Abstract
Oxygen deprivation and excess are both toxic. Thus, the body’s ability
to adapt to varying oxygen tensions is critical for survival. While the
hypoxia transcriptional response has been well studied, the
post-translational effects of oxygen have been underexplored. In this
study, we systematically investigate protein turnover rates in mouse
heart, lung, and brain under different inhaled oxygen tensions. We find
that the lung proteome is the most responsive to varying oxygen
tensions. In particular, several extracellular matrix (ECM) proteins
are stabilized in the lung under both hypoxia and hyperoxia.
Furthermore, we show that complex 1 of the electron transport chain is
destabilized in hyperoxia, in accordance with the exacerbation of
associated disease models by hyperoxia and rescue by hypoxia. Moreover,
we nominate MYBBP1A as a hyperoxia transcriptional regulator,
particularly in the context of rRNA homeostasis. Overall, our study
highlights the importance of varying oxygen tensions on protein
turnover rates and identifies tissue-specific mediators of
oxygen-dependent responses.
__________________________________________________________________
Ambient oxygen levels alter protein turnover rates in tissues,
revealing mediators of the hyperoxia response.
INTRODUCTION
Oxygen deprivation underlies countless pathological conditions
including ischemic heart disease, hypoxic lung disease (e.g.,
emphysema), and stroke ([54]1–[55]3). Oxygen excess is also
pathological in the context of hyperoxic lung injury, retinopathy of
prematurity, and mitochondrial disorders ([56]4–[57]7). Variations in
tissue oxygen levels trigger a multitude of adaptive and maladaptive
signaling cascades. Most prior research in this space has focused on
transcriptional responses mediated by the hypoxia-inducible factors
(HIFs) ([58]3, [59]8, [60]9). However, it is becoming increasingly
clear that post-translational effects are also key regulators of the
hypoxia response ([61]10–[62]12).
There are several known hypoxia adaptations that involve modulating
protein turnover rates. Recently, researchers discovered an
oxygen-sensing mechanism regulated by cysteamine dioxygenase (ADO) that
is conserved from plants to mammals ([63]11, [64]13, [65]14). In
normoxic conditions, oxygen-dependent ADO modifies the N-terminal
cysteines of target proteins, resulting in degradation via the N-end
rule. In hypoxia, this reaction is inhibited, thereby stabilizing
targets ([66]11). Furthermore, hypoxia inhibits mammalian target of
rapamycin (mTOR) signaling and activates the unfolded protein response
(UPR) ([67]15, [68]16). These pathways decrease overall protein
turnover rates in extreme hypoxia to conserve adenosine 5′-triphosphate
(ATP) and alleviate endoplasmic reticulum stress ([69]16).
On the other hand, oxygen-dependent changes in protein turnover can
also be maladaptive. For example, we recently found that specific
iron-sulfur (Fe-S)–containing protein complexes are destabilized and
degraded in hyperoxia, leading to downstream biochemical defects
([70]17). In particular, a substructure of electron transport chain
(ETC) complex 1 (the matrix-facing arm that includes the N and Q
modules) contains eight Fe-S clusters and is destabilized in hyperoxia.
The resulting ETC dysfunction causes progressive hyperoxia, ultimately
damaging additional Fe-S–containing proteins. Thus, the ETC is the
“weakest link” in hyperoxia and is a primary cause of pathophysiology
([71]17). In addition, post-translational disulfide bond formation is
oxygen-dependent. Extreme hypoxia inhibits this reaction, causing
protein misfolding and degradation ([72]18).
Thus, there are clues in the literature that variations in oxygen
levels can affect protein turnover rates in a manner that is adaptive
or maladaptive. However, we lack a comprehensive investigation of
protein turnover rates as a function of oxygen in vivo. More generally,
most in vivo protein turnover studies have focused on baseline
conditions rather than physiological responses to stress. It is also
unknown how different organs cope with proteotoxic stress in hypoxia or
hyperoxia. Lastly, while HIF is a well-studied transcriptional
regulator of the hypoxia response, it is unclear whether there are
analogous mediators of the hyperoxia response.
We set out to answer these questions using organism-level stable
isotope-labeling proteomics ([73]19). We focus on the lung, heart, and
brain because these organs are particularly sensitive to pathologic
conditions related to tissue hypoxia or hyperoxia. Our study highlights
the tissue-specific responses to varying oxygen tensions and nominates
MYB-binding protein 1a (MYBBP1A) as a transcriptional regulator of
hyperoxia response. These findings are of particular relevance to
states of pulmonary oxygen toxicity, including bronchopulmonary
dysplasia in neonates and hyperoxic lung injury in adults.
RESULTS
Protein half-lives are determined by tissue-specific and protein-intrinsic
features in normoxia
To systematically study protein turnover rates in different oxygen
tensions, we conducted stable isotope labeling of amino acids in mice
(SILAM) ([74]Fig. 1A). This approach relies on the assumption that the
proteome is at steady state throughout the experiment. Thus, we first
exposed wild-type mice to three different fractions of inspired oxygen
(F[i]O[2]) for 1 week: chronic hypoxia (8% F[i]O[2]), normoxia (21%
F[i]O[2]), or hyperoxia (60% F[i]O[2]). Throughout this pretreatment
period, we placed mice on a control algae diet containing the standard
^14N isotope. After 1 week, we switched the mice in different oxygen
tensions to a ^15N-containing algae diet and collected all major organs
after 1, 2, 4, 8, 16, and 32 days. Tissue lysates were analyzed by
data-independent acquisition (DIA) proteomics ([75]20) to quantify the
incorporation of labeled amino acids into the proteome of the lung,
heart, and brain.
Fig. 1. Pulsed SILAM reveals variations of protein half-lives across organs.
[76]Fig. 1.
[77]Open in a new tab
(A) Schematic of the pulsed SILAM study design. Mice were acclimatized
with the ^14N algae diet in respective oxygen tensions for 7 days
before the start of the experiment. At day 0, all mice were switched to
the ^15N-labeled algae diet to label all the newly synthesized
proteins. Tissues were harvested at different time points following the
start of the labeling. Lung, heart, and brain samples were analyzed
using LC/MS-MS. (B) Data analysis schematics. The fraction of peptides
made of ^14N amino acids only was quantified and normalized at each
time point to the values of 10 long-lived proteins ([78]22). Following
normalization, ^14N fractions were fitted by linear mixed-effects
models to estimate protein degradation rates for each protein in a
given condition. To compare degradation rates between oxygen tensions,
a linear mixed-effects model was applied where oxygen and time were set
as fixed effects and peptides were random variables. (C) Study
coverage, inter-organ overlap, and correlations. The circles represent
the number of proteins in normoxia (21% O[2]) whose degradation rates
were estimated in the study. The Venn diagrams represent the overlap of
proteins measured in the study across organs. The scatterplots show the
correlations of protein half-lives across organs, and the coefficients
of determination (r^2) are shown in the figures. (D) Density plots of
protein half-lives and median half-lives in lung (3.9 days), heart (7.3
days), and brain (8.0 days) in normoxia across the three organs. (E)
Protein half-lives across oxygen tensions. Statistical analysis was
performed using the Kolmogorov-Smirnov test. *P < 0.05, **P < 0.01, and
****P < 0.0001.
As previously shown, the incorporation of ^15N into the proteome over
time leads to broad MS1 spectra ([79]21), which poses a challenge for
accurate modeling and quantification of peptide turnover rates.
Instead, it has been shown that protein turnover rates can be
calculated by quantifying the decay of ^14N peptides (due to ^15N
incorporation) ([80]19). Therefore, to estimate protein degradation
rates, we first quantified the decay of the ^14N fraction over time. We
then normalized the peptide abundance to that of 10 long-lived
housekeeping proteins (table S1) that have previously been reported to
have half-lives of greater than 2 months in mouse heart ([81]Fig. 1B
and fig. S1A) ([82]22). We fitted these normalized data to a
first-order kinetic model to estimate the protein degradation rate
(K[d]) and applied a linear mixed-effects model to compare turnover
rates across oxygen tensions ([83]Fig. 1B and fig. S1A).
This allowed for the estimation of protein degradation rates and
half-lives for 4152, 2536, and 3877 proteins in lung, heart, and brain,
respectively ([84]Fig. 1C, fig. S1B, and table S2). For greater than
50% of these proteins, the half-life was calculated using at least
three detected peptides (fig. S1C). We observed a strong correlation of
half-lives across biological replicates in each organ (Pearson
correlation coefficient r: median = 0.95; range = 0.73 to 0.99),
demonstrating the reproducibility of the dataset (fig. S1D).
We first examined normoxic tissues and observed marked differences in
the distribution of protein half-lives across organs. The median
half-life was 7.2 days in the heart, 8.0 days in the brain, and 3.9
days in the lung ([85]Fig. 1D and fig. S1B). This may reflect more
direct exposure of lung to the atmosphere. In addition, the lung has
several cell types (e.g., epithelial cells, endothelial cells, etc.)
that show greater overall turnover rates compared to terminally
differentiated cell types in other organs (e.g., neurons and
cardiomyocytes) ([86]23–[87]25).
Regardless of the overall distribution of protein half-lives, it is
possible that the relative half-lives within the proteome are dictated
at least partially by protein-intrinsic features. In line with this, we
observed positive correlations of protein half-lives between normoxic
brain, heart, and lung (r^2 range: 0.35 to 0.59) ([88]Fig. 1C),
consistent with previous findings in primary cells and tissues ([89]21,
[90]26). We further examined the intrinsic biophysical features of
proteins, including peptide length, hydrophobicity, intrinsic
disordered fraction, and charge patterns. We found that a larger
intrinsic disordered fraction, less hydrophobicity, and larger
fractions of negatively charged amino acids are associated with faster
turnover rates in all three tissues (fig. S2A). Together, our data
indicate that protein half-lives are dependent on intrinsic biophysical
features of proteins but vary across organs.
Tissue-specific features are the main determinant of protein half-lives in
hypoxia and hyperoxia
We next determined how variations in oxygen tension affect protein
turnover rates across the three organs. We observed only modest changes
in protein half-lives at the global level ([91]Fig. 1E). Instead, most
changes occurred in specific proteins. The lung had the greatest number
of significantly affected proteins [fold changes > 1.3, false discovery
rate (FDR) < 0.05]: 12.7% for hypoxia and 9.7% for hyperoxia
([92]Fig. 2A). On the other hand, in the heart and brain, less than 4%
of the proteins we monitored had significantly different turnover rates
in hyperoxia or hypoxia ([93]Fig. 2A).
Fig. 2. Protein turnover rates have different sensitivities to varying oxygen
tensions in different organs.
[94]Fig. 2.
[95]Open in a new tab
(A) Volcano plots of log[2](fold changes) and −log[10](FDR) of turnover
rates (K[d]) for each protein calculated using a linear mixed-effects
model. Oxygen and time were set as the fixed variables, and peptides
were set as the random variables. Proteins with fold changes >1.3 and
FDR < 0.05 are highlighted in blue (for proteins with slower turnover)
or red (for faster turnover). Statistical analysis was performed using
the linear mixed-effects model with Benjamini-Hochberg correction. (B)
Venn diagrams of significantly changed proteins (fold change >1.3, FDR
< 0.05). Proteins detected in all three organs were compared. The color
gradients indicate the number of proteins that were significantly
changed in lung (L), heart (H), or/and brain (B).
There are two possible explanations for the marked effects on lung
protein turnover rates: (i) The lung is directly exposed to
environmental oxygen, so the local tissue partial pressure of oxygen
(PO[2]) is most sensitive to changes in inhaled oxygen level, while for
other tissues, oxygen levels are buffered by hemoglobin and partially
normalized by physiological adaptations such as changes in hematocrit
and vascular density; (ii) the overall cellular turnover rates in the
lung may intrinsically be greater, requiring increased protein turnover
([96]27). We observed minimal overlap between the significantly changed
proteins across the three organs ([97]Fig. 2B). In addition, we
assessed the effects of protein-intrinsic properties. We found that the
biophysical properties were not strong predictors of oxygen-dependent
changes in protein turnover (fig. S2B). These results indicate that
oxygen-dependent changes in protein turnover are highly variable across
organs and are likely linked to tissue-specific physiological responses
to hypoxia or hyperoxia.
ECM proteins exhibit oxygen-dependent turnover rates in lung
We performed pathway enrichment analysis for proteins showing
significant changes in turnover rates. In the lung, we found that
proteins with slower turnover rates were enriched for extracellular
matrix (ECM) remodeling pathways in both hypoxia and hyperoxia,
including ECM-receptor interaction and collagen formation ([98]Fig. 3A
and table S3). In particular, collagen proteins (COL1A1, COL1A2, and
COL6A1) and laminin proteins (LAMC2 and LAMB2) exhibit slower turnover
rates in both hypoxia and hyperoxia than in normoxia ([99]Fig. 3, B, C,
and D, and table S3). Collagen and laminin are both critical components
of the ECM, and have important roles in maintaining tissue integrity,
providing structural support, and regulating cell behavior ([100]28).
There are different types of collagen and laminin proteins, with unique
structural properties and functions. Of note, some laminin proteins,
such as laminin subunit alpha 5 (LAMA5), showed higher turnover rates
in hypoxia and hyperoxia than in normoxia ([101]Fig. 3, B and C, and
table S2), suggesting that different laminin subunits are regulated
differently in response to oxygen changes. Our data reveal that these
proteins are likely to be regulated post-translationally in varying
oxygen tensions, potentially contributing to lung ECM remodeling.
Notably, collagen is known to undergo oxygen-dependent prolyl
hydroxylation, which affects protein stability ([102]29).
Fig. 3. Extracellular matrix proteins exhibit oxygen-dependent turnover rates
in lung tissue.
[103]Fig. 3.
[104]Open in a new tab
(A) Enrichment analysis for proteins with significant changes in
protein degradation rates (FDR < 0.05, fold change >1.3) in lung
tissues using the STRING functional analysis method. The size of the
dots represents −log[10](FDR) for each term. (B and C) Protein-protein
physical interaction network of extracellular matrix proteins in lung
constructed using the STRING network function in Cytoscape (confidence
score cutoff at 0.70). The plots show the laminin and collagen proteins
and their first-degree neighbors in the network. Each node represents a
protein that is measured in the lung, and edges represent the physical
interactions. The color represents the fold change in protein
degradation rates, and the size represents the significance level
determined using the linear mixed-effects model. The proteins with FDR
< 0.05 are bolded. (D) Representative scatterplots of ^14N protein
decay over time in the three oxygen tensions. Each dot represents the
average abundance of all peptides in each subject. The fitted lines
were calculated using the linear mixed-effects model, where the
absolute numbers of the slopes correspond to the degradation rates.
Our data revealed distinct changes in protein dynamics in the heart and
brain (fig. S3). In the heart, we observed increased degradation rates
of proteins involved in ubiquinone (coenzyme Q) synthesis in hypoxia
(fig. S3A, including COQ3 and COQ9; and table S3). In the brain, we
found that proteins involved in receptor-mediated endocytosis, such as
EFNB1 and EFNB2, a cell surface transmembrane ligand for Eph receptors,
were stabilized under hyperoxic conditions (fig. S3D and table S3).
Overall, our study provides insights into the tissue-specific responses
to changes in oxygen that will serve as a broadly useful resource for
the field.
Oxygen-dependent effects on protein turnover rates vary across protein
complex subunits
In our previous work, we found that hyperoxia destabilizes specific
protein complexes containing Fe-S clusters ([105]17). For example, the
entire matrix-facing arm of mitochondrial ETC complex 1 was depleted in
hyperoxia, suggesting complex-level effects, not just subunit-level
changes ([106]17). To determine whether this occurs more broadly, we
searched all mammalian protein complexes using the CORUM database
([107]30). This analysis confirmed that most ETC complex 1 subunits had
shorter half-lives in hyperoxia in the lung ([108]Figs. 3D and [109]4,
B and C, and table S4). Among the 17 complex 1 subunits measured in
this study, 11 subunits exhibited faster turnover rates in hyperoxia
(2.1- to 3.7-fold) ([110]Fig. 4, B and C). This was less evident in the
brain and heart (table S2), likely because the inhaled hyperoxia is
buffered by hemoglobin. Of note, tissue hyperoxia can also result from
biochemical ETC deficiency [e.g., genetic mitochondrial disease
([111]4)]. In this case, internal organs such as brain and heart will
likely show more marked hyperoxia-dependent effects. A previous study
demonstrated that subunits in the matrix-facing N-module of complex 1
are degraded at faster rates because of oxidative damage ([112]31,
[113]32). In line with this, we found that the matrix-facing modules
had significantly faster turnover rates than the membrane-bound module
(fig. S4).
Fig. 4. Effects of oxygen on protein complex turnover rates in lung tissue.
[114]Fig. 4.
[115]Open in a new tab
(A and B) Network representation of protein complexes generated using
Cytoscape. The protein complexes were retrieved from the CORUM
database. Protein complexes with four or more subunits measured in the
study are shown. The color gradient represents the differential protein
degradation rates between oxygen tensions. Statistical analysis was
performed using the linear mixed-effects model with Benjamini-Hochberg
correction. Proteins with FDR < 0.05 are highlighted in blue (for
proteins with slower turnover) or red (for faster turnover). The sizes
of the nodes indicate the significance level. The edges represent the
physical interactions among subunits in a protein complex. (C) Crystal
structure and schematic of ETC complex 1 ([116]79, [117]80). The color
gradient represents the fold changes of protein turnover rates in
hyperoxia versus normoxia for the proteins that were significantly
changed (FDR < 0.05) in their turnover rates. Oxygen labile Fe-S
clusters are shown. (D) Crystal structure and schematic of chaperonin
([118]79, [119]81). TCP1, which had faster turnover rates in hypoxia
and hyperoxia than in normoxia, is highlighted in red. Statistical
analysis for all panels was performed using the linear mixed-effects
model with Benjamini-Hochberg correction.
In contrast, in many other protein complexes, only one or two subunits
are sensitive to varying oxygen levels ([120]Fig. 4, A and B, and table
S4). For example, the chaperonin complex contains eight subunits, but
only t-complex 1 (TCP1) has faster turnover rates in both hypoxia and
hyperoxia ([121]Fig. 4D). This may be due to subunit-specific
susceptibility to oxidation or protein-protein interactions. In the
case of chaperonin, TCP1 is known to interact with von Hippel-Lindau
tumor suppressor (VHL), an E3 ligase, to facilitate its folding
([122]33). It is possible that there is increased oxygen-dependent
interaction between TCP1 and VHL, resulting in increased ubiquitination
and proteasomal degradation of TCP1. Alternatively, it is possible that
the protein stability of specific proteins is affected before
incorporation into the larger protein complex. In summary,
oxygen-dependent changes in turnover rate do not always correlate
across all proteins of a given complex.
Hyperoxia-induced protein destabilization may exacerbate monogenic disorders
We previously reported that hyperoxia can worsen the disease
progression of a mouse model of mitochondrial disease caused by genetic
ETC defects, leading to respiratory failure and premature death
([123]4, [124]34). Moreover, hypoxia rescues disease and significantly
improves survival, likely by acting on a fragile complex 1. In this
study, we identified 434 proteins that were destabilized in hyperoxic
lung. To investigate which other monogenic disorders may be exacerbated
by hyperoxia, we crossed this list with the Online Mendelian
Inheritance in Man (OMIM) database, a compendium of human monogenic
disorders. Consistent with our previous findings, we identified many
mitochondrial disease genes as hits ([125]Table 1). In addition, we
identified multiple disease genes that cause glycogen storage disease,
spastic paraplegia, and Charcot-Marie-Tooth disease ([126]Table 1).
Patients with these conditions may be more sensitive to the use of
supplemental oxygen and amenable to rescue by hypoxia exposure. Future
work is needed to investigate the clinical and preclinical significance
of these findings.
Table 1. Monogenic diseases that may be exacerbated by hyperoxia.
Significantly destabilized proteins in hyperoxic lung with known
monogenic disorders and phenotype MIM numbers.
Gene symbol Gene name Phenotypes and MIM numbers
GYG1 Glycogenin 1 Glycogen storage disease XV, 613507 (3), autosomal
recessive; polyglucosan body myopathy 2, 616199 (3), autosomal
recessive
GYS1 Glycogen synthase Glycogen storage disease 0, muscle, 611556 (3),
autosomal recessive
NDUFB10 NADH-ubiquinone oxidoreductase subunit B10 Mitochondrial
complex I deficiency, nuclear type 35, 619003 (3), autosomal recessive
NDUFA13 NADH-ubiquinone oxidoreductase subunit A13 Mitochondrial
complex I deficiency, nuclear type 28, 618249 (3), autosomal recessive
NDUFA2 NADH-ubiquinone oxidoreductase subunit A2 Mitochondrial complex
I deficiency, nuclear type 13, 618235 (3), autosomal recessive
NDUFS8 NADH-ubiquinone oxidoreductase core subunit S8 Mitochondrial
complex I deficiency, nuclear type 2, 618222 (3), autosomal recessive
NDUFA10 NADH-ubiquinone oxidoreductase subunit A10 Mitochondrial
complex I deficiency, nuclear type 22, 618243 (3), autosomal recessive
NDUFA9 NADH-ubiquinone oxidoreductase subunit A9 Mitochondrial complex
I deficiency, nuclear type 26, 618247 (3), autosomal recessive
NDUFA8 NADH-ubiquinone oxidoreductase subunit A8 Mitochondrial complex
I deficiency, nuclear type 37, 619272 (3), autosomal recessive
NDUFV2 NADH-ubiquinone oxidoreductase core subunit V2 Mitochondrial
complex I deficiency, nuclear type 7, 618229 (3), autosomal recessive
NDUFS3 NADH-ubiquinone oxidoreductase core subunit S3 Mitochondrial
complex I deficiency, nuclear type 8, 618230 (3), autosomal recessive
B4GALNT1 β-1,4-N-acetylgalactosaminyltransferase 1 Spastic paraplegia
26, autosomal recessive, 609195 (3), autosomal recessive
CAPN1 Calpain, large polypeptide L1 Spastic paraplegia 76, autosomal
recessive, 616907 (3), autosomal recessive
MTMR2 Myotubularin-related protein 2 Charcot-Marie-Tooth disease, type
4B1, 601382 (3), autosomal recessive
PRX Periaxin Charcot-Marie-Tooth disease, type 4F, 614895 (3),
autosomal recessive; Dejerine-Sottas disease, 145900 (3), autosomal
recessive, autosomal dominant
[127]Open in a new tab
Transcription cofactor MYBBP1A is stabilized in hyperoxic lung tissue
Hypoxia is known to stabilize HIF1α and HIF2α, which are critical
transcription factors that drive hypoxia adaptations. We wondered
whether there are specific transcriptional regulators that are
similarly stabilized in hyperoxia, and may serve as mediators of a
hyperoxia response. We searched for all transcription factors and
cofactors in our dataset using the Animal TFDB 3.0 database ([128]35).
We found three such proteins with decreased turnover, indicating
increased stability in lung tissue from hyperoxic mice: NOTCH2,
MYBBP1A, and SND1 ([129]Fig. 5, A and B, and fig. S5A). To investigate
whether these transcription factors accumulate in hyperoxia, we
reanalyzed an abundance proteomics dataset in lung homogenate from
wild-type mice that were exposed to 21% O[2] or 80% O[2] for 5 days
from our previous study ([130]17). We also included samples from mice
that were exposed to hyperoxia for 5 days and returned to normoxia for
1 or 5 days. We found that MYBBP1A protein levels in the lung
progressively increased over the 5 days of exposure to hyperoxia
([131]Fig. 5C). We validated this finding by Western blot of whole-lung
homogenate from these same samples ([132]Fig. 5D). The protein level
remained elevated 1 day after the mice were transferred from 80 to 21%
O[2], as expected, because the protein’s half-life is ~2 days in
normoxia. However, it normalized after 5 days of recovery in normoxia
([133]Fig. 5C). These results demonstrate that MYBBP1A is stabilized
under hyperoxia in lung tissue, resulting in its progressive
accumulation. Moreover, this stabilization is reversible when the mice
are returned to normoxic conditions. Next, to interrogate whether
MYBBP1A is up-regulated in different lung cell types ([134]Fig. 5E), we
isolated lung fibroblasts and alveolar type II (ATII) cells from mice
exposed to 21 or 80% O[2]. We demonstrated that MYBBP1A level was
elevated in both cell types ([135]Fig. 5, F and G). The timing of
up-regulation varied: ATII cells responded after 3 days of hyperoxia,
whereas fibroblasts showed an increase after 5 days ([136]Fig. 5, F and
G). This suggests varying sensitivities to hyperoxia among cell types.
Furthermore, primary lung fibroblasts and ATII cells isolated from
human patients also displayed increased MYBBP1A expression after being
cultured in hyperoxic conditions ([137]Fig. 5, H and I), highlighting
the clinical relevance and evolutionary conservation of this response.
These features make MYBBP1A a strong candidate for a regulator of the
hyperoxia response. Therefore, we next focused on MYBBP1A and its
downstream consequences.
Fig. 5. MYBBP1A is stabilized and accumulates in hyperoxic lung tissue.
[138]Fig. 5.
[139]Open in a new tab
(A) Volcano plots of proteins with changes in protein turnover rates in
the lung. Transcription factors and cofactors with significant changes
are highlighted in blue (for proteins with slower turnover) or red (for
faster turnover) in respective oxygen tensions. Statistical analysis
was performed using the linear mixed-effects model with
Benjamini-Hochberg correction. (B) ^14N protein decay over time of
MYBBP1A in lung tissues. Each dot represents the average abundance of
all peptides in each subject. The corresponding fitted lines were
calculated using the linear mixed-effects model, where the absolute
numbers of the slopes correspond to the degradation rates. (C)
Abundance proteomics data for MYBBP1A in mouse lung (n = 6 biological
replicates per group) ([140]17). Mice were exposed to 21 or 80% FiO[2].
For the recovery groups, mice were first exposed to 80% FiO[2] for 5
days and brought back to 21% FiO[2] for 1 or 5 days. Data were analyzed
using unpaired t test following log transformation. *P value (with
Benjamini-Hochberg correction) < 0.05. (E) Main cell types in lung
tissues. Illustration made with Biorender. (D, F, and G) Western blots
of MYBBP1A in mouse lung homogenates (D), isolated mouse lung
fibroblasts (F), or mouse ATII cells (G) (n = 3 biological replicates
per group). Mice were exposed to 21 or 80% FiO[2] for up to 5 days. (H
and I) Western blots of MYBBP1A in primary human ATII cells (H) and
lung fibroblasts (I).
MYBBP1A is associated with rRNA processing in hyperoxic lung tissue
MYBBP1A has been studied in several cancer cell lines and Drosophila,
where it is reported to be localized in the nucleolus
([141]36–[142]40). It has been proposed to have two primary roles: (i)
as a cofactor of multiple transcription factors and (ii) as a regulator
of ribosomal RNA (rRNA) processing/transcription ([143]41). In the
former case, MYBBP1A translocates to the nucleoplasm under stress
conditions, where it represses the activity of certain transcription
factors (e.g., PGC1α and MYB) and activates that of others (e.g., p53)
([144]40–[145]42). In the latter case, it is proposed to affect rRNA
levels in the nucleolus ([146]36, [147]37). To better understand
MYBBP1A’s function in an unbiased manner, we searched for genes that
are coessential with MYBBP1A using the DepMap Genetic Dependency
database. This resource identifies essential genes across hundreds of
cancer cell lines using genome-wide CRISPR knockout screens ([148]43).
Genes that are most coessential across these cell lines are likely to
be functionally related. We found that MYBBP1A is coessential with many
other players in the rRNA biogenesis pathway (including WDR18, PELP1,
POP5, BOP1, and TSR1), supporting the findings from previous literature
([149]Fig. 6B and table S5).
Fig. 6. MYBBP1A up-regulation is associated with increased rRNA biogenesis in
hyperoxia.
[150]Fig. 6.
[151]Open in a new tab
(A) Top enriched pathways for up-regulated genes in hyperoxic lung [FDR
< 0.05, log[2] (fold change) > 1.5] using EnrichR. The color gradient
represents the number of significantly up-regulated genes in each
pathway. The Gene Ontology terms are ranked on the basis of the FDR
values. (B) Coessentiality gene network of MYBBP1A in the DepMap
database. The network was generated with the FIREWORKS web tool
([152]44). Top 30 genes positively correlated with MYBBP1A are shown in
red (primary), and the genes with the secondary interactions are shown
in yellow. The thickness of the edges corresponds to Pearson
correlations between two nodes (range = 0.21 to 0.65). Genes involved
in rRNA biogenesis (GO:0042254) are circled with blue outlines. (C)
Schematic of the rRNA precursor and mature ribosomal rRNA. (D) qRT-PCR
of internal transcribed spacers (ITS1 and ITS2) of rRNA and mature
rRNA, including 18S, 5.8S, and 28S. The RNA expression levels relative
to Hprt1 are shown. Each data point represents the average value of
technical duplicates (n = 4 animals per group). Statistical analysis
was performed using unpaired t test. **P < 0.01. (E) Genes in each cell
type were ranked on the basis of the fold changes in the single-cell
RNAseq data ([153]46) (from the most up-regulated to the most
down-regulated in hyperoxia). The genes involved in ribosome biogenesis
(GO:0042254) were analyzed using the GSEA algorithm in each lung cell
type. The cell types with the most significant enrichment are shown.
NES, normalized enrichment score.
To identify the potential effects of MYBBP1A in hyperoxia, we performed
unbiased transcriptomics (using 3′ untranslated region mRNA sequencing,
Quant-seq) to compare gene expression in normoxic versus hyperoxic
mouse lung, matched to our previous abundance proteomics dataset. We
found that genes that were up-regulated in lung tissue after exposure
to 80% oxygen for 5 days were enriched in ribosome processing and rRNA
biogenesis ([154]Fig. 6A and tables S6 and S7). Moreover, we observed a
moderate up-regulation of p53 targets in hyperoxia, including p21 (fig.
S5B). Thus, it is possible that MYBBP1A up-regulation also promotes the
p53-mediated transcriptional response. On the other hand, targets of
MYB or PGC1⍺ were not significantly changed (fig. S5B). Therefore, we
hypothesized that up-regulation of MYBBP1A in hyperoxia primarily
affects rRNA processing in hyperoxia.
The rRNA precursor is a polycistronic sequence that is transcribed from
rDNA and undergoes complex processing to produce three of the four
mature rRNAs (18S, 5.8S, and 28S) in the nucleolus. The mature rRNAs
are then assembled into the ribosome and are required for its function
in protein translation ([155]45). To investigate the effects of
hyperoxia on rRNA biogenesis, we designed polymerase chain reaction
(PCR) primers targeting the three mature rRNA and internally
transcribed sequences (ITS) of the 47S rRNA precursor ([156]Fig. 6C).
We performed quantitative reverse transcription polymerase chain
reaction (qRT-PCR) for pre-rRNA and mature RNA in mouse lung tissues
exposed to 21% O[2] or 80% O[2] for 5 days. We found that lung tissues
had a nearly 10-fold increase in pre-rRNA expression in hyperoxia
compared to normoxia, whereas there were no significant changes in the
expression of mature rRNAs ([157]Fig. 6D). Thus, stabilization of
MYBBP1A is associated with increased rRNA precursors in the hyperoxic
lung.
Next, we set out to understand which lung cell types show
MYBBP1A-mediated transcriptional changes. We analyzed a previously
published single-cell RNA-sequencing (RNAseq) data for mouse lung
exposed to 21 or 85% oxygen from birth to postnatal day 14 ([158]46).
We found that most endothelial, epithelial, and stromal cell types
showed a significant increase in expression of genes involved in
ribosomal biogenesis (fig. S5C), which is in concordance with our bulk
Quant-seq data in hyperoxic lung tissue. In particular, ATII cells,
pulmonary vein endothelial cells, and pericytes showed the most
substantial fold changes ([159]Fig. 6E). In addition, we crossed the
single-cell RNAseq dataset with genes down-regulated in MYBBP1A
knockout K562 cells ([160]47). These genes are likely regulated by
MYBBP1A-mediated transcription responses. In support of our overarching
hypothesis, these genes are enriched in most of the lung cell types in
hyperoxia, including ATII cells, pulmonary vein endothelial cells, and
fibromyocytes (fig. S5D). Together, these analyses support the idea
that MYBBP1A-mediated changes occur in a range of lung cell types
during hyperoxia.
The role of NAD+/NADH in mediating MYBBP1A stabilization in hyperoxia
Destabilization of ETC complex 1 causes reductive stress resulting from
decreased NADH recycling to NAD+ ([161]48). To interrogate the effects
of complex 1 loss on MYBBP1A stabilization, we expressed yeast NADH
dehydrogenase (NDI1) protein in cells to restore the NAD+/NADH ratio in
hyperoxia. Unlike mammalian complex 1, NDI1 does not contain Fe-S
clusters but instead contains flavin adenine dinucleotide (FAD) as the
electron carrier ([162]49). We showed that expressing NDI1 in multiple
cell lines increased the basal mitochondrial respiration rates and
rotenone-resistant respiration rates in hyperoxia, demonstrating that
NDI1 can at least partially complement the decrease in NADH-linked ETC
respiration in hyperoxia (fig. S5, E and F). To test whether impaired
NADH recycling underlies elevated MYBBP1A levels in hyperoxia, we
transduced primary human lung fibroblasts with NDI1 and exposed them to
different oxygen levels. NDI1 expression failed to reverse the
up-regulation of MYBBP1A in hyperoxia ([163]Fig. 7A). In addition, we
treated the cells with sodium pyruvate to restore the NAD+/NADH ratio
by increasing lactate dehydrogenase activity. The addition of pyruvate
did not reverse the up-regulation of MYBBP1A ([164]Fig. 7B). Therefore,
impaired NADH recycling in hyperoxia is not sufficient to explain the
up-regulation of MYBBP1A. Future work will be needed to decipher the
signaling cascades linking hyperoxia to MYBBP1A accumulation and the
downstream consequences.
Fig. 7. Effects of manipulating NADH recycling on MYBBP1A induction in
hyperoxia.
[165]Fig. 7.
[166]Open in a new tab
(A and B) Western blotting of human lung fibroblasts in 21% versus 80%
oxygen for 6 days with NDI1 expression or pyruvate treatment.
Biological duplicates are shown.
DISCUSSION
Oxygen plays a critical role in protein turnover rates via
post-translational modifications. However, it is unclear how varying
oxygen tensions affect protein turnover in vivo. In this study, we
found that oxygen-dependent changes in protein turnover rates under
hypoxic and hyperoxic conditions were highly tissue-specific. The lung,
in particular, showed the most marked changes. We uncovered that
oxygen-dependent changes of turnover rates in protein complexes were
not always correlated. Furthermore, our results nominate MYBBP1A as a
transcriptional regulator in hyperoxic lung, which may have important
implications for lung pathophysiology.
Changes in protein turnover rates reveal tissue-specific proteome
dynamics in response to hypoxia or hyperoxia. The proteins with
significant changes in turnover rates showed minimal overlap across the
three tissues. The top changing pathways were associated with
tissue-specific responses to varying environmental oxygen levels. Our
results demonstrated that in lung tissue, many ECM proteins showed
changes in both hypoxia and hyperoxia, including collagen proteins and
laminin proteins. ECM determines the architecture of lung via
biochemical and biomechanical signals ([167]50). Changes in the ECM
composition and protein modifications occur in several chronic lung
diseases ([168]50). Previous literature has revealed that hypoxia and
hyperoxia can have substantial effects on the ECM remodeling in the
lung, via post-translational regulations. In hypoxia,
collagen-modifying enzymes are up-regulated in a HIF-mediated manner,
including prolyl 4-hydroxylase α-subunit isoform 1 and 2 (P4HA1/2),
procollagen-lysine 2-oxyglutarate 5-dioxygenase 2 (PLOD2), and lysyl
oxidase (LOX) ([169]51, [170]52). On the other hand, hyperoxia
modulates matrix metalloproteinase (MMP) activities via transforming
growth factor–β (TGFβ) signaling, resulting in increased stiffness of
ECM ([171]53, [172]54). It is possible that these enzymes modify
collagen and laminin proteins post-translationally, resulting in their
stabilization in hypoxia or hyperoxia. This may contribute to increased
collagen deposition and basement membrane disruption in lung ECM in
varying oxygen tensions ([173]53, [174]55). Future studies should
explore which post-translational modifications contribute to ECM
protein stabilization in these contexts.
We found that protein turnover changes in protein complexes are not
always correlated. Our previous study showed that most ETC complex 1
subunits are degraded in hyperoxia in cell lines and in mouse lung
tissue ([175]17). Consistent with this, we found that in the hyperoxic
lung, most detected complex 1 subunits exhibited increased protein
turnover rates. However, to our surprise, in other complexes, such as
the chaperonin and proteasome, only specific proteins showed
significant changes. The labile subunits in these complexes may be
regulated by interactions with other proteins or post-translational
modifications ([176]33, [177]56).
In addition, our findings shed light on potential therapeutic targets
for diseases related to chronic hypoxia or hyperoxia, including chronic
lung disease, hyperoxic lung injury, ischemic heart disease, and
mitochondrial disease. We found that ETC complex 1 subunits were
destabilized in hyperoxic lung, which is in accordance with our
previous studies demonstrating that hyperoxia exacerbates and hypoxia
rescues mitochondrial complex 1 disease. In addition, we nominated
three other monogenic diseases that may be worsened by tissue
hyperoxia, including glycogen storage disease, spastic paraplegia, and
Charcot-Marie-Tooth disease.
Furthermore, we nominated MYBBP1A, a transcriptional regulator, as a
dynamic mediator of the hyperoxia response. We found that MYBBP1A was
stabilized in hyperoxic lung tissue, resulting in its accumulation in
multiple lung cell types. Notably, the protein level returned to
baseline levels after return to normoxia. MYBBP1A is predominantly
localized in the nucleolus, where it regulates rRNA biogenesis and
plays a critical role in the assembly of ribosomes ([178]37). Our data
showed an up-regulation of rRNA biogenesis genes in hyperoxic lung
homogenates, as well as several key lung cell types, suggesting that
MYBBP1A may play a role in rRNA regulation in hyperoxia. It has been
shown that knockdown of Mybbp1a in Drosophila affects pre-rRNA and
mature rRNA levels and causes developmental defects ([179]36). In
contrast, Mybbp1a overexpression suppresses RNA polymerase I activities
([180]37), indicating its complex regulatory role in rRNA biogenesis.
Therefore, we investigated the levels of rRNA in varying oxygen
tensions. We found that in hyperoxic lung tissues, the pre-rRNA level
was increased. This increase may lead to MYBBP1A retention and
stabilization in the nucleolus. Overall, our findings shed light on the
critical role of MYBPP1A in rRNA biogenesis in hyperoxic lung. Future
studies will causally delineate the complex role of MYBBP1A in
hyperoxic transcriptional regulation by knocking out the gene in
different lung cell types.
Future studies will also investigate the mechanisms underlying MYBBP1A
stabilization in hyperoxia and test whether this is adaptive or
maladaptive. We demonstrated that impaired NADH oxidation due to
impaired complex 1 activity is not required for hyperoxic stabilization
of MYBBP1A. Follow-up studies will investigate the role of several
known regulators of MYBBP1A turnover in hyperoxia signaling—for
example, PREP1 and USP29 have been shown to stabilize MYBBP1A, whereas
VHL degrades it ([181]57–[182]59).
Limitations of our study include the fact that it spans from 1 day to
32 days, which only allows us to estimate protein half-lives that fall
into that range. If proteins have half-lives shorter than 12 hours or
longer than 32 days, we are not able to accurately estimate half-lives.
As a result, we may miss some proteins with oxygen-dependent
half-lives. The model we use is based on the assumption that protein
abundance is constant over the course of the study. We have previously
shown that acute hypoxia causes marked changes in behavior and tissue
oxygenation, but these functions are normalized after ~1 week
([183]60). Thus, we acclimatized the mice in respective oxygen tensions
for 1 week before the start of the isotope labeling. However, we cannot
rule out that the expressions of some proteins fluctuate (e.g.,
circadian proteins) throughout our experiment. We showed that the
stabilization of MYBBP1A is associated with hyperoxia responses
including alterations in rRNA biogenesis and increased expressions of
p53 targets. However, future studies are needed to causally prove this.
In summary, our study provides a comprehensive analysis of in vivo
protein turnover rates in varying oxygen tensions. This is relevant for
a broad range of clinical conditions including chronic lung disease,
hyperoxic lung injury, ischemic heart disease, and mitochondrial
disease. This work sheds light on hypoxia and hyperoxia-labile protein
complexes and subunits that may be relevant for several monogenic
disorders. Moreover, it serves as a valuable resource for future
studies of hypoxia and hyperoxia responses that are mediated by changes
in protein stability. As a key example, we nominate MYBBP1A as a
regulator of the hyperoxia response that is associated with changes in
rRNA homeostasis. By discovering mediators of oxygen-dependent
responses, we can identify therapeutic targets for states of mismatched
oxygen supply and demand.
MATERIALS AND METHODS
Animal model
C57BL/6J (IMSR_JAX: 000664) male mice (7-week-old) were purchased from
the Jackson Laboratory. Upon delivery, the mice were housed in the
University of California, San Francisco (UCSF) animal facility.
Different oxygen levels (8, 60, or 80% O[2]) were created by mixing
N[2](Airgas), O[2] (Airgas, Praxair), and room air carefully controlled
by gas regulators. The O[2] and CO[2] levels were continuously
monitored with wireless sensors and checked daily. To prevent CO[2]
buildup in the chambers, soda lime (Fisher Scientific, A1935236) was
placed inside the chambers to absorb CO[2]. All experiments were
performed on C57BL/6J mice between the ages of 8 and 12 weeks. All
animal studies were approved by the Institutional Animal Care and Use
Committee Program at UCSF.
Cell models
K562 [American Type Culture Collection (ATCC), CCL-243] and A549 (ATCC,
CCL-185) cells were purchased from ATCC and were maintained in
Dulbecco’s modified Eagle’s medium (DMEM; Gibco/Life Technologies,
11995073) supplemented with 10% fetal bovine serum (FBS; Corning/Fisher
Scientific, MT35015CV) and 1% penicillin-streptomycin (Fisher
Scientific, 15140122). Human lung fibroblasts (HLFs; Sigma-Aldrich
Inc., 506-05A) were maintained in DMEM/F-12 (Thermo Fisher Scientific,
11330032) supplemented with 10% FBS and 1% penicillin-streptomycin.
Human ATII epithelial cells were isolated from human lungs declined for
transplantation by the Northern California Transplant Donor Network as
previously described ([184]61). Cells were plated at 1 × 10^6 cells per
well on collagen I–coated Transwell plates (Corning, CLS3495,
Sigma-Aldrich) in an air-liquid interface. Cells were maintained in 50%
DMEM high-glucose/50% F-12 mix supplemented with 1%
penicillin-streptomycin, 1% fungisome, and 0.1% gentamicin. Mycoplasma
tests were performed quarterly on all cell lines. All cells were
maintained in cell culture incubators (37°C, 5% CO[2]). The oxygen
tension in the hyperoxia cell culture incubator was created by mixing
compressed air and 100% O[2] (Praxair).
Isotope labeling of mice and tissue collection
C57BL/6J male mice (four to six mice per time point) were fed on ^14N
and ^15N mouse chow (obtained from Silantes). Animals were first
acclimatized in respective oxygen tension with ^14N (normisotopic) food
for 1 week. Mice in the hypoxia group were first acclimatized in 11%
O[2] for 1 week before transferring to 8% O[2]. The animals were then
transitioned to ^15N chow throughout the labeling period (for 1, 2, 4,
8, 16, and 32 days). Mice were sacrificed by isoflurane inhalation. The
animals were perfused via the left ventricle with ice-cold PBS for 2
min. Heart, lung, and brain tissues were flash-frozen in liquid
nitrogen.
Proteomics sample preparation
Tissues were processed using the SPEED (sample preparation by easy
extraction and digestion) preparation method ([185]62). Briefly, 40 μl
of trifluoroacetic acid (TFA) was added and incubated for 3 min at
70°C. TFA was neutralized by adding 10 volumes of neutralization buffer
(2 M Trizma base in H[2]O). Cysteines were reduced and alkylated with 5
mM Tris(2-carboxyethyl)phosphine (TCEP) and 10 mM chloroacetamide, and
the samples were incubated at 95°C for 5 min. Protein quantification
was performed via the Bradford assay, and all samples were adjusted at
100 μg. The resulting volumes were diluted 1:1, and 1 μg of trypsin was
added. Proteins were digested overnight at 37°C on a thermo-shaker (600
rpm). Peptides were desalted using a 96-well format C18 plate (Nest
group) following the manufacturer’s instruction and dried under vacuum.
Mass spectrometry acquisition
The lung samples were resuspended in buffer A [0.1% formic acid (FA)],
and approximately 200 ng was analyzed by DIA-PASEF (parallel
accumulation-serial fragmentation combined with data-independent
aquisition) ([186]63) on a Bruker TimsTof Pro 2 interfaced with an
Ultimate3000 UHPLC. The peptides were separated on a PepSep column (15
cm length, 150 μm inner diameter) using a 38-min gradient at 0.5
μl/min. Following loading, the peptides were eluted with a 5 to 30%
buffer B [0.1% FA in acetonitrile (ACN)] in 20 min. The column was then
washed for 5 min at 90% and high flow (1 μl/min) and reequilibrated
with 5% ACN for the next run. The peptides were sprayed on a glass
capillary kept at 1700 V and 200°C. The mass spectrometer was operated
in a positive mode using the DIA-PASEF acquisition ([187]63). Briefly,
four PASEF scans (0.85 1/K0 to 1.30 1/K0) were acquired and divided
each precursor range into 24 windows of 32 Da [500.7502 to 966.67502
mass/charge ratio (m/z)] overlapping 1 Th.
The heart and brain samples (approximately 500 ng) were directly loaded
on an Evosep C18 tip and separated using the Evosep One using the 60
spd method (Evosep, Odense, Denmark). Peptides were eluted and ionized
using a Bruker Captive Spray emitter. A Bruker TimsTof Pro 2 mass
spectrometer running in DIA-PASEF mode ([188]63) was used for
acquisition. The acquisition scheme used for dia-PASEF consisted of 6 ×
3 50 m/z windows per PASEF scan.
Lung cell isolation
C57BL/6J male mice were sacrificed by isoflurane inhalation, soaked in
70% ethanol for 1 min, and then perfused with PBS. The heart tissues
were removed, and the trachea was tied with a loose knot. The lungs
were insufflated with dispase, collagenase, and deoxyribonuclease
I. Next, the lungs were removed and immersed in dispase in 37°C for 45
min on a rocker to digest the lungs. To prepare for sorting, the
suspension was filtered through a 70-μm filter and rinsed with sorting
buffer (DMEM/F12, no phenol red, with 2% FBS, and 1%
penicillin-streptomycin). The cells were spun down at 550g at 4°C for 5
min and resuspended in red blood lysis buffer. The cells were filtered
through a 40-μm filter, rinsed, and spun down at 550g at 4°C for 5 min.
Cells were mixed with rat serum at 4°C for 10 min to block nonspecific
binding. Cells were incubated in biotin-conjugated primary antibodies
(CD45, CD31, CD46, and Ter^199) for 30 min at 4°C and Streptavidin
beads at room temperature (RT) for 3 min to remove immune and
endothelial cells. Cells were then stained with Epcam, integrin B4, and
major histocompatibility complex class II (MHC-II) for 30 min at 4°C.
Live AT2 cells (lin^− Epcam^+ B4^− MHC-II^+) and line fibroblast (lin^−
Epcam^−) were sorted with a flow cytometer ([189]64, [190]65). See
table S8 for the catalog numbers and dilutions of the antibodies.
Biological triplicates were analyzed in the study, and two mice were
pooled for each replicate. Cells were pelleted and stored at −80°C.
Western blotting
C57BL/6J mice were exposed to room air or 80% O[2] (n = 3 animals per
group). Tissues were perfused with chilled Dulbecco’s
phosphate-buffered saline (DPBS; Corning). Lung tissues were collected
and flash-frozen in liquid nitrogen. Lung tissues were homogenized
using the Qiagen Tissue Lyser II (30 Hz, 1 min) in
radioimmunoprecipitation assay (RIPA) buffer (Thermo Fisher Scientific,
PI89901) with cOmplete Protease Inhibitor Cocktail (Roche), followed by
sonication. Cell pellets were lysed in RIPA buffer with cOmplete
Protease Inhibitor Cocktail. Protein concentrations were determined
using the Rapid Gold BCA Protein Assay Kit (Thermo Fisher Scientific,
[191]A53225). Equal amounts of protein were mixed with 6× Laemmli SDS
sample buffer (Fisher Scientific, AAJ61337AD), and samples were boiled
at 95°C for 5 min. Protein lysates were run on SDS–polyacrylamide gel
electrophoresis gels (Mini-PROTEAN TGX Precast Protein Gels) at 200 V
and were transferred onto polyvinylidene difluoride membranes.
Membranes were blocked with 3% nonfat milk (Genesee Scientific
Corporation, 20-241) in Tris-buffered saline, 0.1% Tween 20 (TBST)
(Fisher Scientific, 28360) for 1 hour at RT on a rocker. The membranes
were probed with primary antibodies overnight at 4°C (anti-MYBBP1A,
Proteintech, 14524-1-AP, 1:1000; anti-HK1, Cell Signaling Technology,
2024, 1:1000). The corresponding secondary antibodies were applied to
the membranes for 1 hour at RT (anti-rabbit horseradish peroxidase, VWR
95017-556; 1:5000). Bands were visualized using enhanced
chemiluminescence (Fisher Scientific, PI32106) on the Western Blot
Imaging System (Azure Biosystems) or x-ray films (GE Healthcare).
Uncropped blot images have been deposited to Mendeley Data (doi:
[192]10.17632/jx73jbnj3g.1).
RNA extraction
Mouse lung tissues were perfused with cold DPBS and flash-frozen in
liquid nitrogen. Tissues were homogenized in TRIzol reagent (Thermo
Fisher Scientific, 15596026) using the Qiagen Tissue Lyser II (30 Hz, 2
min). RNA extraction was performed according to the manufacturer’s
instructions. RNA purity and integrity were determined using NanoDrop
One (Thermo Fisher Scientific) and Bioanalyzer (Agilent), respectively.
Quant-seq 3′ mRNA-seq
The Quant-seq library preparation was carried out using the Lexogen
Quant-seq 3′ mRNA-Seq V2 Library Prep Kit with unique dual index (UDI;
#191.96). Equal amounts of RNA (500 ng) from samples (n = 7 for
normoxia and n = 6 for hyperoxia) were used for the preparation,
according to the manufacturer’s guide. Following the PCR amplification,
the complementary DNA (cDNA) concentrations were examined using a Qubit
double-stranded DNA High Sensitivity and Broad Range Assay Kit
(Invitrogen [193]Q32851) on a Qubit 4 fluorometer (Thermo Fisher
Scientific). The average library sizes were determined using
TapeStation 4200 (Agilent). Next, equal amounts of cDNA with UDIs were
pooled and sequenced on an Illumina’s NovaSeq 6000 System in the UCSF
Center for Advanced Technology (CAT).
Quantitative reverse transcription polymerase chain reaction
RNA was reverse-transcribed using the QuantiTect Reverse Transcription
Kit (Qiagen, 205311) according to the manufacturer’s instructions. The
qPCR reactions were performed using qPCR primers targeting different
loci of the rRNA ([194]Fig. 6C and table S8) ([195]66) and Maxima SYBR
Green/ROX qPCR Master Mix (Thermo Fisher Scientific, K02222) on a
QuantStudio 5 real-time PCR machine (Applied Biosystems). Hprt1 was
used as the housekeeping gene. RNA expressions relative to Hprt1 were
calculated using the delta Ct method. The experiments were performed in
four biological replicates and technical duplicates. Unpaired t test
was used for statistical analysis.
Generation of NDI1-expressing cell lines
Retroviruses were generated by transfecting PMXS-NDI1 (Addgene, 72876)
plasmid or empty vector in human embryonic kidney 293T cells (ATCC,
CRL-3216) with the packaging plasmids pVSVg (Addgene, 8454) and psPAX2
(Addgene, 12260). K562, A549, and HLF cells were transduced with NDI1
or empty vector virus by spinfection as previously described ([196]34).
Cells were selected with blasticidin (Gibco, A1113903) at 1 μg/ml for 4
days.
Mitochondrial respiration assay
K562 cells were pretreated in 21 or 50% O[2] for 4 days. Cell-Tak (25
μg/ml) was used for acute adhesion on the Seahorse XFe96 Cell Culture
Plates. On the assay day, 4.5 × 10^4 cells were plated per well. The
plates were centrifuged at 600g for 5 min. A549 cells were pretreated
in 21 or 50% O[2] for 3 days. An equal number of cells (1.5 × 10^4)
were plated on the Seahorse plate and incubated at respective oxygen
tensions for an additional day. The assay medium was composed of DMEM
powder (Sigma-Aldrich, D5030), 5 mM Hepes, 30 mM NaCl, 8 mM glucose, 2
mM pyruvate, and 2 mM glutamine (pH 7.4). Port A of the Seahorse XFe96
Sensor Cartridges was loaded with rotenone (final concentration = 500
nM). Port B was loaded with antimycin A (final concentration = 1 μM).
The experiment was performed in three biological replicates and five
technical replicates. Each well was measured five times at baseline and
after each injection. The basal oxygen consumption rate (OCR) was
determined by subtracting the average OCR by antimycin A injection from
the average OCR in baseline. The rotenone-resistant OCR was determined
by subtracting the average OCR after antimycin A injection from the
average OCR after rotenone injection.
MS data processing
The DIA-PASEF data were searched with DIA-NN v1.7.1 ([197]67) using a
library-centric approach. Identified spectra with MS1 precursors within
10 parts per million (ppm) and MS2 precursor within 15 ppm were
selected, and a second library was generated (double-pass mode).
Quantification was set to robust (high accuracy) while signal was
renormalized as function of the spray stability (RT dependent). Protein
inference was disabled, and library generation was set to smart
profiling. The transition level data were filtered at 1% library
Q-value, and transitions were summed into precursor MS2 abundances, and
precursors were averaged to a single-peptide abundance.
Degradation rate calculation and statistical analysis
The degradation rate (K[d]) is determined using a first-order kinetic
model ([198]19), under the assumption that the protein abundance
remains constant and that protein synthesis is a zero-order process.
The fraction unlabeled peptides (without ^15N) can be modeled as a
first-order kinetic model over the course the labeling time
[MATH: Fraction14N(t)=e−
kd×t :MATH]
The unlabeled peptide abundance was first centered using previously
reported long-living proteins ([199]22) under the assumption of fixed
abundance between different conditions and labeling time. The
normalized abundance was divided by the abundance at time 0 of the
labeling to calculate the fraction of unlabeled peptides at each time
point. Next, a preliminary fitting was performed to filter peptides
with good linear correlations using the Ordinary Least Square (OLS)
function from the Python (v3.11) statsmodels package (v0.13.5). Each
peptide for a given condition was fit into the first-order kinetic
model using the ordinary linear model, and the goodness of fit was
assessed by r^2. For the subsequent analysis, only peptides with r^2 >
0.65 that are detected in more than five samples in a given condition
were used for the fitting.
To calculate the K[d] for each protein, a linear mixed-effects model
was applied using the mixedlm function from the python (v3.11)
statsmodels package (v0.13.5). For each oxygen condition and each
tissue, time was set as the fixed variable, and to account for the
variability of different peptides for a given protein, the peptides
were set as the random variable. K[d] values with P values less than
0.20 were reported. Protein half-lives (t[1/2]) were calculated on the
basis of
[MATH:
t1/2=
ln(2)Kd :MATH]
To compare K[d] between oxygen tensions, a linear mixed-effects model
using mixedlm with oxygen, time, and oxygen:time as fixed effects, and
peptides as random variables was applied. The P values of the
interaction between oxygen and time (oxygen:time) indicated the
differences in the slopes, which assessed the differences of the
degradation rates. Adjusted P values were calculated using the
Benjamini-Hochberg correction.
Quant-seq sequence alignment and data processing
The data alignment and analysis were carried out using the BioJupies
automated analysis pipeline ([200]68). The genes were aligned and the
expressions were quantified using the kallisto pseudoaligner (v0.46.1)
([201]69). The data were normalized using the count per million method.
The statistics analysis was performed using limma (v3.54.2) ([202]70).
Biophysical feature analysis
The UniProt mouse canonical protein sequences (downloaded on 21 August
2022) were used as input to calculate various protein sequence features
as described below. “Disordered fraction” was computed on the basis of
the fraction of total residues with a score above 0.5 using the
DISOPRED3 method ([203]71). “Longest disordered stretch” refers to the
length of the longest disordered domain (defined as a continuous
stretch of sequence with each amino acid having a DISOPRED3 score above
0.5). “Disorder promoting” was computed on the basis of the fraction of
residues that is predicted to be disorder promoting using the
TOP-IDP–scale method ([204]72). “Fraction expanding” is defined by the
fraction of residues that contribute to chain expansion (E/D/R/K/P)
calculated using CIDER ([205]73). “Delta” score is a parameter for
protein charge patterning calculated using the localCIDER algorithm
([206]73). “Fraction charged AA”, “Fraction AA^−”, and “Fraction AA^+”
refer to the fraction of charged residues (H, K, R, D, and E), the
fraction of negatively charged residues (D and E), and the fraction of
positively charged residues (H, K, and R), respectively. “Hydropathy”
is the mean hydropathy as calculated from a skewed Kyte-Doolittle
hydrophobicity scale ([207]74). “Length” refers to the protein amino
acid length.
For enrichment analysis at baseline, the proteins are ranked on the
basis of the degradation rates. The top 5% and bottom 5% of proteins
from each organ were used to compare with the background distribution
(all other proteins). For the analysis for the oxygen-dependent
changes, the proteins were ranked by t-statistics calculated using the
linear mixed-effects model described above. The top 2% and bottom 2% of
proteins from each organ were used to compare with the background
distribution (all other proteins). The enrichment scores were
calculated using the Kolmogorov-Smirnov test.
Pathway analysis
For the pulsed-SILAM dataset, proteins with fold changes >1.3 and FDR <
0.05 in protein turnover rates between different oxygen tensions were
imported into Cytoscape (3.9.1) ([208]75). The enrichment analysis was
performed using the STRING Functional Enrichment in the Cytoscape
(confidence score cutoff: 0.70). Pathways with fewer than 1000 proteins
were selected and were ranked on the basis of P values.
For the Quant-seq dataset, RNA transcripts with absolute number of
log[2](fold change) > 1.5 and FDR < 0.05 were selected, and the
enrichment analysis was performed using EnrichR, and the genes were
mapped against the mouse Gene Ontology database ([209]76).
The transcription factor target lists were curated from the Cistrome
database ([210]77). All the datasets that were generated in mice and
passed the peak quality control were merged. The binding targets that
were found in at least 50% of all datasets were selected for the
analysis. The enrichment scores were calculated using the Gene Set
Enrichment Analysis (GSEA) ([211]78).
For the single-cell RNAseq data ([212]46), genes were ranked by fold
changes for each cell type. The pathway enrichment was performed using
GSEA.
Protein complex analysis
The protein complex data were retrieved from CORUM (on 3 September
2018) ([213]30). The data were imported to Cytoscape (3.9.1) to create
the network representations.
Acknowledgments