A 2019 Cell study by Grandclaudon and colleagues built the first quantitative, multivariate, protein-level input-output model of human dendritic cell (DC) to T helper (Th) cell communication: 36 DC-derived communication molecules measured across 428 observations of 82 stimulatory conditions, regressed against 18 Th output variables with a variable-selection pipeline (MultiVarSel), yielding 56 known and 290 candidate novel mechanisms and revealing a context-dependent function of IL-12 in inducing IL-17F. The study's capstone open question is whether this framework and its learned rules generalize to other sender cell types — whether cells 'speak the same language', i.e. express similar patterns of communication molecules and assign the same function to the same molecule regardless of the sender cell type. No comparative quantitative multivariate model for a second sender cell type has been published: partial progress exists (framework transfer to single-stimulus context-dependency of OX40L on T cells, a mechanistic model of IL-17A/IL-17F differential expression, DC-state clinical association studies), while the field has moved to expression-based single-cell inference tools that do not test protein-level function conservation across senders.

Background

Cells of the immune system exchange information through secreted and surface molecular signals, most prominently cytokines — small proteins that act on other cells through specific receptors. Dendritic cells (DCs) are the professional antigen-presenting cells of the immune system: they sample tissues, respond to microbes and inflammatory cues by entering distinct activation states, and then instruct naive CD4+ T cells, which differentiate into various T helper (Th) effector subsets (Th1, Th2, Th17, and others) defined by the cytokines they produce. Because a DC simultaneously expresses dozens of communication molecules whose joint effect on the receiving T cell is combinatorial and context-dependent — a situation the 2019 study likened to a language whose individual words (molecules) change meaning depending on which other words accompany them — the input-output mapping from DC state to Th cytokine response cannot be resolved one molecule at a time.

Earlier modeling efforts fell into two families: theoretical regulatory-network models of Th differentiation that enumerate many possible T-cell inputs but assume prior knowledge and do not represent naturally occurring combinations and concentrations of signals, and data-driven transcriptomic studies of responses to predefined stimulus cocktails that do not integrate the many simultaneous communication signals. The 2019 Cell study by Grandclaudon, Perrot-Dockès, and colleagues bridged this gap: they measured 36 DC-derived communication signals (cytokines, surface ligands) at the protein level across 428 observations spanning 82 DC stimulatory conditions (cytokines, virus, bacteria, fungi at various doses and combinations), together with 18 output variables in the receiving Th cells (17 cytokines plus a proliferation readout). They related inputs to outputs with MultiVarSel, a statistical pipeline for variable selection in the multivariate linear model that combines the Lasso (a regression method that shrinks most coefficients to zero, selecting a sparse set of explanatory variables) with an estimate of the residual dependence structure between outputs and stability selection (repeated subsampling to keep only stably selected variables; the method is available as the R package MultiVarSel). To capture context-dependency, they introduced composite input variables that take the value of a molecule only when a co-expressed partner signal is present. The resulting model reproduced most known DC–Th mechanisms, predicted 290 candidate novel ones, and revealed that interleukin-12 (IL-12; the interleukins, IL, are the numbered members of the cytokine family), previously considered a Th1-inducing cytokine, induces IL-17F production when acting in an interleukin-1 (IL-1) context — validated experimentally. The model also exposed limitations the authors acknowledged: a small number of poorly predicted outputs (e.g. five observations with very high levels of the cytokine tumor necrosis factor-β, TNF-β), suggesting a still-unmeasured DC-derived input; and overfitting when all 455 pairwise context-dependent composite variables were included, since the input count exceeded the number of observations.

The study closed by abstracting its approach into a five-step framework applicable to any sender-receiver pair: (1) systematically perturb the sender cell to generate communication-state diversity; (2) quantitatively measure the sender's communication molecules at the protein level; (3) quantitatively measure the receiver's responses; (4) model the input-output relationship with multivariate variable selection; (5) validate computationally and experimentally. But the authors stated their belief that cell-type specificities in both the expression and the function of communication molecules would prevent direct generalization of the DC–Th model to other cell types, and posed the comparative question that defines this problem: comparing quantitative models of cell-cell communication built for different cell types will ultimately tell whether cells speak the same language — whether they express similar patterns of communication molecules, and whether the same communication molecule has the same meaning and function when expressed by two different cell types.

Problem Statement

Determine whether the quantitative multivariate rules of intercellular communication established for human dendritic-cell-to-T-helper-cell signaling generalize to other sender cell types — whether cells 'speak the same language'. Concretely, applying the five-step framework of the source study (systematic perturbation of sender states; protein-level quantitative measurement of the sender's communication molecules; quantitative measurement of receiver outputs; multivariate variable-selection modeling of the input-output relationship; computational and experimental validation), build a comparable quantitative multivariate input-output model for at least one sender cell type other than dendritic cells acting on a defined receiver cell (e.g., Th cells or another defined responder), and compare its learned rules with those of the published dendritic-cell-to-T-helper model for the communication molecules and receiver responses that both models cover. The comparison must answer: (i) whether different sender cell types express similar patterns of communication molecules across the sender states sampled by the framework's perturbation panel, which are taken to span the naturally occurring range of communication states; and (ii) whether the same communication molecule has the same meaning/function when expressed by two different sender cell types — identifying which input-output rules are conserved across senders and which are sender-specific. Consistent with the source study's framing that comparing different quantitative models of cell-cell communication 'will ultimately tell us whether cells speak the same language', a first cross-sender comparison constitutes the initial empirical test bearing on this general question rather than a resolution of it for all sender cell types. The question concerns the input-output communication rules and their experimental validation, not the intracellular signaling mechanisms downstream in the receiver, and not clinical deployment of the original model. No pre-assembled second-sender dataset exists, so the comparison rests on the chosen sender's protein-level communication molecules and matched receiver outputs drawn from accessible sources, or, where these cannot be drawn, on the specified perturbation and measurement design that would generate them.

References

  1. Grandclaudon M, Perrot-Dockès M, Trichot C, Karpf L, Abouzid O, Chauvin C, Sirven P, Abou-Jaoudé W, Berger F, Hupé P, Thieffry D, Sansonnet L, Chiquet J, Lévy-Leduc C, Soumelis V. A Quantitative Multivariate Model of Human Dendritic Cell-T Helper Cell Communication. Cell 179(2):432-447.e21, 2019. doi:10.1016/j.cell.2019.09.012. https://doi.org/10.1016/j.cell.2019.09.012
  2. Karpf L, Trichot C, Faucheux L, Legbre I, Grandclaudon M, Lahoute C, Mattoo H, Pasquier B, Soumelis V. A multivariate modeling framework to quantify immune checkpoint context-dependent stimulation on T cells. Cell Discovery 8:1, 2022. doi:10.1038/s41421-021-00352-4. https://doi.org/10.1038/s41421-021-00352-4
  3. Corral-Jara KF, Chauvin C, Abou-Jaoudé W, Grandclaudon M, Naldi A, Soumelis V, Thieffry D. Interplay between SMAD2 and STAT5A is a critical determinant of IL-17A/IL-17F differential expression. Molecular Biomedicine 2:9, 2021. doi:10.1186/s43556-021-00034-3. https://doi.org/10.1186/s43556-021-00034-3
  4. Hoffmann C, Noël F, Grandclaudon M, Massenet-Regad L, Michea P, Sirven P, Faucheux L, Surun A, Lantz O, Bohec M, Ye J, Guo W, Rochefort J, Klijanienko J, Baulande S, Lecerf C, Kamal M, Le Tourneau C, Guillot-Delost M, Soumelis V. PD-L1 and ICOSL discriminate human Secretory and Helper dendritic cells in cancer, allergy and autoimmunity. Nature Communications 13:1983, 2022. doi:10.1038/s41467-022-29516-w. https://doi.org/10.1038/s41467-022-29516-w
  5. Shilts J, Severin Y, Galaway F, Müller-Sienerth N, Chong ZS, Pritchard S, Teichmann S, Vento-Tormo R, Snijder B, Wright GJ. A physical wiring diagram for the human immune system. Nature 608(7922):397-404, 2022. doi:10.1038/s41586-022-05028-x. https://doi.org/10.1038/s41586-022-05028-x
  6. Shilts J, Wright GJ. Mapping the Human Cell Surface Interactome: A Key to Decode Cell-to-Cell Communication. Annual Review of Biomedical Data Science 7:155-177, 2024. doi:10.1146/annurev-biodatasci-102523-103821. https://doi.org/10.1146/annurev-biodatasci-102523-103821
  7. Armingol E, Officer A, Harismendy O, Lewis NE. Deciphering cell-cell interactions and communication from gene expression. Nature Reviews Genetics 22(2):71-88, 2021. doi:10.1038/s41576-020-00292-x. https://doi.org/10.1038/s41576-020-00292-x
  8. Aghamiri SS, Puniya BL, Amin R, Helikar T. A multiscale mechanistic model of human dendritic cells for in-silico investigation of immune responses and novel therapeutics discovery. Frontiers in Immunology 14:1112985, 2023. doi:10.3389/fimmu.2023.1112985. https://doi.org/10.3389/fimmu.2023.1112985
  9. Perrot-Dockès M, Lévy-Leduc C, Sansonnet L, Chiquet J. Variable selection in multivariate linear models with high-dimensional covariance matrix estimation. Journal of Multivariate Analysis 166:78-97, 2018. doi:10.1016/j.jmva.2018.02.006. https://doi.org/10.1016/j.jmva.2018.02.006
  10. Perrot-Dockès M, Lévy-Leduc C, Chiquet J, Sansonnet L, Brégère M, Étienne MP, Robin S, Genta-Jouve G. A variable selection approach in the multivariate linear model: an application to LC-MS metabolomics data. Statistical Applications in Genetics and Molecular Biology 17(5), 2018. doi:10.1515/sagmb-2017-0077. https://doi.org/10.1515/sagmb-2017-0077
  11. Tatsumi N, Tjitropranoto A, Davila-Pagan A, Kumamoto Y. Functional Specialization and Collaboration of cDC2 Subsets in CD4+ T Cell Priming and Differentiation. Immunological Reviews 336(1):e70069, 2025. doi:10.1111/imr.70069. https://doi.org/10.1111/imr.70069
  12. Wahiduzzaman M, Liu Y, Huang T, Wei W, Li Y. Cell-cell communication analysis for single-cell RNA sequencing and its applications in carcinogenesis and COVID-19. Biosafety and Health 4(4):220-227, 2022. doi:10.1016/j.bsheal.2022.03.001. https://doi.org/10.1016/j.bsheal.2022.03.001
  13. Amblard E, Soumelis V. Context-Dependent Effects Explain Divergent Prognostic Roles of Tregs in Cancer. Cancers 14(12):2991, 2022. doi:10.3390/cancers14122991. https://doi.org/10.3390/cancers14122991