Barbara Fazekas de St Groth

Multi-parameter cytometry technologies enable high-dimensional analysis of immune cell populations at single-cell resolution. Deep learning has been widely applied to these datasets, but existing methods often struggle with transferability across datasets due to technical variability, batch effects and identification of biologically relevant cell populations, limiting their utility in clinical research. We present dioscRi, a transferable deep learning framework that integrates a maximum mean discrepancy variational autoencoder for normalization and de-noising, enhancing cross-dataset compatibility. Changes in cell type proportions and marker expression are identified by structuring these features within biologically or empirically derived cell type hierarchies. These hierarchies are incorporated directly into an overlapping group LASSO model, improving the prediction of clinical outcomes. When applied to a coronary artery disease study, dioscRi recapitulated several known immune associations. Benchmarking across multiple datasets demonstrated dioscRi’s ability to transfer across cohorts with compatible marker panels and outperform existing methods on three of four datasets, establishing it as an interpretable tool for cytometry data analysis.

Multi-parameter cytometry technologies enable high-dimensional analysis of immune cell populations at single-cell resolution. Deep learning has emerged as a transformative tool for analyzing these datasets, but existing methods often struggle with transferability across datasets due to technical variability, batch effects and identification of biologically relevant cell populations, limiting their utility in clinical research. We present dioscRi, a transferable deep learning framework that integrates a maximum mean discrepancy variational autoencoder for normalization and de-noising, enhancing cross-dataset compatibility. Changes in cell type proportions and marker expression are identified by structuring these features within biologically or empirically derived cell type hierarchies. These hierarchies are incorporated directly into an overlapping group LASSO model, improving the prediction of clinical outcomes. When applied to a coronary artery disease study, dioscRi recapitulated several known immune associations. Benchmarking across multiple datasets demonstrated dioscRi’s ability to generalize and outperform existing methods, establishing it as a versatile and interpretable tool for cytometry data analysis.

Antibody titration is an important step in every cytometric workflow, with the goal being to determine antibody concentrations that ensure highly reproducible results. When aiming to compare antigen expression between samples using mean or median fluorescence intensity (MFI), reagents should be used at a saturating concentration so that unavoidable variations in staining conditions do not affect the fluorescence signal. The recommended concentrations of commercially available fluorophore‐labeled monoclonal antibodies (mAbs) may not achieve plateau staining, and their saturating concentration may be too high to be experimentally useful. To address these common concerns, we present a novel method to achieve saturation of fluorophore‐conjugated mAbs, by ‘spiking‐in’ unlabelled antibody of the same clone. Here, we demonstrate the application of this workflow to human anti‐CD3 (clone OKT3, mouse IgG2a) and anti‐TCRαβ (clone IP26, mouse IgG1), two mAbs that do not achieve saturation at 2‐fold above their commercially recommended concentrations. First, the saturating concentration of unlabelled (purified) OKT3 and IP26 was determined by detection with a fluorophore‐labeled anti‐mouse IgG (H + L) secondary antibody. Titration curves of unlabelled and labeled mAbs were compared for each clone to determine whether labeling had resulted in any loss in binding activity. Unlabelled antibody was then ‘spiked’ into the labeled antibody at varying ratios, and those that achieved saturation while maintaining an adequate fluorescence signal were identified. We demonstrate that antibody saturation can be achieved with an optimized mixture of labeled and unlabelled antibody, while maintaining a clear signal from the fluorophore. While this workflow has only been applied to OKT3 and IP26, it has potential applicability for any antibody clone for which both labeled and unlabelled preparations are available. This method has significance for robust comparison of biomarker expression when fluorophore labeled reagents do not reach saturation under standard staining conditions.

Objective: Chronic obstructive pulmonary disease (COPD) is a major cause of global illness and death, most commonly caused by cigarette smoke. The mechanisms of pathogenesis remain poorly understood, limiting the development of effective therapies. The gastrointestinal microbiome has been implicated in chronic lung diseases via the gut-lung axis, but its role is unclear. Design: Using an in vivo mouse model of cigarette smoke (CS)-induced COPD and faecal microbial transfer (FMT), we characterised the faecal microbiota using metagenomics, proteomics and metabolomics. Findings were correlated with airway and systemic inflammation, lung and gut histopathology and lung function. Complex carbohydrates were assessed in mice using a high resistant starch diet, and in 16 patients with COPD using a randomised, double-blind, placebo-controlled pilot study of inulin supplementation. Results: FMT alleviated hallmark features of COPD (inflammation, alveolar destruction, impaired lung function), gastrointestinal pathology and systemic immune changes. Protective effects were additive to smoking cessation, and transfer of CS-associated microbiota after antibiotic-induced microbiome depletion was sufficient to increase lung inflammation while suppressing colonic immunity in the absence of CS exposure. Disease features correlated with the relative abundance of Muribaculaceae, Desulfovibrionaceae and Lachnospiraceae family members. Proteomics and metabolomics identified downregulation of glucose and starch metabolism in CS-associated microbiota, and supplementation of mice or human patients with complex carbohydrates improved disease outcomes. Conclusion: The gut microbiome contributes to COPD pathogenesis and can be targeted therapeutically.

There has been growing interest in the role of B cells in antitumour immunity and potential use in adoptive cellular therapies. To date, the success of such therapies is limited. The intrinsic capacity of B cells to specifically activate tumour-specific CD4+ T cells in vivo via TCR-dependent interactions remains poorly defined. We have developed an in vivo tumour model that utilizes MHCII I-E restriction which limits antigen presentation to tumour-specific CD4 T cells to either tumour-specific B cells or host myeloid antigen presenting cells (APCs) in lymphopenic RAG-/-mice. We have previously shown that these naive tumour-specific CD4+ T cells can successfully eradicate established tumours in this model when activated by host APCs. When naïve tumour-specific B cells are the only source of I-E+ APC, very limited proliferation of naïve CD4+ T cells is observed, whereas host I-E+ APCs are potent T cell activators. B cells pre-activated with an anti-CD40 agonistic antibody in vivo support increased T cell proliferation, although far less than host APCs. CD4+ T cells that have already differentiated to an effector/central memory phenotype proliferate more readily in response to naïve B cells, although still 100-fold less than in response to host APCs. This study demonstrates that even in a significantly lymphopenic environment, myeloid APCs are the dominant primary activators of tumour-specific T cells, in contrast to the very limited capacity of tumour-specific B cells. This suggests that future anti-tumour therapies that incorporate activated B cells should also include mechanisms that activate host APCs.

Human cytomegalovirus reactivation is a major opportunistic infection after allogeneic haematopoietic stem cell transplantation and has a complex relationship with post-transplant immune reconstitution. Here, we use mass cytometry to define patterns of innate and adaptive immune cell reconstitution at key phases of human cytomegalovirus reactivation in the first 100 days post haematopoietic stem cell transplantation. Human cytomegalovirus reactivation is associated with the development of activated, memory T-cell profiles, with faster effector-memory CD4⁺ T-cell recovery in patients with low-level versus high-level human cytomegalovirus DNAemia. Mucosal-associated invariant T cell levels at the initial detection of human cytomegalovirus DNAemia are significantly lower in patients who subsequently develop high-level versus low-level human cytomegalovirus reactivation. Our data describe distinct immune signatures that emerged with human cytomegalovirus reactivation after haematopoietic stem cell transplantation, and highlight Mucosal-associated invariant T cell levels at the first detection of reactivation as a marker that may be useful to anticipate the magnitude of human cytomegalovirus DNAemia.

Summary: Human cytomegalovirus (HCMV) reactivation is a major opportunistic infection after allogeneic haematopoietic stem cell transplantation and has a complex relationship with post-transplant immune reconstitution. Here, we used mass cytometry to comprehensively define global patterns of innate and adaptive immune cell reconstitution at key phases of HCMV reactivation (before detection, initial detection, peak and near resolution) in the first 100 days post-transplant. In addition to identifying patterns of immune reconstitution in those with or without HCMV reactivation, we found mucosal-associated invariant T (MAIT) cell levels at the initial detection of HCMV DNAemia distinguished patients who subsequently developed low-level versus high-level HCMV reactivation. In addition, early recovery of effector-memory CD4 ⁺ T cells distinguished low-level and high-level reactivation. Our data describe distinct immune signatures that emerged with HCMV reactivation post-HSCT, and highlight MAIT cell levels at the initial detection of reactivation as a potential prognostic marker to guide clinical decisions regarding pre-emptive therapy.

We sought to determine the effect of time and temperature of blood sample storage before preparation of human PBMCs by Ficoll-hypaque density gradient centrifugation. Blood samples from healthy donors were stored at room temperature (RT) or refrigerated at 4^oC before preparation of PBMCs. Cell yield and viability, and proportions of major cell populations within PBMCs, as determined by fluorescence flow cytometry, were assessed for both fresh and cryopreserved samples. Highly multiparametric mass cytometry was performed on cryopreserved PBMCs. We found that refrigeration had marked negative effects on subsequent PBMC yield. Storage at RT led to co-purification of low density neutrophils with PBMCs, but had no detectable effects on the proportions of multiple cell subsets including, but not limited to, monocytes, NK cells, B cells, Treg cells, and naïve, central memory and effector memory CD4⁺ and CD8⁺ T cells and CD45RA-positive terminal effector CD8⁺ T cells. Expression of a number of cell surface receptors, including CXCR5, CCR6, CXCR3 and TIGIT, but not CD247 was reduced after RT storage before PBMC preparation, and this effect correlated with the degree of low density neutrophil contamination. As such, when PBMC preparation cannot be undertaken immediately after blood draw, storage at RT is far superior to refrigeration. RT storage leads to neutrophil activation, but does not compromise measurement of PBMC subset distribution. However caution must be applied to interpretation of cytometric measurements of surface molecules such as chemokine receptors.

HIV-1 infection rapidly leads to a loss of the proliferative response of memory CD4+ T lymphocytes, when cultured with recall antigens. We report here that CD73 expression defines a subset of resting memory CD4+ T cells in peripheral blood, which highly express the α-chain of the IL-7 receptor (CD127), but not CD38 or Ki-67, yet are highly proliferative in response to mitogen and recall antigens, and to IL-7, in vitro. These cells also preferentially express CCR5 and produce IL-2. We reasoned that CD73+ memory CD4+ T cells decrease very early in HIV-1 infection. Indeed, CD73+ memory CD4+ T cells comprised a median of 7.5% (interquartile range: 4.5–10.4%) of CD4+ T cells in peripheral blood from healthy adults, but were decreased in primary HIV-1 infection to a median of 3.7% (IQR: 2.6–6.4%; p = 0.002); and in chronic HIV-1 infection to 1.9% (IQR: 1.1–3%; p < 0.0001), and were not restored by antiretroviral therapy. Moreover, we found that a significant proportion of CD73+ memory CD4+ T cells were skewed to a gut-homing phenotype, expressing integrins α4 and β7, CXCR3, CCR6, CD161 and CD26. Accordingly, 20% of CD4+ T cells present in gut biopsies were CD73+. In HIV+ subjects, purified CD73+ resting memory CD4+ T cells in PBMC were infected with HIV-1 DNA, determined by real-time PCR, to the same level as for purified CD73-negative CD4+ T cells, both in untreated and treated subjects. Therefore, the proliferative CD73+ subset of memory CD4+ T cells is disproportionately reduced in HIV-1 infection, but, unexpectedly, their IL-7 dependent long-term resting phenotype suggests that residual infected cells in this subset may contribute significantly to the very long-lived HIV proviral DNA reservoir in treated subjects.