mice were injected intravenously with 200 μg of CD40L-blocking antibody (clone MR-1, BioXCell) or subcutaneously with 150 μg of MHC-II (I-A/I-E) blocking antibody (clone M5/114, BioXCell), four hours prior to the first injection of substrate. For JAML blocking experiments, animals were injected intraperitoneally with 100 µg of anti-JAML antibody (clone 4E10, BioLegend) or isotype control (clone HTK888, BioLegend) at the indicated times.
Tamoxifen treatment
For induction of SrtA expression in regulatory T cells and conventional T cells, Foxp3eGFP-CreERT2/Y.Rosa26uLIPSTIC/WT mice and CD4-CreERT2.Rosa26uLIPSTIC/WT mice, respectively, were given two intragastric doses of 10 mg of tamoxifen (Sigma-Aldrich) dissolved in corn oil (Sigma-Aldrich) at 4 and 2 days prior to the end point (day 0). For SrtA expression in germinal center B cells, S1pr2-CreERT2.Rosa26uLIPSTIC/WT mice and AicdaCreERT2/+.Rosa26uLIPSTIC/WT mice, two doses of 10 mg of tamoxifen were administered at 3 and 2 days prior to the end point. SrtA expression in gut epithelial cells was induced by daily intraperitoneal injections of tamoxifen (2 mg per injection) for 5 consecutive days, starting 14 days before the end point.
In vivo substrate administration
Biotin-aminohexanoic acid-LPETGS, C-terminus amide at 95% purity (biotin-LPTEG) was purchased from LifeTein (custom synthesis) and stock solutions were prepared in PBS at 20 mM. For in vivo LIPSTIC and uLIPSTIC labeling experiments in popliteal lymph nodes (pLNs), biotin–LPETG was injected subcutaneously into the hind footpad (20 μl of 2.5 mM solution in PBS) six times 20 min apart, and pLNs were collected 40 min after the last injection, as described60. Mice were briefly anaesthetized with isoflurane at each injection. For in vivo labeling of gut IELs, biotin-LPETG substrate was injected intraperitoneally (i.p.) (100 μl of 20 mM solution in PBS) six times 20 min apart. Small intestines were collected 40 min after the last injection.
Isolation of lymphocytes from pLNs
pLNs were collected into microfuge tubes with 500 μl HBSS (Gibco) supplemented with CaCl2, MgCl2, and collagenase D at 400 U ml-1 (Roche). pLNs were cut into small pieces and incubated for 30 min at 37 °C . After digestion, tissue was forced 5 times through a 21-gauge needle and filtered through a 70 µm strainer into a 15 ml falcon tube with PBE.
Isolation of intraepithelial leukocytes
Intraepithelial leukocytes were isolated as previously described 61. Briefly, small intestines were harvested and washed in PBS. Peyer’s patches were surgically removed prior to incubation with 1 mM dithiothreitol followed by 30 mM EDTA. Intraepithelial cells were recovered from the supernatant of dithiothreitol and EDTA washes and mononuclear cells were isolated by 40% and 80% gradient Percoll centrifugation.
Flow Cytometry and cell sorting
Single-cell suspensions were washed with PBE, incubated with 1 μg ml−1 anti-CD16/32 (2.4G2, BioXCell) for 5 min at room temperature and then stained for cell surface markers at 4 °C for 20 min in PBS using the reagents listed in Supplementary Table 1. Cells were washed with PBE and stained with Zombie fixable viability dye (BioLegend) or fixable Aqua dead cell stain kit (Invitrogen) at room temperature for 15 min, then washed with PBE and filtered through a 40 μm strainer for acquisition. For in vivo JAML staining of IELs, mice were injected i.p. with 100 μg of anti-JAML AF646 antibody 12 or 6 h prior to the end point. For single-cell transcriptomic analysis, stained cells were further incubated with DNA-barcoded anti-biotin and sample hashtag (anti-MHC-I) antibodies (BioLegend) for 20 minutes in PBE, washed three times with PBE, and bulk-sorted. For substrate detection in vivo, an anti-biotin–PE antibody (Miltenyi Biotec) was exclusively used, as described previously60. Samples were acquired on FACS Symphony or Fortessa analyzers or sorted on FACSAriaII or FACSAriaIII cell sorters (BD Biosciences). Data were analyzed using FlowJo v.10.6.2 software.
uLIPSTIC labeling in vitro
HEK293T cells (ATCC) were transfected by calcium phosphate transfection with the indicated expression vectors at high (1 μg μl-1) and low (0.1 μg μl-1) concentrations of Thy1.1-G5 and mSrtA constructs. Forty hours after transfection, cells were detached TrypLE Express cell dissociation solution (ThermoFisher Scientific), washed and resuspended at 106 cell per ml in PBS. Donor cell populations transfected with CD40L and/or mSrtA constructs and acceptor cell populations transfected with CD40 and/or Thy1.1-G5 were mixed at a 1:1 ratio (105 cells of each population) in a 1.5-ml conical tube, to which biotin–LPETG was added to a final concentration of 100 μM. Cells were incubated at room temperature for 30 min and washed three times with PBE to remove excess biotin–LPETG before FACS staining.
uLIPSTIC labeling ex vivo
B cells from Rosa26uLIPSTIC/WT mice and CD4+ T cells from OT-II CD4-Cre.Rosa26uLIPSTIC/WT mice were isolated from mouse spleens as described above. Isolated T cells were activated with CD3/CD28 dynabeads (ThermoFisher Scientific) for 24 h and then co-cultured with isolated B cells (2 × 105 cells per well, 1:1 ratio) in the presence or absence of OVA323–339 peptide in RPMI, 10% FBS supplemented with 0.1% 2 mercaptoethanol (Gibco) in U-bottom 96-well plates for 20 h. Blocking antibodies were added at the beginning of the co-culture at a final concentration of 150 μg ml−1. To label interactions ex-vivo, biotin-LPETG substrate was added 30 min before harvest at a final concentration of 100 μM.
Library preparation for single cell-RNA sequencing of IELs
In addition to fluorescent antibodies, cells were co-stained prior to sorting with hashtag oligonucleotide (HTO)-labeled antibodies to CD45 and MHC-I for sample separation (two hashtags per sample) and HTO-anti-biotin for detection of the uLIPSTIC signal. Sorted cells were collected into a microfuge tube with 300 μl PBS supplemented with 0.4% BSA. After the sort, tubes were topped with PBS 0.4% BSA, centrifuged and the buffer was carefully reduced by removing the volume with a pipette to a final volume of 40 μl. Cells were counted for viability and immediately submitted to library preparation. The scRNA-seq library was prepared using the 10X Single Cell Chromium system, according to the manufacturer’s instructions, at the Genomics Core of Rockefeller University and was sequenced on an Illumina NovaSeq SP flowcell to a minimum sequencing depth of 30,000 reads per cell using read lengths of 26 bp read 1, 8 bp i7 index, 98 bp read 2.
Single-cell RNA sequencing analysis
Gene expression unique molecular identifier (UMI) counts, along with sample and biotin (uLIPSTIC) HTO counts, were generated with CellRanger v6.0.1 “count” using “Feature Barcode” counts and otherwise default parameters, with mm10 reference. TCR data were preprocessed with CellRanger “vdj” with default parameters. Applying default cellranger filtering, this resulted in a filtered gene expression UMI count matrix including 4,607 cells and 32,285 genes.
The scanpy package v1.9.1 was used for all analysis of the gene expression data62. Cell barcodes with unresolved sample HTOs, a low or extremely high number of expressed genes, a large fraction of expressed mitochondrial genes, or likely doublets were removed. Genes expressed in a low number of cells were removed. This resulted in a filtered gene expression matrix of 3,677 cells and 14,332 genes with a matching biotin HTO count in each cell representing uLIPSTIC signal.
Gene counts were normalized using Pearson residual normalization with theta = 1. Principal component analysis (PCA) was run with default parameters, and then k nearest neighbor (kNN) graph was constructed using 40 PCs, k = 30 and otherwise default parameters. Then Leiden clustering was performed with a resolution of 1, resulting in 26 clusters.
uLIPSTIC normalized values were obtained for each cell by dividing the uLIPSTIC HTO counts by the number of sample-encoding HTO read counts in a cell. The 5th percentile of these normalized values was added as a pseudocount, and then log10 applied. These values were then shifted by the minimum log-scaled value, so the scale starts at 0.
Known marker genes as well as TCR data were used to annotate the Leiden clusters. The scirpy package v0.10.1 was used for the TCR data preprocessing and analysis63. Cluster 10 was split into two subclusters that contained cycling T and B cells. Annotations were confirmed by scoring PanglaoDB immune cell marker gene sets64 using the score_genes() function in scanpy and by exploring significantly differentially expressed genes in each cluster as compared with all cells outside the cluster, obtained using a custom script. For differential expression analysis, log2 fold change (log2FC) of expression was calculated as the ratio of pseudobulk raw UMI counts summed over cells within and outside the cluster (then normalized by total amount of UMI counts inside and outside the cluster), p-values were calculated using Mann-Whitney U test applied to Pearson residual normalized expression values in single cells within and outside the cluster, and Bonferroni correction for multiple hypothesis testing applied to all genes with abs(log2FC) > 0.5.
The analysis then focused on the CD4 T cell subset. A new kNN graph was generated for this subset, again using k = 30 neighbors and 40 PCs, and Leiden clustering performed with resolution = 1.3. Clusters were annotated using known marker genes and TCR clonality information. Trajectory analysis and subsequent cell pseudotime calculation were performed using Wishbone v0.5.265 using default parameters as available in scanpy and using as the root the cell in the Naive/Tconv cluster with the highest value of Sell expression.
To identify candidate genes involved in cell-cell interactions, for every gene the Spearman correlation was calculated between the Pearson residual normalized value of expression of that gene and the uLIPSTIC signal across all cells in the CD4 T cell subset. Bonferroni correction was used for multiple hypothesis testing on all genes. This calculation was separately performed when removing Tfh-like and naïve/memory cells, or when restricting to cells from each individual mouse, with consistent results.
For violin plot of scRNA-seq expression of Jaml (Fig. 4I), Pearson residual normalized values were shifted so that the minimum value is zero, bottom 5th percentile of all values (across cell groups) omitted, and then plotted on log scale. The T cell subpopulations for the plot were defined as follows. The subpopulations of CD4 T cells, “Naïve/Tconv”, “Pre-IEL”, “IEL” (Fig. 4D-H, S7), were used as CD4+ Tconv, CD4+ Pre-IEL and CD4+ CD8αα+ IEL, respectively. The “Natural IEL” cells (Fig. 4B, S5, S6) were separated into three groups: CD8αα+ IEL if TCR ab chain was detected (301 cells), otherwise ψ8 IEL if normalized expression of Trdc was above 0 (517 cells), and other (163 cells) which were not included in the plot.
Modelling the SrtA-Thy1.1 complex on cells surfaces
First, structures of G5-Thy1.1 and FLAG-SrtA-PDGFRb were generated using Alphafold266. Next, in the FLAG-SrtA-PDGFRB model, the domain constructing peptide binding domain was substituted with the substrate bound Sortase A structure (PDB: 1T2W). Additionally, the flexible linker connecting SrtA domain to PDGFRB transmembrane helix was rebuilt to an extended conformation using COOT67 to better estimate the maximum distance the protein is able to extend to. The Thy1.1 was aligned to SrtA using the substrate of 1T2W and 5G acceptor motif of G5-Thy1.1. Any resulting interprotein clashes were corrected using GalaxyWEB server68. To build the GPI anchor and the lipid bilayers we used CHARMM-GUI69. The anchor glycolipid was generated based on the human prion protein (PrP) GPI70. Next theFLAG-SrtA-PDGFRB:G5-Thy1.1 complex was modelled in the POPC/cholesterol lipid bilayer using CHARMM-GUI. The GPI anchor was placed in the second bilayer using ChimeraX71. Finally, both bilayers, with the protein complex and the GPI anchor were aligned in ChimeraX. The distance between two bilayers was measured in PyMOL72.
Statistical analysis
Statistical tests were performed in GraphPad Prism 9.0 software. Comparisons between two treatment conditions were analyzed using unpaired, two-tailed Student’s t-test and multivariate data were analyzed by one-way ANOVA with Tukey’s post-hoc tests to further examine pairwise differences.
Supplementary Figures
Acknowledgements
We would like to thank the Rockefeller University Transgenics and Gene Targeting facilities for generating the uLIPSTIC mouse strain, the Comparative Biosciences Center for mouse housing, Kristie Gordon for cell sorting, and all Rockefeller University staff for their continuous support. We thank Caetano Reis e Sousa (Francis Crick Institute, UK), Tomohiro Kurosaki (U. Osaka, Japan), Takaharu Okada (RIKEN-Yokohama, Japan), and Claude-Agnès Reynaud and Jean-Claude Weill (Université Paris-Descartes, France) for mice; and Telmo Catarino (Instituto de Investigação e Inovação em Saúde, Portugal) and Henrique Veiga-Fernandes (Champalimaud Institute, Portugal) for their preliminary contributions to the development of the uLIPSTIC system.