Abstract
The development of vascular networks in microfluidic chips is crucial
for the long-term culture of three-dimensional cell aggregates such as
spheroids, organoids, tumoroids, or tissue explants. Despite rapid
advancement in microvascular network systems and organoid technologies,
vascularizing organoids-on-chips remains a challenge in tissue
engineering. Most existing microfluidic devices poorly reflect the
complexity of in vivo flows and require complex technical set-ups.
Considering these constraints, we develop a platform to establish and
monitor the formation of endothelial networks around mesenchymal and
pancreatic islet spheroids, as well as blood vessel organoids generated
from pluripotent stem cells, cultured for up to 30 days on-chip. We
show that these networks establish functional connections with the
endothelium-rich spheroids and vascular organoids, as they successfully
provide intravascular perfusion to these structures. We find that
organoid growth, maturation, and function are enhanced when cultured
on-chip using our vascularization method. This microphysiological
system represents a viable organ-on-chip model to vascularize diverse
biological 3D tissues and sets the stage to establish organoid
perfusions using advanced microfluidics.
Subject terms: Stem-cell biotechnology, Tissue engineering, Biomedical
engineering, Induced pluripotent stem cells, Microfluidics
__________________________________________________________________
Vascularization remains a significant challenge in organoid technology.
Here, the authors develop a microfluidic platform that enhances
organoid growth, function and maturation, by establishing functional
perfusable vascular networks.
Introduction
The ability to vascularize organoids remains a challenge in the field
of tissue engineering. Indeed, most tissues exceeding 400 µm in
thickness need a functional vasculature to ensure a sufficient supply
of nutrients and oxygen, as well as the ability to remove carbon
dioxide and cellular waste products, preventing the formation of
necrotic inner cores^[66]1. Although significant efforts have been
directed towards creating increasingly complex organoid model systems
in vitro, it remains necessary to transplant such organoids into host
animals to establish functional vascular circulation^[67]2–[68]6.
However, in vivo transplantations are very expensive and lack
scalability for larger scale toxicity or drug screening.
Considerable effort has been directed towards addressing this problem
through the generation of perfusable vascular networks on-chip, either
as standalone^[69]7–[70]9, or by incorporation of other tissues to
develop functional, vascularized organs-on-chips^[71]10–[72]15. In this
in vitro approach, endothelial cells and supportive cells are seeded
into a central microfluidic chamber using hydrogels. These hydrogels
provide structural support to the embedded cells where they can
self-organize into endothelial networks. The media flows continuously
into the lateral channels adjacent to the central microchamber,
providing the nutrients and gas exchange required for long-term cell
culture. However, this conventional geometry does not allow
reproduction of the fluxes observed in vivo, and replicating the in
vivo functional vascularization of iPSC-derived organoids is still an
ongoing challenge^[73]16.
Here, we report a platform to vascularize various biological tissues
on-chip, using an original and user-friendly microfluidic device and
chip loading process. The reliability of our system was validated using
spheroids generated from human fibroblasts and endothelial cells as
well as 3D human blood vessel organoids (BVOs) generated from
human-induced pluripotent stem cells (hiPSCs)^[74]3,[75]17 and human
pancreatic islet spheroids. Importantly, we demonstrate effective
anastomosis and controlled perfusion of the vascular organoids, as well
as enhanced organoid growth, maturation, and vasculature development.
Additionally, we report enhanced functionality of pancreatic islet
spheroids using our platform.
Results
Design of a microfluidic device for precise encapsulation of organoids
We set out to design a microfluidic device that offers reliable and
user-friendly monitoring of flow dynamics. A microfluidic chip was
fabricated using cyclic olefin copolymer (COC), a material that
presents long-term robustness, is suitable for mass production,
exhibits desired optical qualities for imaging, and low absorption of
chemicals^[76]18. Each chip contained 10 microchannels and was
monitored using a 10-channel syringe-pump (Fig. [77]1a–c and
Supplementary Fig. [78]1). We next developed an efficient method,
adapted from existing hydrodynamic trapping principles^[79]19, that
allowed for the encapsulation of organoids in a predefined location
within a serpentine-shaped microchannel (Fig. [80]1d, Supplementary
Fig. [81]2a and Supplementary Note [82]1)^[83]20. The organoids,
embedded in a fibrin hydrogel, were indeed precisely positioned at the
trap site without any apparent morphological alteration (Supplementary
Fig. [84]2b). Of note, the dimensions of the trap site can be adjusted
according to the size of the organoids used (for BVOs of diameter Ø ≈
600 µm, Width = 300 µm and Height = 800 µm, for mesenchymal and
pancreatic islet spheroids of diameter Ø ≈ 300 µm, Width = 200 µm and
Height = 400 µm) (Fig. [85]1d and Supplementary Fig. [86]1d).
Fig. 1. Device design and overview of organoid and cell configurations.
[87]Fig. 1
[88]Open in a new tab
a Computer-aided design (top) and photographs (bottom) of the
microfluidic chip displaying 10 microchannels, with each microchannel
featuring a trapping site. b Top view of the microfluidic device. A
syringe pump was connected to the outlet of the channel to introduce
fluid perfusion. c Schematic diagram and photograph of the
parallelization feature of our setup, showcasing 10 microchannels
controlled simultaneously. d Photograph of the microfluidic chip and
schematic three-dimensional view of the U-cup shaped area functioning
as a trap. Here, the trap site is exemplarily occupied by a cell
aggregate. e Schematic diagram showing an overview of the loading
process. Initially, the hydrogel containing an organoid and HUVEC cells
was introduced. Before polymerization of the hydrogel, air was
introduced to position the hydrogel and the HUVEC cells. Finally,
growth medium was introduced for continuous perfusion of the
microfluidic chamber and the trapped organoid. f Schematic 3D and
cross-sectional views of the microchannel showing the air loading
process and associated hydrogel deposition. g Experimental
cross-sectional view (left) of the microfluidic channel showing the
hydrogel deposition in the trap and in the channel’s corners and 3D
rendering (right), taken with an in-house light sheet fluorescence
microscopy set-up. h Representative images of vascular
spheroid/organoid cultured in fibrin (left), and hiPSC-derived lung
organoids cultured in Matrigel, showing efficient trapping and robust
growth over 2 weeks on-chip (right).
Into the fibrin hydrogel we incorporated a mixture of human umbilical
vein endothelial cells (HUVECs) and fibroblasts, surrounding the
organoid. Subsequently, air was injected to push the hydrogel toward
the exit of the microfluidic channel. After allowing the gel to
polymerize for approximately 5 minutes at room temperature, continuous
microfluidic perfusion with growth medium was established
(Fig. [89]1e). Both the hydrogel and air were injected at
Q = 300 µl/min. Of note, the hydrogel remained inside the trap site due
to capillarity, effectively surrounding the organoid and thereby
minimizing organoid contact with the microchannel walls resulting in a
permanent organoid encapsulation inside the trap site^[90]20. Because
of the Landau-Levich-Bretherton effect^[91]21, a thin layer of hydrogel
remained along the square profile cross-section of the microchannels
after injection of air (Fig. [92]1f). We used this property as means of
achieving complete endothelialization of the serpentine channel with
HUVECs (Supplementary Fig. [93]3a). Experiments using an adapted
in-house light sheet fluorescence microscopy set-up were conducted to
confirm the three-dimensional structure of the hydrogel deposition near
the trap site and along the main serpentine microchannel (Fig. [94]1g
and Supplementary Fig. [95]3b–d). The thickness of the gel layer on the
microchannel walls can be adjusted by varying the flow rate of the air
bubble passage (Supplementary Fig. [96]3e). This localized hydrogel
deposition enabled the development of a self-organized endothelial
network that traverses the trap site and supplies the organoid with
nutrients. One of the strengths of our microfluidic platform is its
robustness, scalability, and adaptability. Whereas other platforms
might achieve comparable vascularization, the presented design promises
enhanced speed and reliability. In practice, we can load a microchannel
in just around 10 seconds, achieving a trapping efficiency that is
close to 100% (Supplementary Fig. [97]4).
Moreover, this chip design can be readily modified to host several
traps in order to study multiple organoids in parallel (Supplementary
Fig. [98]5). In this study, our emphasis was on vascular spheroids and
organoids (generated off-chip) cultured in a fibrin hydrogel. However,
our approach is versatile and applicable to other commonly utilized
extracellular matrices (ECM) like Matrigel. For instance, by adhering
to the same protocol, we cultured hiPSC-derived lung organoids within
Matrigel (Fig. [99]1h). These organoids were not only efficiently
trapped in our device but also exhibited robust growth and bud
formation over a 2-week culture period on-chip. Overall, we have
designed a serpentine geometry chip for precise and controlled
entrapment of organoids and other 3-dimensional cellular constructs in
endothelium-lined microfluidic channels.
Establishment of interconnected perfusable endothelial networks
To interrogate the utility and biological relevance of our system, we
used cell aggregates consisting of human fibroblasts and GFP labelled
HUVEC cells, termed hereafter mesenchymal spheroids, which were seeded
into the microfluidic channels. We first examined the effects of fluid
flow by culturing the mesenchymal spheroids alone under static (media
changed daily) or flow conditions (Fig. [100]2a). We observed an
enhanced formation of endothelial networks under dynamic perfusion,
with a significant 4.4-, 6.5-, 5.0- and 4.8-fold increase in the number
of vessel junctions, number of meshes, number of segments and total
segment length, respectively, as compared to static conditions. Thus,
flow conditions can directly drive the differentiation of these
mesenchymal spheroid into vessel-like structures (Fig. [101]2a–c).
Fig. 2. Generation of anastomosed endothelial networks through functional
vascularization of mesenchymal spheroids.
[102]Fig. 2
[103]Open in a new tab
a Mesenchymal spheroids cultured on-chip under static or flow
conditions. Representative images are shown at day 7 after seeding. b
Angiogenesis Analyzer output of the morphology of the endothelial
networks after a week of culture on-chip (n = 10 (static) and 8
(flow)). c Confocal z-stack maximum intensity projection of the
three-dimensional endothelial network. d Gel embedded RFP-HUVEC cells
in the main channel are shown in red, GFP-HUVEC cells from the
mesenchymal spheroids are shown in green. Note the formation of
structured endothelial networks over time that appear to be stable
until the end of the observation period (day 13 after seeding). In a
close-up view, anastomosis between RFP-HUVEC and GFP-HUVEC cells are
indicated by white arrows (inset). e Schematic drawing depicting the
cell culture configuration, with anticipated open connections at the
fibrin gel interface allowing for microbeads perfusion (created with
BioRender.com). f Maximum of intensity projection over a 71 images
stack, highlighting the tracks of the microbeads passing through the
interconnected network. See Supplementary Movie [104]1 and
Supplementary Fig. [105]5 for raw movies and details. g Sum of the
binarized and frame-color coded images from (f) showing time-resolved
beads perfusion. See Supplementary Movie [106]2 for details. h
Projections of maximum intensity over an image stack showing tracking
of one individual microbead (red) passing through the endothelial
network. The inset shows an assembled projection of three movies taken
at the indicated area at higher magnification. For better
visualization, the RFP-HUVEC cells are not shown in these images. See
Supplementary Movie [107]3 for raw data. i Summary of different flow
parameters measured in our organ-on-chip device as compared to in vivo
physiological flow rates in human capillaries. Data represent mean
values ± s.d. Beads were 1 µm (f), 4.8 µm and 0.5 µm (h and inset) in
diameter. Scale bars, 400 µm (d), 200 µm (f and g), 100 µm (a, c and
h), 50 µm (d (inset)) and 20 µm (h (inset)). Data represents mean ±
s.d. Statistical significance was attributed to values of P < 0.05 as
determined by unpaired t test (two-tailed). ***P < 0.001. Source data
are provided as a Source Data file.
We next incorporated HUVECs in the hydrogel mix (as described above) to
establish functional connections between the HUVEC lined microchannel
and the trapped mesenchymal spheroid. The HUVEC endothelial cells of
the spheroid expressed GFP and the HUVEC endothelial cells suspended
inside the gel prior to injection expressed RFP, allowing for
visualization of the distinct cell populations and their interactions
in real time (Fig. [108]2d). On day 0, the mesenchymal spheroid was
introduced into the microchannel and trapped at the correct location,
where it maintained its spherical shape. On day 3, we observed an
initial organization of the endothelial cells, and by day 7,
network-like structures with a three-dimensional configuration
developed and remained stable up to day 13 (Fig. [109]2d).
Intriguingly, we observed spontaneous anastomosis between the RFP-HUVEC
endothelial bed and the vasculature of the mesenchymal spheroids
(Fig. [110]2d inset), establishing an interconnected network across the
trap site.
To demonstrate the functionality of the interconnected endothelial
networks, we performed perfusion assays using fluorescent microbeads
(Fig. [111]2e). To visualize flow through the network, beads with a
diameter of 1 µm were injected into the microchannel at a flow rate of
Q[+] = 10 µl/min on the day 13 of culture. The bead tracings were
overlaid onto the patterns generated by the endothelial networks from
the entrance to the exit of the trap, indicating the capability of the
formed network to support perfusion (Fig. [112]2f, Supplementary
Fig. [113]6a and Supplementary Movie [114]1). Importantly, the majority
of the beads travelled through the endothelial network along the main
perfusion direction (path A); as expected, some beads also flowed
through the secondary pathway (path B) (Fig. [115]1d, f). Of note, as
the experiment was done at a constant flow rate, the velocity of a
microbead was inversely proportional to the cross-sectional area it
flowed through, hence the velocity decreased in regions without
endothelium. This resulted in an intense fluorescent signal from the
beads’ tracings on the bottom left and top right corners of the imaged
area (Fig. [116]2f).
We next analyzed the distribution of the microbeads within the
endothelial network over time; this was done by counting the numbers of
pixels in each frame of the binarized and frame-color-coded movies that
corresponded to the movements of the microbeads. The microbeads
perfused the whole endothelial network without any apparent
prioritization of any particular area (Fig. [117]2g, Supplementary
Fig. [118]6b and Supplementary Movie [119]2), demonstrating that
apparently the entire endothelial network present in the mesenchymal
spheroids was perfusable. In the microchannels shown in Fig. [120]2h, a
smaller number of beads was introduced at a lower flow rate of
Q[− ]= 0.1 µl/min to visualize the motions of individual microbeads
(Supplementary Movie [121]3). Superposition of the tracked microbeads
(red) and the fluorescent signal from the mesenchymal cell aggregate
endothelium (grey) confirmed perfusion of the network (Fig. [122]2h).
Thus, our serpentine geometry chips enable the establishment of
anastomosed and perfusable endothelial networks.
Lastly, we investigated the physiological relevance of our microfluidic
platform. The observed flows in the recorded networks were
characterized by tracking fluorescent microbeads as they entered the
endothelial network. Depending on the flow rate imposed by the syringe
pump, microbeads moving through the microvessels exhibited fluid
velocities ranging from v[min] = 100 µm/s to v[max] = 7500 µm/s
(Fig. [123]2i). With a calculated Reynolds number (
[MATH: Re :MATH]
) of 10^−3 to 10^−1 (Supplementary Note [124]2), the flow can be
ascertained to be laminar; thus one can deduct from these values the
fluidic shear rate at the vessel wall, which is given by
[MATH: γw°=4v¯R :MATH]
, where R is the radius of the vessel and
[MATH: v¯
:MATH]
the linear fluid velocity (Supplementary Note [125]3). Using this
equation,
[MATH: γw° :MATH]
ranged from 27 to 2000 s^−1. The mean ( ± s.d.) velocity over the
perfused network at Q[−,] and associated shear rate, were
[MATH: v¯
:MATH]
= 502 ± 305 µm/s and
[MATH: γw°= :MATH]
134 ± 81 s^−1, respectively (n = 4 beads individually tracked). The
mean (±s.d.) velocity over the perfused network at Q[+], and the
associated wall shear rate, were
[MATH: v¯=
:MATH]
2689 ± 531 µm/s and
[MATH: γw°= :MATH]
531 ± 142 s^−1, respectively (n = 4 beads individually tracked). The
observed range of flux values was consistent with the flow rates
detected in human blood vessels of similar size^[126]22, corresponding
to velocities of
[MATH: v¯
:MATH]
= 500–1500 µm/s and wall shear rates of
[MATH: γw° :MATH]
= 50–1900 s^−1 (Fig. [127]2i). Thus, the observed perfusion flow in
our organ-on-chip device resembles physiologic flow observed in human
capillaries.
Vascularization of blood vessel organoids
While significant progress has been made in the development of complex
organoids, the convergence of human tissue engineering and
microfluidics is needed to address the technical challenges that
remain, in particular to vascularize 3D tissues to support extended
growth and differentiation^[128]23. To further show the utility of our
device design, we seeded 3D blood vessel organoids (BVOs) generated
from human-induced pluripotent stem cells (hiPSCs) into our chip. We
have previously reported the generation of such organoids^[129]17 and
demonstrated their physiological relevance through the modelling of
diabetic vasculopathy both in vitro and in vivo^[130]3. Slight
modifications were made to the BVO differentiation protocol to ensure
diametric homogeneity for trapping consistency (Fig. [131]3a).
Similarly to BVOs cultured in wells, these BVOs self-organize on-chip
into three-dimensional interconnected networks of bona fide capillaries
containing an endothelial cell lined lumen, a pericyte and smooth
muscle cells coverage, and expression of tight junction proteins ZO-1
and adherens junction protein VE-cadherin (Fig. [132]3b). The organoids
formed capillaries with hollow lumens surrounded by a prototypic basal
membrane (Fig. [133]3c). Because it has been previously reported that
narrow capillaries fail to open at the interface to the media
channel^[134]24, we attempted to link these BVO capillaries with the
HUVEC network of the microchannel.
Fig. 3. Vascularization of Blood Vessel Organoids (BVOs) on-chip.
[135]Fig. 3
[136]Open in a new tab
a Schematic of the protocol used for the differentiation of human
pluripotent stem cells into blood vessel organoids. b Representative
immunofluorescence of BVOs-on-chip, with capillary networks expressing
CD31, the tight junction protein ZO-1, the adherens junction protein
VE-cadherin, and covered by pericytes (PDGFRβ). c Cross-section of
BVO-on-chip showing capillaries with hollow lumens and collagen type IV
basal membrane coverage. d Initial and final cell culture
configuration. e Vessel from the endothelial HUVEC network (stained for
the endothelial marker CD31) with orthogonal views showing hollow lumen
structures. f Time resolved evolution of the cell culture on chip.
GFP-HUVEC cells self-organized from a single cell suspension, into an
endothelial network surrounding the organoid. g Quantification of the
endothelial networks using Angiogenesis Analyzer plugin, at different
time points of the on-chip culture. h–k Angiogenesis Analyzer outputs
of the morphology of the endothelial networks regarding the total
network length (e), the number of segments (f), the number of junctions
(g) and the total isolated branches length (h) in imaged area from
n = 5 independent microchannels denoted as C[i] (where i ranges from 1
to 5). Scale bars, 200 µm (b, d, f, and g) and 20 µm (c, e).
Mature BVOs (day 15), GFP-HUVECs (6 × 10^6 cells/ml) and fibroblasts (2
× 10^6 cells/ml) were embedded within the hydrogel and introduced into
the microchannels as described above (Fig. [137]1d). HUVECs
self-organized into endothelial networks arborizing the seeded BVOs and
were cultured under long-term continuous perfusion on-chip for up to
two weeks (Fig. [138]3d). HUVEC networks with hollow lumen structures
were observed starting around one week of culture (Fig. [139]3e).
Starting from a single-cell suspension, HUVECs begin to self-organize
within the first day of culture. By day 6, we observed the
establishment of stable endothelial networks surrounding the BVOs, with
open connections to the main microchannel (indicated by white arrows),
which remained from that day onwards (Fig. [140]3f). Using the
Angiogenesis Analyzer plugin in ImageJ, the development of the vascular
network was quantified (Fig. [141]3g and Supplementary Fig. [142]7). We
measured an average increase of 110%, 109% and 96% between day 1 and
day 10 in the total network length (Fig. [143]3h), numbers of segments
(Fig. [144]3i) and numbers of junctions (Fig. [145]3j), respectively.
Moreover, between day 2 and day 10, we observed an average decrease of
44% in the total isolated branches length (Fig. [146]3k), indicative of
a progressive organization into an interconnected network.
The networks formed endothelial beds encompassing the BVOs and
exhibited three-dimensional organization that extended across the trap
site (Fig. [147]4a). After 10 to 14 days of on-chip culture, these
cellular structures were fixed and stained using a dynamic on-chip
protocol where PFA 4%, blocking buffer, primary antibody and secondary
antibody solutions were sequentially flowed into the microchannels. All
the GFP-HUVEC vessels were CD31 positive (CD31^+), showing that the
microvascular networks can be readily perfused with the anti-CD31
antibody (Fig. [148]4b). Functionality of the HUVEC networks was
assessed by the live perfusion of fluorescent microbeads through the
microchannels after 10 to 13 days of culture (Fig. [149]4c and
Supplementary Movie [150]4). Note that the beads were nearly completely
flushed out during the immunostaining on-chip process, however, a few
beads remained trapped inside the HUVEC formed endothelial structures
(Fig. [151]4d). While most of the experiments were stopped after around
2 weeks of culture on-chip, the HUVEC networks were found to be stable
up to 30 days of culture (Fig. [152]4e). These findings demonstrate the
robustness of our microfluidic platform to form stable and perfusable
endothelial networks, arborizing the trapped BVOs.
Fig. 4. Establishment of perfusable HUVEC endothelial networks encompassing
blood vessel organoids.
[153]Fig. 4
[154]Open in a new tab
a Imaris 3D rendering from confocal z-stacks of the endothelial network
that has developed in the trapping site after 14 days of culture
on-chip. b Confocal z-stack maximum intensity projection of the
endothelial network at day 14 after staining of the microfluidic chip
for CD31 expression and the nuclear marker Hoechst. Cyan dotted lines
are used to provide the organoid’s location behind the network,
determined through brightfield images. c Projection of maximum
intensity over an image stack showing the tracking of one individual
microbead (red) passing through the endothelial network (green). The
inset shows the perfusion of several microbeads at higher
magnification. See Supplementary Movie [155]4 for raw movies. d 3D
representation from confocal z-stacks using clipping planes to reveal
the presence of fluorescent microbeads located inside the lumen of a
HUVEC vessel. e Angiogenesis Analyzer outputs of endothelial networks
on-chip assessing the total network length evolution over a period of
30 days. Experiments were conducted on n = 4 independent microchannels
denoted as Ci (where i ranges from 1 to 4). Scale bars, 400 µm, (e),
200 µm (a–c), 50 µm (d) and 20 µm (d (inset)).
Functional anastomosis between HUVEC networks and BVOs-on-chip
To demonstrate “functional vascularization”, we examined anastomosis
between the HUVEC endothelial networks and the seeded BVOs. Having
previously shown that BVOs transplanted into the kidney capsule of mice
formed functional connections with the host vasculature^[156]3, we
tested whether we could achieve similar in vitro anastomoses through
the use of an intermediate HUVEC endothelial bed. On-chip anti-CD31
antibody staining revealed the presence of endothelial cells in the BVO
vasculature central to the GFP-labelled HUVEC network (Fig. [157]5a).
Numerous BVOs vessels (CD31^+GFP^-) were observed near HUVEC networks,
suggesting functional anastomosis from the HUVEC endothelial bed into
the organoid vessels (Supplementary Fig. [158]8a). When cultured under
flow conditions, the vascularized BVOs displayed a physiological
hierarchical organization with larger HUVEC vessels upstream and
downstream of the BVO (Fig. [159]5b). The regions upstream and
downstream with arteriole-venule sized vessels exhibited strong
expression of the smooth muscle cell SM22 marker, while the narrow
capillaries in the BVOs had hollow lumen structures surrounded by
pericytes (Fig. [160]5b). The average diameter of the HUVEC vessels in
the areas upstream and downstream of the BVO was measured to be 37 µm,
corresponding to arterioles-venules sizes, whereas the vessels inside
the BVO averaged 8 µm in diameter, aligning with in vivo capillary
sizes (Fig. [161]5c). Of note, BVOs that were cultured with fibroblasts
but no HUVECs (referred to as “w/o vasc.”), exhibited minimal CD31
staining and only few vessels at the periphery of the BVO, adjacent to
the microchannel, were stained (Fig. [162]5d); however, we did detect
endothelial sprouts with hollow lumens in the fibrin hydrogel in the
“w/o vasc.” condition (Fig. [163]5d inset). This data indicates that
the presence of the intermediate endothelial bed is required for the
antibodies to penetrate the BVOs located at the trapping area.
Fig. 5. Assessment of functional anastomosis between a HUVEC endothelial bed
and stem cell derived blood vessel organoids.
[164]Fig. 5
[165]Open in a new tab
a Confocal z-stacks maximum intensity projection after staining of the
microchannels for CD31 expression (red) and the nuclear marker Hoechst.
The BVO vasculature corresponding to CD31^+GFP^- vessels is visible in
the center of the trap. b Hierarchical structure of vascularized
BVO-on-chip emulates the arteriole-capillary-venule transition as seen
in vivo. Schematic representation below (created with BioRender.com). c
Quantification of diameters for HUVEC vessels and BVOs vessels (n = 50
individual vessels were measured for each type). d Confocal z-stacks
maximum intensity projection after staining for CD31 expression of the
organoid cultures under flow conditions with and without HUVECs, and in
wells. The organoid location is shown by the cyan dotted line.
e–g Anastomosis of GFP^+ HUVEC vessels with iPS cell-derived blood
vessels of the organoid (e). Imaris segmentation and distance mapping
revealed direct contact between the two vasculatures (f and g). h
Z-slice (150 µm deep) from a confocal z-stack showing the tracking of
one individual microbead (yellow) passing through the BVO endothelium.
Side panels show red, green and blue color channels separately. i, j
Vascularized BVO imaged live (i) and stained (j) after overnight beads
perfusion. Microbeads accumulation in the BVO’s vasculature is
highlighted by a white arrow (j) and Imaris 3D rendering (j (inset)). k
Projection of maximum intensity over an image stack showing tracking of
microbeads (cyan) passing through the BVO’s vasculature (red). The
perfusion direction is indicated by arrows and the image is an assembly
from two movies taken at different z positions. See Supplementary
Movie [166]5 for raw data. l Schematic diagram of the imaging setup
(created with BioRender.com) and 3D rendering of a resulting confocal
z-stack are shown. A close-up view of an individual microbead lodged
deep within the BVO’s vasculature is presented, approximately 400 µm
away from the bottom of the microfluidic chip. Beads were 2 µm (h) and
1 µm (j, k and l) in diameter. Scale bars, 200 µm (a, b and h), 100 µm
(d–f and h (inset), i, k and l), 30 µm (f (inset) and g). Statistical
significance was attributed to values of P < 0.05 as determined by
unpaired t test (two-tailed). ***P < 0.001. Source data are provided as
a Source Data file.
Importantly, we were also able to identify several interfaces showing
anastomosis between the HUVEC endothelial network and the BVO’s
vasculature (Fig. [167]5e and Supplementary Fig. [168]8). Using
segmentation tools in Imaris on the confocal fine z-stacks, we
confirmed direct connections between HUVEC GFP^+ networks and BVO’s
CD31^+GFP^- capillaries (Fig. [169]5f). Following distance
transformation mapping, the organoid vasculature was further shown to
be in direct contact (purple) with the HUVEC network (white)
(Fig. [170]5g). We also performed microbeads perfusions on-chip.
Considering fundamentals in fluid mechanics, the beads preferentially
flow along the paths of least fluid resistance; therefore, the
likelihood that beads will enter the narrow, lumenized vasculature of
the organoid is low as a result of increased resistance, elevated
intralumenal pressure, and reduction in flow-through. Nevertheless, we
were able to capture microbeads within large vessels of the organoids
in real time (Fig. [171]5h) and we were able to observe beads inside
the internal vasculature of the BVOs after overnight perfusion
(Fig. [172]5i, j). We also captured bead perfusion footage in a subset
of stained organoids (Fig. [173]5k and Supplementary Movie [174]5).
Moreover, we were able to image individual microbeads deep inside the
BVOs’ vasculature (Fig. [175]5l and Supplementary Fig. [176]9a, b),
supporting the notion of effective organoid perfusion. Imaris
segmentation confirmed the presence of beads inside the blood vessels
(Fig. [177]5l inset). In the absence of HUVEC endothelial networks, the
BVO’s vasculature was not perfused as assessed by live microbeads
perfusion, and no beads were detected in the trap site area
(Supplementary Fig. [178]9c). Together, these results show that our
method is suitable for connecting HUVEC endothelial networks with bona
fide blood vessel organoids to organize a perfusable functional
vascular tree.
Enhanced on-chip blood vessel organoids growth and maturation
To demonstrate the applicability of our approach to address critical
questions in the field concerning improved growth and functionality of
3D tissues via vascularization, we conducted experiments comparing
organoid growth under both static and flow conditions, and with or
without a neighbouring HUVEC endothelial bed (termed “with vasc.” and
“w/o vasc.”, respectively). Images in brightfield and GFP fluorescence
were taken repeatedly to observe the development of both the BVOs
(circled in dotted cyan line for ease of visualization) and the
surrounding endothelial networks (Fig. [179]6a, b). BVOs cultured
on-chip in the fibrin hydrogel without HUVECs and fibroblasts showed
impaired growth, resulting in gel degradation, BVO’s shrinkage, and
premature death (Supplementary Fig. [180]10a). The growth and sprouting
of these BVOs, when cultured in wells under identical conditions,
remained unaffected because they were encapsulated in a substantial
volume of gel, which protected against any gel degradation in the
initial days of culture (Supplementary Fig. [181]10b).
Fig. 6. Organoids’ growth and maturation enhancement through flow and
vascularization.
[182]Fig. 6
[183]Open in a new tab
a, b BVO (brightfield, dotted cyan line) and HUVEC network (GFP)
development from day 0 to day 10 on-chip, in static (a) and flow (b)
conditions (diagrams created with BioRender.com). Enhanced ECM
remodeling under flow conditions can be observed in the brightfield
images at day 10, particularly by the presence of black structures
surrounding the trap site, indicative of active cell processes
(highlighted by white arrows). c Angiogenesis Analyzer outputs of the
morphology of the endothelial networks between static and flow
conditions after 10 days of culture on-chip (n = 8 (static) and 10
(flow)). d BVOs’ growth after 10 days of culture on-chip in various
conditions, reported as a percentage increase in their size (average of
minor and major axes from the fitted ellipse) compared to day 0 (n = 7
(static w/o vasc.), 11 (static with vasc.), 7 (flow w/o vasc.) and 12
(flow with vasc.)). e Volcano plot showing differentially expressed
genes between conditions static w/o vasc. and flow with vasc. f Most
enriched (top 15) Gene Ontology (GO) biological process terms resulting
from the comparison between conditions static w/o vasc. and flow with
vasc. g Heatmap of a selection of genes associated with ECM
organization (dotted green boxes) and blood vessel development (dotted
black boxes) pathways. The color-coded representation illustrates the
expression patterns of key genes involved in these biological
processes, enabling a visual comparison of their relative expression
levels between conditions static w/o vasc. and flow with vasc. h
Classification of BVO transcriptomes with a reference tissue simulated
from single-cell RNA-seq data of BVOs matured in mice. Semi-supervised
classification based on the Kruskal-Wallis test of genes significantly
differentially expressed across all conditions, followed by
visualization of the top 50 most significant genes (diagrams created
with BioRender.com). Scale bars, 200 µm (a, b and d). Data represents
mean ± s.d. Statistical significance was attributed to values of
P < 0.05 as determined by unpaired t test (two-tailed). *P < 0.05
(P = 0.05 (static w/ vasc. vs flow w/o vasc.) and P = 0.03 (static with
vasc. vs. flow w/o vasc.)), **P < 0.01 (P = 0.008), ***P < 0.001.
Source data are provided as a Source Data file.
Although static and flow conditions appeared indistinguishable during
the initial days of culture, the flow effects became apparent over
time. HUVEC endothelial networks failed to mature properly under static
conditions in contrast to flow conditions (Fig. [184]6a, b, HUVEC-GFP),
confirmed by the Angiogenesis Analyzer plugin used for network
quantification measurements (Fig. [185]6c). Additionally, ECM
remodeling was substantially enhanced under flow conditions, resulting
in the formation of structures that gradually expanded over time in the
microchannel adjacent to the trap site. In contrast, ECM remodelling
and vascular restructuring was not observed under static culture
conditions (Fig. [186]6a, b, brightfield). Importantly, measurements of
the BVOs’ diameters between day 0 and day 10 of culture on-chip
revealed an enhanced growth of BVOs cultured under flow within an
endothelial bed, as compared to the other three conditions
(Fig. [187]6d). In comparing the flow condition to the static
condition, it is essential to highlight that the distinctions are not
limited to the convective flow and its associated shear; there are also
increase in oxygenation and nutrient availability.
We next performed RNA sequencing to assess the gene expression status
of 3D tissues cultured without and with flow after extraction from the
chip (Supplementary Fig. [188]11). The four conditions on-chip
displayed a notably strong clustering pattern, as evidenced by the
dendrogram and principal component analyses (Supplementary
Fig. [189]12). We found 526 genes differentially expressed with a
log2[fold change] > 1 and adjusted P value below 0.05, comparing the
condition static w/o vasc. with the condition flow with vasc.
(Fig. [190]6e), significantly associated with Gene Ontology terms
(biological process) related to extracellular matrix and blood vessel
development (Fig. [191]6f). We conducted several control analyses to
confirm that these features could be ascribed to our technique, rather
than the residual presence of HUVECs (Supplementary Fig. [192]13).
Genes such as ADAMTS16, CAV1, ANPEP, HMOX1, ACE, known for their key
role in angiogenesis, blood vessel development and blood pressure
regulation processes^[193]25–[194]31, were found upregulated in
condition “flow with vasc.” (Fig. [195]6g). By contrast, genes that are
thought to have anti-angiogenic functions or induce cell death in
endothelial cells and smooth muscle cells (e.g. BMP10, OLR1,
AGT)^[196]32–[197]34, were found downregulated in this condition
(Fig. [198]6g). Moreover, genes of the metalloproteinase family,
involved in ECM remodeling^[199]35, were found to be upregulated under
flow conditions (Fig. [200]6g). To explore the impact of static and
flow conditions on organoid maturation, we next analyzed the RNA
profile of BVOs grown under various conditions on-chip and in
conventional 96-well suspension cultures. We then compared these
profiles to a reference dataset generated from the Nikolova et
al.^[201]36 single-cell RNA-seq data of human BVOs matured in mice for
two months. Remarkably, we found that the transcriptomes from our
samples “flow with vasc.” were the most similar to the in vivo
transplanted human BVO vasculature (Fig. [202]6h and Supplementary
Fig. [203]14). These findings indicate that our setup, providing
perfusion and flow, promotes blood vessel organoid maturation.
Enhanced functionality of on-chip vascularized islet spheroids
To further showcase the power of our microfluidic platform, we
generated pre-vascularized insulin-secreting pancreatic islet spheroids
as a second proof-of-principle of our method. Using the same approach
as with the BVOs, we cultured the islets on-chip under static or flow
conditions, and with or without a neighbouring GFP-HUVEC endothelial
bed (Fig. [204]7a). At day 5, we performed glucose stimulated insulin
secretion (GSIS) assays (Fig. [205]7b). Insulin concentration
measurements at baseline showed no significant differences in all the
conditions tested, but insulin secretion was significantly increased in
response to high glucose challenge (Fig. [206]7c). Remarkably, the
corresponding stimulation index ([high glucose solution]/[low glucose
solution]) was significantly higher for the pancreatic islet spheroids
on-chip with flow and with vascularization, compared to all other
conditions on-chip, as well as compared to islets cultured in
conventional wells (Fig. [207]7d). The stimulation index for the
pancreatic islet spheroids on-chip with both flow and vascularization
showed higher variation, which might be attributed to differences in
the quality or extent of the vascular network. However, the stimulation
index showed no improvement when either only the flow or the HUVEC
endothelial bed was introduced. These data indicate an improved insulin
secretion in response to glucose stimulation of pancreatic islet
spheroids when cultured on-chip under flow conditions and embedded in a
HUVEC vascular bed, creating an improved functional β-cell niche.
Fig. 7. Pancreatic islet spheroids’ function enhancement through flow and
vascularization.
[208]Fig. 7
[209]Open in a new tab
a Generation of pre-vascularized pancreatic islet spheroids for on-chip
culture within a HUVEC endothelial bed and under flow conditions
(diagram created with BioRender.com). b Schematic diagram of glucose
stimulation and collect of secretions performed on-chip (created with
BioRender.com). c, d Comparison of the insulin secretions (c) and
stimulation index = [high glucose solution]/[low glucose solution] (d)
of the pancreatic islet spheroids between the various culture
conditions. Each point on the graph represents the stimulation index of
one pancreatic islet obtained from the measure of low and high glucose
stimulation. Scale bar, 100 µm (a). Data represents mean ± s.d.
Statistical significance was attributed to values of P < 0.05 as
determined by unpaired t test (two-tailed). *P < 0.05 (P = 0.017
(static w/o vasc.), P = 0.029 (static with vasc.), P = 0.037 (flow
without vasc.)), ***P < 0.001. Source data are provided as a Source
Data file.
Discussion
In this paper we present a microfluidic platform which provides simple,
flexible and robust solutions to form vascularized organoids-on-chip
with precise control over the fluxes generated during both short- and
long-term perfusion. Through this approach, we counteract common
barriers in organoid growth, maturation, and organ-on-chip
technologies, such as manual operation and a lack of automation in
construct location and hydrogel handling. First, the device was
manufactured using COC, a material readily adapted by industry,
avoiding fundamental complications of PDMS systems such as
incompatibility with hydrophobic compounds or small molecule
absorption. Second, in most studies, the flow rate is determined by
hydrostatic pressure, a simple and cost-efficient liquid actuation
principle. However, optimal hydrostatic differential pressure cannot be
maintained for a long time, with the consequence that the culture
medium must be changed regularly. This can be avoided by using rocker
perfusion platforms^[210]37,[211]38, at the consequence of a reduction
in physiologically-relevant flow mechanics (i.e. pressure modulation
and/or peristalsis). In our study, this has been overcome through the
use of a syringe pump. Using 20 ml syringes, the microfluidic perfusion
can be maintained for two weeks at a flow rate of 1 µl/min without
disconnecting the syringes. It should be noted that by using a
10-channel syringe pump, as was done in this study, flow rates in 10
different channels can be readily monitored in parallel. Third, the
organoids and hydrogels were passively placed into a hydrodynamic trap,
wherein the cells were subjected to minimal shear stress. These traps
can be put in series, thus hosting multiple vascularized organoids
(including different tissue organoids) interconnected through
endothelial networks.
Cell and hydrogel loading, as well as long-term organoid perfusion with
a defined growth medium, were performed through a single fluidic inlet.
Our automation of fluid handling in this system facilitates widespread
use, distribution, and adaptation to microfluidic devices in both
industries and healthcare settings. The presented microfluidic platform
offers robustness through an innovative encapsulation technique, which
relies on hydrodynamic and capillary effects that are largely
independent of the working pressure. While gel loading is often a
delicate step in conventional chip designs, the process used in this
study is highly reproducible and easy to perform. The loading steps can
be performed in seconds, allowing all operations to be completed at
room temperature. With flexibility in application to a variety of
bioengineering projects, this microfluidic platform supports the
vascularization of nearly any 3D construct simply by adjusting
perfusion channel and trap dimensions. In summary, our organ-on-chip
platform is robust, flexible, and user-friendly, overcoming key hurdles
in the industrialization of organ-on-chip models and providing access
to user-friendly organoid-on-chip technologies.
Several limitations of our platform merit discussion. First, the
encapsulation process of endothelial cells and fibroblasts in the
hydrogel results in a significant cell loss when the excess gel is
expelled before polymerization. This becomes particularly pronounced
for larger designs intended for large organoids, where the gel volume,
and consequently the number of cells required, escalates considerably.
This poses challenges, especially when considering more valuable cell
types like isogenic iPSC-derived vascular cells. Furthermore, while
adjusting the gel layer thickness around the organoid is theoretically
feasible, its practical execution is intricate. We invested significant
effort into modeling the Landau-Levich-Bretherton phenomenon, which
dictates the residual thin gel layer on the microchannels post air
bubble passage. Our setup’s unique design adds to this complexity, and
addressing these challenges remains a priority as we refine the
platform.
Nevertheless, our microfluidic device supports the vascularization and
perfusion of 3D biological tissues. The formation of perfusable
endothelial networks was found to be highly reproducible. We were able
to generate physiological fluxes in these networks, with observed
velocities ranging from v = 100 µm/s to v = 7500 µm/s and shear rates
ranging from
[MATH: γ°
:MATH]
= 25 s^−1 to 2000 s^−1. Thus, physiological blood flow velocities and
shear stress dynamics observed in the human body can be readily
achieved using our microfluidic platform—an advantage over commonly
reported hydrostatic pressure operated devices, that result in lower
flow rates^[212]39. Finally, the deposition of the hydrogel on the
walls of the microchannels, conditioned by the Landau-Levich-Bretherton
effect, offers a robust means of achieving continuous endothelial cell
lining of the perfusion channel, recapitulating aspects of vascular
patterning of the tunica intima in vivo.
The chip configuration described here resulted in anastomosis between a
HUVEC endothelial bed and the capillaries of human blood vessel
organoids. Consequently, our system can generate perfused hierarchical
networks that encompass vessels ranging in size from arterioles to
venules and capillaries. Most importantly, we have now accomplished
intravascular perfusion of blood vessel organoids, thus making our
microfluidic platform the first device to incorporate functional
vasculature throughout microfluidic-trapped embedded 3D organoid
constructs. Our platform now allows us to explore diverse topics, such
as organoid lifespan enhancement through vascularization, exposure to
drugs, nucleic acids or metabolic stress. The device we have developed
also offers the flexibility to vascularize other types of organoids,
spheroids, tumoroids, or human tissue explants, as exclaimed in our
study by improved glucose responsiveness of islet spheroids.
Methods
Microfluidic device fabrication
Computer-aided design (CAD) files of the chip were created using
SolidWorks 2022 (Dassault Systèmes). The microfluidic chips were made
of Cyclic Olefin Copolymer (COC) because of its low autofluorescence,
strong chemical resistance, and low drug absorption. Microfluidic
patterns were directly machined on a COC sheet (TOPAS, USA), using
high-precision milling (DATRON M7HP equipment). The chip (84 mm×54 mm)
contained 10 identical microfluidic circuits. The square-profiled
channels were 400 µm x 400 µm for the experiments with mesenchymal
spheroids, and 800 µm x 800 µm for experiments with blood vessel
organoids. The microfluidic channels were sealed with a MicroAmp
optical adhesive film (Applied Biosystems, cat. no. 4306311).
Trapping mechanism and cell seeding procedure
The microfluidic setup is initiated using a syringe at the channel
entrance, to sequentially introduce: (1) non-polymerized hydrogel
embedding spheroids/organoids and endothelial cells, (2) air, and (3)
growth medium. The channel features a serpentine loop that is bypassed
by a U-cup shaped microchannel, which serves as a cell aggregate trap.
When unoccupied, the hydraulic resistance of this trap (R[1], Path A)
is less than that of the serpentine loop (R[2], Path B), thereby
guiding spheroids/organoids into the trap by flow preference.
To initiate the seeding, a spheroid/organoid was gently extracted from
a 96-well plate via a pipette tip and mixed into the hydrogel with
thrombin. This mixture, amounting to 50 µl and containing the
spheroid/organoid, was placed into the reservoir leading to the
microfluidic system. Using a syringe pump set to a withdrawal mode with
a flow rate of Q = 300 µl/min, the spheroid/organoid progressed through
the channel and was captured by the U-cup microchamber. Subsequent
introduction of air served to position the hydrogel before it
solidified. The Hydrogel remained at the U-cup, securing the
spheroid/organoid and lining the channel corners. Both the hydrogel mix
(prior to thrombin addition) and thrombin were kept at 4 °C until used
for the experiment. Immediately after mixing thrombin with the fibrin
hydrogel, it was loaded onto the chip. This microchannel injection
process takes approximately 10 seconds, largely excluding viscosity
changes in the gel.
Cell culture and generation of spheroids and organoids
Mesenchymal spheroids
Primary human fibroblasts (FMA73) were extracted from skin explants
obtained through the elective breast surgery of a healthy young woman
following informed consent; this tissue was provided by Walid Rachidi,
CEA Grenoble. GFP- and RFP-labelled HUVEC cells (Angio-Proteomie, cat.
no. CAP0001GFP and cat. no. CAP0001RFP, respectively) were cultured in
complete EndoGM medium (Angio-Proteomie, cat. no. CAP02). Passage 5–7
cells were used for the experiments. Fibroblasts cultured in Fibroblast
Growth Medium-2 (Lonza, cat. no. CC-3132), and passage 6–8 cells were
used for the experiments. We prepared fibroblasts and HUVEC co-culture,
termed mesenchymal spheroid model here, in U-shaped 96-well ultra-low
attachment microplates (Corning, cat. no. CLS4515). Fibroblasts and
HUVEC cells were mixed at a ratio of 1:1 (5000 cells per well) in
150 µl of medium consisting of a mix of CnT-ENDO (Cellntec, cat. no.
CnT-ENDO) / CnT-Prime Fibroblast medium (Cellntec, cat. no. CnT-PR-F)
at a ratio 1:1. After pre-culturing for 1 day in the microplate, a
spheroid was introduced into the device. The same medium mix of
CnT-ENDO / CnT-Prime Fibroblast medium was used for the microfluidic
perfusion of the fibroblasts and HUVEC co-culture spheroids. RFP-HUVEC
cells were suspended in the hydrogel at a concentration of 6×10^6 cells
per ml.
Blood vessel organoids (BVOs)
3D human blood vessel organoids were generated from human induced
pluripotent stem cells (hiPSCs) as previously described^[213]17. In
brief, NC8 stem cell colonies were harvested using Accutase (Gibco,
cat. no. A1110501) to get a single cell suspension. To make sure that
the organoids would be of an appropriate size to fit into the 800 µm x
800 µm square profile microchannels of the chip used for BVO
experiments, AggreWell™400 (STEMCELL Technologies, cat. no. 34415)
plates were used. Each well of the plate contains 1200 microwells with
a 400 µm diameter. 600000 single stem cells were seeded per well (500
cells/microwell) in aggregation media with 50 µM Y-27632 (Tocris, cat.
no. 1254/10). Mesoderm induction and sprouting was induced directly in
the AggreWell™400 plates by carefully changing the media with a p1000
pipette making sure to not disturb the cell aggregates in the
microwells. For Collagen I-Matrigel embedding, organoids were harvested
from the AggreWell™400 plate by vigorously pipetting up and down with a
cut p1000 tip close to the bottom of the well. Harvested organoids were
embedded in a 12-well plate (approx. 100 organoids/12-well), and
subsequently cut out and singled into low attachment 96 U-well plates
4-5 days after embedding as previously described^[214]17. No
alterations were made to any of the BVO differentiation media. The BVOs
were maintained in a differentiation medium containing 15% FBS
(Sigma-Aldrich, cat. no. F1051), 100 ng/ml VEGF-A (PrepoTech, cat. no.
100-20) and 100 ng/ml FGF-2 (Miltenyi Biotec, cat. no. 130-093-564).
Organoids with a diameter of 500-600 µm were selected and added to the
microfluidic chip at day 15. A mix of the differentiation medium with
CnT-ENDO and CnT Prime Fibroblast media was used for the long-term
culture on-chip (ratio 1:1:1). To address concerns regarding the
stability of sensitive growth factors in our cell culture media, such
as FGF-2 and VEGF-A, we replenished the syringe reservoirs with fresh
media daily.
Pre-vascularized pancreatic islet spheroids
Commercially available human primary β cells called EndoC-βH5 (Human
Cell Design) were co-cultured immediately after thawing with the
above-mentioned primary fibroblasts FMA73 and HUVEC-RFP in a U-shaped
96-well microplate with an ultra-low attachment surface, in a mix of
CnT-ENDO, EndoC-βH5 culture medium (Human Cell Design, cat. no.
ULTIB1-100) and DMEM/F-12 (Gibco, cat. no. 11330032) with 1%
Penicillin-Streptomycin (Gibco, cat. no. 15140122). A ratio of 6:1:1
(EndoC-βH5:FMA73:HUVEC-RFP, 10 000 cells per well) was used, resulting
in 300 µm spheroids after 6 days. Passage 5-12 cells were used for
these experiments.
Hydrogel preparation
A fibrin-hydrogel made of 6.6 mg/ml fibrinogen (Sigma-Aldrich, cat. no.
9001-32-5), 0.15 TIU/ml aprotinin (Sigma-Aldrich, cat. no. 9087-70-1),
2.5 mM CaCl[2] (Sigma-Aldrich, cat. no. 10035-04-8), and 1 U/ml
thrombin (Sigma-Aldrich, cat. no. 9002-04-4) prepared in HEPES-buffered
saline was used in all experiments. After adding the thrombin into the
mixture, all the procedures were quickly performed to avoid premature
gelation.
On-chip immunofluorescent staining
For immunofluorescent staining, the tissues were fixed by flowing 4%
paraformaldehyde (Boston BioProducts, cat. no. BM-155) for 1 h at room
temperature through the microchannel, and subsequently blocked with 3%
FBS (Sigma-Aldrich, cat. no. F1051), 1% BSA (Sigma-Aldrich, cat. no.
A9647), 0.5% Triton-X-100 (Sigma-Aldrich, cat. no. T8787) and 0.5%
Tween (Sigma-Aldrich, cat. no. P7949) for 2 h at room temperature.
Primary antibodies CD31 (Abcam, ab134168, rabbit anti-human, 1:200),
CD31 (Abcam, ab9498, mouse anti-human, 1:200), PDGFRβ (Cell Signaling
Technology, 3169 S, rabbit anti-human, 1:200), ColIV (Chemicon, AB769,
goat anti-human, 1:50), SM22/TAGLN (Abcam, ab14106, rabbit anti-human,
1:200), ZO-1 (Abcam, ab216880, rabbit anti-human, 1:200), VE-Cadherin
(Abcam, ab33168, rabbit anti-human, 1:200), were diluted in blocking
buffer and flowed overnight into the microfluidic chip at 4 °C. After a
30 min wash in PBST (0.05% Tween), secondary antibodies (donkey
anti-mouse Alexa Fluor 488, Invitrogen, A-21202, donkey anti-rabbit Cy3
Jackson ImmunoResearch Inc., 711-165-152, donkey anti-rabbit Alexa
Fluor 555, Invitrogen, A-31572, donkey anti-rabbit Alexa Fluor 647,
Invitrogen, A-31573, donkey anti-goat Alexa Fluor 555, Invitrogen,
A-21432, donkey anti-goat and Alexa Fluor 647, Invitrogen, A-21447)
were flowed into the microchannels at 1:200 in blocking buffer for 2 h
at room temperature. After a 30 min wash in PBST, nuclear
counter-staining using Hoechst 33342 was carried out according to a
routine protocol.
Imaging
The microfluidic chip was imaged on a daily basis using an inverted
Olympus IX50 microscope for the period of the experiment. Images were
taken in brightfield and fluorescence channels, with 5x and 10x
objectives. After immunostaining, the microchannels were imaged using
either a Nikon Eclipse Ti-E Spinning Disk microscope, a Zeiss LSM880
scanning confocal microscope or a Leica SP8 scanning confocal
microscope, with 10x, 20x, 25x, 40x and 63x objectives. The flow in
vascular networks was assessed in the second week of culture by loading
polystyrene fluorescent microbeads (Thermo Fisher Scientific,
Fluoro-Max Fluorescent Beads) into the serpentine channel. Images were
captured at 15 Hz using the inverted Olympus IX50 microscope described
above. Microbeads were tracked in perfused tissues from separate
microfluidic channels using Fiji (ImageJ). For microbeads perfusion in
the BVOs’ vasculature, since we needed to capture images deep within
the tissue at a high frame rate while minimizing photobleaching,
confocal microscopy was unsuitable. Therefore, we employed a Leica
THUNDER 3D Cell Imager microscope for this purpose.
Light sheet fluorescence microscopy
A homemade light sheet fluorescence microscope was used in this
project, which we adapted to image biological samples inside
microfluidic chambers without interfering with the normal function of
the chip^[215]40. The light sheet was generated with a 488 nm Ar-laser,
focused by a 100 mm focal length cylindrical lens. The fluorescence
signal generated at the illuminated plane was collected by a long
working distance, with the objective (Mitutoyo M Plan APO SL 20X,
0.28 N.A.) placed at 90° to the excitation path. The sample plane was
at 45° from both paths. A tube lens was associated to the objective to
form the image of the fluorescent structure onto a high-sensitive sCMOS
camera (Hamamatsu HPF6 ORCA FLASH 4.0 V3) with a magnification factor
of 12. To filter out the laser excitation, a high pass (cut-off
wavelength of 490 nm) interference filter was used. The sample was
mounted onto a custom-designed holder attached to computer-controlled
xz linear translational stages. In this configuration, the microfluidic
chip was kept horizontal, and the thinner lateral part of the light
sheet was positioned at the surface of the gel. The light sheet
illuminated the sample in the direction perpendicular to that of the
microfluidic channel (Supplementary Fig. [216]3b-d).
Endothelial networks analysis
Confocal z-stacks of the microchannels in various culture conditions
were taken. These stacks were then flattened in ImageJ to a 2D maximum
intensity projection and analysed using the Angiogenesis Analyzer
plugin with default settings^[217]41. Four metric parameters were
selected for this study, namely the number of junctions, the number of
meshes, the number of segments and the total segments length.
Imaris analysis
Confocal z-stacks were opened and 3D-rendered using Imaris imaging
software. HUVEC endothelial networks and BVOs’ vasculature were
rendered as surfaces after masking with the GFP-HUVEC and CD31 signals,
and the fluorescent microbeads as spots of known diameter.
RNA sequencing
We performed bulk RNA sequencing to investigate gene expression
profiles of the samples under scrutiny. Using a needle, BVOs were
meticulously extracted from the chip, with special care taken to
isolate the BVO without including the surrounding proliferating tissue.
RNA extraction was done using the Trizol protocol (Invitrogen). Sample
quality control was performed using the Agilent 2100 Bioanalyzer to
ensure high-quality RNA input. Qualifying samples underwent library
preparation following the standard protocol for the Illumina Stranded
mRNA prep (Illumina). Subsequently, sequencing was performed on the
Illumina NextSeq2000 platform, generating 59 bp × 59 bp Paired End
reads. Post-sequencing, the data was demultiplexed using Illumina’s BCL
Convert, and the resulting de-multiplexed read sequences were aligned
to the Homo sapiens (hg38 no Alts, with decoys) /Mus Musculous (mm10)
reference sequences. Alignment was carried out using the DRAGEN RNA app
on the Basespace Sequence Hub.
Bioinformatics analysis
Pre-processing of bulk transcriptome data
Gene expression profiles measured as raw read counts were first
filtered for low expressed genes keeping only genes whose sum of reads
across all samples was greater than or equal to 10. Then the ENSEMBL
gene ID were converted into gene symbols using the mapIds() function of
the R/Bioconductor package AnnotationDbi (v.1.60.0). Only genes with a
gene symbol and with a total sum of reads across all conditions greater
than or equal to 10 were kept for further analysis. Gene expression
values were then normalized with the VST method from the R/Bioconductor
package DESeq2 (v.1.3.8.3).
Differential gene expression analysis
The identification of differentially expressed genes (p-value < 0.05)
across multiple groups of experimental conditions was performed with
the non-parametric Kruskal-Wallis rank test from normalized gene
expression values using the col_kruskalwallis function of the
R/Bioconductor package matrixTests (v.0.1.9.1). The list of
differentially expressed genes was visualized using the R/Bioconductor
package ComplexHeatmap (v.2.14.0) and these parameters: z-score
normalization by row and by column, use of Pearson as distance
correlation method and use of ward.D2 as clustering method. Sample
pairwise identification of differentially expressed genes was performed
with R/Bioconductor package DESeq2 (v.1.3.8.3) (FDR corrected
p-value < 0.05; absolute Log2 fold change > 1; baseMean > 18).
Pathway enrichment analysis
The ontological enrichments were performed from the lists of
differentially expressed genes with the R/Bioconductor package
clusterProfiler (v.4.6.0) (p-value < 0.05; Benjamini-Hochberg
q-value < 0.2; minGSSize = 30; maxGSSize = 500). The reference gene
vector used as background corresponded to the genes expressed in the
samples considered for the analyses.
Generation of pseudobulk mixture for reference tissue
From a gene read count matrix of single-cell RNA-seq data of 16,410
cells of BVOs transplanted into mice for their maturation, we randomly
selected three times 5000 cells without replacement. For each batch of
5000 cells, the sum of reads per gene across cells was performed to
generate a pseudobulk sample. These 3 pseudobulk samples were then
integrated with the other experimental conditions and filtered in the
same way. To perform comparisons between transcriptomic data of
different types, bulk RNA-seq and pseudobulk mixture generated from
single-cell RNA-seq data, we removed technical biases and adjusted
artefactual variations in gene expression using the RUVg function from
the R/Bioconductor package RUVSeq (v.1.32.0). The list of negative
control genes, having constant expression across all experimental
conditions, was obtained empirically using the non-parametric
Kruskal-Wallis rank test applied to all conditions. Non-differentially
expressed genes, associated with a p-value > 0.10, were used as
negative control genes. Concerning the number of undesirable factors k,
the higher the value of k, the greater the probability of deleting the
biological signal of interest. We therefore selected the value k = 1
because beyond this threshold we could begin to detect the non-grouping
by clustering of some biological replicates.
Glucose stimulated insulin secretion (GSIS) assays
Pre-vascularized pancreatic islet spheroids were individually placed in
a 96 well plate with ultra-low attachment surface, and incubated at
37 °C, 95% O2, 5% CO2 for 1 h in 60 µl KREBS buffer (Sigma-Aldrich,
cat. no. K4002), 1% BSA, 2.8 mM glucose (Gibco, cat. no. A2494001) as
pre-incubation. Thereafter, islets were incubated for 1 h with 2.8 mM
glucose solution (low glucose solution) and for 1 h with 16.7 mM
glucose solution (high glucose solution). After each incubation step,
supernatants were collected and stored at −80 °C. Insulin concentration
in each collected supernatant was measured using STELLUX
Chemiluminescence Human Insulin ELISA (ALPCO, cat. no. 80-INSHU-CH01).
Similar protocol was performed for on-chip experiments, following a
protocol previously described^[218]42. Each sample was measured in
duplicate. A stimulation index was obtained by calculating the ratio of
insulin measured between high and low glucose stimulation, [high
glucose solution]/[low glucose solution].
Statistics and Reproducibility
Results are shown as mean ± s.d. as indicated in the Figure legends.
Statistical analyses were conducted using GraphPad Prism 9 (GraphPad
Software Inc.). Different significance levels (P values) are indicated
in each figure with asterisks (*P < 0.05, **P < 0.01, ***P < 0.001) and
exact P values when possible.
For transparency, we state here the number of biological samples on
which experiments were repeated independently, with similar results
obtained, to produce the data shown: Figs. [219]1h, [220]6;
Fig. [221]2a, b, > 50; Fig. [222]2c, 12; Figs. [223]3b, [224]6;
Fig. [225]3c, 10; Fig. [226]3d, > 50; Fig. [227]4b, 30; Figs. [228]5b,
[229]6; Figs. [230]5e–g, [231]6; Figs. [232]5i, j, [233]6;
Fig. [234]6a, b, at least 8 for each condition; Figs. [235]6e–h, 3
replicates for each condition; Figs. [236]7c, d, 22, 5, 5, 5, and 11
replicates per condition (wells, static w/o vasc., static with vasc.,
flow w/o vasc., and flow with vasc. respectively).
Reporting summary
Further information on research design is available in the [237]Nature
Portfolio Reporting Summary linked to this article.
Supplementary information
[238]Supplementary Information^ (24.6MB, pdf)
[239]Peer Review File^ (5.7MB, pdf)
[240]41467_2024_45710_MOESM3_ESM.pdf^ (103.8KB, pdf)
Description of additional supplementary files
[241]Supplementary Movie 1^ (4.5MB, avi)
[242]Supplementary Movie 2^ (38.6MB, avi)
[243]Supplementary Movie 3^ (9.9MB, avi)
[244]Supplementary Movie 4^ (12.2MB, avi)
[245]Supplementary Movie 5^ (22.8MB, avi)
[246]Reporting Summary^ (3.6MB, pdf)
Source data
[247]Source Data^ (28.9KB, xlsx)
Acknowledgements