Liver transplantation remains the only definitive treatment for patients with end-stage liver disease, acute liver failure, or select liver cancers.¹ Since its introduction in the 1960s, it has become an established and essential component of clinical care, offering sustained survival and improved quality of life.² Despite major advances in surgical techniques, immunosuppression, and perioperative management, key challenges remain.³ Limited organ availability continues to restrict access,⁴ and immune-mediated complications such as rejection and graft dysfunction still contribute to morbidity and threaten long-term outcomes of the liver transplant recipients.^{5, 6, 7} It is thus important to be able to reduce and or eliminate the need for a lifetime immunosuppression in transplant recipients.² The liver is a highly immunologically active organ, and transplantation triggers immediate immune surveillance by the recipient.^{8, 9} Lymph nodes strongly contribute to this process, acting as one of the first sites where donor antigens are first encountered and processed by the immune system.^{10, 11, 12} In particular, hepatic draining lymph nodes (HDLNs) filter lymph from specific peripheral tissues, concentrating both self and non-self antigens to enable their recognition by the immune system.¹³ This localized antigen accumulation allows for efficient presentation to naïve T cells, primarily by professional antigen-presenting cells (APCs) such as dendritic cells, macrophages, and B cells, which are essential for priming antigen-specific immune responses.^{14, 15}

The lymph nodes provide spatially optimal sites that bring APCs into contact with T cells, initiating an adaptive immune response upon exposure to alloantigens.^{11, 16} This activation can occur through multiple, overlapping routes. Canonically, donor “passenger” leukocytes exit the graft and migrate into recipient lymphoid tissues, where they directly present donor MHC–peptide complexes to recipient T cells (direct pathway).^{17, 18} In parallel, more recent work in kidney and liver transplant in humans has identified an alternative and complementary mechanism involving donor-derived extracellular vesicles (EVs).¹⁹ Released almost immediately following reperfusion, these EVs enter the recipient’s circulation and have been shown to rapidly traffic to draining lymph nodes (DLN) in preclinical models of organ transplantation.¹⁵

In the DLN, recipient APCs either acquire intact donor MHC–peptide complexes by membrane transfer (“cross-dressing,” semidirect pathway) or phagocytose donor material and process it for presentation on recipient MHC molecules (indirect pathway).¹⁸ Acting together, donor APCs in the direct pathway, cross-dressed recipient APCs in the semidirect pathway, and recipient APCs engaging the indirect pathway can initiate alloreactive T-cell recognition within minutes to hours,²⁰ causing subsequent T-cell activation over days to weeks.^{21, 22} Activated CD8⁺ T cells then home back to the graft, where they induce apoptosis of donor cells through perforin/granzyme release or Fas–FasL signaling, while CD4⁺ T cells secrete proinflammatory cytokines (such as IFN-γ and TNF-α) and provide help for cytotoxic T-cell and antibody responses, causing graft injury and rejection.²³ These preclinical models support the idea that similar early alloimmune events may also occur within draining lymph nodes in human liver transplantation. In fact, histopathologic analysis of explanted human liver allografts has revealed donor–recipient immune cell interactions within the donor hepatic draining lymph node, suggesting that this site is immunologically active during rejection.²⁴ Clinically, high-dose corticosteroids are used to rapidly quell this immune attack by inducing T-cell apoptosis and suppressing cytokine production; however, their broad immunosuppressive effects can lead to infection, metabolic complications, and hinder tissue repair, making steroid minimization or avoidance a key clinical goal.²To date, most studies of acute rejection have focused on later timepoints, ranging from days to weeks after transplant, and have largely overlooked the recipient lymphatic environment, where the earliest immune decisions are likely made.^{9, 12} From an orthotopic liver transplant (OLT) archive, we obtained paired recipient HDLN samples collected pre-reperfusion and approximately 1-hour post-reperfusion. From this cohort of individuals, we provide one of the first paired studies investigating the initial immune responses within the HDLN that may give rise to acute liver transplant rejection in humans. To characterize this, we applied multiplexed imaging to capture spatial protein expression as well as GeoMx whole transcriptome atlas to quantify transcript abundance within selected regions of interest for each pair of samples. We integrated these data through several computational methods that define spatial immune organization, cell-type interactions, marker expression profiles, and signaling pathway activation. Using these approaches, we identified macrophage–CD4/CD8 + T cell interactions to be associated with acute rejection, while dendritic cell-CD4/CD8 + T cell interactions correlated with non-rejection. This is the first study linking ∼1-hour post-reperfusion changes in the hepatic draining lymph node microenvironment to downstream clinical outcome in humans.

From each pre- and post-reperfusion HDLN pair in the cohort (Fig. 1A; Supp. Table 1), we profiled spatial gene expression using NanoString GeoMx Whole Transcriptome Atlas (WTA). GeoMx WTA measures 18,269 human genes by in situ hybridization with UV-releasable, barcode-tagged probes on FFPE sections(Fig. 1B).²⁷ We defined circular regions of interest (ROIs; 600 µm diameter) centered on germinal centers and cortical T-cell zones, the compartments where early adaptive responses are concentrated, and quantified mRNA counts per ROI (Fig. 1C–D). Within each ROI, we further partitioned signal using specific immunofluorescence-based morphology markers: CD3 for T cells, CD20 for B cells, and CD68 for macrophages (Fig. 1D). While these single markers alone are insufficient to classify each respective cell type, they provide broad transcriptomic information within each distinct immune segment. Gene counts were normalized across samples and segments with DESeq2 to mitigate differences in capture efficiency and overall depth (Supp. Figure 1B).

In parallel, we performed multiplexed immunofluorescence (MxIF) imaging of a 35-marker immune panel on the same samples (Fig. 1B-C; Supp. Table 2). This approach enabled the construction of a high-resolution single-cell spatial proteomic atlas encompassing millions of cells, providing a detailed view of the immune landscape across both time points. We annotated eight major immune cell populations (see methods), including helper and cytotoxic T cells, regulatory T cells (Tregs), B cells, dendritic cells (DC), macrophages, natural killer (NK) cells, and a small population of unclassified cells (Fig. 1D-E).

To connect spatial protein context to local gene activity, we performed k-nearest neighbor (KNN) niche analysis on the MxIF single-cell coordinates (Fig. 1E). Neighborhoods were seeded from four marker-defined zones centered on marker-positive cells: CD3 for T-cell–rich niches, CD20 for B-cell–rich niches, CD68 for macrophage-rich niches, and a fibrotic/adipose zone. These marker-defined zones were used strictly as heuristic spatial anchors for cross-modal alignment with GeoMx WTA and were not treated as definitive cell-type classifications. The fibrotic/adipose class was excluded from downstream analyses and was not represented within the GeoMx ROIs. We then overlaid these KNN neighborhoods onto the GeoMx ROI footprints and compared them to the GeoMx morphology segments. Features from the CD3-, CD20-, and CD68-enriched neighborhoods aligned with the corresponding RNA-seq subdivisions within each ROI, providing a direct cross-modal link that we used to match local protein measurements to the paired transcriptional profiles (Fig. 1D-E).

We next assessed whether pre- vs post-reperfusion status introduced confounding structure. Principal component analysis of the MxIF markers, performed both on unscaled data and after z-score scaling, showed no dominant separation by time point in the unscaled space, indicating that the pairs remain suitable for within-recipient comparisons. After scaling, paired samples moved closer in the reduced space, consistent with shared architecture and donor-independent features. We observed a similar pattern for the GeoMx WTA counts (Supp. Figure 1C-D). Furthermore, global cell-type composition was summarized as stacked bar plots with per-case standard deviations. The distributions of major lineages were comparable between pre- and post-reperfusion nodes, supporting architectural similarity at the level of broad immune compartments (Supp. Figure 1E).

To assess cellular heterogeneity by clinical outcome, we computed t-SNE embeddings from a balanced random sample of 50,000 cells from rejection cases and 50,000 from non-rejection cases (Fig. 1F). The embeddings did not segregate by outcome, indicating that single-cell multiplex marker profiles alone are insufficient to distinguish these groups. We then overlaid curated cell-type labels and intensity maps for CD4, CD8, CD20, CD68, and CD11c. After arcsinh normalization for display, each marker localized to the expected cell-type neighborhoods, consistent with the assigned annotations. Together, these visual checks support the validity of downstream cross-modal analyses.

To identify acute changes in cellular interactions that may correlate with rejection vs. non-rejection outcomes, we analyzed our MxIF dataset to map the spatial organization of the HDLN. We hypothesized that reperfusion-induced, localized changes in the lymph-node microenvironment could distinguish clinical outcomes following allograft reperfusion. To test this, we calculated the likelihood of specific immune cell types being located within a 40-micron radius of one another using SciMap’s Spatial Interaction Analysis.²⁸ Comparing these interactions against a permuted background yielded Z-scores that quantify the strength and direction of spatial associations (Fig. 2A). To capture reperfusion-associated changes, we then computed delta Z-scores by comparing post- versus pre-reperfusion interactions for each case independently and then calculated the average. We then related these spatial changes to clinical outcomes by calculating the log₂ fold change in interaction scores (post- vs. pre-reperfusion) and identifying interactions significantly enriched in rejection versus non-rejection. In parallel, we trained multiple machine learning models—including logistic regression, XGBoost, and random forest—on ROI-level interaction data to predict rejection outcomes (Fig. 2B; Supp. Figure 2A-D). In this case, we focused our analysis on the subsets of cells that were contained within our GeoMx-defined ROIs because they provided increased statistical power due to their higher number per sample. By comparing feature importance across models, we identified a core set of spatial interaction patterns consistently associated with rejection versus non-rejection. This convergence highlights robust, reproducible immune cell relationships that are correlated with downstream acute rejection outcomes (Fig. 2B).

A closer inspection of these spatial interaction patterns revealed that proximity between helper T (CD4+) and cytotoxic T (CD8+) cells and either macrophages or DCs was most strongly associated with rejection versus non-rejection based on the computed spatial graphs and machine learning model feature importances (Fig. 2A–B), aligning with prior observations that APC–T cell contact underpins early alloimmunity. In non-rejection, both macrophages and DCs were initially segregated from both CD4 + and CD8 + T cells; however, following the allograft reperfusion, we observed an increase in DC to CD4+/CD8 + T cells and macrophage to CD4 + T cells, as indicated by positive delta-Z scores. On the other hand, in rejection, we observed no changes in post- compared to pre- reperfusion samples for DC to CD4 + or CD8 + T cell contacts, as these were constitutively present between both timepoints. Instead, we saw a large positive increase in delta-Z scores for macrophage to CD4+/CD8 + T cells in the post- compared to pre-HDLN for rejection, suggesting a role for these interactions in promoting acute rejection (Fig. 2A).

While DCs are conventionally regarded as the principal APC driving acute transplant rejection,^{20, 29, 30} our observations pointing to macrophage–T cell contacts may be important in the early window examined in human OLT.³¹ Canonically, acute transplant rejection is driven by T cell activation in response to donor-derived alloantigens presented on the surface of APCs such as B cells, DCs, and macrophages. On the ~ 1h timescale we are observing, we speculate that this antigen presentation could be mediated by cross-dressing,³² a mechanism that allows host APCs to rapidly present intact alloantigens through mechanisms such as cell contact or EV uptake without prior processing.

We next computed Squidpy co-occurrence probabilities for all pairwise interactions after subsetting to helper T (CD4+), cytotoxic T (CD8+), macrophages, and DCs to align better with the previously computed Z-scores. Each ROI was treated as an independent spatial unit, and scores were averaged within timepoint–outcome strata. Co-occurrence is defined as the conditional probability that a cell of type A has a neighbor of type B in the spatial graph, compared against a permuted label background to assess enrichment. These enrichment scores are computed as a function of spatial distance, providing a more interpretable framework compared to Scimap’s z-scores. After reperfusion, non-rejection samples showed increased DC–CD8 + co-localization, whereas rejection samples showed increased macrophage–CD4 + co-localization, consistent with macrophages’ MHC class II–preferred engagement of Helper T cells. A DC–CD8 + T cell co-occurrence signal was also detectable in the rejection group at both pre- and post-reperfusion HDLN, but at slightly lower magnitude than in non-rejection. These patterns complement the ROI-level features derived from Scimap’s spatial-neighbor Z-scores; however, this analysis prioritized specific APC-T-cell pairs. These discrepancies may be explained by the slightly different approaches employed, with Squidpy focusing on per-cell co-localization, while Scimap aggregates a proximity-enrichment readout (Fig. 2C).

To further investigate mechanisms distinguishing rejection from non-rejection after liver allograft reperfusion, we leveraged the previously described niche analysis (Fig. 1E) as an orthogonal spatial framework. Instead of testing pairwise proximity or co-occurrence, this approach resolves marker-anchored spatial neighborhoods in which we can probe cell-state behavior by region, complementing the Scimap and Squidpy metrics. To illustrate, we subset to either CD4/CD8 + T cells with DCs or CD4/CD8 + T cells with macrophages and computed KNN neighborhoods (K = 2) for each pairing, then plotted cell centroids in representative ROIs across pre- and post-reperfusion and non-rejection/rejection conditions (Fig. 2D). To quantify intermixing between the two niches, we built a kNN graph using the same K and neighborhood definition as in the two-cluster niche analysis and defined a mixing score as the fraction of kNN edges that connect a cell to the opposite niche (Fig. 2E). Consistent with the representative ROIs, DCs are relatively peripheral to CD4/CD8⁺ T cells pre-reperfusion but become more intermixed post-reperfusion. By contrast, we did not detect significant changes in macrophage–T-cell mixing across conditions (Fig. 2E).

Next, we applied the KNN = 4 niche analysis (as in Fig. 1E; CD20, CD68, fibrotic/adipose, CD3) to the post-reperfusion HDLN, where APC–T engagement is most pronounced across outcomes. Within these niches, we quantified a curated panel of markers (HLA-DR, HLA-A, HLA-E, CD40, PD-L1, VISTA, CD14, CD38, CD39, CD44, IDO1, T-bet) and compared expression inside versus outside the CD3 niche for DCs and macrophages (Fig. 2F). While rejection samples showed no significant niche-specific shifts, non-rejection exhibited coordinated upregulation of CD39, IDO1, and HLA-DR (log2FC > 1 and p-value < 0.05) on both DCs and macrophages. CD39 and IDO1 impose local brakes on T cells (via adenosine and kynurenines), and HLA-DR marks active antigen presentation; together, their co-expression may define a regulatory APC interface within the T-cell zone (Fig. 2F). Taken together, these data suggest that CD4/CD8 + T cell–DC contacts in non-rejection occur in an immunoregulatory antigen-presenting context, whereas rejection appears dominated by both CD4/CD8+–DC and CD4/CD8+–macrophage interactions lacking these programs.

What other factors determine whether the immune response ultimately shifts toward rejection versus non-rejection? An additional possibility is that the outcome is not solely dictated by APC–T-cell interactions but also by the inhibitory “tone” of the interacting T cells. In non-rejecting recipients, T cells may be checkpoint-high, such that they are less responsive to pro-inflammatory cues from antigen-presenting macrophages, allowing DC-derived inhibitory cues to shape the immune environment (Fig. 2E).³⁶ In contrast, in rejecting individuals, T cells may remain highly responsive to macrophage-derived activation cues, leading to sustained inflammation and eventual graft injury.³⁷ To test this hypothesis, we quantified PD-1 and LAG-3 on CD4 + and CD8 + T cells and PD-L1 on macrophages and dendritic cells across the whole slide in the pre-reperfusion HDLN. We observed higher PD-1 (4.1-fold on CD8 + and 2.5-fold on CD4+) and LAG-3 (2.2-fold on CD8 + and 1.5-fold on CD4+) on both T-cell compartments in non-rejection recipients compared with those who rejected, together with elevated PD-L1 on macrophages (1.5-fold) in non-rejection. Dendritic cells displayed consistently high PD-L1 in both groups (0.88-fold, not differentially enriched) (Fig. 2H–I). We interpret these features not as evidence of bona fide T-cell exhaustion, which classically reflects prolonged antigenic stimulation and a broader transcriptional/functional program, but rather as contextual indicators of pre-existing inhibitory conditioning within the lymph node.

Together, these findings support a model in which early APC–T cell interactions in the draining lymph nodes might be critical for post-transplant allograft outcome.^{38, 39, 40, 41, 42} In OLT cases with post-transplant rejection, CD4 + and CD8 + T cells remain responsive to alloantigen presented by APCs, with transcriptional evidence of pro-inflammatory effector programs compatible with eventual graft injury. By contrast, non-rejection may be due to reduced T-cell activation and limited early priming, aligning with tolerance-associated programs.

To link shifts in HDLN spatial architecture with cellular signaling, we analyzed GeoMx WTA within defined ROIs across pre- versus post-reperfusion and rejection versus non-rejection groups (Fig. 3A). To evaluate concordance between GeoMx WTA and MxIF in matched regions, we compared nuclei counts for CD3, CD20, and CD68 segments in MxIF-defined neighborhoods with counts from the corresponding GeoMx morphology segments within each ROI (Fig. 3A-B). Estimates for CD3 + cells and CD20 + cells were concordant across platforms. However, CD68 + cells were consistently lower by GeoMx WTA than by MxIF, suggesting underestimation of macrophage-rich areas in the WTA, likely the result of limitations with NanoString’s Digital Spatial Profiler segmentation in certain contexts⁴³.Furthermore, because our spatial analyses highlighted the reorganization of T cells, macrophages, and DCs in relation to rejection outcome, we focused GeoMx WTA profiling on the available CD3 and CD68 segments (T cells and macrophages). B cells, corresponding to CD20 segments, were less prominent in the spatial patterns and therefore were not emphasized in our interpretation of the transcriptomic data.

We conducted three differential-expression comparisons across all GeoMx segments: 1) HDLN Post vs. HDLN Pre to capture rejection-invariant, reperfusion-associated changes, 2) HDLN Pre Rejection vs. HDLN Pre non-Rejection to identify pre-existing features associated with subsequent rejection, and 3) HDLN Post Rejection vs. HDLN Post non-Rejection to identify outcome-specific signals present at ~ 1 h post-reperfusion. Using log2 fold changes and p-values from DESeq2, we ran Qiagen Ingenuity Pathway Analysis (IPA) on each subset and quantified pathway activation using IPA z-scores.

Inspection of the CD68 + compartment showed an AMA phenotype characterized by IL-4/IL-13–linked programs which are traditionally associated with M2 macrophage polarization (Fig. 3C–D). Given the ~ 1 h timescale, we hypothesized that these changes observed in both rejection and non-rejection groups could primarily reflect reperfusion-associated injury and routine perioperative corticosteroid exposure rather than changes in polarization status. Accordingly, we conducted hypothesis-driven over representation analysis (ORA) using curated macrophage program gene sets (glucocorticoid pulse, IL-10/STAT3, PGE₂-mediated immunoregulation, efferocytosis/LXR–PPAR, and scavenger/clearance). The glucocorticoid-response signature was most strongly enriched post- versus pre-reperfusion, with IL-10/STAT3 and PGE₂ modules also prominent, consistent with a rapid, immunoregulatory skew (Fig. 3E; Supp. Table 3). In parallel, we detected JAK/STAT pathway activity and IL-6 signals in CD3 + areas, consistent with early cytokine release (Fig. 3C–D). Similarly, hypothesis-driven ORA comparing pre- vs. post-HDLN in the CD3 + segment showed immediate early gene activation for TCRs, indicating early signs of T cell engagement and activation (Fig. 3F). Together, these findings indicate that CD3 + cells exhibit early signs of heightened inflammatory signaling post-reperfusion relative to pre-reperfusion, while CD68 + cells show a concurrent corticosteroid-linked program that appears to initiate an M2-like phenotype.

Next, to separate ischemia–reperfusion effects from the pre-reperfusion immune landscape, we compared rejection and non-rejection in the pre-reperfusion nodes only. In these pre-HDLN samples, CD3 + segments showed enrichment for IL-4 and IL-13 programs with downstream STAT6 activity. At the same time, CD68 + cells showed increased RHOA signaling, phagosome pathways, and SUMOylation (Fig. 3C–D). These features indicate an “inflamed” lymph-node environment already primed for antigen capture and trafficking, with a bias toward subsequent processing and presentation. Early after reperfusion, macrophages can present processed donor peptides on MHC II to CD4 + T cells, while intact donor MHC–peptide complexes may also transfer via passenger leukocytes, extracellular vesicles, or apoptotic bodies (“cross-dressing”). Together these routes relay alloantigen to professional APCs such as DCs, which initiate alloresponses that can drive downstream cytotoxic T-cell activation and graft injury.

Similarly, we compared rejection and non-rejection in the post-HDLN samples to isolate ischemia–reperfusion–dependent effects on outcome. T cells showed activation of DNA-damage–response pathways (e.g., NER, BER), but the direction and magnitude were inconclusive. By contrast, ISGylation and IL-27 signaling ranked among the next-highest features and together suggest exposure to type I interferon–conditioned signaling. Because segmentation was based on CD3 rather than CD4/CD8 and due to the short timescale between pre and post, we do not ascribe these signals to a Th1 lineage. Meanwhile, CD68 + regions exhibited induction of oxidative-stress programs (NRF2/HMOX1) alongside TGF-β–linked transcripts, a pattern compatible with reperfusion injury (Fig. 3C–D). A targeted ORA of DAMP sensors/responders provided only modest support, which is plausible given the transient, largely protein-level nature of DAMP sensing relative to our sampling window (Fig. 3E). Oxidative injury and DAMP release at reperfusion can unmask latent TGF-β, promoting chemokine production and sustaining antigen handling by myeloid cells.

In this study, we compared paired HDLNs from liver transplant recipients sampled before and after the liver allograft reperfusion, revealing rapid changes in transcriptional/proteomic programs as well as shifts in the spatial organization of immune cells. To our knowledge, this is the first paired pre- and post-reperfusion analysis of hepatic draining lymph nodes human liver transplantation, enabling resolution of immune programs within ~ 1 hour of reperfusion in the compartments where priming occurs. Applying four orthogonal spatial approaches to probe tissue architecture, we observed a consistent pattern in APC–T-cell interactions: rejection was associated with macrophage–T-cell proximity, whereas non-rejection was dominated by DC–T-cell interactions. Notably, within T-cell–rich neighborhoods, both DCs and macrophages exhibited immunoregulatory-leaning features in the non-rejection group. In parallel, GeoMx WTA revealed reperfusion-associated alternative macrophage activation (AMA) and early signs of T cell activation, consistent with both corticosteroid administration and an influx of donor-derived alloantigen. Taken together, rejection was more strongly associated with enhanced macrophage motility and phagocytosis and greater T cell responsiveness, including activation of the IL27 signaling pathway following reperfusion.

Though T cell - macrophage interactions were associated with rejection outcomes, these contacts may be transient and less stimulatory than presentation by classically licensed DCs or B cells. Over the ensuing hours to days, antigen handling is expected to shift toward DCs and B cells, whose presentation and co-stimulatory programs more directly drive acute and longer-term allograft injury. Our results emphasize T cell to macrophage contacts in the very early post-reperfusion window but do not exclude the role for DCs in priming T cell activation. In this context, macrophages are positioned as early mediators of outcome-linked T-cell engagement through antigen capture, support of cross-presentation, and inflammatory cues, rather than as sole drivers of cytotoxic injury. The observed split between rejection and non-rejection may therefore reflect differences in the baseline microenvironment that favor macrophage access to nearby T cells during the first hour, without implying fundamentally different antigen-presentation hierarchies.

B cells were not major drivers in our short-interval sampling. We initially prioritized them given emerging evidence that B cells can acquire intact peptide–MHC (pMHC) from DCs via trogocytosis (“cross-dressing”) and potentially modulate T-cell responses.⁴⁰ However, at this very early time-point we found stronger and more consistent signals from DCs and macrophages. This likely reflects kinetics more than absence of effect: germinal center programs and antibody-dependent mechanisms unfold over longer intervals, and cortical B-cell territories may not be the dominant sites for the first alloantigen contacts in the HDLN.

We also acknowledge that we did not directly visualize the alloantigen itself or the EVs that can carry donor MHC to recipient APCs. Our inference about cross-dressing is therefore indirect, based on timing, spatial patterns, and known biology.^{32, 40} Prior studies have shown donor-derived EVs bearing HLA can rapidly “dress” recipient APCs after liver transplantation.⁴⁴ and that DC- or allograft-derived vesicles can initiate or modulate alloresponses in vivo.¹⁵ We do not exclude the contribution of donor-passenger leukocytes, as both routes are very likely to operate in parallel immediately after reperfusion.

An additional limitation of this study was the relatively small sample size, which may constrain the generalizability of our findings as well as the multiplexed imaging panel being limited to only 35 proteins. As a result, our analysis was incapable of fully capturing the spectrum of physical and chemical cues that contribute to immune cell positioning and function at the single-cell level. Furthermore, we were unable to classify immune cell types at deeper lineages; for example, monocyte-derived DCs may importantly contribute to graft rejection at this timescale, but our more general cell categories failed to disentangle this.³¹ Additionally, our classification failed to assign a substantial number of Tregs, whose contributions in synergy with regulatory DCs may help explain the effects we observed within the non-rejection cohort.

Despite these limitations, this work provides, to our knowledge, the first within-patient, outcome-linked view of the earliest alloimmune events in humans. Within ~ 1 h of liver allograft reperfusion, a coordinated program emerges in the HDLN that biases trajectories toward rejection vs. non-rejection. Using spatial multiomics with orthogonal imaging on primary human tissue, we resolved how APC - T cell relationships reorganize in near “real time”, a level of resolution unattainable from blood or bulk assays. Because HDLNs uniquely concentrate donor antigens and promote cellular cross-talk, yet are seldom profiled, our findings position them as highly informative substrates for validation, biomarker development, and therapeutic hypothesis testing, and motivate thoughtful archiving and prospective study in future cohorts of liver transplantation.