Abstract
Fibrosis is a progressive biological condition, leading to organ
dysfunction in various clinical settings. Although fibroblasts and
macrophages are known as key cellular players for fibrosis development,
a comprehensive functional model that considers their interaction in
the metabolic/immunologic context of fibrotic tissue has not been set
up. Here we show, by transcriptome-based mathematical modeling in an in
vitro system that represents macrophage-fibroblast interplay and
reflects the functional effects of inflammation, hypoxia and the
adaptive immune context, that irreversible fibrosis development is
associated with specific combinations of metabolic and inflammatory
cues. The in vitro signatures are in good alignment with transcriptomic
profiles generated on laser captured glomeruli and cortical
tubule-interstitial area, isolated from human transplanted kidneys with
advanced stages of glomerulosclerosis and interstitial fibrosis/tubular
atrophy, two clinically relevant conditions associated with organ
failure in renal allografts. The model we describe here is validated on
tissue based quantitative immune-phenotyping of biopsies from
transplanted kidneys, demonstrating its feasibility. We conclude that
the combination of in vitro and in silico modeling represents a
powerful systems medicine approach to dissect fibrosis pathogenesis,
applicable to specific pathological conditions, and develop coordinated
targeted approaches.
Subject terms: Systems biology, Computational models, Renal fibrosis,
Monocytes and macrophages
__________________________________________________________________
Renal fibrosis is a progressive process with complex etiopathology,
causing organ failure. Here authors present a mathematical model, based
on an in vitro system faithfully contemplating macrophage-fibroblast
interaction and the metabolic-immunologic signals that are affecting
kidney fibrosis, that is applicable to kidney transplant failure.
Introduction
Fibrosis is the final state of continuous scarring that occurs normally
during healing processes but also significantly contributes reducing
organ function in several chronic diseases^[46]1. Four core mechanisms
are involved: (i) early inflammatory events with involvement of various
immune cells, including T cells and macrophages (Mφ); (ii) activation
of fibroblasts (Fb) to myofibroblasts (collagen-secreting
α-SMA^+activated Fb) and extracellular matrix (ECM) deposition,
generating interstitial scars; (iii) loss of epithelial cells
regenerative properties; (iv) loss of interstitial capillary integrity,
compromising oxygen delivery and leading to a cascade of events related
to hypoxia-oxidant stress, further promoting the fibrotic
process^[47]2. Chemokines and profibrotic cytokines tightly control
recruitment and activation of inflammatory cells at the site of injury.
In a tissue microenvironment dominated by Th2 cytokines (IL-4, IL-13),
Mφ enhance secretion and activation of latent TGFβ1, Fb proliferation
and ECM production, conversion of epithelial cells into
collagen-producing myofibroblasts via epithelial-mesenchymal
transition, and release of proangiogenic factors to promote vascular
remodeling and angiogenesis^[48]3–[49]5. Pathologic angiogenesis and
vessel sprouting in hypoxic tissues worsen fibrosis by leading to
continuous myofibroblast activation and proliferation^[50]6–[51]9.
Fibrosis characterizes the progression of chronic diseases occurring in
many different tissues, including skin, lung, liver, heart, and
kidney^[52]1. In chronic kidney diseases, and in particular, after
renal transplantation, fibrosis can result in glomerulosclerosis with
loss of glomerular filter capacity and interstitial fibrosis and
tubular atrophy (IF/TA), which is associated with impaired tubular
adsorption and secretion processes and regulatory functions and
declining renal function over time^[53]10. Accumulation of ECM
components in glomeruli and in the cortical interstitium collectively
results in progressive loss of renal function^[54]11. The role of Mφ in
acute rejection of kidney transplant, through their contribution to
both T cell-mediated and antibody-mediated rejection processes, is well
established^[55]12,[56]13. On the opposite, their role in IF/TA
development is still under debate, since positive and negative effects
of Mφ infiltration on long-term renal allograft functions were
reported^[57]12,[58]14–[59]16. In IF/TA, Mφ have been shown to switch
from a classical proinflammatory phenotype (also alluded to as M1) to
an alternative phenotype (also alluded to as M2) typically associated
with progression of fibrosis^[60]17,[61]18 prompted by the release of
profibrotic factors, such as TGFβ1, FGF2 and PDGF, that promote
myofibroblasts proliferation and activation, leading to ECM
overproduction^[62]19. However, in different models, chronic
inflammation and subsequent fibrosis have been shown to be promoted by
depletion of Mφ in certain stages of the disease, and in several
diseases an increase in M2 Mφ characterizes the recovery
phase^[63]5,[64]20–[65]22.
Mathematical modeling in biology has been proven an invaluable tool in
investigating existing biological hypotheses in realistic scenarios or
generating experimentally testable ones^[66]23,[67]24. Based on the in
vitro analysis of Fb and Mφ interplay in the context of wound healing
and scarring processes a mathematical model of fibrosis has been
recently developed, which under parameter variations robustly predicted
three functional states: a state of healing associated with modest ECM
production, and two fibrotic states associated with excessive ECM
production and different cellularity where a prominent myofibroblast
infiltration is associated with high or low numbers of Mφ, termed “hot”
and “cold” fibrosis, respectively^[68]25. The model comprehensively
described central aspects of fibrosis and potentially predicted cell
networks leading to clinically observed conditions, but was limited by
the fact that Mφ/Fb interactions always occur in the context of
specific immunological and metabolic features characterizing the tissue
microenvironment^[69]26, an aspect that has not yet been modeled. Here
we develop a mathematical model by translating in vitro observations,
on the relative relevance of different immune and metabolic
microenvironmental cues to the Mφ/Fb interactions, to an in vivo
fibrotic scenario related to IF/TA progression and glomerulosclerosis
in transplanted kidneys (research design is shown in Fig. [70]S1a). We
consider fibrosis after transplantation as a clinically relevant
application case for the model. Large scale evaluation of immune cell
populations in renal biopsies^[71]27,[72]28 showed some potential for
prognosis^[73]14–[74]16 but was insufficient to predict fibrosis and
functional decline^[75]29.
Starting from an in vitro setting that allows transcriptomic dissection
of the effects of different microenvironmental cues on the
macrophage/fibroblast interplay occurring during fibrosis development,
we instruct a mathematical model that predicts key interactions for
fibrosis development and apply it to tissue sections from human
transplanted kidneys with different degree of glomerulosclerosis and
IF/TA. We propose an integrated approach based on a combination of
biological and mathematical models that allow the implementation of
immune and metabolic cues in the current “hot” and “cold” fibrosis
model. By demonstrating the feasibility of this model on transplanted
kidney biopsies, we provide an innovative approach to analyze fibrosis
pathogenesis and guide the development of targeted therapies.
Results
Global transcriptomes associated with Mφ and Fb activation and their
cell-cell interactions
Transcriptional events related to the interaction of Mφ and Fb in the
context of defined immune and metabolic settings were investigated by
RNAseq (see Fig. [76]S1b for experimental design and sample coding of
the in vitro approach). Principal component analysis (PCA) of
transcriptomes associated with the 44 experimental conditions revealed
that the most relevant sources of variability in the system were the
cell types and their response to inflammatory conditions (PC1 and PC2,
respectively, collectively accounting for 88% of total variance;
Fig. [77]1a). Biological hallmarks were investigated by single-sample
Gene Set Enrichment Analysis (ssGSEA), revealing that hypoxia, the
reactive oxygen species (ROS) pathway, glycolysis, oxidative
phosphorylation, fatty acid metabolism, cholesterol homeostasis, and
TGFβ signaling were the major drivers of differential gene expression
across of the entire dataset. In both Mφ and Fb, proinflammatory
conditions were associated with IFN- and TNF-related pathways, whereas
genes related to epithelial-mesenchymal transition and angiogenesis
were enriched in Fb but not in Mφ (Fig. [78]1b).
Fig. 1. Global transcriptome analysis of macrophages (Mφ) and fibroblasts
(Fb) exposed to distinct immune, metabolic and culture conditions.
[79]Fig. 1
[80]Open in a new tab
a PCA on all samples indicates that the 2 components describing most of
the system variability are represented by the cell type and
inflammation (73% - PC1 and 15% - PC2, respectively, in the upper
panel). Other sources of variability are not detectable (see PC3 and
PC4 in the bottom panel). For sample codes refer to Fig. 1b. b Single
sample GSEA (ssGSEA) on the hallmark database of enriched gene
ontologies. Labels indicate 44 sample conditions (columns) and hallmark
categories enriched (rows). Samples are grouped by cell type (Mφ or
Fb), cellular interaction (CC or SC), immune cues (0 or I or F) and
metabolic cue (H or N; Mφ and Fb in different shades of orange or blue,
respectively). Different shape and color indicate different sample
conditions: squares = resting (0), circles = proinflammatory (I) and
triangles = profibrotic (F); empty and full shapes are related to
single culture (SC) and coculture (CC), respectively; different shade
of color are related to time (4 or 24 h) and oxygen status
(N = normoxia, H = hypoxia). c–h Quantification of differentially
expressed genes (DEG; FDR ≤ 0.05) in Mφ and Fb depending on immune
cues, hypoxia, or cellular interaction. c reports the effect of immune
cues comparing proinflammatory or profibrotic Mφ to resting Mφ
(MIvsM0 = red bars, MFvsM0 = blue bars, respectively) in SC or CC
(empty and full bars, respectively) at 4 or 24 h of normoxia (4 N or
24 N) or hypoxia (4H or 24H). d reports the effect of hypoxia on Mφ
comparing hypoxic to normoxic Mφ in resting (MHvsM0/N = gray bars),
proinflammatory (MI/HvsMI/N = red bars) and profibrotic
(MF/HvsMF/N = blue bars) conditions, in SC or CC (empty and full bars,
respectively), both at 4 or 24 h. e reports the effect of Fb on Mφ
comparing cocultivated to single cultivated Mφ in resting
(M0/CCvsM0/SC = gray bars), proinflammatory (MI/CCvsMI/SC = red bars)
and profibrotic (MF/CCvsMF/SC = blue bars) conditions at 4 or 24 h of
normoxia (4 N or 24 N) or hypoxia (4H or 24H). f–h report the same
comparisons of c–e, respectively, but performed on Fb. Source data of
c–h are provided as a Source Data file.
The contribution of the distinct parameters was then investigated by
multi-level analysis to define the impact of each single variable (1st
level analysis) and the combinatorial effects of two (2nd level
analysis) or three of them (3rd level analysis). The differentially
expressed genes (DEG), selected as described in Methods section, were
then clustered based on the main variable taken into account. When the
analysis was focused on the relevance of the immune network,
inflammatory/Th1 conditions, mimicked by exposing cells to the combined
effect of the proinflammatory mediator lipopolysaccharide (LPS) plus
the type 1 cytokine IFNγ, emerged as a major driver for both cell
types, generating a high number of DEGs in both Mφ and Fb. On the
opposite, profibrotic/Th2 conditions, mimicked by exposing cells to the
type 2 cytokine IL-4, showed a divergent effect between Mφ, which
regulated hundreds of DEGs, and Fb, which did not significantly differ
from their untreated counterpart (Fig. [81]1c, [82]f for Mφ and Fb,
respectively). The relevance of taking into consideration the combined
effects of multiple parameters was exemplified by the analysis on
hypoxia, whose effects in both cell types were significantly enhanced
when cells were activated in coculture conditions as compared to single
cell cultures (Fig. [83]1d, [84]g for Mφ and Fb, respectively).
Similarly, Mφ and particularly Fb exposed to proinflammatory conditions
regulated hundreds of additional DEGs specifically when activated in
the presence of the second cell type (Fig. [85]1e, [86]h for Mφ and Fb,
respectively).
Effects of inflammation on Fb and Mφ are highly influenced by the tissue
microenvironment
As inflammation is a potent immune driver, able to induce phenotypic
and functional changes in different cell types^[87]30,[88]31, we first
focused our analysis on its functional effects on Mφ and Fb
(Fig. [89]2a, [90]c, respectively) comparing LPS + IFNγ-stimulated Mφ
(MI) and Fb (FbI) with resting counterparts in single culture/normoxic
conditions (comparisons A[M] and A[F]), in single culture/hypoxic
conditions (comparisons B[M] and B[F]), in coculture/normoxic
conditions (comparisons C[M] and C[F]), and in coculture/hypoxic
conditions (comparisons D[M] and D[F]). As expected, this analysis
revealed that both Mφ and Fb acquired a proinflammatory phenotype
across different metabolic/culture conditions when exposed to an
inflammatory setting (Fig. [91]S2 and Supplementary Data [92]1).
Keeping the focus on the role of inflammation, we then applied a 2nd
level analysis comparing hypoxic versus normoxic conditions in single
culture (comparison AvsB) and in coculture (comparison CvsD)
conditions, and then a 3rd level analysis comparing single culture and
coculture conditions (comparison [(AvsB)vs(CvsD)]). This analysis
revealed that a high amount of DEGs was shared among all Mφ
comparisons, with the remarkable exception of D[M], which was
characterized by the emergence of 1524 specific DEGs, revealing that
even if inflammation per se is a major Mφ activator the concomitant
presence of Fb and hypoxic conditions adds further complexity to its
functional effects on Mφ (Fig. [93]2b). In line with the robustness of
inflammation as Mφ-activating driver, pathway enrichment analysis
revealed that in Mφ most functional pathways were linked to
inflammation, adaptive immune response and cell senescence, and were
shared through different experimental conditions (Fig. [94]2e). A
remarkable exception was represented by the negative regulation of a
set of functional pathways related to cholesterol metabolism in Mφ
inflamed in the presence of Fb (comparison C[M] in Fig. [95]2e).
Fig. 2. Role of inflammation on the global biological response.
[96]Fig. 2
[97]Open in a new tab
a, c Schematic representation of the multi-level approach to study the
effect of inflammation on Mφ (a, b, e) and Fb (c, d, f) in different
metabolic and culture conditions. 1st level compares LPS+IFNγ
stimulated cells to resting cells in four different conditions:
normoxic and single culture (comparison A[M]/A[F] in pink), hypoxic and
single culture (B[M]/B[F] in blue), normoxic and coculture (C[M]/C[F]
in green), hypoxic and coculture (D[M]/D[F] in yellow). 2nd level
(double arrows) compares hypoxic to normoxic treatment (AvsB and CvsD),
3rd level (triple arrows) compares coculture to single culture
conditions [(A+B)vs(C+D)]. b, d Venn diagrams show the distribution of
DEGs in each 1st level comparison in Mφ (b) and Fb (d). e, f Pathways
enrichment analysis in Mφ (e) and Fb (f) comparisons. Columns represent
1st level of comparisons (A-B-C-D), rows report pathways significantly
modulated (|z-score | ≥ 2) in at least one comparison. Color intensity
bar indicates the level of positive (in red) or negative (in blue)
enrichment; dots appear only when pathways are significantly enriched;
gray indicates no modulation. See also Fig. [98]S2. g Schematic
representation of cellular changes in the key condition which feeds
assumption A in the mathematical model. Source data of heatmaps e, f
are provided as a Source Data file.
Similar to Mφ, inflammatory cues induced also in Fb a high number of
DEGs sustaining functional pathways clearly associated with fibrosis
and cell senescence (Fig. [99]2d, [100]f). Unexpectedly, however, most
of these DEGs were significantly modulated only in Fb cocultivated with
Mφ (comparisons C[F] and/or D[F] in Fig. [101]2f). Focusing on the gene
enrichment analysis in Fb in coculture with Mφ, a lower activation of
several pathways related to inflammation and fibrosis was evident in
the concomitant presence of hypoxia (comparison D[F] in Fig. [102]2f),
while Mφ per se had a significant impact on pathways related to cell
cycle regulation and metabolism (comparison C[F] in Fig. [103]2f).
These findings indicate the emergence of a senescent phenotype of both
Mφ and Fb in inflamed tissues (here mimicked by coculture conditions),
with Mφ modifying their metabolic profile and Fb regulating their cell
proliferation properties, and further subtle metabolic adjustments
related to the concomitant presence of hypoxia. These phenotypic
changes lay the ground for assumption A of our mathematical model
(Fig. [104]2g).
Type 2 immune responses have direct effect on Mφ, which then indirectly
affect Fb
During fibrosis development, adaptive immune responses in the site of
lesion influence both immune and non-immune cells. Th2 cytokines in
particular induce in Mφ a phenotypic switching into an alternative
phenotype (M2, here indicated as MF), while on Fb contribute to
promoting their activation into myofibroblasts involved in ECM
components production (here indicated as FbF)^[105]32. As expected,
IL-4 (here used to mimic the contribution of the Th2 immune pathway)
induced a restricted but well-defined transcriptional profile in Mφ,
which was largely insensitive to the hypoxic context or the concomitant
presence of Fb (Fig. [106]3b and comparisons A[M], B[M], and C[M] in
Fig. [107]S3). On the opposite, IL-4 had no direct effect on Fb, nor
was it able to induce Fb activation when hypoxia and Mφ were added as
single variables (Fig. [108]3d and comparisons A[F], B[F], and C[F] in
Fig. [109]S3 and Supplementary Data [110]2). However, a dramatic change
in the transcriptional profile of both Mφ and Fb was observed when IL-4
was applied to both cell types under hypoxic conditions (Fig. [111]3b,
[112]d). Specifically, when cocultivated in hypoxia and in the presence
of Fb, IL-4-conditioned Mφ showed a negative regulation of pathways
linked to ECM deposition. A similar effect was observed in
IL-4-conditioned Fb, which also showed a negative regulation of
inflammation- and cell growth-related pathways specifically when
exposed to hypoxia and in the concomitant presence of Mφ. Under these
experimental conditions, both cell types also showed a significant
regulation of functional pathways involved in cell-cell contacts,
indicating that a direct Mφ/Fb interaction may be involved. Thus, at
least in this in vitro setting, the development of a Th2 immune
response per se has distinct effects on Mφ and negligible effects on
Fb, but when hypoxic conditions are imposed to the tissue (i.e., Mφ and
Fb are allowed to directly interact with each other) IL-4 profoundly
influences both inflammatory and profibrotic potential of both cell
types (Fig. [113]3g).
Fig. 3. Role of adaptive Th2 immune response (IL-4) on the global biological
response.
[114]Fig. 3
[115]Open in a new tab
a, c Schematic representation of the multi-level approach to study the
effect of Th2 cytokine on Mφ (a, b, e) and Fb (c, d, f) in different
metabolic and culture conditions. 1st level compares IL-4 stimulated
cells to resting cells in four different conditions: normoxic and
single culture (comparison A[M]/A[F] in pink), hypoxic and single
culture (B[M]/B[F] in blue), normoxic and coculture (C[M]/C[F] in
green), hypoxic and coculture (D[M]/D[F] in yellow). 2nd level (double
arrows) compares hypoxic to normoxic treatment (AvsB and CvsD) and 3rd
level (triple arrows) compares coculture to single culture conditions
[(A+B)vs(C+D)]. b, d Venn diagrams with DEGs distribution in each 1st
level comparison in Mφ (b) and Fb (d). e, f Pathways enrichment
analysis in Mφ (e) and Fb (f) comparisons. Columns represent 1st level
comparisons (A-B-C-D), rows report pathways significantly modulated
(|z-score | ≥2) in at least one comparison. Bar color intensity
indicates the level of positive (red) or negative (blue) enrichment;
dots appear only when pathways are significantly enriched; gray
indicates no modulation. See also Fig. [116]S3. g Schematic
representation of cellular changes in the condition with higher number
of DEGs. Source data of heatmaps e, f are provided as a Source Data
file.
Hypoxia effects are insensitive to the specific immune microenvironment but
critical to cell-cell interactions
Reduction of oxygen tension induces a metabolic switch in
cells^[117]33–[118]35. In our experimental setting, hypoxia (1% O[2])
regulated a restricted set of DEGs in Mφ and Fb, including well-known
hypoxia-responsive genes such as GLUT1, VEGFA, CXCR4, and BNIP3
(Fig. [119]S4a–c and [120]S4d–f for Mφ and Fb, respectively).When we
compared the global effect of hypoxia in single cell cultures
(Fig. [121]4 and [122]S5; comparisons A[M] and A[F] for Mφ and Fb,
respectively) with its effects in coculture setting (Fig. [123]4 and
[124]S5; comparisons B[M] and B[F] for Mφ and Fb, respectively), we
observed a dramatic increase in hypoxia-regulated DEGs when the two
cell types were stimulated in tissue-like conditions (Fig. [125]4b,
[126]d, respectively). Conversely, the concomitant presence of
proinflammatory conditions had no significant impact on the effects of
hypoxia on neither Mφ nor Fb, neither in single cell nor in coculture
conditions (comparisons in Fig. [127]S5; C[M] and D[M] for Mφ and C[F]
and D[F] for Fb, respectively). Similar findings were obtained when the
effects of IL-4 were analyzed (Fig. [128]S5; comparisons E[M] and F[M]
for MF and E[F] and F[F] for FbF, respectively).
Fig. 4. Role of hypoxia on the global biological response.
[129]Fig. 4
[130]Open in a new tab
a, c Schematic representation of the multi-level approach to study the
effect of hypoxia on Mφ (a, b, e) and Fb (c, d, f) in different immune
and culture conditions. 1st level compares hypoxic cells to normoxic
cells in four different conditions: resting and single culture
(comparison A[M]/A[F] in pink), resting and coculture (B[M]/B[F] in
blue), LPS+IFNγ and single culture (C[M]/C[F] in green), LPS+IFNγ and
coculture (D[M]/D[F] in yellow); 2nd level (double arrows) compares
coculture to single culture conditions (AvsB and CvsD); 3rd level
(triple arrows) compares resting to proinflammatory conditions
[(A+B)vs(C+D)]. b, d Venn diagrams with DEGs distribution in each 1st
level comparison in Mφ (b) and Fb (d). e, f Pathways enrichment
analysis in Mφ (e) and Fb (f) comparisons. Columns represent 1st level
analysis (A-B-C-D), rows report pathways significantly modulated
(|z-score | ≥ 2) in at least one comparison. Color intensity bar
indicates the level of positive (in red) or negative (in blue)
enrichment; dots appear only when pathways are significantly enriched;
gray indicates no modulation. See also Fig. [131]S5. g Schematic
representation of cellular changes in the key condition which feeds
assumption B in the mathematical model. Source data of heatmaps e, f
are provided as a Source Data file.
Functional analysis performed on macrophages hypoxia-responsive DEGs
showed that hypoxia regulated pathways were related to actin
remodeling, extracellular matrix deposition, and proliferation
(Fig. [132]4e and Supplementary Data [133]3). Of note, these effects of
hypoxia were evident only when Mφ were cocultivated with Fb, and were
abolished when cells were exposed to proinflammatory or Th2 immune
stimuli (Fig. [134]4 and [135]S5). A similar regulation of hypoxia
effects on the cell transcriptome was observed in Fb, where hypoxia
regulated DEGs related to inflammation and leukocyte recruitment
(Fig. [136]4f and Supplementary Data [137]3).
We conclude that hypoxia has a deep impact on both Mφ and Fb
specifically when these cells can establish intercellular networks,
with Mφ acquiring an active and proliferative phenotype and Fb assuming
proinflammatory and angiogenic properties. These findings constitutes
the basis for assumption B of our mathematical model (Fig. [138]4g).
Our results also suggest that this effect is mostly relevant in tissues
exposed to hypoxic conditions in the absence of significant immune
reactions, while the specific contribution of hypoxia to the cells
transcriptome is significantly reduced when inflammatory/immune
triggers concomitantly act on Mφ and Fb (Fig. [139]4e, [140]f).
Functional interactions between Mφ and Fb are enhanced in hypoxic and
inflammatory conditions
Cellular communication is a key element to adopt appropriate complex
responses to stimuli of various origin (mechanic, chemical, immune,
metabolic)^[141]36. We took advantage of our in vitro setting to
specifically investigate how an immune cell, such as the Mφ, interacts
with a stromal cell, represented by the Fb, in different immune and
metabolic contexts and how this interaction affects the response to
these cues. The comparison of cocultivated Mφ and Fb to their single
cultivated counterparts revealed that in resting conditions (i.e.,
normoxia, absence of inflammation and immune challenging) cocultivated
cells did not differ from single cultivated counterparts (1st level
analysis; comparisons A[M] and A[F] in Fig. [142]5 and Fig. [143]S6).
Coculture also did not affect Mφ response to inflammatory triggers,
while the presence of Mφ has significant impact on Fb response to
inflammation (comparisons C[M] and C[F] in Fig. [144]5, respectively,
and Fig. [145]S6). At variance, coculture significantly affected the
activation of both cell types in response to hypoxia, both in the
absence (comparisons B[M] and B[F] in Fig. [146]5 and [147]S6) and in
the concomitant presence of hypoxia and inflammation (comparisons D[M]
and D[F] in Fig. [148]5 and [149]S6). Of note, coculture in
Th2-conditioned environment induced no changes nor in Mφ neither in Fb;
however, IL-4-conditioned Mφ in hypoxic and tissue-like environment are
able to block the switch of Fb into proinflammatory phenotype
(Fig. [150]3g). We conclude that Fb have an impact on Mφ activation
only if this interplay takes place in hypoxic conditions, while Mφ
influence Fb response to hypoxia and to inflammation independently from
oxygen tension levels.
Fig. 5. Role of cellular interplay on the global biological response.
[151]Fig. 5
[152]Open in a new tab
a, c Schematic representation of the multi-level approach to study the
effect of Fb on Mφ (a, b, e) and of Mφ on Fb (c, d, f) in a tissue-like
setting with different metabolic and immune conditions.1st level
compares cocultivated cells to single cultivated cells in four
different conditions: in normoxic and resting (comparison A[M]/A[F] in
pink), in hypoxic and resting (B[M]/B[F] in blue), in normoxic and
inflamed (C[M]/C[F] in green), in hypoxic and inflamed (D[M]/D[F] in
yellow). 2nd level (double arrows) compares normoxic to hypoxic
environments (AvsB and CvsD) and 3rd level (triple arrows) compares
resting to proinflammatory conditions [(A+B)vs(C+D)]. b, d Venn
diagrams show the distribution of DEGs in each 1st level comparison in
Mφ (b) and Fb (d). e, f Pathways enrichment analysis in Mφ (e) and Fb
(f) comparisons. Columns represent 1st level of comparisons (A-B-C-D),
rows report pathways significantly modulated (|z-score | ≥ 2) in at
least one comparison. Color intensity bar indicates the level of
positive (in red) or negative (in blue) enrichment; dots appear only
when pathways are significantly enriched; gray indicates no modulation.
See also Fig. [153]S6. g Schematic representation of cellular changes
in the key condition which feeds assumptions B and C in the
mathematical model. Source data of heatmaps e, f are provided as a
Source Data file.
Functional analysis showed that, when cells were challenged with
hypoxia, coculture significantly enriched the expression of genes
related to intercellular communication in both Mφ and Fb (Fig. [154]5e,
[155]f and Supplementary Data [156]4). These functional networks
related to cell-cell interactions are likely responsible for the
enhanced expression of genes related to cell proliferation observed in
Mφ exposed to hypoxia in the presence of Fb and genes related to
inflammation, particularly evident in Fb exposed to hypoxia in the
presence of Mφ. When hypoxia operates in an inflammatory setting, a
reduction in the number of enriched pathways was observed both in Mφ
and Fb. In functional terms, whereas in these experimental conditions
cocultivated Fb maintained an enriched expression of genes related to
inflammation, cocultivated Mφ showed a significant enrichment only in
pathways related to cell-cell interaction and hypoxia, while a gene
signature related to cell proliferation was not evident anymore.
Finally, Fb were susceptible to signals derived from the presence of Mφ
also in inflammatory settings, irrespective of the presence of hypoxia.
In this setting, their response showed an upregulation of genes mainly
related to cell cycle control, cell response to stress, and senescence.
We conclude that in normoxic conditions the establishment of an
inflammatory milieu in the tissue supports the ability of Mφ to induce
cellular stress and activate senescence programs in Fb. On the
opposite, in absence of an immune challenge, the interplay between Fb
and Mφ occurring in tissues in response to hypoxia modifies the
response of both cell types, promoting the acquisition of proliferative
properties in Mφ and inducing a proinflammatory phenotype in Fb. Of
note, these effects of hypoxia are significantly attenuated when the
Mφ/Fb interplay occurs in an inflammatory setting. These changes in
cell response will be applied as assumptions B and C in our
mathematical model (Fig. [157]5g).
Different tissue districts in fibrotic kidneys are characterized by distinct
patterns of immune infiltrate and transcriptional profiles
These in vitro results indicate the interaction between Fb and Mφ
profoundly influence the functional effects of immunological and
metabolic cues operating in tissues, likely affecting their role in
fibrosis development. Therefore, translation of these in vitro findings
to specific human diseases requires considering the fine anatomy of the
involved organ, which influences the likelihood of direct Mφ/Fb
interactions and the degree of hypoxia.
The kidney is composed of different tissue compartments, which in the
cortical area include the glomeruli and the tubulointerstitial
compartments. It also presents a unique type of vascularization, with
each individual glomerulus connected to the circulation by an exclusive
afferent and efferent arteriolar vessels, the latter continuing into
longitudinally arranged capillaries along the tubules. The distinct
anatomy of the interstitial compartment and the presence of
kidney-specific epithelial boundary structures (e.g., the Bowman
capsule confining glomeruli) were our rationale to investigate the
potential relevance of tissue organization for fibrosis development. We
therefore investigated the cellular composition of different
compartments of the cortical tissue by multiplexed immunohistochemistry
(mIHC) and used laser capture microdissection (LCM) to enrich
transcriptional programs in distinct anatomic districts defined as
reported in the Methods section: the glomerulus; the 200–250 µm
surrounding area (from now referred to as “surrounding”), which
represent the range where cytokine/chemokine gradients control motility
of immune cells and their mutual communication^[158]37,[159]38; the
tubulointerstitial area (from now referred to as “interstitium”).
Comprehensive immune cell phenotyping and transcriptional analysis was
performed in these anatomical areas on transplanted kidneys at
different stages of fibrosis and in control samples, represented by
non-fibrotic tissue specimen obtained from tumor-distant areas of
kidneys removed for renal cell carcinomas.
Conventional microscopy and mIHC showed that different stages and
degrees of fibrosis coexist within a specimen. Some morphologically
almost intact glomeruli and normal-appearing tubular structures were
still present even in kidneys with terminal graft failure after
different types of chronic rejection (Fig. [160]6a, b). In an earlier
stage of fibrosis, a broad range from severely affected to almost
intact glomeruli (Fig. [161]6c, right upper panels) and locally
different degrees of inflammation in interstitial areas (Fig. [162]6c,
right lower panels) were present (Fig. [163]6e, f). Spatial
heterogeneity between tissue compartments was also observed for the
composition of the immune cell infiltrate (Fig. [164]6d). CD4^+ and
CD8^+ T cells showed increased density in the interstitium compared to
the glomerular compartment, and CD20^+ B cells were almost exclusively
present outside the glomeruli. Mφ subsets also showed tissue
compartment-specific distribution, with abundant CD206^+/CD68^+
alternatively activated Mφ highly represented in the interstitium and
almost absent in the glomeruli (Fig. [165]6d). In contrast, activated
(FAP^+) Fb were present in all compartments, but significantly enriched
in the glomeruli (Fig. [166]6g). Large-scale neighborhood analysis
revealed that: (i) CD20^+ B cells were largely clustered with
themselves, with few Mφ in the direct neighborhood; (ii) mixed
infiltrates composed of CD8^+ effector and CD4^+helper T cells tended
to cluster together with dispersed Mφ of different types; (iii) FAP^+
Fb were in relatively close contact to other immune cells in the
interstitial area, while being predominantly located in neighborhood
with each other in glomeruli, suggesting that their activation
mechanisms were different in distinct tissue compartments
(Fig. [167]6h–j).
Fig. 6. Different stages of fibrosis and heterogeneity of immune cell
infiltrates characterizing distinct anatomical kidney regions.
[168]Fig. 6
[169]Open in a new tab
a–c Multiplexed immunohistochemistry of transplant nephrectomy samples
in a acute cell-mediated rejection (TCMR; BANFF type IIa), b chronic
antibody-mediated rejection (ABMR), and c active stage of fibrosis (ca.
90% IF/TA; BANFF category 5, grade III) with signs of chronic rejection
consistent with combined TCMR and ABMR. The left panel of a–c shows
fluorescent whole slide scans for overview in tissue context. The right
three panels show region-specific immune cell phenotyping in multiple
fields of view after multispectral unmixing (see color code in right
panel of a–c). While multiplexed immunohistochemistry was performed in
only a single final experiment, all immunohistochemical staining
results for individual markers (FAP, CD206, CD68, CD4, CD8, CD20) have
been systematically controlled by chromogenic single- and duplex
immunohistochemistry on consecutive sequential sectioning levels
(Fig. 6f, g, and ref. [170]27). Representative examples of the
glomerular compartment (Glo) and the tubulointerstitial areas (Interst)
are displayed. Co-existence of almost normal Glo and tubulointerstitial
areas (a, left panel, arrows) and severely inflamed regions (a, left
panel, asterisks) can be observed. Glomeruli were almost devoid of Mφ
with few exceptions (b, right panel, arrows). If present in glomeruli,
Mφ were rarely in direct neighborhood to FAP^+ myofibroblasts (c, right
panel, arrows). d Scatter plot representing cell density of different
cell types: macrophages (CD68^+/CD206^+, CD68^+/CD206^−), T lymphocytes
(CD8^+, CD4^+), B lymphocytes (CD20^+) and activated myofibroblasts
(FAP^+), in the same case as presented in (c) in each anatomical
compartment (glomeruli (G), Points indicate results for each ROI,
N = 79; surrounding (S), N = 79; interstitium (I), N = 31; center line
represents mean values, whiskers depict SD. Two-way ANOVA with Tukey’s
multiple comparisons test were used to evaluate the differences
(p-values are reported on the graph). e Sirius red staining
(corresponding to c) showing co-existence of morphologically almost
intact Glo and sclerotic glomeruli (sGlo), and tubulointerstitial areas
ranging from severe (**) to moderate/low fibrosis (*). f IHC for FAP
(corresponding to c), showing co-existence of almost normal Glo and
areas of massive myofibroblast activation (sGlo), and
tubulointerstitial areas ranging from severe (**) to moderate/low
myofibroblast activation (*). g Duplex immunohistochemistry
(corresponding to c) confirmed the heterogeneous distribution of
activated myofibroblasts (FAP^+), and alternatively activated CD206^+Mφ
in different anatomical regions. h–j Neighborhood analysis on
multiplexed IHC. For each region, glomeruli (h), interstitium (i) and
surrounding (j), each cell of interest indicated on the horizontal axis
(CD68^+/CD206^+,CD68^+/CD206^−, CD8^+,CD4^+,CD20^+, and FAP^+) was
evaluated for the relative frequency of the corresponding cell types
(vertical axis) within a radius of 7.5 μm. Data are presented as mean
values (top of bar graph) with SD (whiskers). The number of analyzed
cells was (h) CD68^+/CD206^+ n = 99, CD68^+/CD206^− n = 50, CD8^+
n = 18, CD4^+ n = 78, CD20^+ n = 0, and FAP^+ n = 462; iCD68^+/CD206^+
n = 15179, CD68^+/CD206^− n = 1356, CD8^+ n = 3659, CD4^+n = 9286,
CD20^+ n = 4756, and FAP^+ n = 1443; j CD68^+/CD206^+ n = 1072,
CD68^+/CD206^− n = 37, CD8^+ n = 265, CD4^+ n = 829, CD20^+ n = 810,
and FAP^+ n = 69. Source data of the multiplex immunohistochemistry
experiment depicted in Fig. 6d, h–j are provided as Source data files
and in Zenodo repository.
As different tissue compartments showed different patterns of immune
infiltrate, we compared the LCM-enriched sub-compartment-specific
transcriptome profiles of the three distinct anatomical regions of
interest (Fig. [171]S7a). PCA showed that the anatomical area was the
first source of variation among all samples, and that for each
anatomical district investigated samples from control and fibrotic
kidneys were clearly distinct (Fig. [172]S7b). Gene ontology (GO) and
pathways enrichment analysis revealed a number of functional categories
significantly enriched in fibrotic samples, most of which related to
immune response and ECM organization (Fig. [173]S7c). We then
investigated the potential relevance in vivo of the different cellular
networks identified in vitro by defining their relative contribution to
the transcriptional profiles detected in fibrotic reactions occurring
in these different anatomical districts. To this purpose, GSEA and
overlap analysis were applied to define the enrichment of specific
signatures identified by in vitro comparisons in the ex vivo signature
defined by the comparison between fibrotic kidneys and controls.
Overall, our in vitro signatures were able to explain 45.3% of total
DEGs that discriminated fibrotic kidneys from controls. This percentage
included genes shared by more than one signature (35.6%) as well as
genes only present in a single signature (9.7% in total). A limited
number of experimental in vitro conditions could explain each a
significant fraction (up to 20%) of the ex vivo gene signature
(Fig. [174]S7d, middle panel). Most of these gene sets were shared
among two or more settings, but some provided a univocal contribution
(i.e., DEGs detected in the ex vivo signature only present in one
specific in vitro experimental comparison) shading light on the
biological processes occurred in the tissue. Gene sets with a univocal
signature were grouped depending on the discriminating variable that
defines in vitro comparison. We observed that univocal signatures that
better describe fibrotic kidneys presented inflammation as a major
trigger, being other variables investigated in the in vitro system
(IL-4, cell-cell interaction, hypoxia) able to explain only a minor
number of genes characterizing fibrotic kidneys (Fig. [175]S7d, left
panel). Considering both univocal and general contribution of in vitro
signatures, we observed that genes associated with leukocyte activation
and inflammatory events occurring in fibrotic kidneys were mostly
overlapping with our in vitro data related to Mφ subjected to
inflammatory stimuli mainly in coculture setting, independently of
oxygen level. On the other side, in vitro signatures describing Fb
subjected to Mφ influence in hypoxic environment provided a description
of fibrotic and angiogenic properties characterizing fibrotic kidneys
compared to control samples (Fig. [176]S7d, right and middle panels).
Enrichment of proinflammatory signatures outside glomeruli during renal
allograft rejection
When distinct anatomical regions were investigated, PCA confirmed the
clustering of fibrotic and control samples in each district
(Figs. [177]7a, [178]8a, and [179]9a for the interstitium, the
surrounding area, and glomeruli, respectively). Both in the
interstitium and in surrounding regions we confirmed the enrichment of
inflammatory pathways in pathological samples, with GO terms related to
leukocyte activation and migration, adaptive immune response, cellular
response to IFNγ, regulation of cytokine production being clearly
enriched in fibrotic samples (Figs. [180]7b and [181]8b). Our in vitro
signatures explained about 60% of the fibrosis-related signatures
emerging in these two anatomical districts, with an evident
predominance of signatures related to inflammation testified by 10% of
DEGs included in functional categories emerging exclusively in response
to inflammatory triggers used in vitro. In both anatomical districts,
the in vitro generated signature contributing the most to DEGs observed
in vivo was related to Mφ response to inflammatory stimuli, requiring
however the concomitant presence of Fb and hypoxia (MIvsM0_CC/24H
signature; Figs. [182]7c and [183]8c). Of note, in the interstitium all
signatures providing significant contribution to explain DEGs observed
ex vivo were related to response to inflammation, either by Mφ or Fb.
At variance, a distinct signature associated with Mφ response to IL-4,
again requiring the concomitant presence of Fb and hypoxia, emerged
only in the surrounding tissue (MFvsM0_CC/24H signature; Fig. [184]8c,
left and middle panels), and appeared to be specifically related to the
GO category ECM organization (Fig. [185]8c, right panel). In vitro
signature enrichment was validated and confirmed also by GSEA analysis
(Fig. [186]S8, violet and orange boxes).We conclude that, both in the
interstitium and in the surrounding area, proinflammatory signatures
explain most of DEGs characterizing fibrotic samples, with many DEGs
provided by inflammatory Mφ activated in the presence of hypoxia
(Figs. [187]7c and [188]8c, right panels). Interestingly, the presence
only in the surrounding area of a signature associated with Mφ response
to IL-4 suggests that different biological networks support fibrosis
development in these two anatomical districts.
Fig. 7. Gene overlap analysis of ex vivo tubulointerstitial area from
fibrotic kidneys with in vitro signatures.
[189]Fig. 7
[190]Open in a new tab
a PCA of ex vivo interstitium samples. Triangles indicate fibrotic
kidneys, dots control kidneys. b Bar graph reports pathways
significantly enriched (log[10] (p-value) ≥1.3) in fibrotic vs control
interstitium samples as calculated by Metascape software (for
statistical type of analysis see ref. [191]45). c Overlap analysis was
performed comparing DEGs of ex vivo fibrotic vs control interstitium
with DEGs of each in vitro comparison (which defined 78 corresponding
signatures). Pie chart reveals that 49.9% of ex vivo signature is
explained by at least one in vitro signature (in light green); 10.8% is
specifically explained by only one in vitro signature and different
colors are referred to the discriminating variable: inflammation in
red, IL-4 stimulation in green, cell-cell interaction in blue, hypoxia
in yellow. The remaining 39.3% of the ex vivo interstitium signature
(633 genes) is not explained by in vitro model (in gray). Dot chart
reports on the vertical axis 19 in vitro signatures enriched in ex vivo
model and their relative contribution with unique expressed genes:
white numbers in each dot represent relative percentages on the
totality of explained and not explained genes. The central heatmap
highlights the overlapping percentage of the same 19 in vitro
signatures on the ex vivo signature. On the diagonal, numbers indicate
the overlapping percentage of each signature; the other numbers explain
the overlapping percentage shared between two in vitro and the ex vivo
signatures. Color intensity indicates the overlapping percentage level
referred to ex vivo signature. Arrows near each in vitro signature
indicate the positive (red) or negative (blue) enrichment of the first
term of in vitro comparison (see also Fig. [192]S8). The heatmap on the
right reports Metascape pathways in ex vivo comparison (b) with the
percentage of enrichment given by each in vitro signature (on the
left).
Fig. 8. Gene overlap analysis of ex vivo glomeruli surrounding area from
fibrotic kidneys with in vitro signatures.
[193]Fig. 8
[194]Open in a new tab
a PCA of ex vivo glomeruli surrounding area samples. Triangles indicate
fibrotic kidneys, dots control samples. b Bar graph reports pathways
significantly enriched (log[10] (p-value) ≥1.3) in pathological vs
control surrounding samples as calculated by Metascape software (for
statistical type of analysis see ref. [195]45). c Overlap analysis was
performed comparing DEGs of ex vivo fibrotic vs control surrounding
areas with DEGs of each in vitro comparison (which defined 78
corresponding signatures). Pie chart reveals that 52.7% of ex vivo
signature is explained by at least one in vitro signature (in light
green); 10.1% is specifically explained by only one in vitro signature
and different colors are referred to the discriminating variable of in
vitro signatures: inflammation in red, IL-4 stimulation in green,
cell-cell interaction in blue, hypoxia in yellow. The remaining 37.2%
of the surrounding ex vivo signature (2685 genes) is not explained by
in vitro model (in gray). Dot chart reports on the vertical axis 24 in
vitro signatures enriched in ex vivo model and their relative
contribution with unique expressed genes: white numbers in each dot
represent relative percentages on the totality of explained and not
explained genes. The central heatmap highlights the overlapping
percentage of the same 24 in vitro signatures. On the diagonal, numbers
indicate the overlapping percentage of each signature; the other
numbers explain the overlapping percentage shared between two in vitro
and the ex vivo signatures. Color intensity indicates the overlapping
percentage level referred to ex vivo signature. Arrows near each in
vitro signature indicate the positive (red) or negative (blue)
enrichment of the first term of in vitro comparison (see also
Fig. [196]S8). The heatmap on the right reports Metascape pathways in
ex vivo comparison (b) with the percentage of enrichment given by each
in vitro signature (on the left). Orange stars highlight pathways that
are specifically enriched in surrounding nephrectomies and are not
shared with other renal districts (interstitium, glomeruli).
Fig. 9. Gene overlap analysis of ex vivo glomeruli from fibrotic kidneys with
in vitro signatures.
[197]Fig. 9
[198]Open in a new tab
a PCA of ex vivo glomeruli samples. Triangles indicate fibrotic
kidneys, dots control samples. b Bar graph reports pathways
significantly enriched (log[10] (p-value) ≥1.3) in fibrotic vs control
glomeruli samples as calculated by Metascape software (for statistical
type of analysis see ref. [199]45). c Overlap analysis was performed
comparing DEGs of ex vivo fibrotic vs control glomeruli with DEGs of
each in vitro comparison (which defined 78 corresponding signatures).
Pie chart reveals that 42.3% of ex vivo signature is explained by at
least one in vitro signature (in light green); 10.2% is specifically
explained by only one in vitro signature and different colors are
referred to the discriminating variable of in vitro signatures:
inflammation in red, IL-4 stimulation in green, cell-cell interaction
in blue, hypoxia in yellow. The remaining 47.5% of the ex vivo
glomeruli signature (255 genes) is not explained by in vitro model (in
gray). Dot chart reports on the vertical axis 12 in vitro signatures
enriched in ex vivo model and their relative contribution with unique
expressed genes: white numbers in each dot represent relative
percentages on the totality of explained and not explained genes. The
central heatmap highlights the overlapping percentage of the same 12 in
vitro signatures on the ex vivo signature. On the diagonal, numbers
indicate the overlapping percentage of each signature; the other
numbers explain the overlapping percentage shared between two in vitro
and the ex vivo signatures. Color intensity indicates the overlapping
percentage level referred to ex vivo signature. Arrows near each in
vitro signature indicate the positive (red) or negative (blue)
enrichment of the first term of in vitro comparison. (See also
Fig. [200]S8). The heatmap on the right reports Metascape pathways in
ex vivo comparison (b) with the percentage of enrichment given by each
in vitro signature (on the left). Green stars highlight pathways that
are specifically enriched in glomeruli nephrectomies and are not shared
with other renal districts (interstitium, surrounding area).
Enrichment of IL-4 conditioned signature in fibrotic glomeruli during renal
allograft rejection
Pathological glomeruli from fibrotic kidneys were clearly distinct from
the corresponding control samples (Fig. [201]9a) and showed enrichment
in pathways related to ECM organization, blood vessel morphogenesis,
and cellular adhesion (Fig. [202]9b). In vitro signatures were clearly
different from signatures that enriched the other renal districts
previously described. Indeed, the contribution of inflammation-related
pathways was reduced (1.2%; Fig. [203]9c, left panel), and a unique
signature corresponding to IL-4-activated Mφ in the concomitant
presence of Fb and hypoxia emerged, which contributed the most to
describe DEGs enriched in the glomeruli area from fibrotic kidneys
(MFvsM0_CC/24H; Fig. [204]9c, left and central panels). Of note, this
signature explained more than 50% of DEGs that contributed to GO terms
related to ECM deposition and blood vessels in these pathological
samples (Fig. [205]9c, right panel). Signatures related to Fb
interacting with Mφ, in hypoxic conditions but independently from
inflammation, also significantly contributed to the description of
glomeruli from fibrotic kidneys (CCvsSC_FbI/24H, 24Hvs24N_Fb0/CC,
CCvsSC_Fb0/24H; left and central panels in Fig. [206]9c, green box in
Fig. [207]S8). Interestingly, these Fb-related in vitro signatures
explained a large fraction (from 30% to 70%) of DEGs associated with
different GO categories related to cytokine signaling, immune response,
cell adhesion, and chemotaxis (Fig. [208]9c, right panel). However,
while all Fb-related GO categories were not specific for pathological
glomeruli, as genes that enriched those pathways were also contributing
to the interstitium and surrounding areas, most Mφ-related signatures
were only evident in this anatomical compartment, suggesting that IL-4
conditioned Mφ interacting with Fb in hypoxic areas provide a specific
spatial-restricted driver in the glomeruli area in fibrotic kidneys.
A mathematical model based on in vitro signatures identifies three dynamic
states in fibrosis development
The combination of our in vitro and ex vivo transcriptomic approaches
identified the key parameters driving the interplay between Mφ and Fb
in different biological settings, their specific functional
implications, and their correspondence with fibrotic events occurring
in vivo in different anatomical regions (Fig. [209]S1a). Here, we
exploited this information to develop a mathematical model able to
translate the Mφ/Fb interactions in vitro identified into their
corresponding in vivo dynamics (Fig. [210]10a). Specifically, the model
was based on literature information showing that inflammatory Mφ are
mainly sustained by local recruitment in inflamed tissues while
alternative Mφ accumulation can be sustained by their proliferative
abilities^[211]39, and that Mφ phenotypic switching can occur
relatively fast and at shorter time point compared to their
proliferation/death rate^[212]40, integrated by the three key
observations emerged in our in vitro analysis related to inflammation
and hypoxia:
* A.
when inflammatory conditions develop in the absence of significant
hypoxia, both Mφ and Fb develop a proinflammatory phenotype but
also activate a senescence program (comparisons C[M] and C[F] in
Fig. [213]2e, [214]f);
* B.
in hypoxic conditions, Mφ acquire proliferative properties while Fb
assume a proinflammatory phenotype (comparisons B[M] and B[F] in
Fig. [215]4e, [216]f, comparisons B[M] and B[F] in Fig. [217]5e,
[218]f);
* C.
when inflammatory conditions and hypoxia are combined, Fb remain
proinflammatory while Mφ become involved in regulation of
epithelial-mesenchymal transition and angiogenesis (comparisons
D[M]and D[F] in Fig. [219]5e, [220]f).
Fig. 10. Mathematical model based on in vitro data and exemplary clinical
observations reflecting the predicted outcomes.
[221]Fig. 10
[222]Open in a new tab
a Schematic representation of Mφ (MI in red and MF in green) and Fb
interactions under hypoxia (H) and inflammation (Y). Details are
provided in Methods section. b Phase diagram of mathematical
prediction. Three phases (I-II-II) were identified on different levels
of inflammation (horizontal axis) and hypoxia (vertical axis). The red
line between phases II and III is the separatrix. c Nine exemplary
cases (cases 1–9, for clinical details see Supplementary Table [223]1)
reflecting the stable “Phase I”: The curves depict the approximate
glomerular filtration rate (GFR) over time (for exact GFR measurements
and time points of biopsies, see Fig. [224]S11). Pie charts describe
the individual immune cell infiltrate (lymphocytes and Mφ) evaluated by
quantitative digital pathology after multiplex immunohistochemistry
(Mplx) (source data are provided as Source data file). The percentage
of immune cells (color code in the figure legend indicates the immune
phenotype) in comparison to all nucleated cells in the biopsy as
defined by DAPI nuclear staining (gray area in the pie charts) can
massively vary between almost no inflammation (e.g., case 3, 8) and
abundant infiltrates (e.g., cases 5, 6, 9). d Graphical depiction of in
silico simulations (left panels) show the solution trajectory and the
predicted dynamics of Mφ and Fb depending on time (horizontal axis) and
cell fraction (vertical axis) for each of the three phases as predicted
by the mathematical model in more detail. The right panel shows
corresponding representative clinical constellations that confirmed the
hypotheses generated by computer simulations, based on the exemplary
cases 10 (unfavorable outcome with irreversible “hot” and “cold”
fibrosis) and case 11 (favorable outcome despite “borderline” immune
infiltration). The color legend indicates the different cell types as
detected my multiplexed immunohistochemistry. Note the abundant CD206^+
macrophages in the indicated biopsy of case 11 in “Phase II” in the
surrounding area (upper image), inside the glomerulus (middle image)
and in the interstitial compartment (bottom image), and in case 10 in
“Phase III” the co-existence of areas with many CD206^+ macrophages
(“hot” fibrosis) and low density of this phenotype (“cold” fibrosis)
co-localized with myofibroblast activation (FAP^+ cells) in the same
biopsy as predicted by in silico modeling.
The above has been translated into the assumptions A1–A5 (see Methods
section), which allowed us to build our mathematical model. It is worth
mentioning that as in vitro parameters can largely deviate from the in
vivo (e.g., human macrophages do not proliferate in vitro but in vivo
they do so), the model parameters have not been calibrated to the in
vitro kinetics. Instead, we have non-dimensionalised the model and
exhaustively explored the corresponding parametric space. After model
analysis, we identified three dynamic phases/regimes resulting from the
inflammation/hypoxia interplay (Fig. [225]10b), which were reflected in
renal biopsies representative of cases with different clinical courses
(Fig. [226]10c) and in cases experiencing clinically manifest rejection
(Fig. [227]10d). Interestingly, qualitatively these dynamics regimes
were robust under large variations of the model parameters (for details
see Figs. [228]S9 and [229]S10).
Phase I is characterized by a low degree of hypoxia, independently of
inflammation that could range from low to high levels (green area/panel
in Fig. [230]10b, [231]d, respectively). This regime is characterized
by model conditions where the predictive cell trajectory starts with an
arbitrary number of cells and ends with a particular amount of Mφ
dependent on the local inflammatory signals, while all conditions are
characterized by a very low number of Fb. A specific example of a
clinical situation reflecting this stable quasi-healthy state can be
found in “protocol” biopsies, which are obtained from patients at a
defined time point after uncomplicated renal transplantation, as
exemplified by cases of variable inflammation in the protocol biopsy 6
months after renal transplantation and favorable long-term clinical
course (stable renal function over 2–5 years as shown in nine exemplary
cases (cases 1–9, Fig. [232]10c)): various levels of inflammation can
be present, but since there is no relevant hypoxia the system can
always return to normal. Another example consistent with phase I could
be considered “time-point zero” biopsies taken at very early time
points during or shortly after the uncomplicated surgical procedure as
shown in two exemplary cases (case 10 and 11, Fig. [233]10d, green
upper panel). In this setting, very close to the physiological
situation of the kidney in the donor, the recipient’s immune system did
not yet have time to respond to the allograft, and consequently we do
not yet observe severe inflammation.
Phase II, a bistable dynamic regime that can lead either to healthy or
fibrotic states, is characterized by wide range of tolerable
inflammation and hypoxia levels (blue area/panel in Fig. [234]10b,
[235]d, respectively). The critical quantity for this regime is the
initial condition of Mφ and Fb. In particular, there is a separatrix
indicating that only high Mφ/low Fb or vice versa may lead to a healthy
renal state, with all the other initial conditions with moderate or
high Mφ and Fb levels are predicted to evolve into fibrosis
development. Clinical examples reflecting such bistable condition
include so-called “borderline” states, mostly observed in “indicated”
biopsies that are obtained for diagnostic purposes in clinical
situations requiring a decision on potential adaptation of the current
immunosuppressive therapy. “Borderline” is a microscopic pattern
defined by the Banff classification as evident inflammation that is
suspicious, but not sufficient to justify the diagnosis of rejection.
Borderline states, but also some cases of manifest rejection can still
resolve, e.g., case 11 that resolved into long-term sufficient renal
function after “borderline” diagnosis in an indicated biopsy after 65
months in (Fig. [236]10d). Both clinical conditions, borderline and
some cases of manifest rejection can be successfully treated and may
return to a healthy state, but a chance of turning into a profibrotic
state remains.
Phase III (red area/panel in Fig. [237]10b, [238]d, respectively) is
the fully pathological, compromised state with an unfavorable
combination of hypoxia and inflammation. It can be associated with
“hot” or “cold” fibrosis (with or without persisting presence of immune
cells, respectively). In case of “hot” fibrosis, both Mφ and Fb are
highly concentrated, whereas in case of “cold” fibrosis only the number
of Fb remains high. Exemplary cases that clinically represent such
states include persistent viral infection with BK virus (Fig. [239]10d,
case 10, left panel of the clinical examples) and irreversible
end-stages after chronic rejection (Fig. [240]6a, b).
The mathematical model predicts a potential favorable outcome for
phases I and II, which is consistent with the clinical observation of
successful transplants in case of high Mφ infiltration but low Fb
number, even in the presence of dense immune cell infiltrates
(Fig. [241]10c). Of note, we found an explicit equation separating
stable (phases I and II) from clinically relevant conditions (phase
III) which dictates:
[MATH: Hypoxia<Pro−inflammatory activityFibroblast viability :MATH]
where proinflammatory activity corresponds to the combined effect of
proinflammatory Mφ polarization (λ), proinflammatory T- and B-cell
presence (y) and anti-inflammatory decay rate and Fb viability is the
ratio of Fb proliferation/apoptosis (mathematical details are in
Methods section).
Discussion
Although current treatments for fibrotic diseases typically target the
inflammatory response, there is accumulating evidence that the
mechanisms driving fibrogenesis are distinct from those regulating
inflammation^[242]1. The key cellular mediator of fibrosis is the
myofibroblast, which is activated by a variety of mechanisms including
autocrine factors and paracrine signals derived from immune cells. In
particular, in tissue repair and fibrosis Mφ exert relevant effects,
which go beyond their proinflammatory activities and include Fb
activation^[243]41. As highlighted by ref. [244]25, the reciprocal
Mφ-Fb interaction is a key component for understanding observations
such as the fibrotic time-window, the long timescale of scar
maturation, and the paradoxical effect of Mφ depletion. Equally
important are the environmental metabolic status and the
cytokines/growth factors networks, being these parameters capable to
influence Mφ functions and their ability to interact with neighboring
cells, including Fb, as well as other immune cells such as lymphocytes.
In our study, we implemented these immune and metabolic stimuli in a
simplified in vitro system where cell circuits recreated a tissue-like
scenario, giving at the same time the possibility to discriminate the
weight of single variables and their combinatorial effects in the
progression of the fibrotic process.
Mφ and Fb were not influencing each other when interacting in resting
conditions, suggesting that cellular contacts without external stimuli
are insufficient to induce major phenotype/functional changes. As
reported, hypoxic conditions per se were able to activate a specific
signature in both cell types, but surprisingly when the hypoxic
environment acted on the two cell types directly interacting with each
other, both Mφ and Fb regulated a significantly larger set of DEGs.
This effect constitutes the first observation used as basis of our
mathematical model. Similarly, the well-known strong response of Mφ to
inflammatory mediators was highly influenced by the presence and
absence of hypoxia and Fb. The concomitant exposure to inflammatory
stimuli and hypoxic coculture conditions also influenced Fb, which
acquired a clear proinflammatory phenotype, while Mφ remained involved
in ECM remodeling and angiogenesis. This effect constitutes the second
observation at the basis of our mathematical model. Instead, when Mφ
and Fb were cocultivated in normoxic but proinflammatory environment,
they both acquired proinflammatory properties, suggesting that in this
context proinflammatory factors were sufficient to promote phenotypic
changes without the contribution of metabolic switch induced by
hypoxia. Moreover, both cell types downregulated oxidative
phosphorylation and the EIF2 signaling pathways, indicating the
activation of senescence processes that we suppose necessary to the
exhaustion of inflammation. This effect constitutes the third
observation included in our mathematical model. Finally, while
profibrotic stimulation by IL-4 promoted in Mφ the alternative
activation phenotype, which was not significantly influenced by neither
the direct contact with Fb nor the presence of hypoxia, Fb conditioned
with IL-4 did not respond in hypoxia, neither when in contact with Mφ.
However, IL-4 conditioned Fb differed from the resting ones, which
changed their phenotype in hypoxic environment when in contact with Mφ.
These results suggest that IL-4 per se has no impact on Fb, but through
conditioning of Mφ it induces an inhibitory mechanism preventing Fb
from acquiring a proinflammatory phenotype. Indeed, we can suppose that
though IL-4 is not sufficient to induce the acquisition of profibrotic
properties in Fb, through Mφ it is able to influence Fb and their
response to hypoxia.
Taken together, these observations obtained in an integrated
tissue-like context candidate hypoxia, inflammation, and the Th2
cytokine IL-4 as key regulators of fibrosis. Consistent with this, the
in vitro signatures generated in our in vitro model were able to
describe up to 50% of the ex vivo signatures generated from laser
microdissected fibrotic transplant nephrectomies, a surprisingly high
percentage considering the variety of cell types and factors acting in
the tissue that were not taken into account in the model. Moreover, we
were able to distinguish different signatures, suggesting enrichment
for variable functions, in three distinct anatomical regions of
kidneys: interstitium, glomeruli surrounding area, and the compartment
of glomeruli themselves. A recent work by Barwinska et al. has reported
a 15-gene signature list characterizing different kidney anatomical
regions^[245]42. We observed a good overlap of these signatures with
our results on healthy tissues, which validated all the 15 markers
specifically associated with glomeruli and 7 out of the 15 markers,
likely due to higher complexity in delineating this second region.
Specifically, our approach of micro-dissecting a fixed radius around
glomeruli results in bulk signatures composed of epithelial and
interstitial elements, which can be expected to be distinct from
profiling the tubular structures separately. In contrast to the
considerable overlap in healthy samples, the transcriptomic profile
associated with the IF/TA interstitium of our kidney transplant
rejection cases showed only a limited overlap (4.9%) with the IF/TA
interstitium of diabetic patients reported by Barwinska et al.,
indicating that IF/TA can be driven by fundamentally different
pathological processes in different clinical conditions. Unfortunately,
we could not evaluate whether this also applies to glomeruli as this
anatomical compartment has not been molecularly characterized in the
diabetic patients study.
A prominent proinflammatory feature of the interstitial compartment,
with immune response and leukocyte activation, was already evident from
the GO analysis performed on the transcriptome of the nephrectomy ex
vivo samples. In addition, the in vitro signatures of Mφ cocultivated
with Fb in hypoxic and proinflammatory conditions shared the
proinflammatory features as major source of variation and were able to
explain 2.7% of differentially expressed genes in this anatomical
region univocally. Moreover, interstitium showed a higher density of
Mφ, T and B lymphocytes than the other regions suggesting a massive
inflammatory response in this area. In the glomeruli surrounding
region, inflammation remained an important feature, similarly but at
lower levels than in the interstitium, but in addition the
transcriptomic profile of this compartment showed also a slightly
increased relevance of processes linked to ECM remodeling. This
corresponded to enrichment of transcriptomic signatures connected to Mφ
IL-4 stimulation that were obtained by in vitro experiments under
hypoxic conditions with Fb coculture. Finally, the transcriptome of the
glomerular compartment showed a strikingly different composition: in
these areas we found predominant signatures of Fb and enrichment for
gene sets clearly suggesting profibrotic behavior with a reduction of
proinflammatory properties. The trend towards gradually declining
levels of inflammation from high in the interstitium, lower in
glomeruli surrounding area, and lowest in glomeruli was paralleled by
an opposite for signatures of fibrosis for which enrichment was least
prominent in the interstitium, increased in the glomeruli surrounding
area and was highest in the glomeruli. Corresponding to the
transcriptomic profiles, IHC showed in some glomeruli aggregates of
FAP^+ activated Fb in largely isolated localization, suggesting regions
of “cold” fibrosis as proposed by ref. [246]25, where no direct
interaction with Mφ or other immune cells is required to maintain a
profibrotic state. Moreover, the same microscopic section of renal
tissue could concomitantly include glomeruli with different density of
FAP^+ cells, and different degrees of fibrosis. This spatial
heterogeneity of the extent of fibrosis and profibrotic states within a
kidney can be predicted, based on the mathematical model that we have
developed on the basis of in vitro observations. Given that the model
predicts the existence of three dynamic regimes that can modify the
interplay between Mφ and Fb depending on local hypoxia and
inflammation, it can be expected that the regional outcome in a kidney
with its complex vascular system, potential pre-existing pathological
conditions and variable presence of Mφ and Fb due to locally different
of focal inflammatory infiltrates will be heterogeneous. The predicted
distinct cellular trajectories, which could lead to different degrees
of stability ranging from stable healthy to irreversible pathological
conditions, is reflected by immunohistological data confirming spatial
heterogeneity with the co-existence of hot and cold fibrosis regions.
Though based on a single case, the analysis of renal biopsies from a
SV40-positive case with reactivation of the polyomavirus BK suggests
that the mathematical prediction is not necessarily specific for graft
rejection.
Taken together, the findings confirm that Fb and Mφ are central to the
development of fibrosis, and show that additional local
microenvironmental cues such as inflammation, Mφ polarization, and
hypoxia have relevant impact on the regional outcome of profibrotic
constellations. Consequently, therapeutic interventions in order to
avoid or reduce fibrosis may have variable effects on the delicate
balance between different factors, including oxygenation, the degree
and type of inflammation, and local anatomical conditions in different
kidney compartments. This may have some intriguing therapeutic
implications. On a first note, our results confirm the plausibility of
anti-inflammatory drug usage, which is the major strategy in the
current Standard of Care. However, the model-based predictions reveal
that a critical inflammation intensity highly depends on local hypoxia
and cell interactions, which may explain why anti-inflammatory
treatments may fail to fully prevent fibrosis. Moreover, interventions
controlling Fb proliferation, for example, PDGFR-targeted drugs such as
Nintedanib, may be helpful against fibrosis development. However, the
inhibitory effect on Fb proliferation rates and the overall outcome may
again depend on local inflammation and hypoxia, implying that
multimodal approaches optimizing immunological and metabolic cues may
be the most promising application of such drugs. The same holds true
for drugs modifying Mφ reprogramming, which could potentially
contribute to effective treatment strategies, but only under certain
microenvironmental circumstances. For example, the present model would
suggest that this could only be successful when Mφ phenotypic switch
rate (λ) is larger than the Fb viability rate β/δ2
(proliferation/death). Overall, therapeutic interventions avoiding or
significantly reducing development of fibrosis in anatomically complex
organs like the kidney are unlikely to be successful in the setting of
unimodal treatment.
Results reported here indicate that combining in vitro/ex vivo
transcriptomics and in silico modeling is a powerful approach to
dissect molecular and cellular mechanisms underlying complex biological
processes and possibly support the discovery and development of new
therapeutic strategies. It is equally important however to define the
main limitations of this approach. First, the restricted availability
of biopsy-based clinical data largely limited the parametrization of
cell kinetic to human in vivo cell dynamics. On the other hand,
calibrating the model to in vitro parameters would have had little
relevance to any clinical scenario. Therefore, we chose an alternative
route where we explored model dynamics for large biologically relevant
parametric regimes. In turn, we focused on the robustness of the three
identified dynamic fibrotic states under parametric variations (see
Figs. [247]S1a, [248]S9, [249]S10). Second, the model indirectly
included the impact of main immunological cell populations, such as T
cells. Although their lump effect has been taken into account by the
inflammation parameter (y), the full complexity of immune interactions
is not yet reflected and could confer further interesting implications.
This is a limitation, but it also suggests a clear path for extending
the model. Third, our model involved only mean-field non-spatial
dynamics. The importance of space/anatomy has been exemplified in the
analysis of different tissue specimens in terms of immune constitution
and transcriptomic profile. In our model each tissue district would
correspond to a different inflammation value (y) and hypoxia dynamics
(H). This could not be realized within the given limitations in
resources and represents a promising extension of this model. Fourth,
although including general aspects of fibrosis, the clinical use case
for this model is specifically limited to ex vivo findings related to
IF/TA progression and glomerulosclerosis in transplanted kidneys. We
present this as a relevant clinical example, as any transplantation
inevitably involves severe hypoxia (e.g., the cold ischemia during
surgery and transport) and inflammation through rejection (except
genetically identical twin transplantation). The qualitative relevance
of the model itself may be broader than this clinical use case, as it
has been developed based on in vitro data generated without any
reference to a specific tissue. Nevertheless, its application to
distinct diseases will require further consideration of specific
pathogenic mechanisms.
Methods
Cell cultures
Human monocytes were obtained from healthy blood donor buffy coats,
upon approval by the local ethical committee. Monocytes were isolated
by two-step density gradient centrifugations using Lympholyte H
(Cederlane) and 46% Percoll (Lonza) followed by incubation of purified
cells in RPMI 1640 (Lonza) without serum, for 20 min at RT. Adherent
monocytes were washed twice with PBS and then cultured in RPMI medium
supplemented with 10% fetal bovine serum (FBS; Lonza), 100 U/mL
penicillin/streptomycin (Lonza), and 2 mM L-glutamine (Lonza). Monocyte
purity was >90% as assessed by CD14/CD16 FACS analysis^[250]43. Mφ were
obtained by culturing monocytes for 7 days in complete RPMI
supplemented with human M-CSF (100 ng/ml; Miltenyi). The human dermal
BJ fibroblast cell line (CRL2522; ATCC) was cultivated in high glucose
D-MEM (Lonza) 10% FBS, 100 U/mL penicillin/streptomycin, and 2 mM
L-glutamine. When cultivated in normoxic conditions, Mφ and Fb were
maintained at 37 °C in a humidified incubator settled at 20% O[2], 5%
CO[2] in air, while hypoxic treatment was performed moving cells at
37 °C in a humidified incubator with a mixture of 1% O[2], 5% CO[2] and
94% N[2]. Hypoxic and normoxic cells are labeled “H” and “N”,
respectively. Mφ were polarized toward a proinflammatory phenotype (MI)
by incubation with 100 ng/ml LPS (Sigma) plus 20 ng/ml IFNγ (R&D
Systems) or into an alternative phenotype (MF) by incubation with
20 ng/ml IL-4 (Miltenyi)^[251]44. Resting Mφ (M0) were left
unstimulated for the same period. Fb were stimulated as Mφ, with cells
treated with LPS + IFNγ (FbI), IL-4 (FbF) or left unstimulated (Fb0).
Polarizing stimuli and hypoxia were applied simultaneously. For
coculture experiments, differentiated Mφ were replated directly onto
adherent Fb (plated 16 h before) with a 2:1 ratio. After 24 h of
coculture in basal conditions (normoxia without stimuli), cells were
stimulated as described above, detached, and FACS sorted (FACS Aria
III; BD Bioscience) based on staining with anti-human CD45 clone 2D1
(dilution 1:1000; Cat.No: 560178; BD Bioscience) to distinguish CD45^+
Mφ from CD45^− Fb using a FACS Aria III cell sorter (BD Bioscience).
Zombie Aqua Fixable Viability kit (Cat.No: 423101; BioLegend) was used
to exclude dead cells (L/D^+) (see Fig. [252]S1c for the gating
strategy and Source Data file for sorted cell numbers). Single cell and
FACS-sorted cocultivated cells are labeled “SC” and “CC”, respectively.
The combination of different parameters applied to Mφ and Fb generated
a total of 44 different experimental conditions, detailed in
Fig. [253]S1b. The experiment was performed in triplicate and the
transcriptomic profiles were then investigated.
Histological samples
Immune cell phenotyping and transcriptional analysis were performed on
formalin-fixed paraffin-embedded 3 µm thick consecutive sections from
archival material of four transplant nephrectomy specimens (Cases A-D,
see Supplementary Table [254]1) that were selected to represent
different stages and underlying causes of fibrosis. The cases include
three explanted kidneys that lost function due to previous TCMR and/or
ABMR, and one nephrectomy specimen surgically removed because of the
development of renal cell carcinoma within the transplanted
organ^[255]27. As non-fibrotic control samples we used tissue selected
as distant as possible from the tumor margin form four tumor
nephrectomy specimens. Protocol biopsies and indicated biopsies from
kidney transplants were obtained from a previous study^[256]28, and the
clinical follow-up (development of GFR over time) was obtained in the
context of the SYSIMIT systems medicine study ([257]www.sysim.it). The
study was approved by the local institutional review board (IRB) the
Ethikkommission (Ethics Commission) of Hannover Medical School;
approval number #2063-2013 and its amendment #2968-2015. Additional
ethical approval was obtained according to the guideline for
ERACoSysMed- funded translational projects, including Comité de
Protection des Personnes, Est-IV (Ethical Research Committee),
Strasbourg, France and the Comitato Etico Indipendente (Independent
Ethics Committee) of IRCCS, Milan, Italia. The approvals cover (1)
research use of surplus archivial material from nephrectomy and
indicated biopsy samples after completion of the diagnostic workup and
associated anonymized (non-identifiable) clinical information by a
waiver for individual informed consent, and (2) research use of surplus
archivial biopsy material and associated pseudonymised clinical
information of patients who gave their written informed consent when
entering the Hannover Medical School protocol biopsy registry program.
Sections were stored at 4 °C in the dark to minimize antigen aging over
time until further processing.
Immunohistochemistry (IHC)
Standard histology (H&E, PAS, sirius red) and IHC to detect
(fibroblast-activating protein FAP, CD206) as well as control
experiments to calibrate the multiplexing IHC for CD4, CD8, CD20, CD68,
CD206, MS4A4A, and FAP) were performed using an automated staining
instrument (Ventana Benchmark Ultra) following the manufacturer’s
recommendations, and using 3,3 diaminobenzidineor Fast-Red as
chromogens. Primary antibodies are listed below, except polyclonal
anti-MS4A4A (Sigma Life Science, 1:200, #HPA029323), and anti-SV40
(clone MRQ-4; Cell Marque, 1:750,#351M-16). Multiplexed IHC analysis
was optimized following the manufacturer’s instructions (OPAL Multiplex
IHC Assay Development Guide, Akoya Bioscience). 2–4 validation
experiments per IHC assay were performed on subsequent nearly
consecutive sections with identical primary antibodies in automated
chromogenic single or duplex immunohistochemical staining experiments,
confirming identical staining patterns in comparison with
single-channel view after multiplexed immunohistochemistry. Some of
these multiple stainings on sequential tissue sections were used for
robustness testing and development of image analysis algorithms
[ref. [258]27]. Slides were deparaffinized, initial antigen retrieval
was performed by microwave cooking in Tris-buffered saline (TBS) at pH
9, and blocking of unspecific protein binding was performed using
Protein Block Serum-free solution (Agilent/Dako). Subsequent antigen
retrieval and deactivation of the preceding staining step was performed
either in TBS at pH 9 or citrate buffer at pH 6. Consecutive IHC
staining using the OPAL 7-plex fluorescence system was performed using
the following primary antibodies diluted in REAL Antibody Diluent
Agilent/Dako #S2022: anti-CD4 (clone SP35; Zytomed Systems, 1:50,
#503–3354), anti-CD8 (clone C8/144B; Agilent/Dako, 1:350, #M7103),
anti-CD20 (clone L26; Dako, 1:500, #M0755), anti-CD68 (clone PG-M1;
Dako, 1:1000, #M0876), anti-CD206 (clone 5C11; Bio-Rad, 1:1500,
#MCA5552Z), anti-FAP (clone D8; Vitatex, #MABS1001, 1:1500). Cell
nuclei were stained with DAPI. The following fluorophores were used in
the tyramide signal amplification-based multiplexed system to detect
bound antibodies: Opal 520, Opal 540, Opal 570, Opal 620, Opal 650 or
Opal 690. Fluoromount-G mounting medium (Thermo Fisher Scientific) was
applied to cover slides before imaging.
Multispectral image analysis and quantitative evaluation
Whole slide image scanning was done at 20x magnification using the
Vectra Polaris instrument (Akoya Bioscience). Three-channel fluorescent
whole slide images were used to select regions of interest in
nephrectomies (466 × 349 µm size) for subsequent targeted scanning of
image stacks at 40x magnification across the visible spectrum
(420–720 nm) for multispectral imaging, containing glomeruli and the
surrounding region, or representative areas of the interstitial
compartment. Biopsies were fully scanned and spectral libraries were
generated using single stained scans of tonsil tissue for each reagent.
Deconvolution and training for the machine learning-based segmentation,
tissue classification and cell phenotyping algorithms were performed
using the inForm v2.4.8 software (Akoya Bioscience). Visual quality
control of results was performed by comparing all composite images and
selected single-channel images and phenotyping results with
corresponding chromogenic single and duplex staining of adjacent
consecutive sections.
Cell density and neighborhood analysis
Cell phenotypes and individual coordinates of each cell were exported
in text (.csv) format and further processed in GraphPad and MS Excel
(Source data is provided as Source data file). In the context of
neighborhood analysis the term “glomerulus” refers to the area within
the Bowman membrane. For this analysis, the term “surrounding” refers
to the area within a ROI (see above) of 466 × 349 µm size, where the
glomerulus is in the center (Fig. [259]6a–c, panels labeled with
“Glo”), subtracting the glomerulus area from each ROI. The term
“interstitium” refers to ROIs of the same 466 × 349 µm size that are
taken exclusively in interstitial areas, strictly avoiding areas
meeting the “glomerulus” or “surrounding” criteria.
Laser-captured microdissection (LCM) and RNA sequencing
Formalin-fixed paraffin-embedded 5 μm thick tissue sections adjacent to
the sectioning levels used for multiplexed IHC were mounted on a
poly‐l‐lysin‐coated membrane attached to a metal frame under RNAse-free
conditions. LCM was performed on deparaffinized, hemalaun-stained
sections using the Cell Cut Plus System (MMI Molecular Machines &
Industries). In the context of LCM analysis, the term “glomerulus”
refers to the area within the Bowman capsule; the term “surrounding”
refers to a circle with a diameter of 550 µm with the glomerulus in the
center, obtained after the glomerulus had been removed, and considering
that in paraffin sections glomeruli show a diameter of ∼200 µm (range
150–250 µm) and some tissue loss caused by the laser beam, in practical
terms this area corresponds to a ring area of 200–250 µm around each
glomerulus adopted for the cell density and neighborhood analysis; the
term “interstitial” refers to areas of variable shape selected by
strictly avoiding the other two compartments (Fig. [260]S7). We chose
this pragmatic terminology to make the LMD and multispectral image
analysis results comparable using similar distances/areas. RNA
isolation was performed using the RNeasy Mini Kit (Qiagen) in
combination with Qiashredder columns (Qiagen) according to the
manufacturer’s instructions. Obtained RNA quality was checked using the
RNA Nano Kit (Agilent Technologies) on an Agilent BioAnalyzer 2100
(Agilent Technologies). Prior to library generation, nucleic acid
samples were pretreated with DNase (Qiagen). 5 ng equivalents from
DNase pretreated samples were used for library preparation with the
SMARTer Stranded Total RNA-Seq Kit – Pico Input Mammalian (#635006;
Takara/Clontech) according to conditions recommended in user manual
#101215. Generated libraries were barcoded by dual indexing approach
and were finally amplified with 15 cycles of PCR. Fragment length
distribution of generated libraries was monitored using a BioAnalyzer
High Sensitivity DNA assay (5067–4626; Agilent Technologies).
Quantification of libraries was performed by use of the Qubit® dsDNA HS
Assay Kit ([261]Q32854; Thermo Fisher Scientific). Pooled libraries
were denatured with NaOH and finally diluted to 1.8pM according to the
Denature and Dilute Libraries Guide (document # 15048776 v02;
Illumina). 1.3 ml of denatured pool was loaded on an Illumina NextSeq
550 sequencer using a High Output Flowcell for 75 bp single reads
(#FC-404-2005; Illumina). BCL files were converted to FASTQ files using
bcl2fastq Conversion Software version v2.20.0.422 (Illumina). Raw data
processing and quality control were conducted by use of nfcore/rnaseq
(version 1.5dev). The genome reference and annotation data were taken
from GENCODE.org (Homo sapiens; GRCh38.p12; release 28). Normalization
and differential expression analyses were performed with DESeq2 (Galaxy
Tool Version 2.11.40.2) with default settings except for “Output
normalized counts table” which was set to “Yes”. FACS sorted samples
were collected and cells were lysed by TRIzol reagent. Total RNA was
isolated using DirectZOL RNA miniprep kit (ZymoResearch) according to
the manufacturer’s instructions. Quantification and quality check (RNA
integrity number RIN > 7) were assessed by using Qubit4 (Invitrogen)
instrument. Libraries preparation and processing were performed with
Lexogen protocol, using the QuantSeq 3’ mRNA-Seq Library Prep kit to
generate Illumina-compatible libraries of sequences close to the 3’-end
of poly(A) RNA. Sequencing was performed using a NextSeq 500
(Illumina), producing an average of 5 × 10^6 reads/sample (single-end,
75 bp).
Bioinformatics analysis of transcriptomic data generated on the in vitro
model
Reads from RNA-sequencing were subjected to quality check and trimming
using the FastqQC and BBduk tools and to alignment using the STAR
method. The Phread quality score was greater than 20, and the
percentage of alignment along the reference genome was higher than 80%
along all the samples. Reads were aligned along genes using the HTseq
count tool. Counts matrix was normalized with the TMM method (EdgeR
3.24.3). Gene expression TMM matrix counted a total of 17,650 genes for
each of the 132 samples. Principal Component Analysis (PCA) was
performed using Variance Stabilizing Transformation (VST, DESeq2
1.22.2) on all genes under investigation. The single sample GSEA was
performed starting from the normalized counts matrix with Hallmark
collection gene set. For this analysis, a new gene expression matrix
was defined, considering the mean expression of replicates (matrix
17.650 × 44). In the ssGSEA algorithm, averaged replicates were
considered one by one. A score for each signature for each sample was
calculated determining an absolute values score matrix (signature x
samples). The score matrix was normalized considering the absolute
value of the maximum value and the absolute value of the minimum one.
The normalized matrix was plotted in a heatmap. The relative
contribution of each variable and multiple combined effects was
evaluated by supervised analysis at three levels of increasing
complexity, as reported in the Supplementary methods section.
Bioinformatics analysis of transcriptomic data generated on ex vivo samples
Ex vivo data were analyzed with the same protocol by comparing
pathological and control samples. Signatures generated by in vitro and
ex vivo datasets were compared by using a multi-step method that we
have developed. Pathway enrichment in the pathological samples was
analyzed by Metascape online tool^[262]45, considering DEGs in
nephrectomy ex vivo data (logFC ≥ 1 and p ≤ 0.01). Only pathways with a
–log[10] (p-value) that exceeds the 1.3 threshold value were selected
and represented by bar plots. Pre-ranked GSEA were analyzed by ranking
ex vivo data on logFC, using in vitro data as input gene set signature.
For each differential analysis a signature was obtained and the overlap
between ex vivo and in vitro data was tested. Only in vitro signatures
significantly enriched and overlapped with ex vivo signature were
selected. See the Supplementary methods section for a detailed
description of pre-ranked GSEA and gene overlap analysis.
Mathematical modeling
An ordinary differential equation model was constructed to describe the
dynamics of activated Mφ and fibroblasts and analyze the corresponding
qualitative behavior in potential in vivo scenarios. The equations
consist of terms related to the in vitro experimental observations and
biological assumptions:
A1. Proinflammatory Mφ are recruited proportionally to the inflammation
level.
A2. Profibrotic Mφ proliferate in the presence of Fb, in hypoxic and
proinflammatory environments, while in normoxic environments, they
become senescent (A, B).
A3. Fb become activated and promote cell growth in the presence of Mφ
in hypoxic conditions.
A4. All cell types become deactivated at a constant rate.
A5. Due to the short phenotypic switch rates compared to the system’s
characteristic time scale (inverse Mφ deactivation rate), Mφ were
assumed to be in equilibrium with respect to their phenotypic switch.
Under these assumptions, we define the following system of ordinary
differential equations:
[MATH: M1°=lY
+E1,<
mn>2−d1M1, :MATH]
1
[MATH: M2°=rH
M2<
/msub>FY+Y<
mn>0M<
/mi>2+K−E1,<
mn>2−d2M2, :MATH]
2
[MATH: F°=gHM2FF0−F−d3F, :MATH]
3
where
[MATH: M1
:MATH]
represents the number of proinflammatory Mφ,
[MATH: M2
:MATH]
the number of profibrotic Mφ,
[MATH: F :MATH]
the number of activated Fb,
[MATH: Y :MATH]
the inflammation level,
[MATH: H :MATH]
the hypoxia level,
[MATH: l :MATH]
the level of macrophage recruitment due to local inflammation,
[MATH: d1
:MATH]
the death rate of proinflammatory Mφ,
[MATH: r :MATH]
the profibrotic macrophage proliferation rate,
[MATH: Y0
:MATH]
a reference level of inflammation,
[MATH: K :MATH]
the base inhibition of profibrotic macrophage proliferation,
[MATH: d2
:MATH]
the death rate of profibrotic Mφ,
[MATH: g :MATH]
the activation rate of Fb,
[MATH: F0
:MATH]
the mean maximum number of Fb in fibrosis, and
[MATH: d3
:MATH]
the death rate of Fb. Phenotypic switch between profibrotic and
proinflammatory Mφ was considered via the switching term:
[MATH:
E1,2=−k1→2M
1+k2→1M
2Y, :MATH]
4
where
[MATH:
k1→2 :MATH]
and
[MATH:
k2→1 :MATH]
are the switching rates from proinflammatory to profibrotic phenotypes
and vice versa.
Due to the short phenotypic switch rates compared to the rates of cell
proliferation and deactivation, Mφ were assumed to be in equilibrium
with respect to phenotypic switch. Assumption A5 allow us to link the
number of profibrotic and proinflammatory Mφ as:
[MATH:
Y=λ−1<
/mn>M1
M
2, :MATH]
5
where
[MATH:
λ=k2→1k
1→2 :MATH]
is the ratio of the switching rates. Using this expression, we can
combine the equations for both macrophage phenotypes into a single
equation for the total macrophage number,
[MATH:
M=M1+M2<
/msub>. :MATH]
The model is simplified further by non-dimensionalizing all quantities
(Box [263]1). The fully parameterized model has been deposited in the
Biomodels database ([264]MODEL2209160001).
When the cytokines PDGF and CSF1 are assumed to be in a quasi-steady
state, the reduced model (Box [265]1) has a similar structure to the
one developed by ref. [266]25. For a comparison of the two models refer
to the mathematical modeling appendix in the Supplementary information
section. In contrast to ref. [267]25, the activation rates of the two
cell populations are modified by the impact of inflammation and
hypoxia. Under normoxic conditions (
[MATH: H=0 :MATH]
), there is a single stable steady state:
[MATH: m=y1+λyλy+δ1, :MATH]
6
[MATH: f=0. :MATH]
7
In hypoxia, this steady state exists along with two additional steady
states. Linear stability analysis reveals that the steady state common
to hypoxic and normoxic conditions (the “healthy” steady state) is a
stable node as long as:
[MATH: H<δ2βλ+δ<
/mrow>1y, :MATH]
8
or written in terms of the inflammation level, which defines a critical
inflammation, above which the system cannot return to the healthy
state:
[MATH: y<ycr=
δ1βδ2<
/mrow>H−λ. :MATH]
9
It should be noted that, in the limit of very high inflammation (
[MATH:
δ1≪λ :MATH]
), stability of the healthy state is still possible, as the stability
condition reduces to:
[MATH: H<λδ2β. :MATH]
10
One of the two steady states found only in hypoxic conditions is
numerically found to correspond to a saddle point, while the other can
be a stable or unstable node (“fibrotic” steady state) with
[MATH: f≈1 :MATH]
and
[MATH: m≫1 :MATH]
. The inflammation-hypoxia space can be partitioned into three regions
depending on the stability of the two nodes: (I) the healthy state is
stable, the fibrotic state is unstable; (II) both states are stable;
(III) the healthy steady state is unstable, the fibrotic state is
stable (Fig. [268]10b).
Box 1 Mathematical model.
The mathematical model describes the change in total number of
macrophages and fibroblasts through the following system of coupled
ordinary differential equations:
[MATH: m°=y−mλy<
mo>+δ11+λy+<
/mo>αHy+1mm+1+λyf, :MATH]
[MATH: f°=βHm1+λyf1−f−δ2f<
mo>, :MATH]
where
[MATH: m :MATH]
and
[MATH: f :MATH]
are the non-dimensional macrophage and fibroblast populations,
respectively,
[MATH: y :MATH]
represents the non-dimensional inflammation,
[MATH: δ1
:MATH]
is related to the ratio of profibrotic to proinflammatory macrophage
deactivation rates,
[MATH: δ2
:MATH]
is the ratio of fibroblast to proinflammatory deactivation rates,
[MATH: α :MATH]
is proportional to the profibrotic macrophage proliferation rate,
[MATH: β :MATH]
is proportional to the fibroblast proliferation rate,
[MATH: H :MATH]
is the hypoxia level, and
[MATH: λ :MATH]
is the ratio of macrophage switching rates from proinflammatory to
profibrotic phenotypes and vice versa. All quantities are
non-dimensional.
Statistical analysis
Statistical analysis was performed using Prism version 7.0 (GraphPad
software). Comparisons were calculated by two-way ANOVA test applying
Tukey’s multiple comparisons correction. The level of statistically
significant difference was defined as p < 0.05.
Reporting summary
Further information on research design is available in the [269]Nature
Research Reporting Summary linked to this article.
Supplementary information
[270]Supplementary Information^ (37.3MB, pdf)
[271]41467_2022_34241_MOESM2_ESM.pdf^ (101.1KB, pdf)
Description of Additional Supplementary Files
[272]Supplementary Data 1^ (39.5KB, xlsx)
[273]Supplementary Data 2^ (25.6KB, xlsx)
[274]Supplementary Data 3^ (24.3KB, xlsx)
[275]Supplementary Data 4^ (40KB, xlsx)
[276]Reporting Summary^ (97.7KB, pdf)
Acknowledgements