Biofabrication

PAPER • OPEN ACCESS You may also like

- Cell-printed 3D liver-on-a-chip possessing
Liver-on-Micropillar: a humanized, animal-free a liver microenvironment and biliary
system
platform for high-throughput assessment of drug- Hyungseok Lee, Suhun Chae, Jae Yun
Kim et al.

induced liver injury - Gastrointestinal organs and organoids-on-

a-chip: advances and translation into the
clinics
To cite this article: Ahmed S M Ali et al 2025 Biofabrication 17 045022 Mariana R Carvalho, Le-Ping Yan, Bo Li et
al.

- hiPSC-derived macrophages improve drug

sensitivity and selectivity in a macrophage-
incorporating organoid culture model
View the article online for updates and enhancements. Seongyea Jo, Sung Bum Park, Hyemin
Kim et al.

This content was downloaded from IP address [Link] on 26/10/2025 at 12:50

 Biofabrication 17 (2025) 045022 [Link]

Biofabrication

PAPER

OPEN ACCESS
Liver-on-Micropillar: a humanized, animal-free platform for
high-throughput assessment of drug-induced liver injury
RECEIVED
12 August 2025
Ahmed S M Ali1,3, Dongwei Wu1,3, Viola Roehrs1, Duy Bao Tran Nguyen1, Mathias Ziersch1,
REVISED
24 September 2025
Albert Braeuning2 and Jens Kurreck1,∗
1
ACCEPTED FOR PUBLICATION Technische Universität Berlin, Chair of Applied Biochemistry, Berlin, Germany
2
7 October 2025 Department Chemical and Product Safety, German Federal Institute for Risk Assessment (BfR), Berlin, Germany
3
PUBLISHED
These authors contributed equally to this work.
∗
23 October 2025 Author to whom any correspondence should be addressed.
E-mail: [Link]@[Link]
Original content from
this work may be used Keywords: 3D bioprinting, drug-induced liver injury (DILI), animal-free testing, xeno-free medium (XFM),
under the terms of the Liver-on-Micropillar (LoM)
Creative Commons
Attribution 4.0 licence. Supplementary material for this article is available online
Any further distribution
of this work must
maintain attribution to
the author(s) and the title Abstract
of the work, journal
citation and DOI.
Drug-induced liver injury (DILI) remains a major cause of acute liver failure, clinical trial attri-
tion, and post-marketing drug withdrawal, yet predictive in vitro models are limited in accuracy,
scalability, and human relevance. Here, we present a Liver-on-Micropillar (LoM) platform a fully
animal-free, high-throughput, miniaturized human liver model designed for early-stage hepato-
toxicity screening. The system combines a xeno-free medium with a xeno-free bioink to support
co-culture of four human liver-relevant cell types: differentiated HepaRG, LX-2, HMEC-1, and
differentiated THP-1 cells. Microlivers are bioprinted onto micropillar arrays compatible with
standard 96-well plate formats. Functional characterization confirmed stable cell viability, albumin
and urea production, as well as inducible CYP expression. To evaluate DILI predictivity, ten refer-
ence drugs were tested using assays to measure ATP content, XTT metabolic activity, and albumin
secretion. Half-maximal inhibitory concentrations (IC50 ) were experimentally determined, and
margins of safety (MOS) were calculated by dividing IC50 by clinical maximum plasma concen-
tration (Cmax ). The LoM platform correctly classified 90% of the tested compounds using a MOS
threshold of 100. This scalable and reproducible model provides a human-relevant, regulatory-
aligned alternative to animal testing and supports broader efforts to implement non-animal meth-
odologies in drug safety evaluation.

1. Introduction human hepatic responses [5, 6]. Traditional two-

dimensional (2D) monocultures of hepatocyte-like
Drug-induced liver injury (DILI) remains a persist- cells lack architectural fidelity and rapidly lose liver-
ent and unresolved challenge in both drug develop- specific functions, including cytochrome P450 (CYP)
ment and clinical medicine. It is the leading cause of activity and albumin synthesis [7, 8]. Moreover,
acute liver failure in western countries and accounts reliance on fetal bovine serum (FBS) and other
for a substantial proportion of late-stage clinical trial animal-derived supplements introduces undefined
failures and post-marketing drug withdrawals [1–3]. factors, reduces reproducibility, and complicates
The pathophysiology of DILI is multifactorial and regulatory approval particularly for pharmaceutical
includes both concentration-dependent and idiosyn- applications [9, 10].
cratic mechanisms [4]. Given this complexity, robust Recent advances in tissue engineering have
and predictive preclinical models are essential for yielded three-dimensional (3D) liver models with
accurate risk stratification of drug candidates. Despite improved phenotypic stability and functional output
decades of research, existing in vitro and in vivo [11, 12]. However, many of these platforms suffer
models exhibit poor translational concordance with from technical complexity, low throughput, high

© 2025 The Author(s). Published by IOP Publishing Ltd

Biofabrication 17 (2025) 045022 A S M Ali et al

Figure 1. Workflow of the Liver-on-Micropillar (LoM) platform. Schematic representation of the LoM platform, outlining each
step from xeno-free culture medium formulation and human cell line adaptation to micropillar fabrication via high-resolution
3D bioprinting, preparation of an animal-free ECM-based bioink, and 3D bioprinting of liver constructs onto micropillars. The
microlivers are then cultured under static conditions and assessed for viability, functionality, and drug-induced liver injury using
multiple assay endpoints (ATP, XTT, albumin). Created in BioRender. Kurreck, J. (2025) [Link] b8gby8x.

operational costs, or continued reliance on animal- of truly animal-free, 3D-bioprinted hepatic mod-
based components [13–16]. Furthermore, most mod- els remains limited, due to the lack of effective
els omit critical cellular contributors to the hepatic media without animal components and efficient
microenvironment, such as hepatic stellate cells, liver bioprinting strategies, hindering their broader use in
sinusoidal endothelial cells, and liver-resident mac- pharmaceutical screening [26].
rophages (Kupffer cells). In parallel, there is growing To address these limitations, we have developed
regulatory and societal pressure to reduce animal test- a Liver-on-Micropillar (LoM) platform, a miniatur-
ing and transition toward human-based, animal-free ized, 3D-bioprinted human liver model designed for
methodologies [17, 18]. A major step in this direction high-throughput, animal-free toxicity testing. The
came in April 2025, when the U.S. Food and Drug platform utilizes a xeno-free medium (XFM) com-
Administration (FDA) announced plans to replace patible with co-culture of four human cell types: dif-
animal testing in preclinical toxicity studies, partic- ferentiated HepaRG hepatocyte-like cells, LX-2 hep-
ularly for biologics such as monoclonal antibodies atic stellate cells, HMEC-1 endothelial cells, and dif-
[19]. These initiatives align with the broader 3Rs ferentiated THP-1 macrophage-like cells. These cells
principles (replacement, reduction, refinement) and are incorporated in a bioink composed exclusively of
underline the urgent need for human in vitro models animal-free extracellular matrix. The liver microtis-
with high predictivity [20, 21]. sues were fabricated atop an array of miniature posts,
Among the emerging technologies in this con- herein referred to as micropillars, which are engin-
text, 3D bioprinting has gained considerable trac- eered to be geometrically compatible with standard
tion due to its ability to spatially organize mul- 96-well plate formats, facilitating high-throughput
tiple human cell types within customized extracellu- drug screening. This LoM platform was functionally
lar matrix scaffolds [22–24]. This technique enables validated using ten reference drugs spanning a range
the fabrication of physiologically relevant, multicel- of DILI risks and clinical outcomes. Margins of safety
lular liver constructs with enhanced reproducibility, (MOS) were calculated by normalizing half-maximal
scalability, and functional performance. Moreover, inhibitory concentration (IC50 ) values to clinical
the flexibility of 3D bioprinting allows the fabrica- maximum plasma concentration (Cmax ), providing a
tion of miniaturized tissue constructs in standard- pharmacologically relevant metric for evaluating clin-
ized formats, such as 96-well plates, which enables ical risk. The results demonstrate the LoM system as
automated handling and high-throughput screen- a reproducible, pharmacologically relevant, and fully
ing. This makes bioprinting particularly well-suited animal-free platform for early-stage hepatotoxicity
for generating reproducible liver models for drug evaluation and prioritization of drug candidates. The
testing applications [25]. However, the development workflow in the current study is delineated in figure 1.

2
Biofabrication 17 (2025) 045022 A S M Ali et al

2. Results cell line was cultured in basal medium (William’s

E + ITS + FN for adherent cells or +2% HS for
2.1. Optimization of XFM THP-1), supplemented individually with candidate
To establish a humanized in vitro liver model suit- additives (table S1), and proliferation was quantified
able for DILI assessment, we employed a multi- after 48 h of culture. The impact of each additive was
cellular approach incorporating four liver-relevant assessed based on cell count compared to the basal
human cell types: differentiated HepaRG cells as medium control (figure S3). Several components
metabolically competent hepatocyte-like cells, LX-2 exhibited a broadly beneficial effect and were retained
cells as hepatic stellate cells, HMEC-1 as microvas- in the final medium formulation. For example, L-
cular endothelial cells, and differentiated THP-1 glutamine significantly enhanced proliferation across
as macrophage-like cells (M0) representing resid- all four cell types and was thus considered essential.
ent Kupffer cells. The first challenge was to define Insulin and hydrocortisone stimulated the prolifera-
culture conditions that could support the pheno- tion of HepaRG, HMEC-1, and LX-2, with no detect-
typic maintenance and proliferation of all four cell able negative effects on THP-1 cells, and were there-
types in the absence of FBS or any animal-derived fore included despite cell-type-specific variation in
component. Among the four cell types, HepaRG, responsiveness. Similarly, sodium pyruvate was bene-
LX-2, and HMEC-1 require substrate attachment ficial for THP-1 cell count but had a neutral effect
for proliferation and morphogenesis, whereas THP-1 on the adherent lines; it was included to support
cells are maintained in suspension during expansion THP-1 metabolism without compromising compat-
and adhere only upon differentiation. To optimize ibility. Conversely, certain components demonstrated
adhesion under chemically defined conditions, we cell-type-specific cytotoxicity or undesired pheno-
screened several human-based or synthetic attach- typic effects and were excluded despite having posit-
ment factors: polyethyleneimine (PEI), human serum ive effects on some cell types. For instance, a chem-
albumin (HSA), human placenta collagen type 1 ically defined lipid mixture improved proliferation in
(COL), fibronectin (FN), and fibrinogen (FIB). LX-2, HMEC-1, and THP-1 cells but induced detach-
FN and FIB significantly improved attachment of ment and morphological deterioration in HepaRG
HepaRG and HMEC-1 cells compared to other coat- cells. Epidermal growth factor supported prolifera-
ings. LX-2 cells, which require minimal attachment tion across all four cell types but led to an aber-
conditions, showed selective enhancement with FN. rant spindle-like morphology in HepaRG cells (figure
Based on these results, FN at 1.5 µg ml−1 was selec- S4); it was therefore excluded from the final formu-
ted for inclusion in the final XFM formulation due to lation. Additives that had no measurable benefit or
its consistent performance across all three adherent demonstrated inconsistent results were also removed,
cell types (figure S1). PEI and HSA exhibited inferior with the exception of penicillin/streptomycin (P/S),
performance, and FIB was excluded due to lot-to- which showed neutral effects on proliferation but
lot variability and observed precipitation at working was retained for antimicrobial protection. This select-
concentrations. ive strategy ensured that only components with uni-
In contrast, THP-1 cells cultured under serum- formly beneficial or neutral effects across all cell types
free conditions exhibited an unintended shift from were included in the final formulation. Importantly,
suspension to sedimentation and aggregation at the the formulation avoided any factors that could com-
bottom of the culture vessel (figure S2). To address promise phenotype or viability in a single cell type,
this, we investigated multiple approaches to maintain thereby enabling a unified animal-free culture system
suspension behavior under animal-free conditions. suitable for multicellular liver modeling.
Cells were cultured in William’s E medium supple-
mented with insulin-transferrin-selenium (ITS), with 2.2. Growth kinetics in XFM
or without 10% FBS, 2% human serum (HS), HSA, To assess the proliferative capacity of individual
or plating on anti-adherence-treated culture dishes. cell types in the optimized XFM, we monitored
FBS fully prevented aggregation, while 2% HS yielded their growth over a 6 d period. As shown in
similar suspension quality without animal-derived figure 2(a), all adherent cell types exhibited a con-
components. HSA alone slightly reduced aggregation, tinuous and robust increase in cell number, indic-
and anti-adherence coatings reduced cell-surface ating that the XFM supports active proliferation in
interactions but promoted the formation of larger the absence of serum. Among the adherent lines,
aggregates. Only 2% HS maintained THP-1 cells in a HepaRG cells demonstrated the most pronounced
dispersed, suspended state comparable to FBS-based expansion, increasing over 10 folds from an initial
controls. Therefore, 2% HS was selected for THP- seeding density of 10 000 cells cm−2 to approxim-
1 expansion and differentiation under animal-free ately 110 000 cells cm−2 by day 6. LX-2 and HMEC-1
conditions (figure S2). cells also proliferated steadily, with HMEC-1 show-
Afterwards, we conducted a systematic screening ing a slightly higher growth rate than LX-2. Both cell
of media components to identify a formulation that types expanded to approximately four- to sixfold their
was broadly supportive across all four cell types. Each original seeding densities by the end of the culture

3
Biofabrication 17 (2025) 045022 A S M Ali et al

Figure 2. Characterization and functional validation of four liver-relevant human cell types cultured in xeno-free medium
(XFM). (a) Growth kinetics of adherent cells (HepaRG, LX-2, HMEC-1) over 6 d in XFM, presented as cells cm2 . (b) Growth
curve of THP-1 suspension cells in XFM, expressed as cells ml−1 . (c) Composition of the chemically defined freezing medium
(CDFM). (d) Post-thaw viability of all four cell types cryopreserved in CDFM, showing survival rates ranging from 85% to 95%.
(e) Immunofluorescence staining of actin filaments (phalloidin, green) and nuclei (DAPI, blue) in XFM (left side), showing pre-
served morphology compared to cells cultured in FBS-based media (right side). (f) Lineage-specific marker expression: vimentin
in LX-2, VE-cadherin in HMEC-1, and CD68 in PMA-differentiated THP-1 macrophages. (g) Basal gene expression of CYP1A2,
CYP2C9, CYP3A4, CYP7A1, and ALB in HepaRG cultured and differentiated in XFM versus FBS-based medium. (h) CYP induc-
tion profile following 72 h rifampicin treatment in XFM and FBS-medium normalized to untreated cells, demonstrating signi-
ficant upregulation of CYP3A4. (i) Endothelial tube formation by HMEC-1 cells on Matrigel, visualized via calcein AM staining.
(j) Phagocytic activity of differentiated THP-1 cells, indicated by uptake of fluorescent microbeads, the control (CTL) cells are
differentiated cells in XFM without incubation with microbeads. Statistical comparisons were performed using unpaired two-
tailed t-tests; p < 0.01 (∗∗ ), p < 0.0001 (∗∗∗∗ ), ns = not significant. Scale bars: 20 µm (e, f, j); 100 µm (i). Data are presented as
mean + standard deviation (SD). (n = 3).

period, maintaining normal morphology through- number increased from 10 000 cells ml−1 to over
out. For THP-1 cells cultured in suspension, XFM 30 000 cells ml−1 within 6 d, corresponding to more
supplemented with 2% HS also supported sustained than a threefold expansion, consistent with expec-
proliferation. As shown in figure 2(b), THP-1 cell ted behavior under FBS-based conditions. These data

4
Biofabrication 17 (2025) 045022 A S M Ali et al

confirm that the XFM formulation enables stable with a cobblestone-like arrangement and fine actin
and cell-type-appropriate growth kinetics across all filaments outlining intercellular junctions, reflect-
four human liver-relevant cell lines. A comparat- ing typical endothelial cell behavior. These observa-
ive analysis of growth kinetics in XFM versus FBS- tions confirm that XFM supports the maintenance of
based media is provided in supporting information structural phenotype and cytoskeletal integrity across
(figure S6). multiple liver-associated cell types in the absence of
To evaluate the long-term suitability of the XFM, animal-derived supplements.
we assessed cell proliferation across five consecut- To confirm lineage-specific marker expression
ive passages. As shown in the supporting informa- with XFM, immunofluorescence staining was per-
tion (figure S7), all three adherent cell types HepaRG, formed for representative phenotypic markers. As
LX-2, and HMEC-1 exhibited stable expansion over shown in figure 2(f), LX-2 cells cultured in XFM
the passage series. THP-1 cells also maintained pro- exhibited strong cytoplasmic expression of vimentin,
liferative capacity, with only a slight decrease at pas- consistent with their mesenchymal identity and
sage three. Although overall cell proliferation in XFM activated hepatic stellate phenotype. HMEC-1 cells
was lower compared to traditional FBS-based media, demonstrated robust membrane-associated VE-
the trends were consistent. Relative to FBS condi- Cadherin staining, indicating intact endothelial junc-
tions, HepaRG and HMEC-1 cultured in the XFM tions and preserved vascular identity. Differentiated
maintained approximately 65% of relative cell count, THP-1 macrophages expressed the monocyte/mac-
while THP-1 and LX-2 retained 80%–90% of their rophage marker CD68 with characteristic perinuclear
FBS-based proliferation rates. These results confirm and vesicular localization, indicative of activation and
the ability of the XFM to sustain stable expansion phagocytic potential.
of all four cell types over multiple passages under
animal-free conditions.
To evaluate the feasibility of fully animal-free 2.4. Functional characterization
cell banking, we tested a chemically defined freez- To assess hepatic functionality under animal-free
ing medium (CDFM) previously developed for Huh- conditions, we analyzed the expression of key liver-
7 cells [27], consisting of XFM supplemented with specific genes in differentiated HepaRG cells cultured
dimethyl sulfoxide (DMSO) and Pluronic F-68 as in XFM relative to FBS-based medium. As shown in
cryoprotectants (figure 2(c)). All four cell types figure 2(g), differentiated HepaRG cells maintained in
were cryopreserved in CDFM for 7 d. As shown XFM demonstrated significantly higher basal mRNA
in figure 2(d), post-thaw viability was consistently expression of ALB and major cytochrome P450
high across all cell types, ranging from 85% to 95%. enzymes CYP2C9, CYP3A4, and CYP7A1, compared
Comparative analysis with conventional FBS-based to those cultured in FBS-based medium. Notably,
freezing medium revealed no appreciable superiority CYP3A4 expression was enhanced nearly 14-fold
of the serum-based protocol. A slight improvement in XFM, while CYP2C9 and CYP7A1 also showed
in survival was observed for HMEC-1 cells under FBS marked upregulation. CYP1A2 expression remained
conditions (figure S8). These results confirm that the comparable between the two conditions. To further
combination of XFM and CDFM enables effective, evaluate metabolic inducibility, cells were treated
serum-free cryopreservation of all four liver-relevant with rifampicin (10 µm, 72 h), a prototypical inducer
cell types, thereby supporting animal-free workflows of CYP3A4 via the pregnane X receptor. As shown
from cell banking to downstream applications. in figure 2(h), rifampicin treatment led to a sig-
nificant increase in CYP3A4 expression in differ-
2.3. Morphological evaluation and cell identity entiated HepaRG cells in both media, with higher
To evaluate morphological characteristics and cyto- induction observed in cells cultured in FBS-based
skeletal organization under chemically defined con- medium. However, the absolute expression level of
ditions, the three adherent cell types, HepaRG, LX-2, CYP3A4 remained higher in cells cultured in XFM.
and HMEC-1, were cultured in XFM and stained These findings demonstrate that XFM supports not
with DAPI (nuclear marker) and phalloidin (F-actin only the basal hepatic function of HepaRG cells but
marker). As shown in figure 2(e), all three cell also preserves drug-inducible expression of CYP3A4,
types maintained their characteristic morphologies an essential feature for modeling hepatic drug
and exhibited well-organized actin cytoskeletal struc- metabolism in vitro.
tures. HepaRG cells displayed a polygonal epithelial- To evaluate endothelial function, HMEC-1 cells
like morphology with prominent cortical actin loc- were seeded on Matrigel-coated wells in XFM and
alized along the cell borders and well-defined nuc- assessed for tube formation after 18 h. Cells were
lei, indicative of epithelial polarity. LX-2 cells exhib- stained with calcein AM and visualized by fluores-
ited an elongated, fibroblast-like morphology with cence microscopy. As shown in figure 2(i), HMEC-1
dense cytoplasmic actin stress fibers aligned along the cells formed an extensive network of interconnected
major cellular axis, consistent with a hepatic stellate tube-like structures characteristic of early angiogenic
phenotype. HMEC-1 cells formed a dense monolayer organization. The formation of multicellular cords

5
Biofabrication 17 (2025) 045022 A S M Ali et al

and junctions suggests preserved endothelial differ- 2.6. Optimization of bioink composition and
entiation and functionality under serum-free, chem- photoinitiator concentration
ically defined conditions. These results further valid- A critical step in establishing a robust 3D bioprinted
ate the suitability of XFM for maintaining vascular liver model was the development of a fully animal-
phenotypes in vitro. free, cell-compatible bioink that supports all four
Furthermore, we assessed the phagocytic capacity liver-relevant human cell types in co-culture. The
of THP-1 cells differentiated into macrophage-like optimized bioink formulation consisted of sodium
cells using PMA in XFM supplemented with 2% HS. hyaluronate (HA, 6 mg ml−1 ), methacrylated hyalur-
Differentiated cells were incubated with red fluores- onic acid (HAMA, 12 mg ml−1 ), FIB (2.5 mg ml−1 ),
cent latex beads for 24 h and visualized using fluor- and human placenta collagen type I (50 µg ml−1 ),
escence microscopy. As shown in figure 2(j), THP-1- forming a composite matrix that mimics key fea-
derived macrophages cultured in animal-free condi- tures of the human extracellular matrix (ECM)
tions exhibited pronounced bead uptake, with wide- and enables stabilization via photo-polymerization
spread internalization of fluorescent particles across and enzymatic crosslinking. Lithium phenyl-2,4,6-
the cell population. Beads were visible in perinuc- trimethylbenzoylphosphinate (LAP) was selected as
lear vesicular compartments, consistent with act- the photoinitiator due to its low cytotoxicity and
ive phagocytosis. These results confirm that differ- rapid reactivity under light exposure, allowing pre-
entiated THP-1 cells retain functional phagocytic cise crosslinking with minimal thermal or phototoxic
capacity when cultured and differentiated under stress. The cells were encapsulated at a high density
animal-free conditions. of 20 × 106 cells ml−1 using a defined physiological
ratio (HepaRG:LX-2:HMEC-1:THP-1 at 65:10:15:10)
to recapitulate the cellular complexity of native liver
2.5. Design and fabrication of the micropillar tissue (figure 4(a)).
culture platform To identify the optimal LAP concentration bal-
To enable high-throughput and miniaturized biofab- ancing print fidelity and cytocompatibility, we sys-
rication of liver microtissues, we developed a custom tematically tested a range of LAP concentrations
modular micropillar platform geometrically aligned (0.10%–0.50%, w/v) using a live/dead viability assay
with the standard 96-well plate format. The plat- on cultured HepaRG cell in monolayer 2D culture
form consists of a slotted rack base housing twelve (figure S9). Increasing LAP concentrations resul-
removable comb-like modules, each bearing eight ted in a concentration-dependent rise in ethidium-
vertically oriented tines tipped with shallow con- positive (dead) cells, particularly above 0.30%, indic-
cave discs engineered to retain 3 µl bioink droplets. ating photoinitiator-related toxicity. At 0.10%, cell
The micropillar modules were fabricated using high- viability was comparable to the non-treated control,
resolution liquid crystal display (LCD)-based 3D indicating minimal phototoxic stress. Based on these
printing (Original Prusa SL1S) with a transparent results, 0.10% LAP was selected as the optimal con-
photopolymer resin, providing fine structural defin- centration for all subsequent bioprinting, providing
ition (∼25–50 µm) and optical clarity for imaging sufficient crosslinking while preserving high cellular
through the well bottom. The rack itself was prin- viability across all cell types.
ted via fused deposition modeling (FDM) using poly- Uniform cell distribution within the bioink prior
lactic acid (PLA) or polyethylene terephthalate glycol to printing is critical for reproducible tissue architec-
(PETG, only the upper springs were printed with ture and biological readouts. To evaluate sediment-
transparent PETG to reduce the impact of block- ation and homogeneity, GFP-expressing HEK cells
ing light under microscope) and assembled from five were embedded in the complete bioink formulation
parts to avoid support artifacts and ensure robust and monitored in vertically placed syringes over a
alignment. When inserted into a 96-well plate, the 2 h period (figure 4(b)). Fluorescence intensity across
tines descend centrally into each well, allowing micro- defined height segments remained consistent, with
tissues to form in an ‘inverted’ configuration, sus- no visible sedimentation or layering after including
pended above the well bottom. This architecture HA with high molecular weight, confirming adequate
offers multiple functional advantages, including com- hydrogel viscosity and rheological stability to main-
patibility with automated brightfield and fluores- tain homogenous cell suspensions during multi-well
cence imaging, minimal meniscus distortion, reduced plate printing. This homogeneity window comfort-
reagent volumes, and simplified media exchange by ably exceeds the time required to print an entire
decoupling tissue supports from the well plate. As 96-well plate, as demonstrated in the accompanying
shown in figure 3, the micropillar platform sup- video (Supporting Video S1), which shows repro-
ports uniform construct formation, is reusable fol- ducible droplet placement. A representative photo-
lowing sterilization, and is fully compatible with auto- graph of the microliver model atop of micropillar is
mated handling systems and high-content screening denoted in figure 4(c). To confirm retention of differ-
workflows. entiated THP-1 macrophage-like cells in the bioink,

6
Biofabrication 17 (2025) 045022 A S M Ali et al

Figure 3. Design and fabrication of the micropillar-based bioprinting system. (a) 3D-printed modular micropillar platform:
assembled views showing slotted rack, micropillar strips with concave tips, and full plate compatibility with 96-well plate format.
(b) Bioprinting process: cell-laden bioink extruded onto micropillars, followed by photocrosslinking and fibrin-mediated stabiliz-
ation. The platform is optimized for media exchange, reproducibility, and automated imaging.

monoculture constructs were printed and analyzed increased progressively over time, indicating sus-
after 24 h. Live/dead staining demonstrated high cell tained biosynthetic activity. Similarly, urea produc-
viability and homogeneous distribution within the tion (figure 4(g)), a key marker of nitrogen meta-
hydrogel (figure S10), supporting stable incorpora- bolism, also rose cumulatively over the 7 d culture
tion of THP-1 cells into the LoM platform. period.
Following bioprinting and crosslinking,
microliver constructs were cultured for up to 15 d 2.7. Drug-induced hepatotoxicity profiling using
and evaluated for cell viability, morphology, and ATP, XTT, and albumin secretion assays
hepatic functionality. As shown in figure 4(d), XTT To evaluate the predictive performance of the LoM
assay results confirmed that metabolic activity was system for DILI, we tested a panel of ten reference
maintained over time, reflecting sustained cell viab- drugs representing a spectrum of clinical hepatotox-
ility and mitochondrial function. Live/dead stain- icity profiles, as defined by the FDA’s DILIrank classi-
ing revealed excellent viability through day 7, with fication (table 1). Each compound was tested across
predominantly calcein-positive (viable) cells form- a 7-point concentration range up to 100× human
ing compact, hemispherical tissues (figure 4(e)). By Cmax (or solubility limit) in three independent bio-
day 15, modest ethidium homodimer staining was logical replicates. Hepatotoxicity was quantified by
observed at the construct periphery, likely due lim- assessing cellular ATP content, mitochondrial activity
itation of non-perfused hydrogel systems, which are (XTT assay), and functional protein secretion (albu-
known constraints in static 3D cultures [28]. min enzyme-linked immunosorbent assay, ELISA)
Functional performance of the microliver con- after 72 h of drug exposure.
structs was assessed by quantifying hepatocyte- Representative concentration-response curves for
specific outputs. Albumin secretion (figure 4(f)) three compounds, ibufenac (high DILI concern,

7
Biofabrication 17 (2025) 045022 A S M Ali et al

Figure 4. Bioink optimization and functional validation of bioprinted Liver-on-Micropillar (LoM) constructs. (a) Composition
of the human-compatible bioink used for 3D bioprinting. (b) Sedimentation assessment of GFP-expressing HEK cells embed-
ded in static bioink over time, confirming bioink homogeneity and minimal cell sedimentation. (c) Photograph of representative
bioprinted microliver construct atop micropillar array. (d) Metabolic activity of LoM constructs over 15 d quantified by XTT
assay. (e) Viability of constructs visualized via calcein AM/ethidium homodimer-1 staining over 15 d. (f) Cumulative albumin
secretion and (g) normalized cumulative urea production by LoM constructs over 7 d, demonstrating preserved hepatic function-
ality. Data are presented as mean + standard deviation (SD). Statistical significance: ns = not significant, ∗ P < 0.05, ∗∗ P < 0.01,
∗∗∗ P < 0.001, ∗∗∗∗ P < 0.0001 (n = 3). Scale bar = 200 µm.

withdrawn), busulfan (high DILI concern, clinical use IC50 values for all ten compounds, along with
with hepatotoxic risk), and lidocaine (no DILI con- their corresponding Cmax values, were used to cal-
cern), are shown in figures 5(a)–(c). Ibufenac exhib- culate MOS scores for each assay. Drugs with
ited strong concentration-dependent toxicity across MOS < 100 were considered potentially hepatotoxic
all assays, with ATP and albumin readouts showing in vitro [30, 31] and based on the best prediction
particularly steep declines. Busulfan induced mod- accuracy at different MOS cutoffs (table S2), as a
erate cytotoxicity and functional suppression, with 100-fold difference between in vitro IC50 and clinical
greater sensitivity detected in albumin secretion com- plasma concentrations provides a safety margin that
pared to XTT. In contrast, lidocaine, a drug with well- minimizes false negatives in hepatotoxicity predic-
established hepatic safety, showed minimal reduction tion. As summarized in table 1, the ATP assay yielded
in any assay even at supraphysiological concentra- correct classification for 9 of 10 drugs (90%), missing
tions. The concentration-dependent toxicity curves only trimethoprim (false negative). Albumin-based
for the 10 drugs are shown in figures S11–S13 for XTT, assessment also correctly predicted 9 of 10 drugs,
ATP and Albumin assays, respectively. misclassifying neostigmine (false positive). The XTT

8
Biofabrication 17 (2025) 045022 A S M Ali et al

Figure 5. Drug-induced hepatotoxicity assessment in the LoM platform using ATP, XTT, and albumin assays. (a)–(c)
Representative concentration-response curves for busulfan, lidocaine, and ibufenac using ATP, XTT, and albumin assays. IC50
values were calculated for each readout. (d) Summary of Margins of Safety (MOS = IC50 /Cmax ) for ten test compounds. Albumin
and ATP assays demonstrated the highest sensitivity and correctly classified 9/10 compounds at a MOS threshold of 100. The red
dashed line indicates an MOS cutoff of 100. Data are presented as mean ± standard deviation (SD) (n = 3).

assay was slightly less sensitive, correctly classifying in figure 5(d). To further validate the classification
8 of 10 drugs and generating two false negatives outcomes, we listed published DILI annotation sys-
(trimethoprim, busulfan). tems of Sakatis et al [32], Xu et al [33], and Zhu
As shown in table 1, drugs known to cause severe and Kruhlak [34] in addition to the FDA’s DILIrank
or idiosyncratic liver injury in humans (e.g. ben- dataset [29] of the 10 drugs in supplementary table
zbromarone, lumiracoxib, azathioprine, busulfan) S3. Overall, the LoM predictions were consistent
consistently exhibited MOS values well below the across these complementary datasets, supporting the
threshold of 100 in at least two assays. In contrast, robustness of the platform’s predictive performance.
drugs with no reported DILI concern (e.g. lidocaine, Collectively, these results underscore the import-
neostigmine) generally exceeded the safety threshold ance of using complementary functional and viability
across all assays, confirming the specificity of the endpoints in complex liver models. Notably, integrat-
system. A summary of MOS values is delineated ing multiple assays improves detection sensitivity for

9
 Biofabrication 17 (2025) 045022
Table 1. List of ten reference compounds used to evaluate hepatotoxicity in the Liver-on-Micropillar (LoM) platform. Drugs were selected to span a spectrum of liver injury risk based on the FDA’s DILIrank system, vDILI-Concern
classification, and regulatory labeling [29]. Margin of Safety (MOS) values were calculated for each drug by dividing the IC50 values obtained from ATP, XTT, and albumin assays by the corresponding human Cmax . A MOS threshold
of 100 was used for DILI risk classification. False negatives refer to misclassified DILI drugs, and false positives refer to drugs incorrectly flagged as toxic.

MOS Predicted False negative False positive

Drugs Severity class vDILI concern Label ATP XTT ALB ATP XTT ALB ATP XTT ALB ATP XTT ALB
10

Benzbromaron 8 Most-DILI-Concern Withdrawn 0.3 0.1 0.2 4 4 4

Ibufenac 8 Most-DILI-Concern Withdrawn 15.6 12.9 3.6 4 4 4
Trimethoprim 4 Less-DILI-Concern Adverse reactions 84.7 128.3 19.4 4 4 4
Lidocaine 0 No-DILI-Concern No match 114.7 105.6 21.1 4 4 4
Neostigmine 0 No reported toxicity No match 540 291.1 231.1 4 4 4
Azathioprine 5 Most-DILI-Concern Warnings and precautions 6.8 17.1 6.6 4 4 4
Busulfan 8 Most-DILI-Concern Warnings and precautions 149 179.2 29.8 4 4 4
Carbamazepine 7 Most-DILI-Concern Warnings and precautions 9.9 11.4 2.6 4 4 4
Dasatinib 4 Less-DILI-Concern Adverse reactions 52.5 55.1 26.3 4 4 4
Lumiracoxib 8 Most-DILI-Concern Withdrawn 14.3 14.6 10.5 4 4 4

A S M Ali et al
Biofabrication 17 (2025) 045022 A S M Ali et al

both cytotoxic and subcytotoxic hepatotoxic effects, a high-throughput front-end screen for identifying
providing a more nuanced toxicity profile. The DILI using a mechanistically diverse panel of clinic-
LoM platform’s predictive accuracy, especially using ally annotated compounds. By combining three dif-
ATP and albumin readouts, highlights its potential ferent assays, ATP content, XTT metabolic activ-
as a preclinical screening tool for early-stage DILI ity, and albumin secretion, we aimed to capture
risk stratification in an animal-free, high-throughput both cytotoxic and subcytotoxic hepatic responses,
format. and to calculate translationally relevant MOS using
clinical Cmax values as denominators. Among the
3. Discussion three endpoints, ATP content and albumin secre-
tion exhibited the highest predictive accuracy, cor-
In this study, we developed the LoM platform, a mini- rectly identifying 9 out of 10 compounds when apply-
aturized, bioprinted human liver model optimized ing a MOS threshold of 100. This aligns with recent
for high-throughput assessment of DILI. By integ- findings suggesting that ATP depletion is an early
rating an XFM, an animal-free ECM-based bioink, marker of mitochondrial and metabolic stress, and
and a four-cell-type co-culture, the LoM addresses that loss of albumin secretion reflects functional
major limitations of current hepatotoxicity models impairment prior to overt cell death [40–42]. In con-
related to scalability, cell-type diversity, and human trast, XTT, while a reliable marker of general meta-
relevance. The custom-engineered micropillar array bolic activity, showed reduced sensitivity in detect-
and slotted rack were designed to fit the 96-well plate ing moderate hepatotoxicity or drugs with primary
format, ensuring compatibility with common auto- effects on liver-specific functions rather than viabil-
mated instruments such as microplate readers and ity. The differential classification of compounds such
automated microscopy systems, and enabling seam- as busulfan and trimethoprim across assays under-
less integration into standard screening pipelines. scores the importance of multiplexed endpoints for
Validation with ten clinically annotated drugs nuanced toxicity profiling. Drugs with no reported
demonstrated the system’s predictive capability using DILI concern (e.g. lidocaine, neostigmine) showed
three different endpoints, ATP content, XTT meta- high MOS values across all assays, reinforcing the
bolic activity, and albumin secretion. Notably, both system’s specificity. Together, these results support
ATP and albumin assays correctly classified 9 of 10 the utility of the LoM platform for early-stage com-
compounds using a MOS threshold of 100, high- pound triaging, enabling both safety de-risking and
lighting their sensitivity and translational utility. mechanistic insight. High throughput bioprinting
This performance compares favorably to traditional platforms based on micropillar technologies have
monocultures and spheroid models, which often recently gained momentum in organotypic culture
lack multicellular context and rely on animal-derived and drug screening. Kang et al developed a pil-
components [35]. lar/perfusion plate system for the culture of liver
A key innovation lies in the formulation of a uni- and intestinal organoids derived from induced pluri-
fied XFM capable of supporting four phenotypic- potent stem cell progenitors embedded in Matrigel,
ally and metabolically distinct liver-relevant human demonstrating its potential for disease modeling
cell types, HepaRG, LX-2, HMEC-1, and THP-1- and pharmacological studies [43, 44]. In parallel,
derived macrophages. Achieving this without FBS researchers also developed other formats such as
is significant, given the historical reliance on FBS 144-pillar array [45] and 384-pillar array [46] to
to compensate for divergent cellular requirements. achieve high content ability for drug screening. While
However, FBS introduces variability and undefined these studies underscore the versatility of pillar-
components that limit reproducibility and trans- based formats for high-throughput applications, they
lational alignment [36, 37]. Our rational design require surface coating of pillars with compon-
included screening of defined supplements to main- ents like poly (maleic anhydride-alt-1-octadecene) or
tain both adherent and suspension phenotypes under alginate poly-l-lysine, and rely on Matrigel or tumor-
animal-free conditions. XFM supported robust pro- derived cell lines and do not address hepatic toxicity
liferation across five passages and maintained pheno- mechanisms. In contrast, our LoM platform does not
typic markers, including VE-cadherin, vimentin, and need to be pre-coated, and integrates a fully animal-
CD68 in their respective cell types. Differentiated free bioink, defined co-culture of four liver-relevant
HepaRG cells retained basal expression of ALB and human cell types, and a miniaturized format and
CYP genes, further induced by rifampicin treatment, large exposure area of the models specifically tailored
confirming functional competence without serum for DILI screening. This design not only enables
supplementation [38, 39]. Furthermore, cryopreser- sensitive detection of DILI-related cytotoxicity and
vation in CDFM resulted in high post-thaw viab- functional impairment but also aligns with cur-
ility, offering a practical alternative to FBS-based rent regulatory initiatives promoting the replacement
preservation methods. of animal testing with human-relevant, non-animal
A central objective of this study was to eval- methodologies in chemical safety and pharmaceutical
uate the predictive utility of the LoM platform as development.

11
Biofabrication 17 (2025) 045022 A S M Ali et al

While the LoM platform shows strong prom- pool of drug candidates for in-depth characterization
ise, several limitations remain. The current static using more complex dynamic or perfused systems.
culture format restricts nutrient and oxygen dif- Together, these features offer a practical solution
fusion, which may compromise long-term stabil- for advancing non-animal, human-relevant method-
ity. It is worth noted that the current micropillar ologies in preclinical safety evaluation, fully aligned
platform is ready to incorporate the dynamic cul- with current scientific, ethical, and regulatory trends,
ture conditions, such as placing the micropillar/- as reflected by recent FDA initiatives to replace animal
plate assembly under gentle rocking for low-shear testing in toxicity studies.
media exchange, or coupling the removable pil-
lar combs to a perfused lid/manifold. Future iter- 5. Methods
ations integrating dynamic perfusion could better
recapitulate hepatic microcirculation and zonation, 5.1. Cell lines and culture conditions
thereby enabling chronic exposure studies and more Four human cell lines were used to represent hepatic,
physiologically relevant drug delivery. Although our stromal, endothelial, and immune compartments of
drug panel encompassed compounds with diverse the liver: HepaRG (Biopredic International), LX-2
mechanisms and DILI risks, the addition of biomark- (SCC064, Merck Millipore), HMEC-1 (CRL-3243™,
ers such as oxidative stress, mitochondrial potential, ATCC), and THP-1 (ACC16, DSMZ). All cells were
or cytokine release in addition to cell type–specific cultured in animal component-free conditions unless
immunostaining—e.g. ALB, CD31, α-SMA/desmin, otherwise indicated. The culture conditions in FBS-
CD68—or nuclei-barcoded imaging to track com- based media were applied only for comparative ana-
position over time, would provide greater mechan- lysis during the development phase of the XFM and
istic insight. Busulfan exemplifies these limitations: are detailed in the Supplementary Methods S1. All
in clinical settings, its hepatotoxicity typically arises adherent cell lines, including HepaRG, LX-2, and
only after months to years of therapy or within 10– HMEC-1, were maintained in a XFM composed of
20 d following high-dose conditioning regimens, and William’s E Medium (12551-032, Gibco) supplemen-
manifests predominantly as vascular pathologies such ted with 2 mm L-glutamine (X0550, Biowest), 1×
as sinusoidal obstruction syndrome rather than acute ITS (41400045, Gibco), 5 µg ml−1 insulin (P07-
hepatocellular toxicity [47, 48]. Such delayed, non- 04300, PAN-Biotech), 50 µm hydrocortisone 21-
parenchymal mechanisms are not captured within the hemisuccinate sodium salt (H2270-100 MG, Sigma),
72 h hepatocyte-focused assay window, accounting 1% penicillin/streptomycin (P/S, L0022, Biowest),
for its false negative classification. While the LoM 1 mm sodium pyruvate (S8636-100ML, Sigma),
system is well suited for acute hepatotoxicity screen- 55 µm 2-mercaptoethanol (21985-023, Gibco), and
ing, chronic applications will require incorporation of 1.5 µg ml−1 human fibronectin (1918-FN, R&D
perfusion or microfluidic approaches to enhance con- Systems). Cells were seeded and passaged under
struct longevity and capture long-term or vascular- standard conditions at 37 ◦ C, 5% CO2 in a humidi-
mediated injury pathways. fied incubator. HepaRG cells were initially cultured in
XFM for 14 d with medium replacement every 2–3 d.
Following this initial culture period, cells were incub-
4. Conclusion ated for 24 h in XFM supplemented with 1% (DMSO,
D2650, Sigma), then maintained for an additional
In summary, we present a LoM platform as a repro- 14 d in XFM containing 1.7% DMSO to induce full
ducible, miniaturized, and animal-free human liver hepatic differentiation.
model tailored for high-throughput DILI screening. THP-1 cells were cultured as non-adherent sus-
By integrating an XFM, an animal-free ECM-based pensions in a modified animal-free medium com-
bioink, and a standardized micropillar array, the posed of XFM without FN (XFM- ) supplemented
LoM system supports co-culture of four hepatic cell with 2% HS (H4522, Sigma) and maintained under
types and enables multiplexed cytotoxicity and func- the same atmospheric conditions. For differentiation
tional readouts. The platform demonstrated high pre- into macrophage-like phenotype, THP-1 cells were
dictive accuracy across a panel of reference drugs, treated with 100 ng ml−1 phorbol 12-myristate 13-
particularly when using ATP and albumin-based acetate (PMA, 524400, Sigma) in XFM supplemented
assays for MOS calculations. Its compatibility with with 2% HS for 72 h, followed by a 24-h rest period
standard 96-well plate formats, coupled with fully in PMA-free medium prior to experimentation.
defined and animal-free components, positions LoM For routine subculturing of all adherent lines, cells
as a scalable and regulatory-aligned alternative to were passaged upon reaching 70%–90% confluency.
conventional liver models. Cultures were rinsed with Dulbecco’s phosphate-
As a primary screening tool, LoM is well suited buffered saline (DPBS, L0615, Biowest), followed
to rapidly triage large compound libraries, enabling by incubation with TrypLE Express (animal-free,
the early identification of hepatotoxic liabilities. This 12604021, Gibco) for 3–5 min to facilitate cell
helps focus downstream resources on a narrower detachment.

12
Biofabrication 17 (2025) 045022 A S M Ali et al

Details regarding the optimization of cell adhe- 50 µg ml−1 human placenta collagen type I (gen-
sion, THP-1 suspension maintenance, and screening erously provided by THT Biomaterials, Vienna,
of individual medium components are presented in Austria), 12 mg ml−1 methacrylated hyaluronic acid
the supporting methods S2, S3, and S4 respectively. (HAMA, 5212, Advanced Biomatrix, Cellink), and
lithium phenyl-2,4,6-trimethylbenzoylphosphinate
5.2. Design and fabrication of the micropillar plate (LAP, 900889, Sigma). Bioink preparation was per-
The micropillar plate was modeled using Creo formed using two 1 ml BD syringes (BD) to maintain
Parametric program. A micropillar strip is the basic component integrity prior to mixing.
unit of the micropillar plate, each features a comb-like The first syringe was loaded with 450 µl of 2%
shape with a handle and eight micro pillars. At the end (w/v) HAMA solution and 300 µl of 1.5% (w/v) HA
of each pillar is a small concave disc with a pattern of solution. The second syringe contained a cellular sus-
quartered circle of four outward ticks, facilitating pre- pension in a total of 250 µl composed of 125 µl
cise capture of a mini organ model during printing XFM containing 20 × 106 cells ml−1 (HepaRG:LX-
process and ensuring stable placement within the 96- 2:HMEC-1:THP-1 in a ratio of 65%:10%:15%:10%),
well plate in culture. Designed for the 96-well plate 50 µl of 1 mg ml−1 collagen type I solution, 25 µl
format, this configuration is compatible with high- of 4% (w/v) LAP photoinitiator in PBS, and 50 µl of
throughput measurements on standard automated 50 mg ml−1 FIB solution in ddH2 O. The two syringes
instruments, such as microplate readers, automated were connected using a female–female luer-lock con-
microscopes, and liquid-handling robots. nector and gently mixed before transferring the com-
The micropillar strips were inserted into a slotted posite bioink into a bioprinting cartridge.
rack, accommodating 12 micropillar strips arranged Bioprinting was performed using a multi-
in a fixed array with uniform spacing. The assembled printhead 3D Discovery bioprinter (regenHU, Villaz-
micropillar plate can be securely mounted onto a 96- St-Pierre, Switzerland). The modified cartridge (dia-
well plate, with the pillars protruding into the center meter 4.734 mm) was equipped with a 22 G needle
into each well. and mounted onto the syringe pump printhead.
To ensure the detailed structure of the concave Printing was conducted at RT using a plunger velo-
disc, the micropillar strips were produced using an city of 0.4 mm s−1 and a corresponding flow rate of
LCD-based 3D printer (Original Prusa SL1S, Prusa) 7.04 µl s−1 . A pressurize/retract distance of 0.05 mm
with a transparent resin (tough clear VALUE resin, was applied. Each micropillar construct was printed
PrimaCreator). After printing, the micropillar strips with a bioink volume of 3 µl and immediately photo-
were washed with 96% ethanol and cured for 20 min crosslinked with light for 1 s (wavelength: 365 nm).
in the curing and washing machine (CW1S, Prusa). A total of 1 ml of bioink was sufficient to fabricate
To remove the residual unreacted resin, the micropil- approximately three 96-well plates.
lar strips were then immersed in 96% ethanol at room Post-printing, the micropillar constructs were
temperature (RT) for 90 min, followed by immersion incubated at RT in 100 µl XFM supplemented with
in distilled water at 37 ◦ C for 24 h. 0.5 U ml−1 human thrombin for 15 min to crosslink
In contrast, to avoid support structure and the FIB component. After incubation, the thrombin-
achieve a smooth surface, the slotted rack was divided containing medium was replaced with thrombin-free
into 5 parts for printing using a FDM 3D printer (X1 XFM (200 µl well−1 ), and constructs were cultured
Carbon, Bambu Lab) with PLA or clear PETG mater- under standard conditions at 37 ◦ C with 5% CO2 .
ial (DAS FILAMENT). These 5 parts, comprising a Medium changes were performed every three days.
total of 27 pieces, were subsequently assembled and
covered with a transparent adhesive film (Bio-Rad). 5.4. Functional characterization of the microliver
Detailed description of the assembling way is shown models
in figure S14. 5.4.1. Cell viability by XTT assay
Subsequently, the micropillar plate was assembled The cell viability of cultured microliver mod-
by inserting 12 micropillar strips into each slotted els was assessed using the XTT assay (2,3-bis-(2-
rack. The assembled micropillar plates were then methoxy-4-nitro-5-sulfophenyl)-2 H-tetrazolium-
further disinfected with UV radiation for 30 min 5-carboxanilide, J61726, Alfa Aesar) at the indic-
in a biosafety cabinet. The sterile micropillar plates ated time points. XTT stock solution (1 mg ml−1 )
were used for bioprinting immediately or stored for was diluted 1:3 in XFM without phenol red, and
later use. The 3D designs files are supplemented phenazine methosulfate (PMS, 3.83 mg ml−1 ;
(STL_files.zip). A2212005, AppliChem) was added to the mixture
at a volume ratio of 1:500 (PMS solution: final
5.3. Biofabrication of the microliver models volume). The resulting reaction solution was applied
The 3D liver model was biofabricated using a to the cell cultures and incubated for 4 h at 37 ◦ C.
custom-formulated, animal-free bioink. The bioink Absorbance was then measured at 450 nm with a
is composed of 6 mg ml−1 sodium hyaluronate reference wavelength of 620 nm using a microplate
(HA, 251770050, ThermoFisher), 2.5 mg ml−1 FIB, reader (Sunrise, Tecan, Männedorf, Switzerland).

13
Biofabrication 17 (2025) 045022 A S M Ali et al

5.4.2. Cell viability by live/dead staining reader (Sunrise, Tecan). A standard curve was gen-
Additional to the XTT assay, cell viability of the erated using the supplied urea nitrogen standard at
microliver models was assessed qualitatively using concentrations of 5, 2.5, 1.25, 0.625, 0.3125, 0.156,
live/dead staining with calcein AM (ABD-22002, and 0 mg dl−1 . Urea concentrations in the samples
AAT Bioquest, Biomol) and ethidium homodimer- were calculated based on this curve and expressed
1 (CDX-E0012-M001, Biomol), which respectively in mg dl−1 . Each condition was assessed in three
label viable and non-viable cells. The micropillars biological replicates.
combs were transferred to another 96-well plate and
incubated in 100 µl phenol red-free XFM supple- 5.4.5. Hepatotoxicity testing
mented with 2 µm calcein AM and 2 µm ethidium A panel of ten drugs with varying degrees of DILI
homodimer-1 for 30 min at 37 ◦ C. Following incuba- severity was selected based on the FDA DILIRank
tion, samples were images by fluorescence microscopy classification and associated clinical hepatotox-
(Observer Z1, Zeiss), with green fluorescence indicat- icity data. The compounds included benzbroma-
ing viable cells and red fluorescence indicating dead rone (B5774-1 G, Sigma), ibufenac (SML1452,
cells. Sigma), trimethoprim (92131-1 G, Sigma), lido-
caine (PHR1034-1 G, Sigma), neostigmine methyl
5.4.3. Albumin secretion assay sulfate (N2126-1 G, Sigma), azathioprine (PHR1282,
Human albumin levels in the culture supernatants Sigma), busulfan (B2635, Sigma), carbamazepine
were quantified using the Human Albumin ELISA Kit (C4024-1 G, Sigma), dasatinib (SML2589-50 MG,
(E88-129, Bethyl Laboratories), following the man- Sigma), and lumiracoxib (SML2928, Sigma). These
ufacturer’s instructions. Briefly, culture supernatants drugs were chosen to represent a broad spectrum of
were collected and stored at −20 ◦ C until further DILI concern and severity classes, ranging from no-
analysis. Prior to the assay, samples were diluted in DILI-concern (e.g. lidocaine) to most-DILI-concern
sample diluent to fall within the dynamic range of the drugs that have been withdrawn due to hepatotox-
standard curve. Samples and standards were loaded icity (e.g. benzbromarone, ibufenac, lumiracoxib).
into wells, pre-coated with anti-human albumin cap- Classification of the tested drugs is depicted in table 1.
ture antibody, and incubated for 1 h at RT. Following All compounds were dissolved in DMSO at 100×
washing, a horseradish peroxidase -conjugated detec- their highest working concentration or at their max-
tion antibody was added, and plates were incub- imal solubility in DMSO if the solubility limit was
ated for an additional 1 h. After final washes, the below 100×. The highest working concentration for
TMB substrate solution was added to initiate color each drug was based on 100 times the reported max-
development. The reaction was stopped with 0.18 N imum plasma concentration (Cmax ) in humans, as
sulfuric acid after the 30 min incubation period, compiled from literature (table S4). Serial 1:2 dilu-
and absorbance was measured at 450 nm using a tions were prepared in XFM to yield seven concentra-
microplate reader (Sunrise, Tecan). Albumin con- tions per drug (table S5). Untreated controls received
centrations were calculated based on the standard XFM containing the same final DMSO concentration
curve generated using serial dilutions of the provided as the drug-treated conditions.
human albumin standard. Results were expressed Following 48 h of stabilization post-bioprinting,
as ng ml−1 . the micropillar liver models were transferred into a
new 96-well plate containing 150 µl of the respect-
5.4.4. Urea secretion assay ive drug dilutions or control medium. The cul-
Urea secretion was evaluated as a functional marker tures were maintained under standard conditions
of hepatic metabolism in the microliver models using (37 ◦ C, 5% CO2 ).
the urea nitrogen (BUN) Colorimetric Detection Hepatotoxicity was assessed following 72 h drug
Kit (EIABUN, Thermo Fisher Scientific), according exposure using three complementary endpoints:
to the manufacturer’s instructions. Culture super- mitochondrial metabolic activity (XTT assay), intra-
natants were collected on days 0, 1, 3, 5, and 7, cellular ATP content (ATP assay), and hepatic func-
and stored at −20 ◦ C until analysis. On the day tion via albumin secretion (ELISA). These assays
of analysis, samples were thawed on ice and used provided a multiparametric evaluation of cytotoxic
without dilution. In a 96-well flat-bottom plate, 50 µl and functional responses in the microliver models.
of each sample or standard was pipetted per well. The half-maximal IC50 was determined for each com-
To each well, 75 µl of Color Reagent A and 75 µl pound using nonlinear regression, and the MOS was
of Color Reagent B were sequentially added. The calculated by dividing the IC50 by the corresponding
plate was mixed gently and incubated for 30 min at human maximum plasma concentration (Cmax ).
RT, protected from light. After incubation, absorb- Intracellular ATP content was quantified using
ance was measured at 450 nm using a microplate the CellTiter-Glo® Luminescent Cell Viability Assay

14
Biofabrication 17 (2025) 045022 A S M Ali et al

(G7571, Promega). This assay served as a primary Data availability statement

viability readout for IC50 determination and DILI risk
classification. Prior to use, both the CellTiter-Glo® All data that support the findings of this study are
Buffer and lyophilized Substrate were equilibrated included within the article (and any supplementary
to RT and combined by reconstituting the Substrate files).
with 10 ml of Buffer. The reagent was gently mixed Supplementary Information available at
to ensure complete dissolution. On day 3 of drug [Link]
treatment, LoM constructs cultured in 96-well plates Supporting Video S1.mp4 available at
were equilibrated to RT for approximately 30 min. [Link]
From each well, 50 µl of medium were transferred Micropillar CAD Files available at [Link]
to a white F-bottom 96-well plate (655075, Greiner), 10.1088/1758-5090/ae1063/data3.
followed by the addition of 50 µl of CellTiter-Glo®
reagent. The corresponding LoM construct was then Acknowledgment
transferred into the same well, ensuring full immer-
sion in the medium-reagent mixture. Plates were agit- We are thankful for the financial support of the
ated for 2 min on an orbital shaker to facilitate cell research project by the Einstein Foundation Berlin
lysis, followed by a 10 min incubation at RT to stabil- (Einstein Center 3R, EZ-2020-597-2). We gratefully
ize the luminescent signal. Luminescence was meas- acknowledge THT Biomaterials (Vienna, Austria) for
ured using a Tristar 5 microplate reader (Berthold the generous gift of human placenta collagen type I.
Technologies). Background luminescence from con-
trol wells (medium + reagent without cells) was sub- Conflict of interest
tracted. Each experimental condition was evaluated
in three biological replicates. Resulting luminescence The authors declare no competing interests.
values were normalized to the vehicle control and
used for IC50 curve fitting. Declaration of generative AI and
Mitochondrial metabolic activity was assessed AI-assisted technologies in the writing
using the XTT assay, and hepatic albumin secre- process
tion was quantified using a human albumin ELISA
kit, as described in detail previously. Concentration- During the preparation of this work the authors used
response curves were fitted using nonlinear regres- ChatGPT in order to improve language and readab-
sion (log[inhibitor] vs. response—variable slope) in ility. After using this tool, the authors reviewed and
GraphPad Prism v8.0 (GraphPad Software). IC50 val- edited the content as needed and took full responsib-
ues were calculated from the fitted curves. To eval- ility for the content of the publication.
uate translational relevance, MOS values were com-
puted as IC50 /Cmax for each drug and assay. An
Author contributions
MOS ⩽ 100 was used as a threshold to flag com-
Ahmed S M Ali  0000-0002-0001-8222
pounds as potentially hepatotoxic, in line with pre-
Conceptualization (equal), Data curation (equal),
viously published criteria [30, 31], and based on
Investigation (equal), Methodology (equal), Writing
the accuracy of prediction at different MOS cutoffs
– original draft (equal), Writing – review &
(table S2).
editing (equal)
Dongwei Wu  0000-0001-5446-7376
5.4.6. Statistical analysis
Conceptualization (equal), Data curation (equal),
All experiments were conducted in biological triplic-
Investigation (equal), Methodology (equal), Writing
ates (n = 3) unless otherwise stated. Data are presen-
– original draft (equal), Writing – review &
ted as mean ± standard deviation (SD). Statistical
editing (equal)
significance was evaluated using one-way analysis
of variance followed by Dunnett’s post hoc test to Viola Roehrs
compare drug-treated groups with their correspond- Methodology (supporting), Writing – review &
ing vehicle DMSO controls. A p-value less than 0.05 editing (equal)
was considered statistically significant, with levels
Duy Bao Tran Nguyen  0009-0003-1490-1766
of significance indicated as follows: p < 0.05 (∗ ),
Methodology (supporting), Writing – review &
p < 0.01 (∗∗ ), p < 0.001 (∗∗∗ ), and p < 0.0001 (∗∗∗∗ ).
editing (equal)
All analyses were performed under the assumption
of normal distribution and equal variance, and no Mathias Ziersch
data were excluded unless attributable to technical Methodology (supporting), Writing – review &
error. editing (equal)

15
Biofabrication 17 (2025) 045022 A S M Ali et al

Albert Braeuning [17] Zuang V et al Non-Animal Methods in Science and

Conceptualization (equal), Funding Regulation—EURL ECVAM status report 2023
(Publications Office of the European Union2024) (available
acquisition (equal), Writing – review &
at: [Link]
editing (equal) JRC136460)
[18] FDA predictive toxicology roadmap (FDA, 2017) (U.S. Food
Jens Kurreck  0000-0002-1469-0052 and Drug Administration) (avaiable at: [Link]
Conceptualization (equal), Funding science-research/about-science-research-fda/fdas-predictive-
acquisition (equal), Supervision (equal), Writing – toxicology-roadmap/)
review & editing (equal) [19] FDA pushes to replace animal testing 2025 Nat. Biotechnol.
43 655
[20] Russell W M S, Burch R L and Hume C W 1959
The Principles of Humane Experimental Technique
References (Wheathampstead (UK): Universities Federation for Animal
Welfare) vol 238
[1] Andrade R J, Aithal G P, Björnsson E S, Kaplowitz N, Kullak- [21] Hubrecht R C and Carter E 2019 The 3Rs and humane
Ublick G A, Larrey D and Karlsen T H 2019 EASL clinical experimental technique: implementing change Animals
practice guidelines: drug-induced liver injury J. Hepatol. 9 754
70 1222–61 [22] Mirshafiei M, Rashedi H, Yazdian F, Rahdar A and Baino F
[2] Kaplowitz N 2005 Idiosyncratic drug hepatotoxicity Nat. 2024 Advancements in tissue and organ 3D bioprinting:
Rev. Drug Discovery 4 489–99 current techniques, applications, and future perspectives
[3] Dey S, Bhat A, Janani G, Shandilya V, Gupta R and Mater. Des. 240 112853
Mandal B B 2024 Microfluidic human physiomimetic liver [23] Sun X, Ren W, Xie L, Ren Q, Zhu Z, Jia Q, Jiang W, Jin Z and
model as a screening platform for drug induced liver injury Yu Y 2024 Recent advances in 3D bioprinting of tissues and
Biomaterials 310 122627 organs for transplantation and drug screening Virtual Phys.
[4] Uetrecht J and Naisbitt D J 2013 Idiosyncratic adverse drug Prototyp. 19 e2384662
reactions: current concepts Pharmacol. Rev. 65 779–808 [24] Cuvellier M, Ezan F, Oliveira H, Rose S, Fricain J-C,
[5] Vernetti L A, Vogt A, Gough A and Taylor D L 2017 Langouët S, Legagneux V and Baffet G 2021 3D culture of
Evolution of experimental models of the liver to predict HepaRG cells in GelMa and its application to bioprinting of
human drug hepatotoxicity and efficacy Clin. Liver Dis. a multicellular hepatic model Biomaterials 269 120611
21 197–214 [25] Fang Y, Guo Y, Liu T, Xu R, Mao S, Mo X, Zhang T,
[6] Marshall L J, Bailey J, Cassotta M, Herrmann K and Ouyang L, Xiong Z and Sun W 2022 Advances in 3D
Pistollato F 2023 Poor translatability of biomedical research bioprinting Chin. J. Mech. Eng. 1 100011
using animals—A narrative review, alternatives to laborat- [26] Ali A S M et al 2024 3D bioprinting of liver models: a sys-
ory animals ATLA 51 102–35 tematic scoping review of methods, bioinks, and reporting
[7] Shao W, Xu H, Zeng K, Ye M, Pei R and Wang K 2025 quality Mater. Today Bio 26 100991
Advances in liver organoids: replicating hepatic complex- [27] Ali A S M, Berg J, Roehrs V, Wu D, Hackethal J,
ity for toxicity assessment and disease modeling Stem Cell Braeuning A, Woelken L, Rauh C and Kurreck J 2024 Xeno-
Res. Ther. 16 27 free 3D bioprinted liver model for hepatotoxicity assessment
[8] LeCluyse E L, Witek R P, Andersen M E and Powers M J Int. J. Mol. Sci. 25 1811
2012 Organotypic liver culture models: meeting current [28] Wu D et al 2024 Bioprinting of perfusable vascularized
challenges in toxicity testing Crit. Rev. Toxicol. 42 501–48 organ models for drug development via sacrificial-free direct
[9] van der Valk J et al 2018 Fetal bovine serum (FBS): past— ink writing Adv. Funct. Mater. 34 2314171
present—future Altex 35 99–118 [29] Chen M, Suzuki A, Thakkar S, Yu K, Hu C and Tong W 2016
[10] Subbiahanadar Chelladurai K, Selvan Christyraj J D, DILIrank: the largest reference drug list ranked by the risk
Rajagopalan K, Yesudhason B V, Venkatachalam S, for developing drug-induced liver injury in humans Drug
Mohan M, Chellathurai Vasantha N and Selvan Discovery Today 21 648–53
Christyraj J R S 2021 Alternative to FBS in animal cell [30] Sun B, Liang Z, Wang Y, Yu Y, Zhou X, Geng X and Li B
culture—An overview and future perspective Heliyon 2024 A 3D spheroid model of quadruple cell co-culture
7 e07686 with improved liver functions for hepatotoxicity prediction
[11] Yang S, Ooka M, Margolis R J and Xia M 2023 Liver three- Toxicology 505 153829
dimensional cellular models for high-throughput chemical [31] Proctor W R et al 2017 Utility of spherical human liver
testing Cell Rep. Methods 3 100432 microtissues for prediction of clinical drug-induced liver
[12] Wang Q, Feng Y, Wang A, Hu Y, Cao Y, Zheng J, Le Y and injury Arch. Toxicol. 91 2849–63
Liu J 2024 Innovations in 3D bioprinting and biomateri- [32] Sakatis M Z et al 2012 Preclinical strategy to reduce clinical
als for liver tissue engineering: paving the way for tissue- hepatotoxicity using in vitro bioactivation data for >200
engineered liver iLIVER 3 100080 compounds Chem. Res. Toxicol. 25 2067–82
[13] Carolina E et al 2024 Generation of human iPSC-derived [33] Xu J J, Henstock P V, Dunn M C, Smith A R, Chabot J R
3D bile duct within liver organoid by incorporating human and de Graaf D 2008 Cellular imaging predictions of clinical
iPSC-derived blood vessel Nat. Commun. 15 7424 drug-induced liver injury Toxicol. Sci. 105 97–105
[14] Nguyen D G, Funk J, Robbins J B, Crogan-Grundy C, [34] Zhu X and Kruhlak N L 2014 Construction and analysis
Presnell S C, Singer T and Roth A B 2016 Bioprinted of a human hepatotoxicity database suitable for QSAR
3D primary liver tissues allow assessment of organ-level modeling using post-market safety data Toxicology
response to clinical drug induced toxicity in vitro PLoS One 321 62–72
11 e0158674 [35] Rose K A, Holman N S, Green A M, Andersen M E and
[15] Grix T, Ruppelt A, Thomas A, Amler A K, Noichl B P, LeCluyse E L 2016 Co-culture of hepatocytes and Kupffer
Lauster R and Kloke L 2018 Bioprinting perfusion-enabled cells as an in vitro model of inflammation and drug-induced
liver equivalents for advanced organ-on-a-chip applications hepatotoxicity J. Pharm. Sci. 105 950–64
Genes 9 176 [36] Ireland R G, Kibschull M, Audet J, Ezzo M, Hinz B, Lye S J
[16] Hiller T et al 2018 Generation of a 3D liver model compris- and Simmons C A 2020 Combinatorial extracellular mat-
ing human extracellular matrix in an alginate/gelatin-based rix microarray identifies novel bioengineered substrates
bioink by extrusion bioprinting for infection and transduc- for xeno-free culture of human pluripotent stem cells
tion studies Int. J. Mol. Sci. 19 3129 Biomaterials 248 120017

16
Biofabrication 17 (2025) 045022 A S M Ali et al

[37] Liu S, Yang W, Li Y and Sun C 2023 Fetal bovine serum, an [43] Kang S-Y et al 2024 A pillar and perfusion plate platform for
important factor affecting the reproducibility of cell experi- robust human organoid culture and analysis Adv. Healthcare
ments Sci. Rep. 13 1942 Mater. 13 2302502
[38] Yamashita F, Sasa Y, Yoshida S, Hisaka A, Asai Y, Kitano H, [44] Acharya P, Joshi P, Shrestha S, Choi N Y, Jeong S and
Hashida M and Suzuki H 2013 Modeling of rifampicin- Lee M Y 2024 Uniform cerebral organoid culture on a pil-
induced CYP3A4 activation dynamics for the prediction of lar plate by simple and reproducible spheroid transfer from
clinical drug-drug interactions from in vitro data PLoS One an ultralow attachment well plate Biofabrication 16 025005
8 e70330 [45] Joshi P, Nascimento H S D, Kang S-Y, Lee M, Vanga M G,
[39] Tucker G T 2000 Advances in understanding drug metabol- Lee S-H, Ku B, Miranda M D S and Lee M-Y 2025 Dynamic
ism and its contribution to variability in patient response culture of bioprinted liver tumor spheroids in a pillar/per-
Ther. Drug Monit. 22 110–3 fusion plate for predictive screening of anticancer drugs
[40] Crouch S P, Kozlowski R, Slater K J and Fletcher J 1993 The Biotechnol. Bioeng. 122 995–1009
use of ATP bioluminescence as a measure of cell prolifera- [46] Jun H R et al 2023 High-throughput organo-on-pillar (high-
tion and cytotoxicity J. Immunol. Methods 160 81–88 TOP) array system for three-dimensional ex vivo drug test-
[41] Petty R D, Sutherland L A, Hunter E M and Cree I A 1995 ing Biomaterials 296 122087
Comparison of MTT and ATP-based assays for the meas- [47] Morris L E and Guthrie T H Jr 1988 Busulfan-induced hep-
urement of viable cell number J. Biolumin. Chemilumin. atitis Am. J. Gastroenterol. 83 682–3 (available at: https://
10 29–34 [Link]/3376924/)
[42] Dvořák Z, Štarha P and Trávníček Z 2011 Evaluation of [48] Key N S, Kelly P M, Emerson P M, Chapman R W,
in vitro cytotoxicity of 6-benzylaminopurine carboplatin Allan N C and McGee J O 1987 Oesophageal varices asso-
derivatives against human cancer cell lines and primary ciated with busulphan-thioguanine combination therapy for
human hepatocytes Toxicol. Vitro 25 652–6 chronic myeloid leukaemia Lancet 2 1050–2

17