Inflammation-induced Tissue Damage Mimicking GvHD in Human Skin Models as Test-platform for Immunotherapeutics

With cellular products being on the front run there is a rising demand for non-animal-based test platforms to predict, study and treat undesired immunity. Here, we generated human organotypic skin models from human biopsies isolating and expanding keratinocytes, fibroblasts and microvascular endothelial cells finally allowing to seed these components on a collagen matrix or a biological vascularized scaffold matrix in a bioreactor. Afterwards, we were able to induce inflammation-based tissue damage by pre-stimulated mismatched allogeneic lymphocytes and/or inflammatory cytokine containing supernatants histomorphologically mimicking severe graft versus host disease (GvHD) of the skin. The effects could be prevented by the addition of immunosuppressants to the models. Consequently, these models would harbor a promising potential to serve as a test platform for the prediction, prevention and treatment of GvHD. This would also allow functional studies of immune effectors and suppressors including but not limited to allodepleted lymphocytes, gamma-delta T cells, regulatory T cells and mesenchymal stromal cells which would otherwise be limited to animal models. Thus, the current test platform developed with the limitation given that no professional APC are in place could highly reduce animal testing for investigation of novel immune therapies.


Introduction
The rise of cellular therapies during the last decade has not only been challenging for the healthcare providers but also for the regulators and GMP facilities around the globe. Pre-clinical testing requires extensive use of animal models or in vitro tests although those have proven their limited overall validity several times (Perel et al., 2007;Cook et al., 2012;Johnson et al., 2006;Hunig, 2012). Although animal models are a prerequisite for safe drug development translation of results to humans may be complicated and potentially delay drug development and even affect patients' safety. Therefore, there is a strong medical need for non-animal preclinical test platforms beyond simple human cell cultures systems. We have developed authentic threedimensional human organotypic models of the skin (Groeber et al., 2016a,b;Rossi et al., 2015) that find themselves already applied as human test platforms for risk assessment or efficacy testing in the cosmetic industry and research and gain rising attention from the regulators (Groeber et al., 2011(Groeber et al., , 2015. Here, we were able for the first time to induce inflammatory damage using mismatched lymphocytes as a surrogate for an incompatible allograft. The histomorphological results were highly authentic, clearly mimicking graft versus host disease (GvHD) known from patients' skin biopsies. GvHD is a severe, potentially life threatening complication of allogeneic hematopoietic stem cell transplantation (SCT) (Zeiser and Blazar, 2017) requiring immediate local or systemic immunosuppression. Both acute and chronic GvHD affect roughly every second patient being transplanted. Immunosuppression with calcineurin inhibitors or tacrolimus in combination with methotrexate has reduced severe and life-threatening acute GvHD rates and can be regarded as a broadly accepted clinical standard in GvHD prophylaxis ALTEX preprint published April 14, 2020 doi:10.14573/altex.1907181 while steroids are recommended for the treatment of GvHD . Severe GvHD of the skin is characterized by apoptotic epithelial damage that can lead to complete loss of the protective epithelial layer (Rowlings et al., 1997;Horowitz et al., 1990;Appelbaum, 2001;Falkenburg et al., 2002). GvHD has been extensively studied in several well-established mouse models such as the full MHC class I mismatch model C57BL/6 to BALB/c. Both transgenic and gene-deficient mouse strains have contributed significantly to the mechanistic understanding of GvHD and contributed to the development of allogeneic stem cell transplantation as we know it today (Blazar et al., 1991;van Leeuwen et al., 2002). However, cell populations and their expansion patterns as well as drug metabolism and pharmacology differ between men and mice (Boieri et al., 2016). Thus, there is high medical need for a predictive human in vitro test system (Hofmeister et al., 2004). Unfortunately, current in vitro test systems are weak predictors of GvHD due to a lack of physiological network interactions and organotypic structures. Similarly, humanized animal models have proven their limitation in predicting the GvHD potential of cellular products (Schroeder and DiPersio, 2011). So far, the only reliable predictor of skin GvHD has been developed by Vogelsang and colleagues who used skin biopsies as test platform for allogeneic lymphocytes and were indeed able to predict the appearance of skin GvHD (Dickinson et al., 2002;Sviland and Dickinson, 1999;Vogelsang et al., 1985). One limitation of this valuable model is certainly the limited number of tests it can be used for. Here, we present a novel type of skin model, which can be rebuilt from expanded cellular components of various skin samples and thus allows multiple testing thereby satisfying the needs of extensive GMP testing as well as mechanistic studies under changing conditions involving various types of immune suppressors in their ability to prevent or treat both GvHD and cytokine mediated damages.
Hence, the models could serve as more predictive alternatives to animal-based test procedures and might lead to the reduction or replacement of animal testing in GvHD research not only in mice but also in non-human primates, dogs, pigs and other large animal models.

2
Material and methods

Cell isolation and culture
Primary skin cells (keratinocytes, fibroblasts and microvascular endothelial cells) were isolated from human juvenile foreskin biopsies (donor age ≤ 7 years) or from adult skin biopsies of plastic surgeries (abdominoplasty, brachioplasty, dermolipectomy surgeries from upper arms and abdomen) according to previously published protocols (Rossi et al., 2015;Pudlas et al., 2011;Groeber et al., 2013). Juvenile foreskin biopsies were used to establish the skin models in the first place as the proliferative capacity of keratinocytes was expected to be higher as shown previously by our group (Lange et al., 2016). Indications for the circumcision were not available, however all samples underwent a visual inspection before being used in the models so that inflamed or infected tissues could be sorted out. The use of adult skin, however, was expected to be more challenging but allowed on the other hand the availability of peripheral blood samples from the volunteer donor at the same time. All volunteers or their legal representatives provided their written informed consent and all personal information was pseudonymized. The study protocol was approved by our local ethics committee (AZ 182/10 and 2/12). With later translation to the clinic in mind we also explored for the first time the use of 4 mm punch biopsies from the above described adult tissues. However, due to the limitation in size of these biopsies only fibroblasts and keratinocytes but no endothelial cells could be isolated. These punch biopsies were washed and treated with dispase (2 U/ml, Life Technologies GmbH, Darmstadt, Germany) for 60 min at 37 °C. Afterwards, the epidermal and dermal parts were separated and washed. The epidermis was incubated with 0.05 % (v/v) Trypsin/EDTA for 5 min at 37 °C. After addition of 10 % (v/v) heat-inactivated fetal bovine serum (FBS, Sigma-Aldrich ® Chemie GmbH, Schnelldorf, Germany), the cell suspension was mixed thoroughly for 5 min. Cells were washed and seeded in keratinocyte culture medium containing EpiLife ® basal medium supplemented with Human Keratinocyte Growth Supplement, 100 U/ml penicillin and 100 µg/ml streptomycin (PenStrep, all from Life Technologies GmbH) on juvenile human fibroblasts as feeder cells treated with Mitomycin C (4 µg/ml, Sigma-Aldrich ® Chemie GmbH) (Aasen and Izpisua Belmonte, 2010). These cells were derived from the foreskin biopsies. Dermal parts of the punch biopsies were incubated for 45 min at 37 °C in collagenase (500 U/ml, Roche Diagnostics GmbH, Mannheim, Germany) and washed twice. The suspension with dermal tissue residues was resuspended in fibroblast culture medium containing Dulbecco's Modified Eagle Medium (Life Technologies GmbH) supplemented with 10% (v/v) FBS and PenStrep. After fibroblasts had grown out, remaining tissue residues were washed away. Isolated keratinocytes were cultured in keratinocyte culture medium and fibroblasts in fibroblast culture medium. Microvascular endothelial cells were isolated as described before (Groeber et al., 2013) and cultured in VascuLife ® (VascuLife ® VEGF Endothelial Medium Complete Kit, LifeLine Cell Technology, Frederick, MD, USA) supplemented with PenStrep. For experiments, keratinocytes were used from 1 st to 3 rd passage, fibroblasts from 1 st to 5 th passage and endothelial cells from 1 st to 4 th passage.
Peripheral blood mononuclear cells (PBMCs) were isolated from leukoreduction system chambers (Trima Accel apheresis apparatus, Terumo BCT, Lakewood, CO, USA) (Bock et al., 2002) of healthy platelet donors (personal data fully anonymized) or from adult skin donors (personal data pseudonymized) by density-gradient centrifugation using Biocoll Separating Solution (1.077 g/ml, Biochrom GmbH, Berlin, Germany). For experiments, PBMCs were either used directly after isolation or cryopreserved and thawed the day before the experiment.

Generation of human skin models
Two types of human skin models were used in this study as illustrated in Figure 1. Every skin model was built from only one individual tissue donor. Human full-thickness skin models were generated based on a collagen hydrogel in 24-well-inserts (Polycarbonate Membrane Inserts in Multidish 24, Nunc TM , Roskilde, Denmark) according to a previously published protocol using the EpiLife ® medium system (Lange et al., 2016). Briefly, 500 µl dermal equivalents were prepared by mixing 5x10 4 fibroblasts per insert with the collagen solution (final collagen concentration 4 mg/ml). After 24 h, keratinocytes were seeded on the dermal equivalents with a density of 5x10 5 cells/cm² in keratinocyte culture medium supplemented with 1.5 mmol/l ALTEX preprint published April 14, 2020 doi:10.14573/altex.1907181

Fig. 1: Human artificial skin models
Human primary skin cells (keratinocytes, fibroblasts and microvascular endothelial cells) isolated from human juvenile foreskin or adult 4 mm skin biopsies were used for generation of artificial skin models. Every skin model was built from only one individual tissue donor. (A) For collagen based full-thickness skin models, dermal equivalents were generated by mixing human fibroblasts with rat-tail collagen type I. To form the epidermal layer, keratinocytes were seeded on top of the dermal part. (B) For the vascularized skin models (SkinVaSc-TERM ® ) endothelial cells were injected into the vascular system of the matrix. Fibroblasts and keratinocytes were seeded on top. Vascularized skin models were cultured with physiological perfusion. For both types of skin models, surface of the epidermal layer was exposed to the air to allow epidermal differentiation.
CaCl2 (Sigma-Aldrich ® Chemie GmbH) and cultured overnight. Then medium was replaced and supplemented with 50 µg/ml vitamin C (Caeser & Loretz GmbH, Hilden, Germany) followed by exposure of the surface to air for induction of epidermal differentiation. Medium change was performed every two or three days.
Furthermore, human vascularized skin models (SkinVaSc-TERM ® ) were generated in-house on a biological vascularized scaffold (BioVaSc-TERM ® ) (Mertsching et al., 2009;Scheller et al., 2013) in a previously published perfusable bioreactor system (Groeber et al., 2013(Groeber et al., , 2016a on the basis of the EpiLife ® medium system. The matrix was prepared from porcine jejunal segments of German Landrace. Scaffolds used in here described experiments were explanted from dead animals between 2014 and 2016, which did not require approval according to animal care guidelines. Briefly, vessels were saturated with VascuLife ® overnight and reseeded with 1x10 7 endothelial cells. The matrix was connected to the circulation of a bioreactor system (Groeber et al., 2013). Then 8.9x10 5 fibroblasts were seeded on the matrix surface. The perfusion was started and the pressure was increased stepwise to 100 mmHg finally implementing a physiological pulsatile pressure profile with a 20-mmHg amplitude. After 5 days, medium was changed to keratinocyte culture medium supplemented with 1.5 mmol/l CaCl2 and 4x10 6 keratinocytes were seeded on the matrix surface. Dynamic culture was continued for 3 days, when medium was replaced and supplemented with 50 µg/ml vitamin C. Following, the surface was exposed to air and skin models were cultured for further 11-14 days under dynamic conditions with medium change every 6 days.

Induction of tissue damage in human skin models
For simulation of GvHD and induction of tissue damage, autologous or allogeneic PBMCs pre-stimulated in an MLR were injected into the collagen skin models. Therefore, MLRs were prepared in an autologous stetting (PBMCs from the adult tissue donor stimulated with irradiated PBMCs of the same individual) or an allogeneic setting (randomly mismatched allogeneic PBMCs stimulated with irradiated PBMCs from the autologous tissue donor). MLR was set up with 1x10 6 responder cells/ml in CellGro ® GMP DC Medium for seven days. Furthermore, MLRs were prepared by addition of 200 ng/ml cyclosporin A (CsA, Sandimmun ® Injection, Novartis Pharmaceuticals Corporation, East Hanover, NJ, USA) to test whether tissue damages could be prevented. For enrichment of CsA in the model, collagen skin models were pretreated with 200 ng/ml CsA one day before injection.
Additionally, allogeneic PBMCs were pre-stimulated with 25 ng/ml PMA and 1 µg/ml ionomycin overnight or left unstimulated. Afterwards, cells were washed twice with CellGro ® GMP DC Medium. Collagen skin models were used at day 13 of culture at the air-liquid interface. Both pre-stimulated and non-stimulated PBMCs with or without pre-treatment of CsA were injected with insulin syringes at two locations into the dermal part of the skin models (2x10 6 cells per skin model). Injection of CellGro ® GMP DC Medium alone served as control. EpiLife ® medium was replaced by fresh CellGro ® GMP DC Medium as culture medium. Furthermore, conditioned medium supernatants of unstimulated or pre-stimulated PBMCs were collected and injected directly into the skin models and also used as culture medium under these conditions. Using at least two ALTEX preprint published April 14, 2020 doi:10.14573/altex.1907181 different skin cell donors, triplicates were prepared for each condition. After incubation for 48 hours, medium supernatants were collected from collagen skin models and stored at -80°C. The collagen skin models were fixed in paraformaldehyde (Roti ® -Histofix, Carl Roth GmbH & Co.KG, Karlsruhe, Germany) for two hours and further processed for histological analysis.
For the vascularized skin models, PBMCs were first labeled with 2.5 µM carboxyfluorescein succinimidyl ester (CFSE, CellTrace TM CFSE Cell Proliferation Kit, Life Technologies GmbH) and then pre-stimulated with 50 ng/ml PMA and 0.8 µg/ml ionomycin overnight. Afterwards, PBMCs were washed twice with CellGro ® GMP DC Medium. Vascularized skin models were used after 11 to 14 days of culture at the air-liquid interface. Two conditions were tested. CFSE-labeled, prestimulated PBMCs were injected into the arterial access of the vascularized skin models and also added into the pressure flask (1.5 2x10 8 cells in each case). Injection of CellGro ® GMP DC Medium alone into the arterial access and into the pressure flask of a second bioreactor served as control. Three experiments were performed with different end points: 48, 72 or 96 hours. Medium supernatants were collected and stored at -80°C after 0, 4, 8, 24, 48, 72 (if available) and 96 hours (if available). PBMCs were recollected from circulation and prepared for flow cytometry. Vascularized skin models were fixed in paraformaldehyde overnight and further processed for histological analysis. All vascularized skin models were prepared with cells from one skin cell donor, but with PBMCs from three different donors.

Statistical analysis
Statistical analyses and graphs were conducted applying Prism software v5.04 (GraphPad Software Inc., San Diego, CA, USA). Normal distribution was evaluated using the Kolmogorov Smirnov normality test. For comparison of two groups the unpaired t-test for values with Gaussian distribution was performed. For comparison of more than two groups the one-way analysis of variances (ANOVA) with Bonferroni's Multiple Comparison test for values with Gaussian distribution was performed. A p value of 0.05 or less was considered significant.

Data sharing
For original data, please contact stephan.mielke@ki.se.

Immunogenic properties of human primary skin cells
As a first step, we investigated the ability of the individual cellular components such as keratinocytes, fibroblasts and endothelial cells isolated from adult skin biopsies to stimulate or suppress previously activated or non-activated randomly mismatched PBMCs. All cell types investigated showed marked expression of HLA class I molecules and HLA DR upon stimulation with IFNγ. However, the costimulatory molecule CD80 was expressed at low levels in keratinocytes, fibroblasts and endothelial cells whilst CD86 expression lacked completely even following previous stimulation. Notably, the expression of the inhibitor PD L1 was upregulated in all three cell types after stimulation with IFNγ ( Fig. S1 1 ). Irradiated endothelial cells, keratinocytes and fibroblasts proved their general ability to stimulate allogeneic PBMCs (Fig. S2 1 ). However, attempting to increase their antigen-presenting capacity by unspecific pre-stimulation prior to irradiation did not increase overall proliferative activity as measured in one-way MLRs (Fig. S2 1 ). Consequently, allogeneic PBMCs were also pre-stimulated before including them into the tissue models to overcome a missing lymph node function.

Pre-stimulated PBMCs induce tissue damage in collagen skin models
As a next step, we exposed the collagen skin model built-up from juvenile skin cells to allogeneic PBMCs with and without pre-stimulation. Here, previously unstimulated PBMCs did not cause significant alterations as compared to medium controls ( Fig. 2A). In contrast, PMA/ionomycin pre-stimulated PBMCs resulted in disruption of the epidermal cellular organization and complete detachment of the epidermal layer from the dermal part (mimicking histological grade IV acute GvHD). Once, tissue damage was observed the same damage could be induced in other models by the medium supernatant from lymphocytes previously stimulated with PMA/ionomycin. We could show that these supernatants contained high levels of pro-inflammatory cytokines such as IFNγ, IL2, IL17A and TNFα (Fig. 2B). Immunochemical staining of the lymphocytic infiltrates revealed their identity as CD4 + and CD8 + T lymphocytes (Fig. 2C). Additionally, we exposed collagen skin models built-up from adult skin cells, isolated from 4 mm punch biopsies, to autologous and allogeneic PBMCs, which were pre-stimulated in respective MLRs. The injection of allogeneic stimulator cells resulted in disruption of the epidermal cellular organization and partly detachment of the epidermal layer from the dermal part

Fig. 2: Inflammation-induced damage in collagen skin models (juvenile foreskin skin cells)
Collagen skin models were generated using primary keratinocytes and fibroblasts isolated from human juvenile foreskins. For each experiment, every skin model was built from only one individual tissue donor. Allogeneic PBMCs were used from another donor. Culture medium, unstimulated PBMCs or allogeneic pre-stimulated PBMCs (unspecific stimulation with PMA/ionomycin) were injected into the matrix of collagen skin models and co-cultured for 48 hours. Three experiments were performed in triplicates (n=3). (A) HE-stained histological cross sections of collagen skin models are presented. Representative sections from skin models derived from a single skin donor are shown. The arrow points disruption of the epidermal cellular organization and detachment of the epidermal layer from the dermal part. Scale bars indicate 100 µm. (B) Cytokine levels of IFNγ, IL2, IL17A and TNF in the supernatants of the culture medium of the skin models were measured using Luminex ® technology. Results are expressed as mean and SEM (n=3). (C) Representative immunochemical staining of CD3, CD4 and CD8 molecules in PBMC infiltrates of one collagen skin model treated with pre-stimulated PBMCs is shown. Scale bars indicate 100 µm. * = p<0.05, ** = p<0.005, *** = p<0.0001. # = cytokine level not detectable.

Fig. 3: Inflammation-induced damage in collagen skin models (adult skin cells from punch biopsies)
Collagen skin models were built up from adult skin cells isolated from 4 mm punch biopsies. For each experiment, every skin model was built from only one individual tissue donor. Culture medium (data not shown), autologous, allogeneic or PMA/ionomycin pre-stimulated PBMCs were injected into the matrix of collagen skin models. Additionally, medium supernatants were collected from unstimulated or pre-stimulated PBMCs, injected into collagen skin models and used as culture medium. Furthermore, autologous and allogeneic PBMCs pre-stimulated in an MLR cultured with 200 ng/ml CsA were injected into the matrix. Collagen skin models were pre-treated with CsA one day before injection. Each condition was co-cultured for 48 hours. Two experiments were performed in triplicates (n=2). (A) Representative sections from skin models derived from a single donor are shown. The arrow points disruption of the epidermal cellular organization and detachment of the epidermal layer from the dermal part. HEstained histological cross sections of collagen skin models are presented. (B) Cytokine levels of IFNγ, IL2, IL17A and TNF in the supernatants of the culture medium of the skin models were measured using Luminex ® technology. Results are expressed as mean and SEM (n=2). (C) Representative immunochemical staining of CD3, CD4 and CD8 molecules in PBMC infiltrates of one collagen skin model treated with allogeneic PBMCs is shown. Scale bars indicate 100 µm.

ALTEX preprint published April 14, 2020 doi:10.14573/altex.1907181
(mimicking histological grade IV acute GvHD) (Fig. 3A). No such changes were observed in the autologous setting (negative control). PBMCs pre-stimulated with PMA/ionomycin served as a positive control, where lymphocytes infiltrated the model and induced dermal-epidermal detachment. Tissue damage was not observed after addition of medium supernatants from allogeneic or autologous PBMCs to the skin models, only from PBMCs pre-stimulated with PMA/ionomycin probably due to the high concentration of different cytokines (Fig. 3B). Addition of 200 ng/ml CsA to the collagen skin models as well as to the pre-stimulating MLR lead to prevention of tissue damage in the collagen skin models (Fig. 3A). Furthermore, we could show that supernatants contained pro-inflammatory cytokines such as IFNγ, IL2, IL17A and TNFα (Fig. 3B). Immunochemical staining of the lymphocytic infiltrates revealed their identity as CD4+ and CD8+ T lymphocytes (Fig. 3C).

Pre-stimulated PBMCs induce tissue damage in vascularized skin models
As a final step, we anticipated to induce the same inflammation in even more sophisticated vascularized bioreactor-based skin models built-up from juvenile skin cells, where we injected PMA/ionomycin pre-stimulated PBMCs in the medium circulation within the skin vessels. Skin models treated solely with medium served as a control. Bioreactors were stopped after 48, 72 or 96 hours. Controls showed intact multilayered, cornified epidermis attached to the dermal part (Fig. 4A). In contrast, vascularized skin models treated with pre-stimulated PBMCs showed significant signs of cellular damage in the epidermal layer beginning after 48 hours (mimicking histological grade II acute GvHD). Disruption of the epidermal organization as well as detachment of the epidermis from the dermis became more severe after 72 hours (mimicking histological grade III acute GvHD). After 96 hours, the epidermal layer was completely detached and infiltrates of PBMCs were found within the matrix (mimicking histological grade IV acute GvHD) (Fig. 4B).
After 96 hours, these infiltrates were shown to consist of high proportions of CD4 + helper T lymphocytes and CD8 + cytotoxic T lymphocytes by immunochemical staining (Fig. 5A). We found a pro-inflammatory cytokine profile including IFNγ, IL2, IL17A and TNFα (Fig. 5B). In contrast, these cytokines were not detectable in the medium control at relevant levels. Most interestingly, PBMCs drawn from the bioreactors vessels after 96 hours kept on proliferation without re-stimulation as shown in Figure 5C.

Fig. 4: Inflammation-induced damage in vascularized skin models (juvenile skin cells)
Vascularized skin models (SkinVaSc-TERM ® ) were generated in a perfusable bioreactor system. Every skin model was built from only one individual tissue donor. Pre-stimulated, CFSE-labeled PBMCs were introduced into the circulation of the bioreactor system. As a control, culture medium only was injected accordingly. Bioreactors were stopped after 48, 72 or 96 hours. Representative HE-stained histological cross sections of vascularized skin models are presented for (A) medium control and (B) PBMCs pre-stimulated with PMA/ionomycin. Vascularized skin models were prepared from one skin cell donor, but with PBMCs from different donors for each experiment. The arrow points detachment of the epidermal layer from the dermal part. Scale bars indicate 100 µm.

Discussion
With view to the worldwide rise of cellular therapies particularly in cancer immunotherapy there is a strong upcoming demand for test platforms allowing to predict immunologically induced tissue damage caused by novel cellular products such as genetically modified T lymphocytes (e.g. suicide genes, chimeric antigen receptors), expanded and/or modified autologous or ALTEX preprint published April 14, 2020 doi:10.14573/altex.1907181

Fig. 5: Inflammation-induced damage in vascularized skin models (juvenile skin cells)
Vascularized skin models (SkinVaSc-TERM ® ) were generated in a perfusable bioreactor system. Every skin model was built from only one individual tissue donor. Pre-stimulated, CFSE-labeled PBMCs were introduced into the circulation of the bioreactor system. As a control, culture medium only was injected accordingly. Vascularized skin models were prepared from one skin cell donor, but with PBMCs from different donors for each experiment. (A) Immunochemical staining of histological cross sections for CD3, CD4 and CD8 molecules is presented. Representative sections from vascularized skin models with prestimulated PBMCs after 96 hours are shown. Scale bars indicate 100 µm. (B) Cytokine levels of IFNγ, IL2, IL17A and TNFα in supernatants of the culture medium after 0, 4, 8, 24, 48, 72 and 96 hours were measured using Luminex ® technology. Cytokine profile for the bioreactor that was stopped after 96 hours is shown. The results are depicted as timelines for vascularized skin models with medium only (blue symbols) and vascularized skin models with pre-stimulated PBMCs (red symbols). (C) After bioreactors were stopped, pre-stimulated, CFSE-labeled PBMCs were recollected from circulation, prepared for flow cytometric analysis and compared to unstimulated PBMCs cultured for the same period of time in a culture flask. Representative results are shown for the bioreactor stopped after 96 hours. Cells were gated for lymphocyte properties (forward and side light scatter). CD3 positive cells were analyzed for CD25 expression and proliferation (CFSE dilution, dot plots). Percentages in CFSE histograms correspond to the proportion of cells having undergone at least one cell division. published April 14, 2020 doi:10.14573/altex.1907181 allogeneic NK cells, expanded tumor infiltrating lymphocytes (TILs), allodepleted lymphocyte products as well as allogeneic or autologous antigen specific T lymphocytes against viruses and tumor associated autoantigens (TAAs) (Schuster et al., 2017;Mielke et al., 2008;Barkholt et al., 2009;Maude et al., 2014;Stevanovic et al., 2015;Lutz et al., 2015;Bjorklund et al., 2018). With allogeneic SCT representing the most commonly applied cellular immunotherapy, to date GvHD remains as one of the most important complications that can neither be fully predicted nor be avoided. However, advances in high-resolution HLAtyping for unrelated donors (Fleischhauer et al., 2018), novel immunosuppressive drugs (Zeiser et al., 2015;Mielke et al., 2014;Lutz and Mielke, 2016) and advanced graft manipulation techniques (Handgretinger et al., 2016;Montero et al., 2006;Barrett et al., 1998) have certainly contributed to reduce these risks. Furthermore, more recent and highly promising developments in the field of biomarkers may allow early detection and treatment of clinically relevant GvHD (Major-Monfried et al., 2018;Ahmed et al., 2015). Ex vivo graft manipulation techniques such as selective allodepletion prevented or reduced the likelihood of high-grade GvHD in HLA-mismatched allotransplantation allowing to cut down on immunosuppression thereby enhancing graft-versus-leukemia effects (Roy et al., 2019(Roy et al., , 2020. Thus, tests predicting severe GvHD after ex vivo selection or manipulation of allogeneic immune cells could be of significant value for patient safety. MLRs between donor and recipient have been proven to have predictive value in mismatched transplantation (Mielke et al., 2008;Amrolia et al., 2003) whilst in matched allogeneic SCT the skin explant assay and the helper T-lymphocyte precursor frequency (HTLp) assay were the only assays able beyond HLA typing to predict for GvHD based on minor antigen differences (Kaminski et al., 1989;Sharrock et al., 1990;Schwarer et al., 2004;Fleischhauer et al., 2006). HTLps are very much dependent on the type of antigen-presenting cells (APCs) being used in the assay. However, using PBMCs or expanded T lymphocytes as stimulators have proven to be predictable for overall GvHD (Solomon et al., 2005). The skin explant system requires a skin biopsy from the recipient and lymphocytes from the donor that are then co-cultured. The advantage of this system is of course that a representative fraction of recipient skin is investigated and that this skin biopsy contains professional APCs. The sole but important drawback however is the very limited number of experiments that can be done on a single biopsy. Here, we were able to mimic for the first time GvHD-like changes in artificial human skin models induced by randomly mismatched and pre-stimulated lymphocytes.
Comparable to the skin explant assay, which served as a theoretical template for our GvHD-model, the great advantage in contrast to animal models is the use of human cells in a realistic three-dimensional organotypic setting. With view to the testing of immunotherapeutic drugs the ability to use cells from the same species is an important prerequisite as HLA molecules across borders of species can induce severe immune reactions. In the past, artificial human skin models were built-up from preputium or major parts of skin derived from surgical interventions. Here, we succeeded to build-up several collagen skin models out of a single 4 mm punch biopsy followed by isolation, expansion and structural reshaping of the three-dimensional organ. Therefore, the advantage of the organ models presented here would be the ability to generate several of these models (at least 30 skin models per donor) allowing then multiple and complex investigations in a human test platform thereby facilitating mechanistic research that otherwise would not be feasible in a human setting. In the future, FBS could be replaced by the use of serum-free media or normal AB serum to work in a truly human environment. Certainly, these models need to undergo further optimization to reach this stage. However, such research is deeply warranted as the exact mechanistic pathways of several clinically-applied immune effector cells such as mesenchymal stromal cells remain unknown to date as such basic research has been so far limited to animal models and simple in vitro test systems. Due to the complexity of the human body system such mechanistic pathways are difficult to reveal in clinical trials with uncountable variables being involved. We employed both a collagen skin model under static conditions and a vascularized skin model under dynamic conditions in a bioreactor system (Groeber et al., 2016a;Rossi et al., 2015;Lange et al., 2016). Pre-stimulated or unstimulated randomly HLA-mismatched lymphocytes were either injected directly into the model or into the medium circulation and could either way induce a histomorphological pattern known for GvHD in the final read out. Results from cytokine measurements confirmed a GvHD typical profile of strong inflammation whilst immunohistochemical analyses revealed the presence of CD3+CD8+ lymphocytes. Nonetheless, it is worthwhile mentioning in this context that a GvHD-like pattern of damage could be also induced by the sole use of supernatants of PBMC cultures having been pre-stimulated with an artificial, unphysiological polyclonal stimulator (PMA/ionomycin) and consequently containing extremely high levels of pro-inflammatory cytokines. Consequently, such models may indeed harbor the potential to study the damaging effects of high-level cytokines on the skin as observed for example in different forms of cytokine release syndromes.
On a more critical note, it is obvious that the here presented human artificial skin models cannot mimic the full pathway of GvHD development (Ferrara and Reddy, 2006). Comparable to the skin explant assay, these models lack lymph nodes and consequently the natural priming of T lymphocytes by activated professional APCs in the periphery. In our experiments, all different types of skin cells lacked sufficient expression of co-stimulatory molecules even after pre-stimulation and therefore professional APC features. However, sufficient pre-stimulation of lymphocytes can certainly overcome this limitation mimicking the function of the lymph nodes even if such a scenario can never be fully reflective of the original pathways and complexity in the human body. Although keratinocytes, fibroblasts and endothelial cells do not belong to professional APCs, it has been reported that non-hematological APCs are sufficient for induction of GvHD in mice . However, to date the individual ability of fibroblasts, keratinocytes and endothelial cells to exert APC-like functions is discussed controversially in the literature van Dijk et al., 2000;Teshima et al., 2002;Takagi et al., 2006;Haniffa et al., 2007;Dominguez-Castillo et al., 2008). In our study, unstimulated fibroblasts, keratinocytes and endothelial cells were able to induce proliferation without prior stimulation. Nonetheless, previous exposure to other GvHDrelevant cytokines such as TNFα, IL17A and IL6 or combinations of such could have had a much stronger effect (Pechhold et al., 1997;Tawara et al., 2011;Ju et al., 2014). This finding is line with accumulating evidence that endothelial cells play a significant role in GvHD pathogenesis Zeiser et al., 2011) and may develop into sufficient APCs as a consequence of an exposure to a pro-inflammatory environment thereby enhancing the GvHD cascade (Dietrich et al., 2013;Schmid et al., 2014). Therefore, the vascularized model may indeed exert advantages in simulation of secondary GvHD effects over the collagen model. On the other hand, construction of vascularized skin model was technically demanding and time consuming. Taking into account that a decent number of bioreactors needs to be available for one experiment, we believe that for daily clinical purposes of product testing and multiple comparisons in complex experiments the collagen model appears to ALTEX preprint published April 14, 2020 doi: 10.14573/altex.1907181 be the more feasible one although the vascularized model harbors unique opportunities such as the possibility of drug delivery directly into a vessel-like circulation system. Finally, ease of use and reliability will be pre-requisites for clinical applicability. Our results clearly show that allogeneic PBMCs pre-activated in an MLR were an essential requirement to exert skin destruction in our skin models thereby overcoming the primary absence of professional APCs such as Langerhans cells. Nonetheless introduction of professional APCs into these models would certainly enhance the authenticity of simulated immunological processes (Chau et al., 2013). Taken together, we were able to induce inflammation based histomorphological patterns of tissue damage to both collagen and vascularized human skin models by pre-stimulated lymphocytes. The histopathological read out revealed alterations clearly mimicking GvHD. Nonetheless, these models of GvHD-like tissue damage are limited in their ability to reflect all pathophysiological aspects of GvHD. To be able to assess these models' ability in predicting GvHD we are planning to investigate the residual skin GvHD potential of ex vivo manipulated, selectively allodepleted lymphocytes in haploidentical blood SCT using full-collagen models built-up from recipients' skin samples. With regard to the world-wide rise of haploidentical SCT (Passweg et al., 2017), methods to allow ex vivo testing of manipulated T lymphocytes are likely to be highly demanded. Currently, we are planning to expand these models to other GvHD organs such as gut and liver. Nonetheless, the predictive value of the artificial skin models for GvHD in other organs than skin may still be sufficient enough as known from the skin explant assay and HTLPs where both assays were lacking original APCs from liver and gut and still could predict overall GvHD development (Sviland and Dickinson, 1999;Schwarer et al., 2004;Zhang et al., 2002). These models will allow sophisticated mechanistical studies of mesenchymal stromal cells, regulatory T cells or novel immunosuppressive drugs such as ruxolitinib for the prevention and/or treatment of GvHD in a fully human organ model. So far, only cyclosporine A has been tested but other drugs may result in different findings. This includes a unique opportunity to perform time-dependent studies of biomarkers heralding GvHD development and severity (Major-Monfried et al., 2018;Ahmed et al., 2015).
Developing these models into a sophisticated, reliable test-system would not only reduce animal testing but also enhance accuracy for the prediction of GvHD. A further development would be to incorporate professional APCs and add lymph node-like structures to these models. All of this would then certainly contribute to our overall understanding of underlying mechanisms of action with respect to the development of novel immunotherapeutics in GvHD treatment.