Soluble immunosuppressive checkpointome of mesenchymal stromal cells: Immunoregulatory mechanisms and a conceptual framework for potency assessment

Abstract

Mesenchymal stromal cells (MSCs) are among the most widely investigated platforms in regenerative and immune-directed therapy, yet the molecular basis of their immunosuppressive activity remains incompletely defined. Classical accounts emphasise indoleamine 2,3-dioxygenase, prostaglandin E2, nitric oxide and extracellular vesicles (EVs), while a parallel secretory output has attracted comparatively little attention: The release of soluble immune checkpoint molecules. Soluble programmed death ligand 1 and 2, galectin-9, galectin-1 and the secreted human leukocyte antigen-G5 isoform are each documented in MSC conditioned medium, and each has been linked functionally to suppression of T-cell proliferation, induction of effector-cell apoptosis, or expansion of regulatory T cells. These proteins are generated through proteolytic ectodomain shedding, alternative splicing, or association with small EVs, and their abundance is strongly conditioned by inflammatory licensing. This review consolidates fragmented, single-molecule evidence into a proposed conceptual immunoregulatory framework, the MSC soluble immunosuppressive checkpointome. The framework is explicitly a conceptual grouping and not a physically interacting or experimentally validated synergistic protein complex; coordinated or synergistic action among its members has not been demonstrated. For clarity, the framework is defined operationally to encompass three physically distinct extracellular species, namely shed ectodomains, secreted splice isoforms and extracellular-vesicle-associated checkpoint proteins, and this distinction is applied consistently in the text, tables and figures. It groups molecules with experimentally demonstrated MSC release and immunoregulatory activity while distinguishing them from biologically plausible candidates whose secretion by MSCs has not yet been directly demonstrated. We then consider a translational application that has not been systematically explored, namely the use of secreted checkpoints as pharmacodynamic and potency markers, addressing a long-standing obstacle to standardising MSC product potency for regulatory release. We further discuss how a composite, multi-marker checkpoint signature, integrated with predictive and decision-support modelling, could inform donor selection, batch qualification and patient stratification. Because current evidence is largely associative and derives from heterogeneous platforms, we set out the methodological requirements that must be met before clinical translation.

Key Words: Mesenchymal stromal cells; Soluble immune checkpoints; Programmed death ligand 1; Galectin-9; Human leukocyte antigen-G5; Potency assay; Cell therapy; Immunomodulation

Core Tip: Mesenchymal stromal cells release a defined set of soluble immune checkpoint proteins, notably soluble programmed death ligand 1 and 2, galectin-9, galectin-1 and the secreted human leukocyte antigen-G5 isoform. Treated separately until now, they are here proposed as a conceptual inducible module termed the soluble checkpointome. Because its abundance rises with inflammatory licensing and tracks with suppressive capacity, it offers a licensed, cell-free, multi-analyte panel well suited to the potency assays regulators require, and amenable to composite scoring supported by machine learning.

INTRODUCTION

Mesenchymal stromal cells (MSCs) occupy an unusual position in translational medicine. Defined operationally by plastic adherence, a characteristic surface phenotype and trilineage differentiation capacity[1], they have been administered to many thousands of patients across a broad range of inflammatory and degenerative indications with a consistently reassuring safety record, and yet efficacy has been difficult to reproduce across trials. The tension between an enormous clinical literature and an unsettled mechanistic foundation has been articulated repeatedly, most pointedly in the argument that the field advanced faster than the biology that was meant to justify it[2]. A central reason for this difficulty is that MSCs are not a fixed pharmacological entity. They behave instead as environmental sensors that reconfigure their secretory output according to the inflammatory signals they encounter, so that the same preparation may be suppressive in one milieu and comparatively inert in another[3,4]. Immune checkpoint networks comprise both inhibitory and stimulatory co-signaling pathways. The present review intentionally focuses on the inhibitory/immunosuppressive arm because the proposed translational application concerns the immunosuppressive potency of MSC products. Accordingly, ‘soluble checkpointome’ is used throughout in this restricted context rather than as a comprehensive inventory of all stimulatory and inhibitory checkpoint molecules associated with MSCs.

The mediators through which MSCs restrain immune effector function have been catalogued in considerable detail. Indoleamine 2,3-dioxygenase degrades tryptophan and thereby arrests allogeneic T-cell responses[5]. Prostaglandin E2 reprogrammes dendritic cells and shifts cytokine balance towards an anti-inflammatory profile[6]. Nitric oxide, acting in concert with chemokines that draw lymphocytes into proximity with the stromal cell, suppresses proliferation at short range[7]. Each of these pathways is licensed rather than constitutive, and interferon gamma (IFN-γ) is the canonical licensing signal[8]. What these mediators share is that they are enzymes, small lipid or gaseous messengers, or chemotactic proteins. What they largely omit is a distinct protein class that MSCs also release and that immunologists elsewhere regard as central to the control of lymphocyte activation, namely the ligands and receptors of the immune checkpoint system.

The omission is not for lack of primary data. Human MSCs secrete programmed death ligand 1 (PD-L1) and PD-L2 into conditioned medium, and blockade of these soluble ligands partially restores T-cell interleukin-2 (IL-2) production and activation marker expression under transwell conditions in which cell contact is excluded[9]. MSCs constitutively release galectin-1, and knockdown of galectin-1 measurably impairs their capacity to inhibit allogeneic T cells[10]. Galectin-9, the canonical ligand of T cell immunoglobulin and mucin domain-containing protein 3 (TIM-3), is not constitutive but is strongly induced and released upon exposure to IFN-γ, tumor necrosis factor alpha (TNF-α) or Toll-like receptor ligands[11]. Secretion of the soluble human leukocyte antigen (HLA)-G5 isoform is required both for suppression of T and natural killer (NK) cell function and for the expansion of CD4+CD25^highFOXP3+ regulatory T cells[12]. Each of these observations is well established. What has not been done is to read them together.

We propose the MSC soluble checkpointome as a conceptual framework that organizes a group of soluble immunoregulatory checkpoint molecules released by MSCs. The term ‘module’ is used here functionally rather than structurally: Current evidence does not demonstrate that these molecules form a physical complex or act synergistically as an experimentally validated unit. Instead, the framework integrates independently demonstrated checkpoint pathways that may collectively contribute to the immunoregulatory phenotype of MSCs. Furthermore, Table 1 contrasts it with the classical mediator set.

Table 1 The classical mediator set and the soluble checkpointome compared.

Property	Classical mediators (IDO, PGE2, nitric oxide)	Soluble checkpointome
Molecular class	Enzymes, lipid and gaseous messengers[5-7]	Immunoglobulin superfamily ligands, lectins and MHC class Ib isoforms[9-12]
Receptor engaged	No dedicated inhibitory receptor; act on metabolism or signaling tone	Defined inhibitory receptors: PD-1, TIM-3, ILT2 and ILT4[9,11,12]
Inducibility	Licensed by IFN-γ and TNF-α[5,8]	Mixed: PD-L1 and galectin-9 induced, galectin-1 constitutive[9-11]
Measured in the product	Indirectly, through kynurenine or nitrite	Directly, as protein in cell-free conditioned medium[9,11]
Precedent as a potency marker	IDO proposed and used[39]	Galectin-9 predicts preparation-level potency[16]; PD-L1 proposed alongside IDO1[39]
Suitability for a release assay	Enzymatic readouts are stable but reflect a single pathway	Multi-analyte panel, matches the assay matrix logic already advocated[14,15]

IDO: Indoleamine 2,3-dioxygenase; PGE2: Prostaglandin E2; MHC: Major histocompatibility complex; PD-1: Programmed death 1; TIM-3: T cell immunoglobulin and mucin domain-containing protein 3; ILT: Immunoglobulin-like transcript; IFN-γ: Interferon gamma; TNF-α: Tumor necrosis factor alpha; PD-L1: Programmed death ligand-1.

That last point addresses an acknowledged bottleneck. Regulatory authorities require potency assays as part of release criteria for advanced-phase clinical trials, yet defining such assays for MSCs has proved persistently difficult, in large part because the mechanism of action in humans is not reducible to a single pathway and because reference standards are lacking[13]. The International Society for Cellular Therapy responded by proposing a matrix of immune functional assays rather than any single readout[14], and subsequent work demonstrated that a combinatorial approach, coupling secretome analysis with transcriptional profiling, captures MSC potency more faithfully than any individual assay[15]. The soluble checkpointome fits this matrix logic naturally. It is multi-analyte by construction, it is measurable by routine immunoassay, and at least one of its members has already been shown to predict the immunomodulatory potential of independent MSC preparations[16].

This review is organised around that dual thesis. We first describe how soluble checkpoint proteins are generated and why the route of generation determines what an assay actually measures. We then define the membership of the module, separating molecules for which MSC release is experimentally demonstrated from candidates that remain unproven, and we set out the signaling routes through which each member is induced and acts. We next consider the determinants that shape the module, before turning to its translational deployment as a potency and pharmacodynamic marker, including the role of predictive modelling. We close with the methodological prerequisites that must be satisfied before any of this reaches the clinic. Throughout, we note that the available human evidence is predominantly associative and derived from heterogeneous in vitro platforms, and we frame our conclusions accordingly.

Prior reviews have treated the components of this problem separately. Reviews of the MSC secretome catalogue soluble mediators such as indoleamine 2,3-dioxygenase, prostaglandin E2 and extracellular vesicles (EVs) but do not organise the checkpoint fraction as a class; reviews of immune checkpoints examine them chiefly on tumor cells or in patient serum rather than as MSC products; and the MSC potency literature argues for multi-assay matrices without specifying a coherent, secretion-based analyte set. Recent work has catalogued MSC-expressed checkpoint molecules and their membrane receptors[17], and has mapped surface-receptor mediated immunosuppression[18], but neither reframes the secreted, soluble fraction as a measurable potency dimension. The contribution of the present review is to join these strands: To define the soluble immunosuppressive checkpoint output of MSCs as a single operational class, to separate established from candidate members by an explicit evidentiary rule, and to connect that class to the unmet need for a secretion-based potency readout.

LITERATURE SEARCH STRATEGY

This is a narrative review structured around a defined conceptual question rather than a systematic review with a pooled effect estimate. PubMed and Web of Science were interrogated from database inception to January 2026 using combinations of the terms “mesenchymal stromal cell”, “mesenchymal stem cell”, “soluble immune checkpoint”, “soluble PD-L1”, “PD-L2”, “galectin-1”, “galectin-9”, “HLA-G5”, “ectodomain shedding”, “potency assay”, and “release criteria”. Reference lists of retrieved articles were screened manually. Primary experimental studies in which release of a checkpoint protein into MSC-conditioned medium was demonstrated, together with regulatory and methodological literature on MSC potency assessment, were prioritised. Studies reporting membrane expression without evidence of release were retained but are designated throughout as supporting candidate rather than established membership of the module. The final search was run on January 15, 2026. Records were screened first by title and abstract and then by full text. Selection followed explicit principles: Priority was given to primary experimental studies demonstrating release of a checkpoint protein into MSC-conditioned medium, followed by regulatory and methodological literature on MSC potency, with reviews used only for context. No date, language or design filter was applied at retrieval; however, only English-language full-text reports were included in the final synthesis, and no non-English record met the inclusion criteria for a primary mechanistic claim. Preprints and conference abstracts were not used to support any primary mechanistic or evidence-classification statement and were cited, if at all, only as signposts to peer-reviewed work. Where sources conflicted, greater weight was given to studies using MSC-derived material, transwell or genetic-perturbation designs and independent replication, and unresolved conflicts are reported as such in the text rather than reconciled. Because the review is conceptual rather than systematic, no formal risk-of-bias scoring or quantitative synthesis was performed, and this is acknowledged as a limitation.

HOW SOLUBLE CHECKPOINTS ARE GENERATED, AND WHY THE ROUTE MATTERS

A soluble checkpoint protein is not a single biochemical species. The extracellular form of a given checkpoint can arise by at least three distinct routes, and these routes have different biological meanings and different analytical consequences. Figure 1 summarises them.

Figure 1 Three routes by which immune checkpoint proteins enter the extracellular space. A: Proteolytic ectodomain shedding. Membrane-anchored checkpoints are cleaved within the juxtamembrane stalk by a disintegrin and metalloproteinase 10 (ADAM10), ADAM17 or matrix metalloproteinases, releasing an active ectodomain while reducing surface density; B: Alternative splicing. Transcripts lacking the transmembrane exon yield an isoform that is secreted constitutively, without any requirement for sheddase activity and without depleting the surface pool; C: Vesicle-associated release. The checkpoint remains membrane-anchored but is displayed on small extracellular vesicles. Because conventional sandwich immunoassays capture epitopes irrespective of the physical state of the analyte, a single reported concentration typically represents the sum of all three pools. ADAM10: A disintegrin and metalloproteinase 10; MMPs: Matrix metalloproteinases; MSC: Mesenchymal stromal cell; sEV: Small extracellular vesicle.

The first route is proteolytic ectodomain shedding. For PD-L1 this has been characterised in detail: A disintegrin and metalloproteinase 10 (ADAM10) and ADAM17 cleave the ectodomain from the surface of malignant cells and from EVs, and the liberated fragment is not inert. It retains the capacity to induce apoptosis in human CD8+ T cells and to impair cytotoxic killing, and elevated expression of these sheddases correlates with poor outcome and with resistance to programmed death-1 (PD-1) pathway blockade[19]. Shedding therefore performs two operations at once. It removes ligand from the membrane and it introduces a diffusible mediator into the surrounding space. An important corollary follows for cell therapy: A preparation with low surface PD-L1 is not necessarily a preparation with low PD-L1 activity, because the protein may have been redistributed rather than lost. It must be emphasised that this shedding mechanism has been defined in tumor-cell and non-MSC systems; direct demonstration of ADAM10- or ADAM17-mediated PD-L1 shedding by MSCs is presently lacking, so the mechanism is cited here as indirect, analogy-based evidence rather than as established MSC biology, and the same caveat applies wherever tumor-derived shedding data are invoked below.

The second route is alternative splicing. Transcripts lacking the exon that encodes the transmembrane domain give rise to isoforms that are secreted by default. HLA-G5 is the paradigmatic example within MSC biology. It is a genuinely secreted product rather than a cleavage remnant, and its release by MSCs is functionally necessary for suppression of alloreactive T-cell proliferation, for inhibition of NK-cell cytolysis and IFN-γ secretion, and for the subsequent expansion of regulatory T cells[12]. Notably, full HLA-G5 secretion in that system required contact between MSCs and allostimulated T cells, which illustrates that soluble output and contact dependence are not mutually exclusive categories.

The third route is vesicular. Here the checkpoint is not soluble in any strict sense; it remains membrane-anchored but is presented on the surface of small EVs. Clinical-grade Wharton jelly MSCs secrete vesicles enriched in PD-L1, antibody blockade of vesicle-associated PD-L1 restores T-cell activation, and vesicles from PD-L1-deficient cells generated by clustered regularly interspaced short palindromic repeats editing lose their inhibitory capacity[20]. In the same study, circulating vesicle-associated PD-L1 rose in patients with acute graft-vs-host disease (GvHD) in temporal association with MSC infusion, which is the closest the field has come to an in-human pharmacodynamic signal for this axis.

The analytical consequence is easy to state and awkward to solve. A sandwich immunoassay recognises epitopes, not physical states. Unless conditioned medium is fractionated before measurement, a reported concentration of soluble PD-L1 conflates shed ectodomain, secreted splice variant and vesicle-displayed protein, three species with different half-lives, different diffusion behaviour and potentially different potency per mole. Table 2 sets out these distinctions and their implications for assay design. This pooling is not an invariant property of immunoassay. Whether a given assay reports free protein only, vesicle-associated protein only, or their sum depends on its capture and detection configuration, on sample pre-processing such as high-speed centrifugation or size-exclusion separation, and on the validation performed; the pooled read-out described here is the default behaviour of an unfractionated sandwich assay rather than a universal one. This is not a pedantic distinction. If a checkpoint-based potency assay is to satisfy regulatory scrutiny, the identity of the analyte has to be defined, and at present the literature reports an operational quantity rather than a molecular one.

Table 2 Routes of generation of soluble immune checkpoint proteins and their analytical implications.

Route	Molecular mechanism	Effect on surface pool	Representative evidence	Implication for assay design
Ectodomain shedding	Cleavage of the juxtamembrane stalk by ADAM10, ADAM17 or matrix metalloproteinases	Reduced	ADAM10 and ADAM17 cleave PD-L1 from tumour cells and vesicles, yielding a fragment that induces CD8+ T-cell apoptosis[19] (note: This evidence derives from tumour-cell and non-MSC systems and is therefore indirect for MSC biology)	Surface staining alone underestimates total checkpoint output; sheddase activity should be reported alongside the analyte
Alternative splicing	Transcript lacking the transmembrane exon, secreted constitutively	Unchanged	HLA-G5 secretion by MSCs is required for T-cell and NK-cell suppression and for regulatory T-cell expansion[12]	Secretion may proceed independently of protease inhibitors, so pharmacological blockade of sheddases is not a valid specificity control
Vesicle association	Checkpoint retained in the vesicle membrane	Redistributed	Wharton jelly MSC vesicles are enriched in PD-L1; blockade or gene editing abolishes their suppressive effect[20]	Requires ultracentrifugation or size-exclusion fractionation before immunoassay if free and vesicle-bound pools are to be distinguished

ADAM: A disintegrin and metalloproteinase; PD-L1: Programmed death ligand-1; MSC: Mesenchymal stromal cell; HLA: Human leukocyte antigen; NK: Natural killer.

Biological consequences of the physical state of soluble checkpoints

The physical state of an extracellular checkpoint is likely to influence its biological behaviour beyond its detectability by immunoassay. Freely soluble ectodomains and secreted splice isoforms can diffuse independently through extracellular fluid and may therefore achieve relatively broad spatial distribution, although their effective half-life will depend on proteolytic stability, receptor binding and systemic clearance. In contrast, EV-associated checkpoints remain embedded in a lipid bilayer, which may protect them from proteolytic degradation, alter their biodistribution and prolong their persistence in biological fluids. Vesicular presentation may also create locally high ligand density and permit multivalent receptor engagement, potentially producing signaling properties that differ from those of monomeric soluble proteins.

These differences remain particularly uncertain in MSC biology. No direct head-to-head study has, to our knowledge, compared matched freely soluble, splice-derived and EV-associated forms of an MSC-derived checkpoint for pharmacokinetics, tissue penetration, receptor occupancy or immunosuppressive potency. Consequently, differences inferred from other biological systems should not be assumed to apply quantitatively to MSC products. Resolving this issue will require physical fractionation of conditioned medium followed by molecular characterization and functional testing of each fraction.

DEFINING THE MSC SOLUBLE CHECKPOINTOME

We restrict membership of the module to proteins that engage an inhibitory receptor on immune cells and for which release into MSC conditioned medium has been demonstrated experimentally. Molecules that are expressed on the MSC surface but for which no soluble MSC-derived form has been reported are treated as candidates rather than members. This distinction is deliberate and is maintained in Figure 2 and Table 3, because conflating the two is precisely the error that has allowed the field to overstate what is known.

Figure 2 The mesenchymal stromal cell soluble checkpointome and its cellular targets. Inflammatory licensing reconfigures the mesenchymal stromal cell (MSC) secretory programme. Established members (shaded) are proteins whose release by MSCs has been experimentally demonstrated and linked to a functional immunological outcome. Candidate members (unshaded) are checkpoint molecules of established immunological importance for which MSC-derived soluble forms have not yet been documented. Arrows denote engagement of the indicated receptor system and the downstream consequence. The net effect is tolerogenic conditioning of the local niche rather than the elimination of any single effector population. Each depicted outcome is bounded by the cited primary evidence: Galectin-1 knockdown restores T-cell proliferation but leaves non-alloreactive natural killer (NK) cells unaffected, so no NK or B-cell effect is attributed to galectin-1; galectin-9 is shown to suppress T and B cells as reported, and human leukocyte antigen-G5 to affect T and NK cells and regulatory T cells. Arrows denote reported functional associations, not proven direct molecular interactions. IFN-γ: Interferon gamma; TNF-α: Tumor necrosis factor alpha; IL: Interleukin; MSC: Mesenchymal stromal cell; PD-L1: Programmed death ligand 1; HLA-G5: Human leukocyte antigen-G5; HVEM: Herpesvirus entry mediator; BTLA: B- and T-lymphocyte attenuator; PD-1: Programmed death-1; TIM-3: T cell immunoglobulin and mucin domain-containing protein 3; ILT: Immunoglobulin-like transcript; NK: Natural killer.

Table 3 Membership of the Mesenchymal stromal cell soluble checkpointome and the strength of supporting evidence.

Molecule	Principal receptor	Regulation in MSCs	Documented functional consequence	Evidence status
Soluble PD-L1 and PD-L2	PD-1	Induced by IFN-γ and TNF-α synergistically; further enhanced by interleukin-1 beta[9,21,22]	Neutralisation restores T-cell interleukin-2 production and PD-1 surface expression; PD-1-dependent regulatory T-cell induction[9,22]	Established; demonstrated in transwell and in vivo models. Evidence level: Human and murine in vivo plus in vitro; applies to secreted PD-L1 and PD-L2 (free forms) as distinct from vesicle-associated PD-L1 (separate row)
Galectin-9	TIM-3	Absent at rest; strongly induced by IFN-γ, TNF-α and Toll-like receptor 2, 3 and 4 ligands[11]	Knockdown removes a significant fraction of antiproliferative activity; suppresses T and B cells; mediates benefit in experimental endotoxaemia[11,16,23,24]	Established; already shown to predict preparation-level potency[16]. Evidence level: In vitro (human MSC); in vivo support in murine endotoxaemia
Galectin-1	CD45 and CD43 glycan ligands	Constitutive; released into supernatant[10]	Knockdown partially restores CD4+ and CD8+ proliferation and alters cytokine release; no effect on non-alloreactive NK cells[10]	Established; effect is population-selective. Evidence level: In vitro (human MSC)
HLA-G5	ILT2 and ILT4	Secreted isoform; full secretion requires contact with allostimulated T cells[12]	Necessary for suppression of alloproliferation and NK cytolysis and for regulatory T-cell expansion[12]	Established; antibody neutralisation abrogates suppression. Evidence level: In vitro (human MSC)
Vesicle-associated PD-L1	PD-1	Enriched on small extracellular vesicles from clinical-grade Wharton jelly MSCs[20]	Blockade or gene editing abolishes suppression of T-cell activation; rises in patient circulation after infusion[20]	Established as a vesicular rather than a freely soluble species. Evidence level: In vitro (human MSC) plus correlational human data after infusion; vesicle-associated, not freely soluble
Soluble CD25	Interleukin-2 (decoy capture)	Not characterised in MSCs	Sequesters interleukin-2 and exacerbates autoimmunity when supplied exogenously[25,26]	Candidate; MSC release unproven. Evidence level: No MSC data; in vivo evidence is from non-MSC systems
Soluble HVEM and BTLA	BTLA, CD160 and HVEM	Not characterised in MSCs	Not determined for an MSC-derived protein	Candidate; MSC release unproven. Evidence level: No MSC data
Soluble CD200 and CD276	CD200R and cognate receptors	Membrane expression reported; shedding not established	Not determined for an MSC-derived protein	Candidate; MSC release unproven. Evidence level: MSC membrane expression only; no release data
CD80/CD86	CTLA-4; CD28	CD80/CD86 expression is inconsistent among MSCs preparations, and there is no direct evidence for soluble CD80/CD86	Potentially context-dependent: CTLA-4 engagement is inhibitory, whereas CD28 engagement is costimulatory	Candidate; soluble MSC-derived forms require direct validation. Evidence level: MSC membrane expression and receptor-context data only; no MSC-derived soluble form demonstrated
CD155 (PVR)	TIGIT; CD96; CD226	Relevant to MSC immune co-signaling, but direct evidence defining an MSC-derived soluble CD155 pool and its contribution to MSC immunosuppression remains insufficient	Receptor-dependent; TIGIT-associated signaling is inhibitory, whereas CD226 can support immune activation	Candidate; MSC release and soluble-form function require validation. Evidence level: MSC membrane expression and receptor-context data only; no MSC-derived soluble form demonstrated

MSC: Mesenchymal stromal cell; PD-L1: Programmed death ligand-1; PD-1: Programmed death 1; IFN-γ: Interferon gamma; TNF-α: Tumor necrosis factor alpha; TIM-3: T cell immunoglobulin and mucin domain-containing protein 3; MSC: Mesenchymal stromal cell; NK: Natural killer; HLA: Human leukocyte antigen; ILT: Immunoglobulin-like transcript; HVEM: Herpesvirus entry mediator; BTLA: B- and T-lymphocyte attenuator; CTLA-4: Cytotoxic T-lymphocyte antigen-4; PVR: Poliovirus receptor; TIGIT: T cell immunoreceptor with immunoglobulin and tyrosine-based inhibitory motif domain.

Established members

Soluble PD-L1 and PD-L2 are the best characterised. Davies et al[9] showed that MSCs both express and secrete these ligands, that expression and secretion are potentiated by IFN-γ and TNF-α acting synergistically, and that neutralisation of the soluble ligands under transwell conditions restores T-cell IL-2 production and reverses the MSC-driven downregulation of surface PD-1. PD-L2 deserves special attention since it does not necessarily serve as an inhibitory immune checkpoint ligand but may exhibit different functions depending on the species and cell type it is expressed in. Thus, PD-L2 is known to serve as an inhibitory immune checkpoint in humans via its interaction with PD-1, resulting in inhibition of T-cell receptor signaling and suppression of cell proliferation and cytokine release. On the other hand, experiments performed on mice have shown that, under certain circumstances, PD-L2 is able to exert costimulatory functions such as enhancing T-cell proliferation and cytokine release. This means that the biological function of the PD-L2 interaction depends on the species, receptor, cell type, and experiment conditions. Accordingly, PD-L2 being mentioned here as part of the soluble immunosuppressive checkpointome means its known inhibitory effect in the MSC-associated PD-1 pathway[20-22].

The transwell design is what gives this observation its force, since it excludes the contact-dependent interpretation that had previously dominated. The pathway is inducible in a manner consistent with a canonical IFN response. In human umbilical cord MSCs, IFN-γ upregulates PD-L1 through Janus kinase, signal transducer and activator of transcription 1 and IFN regulatory factor 1 (IRF1) signaling, and combined IFN-γ and TNF-α exposure synergistically enhances immunosuppressive capacity[21]. Priming with IFN-γ, TNF-α and IL-1β upregulates PD-1 ligands on MSCs and on their vesicles, increases regulatory T-cell induction in a PD-1-dependent fashion, and improves clinical scores and survival in a murine GvHD model, with the benefit reversed by neutralising antibodies against PD-L1 and PD-L2 in vitro and in vivo[22].

Galectin-9 has the strongest claim to biomarker status. It is essentially absent from resting MSCs, is strongly induced upon interaction with activated peripheral blood mononuclear cells or upon exposure to IFN-γ, TNF-α or ligands of Toll-like receptors 2, 3 and 4, and is released into the supernatant in substantial quantity; knockdown of galectin-9 removes a significant fraction of the antiproliferative effect of MSCs on T cells[11]. Its receptor TIM-3 was identified as a negative regulator of T helper type 1 immunity, which supplies the immunological rationale for the effect[23]. Independently, Ungerer et al[16] reported that galectin-9 suppresses both T and B lymphocytes and, critically, that its expression predicts the immune modulatory potential of individual MSC preparations. That last finding is the empirical seed of the thesis advanced here: A member of the checkpointome has already been shown to function as a potency-predictive analyte. Galectin-9 also mediates the therapeutic effect of MSCs in experimental endotoxaemia, which extends its relevance from proliferation assays to in vivo outcome[24].

Galectin-1 differs from galectin-9 in being constitutive rather than induced. It is highly expressed by MSCs, detectable intracellularly, on the cell surface and in the supernatant, and stable knockdown produces a significant loss of immunomodulatory function with partial restoration of CD4+ and CD8+ T-cell proliferation and altered release of TNF-α, IFN-γ, IL-2 and IL-10[10]. Its effect on non-alloreactive NK cells was unaffected by knockdown, indicating that the members of the module are not functionally redundant and that they partition across effector populations.

HLA-G5 completes the established set. Its secretion by MSCs is necessary for suppression of alloproliferation, for inhibition of NK-mediated cytolysis and IFN-γ output, and for generation of regulatory T cells, and neutralising anti-HLA-G antibody abrogates these effects[12]. Because HLA-G engages the inhibitory receptors immunoglobulin-like transcript 2 (ILT2) and ILT4, it is a checkpoint ligand in every functional sense, even though it is rarely classified as one in the MSC literature.

Candidate members and the limits of the evidence

Several checkpoint molecules are plausible members but do not yet meet the evidentiary threshold. Soluble CD25 acts as a decoy receptor that sequesters IL-2 in the T-cell microenvironment, an activity demonstrated directly in vivo[25], and its generation by proteolytic cleavage and by alternative splicing has been reviewed in detail[26]. A stromal cell capable of capturing IL-2 in this manner would possess a potent additional brake, yet MSC-derived soluble CD25 has not been documented. Soluble herpesvirus entry mediator (sHVEM) and soluble B- and T-lymphocyte attenuator (BTLA) are central to bidirectional co-signaling in lymphocytes, but MSC-derived soluble forms have not been convincingly reported. CD200 and the B7 family member CD276 are expressed on MSC membranes, yet whether they are shed in functionally relevant amounts remains open. Listing these as candidates is not a rhetorical hedge. It defines the experimental programme that the concept demands, and it prevents the module from expanding into an unfalsifiable category. Table 3 records the current state of evidence member by member.

Furthermore, candidate status is assigned on mechanistic plausibility rather than evidence of MSC secretion. Specifically, candidate molecules were considered when: (1) Their membrane-bound or soluble forms participate in established immune checkpoint or immunoregulatory pathways; (2) A biologically plausible mechanism exists through which a soluble form could influence immune-cell function; and/or (3) Related membrane-associated molecules have been reported in MSC biology. These criteria are intended to identify testable candidates rather than to imply established membership.

Other immune co-signaling molecules that are related to MSCs can be considered in the list of candidates for inclusion, including CD80/CD86 and the CD155 (poliovirus receptor) family. The problem with CD80 and CD86 lies in the fact that their functionality is dependent on the identity of the receptors with which they interact; in the case of engagement of cytotoxic T-lymphocyte antigen-4, they have a negative function and are part of the inhibitory immune signaling process, whereas CD28 engagement is responsible for a costimulatory response[17]. This means that the presence of CD80 and CD86 on the surface of the MSC is not enough to classify them as inhibitors of the immune system. CD155 is involved in bidirectional immune co-signaling with inhibitory receptors T cell immunoreceptor with immunoglobulin and tyrosine-based inhibitory motif domain and CD96, and an activatory receptor CD226, thus its influence on the immunological process depends on the receptor composition and cell type[18]. While the relevance of these molecules in the process of regulation of the immune system by MSCs needs to be considered, membrane expression or co-signaling capabilities should be separated from the fact of the MSC-synthesized soluble checkpoint[27].

The discussion here has also alluded to CD122 (IL-2Rβ); however, CD122 should be treated separately, since it is a subunit of the IL-2/IL-15 receptor complex and not a traditional checkpoint molecule similar to CD80/CD86 and CD155. Currently, there is no strong evidence that would allow the classification of CD122 in a soluble form as one of the checkpointome members[28]. Consequently, the list of CD80/CD86 and CD155 as candidates for further investigation remains unchanged, while CD122 is excluded from the list.

Accordingly, sCD25 and sHVEM/BTLA should be regarded strictly as hypothesis-generating candidates. At present, direct MSC secretomic evidence demonstrating their release is lacking, and their inclusion in the proposed checkpointome requires experimental validation by targeted proteomics, orthogonal immunoassays and functional perturbation studies.

The candidate landscape is broader than the molecules emphasized above. Other co-signaling molecules reported in association with MSC immunobiology, including CD80/CD86 and members of the CD155-associated axis, merit consideration because they can participate in inhibitory as well as stimulatory receptor networks depending on receptor context. Their inclusion in the soluble checkpointome, however, requires a stricter criterion than membrane expression alone. Specifically, demonstration of an MSC-derived extracellular form, biochemical definition of whether it is freely soluble or vesicle-associated, and evidence of functional receptor engagement are required. At present, the available evidence is insufficient to classify these molecules as established soluble checkpointome members, and they are therefore better positioned as candidates for systematic secretomic investigation. Therefore, sCD25 and sHVEM/BTLA have been selected as hypothesis-generating candidates on the basis of mechanistic plausibility rather than secretomics data demonstrating secretion by the MSCs.

MOLECULAR MECHANISMS OF INDUCTION AND ACTION

The four established members are not induced by the same signal, and they do not converge on the same receptor. Setting their signaling routes side by side, as in Figure 3, clarifies why the module behaves as a coordinated system rather than as a set of interchangeable parts.

Figure 3 Parallel signaling routes through which the four established members of the soluble checkpointome are induced and act. Each lane runs from the initiating signal at the top to the immunological outcome at the bottom, and the lanes converge only at the level of effector suppression. The interferon lane and the inflammation lane are licensed, the galectin-1 lane is constitutive, and the human leukocyte antigen-G5 lane requires cell contact. Terminal boxes denote the receptor engagement and its consequence. Pathway steps shown for each axis summarise the cited evidence; where a step is inferred from non-mesenchymal stromal cell (MSC) systems rather than demonstrated in MSCs it should be read as mechanistic context rather than established MSC biology, and the terminal outcomes do not exceed the effects reported in the cited studies. PD-L1: Programmed death ligand 1; IFN-γ: Interferon gamma; IFNGR: Interferon gamma receptor; JAK: Janus kinase; STAT1: Signal transducer and activator of transcription 1; IRF1: Interferon regulatory factor 1; PD-1: Programmed death-1; TNF-α: Tumor necrosis factor alpha; TLR: Toll-like receptor; TIM-3: T cell immunoglobulin and mucin domain-containing protein 3; NF-κB: Nuclear factor kappa B; NK: Natural killer; IL: Interleukin; HLA-G5: Human leukocyte antigen-G5; ILT: Immunoglobulin-like transcript; Treg: Regulatory T cell.

The IFN-driven PD-L1 axis

Engagement of the IFN-γ receptor activates Janus kinase 1 and Janus kinase 2, leading to phosphorylation of signal transducer and activator of transcription 1 and induction of IRF1, which transactivates CD274 and raises surface PD-L1. This cascade has been demonstrated directly in human umbilical cord MSCs using kinase inhibition and small interfering RNA[21]. Surface protein is subsequently released, and both the soluble and the vesicle-associated forms engage PD-1 on activated T cells[9,20]. Two features of this axis are relevant to potency measurement. It is graded rather than binary, since output scales with cytokine concentration, and it is synergistic with TNF-α, so that the concentration recorded in conditioned medium depends on the precise composition of the licensing cocktail[9,22].

The inflammation-driven galectin-9 axis

Galectin-9 induction is broader in its triggers. IFN-γ and TNF-α induce it, but so do ligands of Toll-like receptors 2, 3 and 4, implying that pathogen-associated molecular patterns can license the module in the absence of an adaptive cytokine signal[11]. The protein accumulates intracellularly and is then released, and it engages TIM-3, a receptor whose ligation contracts T helper type 1 responses[23]. Because Toll-like receptor ligands are common contaminants of manufacturing reagents, this axis also identifies a plausible source of unexplained inter-batch variation.

The constitutive galectin-1 axis

Galectin-1 requires no licensing. It is transcribed constitutively, resides intracellularly and at the cell surface, and appears in the supernatant without stimulation[10]. Its ligands are glycan structures on CD45 and CD43 rather than a classical immunoglobulin superfamily receptor. The functional signature is correspondingly distinct, since knockdown restores T-cell proliferation while leaving NK-cell inhibition intact[10]. For assay design this axis supplies a useful internal control, because a licensed and an unlicensed sample should differ in galectin-9 and PD-L1 while remaining comparable in galectin-1.

The contact-dependent HLA-G5 axis

HLA-G5 is a splice isoform and is therefore secreted without proteolysis, yet its output is not cell-autonomous. Contact between MSCs and allostimulated T cells induces IL-10, which is required for full HLA-G5 release; the secreted protein then engages ILT2 and ILT4 to inhibit T and NK cells and to expand regulatory T cells[12]. This axis demonstrates that a soluble output can nevertheless carry a contact-dependent requirement, which matters for any assay that substitutes recombinant IFN-γ for a cellular responder population. Such a substitution is analytically attractive[15], but it would be expected to underestimate the HLA-G5 component of the module.

IMMUNOLOGICAL LOGIC OF A SOLUBLE CHECKPOINT MODULE

Why would a stromal cell deploy soluble rather than membrane-bound checkpoints? Three features distinguish the soluble mode, and together they explain a set of otherwise puzzling observations about MSC pharmacology. Soluble checkpoints act at a distance. A membrane ligand can only engage a receptor on a cell it touches, whereas a secreted ligand conditions a volume. This offers an economical explanation for the capacity of MSCs to suppress lymphocytes across transwell membranes and to exert effects disproportionate to the number of cells that survive infusion[9]. It also sits comfortably alongside, rather than in competition with, the observation that infused MSCs undergo perforin-dependent apoptosis induced by recipient cytotoxic cells and that this apoptosis is necessary to initiate immunosuppression, with clinical response in GvHD confined to patients whose cells could kill MSCs efficiently[29]. A cell destined to die within hours has limited opportunity to signal by contact; a secretory module discharged during licensing does not have that constraint. Consistent with this, apoptotic MSCs retain functional efficacy after cryopreservation in a murine model of allergic inflammation[30].

Soluble checkpoints are inducible in a graded fashion. The dependence of galectin-9 and PD-L1 on IFN and TNF signaling means that output scales with the intensity of the inflammatory signal the MSC encounters[9,11]. The module therefore encodes a proportional response rather than a switch, which is consistent with the sensor-and-switcher framing of MSC biology[3] and with the observation that IFN-γ licensing suppresses T-cell effector function through pathways independent of indoleamine 2,3-dioxygenase[31].

Soluble checkpoints are functionally partitioned. Galectin-1 knockdown restores T-cell proliferation but leaves NK-cell inhibition intact[10], whereas HLA-G5 neutralisation impairs both T and NK suppression and additionally removes regulatory T-cell expansion[12]. The module is therefore not a redundant safety net but a division of labour across effector compartments. This has a direct implication for potency testing: Any single-analyte assay will be blind to part of the mechanism, which is the same conclusion reached empirically by advocates of the assay matrix approach[14,15].

Two caveats attach to this section. First, most of the causal evidence rests on knockdown or antibody blockade in vitro, and the partial nature of the reversals in every study cited above indicates that no member of the module accounts for suppression on its own. Second, the human data are cross-sectional and associative. The rise in circulating vesicle-associated PD-L1 after MSC infusion in patients with GvHD is temporally associated with treatment[20]; it has not been shown to cause the clinical response.

DETERMINANTS THAT SHAPE THE CHECKPOINTOME

If the module is to be used as a measurement, the variables that move it must be enumerated and controlled. Four dominate, and they are summarised in Table 4. Tissue source is the first. Bone marrow, adipose tissue and birth-associated tissues yield MSCs with broadly comparable phenotypes but subtly different immunomodulatory behaviour, and direct comparisons remain relatively scarce[32]. Because the checkpointome is inducible, source-related differences may reside less in the resting secretome than in the responsiveness of each source to a given licensing signal, a distinction that has rarely been tested explicitly.

Table 4 Determinants shaping the composition of the mesenchymal stromal cell soluble checkpointome and the corresponding standardisation requirement.

Determinant	Reported influence	Standardisation requirement
Tissue source	Bone marrow, adipose and birth-associated MSCs differ in immunomodulatory behaviour; direct comparative data remain limited[32]	Report source explicitly; do not pool analyte reference ranges across sources
Licensing regimen	IFN-γ is the canonical licensing signal[8]; IFN-γ and TNF-α act synergistically on PD-L1[9,21]; interleukin-1 beta further enhances ligand expression and in vivo benefit[22]; licensed cells acquire a distinct potency marker profile[36]	Fix cytokine identity, concentration and exposure duration; treat licensing as part of the assay, not as pre-treatment
Culture and manufacturing	Medium composition, supplement choice and passage number alter MSC function; standardisation is a recognised prerequisite for immunotherapeutic use[37]	Define passage window and medium; qualify reagents, since biological variability of assay reagents degrades reproducibility[38]
Cryopreservation and cell state	Freezing alters functional properties; apoptotic MSCs nevertheless retain efficacy[29,30]	Report the interval between thaw and sampling; viability alone is an inadequate surrogate for potency
MSC subpopulation heterogeneity	Distinct MSC subpopulations may differ in transcriptional state, secretory activity and immunomodulatory function	Characterize cellular composition using subpopulation-resolved/single-cell approaches and determine its relationship with checkpoint output

MSC: Mesenchymal stromal cell; IFN-γ: Interferon gamma; TNF-α: Tumor necrosis factor alpha; PD-L1: Programmed death ligand-1.

Cellular heterogeneity within a nominal MSC population represents an additional determinant that may be at least as important as tissue source[33]. MSC preparations are not composed of a single transcriptionally and functionally uniform cell type; rather, they contain subpopulations with distinct differentiation states, proliferative capacities, secretory profiles and immunomodulatory phenotypes. Single-cell and subpopulation-resolved studies have increasingly demonstrated this heterogeneity across MSC sources[34]. It is therefore plausible that soluble checkpoint output reflects not only the average response of the culture but also the relative abundance of checkpoint-high and checkpoint-low subpopulations.

This possibility has direct implications for potency testing. Two MSC batches with similar conventional identity markers may produce different soluble checkpoint profiles because their underlying cellular compositions differ. Future checkpointome studies should therefore combine secretome measurements with single-cell or subpopulation-resolved phenotyping to determine whether specific MSC states preferentially produce PD-L1/PD-L2, galectins or HLA-G5. Such analyses may also help distinguish true donor-to-donor variation from shifts in subpopulation composition introduced during expansion, licensing or cryopreservation[35].

Several recent findings from single-cell and subpopulation-resolved investigations support the view that MSC heterogeneity transcends tissue differences. Functional states characterized by transcriptional profiles related to lineage commitment, extracellular matrix organization, proliferation, metabolism, and immune regulation have been identified. Notably, the proportion of each state can be altered by ex vivo culture and may differ among MSC preparations even when standard phenotypic assays yield equivalent results. It is especially relevant for the checkpointome concept, since bulk quantification of immunoregulatory soluble factors could underestimate the role of rare but highly effective MSC subpopulations. Single-cell analysis combined with secretome profiling might be useful in establishing whether checkpoint production is a widespread property of all MSCs or a feature of certain subpopulations of functional cells[33-35].

Licensing regimen is the second and probably the most consequential. IFN-γ is the canonical licensing cytokine[8], its combination with TNF-α acts synergistically on PD-L1 in both bone marrow and umbilical cord MSCs[9,21], and the further addition of IL-1β strengthens PD-1 ligand expression on cells and vesicles and improves outcome in experimental GvHD[22]. Licensed MSCs also acquire a characteristic potency marker expression profile that correlates with their immunomodulatory capacity[36]. Because different laboratories use different cytokine concentrations and exposure times, cross-study comparison of absolute analyte concentrations is currently unsafe.

Manufacturing and culture conditions constitute the third. Good manufacturing practice-compliant production imposes constraints on medium composition, passage number and supplement choice that are known to alter MSC behaviour, and standardisation of these steps has been recognised as a prerequisite for immunotherapeutic use[37]. Cryopreservation is the fourth. It affects functional properties in a context-dependent manner, and the observation that apoptotic MSCs retain efficacy after freezing[30] indicates that viability-based release criteria may be poorly aligned with the mechanism that actually matters.

FROM MECHANISM TO MEASUREMENT: A CONCEPTUAL FRAMEWORK FOR CHECKPOINT-BASED POTENCY ASSESSMENT

The regulatory problem is precisely stated. Culture-expanded MSCs are classified as more than minimally manipulated cell and gene therapy products, and for advanced-phase trials the authorities require potency assays as part of release criteria. Meeting that requirement has been obstructed by variability of tissue source, by mechanisms of action that are undefined in humans, and by the absence of reference standards[13,15]. The response of the field has been to abandon the search for a single readout in favour of an assay matrix that captures overlapping effector pathways[14,15], with functional co-culture assays serving as an imperfect gold standard whose reproducibility depends heavily on reagent qualification[38].

The soluble checkpointome offers a matrix that is already biologically coherent. It has four properties that a release assay requires. It is inducible by a defined and reproducible stimulus, which allows IFN-γ to be used as a surrogate for activated mononuclear cells and thereby removes the need for a variable human responder population[15]. It is multi-analyte, and its members partition across effector compartments so that a panel carries more information than any single member. It is measurable in cell-free conditioned medium by multiplex immunoassay, so the product itself is not consumed. And at least two of its members have independent precedent as potency markers: Galectin-9 expression predicts the immune modulatory potential of MSC preparations[16], and inducible indoleamine 2,3-dioxygenase 1 together with PD-L1 has been proposed explicitly as a potency marker for MSCs[39]. These properties describe analytical suitability, not demonstrated performance. The composite score proposed here has no validated analyte weights, no defined threshold, and no established diagnostic or predictive accuracy, incremental value or prospective clinical association. Every downstream application discussed, including donor screening, batch release, patient stratification and any machine-learning model, is therefore framed strictly as a future validation study rather than as a current or ready-to-use capability.

Figure 4 sets out the workflow we propose. A licensed conditioned medium sample is assayed for the panel, the analyte concentrations are combined into a composite score normalised against a reference batch, and the score is then used for donor screening, batch release and, prospectively, patient stratification. We deliberately avoid proposing fixed weights or cut-off values. No dataset currently exists that would justify them, and premature specification of thresholds is one of the ways in which biomarker programmes fail.

Figure 4 Proposed translational workflow for a checkpoint-based potency assay presents a hypothetical development framework rather than a validated clinical workflow. It illustrates the experimental and analytical steps that would need to be evaluated before a checkpoint-based composite score could be considered for donor screening, batch qualification or patient stratification. No element of this workflow is validated; it defines future validation studies rather than a usable clinical procedure. IFN-γ: Interferon gamma; TNF-α: Tumor necrosis factor alpha.

Furthermore, the patient stratification deserves separate comment, because it is where the concept could matter most and where it is most speculative. Clinical response to MSC therapy in GvHD depends on the recipient rather than on the product alone, since only patients whose cytotoxic cells could induce MSC apoptosis responded to infusion[29]. A potency score computed solely from the product would therefore be incomplete. A pharmacodynamic readout taken from the patient after infusion, for which circulating vesicle-associated PD-L1 is the existing precedent[20], addresses the recipient side of the equation. The two measurements are complementary, and neither has been validated prospectively.

Importantly, no paired dataset currently links a composite MSC checkpoint profile to prospective clinical outcomes. The proposed score therefore has no validated weighting scheme, threshold, predictive accuracy or demonstrated incremental value over existing potency assays. Its current purpose is to define a testable biomarker-development strategy rather than to recommend implementation as a release criterion.

The proposed composite checkpoint score should therefore be regarded as a design for future validation studies, not a clinically usable potency score or decision-making tool. Its use for batch release, donor screening, or patient selection will require paired product-outcome data, prospective validation, and independent verification before it can have any clinical value.

If a single first setting is to be prioritised for prospective evaluation, steroid-refractory acute GvHD is the most defensible choice. It is the indication in which MSC therapy is most advanced, in which the only existing human pharmacodynamic signal for this axis has been observed, namely a rise in circulating vesicle-associated PD-L1 after infusion, and in which recipient-side determinants of response are already partly characterised. A correlative study nested within an existing trial in this population, pairing a standardised checkpoint panel measured on the product with the established clinical response criteria, would provide the first test of whether the panel carries information beyond current release assays, without exposing patients to any change in their treatment.

PREDICTIVE MODELLING AND DECISION SUPPORT

Composite scores require weights, and weights assigned by intuition are indefensible under regulatory review. This is the natural point of entry for computational methods. The argument has recently been made explicitly for MSC potency: Because MSC function rests on overlapping and cumulative effector pathways rather than a single mechanism, an assay matrix is required, and artificial intelligence tools can enhance that matrix by generating potency scores that unify critical quality attributes not readily discernible by inspection, potentially allowing patients to be matched to products with an appropriate potency profile[40]. The same logic has already been applied within MSC potency analysis, where matrix-based analytical strategies capture multidimensional responses that single assays miss[15,41].

The broader landscape supports this direction while also cautioning against enthusiasm. Artificial intelligence methods applied to MSC imaging have been surveyed systematically and judged effective, yet the same survey identifies subjective assessment and the absence of standardised evaluation criteria as unresolved constraints[42]. In systemic lupus erythematosus, artificial intelligence has been proposed as a means of predicting MSC behaviour under different priming conditions and of designing personalised therapy, precisely because efficacy is inconsistent across donors, culture conditions and host immune contexts[43]. Neither body of work has yet been applied to a soluble checkpoint panel.

Three conditions govern whether such modelling would be legitimate here. The training data must pair analyte concentrations with a clinically meaningful outcome, not merely with an in vitro suppression percentage, since the latter is itself a surrogate of uncertain validity. The number of donors must substantially exceed the number of analytes, or the model will describe noise. And the resulting score must be shown to add information beyond the assays already in use, because a composite that merely recapitulates a mixed lymphocyte reaction offers convenience rather than knowledge. None of these conditions has been satisfied for the checkpointome, and we present the modelling component as a design specification rather than as a validated method.

Accordingly, artificial intelligence and machine-learning approaches are discussed here as potential future analytical tools rather than validated components of MSC potency testing. Their value can only be established after sufficiently large, independent datasets containing paired checkpoint measurements, conventional potency assays and clinically meaningful outcomes become available.

LIMITATIONS, OPEN QUESTIONS AND A RESEARCH AGENDA

The concept advanced in this review is a synthesis, and syntheses inherit the weaknesses of the studies they combine. Four limitations are material. The evidence is predominantly in vitro and mechanistically partial. Every functional study cited above reports partial rather than complete reversal of suppression after removal of a single member, which is consistent with a module but which also means that no member has been shown to be sufficient. The human evidence is associative. Circulating vesicle-associated PD-L1 rises after MSC infusion[20]; no study has demonstrated that manipulating a soluble checkpoint in patients alters clinical outcome. The analyte is operationally rather than molecularly defined, for the reasons discussed in “Defining the MSC Soluble Checkpointome”. Until free and vesicle-bound pools are separated before measurement, reported concentrations of soluble PD-L1 are composite quantities of uncertain interpretation. Pre-analytical variables are unharmonised. Licensing regimen, medium, passage number, cryopreservation and the interval between thaw and sampling all influence the module, and none is standardised across the studies from which the concept is assembled[30,37]. Finally, the direction of the effect cannot be assumed. Shed PD-L1 induces apoptosis in activated CD8+ T cells in the tumor context[19], where it is unambiguously detrimental to the host. In the cell therapy context the identical activity would be therapeutically desirable. The molecule has no intrinsic valence, and a high checkpointome score is favourable only relative to a defined clinical objective. Table 5 translates these limitations into an actionable agenda.

Table 5 Research agenda for qualifying the mesenchymal stromal cell soluble checkpointome as a potency and pharmacodynamic marker.

Priority	Open question	Proposed approach	Criterion for success
Analyte definition	What fraction of measured soluble checkpoint is free protein rather than vesicle-associated?	Size-exclusion chromatography or ultracentrifugation of licensed conditioned medium before multiplex immunoassay	Reproducible partition ratio across donors and licensing conditions
Candidate membership	Do MSCs release soluble CD25, HVEM, BTLA, CD200 or CD276 in functionally relevant amounts?	Targeted proteomic interrogation of licensed and unlicensed secretome with orthogonal immunoassay confirmation and functional blockade	Detection above a pre-specified limit of quantification, with a demonstrable functional effect
Donor-level correlation	Does a composite checkpoint score correlate with independent measures of suppressive capacity?	Paired measurement of the panel and a qualified immunopotency assay across a donor panel of adequate size	Correlation demonstrated in a pre-registered analysis with correction for multiple comparisons
Pre-analytical harmonisation	How much of the observed inter-study variance is attributable to licensing and handling?	Multi-laboratory ring study using a shared protocol and a common reference batch	Inter-laboratory coefficient of variation within a pre-defined acceptance limit
Clinical qualification	Does the score predict clinical response, and does it add information beyond existing assays?	Prospective correlative study nested within an ongoing MSC trial, with pharmacodynamic sampling after infusion	Incremental predictive value over the existing release assay, assessed prospectively

MSC: Mesenchymal stromal cell; HVEM: Herpesvirus entry mediator; BTLA: B- and T-lymphocyte attenuator.

This valence problem has concrete, disease-specific consequences that any translational use of the framework must weigh. Amplifying soluble checkpoint activity may be advantageous in GvHD or in autoimmune indications, where restraint of effector lymphocytes is the therapeutic aim, yet the identical activity could be harmful in settings where immune surveillance must be preserved. In patients with malignant disease, or those harbouring occult tumors, enhanced PD-L1 or galectin-9 signaling could in principle blunt antitumor immunity, and in patients with active or latent infection it could impair pathogen clearance[44]. A high checkpointome reading is therefore not a universal marker of a better product; its desirability is conditional on the clinical objective, and a potency-enhancement strategy in one indication may constitute a safety liability in another.

A particularly important limitation is the hierarchy of the available evidence. Mechanistic support for the proposed checkpointome is derived predominantly from in vitro systems, including transwell experiments, neutralizing-antibody studies and gene knockdown, supplemented by animal models such as experimental GvHD and endotoxaemia. These approaches establish biological plausibility but cannot establish that the same pathways determine therapeutic efficacy in humans. The principal human observation currently available is the increase in circulating EV-associated PD-L1 following MSC infusion in patients with GvHD. This temporal association provides pharmacodynamic evidence of pathway engagement but does not establish causality, because neither selective manipulation of the checkpoint pathway nor prospective demonstration of an association with clinical response has been performed. The clinical relevance of the soluble checkpointome must therefore remain provisional until prospective interventional or rigorously designed correlative studies are available.

One notable limitation of the existing evidence is that there is a disconnect between experimental results and clinical validation. Most of the evidence pointing to the participation of soluble checkpoint molecules in the MSC-based immunomodulatory mechanism comes from in vitro work and animal models using transwell experiments and pathway inhibition. While such work offers useful mechanistic information, it does not automatically mean that the same pathways are responsible for the observed therapeutic effect of MSC therapy in humans[20]. Evidence in humans remains scarce and mainly associative. For instance, the increase in EV-bound PD-L1 levels in the plasma of GvHD patients following the administration of MSCs only indicates the activation of this pathway, but it does not provide a causal link between its activation and clinical outcomes[20].

CONCLUSION

Soluble PD-L1 and PD-L2, galectin-9, galectin-1 and HLA-G5 have each been studied in isolation within MSC biology, and each has been shown to contribute measurably to immunosuppression. Read together, they describe a coherent secretory module that is inducible by inflammatory licensing, distributed across effector compartments, and quantifiable in cell-free medium. Naming this module the soluble checkpointome is useful only if the name does work, and the work it does is translational. The field has struggled for more than a decade to define a potency assay that regulatory authorities will accept, and the obstacles have been the multiplicity of mechanisms, the absence of reference standards, and the impracticality of functional co-culture assays as release criteria. A licensed, multi-analyte, cell-free checkpoint panel addresses each of these obstacles directly, and it does so using measurements that most cell therapy laboratories can already perform. We emphasise what has not been shown. No prospective study has demonstrated that a checkpoint score predicts clinical response. The analyte itself remains operationally defined. The candidate members of the module may prove not to be members at all. These are not reasons to defer the question; they are the question. Framing the soluble checkpointome explicitly, and stating the conditions under which it would be falsified, converts a scattered literature into a testable programme, and it is on that basis that we offer the concept. The MSC soluble checkpointome should therefore be regarded as a testable conceptual framework rather than an experimentally demonstrated synergistic protein complex. Its proposed members have different levels of evidentiary support, and candidate molecules should not be interpreted as established components until MSC-derived secretion and functional activity are directly demonstrated. The concept encompasses separately discovered soluble immunological checkpoints, but experimental evidence of interactions or synergy among them is still needed.