A Human‐Based Skin‐Lymphoreticular Model‐on‐Chip to Emulate Inflammatory Skin Conditions

ABSTRACT

Despite fast‐paced advances in bioengineering human‐based models, emulating fine‐tuned and interdependent immunoregulations of the human body remains challenging. While many bioengineered models contain immune cells, secondary lymphoid tissues are typically missing.

Here, we report a self‐assembled, minimally functional lymphoreticular unit composed of fibroblastic reticular, lymphatic endothelial and CD4⁺ T cells that emulate structural and functional interactions with the skin under inflammatory conditions. Following the optimization of the culture conditions, a layer‐by‐layer approach combining the relevant cell types yielded a self‐organized and compartmentalized lymphoreticular model mirroring human lymphoid tissue. Following verification of T cell motility, we established a skin‐lymphoreticular co‐culture within a microfluidic organ‐on‐chip to mimic interactions between skin and its draining lymph nodes. Co‐cultivation with atopic‐like skin models and the topical application of the skin sensitizer 2,4‐dinitrochlorobenzene stimulated strong migration and T cell infiltration from the lymphoreticular models while maintaining tissue integrity. The addition of anti‐inflammatory drugs abrogated the T cell mediated effects verifying the functionality of the model system. Taken together, the presented human‐based skin‐lymphoreticular model resembles a minimally functional unit that allows to study interactions between the skin and adjacent lymphoid tissues in a human‐based setup.

A human‐based lymphoreticular (LR) model was developed through guided self‐assembly and integrated in a skin‐lymphoreticular co‐culture in a microfluidic organ‐on‐chip to mimic interactions between skin and its draining lymph nodes. The human‐based skin‐lymphoreticular model resembles a minimally functional unit that allows to study interactions between the skin and adjacent lymphoid tissues in a human‐based setup.

1. Introduction

Biomedical research is currently undergoing a paradigm shift towards human‐centered approaches. Major drivers for this transition are the limitations of animal models, which often yield poor predictions of human (patho)physiology due to distinct interspecies‐related differences, i.e., different anatomical layouts and biological barriers as well as divergent receptor expression and immune responses [1]. Hence, the development of increasingly complex, human‐based models ranging from self‐assembled organoids to 3D bioengineered models, organ‐on‐chip (OoC), and human‐on‐chip setups now provide new tools for basic and translational research [1, 2, 3]. Yet, several challenges persist. In fact, recapitulating the fine‐tuned and highly interdependent immunoregulations of the human body in in vitro remains a major hurdle and our current mechanistic understanding of immune cell trafficking and responses still mostly relies on rodents [4].

3D skin models are amongst the most advanced human‐based models; yet, they also still lack robust and physiologically relevant immunoregulation. Although functional integration of T cells and macrophages has been demonstrated [5, 6, 7], key secondary immunocompetent structures, such as skin‐draining lymph nodes, are generally absent. This limitation extends to many other tissue models, prompting global efforts to bioengineer secondary lymphoid tissues through approaches including 3D printing and microfabrication to reproduce their native architecture [8, 9, 10, 11]. Hydrogels are frequently employed as scaffolds that support immune cell proliferation and aggregation [4, 12, 13, 14, 15, 16], while decellularized ECM can alternatively provide authentic structural and biochemical cues essential for appropriate immune function [17, 18]. In addition, microfluidic systems are used to mimic the dynamic microenvironments characteristic of secondary lymphoid organs [11, 19, 20, 21]. Despite these advances, fully recreating the intricate architecture and microenvironment of lymphoid tissues – arising from the highly organized interplay between stromal cells, immune cells, and ECM components – remains a major challenge [14, 22]. Self‐assembled structures may offer a promising alternative, enabling physiologically relevant modeling without the need for synthetic scaffolds [23, 24].

New approaches are also necessary to recapitulate complex immunological processes such as immune cell recruitment from secondary lymphoid tissues in vitro. This would significantly expand the toolbox to study disease‐related mechanisms and conditions that involve T cell recruitment such as skin sensitization. As of today, only T cell migration into diseased skin can be mimicked using skin (disease) models in OoC setups [7, 25]. Yet, during inflammation, skin resident antigen‐presenting cells such as dendritic cells (DCs) sample antigens and present them as peptide:MHC complexes to T cells in the lymph nodes triggering their activation and expansion. T cells then emigrate from the lymph nodes to the tissue to promote inflammation and other immune‐related effects [26, 27]. This is also a key interaction in chronic skin diseases such as atopic dermatitis (AD). AD is caused by a complex interplay of genetic predisposition, disruption of the epidermal barrier, and dysregulated immune responses. The impaired skin barrier triggers inflammation and the recruitment of Th2 cells, a T cell subtype that is central for pathogen defense but also for inflammatory responses [28]. This interplay is further aggravated by skin‐derived cytokines such as thymic stromal lymphopoietin (TSLP) [29, 30, 31] which activates DCs, facilitates T cell priming [32, 33], and skews Th2 cells towards pathogenic phenotypes [34, 35]. Historically, AD has mainly been studied in mice, yet most mouse strains do not spontaneously develop AD [36, 37]. Hence, a range of human‐based tissue models emulating atopic conditions have been designed to tackle this shortcoming [7, 38, 39, 40].

Striving to close current translational gaps, here, we report on a self‐assembled human lymphoreticular (LR) tissue model that enables direct investigation of T cell recruitment to inflamed skin. Guided self‐assembly of primary fibroblastic reticular cells, lymphatic endothelial cells, and CD4⁺ T cells yielded a 3D LR construct exhibiting tissue‐typical compartmentalization and T cell clustering. Functional assays confirmed responsiveness to sphingosine‐1‐phosphate, a regulator of lymphocyte trafficking, and TSLP, a key cytokine in atopic inflammation. Integration of the LR model with an atopic‐like skin disease model in a microfluidic two‐organ chip recreated the crosstalk between skin, lymphoid tissue, and circulating T cells. T cell emigration occurred only after exposure to the skin sensitizer 2,4‐dinitrochlorobenzene or during co‐culture with atopic‐like skin, and was abolished by a small‐molecule TSLP inhibitor, but not by tacrolimus. These results demonstrate a high level of functional biomimicry and establish a human‐based platform for dissecting immune mechanisms in inflammatory skin disease.

2. Results and Discussion

2.1. Fibroblastic Reticular Cell‐Derived ECM Provides a Physiological Environment for T Cells Ex Vivo and Promotes T Cell Clustering

To bioengineer secondary lymphoid tissue models, we opted for the integration of critical lymphoreticular cell types such as fibroblastic reticular (FRCs) and lymphatic endothelial (LECs) cells. FRCs produce and regulate the ECM of a lymphoreticular tissue which is essential for a proper immune response [41]. LECs form the structural framework of lymphatic vessels and lymph node sinuses. In response to inflammatory stimuli, LECs also produce chemokines that attract immune cells to the lymph nodes [42].

In a first step, we tested different culture media compositions by combining complete RPMI and fibroblast growth medium (FGM), with or without IL‐2, to support the co‐culture of CD4⁺ T cells and FRCs. High cell viability (> 90%) was maintained when mixing complete RPMI/FGM at a 1:1 ratio (Figure S1A). Next, we assessed which scaffold or matrix provides a suitable microenvironment for CD4⁺ T cells to mimic secondary lymphoid tissue. Hence, the CD4⁺ T cell viability was determined following cultivation in the commercially available matrices Matrigel, BME, and BME Type 2 over 10 days. Similar to Matrigel, BME and BME Type 2 are ECMs derived from Engelbreth‐Holm‐Swarm tumors, with BME containing fewer growth factors and BME Type 2 exhibiting increased tensile strength [43]. Interestingly, CD4⁺ T cell viability decreased drastically to <5%, respectively, in Matrigel, BME, and BME Type 2 rendering these matrices non‐suitable for our application. In contrast, cell viability remained at >50% when cultured in complete RPMI medium (Figure S1B). While the underlying reason remains ambiguous, it may be due to the fact that matrices like Matrigel facilitate cell adhesion and migration while inhibiting cell proliferation [44, 45].

We therefore next opted for a guided self‐assembly approach by stimulating the FRCs to produce their own ECM. First, we assessed their capacity for ECM deposition using different cell numbers, ratios, and supplements comparing FRCs only, FRCs + IL‐4, and FRCs + naïve or activated CD4⁺ T cells. IL‐4 promotes T cell priming into Th2 cells and was therefore added as a reference [46]. We consistently observed a more uniform, spherical ECM formation and cell clustering with higher FRC numbers (Figure 1A). Interestingly, when co‐cultivating FRCs with CD4⁺ T cells at a 1:3 ratio, spherical, highly self‐organized structures formed exhibiting compartmentalization similar to actual lymphoid tissue (Figure 1A).

FIGURE 1.

Guided self‐assembly of extracellular matrix (ECM) as a scaffold for secondary lymphoid tissue models. A) Self‐assembled ECM generated by 5 × 10⁴ FRCs, 1 × 10⁵ FRCs, 5 × 10⁴ FRCs + IL‐4, 1 × 10⁵ FRCs + IL‐4, 5 × 10⁴ FRCs + 1.5 × 10⁵ naïve CD4⁺ T cells, 1 × 10⁵ FRCs + 3 × 10⁵ naïve CD4⁺ T cells, 5 × 10⁴ FRCs + 1.5 × 10⁵ activated CD4⁺ T cells, and 1 × 10⁵ FRCs + 3 × 10⁵ activated CD4⁺ T cells. IL‐4 was added at 10 ng/mL. Scale bars = 100 µm. B) Glucose levels and lactate dehydrogenase (LDH) release from the self‐assembled ECMs over 14 days; n = 3. Two‐way ANOVA followed with Tukey post‐hoc tests were performed for the glucose data.

Glucose consumption remained stable over 14 days with no statistically significant differences between the culture conditions (Figure 1B). Lactate dehydrogenase (LDH) and TUNEL assays were performed to monitor cell viability and apoptosis over time. LDH levels remained <20% across most samples, indicating good cell viability [52] and rendering most of them suitable to proceed (Figure 1B). Similarly, very few apoptotic cells were detected via the TUNEL assay (Figure S2), while a uniform distribution of T cells and T cell clustering with 1:3 FRCs:CD4⁺ T cells was observed (Figure S3A).

To verify the motility and functionality of the CD4⁺ T cells within the self‐assembled matrices, sphingosine‐1‐phosphate (S1P), a lipid mediator known to stimulate T cell migration, was added to the culture media. While we also observed unstimulated T cell emigration, S1P addition significantly increased this effect mainly from the self‐assembled matrices (^*** p ≤ 0.001, Figure S3B). Significantly fewer T cells left the Matrigel‐based scaffolds (^**** p ≤ 0.0001, Figure S3B) compared to the self‐assembled matrices. The latter is likely due to the density of Matrigel, which may restrict cell motility. In general, emigration was more pronounced for naïve than activated T cells. Naïve cells tend to be more mobile than activated T cells which migrate in a directed manner upon stimulation [47, 48, 49, 50]. It should be further noted that we chose the S1P concentrations based on animal studies [51], which may explain the lack of a classic dose‐response curve. Nevertheless, the primary objective of demonstrating T cell emigration upon stimulation was achieved.

Taken together, we decided to proceed with the higher FRC number (1 × 10⁵ cells) and the 1:3 FRC:T cell ratio (1 × 10⁵ FRCs + 3 × 10⁵ activated CD4⁺ T cells) as the basis for the LR model due to superior self‐organization and stable culture conditions.

2.2. 3D Human‐Based Lymphoreticular (LR) Models Closely Resemble the Structure of Human Lymphoreticular Tissue

To further enhance the biomimicry of the self‐assembled model, we next added primary human LECs. We tested a variety of layering approaches striving to resemble human lymphoid tissue structure (Figure 2A). First, we seeded FRCs only to allow them to produce lymphoid tissue‐specific ECM as a scaffold. After 7 days, we added a layer of LECs as another critical scaffold‐producing cell type. After additional 7 days, we added the T cells either alone or mixed with FRCs for additional 7 days to avoid mere T cell accumulate on top of the previously secreted ECM (Figure 2B). The five differential experimental groups therefore were:

• ‐FRC + LEC only → FLF

• ‐FRC + LEC + Naïve T → FLN

• ‐FRC + LEC + Activated T → FLA

• ‐Complex mixtures (additional FRC + activated or naïve T cell layer) → FLFN, FLFA

FIGURE 2.

Integration of lymphatic endothelial cells (LEC) increases the biomimicry of a self‐assembled lymphoreticular (LR) model. A) Exemplary histological staining of a human lymph node. Scale bar = 1 mm. Copyright permission has been obtained from Xinxiang Happy Science Co., Ltd. B) Schematic illustration of the layering approach. C) Brightfield image of a self‐assembled LR model. D) Representative H&E staining and E) IF images showing the FRC‐derived ECM marker ER‐TR7 (red), DAPI (blue) and the localization of GFP‐tagged LECs of LR models generated using different layering approaches: FLF (FRCs + GPF‐LECs + FRC), FLFN (FRC + GPF‐LECs + (FRC + naïve CD4⁺ T cells)), FLFA (FRC + GPF‐LECs + (FRC + activated CD4⁺ T cells)), FLN (FRC + GPF‐LECs + naïve CD4⁺ T cells), and FLA (FRC + GPF‐LECs + activated CD4⁺ T cells). Circles and arrows indicate the compartmentalization within LR models. Semi‐quantification of F) ER‐TR7 expression and G) GFP‐LEC distribution in the LR models generated using the different layering approaches. n = 3. One‐way ANOVA followed with Tukey post‐hoc tests were performed. Mean ± SD. ^* p≤ 0.05, ^** p ≤ 0.01, ^*** p ≤ 0.001, ****p ≤ 0.0001. H) IF staining against collagen type 1 and fibronectin as key ECM components of lymphoid tissue in FLFA models. Cell nuclei are counterstained with DAPI (blue). Scale bars = 100 µm.

As confirmed by histological analysis, we observed a distinct, lymphoreticular tissue‐like compartmentalization (Figure 2D). This was most pronounced for both complex mixtures, FLFN and FLFA (Figure 2D), and also yielded the highest ER‐TR7 expression (Figure 2F). The latter is indicative of increased ECM deposition as ER‐TR7 is a classic FRC‐derived lymphoreticular ECM component. It is closely associated with reticular fibers composed of collagen and other ECM proteins [53].

LEC distribution was higher in FLFA and FLFN (Figure 2G) compars to FLF, FLN and FLA. Abundant ECM deposition was further confirmed via fibronectin and collagen 1 IF staining (Figure 2H; Figure S4). We also noted a significantly higher CD4⁺ signal for both the FLFA and FLFN approaches indicative of excellent T cell clustering (Figure 3). Overall, these results demonstrate that incorporating a second FRC layer together with the T cells yields superior results. A TUNEL assay showed very few apoptotic cells across all experimental groups (Figure S5) which demonstrates close resemblance to native human lymph node tissue (Figure 2A), an average size of 0.8 mm (Figure 2C), and a thickness of ∼500 µm (Figure S6; Videos S1 and S2).

FIGURE 3.

CD4⁺ T cell clustering in LR models generated using different layering approaches. A) Representative IF images showing CD4+ expression and GFP‐LEC in self‐assembled LR models generated using different layering approaches. FLFN (FRC + GPF‐LECs + (FRC + naïve CD4⁺ T cells)), FLFA (FRC + GPF‐LECs + (FRC + activated CD4⁺ T cells)), FLN (FRC + GPF‐LECs + naïve CD4⁺ T cells), and FLA (FRC + GPF‐LECs + activated CD4⁺ T cells). Blue = DAPI. Scale bar = 100 µm. B) Semi‐quantification of CD4+ signal normalized to the DAPI count representative for T cell clustering in the different LR models as determined using ImageJ; n = 3. One‐way ANOVA followed with Tukey post–hoc tests were performed. Mean ± SD. ^** p ≤ 0.01, ^**** p ≤ 0.0001.

2.3. Atopic‐Like Skin Conditions Trigger CD4⁺ T Cells Migration from the LR Models

We next sought to evaluate the biomimicry and functionality of our LR models with a focus on T cell‐mediated processes in inflammatory skin diseases. Here, we employed a microfluidic two‐organ chip platform (Figure 4A,B) to replicate the dynamic inter‐tissue crosstalk. OoC platforms are perfused microfluidic systems that integrate bioengineered tissue models via 3D microchannels, recapitulating blood circulation, biomechanics, and (patho)physiological processes such as immune cell infiltration in a human‐relevant context [1]. Immune cells introduced into the system are exposed to shear stress and flow conditions similar to those in vivo. OoCs can accommodate single or multiple tissue models, with the latter allowing the study of inter‐tissue communication. We selected this OoC configuration because its recirculating media flows beneath transwell inserts, facilitating the drainage of signals from the atopic‐like skin model to the LR tissue and enabling T cell trafficking from the LR model back to the skin.

FIGURE 4.

Microfluidic organ‐on‐chip (OoC) setup enabling co‐cultivation of skin and LR models. A) Design of a HUMIMIC Chip3plus and B, C) schematic depiction of the OoC setups. D, E) Total number of CD4⁺ T cells collected from circulating culture media in the OoC channels following LR exposure to TSLP (D) and atopic‐like skin models (E) after 24 and 48 h, respectively. n = 3; Mean ± SD. ^* p ≤ 0.05, ^**** p ≤ 0.0001, ns = not significant. Two‐way ANOVA followed with Tukey post‐hoc tests were performed. F) Representative H&E staining of normal (FLG+), atopic‐like skin models (FLG‐), and LR models following co‐culture over 48 h in the OoC platform. Circles and arrows highlight LR model compartmentalization. Scale bars = 50 µm. SC = stratum corneum.

First, we examined the effect of thymic stromal lymphopoietin (TSLP) on CD4⁺ in vivo T cell migration from the LR models. TSLP is a key pro‐inflammatory mediator that orchestrates immune responses such as T cell migration in atopic skin [25, 54]. Following the addition of TSLP to the circuit of the organ chip, we observed a general increase in CD4⁺ T cell migration from the LR models over 48 h (Figure 4D). Interestingly, TSLP triggered significantly more naïve T cell migration from the LR model (FLFN approach), whereas it did not further add to the migration of activated T cells (FLFA; Figure 4D).

Next, we tested if T cell migration is also initiated when co‐cultured with an atopic‐like skin disease model mimicking the interactions between skin and its draining lymph nodes in vivo. Here, we used a previously established and well‐characterized atopic‐like skin model that closely emulates key features of atopic skin such as an impaired skin barrier function and differentiation and increased TSLP expression [7, 55]. The impaired skin barrier function is induced by the knock‐down of the critical skin barrier protein filaggrin (FLG). In fact, loss‐of‐function mutations in the FLG gene are a major predisposing factor for the manifestation of AD [56].

The skin (disease) models were integrated with LR models in the OoC setup for 48 h (Figure 4C). Similar to TSLP stimulation, a general increase in T cell migration from the LR models was observed, emulating established interactions between human skin and LR tissue in an ex vivo setup. Importantly, significantly higher CD4⁺ T cells emigration was observed for both FLFN and FLFA co‐cultivated with the atopic‐like skin models already after 24 h. This effect was again most pronounced for naïve T cells (Figure 4E). In vivo, naïve T cells migrate primarily through lymphoid tissues. Effector T cells, however, show significantly different migration patterns, favoring movement through non‐lymphoid tissues [57, 58, 59] which may explain the increased egress of the naïve T cells from the LR models. Histological assessments showed no major detrimental effects of the OoC culture conditions in both tissue models (Figure 4F,G) except for the expected histological changes in the atopic‐like skin model due to the disease phenotype.

2.4. The Skin‐LR Model‐on‐Chip Emulates T Cell‐Mediated Effects in Response to the Skin Sensitizer 2,4‐Dinitrochlorobenzene and Anti‐Inflammatory Drugs

Skin sensitizers such as nickel pose a major problem for the general population and there is a general lack of models that reliably identify skin sensitizers. While none of the currently available in vitro models capture all these events [60], immunocompetent skin models could help to close the gap. Hence, we next assessed if the LR model‐on‐chip can replicate certain events related to skin sensitization and anti‐inflammatory drugs in human skin in vitro. We applied the strong skin sensitizer 2,4‐dinitrochlorobenzene (DNCB), a small molecule TSLP inhibitor BP79 [7], and the clinically approved anti‐inflammatory drug tacrolimus topically onto normal and atopic‐like skin‐on‐chip. We hypothesized that the skin models would strongly react to the skin sensitizer DNCB (e.g. high TSLP expression and T cell infiltration) and that TSLP‐mediated effects such as enhanced T cell migration can be abolished by BP79. Due to the different mode‐of‐action of the calcineurin inhibitor tacrolimus, which occurs TSLP independent, we did not expect major anti‐inflammatory effects in these particular experimental settings.

In line with our hypothesis, TSLP expression was massively upregulated following the application of DNCB and in untreated atopic‐like skin models. Also, topical application of BP79 significantly downregulated TSLP expression in the treated samples, while a topical administration of tacrolimus did not alleviate TSLP expression in line with their respective mode of action and previous data (Figure 5A,B) [7].

FIGURE 5.

Skin compartment: Functional validation of the human skin‐LR model‐on‐chip. A) Representative immunohistology staining against TSLP and CD4⁺ in healthy control skin models, DNCB‐treated skin models, untreated atopic‐like skin models, and BP79 and tacrolimus‐treated atopic‐like models. Red circles and arrow highlight positive CD4⁺ signals indicative of T cells migration. Semi‐quantification of the fluorescence signals of B) TSLP expression and C) CD4 expression in the skin models was performed using ImageJ. One‐way ANOVA followed with Tukey post‐hoc tests were performed. n = 3 D) Total number of viable CD4⁺ T cells collected from the culture medium circulating in the OoC platform after 24, 48, and 72h. Two‐way ANOVA with repeated measures and multiple comparisons between the conditions followed with Tukey post‐hoc tests were performed. Data are shown as mean ± SD. ^** p ≤ 0.01, ^*** p ≤ 0.001, ^**** p ≤ 0.0001. n = 3.

Significant T cell infiltration into the skin models was observed in both DNCB‐treated and atopic‐like untreated skin models which was reduced following BP79 treatment. Notably, DNCB exposure yielded the strongest effects (Figure 5A,C). Histologically no major detrimental effects were observed except for anticipated treatment‐ or condition‐related changes (Figure S7). Similarly, the expression of skin differentiation markers such as filaggrin occurred as expected and dependent on the condition: Normal skin model: high expression, atopic‐like skin model: reduced expression, atopic‐like skin BP79 treated: increased expression, tacrolimus, and DNCB treated skin: largely unchanged filaggrin expression (Figure S7).

In line with the increased T cell infiltration observed into the tissues, we detected significantly higher numbers of T cells circulating in the OoC platform after DNCB administration and in untreated atopic‐like skin models, which was reduced following topical treatment with the anti‐TSLP compound BP79. Interestingly, tacrolimus did not abrogate T cell migration (Figure 5D) which is in line with its mode of action. In fact, the calcineurin inhibitor tacrolimus inhibits IL‐2 mediated T cell proliferation and expansion, not T cell migration. Taken together, these data further demonstrate the biomimicry of the skin‐LR model‐on‐chip.

Notably, while clear treatment‐ and condition‐dependent effects were observed, increased T cell emigration was noted over time also for the normal skin model indicating that the models may not be suited for longer‐term applications although further studies are warranted. Glucose consumption remained stable throughout the experiment, whereas increased LDH release was observed following topical application of DNCB and in the atopic‐like conditions irrespective of the treatment (Figure S8).

The LR model remained structurally intact and no major changes in the expression pattern of ER‐TR7 or GFP‐LECs were observed after DNCB treatment. Interestingly, however, ER‐TR7 and GFP‐LECs expression increased following co‐cultivation with the atopic‐like skin model (Figure 6A,B). The implications and underlying mechanism of this, however, require further investigation. CD4⁺ signal was distinctly reduced in all LR models treated with DNCB or co‐cultivated with disease conditions in line with the increased T cell numbers in the OoC platform and the tissue models (Figure 6A,D).

FIGURE 6.

LR compartment: Functional validation of the human skin‐LR model‐on‐chip. A) Representative H&E and immunohistology staining against ER‐TR7, GFP‐LEC and CD4 in healthy control skin models, DNCB‐treated skin models, untreated atopic‐like skin models, and BP79 and tacrolimus‐treated atopic‐like models. Semi‐quantification of the fluorescence signals of B) ER‐TR7 expression C) GFP‐LEC expression and D) CD4⁺ expression in the LR model using ImageJ. One‐way ANOVA followed with Tukey post‐hoc tests were performed. n = 3; Data are shown as mean ± SD. ^* p ≤ 0.05, ^*** p ≤ 0.001, ^**** p ≤ 0.0001. ns = not significant. Scale bars  =  100 µm.

Taken together, we were able to recapitulate compound‐specific effects in our skin‐LR setup, with a focus on T cell migration from the LR to the skin model, which was most pronounced following the topical application of DNCB. This verifies the applicability of the skin‐LR model‐on‐chip to monitor T cell migration and indicates its potential usefulness in identifying skin sensitizers. This remains a major concern. Although simple immune activation assays can detect certain skin sensitizers, no in vitro method currently exists that reliably identifies sensitizers or classifies them as strong, moderate, or weak. A potential advantage of the skin‐LR model‐on‐chip is that it allows monitoring multiple adverse outcome pathways of skin sensitization, specifically key events 1 (covalent interactions with skin proteins), 2 (keratinocyte activation), and 3 (T cell migration). It should be noted, that we did not investigate if the integration of the LR model has any secondary effect on the skin model. This as well as head‐to‐head comparisons with other tissue models might provide important insight, however was beyond the scope of the present study.

3. Conclusion and Limitations

In conclusion, we present a self‐assembled human‐based skin‐LR model‐on‐chip that resembles a minimally functional unit. As such, it allows to study directed T cell migration between human skin and adjacent secondary lymphoid tissues ex vivo. A major advantage of our approach is its ease of production which neither requires any specialized equipment nor engineering facilities. This sets our approach apart from many other bioengineering methods of lymphoid tissue models. For example, a recent publication reported on highly sophisticated in‐house synthesized hydrogels to support lymphoid organoid formation. The organoids were then integrated in a custom‐made microfluidic setup to ensure sustained culture and compartmentalization of germinal centers [16]. Another study engineered ectopic lymphoid follicles (LF) by introducing high cell numbers of human B and T lymphocytes in an ECM matrix on a 2‐channel microfluidic organ chip that was continuously perfused for 4 days [21]. This microfluidic form of superfusion culture both supported ectopic LF formation and prevented autoactivation of high‐density cultures of human B cells. Finally, bioreactor‐based systems combining PBMCs, monocyte‐derived dendritic cells, and mesenchymal stromal cells in perfused 3D hydrogel matrices enable long‐term in vitro cultivation [14]. While existing approaches have shown promise, their complexity limits reproducibility across laboratories. Our method offers a simplified protocol suitable for higher‐throughput studies. Although advanced organ‐on‐chip platforms are needed to study multi‐tissue interactions, our focus was on recapitulating LR extracellular matrix and T cell migration, rather than B cell–mediated responses. The current model lacks B lymphocytes, germinal centers, and dendritic cell–driven T cell activation – features that will be critical for modeling complex immune regulation, including DC extravasation, migration, and T cell priming. Future work aims to incorporate B cells and dendritic cell–mediated activation, further enhancing the model's translational relevance.

4. Materials and Methods

4.1. Cell Culture

Primary human fibroblastic reticular cells (FRCs) (Cat. #2530, male donor) and the poly‐L‐lysine stock solution (Cat. #0413) were purchased from ScienCell. A poly‐L‐lysine‐coated culture vessel (2 µg/cm², T‐75 flask) was prepared before cell seeding. Green fluorescent protein‐expressing dermal primary human lymphatic endothelial cells (GFP – LECs) (cAP‐0003GFP, male donor) and the Quick Coating Solution (cAP‐01) were purchased from Angio‐Proteomie. Primary human dermal fibroblasts (FBs) and keratinocytes (KCs) were isolated from juvenile foreskin according to standard procedures (CREB approval #H19‐03096). Written informed consent was obtained from all subjects in accordance with the Declaration of Helsinki and the Department of Health and Human Services Belmont Report. Peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats of single donors that were purchased from the Canadian Blood Services (approval obtained from CBS REB# 2024.023 and UBC CREB H19‐00446) using Lymphoprep density gradient centrifugation (STEMCELL, Catalog #07851). CD4⁺ T cells were then separated from PBMCs using the EasySep Human CD4⁺ T Cell Isolation Kit (STEMCELL, Catalog #17912), following the manufacturer's instructions. CD4⁺ T cells were cultured for 48 h, followed by a medium change. To activate the CD4⁺ T cells, ImmunoCult Human CD3/CD28 T Cell Activator (25 µl per million cells; STEMCELL, Catalog #10991) was added in accordance with the manufacturer's instructions.

FRCs and FBs were cultured in fibroblast growth medium (FGM) (DMEM – high glucose (Sigma, D6429) supplemented with 10% (v/v) of fetal bovine serum (FBS) (Sigma, F1051) and 1% (v/v) of Penicillin‐Streptomycin (Sigma, P4333). GFP‐LECs were cultured in EBM‐2 Endothelial Cell Growth Basal Medium (Lonza, Cat #: CC‐3156) supplemented with the EGM‐2 Endothelial SingleQuots Kit (Lonza, Cat #: CC‐4176). KCs were cultured with EpiLife medium (Gibco, MEPI500CA) supplemented with human keratinocyte growth supplement (Gibco, s‐001‐5). CD4⁺ T cells were cultured in complete RPMI medium (RPMI‐1640 Medium (Sigma, R8758) supplemented with 10% (v/v) of FBS (Sigma, F1051) and 1% (v/v) of Penicillin‐Streptomycin (Sigma, P4333).

4.2. Reagents and Antibodies

Human recombinant TSLP (Cat# SRP4896), 1‐Chloro‐2,4‐dinitrobenzene (DNCB) (Cat# 237329), and sphingosine 1‐phosphate (S1P) (SML2709) were purchased from Sigma–Aldrich. Tacrolimus monohydrate (Cat# B3579) was purchased from APExBIO. Propidium iodide (P1304mp) was purchased from ThermoFisher. Matrigel matrix (Cat# 354248) was purchased from Corning. Both Cultrex basement membrane extracts (BME) (Cat# 3432001‐01) and Cultrex BME type 2 (Cat# 3532‐001‐02) were purchased from R&D systems. Recombinant human IL‐2 protein (cAT# 589108) was purchased from BioLegend and recombinant human IL‐4 protein (Z100545) was purchased from Applied Biological Materials Inc. ImmunoCult Human CD3/CD28 T Cell Activator (Cat# 10991) was purchased from STEMCELL.

4.3. Viability of Human CD4⁺ T Cells in Different ECM Scaffolds

3 × 10⁵ activated CD4⁺ T cells were resuspended in 175 µl of one of the three types of ECMs (Matrigel Matrix, Cultrex basement membrane extracts (BME) and Cultrex BME type 2), then added onto the 6.5 mm transwell inserts (Costar, Cat# 3422) and placed in a 24‐well plate (Corning, Cat# 353504), respectively. CD4⁺ T cells seeded in a 24‐well plate with 375 µl of complete RPMI served as control. IL‐2 was added at 100 ng/mL. Cell viability was tested using flow cytometry at days 1 and 3. Cells in Matrigel, BME, and BME type 2 were recovered using the cell recovery solution (Corning, CLS 354253) following the manufacturer's instructions. At days 3, 7, and 10, respectively, CD4⁺ T cells were stained with propidium iodide, resuspended, and subjected to flow cytometry. A total of 1×10⁴ events were counted and examined using CytoFLEX LX research flow cytometer (Beckman Coulter, IN, USA). Debris and dead cells were excluded by population gating within forward scatter‐by‐side scatter plots. FlowJo (BD Ashland, OR, USA) was used for further analyses.

4.4. Generation of the 3D (Atopic‐Like) Human‐Based Skin Models

An atopic‐like skin disease model was prepared according to previously established procedures [7, 55, 61]. In brief, the filaggrin gene (FLG) was knocked down in KCs using a siRNA (Invitrogen, Cat# 87663809) with HiPerFect transfection reagent (Qiagen, Cat# 301705) 24 h before the model generation. For constructing the 24‐well skin models, type I collagen (3 ug/mL; Advanced BioMatrix, Cat# 5005) was used. For each model, 162 µL of collagen was combined with 21 µL of HBSS, titrated to a neutral pH, and subsequently mixed with 21 µL of the dermal fibroblast/FBS suspension (2.1 × 10⁴ cells/well). Subsequently, 177 µL of this mixture was dispensed into a cell culture insert (6.5 mm transwells, 8 µm pores, Costar, Cat# 3422; growth area of 0.33 cm²) and allowed to solidify. After 2 h, 200 µL of EpiLife culture medium was added, and the system was transferred to a 37°C incubator with 5% (v/v) CO₂ and 95% (v/v) humidity for another 2 h. Subsequently, 3 × 10⁵ cells/cm² normal or FLG knockdown KCs were added on top of the collagen matrix. After 24 h, the model was lifted to the air‐liquid interface and the medium was changed to 460 µL of keratinocyte differentiation medium. The skin models were cultured for 11 days (at 37°C, with 5% CO₂ and 95% humidity) with medium change every other day.

4.5. Generation and Characterization of Self‐Assembled Human‐Based Lymphoreticular (LR) Model

1 × 10⁵ FRCs were seeded into a 6.5 mm transwell inserts using only FGM (6.5 mm transwell inserts, 8 µm pores, Costar, Cat# 3422), for 7 days while media was changed every other day. On day 7, 8.1 × 10⁴ LECs were seeded on top of the FRCs followed by 7 days of cultivation using a 1:1 FGM:EBM2 culture medium mix. At day 14, additional 1 × 10⁵ FRCs and 3 × 10⁵ viable CD4⁺ T cells were seeded on top, and the models were further cultivated until day 21 with media changes every other day. At this point, the medium was switched to 1:1:1 FGM:EBM2:complete‐RPMI.

4.6. LDH and Glucose Assay

Cell culture media was collected at day 7, day 14 and day 21 and LDH concentration was determined using the cytotoxicity detection kit PLUS (Roche, Cat# 04744934001) following the manufacturer's instructions. In brief, the LDH levels were normalized to the low control (LDH release from FRCs in 2D) and the high control (LDH release from a fully disrupted model) using the following calculation: Cytotoxicity (%) = (exp. Value – low control)/(high control – low control) x 100, whereas 100% cytotoxicity refers to a completely disrupted model. LDH levels below 30% indicate good cell viability. LDH levels were measured using the MQX200 UV/Vis/IR microplate reader from BioTek Instruments (USA). Glucose consumption was measured using the glucose LiquiColor (Stanbio, Cat# 1070–125) following the manufacturer's instructions. Glucose measurement was performed with the Cary 50 Bio UV–vis spectrophotometer from Varian Inc. (USA).

4.7. TUNEL Assay

Apoptosis within the tissue model was determined using the terminal deoxynucleotidyl transferase‐mediated dUTP nick end labeling method. TUNEL assay kit – BrdU‐Red (Abcam, ab66110) was used following the manufacturer's instructions with the sections then being imaged using an EVOS M5000 fluorescence microscope.

4.8. H&E and Immunofluorescence (IF) Staining and Whole Tissue Imaging

6 µm tissue sections were obtained using a cryotome or microtome for further histological analyses. H&E staining was performed according to standard procedures. For IF staining, tissue sections were fixed with 4% formaldehyde for 10 min at room temperature. The tissues were then permeabilized with 0.5% Triton X‐100 in PBS, washed with PBS containing 0.0025% BSA and 0.025% Tween 20, and blocked with normal goat serum (1:20 in PBS). The sections were incubated overnight at 4°C with primary antibodies: human ER‐TR7 antibody (1:100; Invitrogen, MA1‐40076), human anti‐fibronectin antibody (1:100; Abcam, ab2413), human anti‐collagen I alpha 1 antibody (1:200; Novus Biologicals, BN600‐450) or human anti‐CD4 antibody (1:200; Sigma–Aldrich, Cat# MABF419) (in PBS, 0.0025% BSA, 0.025% Tween 20). Subsequently, the sections were incubated for 1h at room temperature with secondary antibodies: donkey anti‐mouse Alexa Fluor 594 (Invitrogen, A32744), Goat anti‐rabbit IgG (H+L) Cross‐Absorbed Secondary Antibody, Alexa Fluor 488 (Invitrogen, A11008) or Goat anti‐rat IgG (H+L) Cross‐Absorbed Secondary Antibody, Alexa Fluor 594 (Invitrogen, A11007) (1:400 in PBS, 0.0025% BSA, 0.025% Tween 20) and counterstained with DAPI mounting medium. Finally, the slides were imaged using an EVOS M5000 fluorescence microscope. Fluorescent images were obtained with a 20x objective, and fluorescent signals from DAPI, GFP, and Alexa 594 were detected using 405 nm, 488 nm, and 561nm excitation lasers, respectively, with an exposure time of 100 ms. Emission wavelengths were 417–477 nm for DAPI, 500–550 nm for GFP, and 581–654 nm for Alexa 594. The mean intensities of IF images were quantified using FIJI ImageJ by counting the pixels representing fluorescent signals and normalization to DAPI counts. The semi‐quantification was performed per slide and then averaged.

Z‐stack images were captured using the Olympus IXplore SpinSR system (Evident, Tokyo, Japan), featuring an inverted microscope (IX83; Evident, Japan), a CSUW1‐Sora spinning disk confocal unit (Yokogawa, Japan), and a Hamamatsu ORCA Fusion BT camera. Imaging was conducted with a 10 × objective (UPLXAPO 10X/0.4NA) with an exposure time of 300 ms, and fluorescence confocal images were acquired using the cellSens software (Evident, Tokyo, Japan). Image processing and 3D reconstruction were performed using Imaris 10.2 (Oxford Instruments, UK).

4.9. Skin‐LR Model Co‐Cultivation in an Organ‐on‐Chip Setup

HUMIMIC Chip3plus were purchased from TissUse (Berlin, Germany). A single 2 mm‐high polydimethylsiloxane (PDMS) layer, incorporating channels, micropumps, and openings for culture compartments, was permanently bonded to a glass microscope slide (75 × 25 mm, Menzel, Braunschweig, Germany) using low‐pressure plasma oxidation (Femto; Diener, Ebhausen, Germany). This process formed a fluid‐tight microfluidic device with standard channel heights of 100 µm and width of 500 µm. The on‐chip peristaltic micropump, adapted from Wu et al., [62] consists of three 500 µm‐thick PDMS membranes arranged in a row, which are sequentially actuated by applying pressure. This design demonstrates a microfluidic volume of 11 µl and a microfluidic surface area of 255 mm² (Figure 6A) [63]. Prior to chip connection, the HUMIMIC starter control unit (TissUse, Berlin, Germany) was configured to 30 bpm (flow rate 5.32 ± 1.46 µL/min) with +500 mbar pressure‐out and −500 mbar vacuum‐out and the chips were prepared according to the manufacturer's standard procedures and these parameters were used during all chip experiments. The fluid flow in the OoC platform was recirculating below the cell culture inserts. The LR models were then transferred into the chip and 10 µL of a 100 ng/mL TSLP was added into the culture media. For the skin‐LR OoC setup, both tissue models (LR models and atopic‐like/normal skin models) were transferred into the Chip3plus and perfusion was initiated. The day of model transfer is considered day 1 of the OoC cultivation. Number of viable CD4⁺ T cells in the chip were counted using an Olympus model R1 cell counter at 24, 48 and 72 h.

To verify the functionality of the skin‐LR model‐on‐chip, 10 µL of a 5 µM DNCB stock, 10 µL of the small molecule TSLP inhibitor BP79 (20 µm stock) or 10 µL of a 19 µg/mL tacrolimus stock were topically applied onto the skin model on day 2. After 24 h [7], media change was performed, and DNCB, BP79 or tacrolimus were re‐applied to the skin models. 24 h later, final DNCB, BP79 and tacrolimus application was performed and both skin and LR models were eventually harvested 4 h later for further characterization. The number of emigrated CD4⁺ T cells in the chip media was counted using the Olympus model R1 cell counter.

4.10. Statistical Analysis

Statistical analyses included one‐way and two‐way ANOVA followed by Tukey's post hoc test for comparison of conditions with one variable, one‐sample Wilcoxon test for estimating overall viability of different models, and two‐way ANOVA with repeated measures and multiple comparisons for experiments with two variables. The respective text performed is stated in the figure legends. Outliers were identified using the ROUT (Q = 1) method in GraphPad Prism (version 9.0, GraphPad Software). Data are presented as mean ± standard deviation (SD) from at least three independent experimental replicates. Statistical significance was defined as p < 0.05. All analyses and data visualizations were performed using GraphPad Prism (version 9.0).

Conflicts of Interest

The authors declare no conflict of interest.

Supporting information

Supporting File 1: adhm70682‐sup‐0001‐SuppMat.docx.

Supporting File 2: adhm70682‐sup‐0002‐VideoS1.mov.

Supporting File 3: adhm70682‐sup‐0003‐VideoS2.mov.

Supporting File 4: adhm70682‐sup‐0004‐FigureS1–S8.zip.