Datasets:
pmcid
large_stringlengths 8
11
| pmid
large_stringlengths 0
8
| authors
large_stringlengths 0
17.7k
| title
large_stringlengths 7
1.37k
| publication_date
large_stringlengths 10
10
| keywords
large_stringlengths 0
5.03k
| abstract
large_stringlengths 0
21.7k
| text
large_stringlengths 282
5.89M
| high_quality
bool 1
class | medium_quality
bool 1
class | low_quality
bool 2
classes |
|---|---|---|---|---|---|---|---|---|---|---|
PMC10000225 | Cristián A. Valenzuela,Marco Azúa,Claudio A. Álvarez,Paulina Schmitt,Nicolás Ojeda,Luis Mercado | Evidence of the Autophagic Process during the Fish Immune Response of Skeletal Muscle Cells against Piscirickettsia salmonis | 28-02-2023 | autophagy,skeletal muscle,fish,immune response,Piscirickettsia salmonis | Simple Summary In mammals, autophagy plays a fundamental role in the defense against intracellular pathogens; however, in fish, this noncanonical function has not been described. In this context, it was proposed to study whether autophagy was modulated/activated in muscle cells challenged with a bacterial pathogen. Muscle cell cultures were performed and challenged with Piscirickettsia salmonis, the main threat to the salmon industry. Genes associated with immune response and autophagy were evaluated. In addition, the protein content of the LC3-II-specific marker of the autophagic process was evaluated via Western blot. Additionally, genes associated with vesicular traffic and endocytosis were evaluated, finding that P. salmonis promotes these processes. The results show a concomitant modulation of the genes associated with the immune response, vesicular trafficking, and autophagy, suggesting an early intracellular response by the muscle cell against this bacterium. Due to the necessity of seeking and discovering new alternatives and strategies to fight intracellular pathogens in the salmon industry, a better understanding of how autophagy participates in immune system responses may lead to the development of new technologies that allow for the effective control of intracellular pathogens, improving animal welfare and contributing to the sustainability of the global fish industry. Abstract Autophagy is a fundamental cellular process implicated in the health of the cell, acting as a cytoplasmatic quality control machinery by self-eating unfunctional organelles and protein aggregates. In mammals, autophagy can participate in the clearance of intracellular pathogens from the cell, and the activity of the toll-like receptors mediates its activation. However, in fish, the modulation of autophagy by these receptors in the muscle is unknown. This study describes and characterizes autophagic modulation during the immune response of fish muscle cells after a challenge with intracellular pathogen Piscirickettsia salmonis. For this, primary cultures of muscle cells were challenged with P. salmonis, and the expressions of immune markers il-1β, tnfα, il-8, hepcidin, tlr3, tlr9, mhc-I and mhc-II were analyzed through RT-qPCR. The expressions of several genes involved in autophagy (becn1, atg9, atg5, atg12, lc3, gabarap and atg4) were also evaluated with RT-qPCR to understand the autophagic modulation during an immune response. In addition, LC3-II protein content was measured via Western blot. The challenge of trout muscle cells with P. salmonis triggered a concomitant immune response to the activation of the autophagic process, suggesting a close relationship between these two processes. | Evidence of the Autophagic Process during the Fish Immune Response of Skeletal Muscle Cells against Piscirickettsia salmonis
In mammals, autophagy plays a fundamental role in the defense against intracellular pathogens; however, in fish, this noncanonical function has not been described. In this context, it was proposed to study whether autophagy was modulated/activated in muscle cells challenged with a bacterial pathogen. Muscle cell cultures were performed and challenged with Piscirickettsia salmonis, the main threat to the salmon industry. Genes associated with immune response and autophagy were evaluated. In addition, the protein content of the LC3-II-specific marker of the autophagic process was evaluated via Western blot. Additionally, genes associated with vesicular traffic and endocytosis were evaluated, finding that P. salmonis promotes these processes. The results show a concomitant modulation of the genes associated with the immune response, vesicular trafficking, and autophagy, suggesting an early intracellular response by the muscle cell against this bacterium. Due to the necessity of seeking and discovering new alternatives and strategies to fight intracellular pathogens in the salmon industry, a better understanding of how autophagy participates in immune system responses may lead to the development of new technologies that allow for the effective control of intracellular pathogens, improving animal welfare and contributing to the sustainability of the global fish industry.
Autophagy is a fundamental cellular process implicated in the health of the cell, acting as a cytoplasmatic quality control machinery by self-eating unfunctional organelles and protein aggregates. In mammals, autophagy can participate in the clearance of intracellular pathogens from the cell, and the activity of the toll-like receptors mediates its activation. However, in fish, the modulation of autophagy by these receptors in the muscle is unknown. This study describes and characterizes autophagic modulation during the immune response of fish muscle cells after a challenge with intracellular pathogen Piscirickettsia salmonis. For this, primary cultures of muscle cells were challenged with P. salmonis, and the expressions of immune markers il-1β, tnfα, il-8, hepcidin, tlr3, tlr9, mhc-I and mhc-II were analyzed through RT-qPCR. The expressions of several genes involved in autophagy (becn1, atg9, atg5, atg12, lc3, gabarap and atg4) were also evaluated with RT-qPCR to understand the autophagic modulation during an immune response. In addition, LC3-II protein content was measured via Western blot. The challenge of trout muscle cells with P. salmonis triggered a concomitant immune response to the activation of the autophagic process, suggesting a close relationship between these two processes.
Autophagy is a highly conserved catabolic process essential for cellular homeostasis, degrading cytosolic macromolecules and damaged organelles to guarantee energy supply and nutrient availability for the cell. Furthermore, autophagy can modulate innate and adaptative immunity, contributing to the clearance of pathogens from the cell [1]. Under normal conditions, autophagy occurs at low levels in almost all cells, but is strongly induced under nutrient deprivation, amino acid starvation, cellular stress, cytokine exposure and pathogen infections [2,3]. Additionally, autophagy can act through a noncanonical pathway in antigen processing and presentation for MHC-I and MHC-II, restricting intracellular pathogens and regulating adaptive immunity [4]. At the molecular level, the autophagic pathway involves the concerted action of evolutionarily conserved ATG gene products involved in the initiation (atg9, atg13 and becn1), autophagosome formation (atg5, atg12 and atg16), the elongation of autophagic processes (atg7 and atg4), and autophagosome closure (gabarap and lc3) [5]. Autophagy is triggered by activating a regulatory protein complex that recruits microtubule-associated light chain protein 3 (LC3) to the nascent autophagosome [6]. The conversion of LC3 into its active form, LC3-II, through conjugation to phosphatidylethanolamine (PE), allows for the transformation of the membrane into a closed double-membrane system [7]. Lastly, the autophagosome binds to the endosomal or lysosomal compartment, forming the autolysosome where organelles and macromolecules are degraded by a pool of proteolytic enzymes [8]. In mammals, autophagic activation is associated with immune signals and molecular pathways, involving toll-like receptor (TLR) activation through different pathogen-associated molecular patterns (PAMPs) [9]. For instance, autophagy activated by the endosomal TLR signaling pathway in macrophages plays a key role as a mechanism for Leishmania major infection resistance [10]. Mammalian skeletal muscle, a nonclassical immune tissue, possesses several membrane-bound TLRs with the ability to detect different PAMPs, triggering signaling pathways to activate autophagy [11]. These findings associate TLR signaling and autophagy, providing a potential molecular mechanism for the induction of autophagy in response to pathogen invasion. Therefore, the ability of TLR ligands to stimulate autophagy may be used to treat intracellular pathogens in mammals [12]. Moreover, several proinflammatory cytokines, such as IFN-γ, TNF-α, IL-1, IL-2, IL-6, and TGF-β induce autophagy, while anti-inflammatory cytokines such as IL-4, IL-10, and IL-13 display an inhibitory effect [13]. All these data suggest a close relation between the immune response and autophagy in mammalian skeletal muscle. However, this interaction is unknown in fish, and only a few reports related to autophagy exist. Nevertheless, due to the direct relation of autophagy and growth, a key parameter for fish-industry productivity, autophagic markers have been studied under different diets and nutritional trials. Results showed a direct association between fish’s energy requirement and the activation of autophagy [14]. For instance, the autophagic flux and the overexpression of autophagic genes atg4, atg9, atg12, lc3, gabarap and becn1 was observed in vitro in epithelial gill cell line RT-gill-W1 of rainbow trout under nutrient restriction, and in the muscle of trout challenged with Flavobacterium psychrophilum [15,16]. Nevertheless, the modulation of autophagy in fish muscle cells during an immune-like response against an intracellular pathogen has not been elucidated. This study aims to describe and characterize autophagic modulation through the immune response of fish muscle cells after a challenge with a critical bacterial pathogen for the salmon industry. Through the immune challenge of primary cultures of muscle cells with Piscirickettsia salmonis, a fastidious intracellular pathogen, we revealed the transcript upregulation and protein production of several immune markers and genes involved in autophagy. Results obtained in the present work give new insight into the role of autophagy in the immune response against bacterial pathogens in muscle cells.
The present research was approved by the bioethics committee of Pontificia Universidad Católica de Valparaíso (PUCV) BIOEPUCV-B 334-2020 and the Agencia Nacional de Investigación y Desarrollo (ANID) of the Chilean government, and adhered to all animal welfare procedures.
In total, 24 rainbow trout (Oncorhynchus mykiss) with body weight of 10 ± 1.7 g and body length of 10 ± 1.5 cm were obtained from the Río Blanco Aquaculture Center (Valparaíso, Chile). These fish were kept in 3 tanks at a density of 8 fish per 25-L tank, using 1 tank per trial. Fish were maintained at a temperature of 16 °C, fed daily with commercial feed, and provided with a photoperiod of 13 h of light and 11 h of darkness.
Muscular cells were prepared from 6-8 g of the skeletal muscle of four rainbow trout as previously reported [17]. Briefly, the dorsal white muscle was collected and disintegrated in Dulbecco’s Modified Eagle’s Medium (DMEM) containing 9 mM NaHCO3, 20 mM HEPES, 10% horse serum at pH 7.4, 100 U/mL penicillin, and 100 µg/mL streptomycin (Sigma, St. Luis, MO, USA). After mechanical dissociation, the muscle was washed with DMEM and then digested with 0.2% collagenase solution in DMEM for 1 h at 18 °C. The suspension was centrifuged at 500 g for 5 min at 15 °C, and the resulting pellet was digested with a 0.1% trypsin solution in DMEM for 30 min at 18 °C. Trypsin was deactivated with complete medium (DMEM containing 10% fetal bovine serum, 100 U/mL penicillin, and 100 µg/mL streptomycin at pH 7.4), the suspension was filtered and centrifuged at 5000× g for 10 min at 18 °C, and 5 mL of DMEM was lastly added to the resulting pellet. The cell suspension was collected to count cells and evaluate viability with Trypan Blue 0.2%. Cells were seeded at a density of 5 × 105 per well in plates previously treated with poly-L-lysine (2 μg/cm2) and laminin (20 μg/mL). The cells were incubated at 18 °C for 14 days, 7 days in a proliferation medium containing DMEM, 9 mM NaHCO3, 20 mM HEPES, 10% fetal bovine serum, 100 U/mL penicillin, and 100 µg/mL streptomycin at pH 7.4, and 7 days in a differentiation medium composed of DMEM (9 mM NaHCO3, 20 mM HEPES, 100 U/mL penicillin, and 10 mg/mL streptomycin). This procedure was repeated three times (n = 3) per experimental trial described in the next sections.
A P. salmonis field isolate, Psal-104a, was obtained from the Chilean National Piscirickettsia salmonis Strain Collection at Pontificia Universidad Católica de Valparaíso (PUCV). The bacteria were cultured in basal medium (BM) broth composed of yeast extract (Merck, St. Louis, USA) 2.0 g L−1, peptone from meat (peptic digested, Merck) 2.0 g L−1, (NH4)2SO4 1.32 g L−1, MgSO4·7H2O 0.1 g L−1, K2HPO4 6.3 g L−1, NaCl 9.0 g L−1, CaCl2·2H2O 0.08 g L−1, FeSO4·7H2O 0.02 g L−1, and glucose 5 g L−1 at 18 °C at 100 rpm. Exponentially growing P. salmonis (O.D. 600 = 0.3) was recovered via centrifugation at 5000× g, washed three times with saline-buffered solution (SBS: 0.15 M NaCl, 7.3 mM KH2PO4, 11.5 mM K2HPO4, pH 6.0) supplemented with 4% lactose, and resuspended in DMEM containing 9 mM NaHCO3, 20 mM HEPES, at pH 7.4. Bacterial cells were counted using a Petroff–Hausser chamber. Muscular cells were infected at a multiplicity of infection (MOI) of 10 for 4, 6, 8, and, 24 h. After the experimental times, the pathogen was removed and then sampled. As a positive control for autophagic activation, muscle cells were treated with rapamycin (50 nm), and DMEM was used as a vehicle for control, sampled at the beginning of the experiment (0 hpi).
The total RNA from the primary culture of skeletal muscle was extracted according to the manufacturer’s guidelines for E.Z.N.A total RNA Kit I (Omega Bio-tek, Norcross, GA, USA), and quantified with NanoDrop LITE spectrophotometer (Thermo Scientific, Rockford, IL, USA). Residual genomic DNA was removed, and 1 µg of RNA was used for cDNA synthesis via RevertAid First strain cDNA Synthesis Kit (Thermo Scientific, Rockford, IL, USA) according to the manufacturer’s protocol. Real-time quantitative PCR (qPCR) assays were performed in a AriaMx Real-Time PCR System (Agilent Technologies, Santa Clara, CA, USA) with 15 µL volume reactions, of which 1 µL was cDNA (a 2-fold dilution), 0.2 μl of each primer (250 nM), and 8.8 µL of Takyon master mix (Eurogentec, Seraing, Belgium). The primers used in this study were designed and validated in our laboratory, and the sequence and efficiency of each primer are listed in Supplementary Table S1. Thermal cycling conditions were as follows: initial activation 95 °C for 10 min, 40 cycles of 30 s denaturation at 95 °C, 30 s annealing at 60 °C, and 30 s elongation at 72 °C, and each sample was loaded in duplicate. Relative expression analysis was conducted using geNorm software (https://genorm.cmgg.be/ (accessed on 30 March 2021)), and the results are expressed as fold changes using EF-1α and b-actin as housekeeping reference genes.
Total protein was extracted from primary culture cells in 150 µL Piercetm RIPA buffer (Thermo) supplemented with 8 mM EDTA, PMSF 1 mM, and a protease inhibitor cocktail (Calbiochem, San Diego, CA, USA), centrifuged at 12,000× g, and solubilized at 4 °C. Protein concentration was determined using a Pierce BCA Protein Assay Kit (Thermo Scientific, Hanover Park, IL, USA). The total protein (10 µg) was loaded into each line, separated with 18% SDS-PAGE, transferred to a nitrocellulose membrane, and blocked for 1 h at 37 °C in Tris-buffered saline (TBS-Tween) with 3% BSA. Primary antibody incubations (LC3A/B (D3U4C) XP® Rabbit mAb #12741, Actine (INVITROGEN), and ant-Rab7a, diluted 1:1000, 1:2000 and 1:500, respectively) were performed overnight at 4 °C. The anti-Rab7a polyclonal antibody was developed in our lab. Briefly, a recombinant Rab7a was used for specific antibody production (Supplementary Figure S1). CF-1 mice (5 weeks old) were subcutaneously injected at 1, 14, and 28 days with 30 μg of peptide diluted 1:1 in FIS, as a T helper cell activator, and in 1:1 Freund’s adjuvant (Thermo Scientific). The antiserum was collected on Day 42 and centrifuged at 700 g for 10 min, and the supernatant was stored at −20 °C. Antibody validation was determined via Western blot (Supplementary Figure S1). Membranes were washed with 1X TBS and incubated for 1 h at room temperature with the appropriate secondary antibody. After washing, the membranes were visualized using commercial kit WESTAR SUPERNOVA from Cyanagen. Then, the membrane was visualized with the ChemiDoc Imaging Systems Bio-Rad system (Bio-Rad Laboratories, Hercules, CA, USA). Digitized images were used for the densitometric analyses of the bands in Image J software(National Institutes of Health, Bethesda, Maryland, USA). Differences are expressed as fold changes in protein content. Western blots were performed for all individual samples per experimental treatment; however, only one representative blot film is shown in the figures (Supplementary Figures S2–S4).
Adherent cells were washed with 1X PBS and fixed with 4% paraformaldehyde (PFA) (in 1x PBS) for 10 min and permeabilized/blocked with 3% BSA with 0.3% Triton in 1X TBS for 30 min, and the quenching solution (50 mM ammonium chloride) was added for 10 min. Then, the cells were incubated overnight with the anti-LC3 antibody diluted at 1:100 (LC3A/B (D3U4C) XP® Rabbit mAb #12741 in 1% BSA at room temperature. Samples were washed with 1X PBS and 0.02% TBST, and incubated with a commercial secondary antibody (1:750) (Invitrogen, Thermo Scientific) for 90 min at room temperature. Nuclear staining was performed with DAPI (Vector Laboratories) for conventional fluorescence microscopy, following the manufacturer’s instructions. For image capture, the samples were placed in a Leica CTR5000fluorescence microscope (Leica Microsystems, Wetzlar, Germany). Western blots were analyzed using the LC3-II/actin and Rab7a/actin ratios.
All generated data were evaluated with one-way ANOVA and Student’s t test. Data are expressed as the mean (n = 3 per treatment) ± standard error of the mean (SEM), and p < 0.05 was considered statistically significant. All analyses were performed using general linear models on Graph Prism 7.0 software (GraphPad Company, Boston, MA., USA).
To understand and characterize the response of muscle cells against bacterial pathogens, rainbow-trout skeletal muscle cells were challenged with P. salmonis for 6, 8, and 24 h. Next, immune molecular markers tlr3, tlr9, il-1β, tnfα, il-8, hepcidin, mhc-I and mhc-II were measured with RT-qPCR, and results showed a concomitant overexpression of toll-like receptors 3 and 9 after 6 h of challenge (Figure 1A). Proinflammatory cytokines il-1β, tnfα and il-8 were modulated compared to the noninduced controls, where tnfα and il-8 were upregulated significantly after 6 and 8 h after the challenge, respectively (Figure 1B,C). Antimicrobial peptide hepcidin increased after 6, 8, and 24 h after the challenge, but no statistical significance was observed (Figure 1C). No changes were observed in the mRNA levels of mhc-I and mhc-II during the entire trial (Figure 1D).
To describe the modulation of autophagy in muscle cells after bacterial challenge, autophagic markers were analyzed via RT-qPCR, Western blot, and immunofluorescence. Rapamycin was used as the positive control of autophagic activation. Autophagic genes associated with the induction of this process were upregulated, particularly becn1, of which the mRNA level was significantly increased after 6 and 24 h of the challenge (Figure 2A). In the case of autophagosome formation genes, the overexpression of atg5 was observed after 6 h of treatment, while the transcript level of atg12 was downregulated after 8 h (Figure 2B). The mRNA levels of atg4, an elongation of an autophagosome molecular marker, was downregulated after 8 h of challenge (Figure 2C). Lastly, the transcript level of lc3 was significantly downregulated after 8 h of challenge (Figure 2D). At the protein level, protein autophagic marker LC3-I/LC3-II was evaluated via immunofluorescence and Western blot. Agreeing with the gene expression of autophagic induction gene analysis, the protein content of LC3 showed an increase in fluorescence signal in the muscular cells challenged with P. salmonis compared with the control group at 6 h after challenge (Figure 3A). An increase in the LC3-II protein content was observed through Western blot analysis after 6 h of stimuli (Figure 3B). In addition, the mRNA levels of small GTPase proteins rab5a and rab7a were upregulated at 6 h after challenge, while the transcript level of rab11a was increased after 24 h (Figure 3C). Lastly, the protein content of Rab7 showed an increasing trend through Western blot (Figure 3D).
In mammals, the ability of muscles to respond and participate in the immune response against pathogens and in autoimmune events is well-characterized [18]. This ability is represented by the immunocompetent attributes of muscle cells, such as the expression of costimulatory molecules, antigen-presenting machinery, proinflammatory cytokines, and pathogen recognition receptors [19,20,21,22,23]. Moreover, the interaction between muscle cells and resident immune cells is an essential step for wound healing and damage regeneration in this tissue [24,25]. In fish, the skeletal muscle appears as an active immunological organ, where immune reactions occur after an in vivo challenge with bacterial pathogens [26,27]. However, the ability of a muscle cell to express classical immune-like molecules by itself is still under study, and only few reports address how this type of cells may respond against PAMPs and pathogens [28,29]. For example, rainbow-trout myotubes responded against Piscirickettsia salmonis through the upregulation of tlr1, tlr22, and il-8 at 8 h after a challenge [17]. In the present study, the immune response against P. salmonis generated a concomitant overexpression of tnfα, tlr3, and tlr9 after 6 h of challenge. Toll-like receptors 3 and 9 are associated with an intracellular/antiviral response where these receptors can detect double-strand RNA and CpG DNA, respectively [30]. Reports in mammals showed how these receptors can be involved in diminishing the growth of intracellular bacterium Salmonella typhimurium [31]. Salmon head kidney cells (SHK-1) respond to an infection with P. salmonis through the overexpression of tlr1 and tlr5s [32]. Furthermore, P. salmonis could induce a flagellin-dependent TLR5 activation in Atlantic salmon, which resulted in the upregulation of proinflammatory genes [33]. The overexpression of intracellular receptors suggests that intracellular defense mechanisms could be activated in response to P. salmonis infections. Despite the overexpression of markers associated with an intracellular response, the RNA levels of molecular markers DotA and DotB of P. salmonis were not detected via RT-qPCR (data not shown), which could indicate that, even though the bacterium does not enter the cell, the intracellular defense mechanisms are stimulated. On the other hand, proinflammatory cytokine tnfα represents an early response of muscular cells against bacterial pathogens. Indeed, an increase in NF-κB and TNFα at 48 h after a challenge against Vibrio anguillarum was observed on fish muscle [34]. Additionally, at 8 h after a challenge, the mRNA level of il-8 was upregulated. In higher vertebrates, TNFα induces the expression of IL-8 [35]. This kind of response may be associated with the need for muscle cells to recruit immune cells to the site of infection, since IL-8 is a chemoattractant for leukocytes [36]. These results contribute to existing data confirming that the muscle cell responds to a bacterial challenge by itself. However, communication with the immune resident’s cells is needed to deploy an effective and robust response. In mammals, the participation of autophagy in the immune response is well-characterized, where autophagy-related gene (ATG) proteins mediate direct pathogen degradation, inflammation restriction, antigen presentation on MHC molecules and survival of memory lymphocyte populations [37]. This process has mainly been studied for its role in the internal quality control of the cell, and for its implication in metabolism and cancer. However, autophagy can participate through a noncanonical pathway to respond against other dangerous stimuli, such as pathogen infection [38]. In fish, the modulation of this process in a pathological context is unclear, and only a few reports show how autophagy is modulated by nutrient restriction and pharmacological treatments [14,39]. Transgenic zebrafish and zebrafish cell lines were used as research tools on the autophagic regulation process, highlighting its role on the protection against pathogen infection, development, and lipid degradation [14]. Autophagy was also linked to development and extracellular matrix formation in zebrafish [40]. Recently, the modulation of autophagic genes atg4, becn1, atg7, lc3, atg12, and gabarap has been evaluated in rainbow-trout liver and muscle under different functional treatments, fasting, and a posteriori F. psychrophilum infection, suggesting a possible role of this process in the resistance against pathogens [16]. In the present work, autophagic genes were up- and downregulated in a time-dependent and coordinated way. The upregulation of becn1 was observed at 6 h after challenge; becn1 is a protein that facilitates the de novo formation of the phagophore, and is critical for the autophagic initiation in embryogenesis and antiviral responses in fish [41,42]. The next step in autophagy is autophagosome formation, where molecules atg5 and atg12 are directly involved [43]. We observed the upregulation of atg5 at 6 h after a challenge with the subsequent downregulation of atg12 after 8 h. Both genes are associated with intracellular immune response against viruses in fish, where atg5 plays crucial roles in viral replication via promoting autophagy in orange-spotted grouper [44], while atg12 can induce both autophagy and the Type I IFN response in large yellow croaker [45]. These two molecules are regulated by RAB5, a small GTPase, which facilitates the early formation of the autophagosome [46]. Six hours after a challenge with P. salmonis, the mRNA levels of rab5a and rab7 increased significantly, suggesting the activation of vesicle traffic in response to the bacteria. In mammals, Rab7a is associated with the endocytic pathway [47]; in fish, this molecule is related with an antiviral response [48], suggesting the activation of different intracellular mechanisms in response to P. salmonis. Lastly, the genes involved in the elongation and closure of the autophagosome were downregulated after 8 h after challenge, particularly atg4 and lc3. Altogether, the present data suggest a tight relation between autophagy and intracellular innate immune response, and are consistent with the used intracellular bacterial pathogen and the evaluated immune markers in our study. A crosstalk between the immune response and autophagy has been reported in mammals. For example, TLRs were directly involved with autophagic activation after TLR ligand stimulations with poly I:C and LPS, among other PAMPs [9]. In addition, the stimulation of TLR7 with different ligands activated autophagy, eliminating intracellular microorganisms even when the target pathogen had not been canonically associated with TLR7 signaling [12]. In this context, the recognized ability of TLR ligands to stimulate autophagy was proposed as a prophylactic alternative to treat intracellular pathogens [12]. In the present study, tlr3, tlr9, tnfα and autophagic genes becn1 and atg5 were upregulated at the same experimental time (6 h after challenge), suggesting a relation between the autophagy and immune response of muscle cells against P. salmonis. This relation was confirmed in mammals where TNFα and other cytokines could induce autophagy [13]. Lastly, to corroborate the activation of autophagy in rainbow-trout muscle cells, we assessed the protein level of LC3-II, a classical protein marker for autophagy. Finding higher protein content of LC3-II in the treated cells added to the presented overall results and the background information, suggesting that the muscle cell deploys an immunelike response where autophagy may play an important role for the clearance of the pathogen. Deeper functional analysis and characterization are required to corroborate the dialogue and regulation between the two essential cellular functions.
The skeletal muscle of fish is a fundamental energy source for the organism and a recurrent site of infection where immunelike responses against different bacterial pathogens occur. The present study described for the first time the modulation of the immune response and its possible connection with autophagy in rainbow-trout muscle cells. The challenge of trout muscle cells with P. salmonis, an intracellular pathogen, triggered an immune response concomitant to the activation of the autophagic process. Results suggest an early response by the muscle cells against the bacteria just after only 6 h of challenge, where both immune and autophagic markers were upregulated. Considering the obtained data and the available research information, two important questions arise regarding the possible implication of P. salmonis challenge in fish muscle: (1) is autophagy acting as an intracellular immune mechanism to fight this pathogen or is the bacterium using this mechanism to enter the cell and evade the immune response? These questions are relevant and represent a new line of research that must be answered. Due to the necessity of seeking and discovering new alternatives and strategies to fight intracellular pathogens in the salmon industry, a better understanding of how autophagy participates in immune system responses may lead to the development of new technologies that allow for the effective control of intracellular pathogens, improving animal welfare and contributing to the sustainability of the global fish industry. | true | true | true |
|
PMC10000373 | Samrat Roy Choudhury,Billie Heflin,Erin Taylor,Brian Koss,Nathan L. Avaritt,Alan J. Tackett | CRISPR/dCas9-KRAB-Mediated Suppression of S100b Restores p53-Mediated Apoptosis in Melanoma Cells | 24-02-2023 | melanoma,S100b,CRISPR,dCas9-KRAB,apoptosis,cell death | Overexpression of S100B is routinely used for disease-staging and for determining prognostic outcomes in patients with malignant melanoma. Intracellular interactions between S100B and wild-type (WT)-p53 have been demonstrated to limit the availability of free WT-p53 in tumor cells, inhibiting the apoptotic signaling cascade. Herein, we demonstrate that, while oncogenic overexpression of S100B is poorly correlated (R < 0.3; p > 0.05) to alterations in S100B copy number or DNA methylation in primary patient samples, the transcriptional start site and upstream promoter of the gene are epigenetically primed in melanoma cells with predicted enrichment of activating transcription factors. Considering the regulatory role of activating transcription factors in S100B upregulation in melanoma, we stably suppressed S100b (murine ortholog) by using a catalytically inactive Cas9 (dCas9) fused to a transcriptional repressor, Krüppel-associated box (KRAB). Selective combination of S100b-specific single-guide RNAs and the dCas9-KRAB fusion significantly suppressed expression of S100b in murine B16 melanoma cells without noticeable off-target effects. S100b suppression resulted in recovery of intracellular WT-p53 and p21 levels and concomitant induction of apoptotic signaling. Expression levels of apoptogenic factors (i.e., apoptosis-inducing factor, caspase-3, and poly-ADP ribose polymerase) were altered in response to S100b suppression. S100b-suppressed cells also showed reduced cell viability and increased susceptibility to the chemotherapeutic agents, cisplatin and tunicamycin. Targeted suppression of S100b therefore offers a therapeutic vulnerability to overcome drug resistance in melanoma. | CRISPR/dCas9-KRAB-Mediated Suppression of S100b Restores p53-Mediated Apoptosis in Melanoma Cells
Overexpression of S100B is routinely used for disease-staging and for determining prognostic outcomes in patients with malignant melanoma. Intracellular interactions between S100B and wild-type (WT)-p53 have been demonstrated to limit the availability of free WT-p53 in tumor cells, inhibiting the apoptotic signaling cascade. Herein, we demonstrate that, while oncogenic overexpression of S100B is poorly correlated (R < 0.3; p > 0.05) to alterations in S100B copy number or DNA methylation in primary patient samples, the transcriptional start site and upstream promoter of the gene are epigenetically primed in melanoma cells with predicted enrichment of activating transcription factors. Considering the regulatory role of activating transcription factors in S100B upregulation in melanoma, we stably suppressed S100b (murine ortholog) by using a catalytically inactive Cas9 (dCas9) fused to a transcriptional repressor, Krüppel-associated box (KRAB). Selective combination of S100b-specific single-guide RNAs and the dCas9-KRAB fusion significantly suppressed expression of S100b in murine B16 melanoma cells without noticeable off-target effects. S100b suppression resulted in recovery of intracellular WT-p53 and p21 levels and concomitant induction of apoptotic signaling. Expression levels of apoptogenic factors (i.e., apoptosis-inducing factor, caspase-3, and poly-ADP ribose polymerase) were altered in response to S100b suppression. S100b-suppressed cells also showed reduced cell viability and increased susceptibility to the chemotherapeutic agents, cisplatin and tunicamycin. Targeted suppression of S100b therefore offers a therapeutic vulnerability to overcome drug resistance in melanoma.
Melanoma is one of the deadliest forms of skin cancer, is particularly prevalent (2.5%) among the Caucasian population, and has almost doubled in incidence over the past three decades in the United States alone [1]. Several serological biomarkers have been identified for early detection, staging, prognosis, and therapeutic determination of melanoma. Among them, a high serum level of S100B (S100 calcium-binding protein B) has emerged as the most reliable biomarker of progression and survival outcome of the disease [2,3]. S100B (10.7 kDa) is a member of the S100 protein family that binds to its molecular targets by undergoing conformational changes at the carboxy-terminal EF-hand motif. The affinity of S100B-mediated protein–protein interactions is strengthened with the increase in intracellular Ca2+ concentrations. Ectopic upregulation of the protein has been noted in metastatic melanomas, as well as proneuronal, neural, or classic types of gliomas [3,4,5]. In melanoma, intracellular S100B interacts with proteins of different signaling pathways [6], such that S100B activates the glycolytic enzyme fructose 1,6-biphosphate (aldolase) and increases metabolism of melanoma cells. The protein also interacts with cytoskeletal components such as tubulin, Rac1 (GTPase), or IQGAP1 (cdc42 effector), which alters motility of melanoma cells toward enhanced migration and invasion [7]. In contrast, when extracellularly secreted via the receptor for advanced glycation end products (RAGE) signal transduction pathway, S100B facilitates tumor development in a mouse model of melanoma [8]. Additionally, S100B triggers melanoma tumor growth by interacting with the C terminus of wild-type (WT)-p53, preventing protein tetramerization and covalent modifications (e.g., phosphorylation or ubiquitination) [9]. Therefore, S100B–p53 interaction lowers the intracellular availability of free WT-p53 and limits the tumor-suppressing function of TP53 [6,10,11]. Previously, siRNA-mediated cell-type-specific targeted inhibition of the S100B–p53 complex was shown to rescue protein levels of WT-p53, phosphorylated p53, and downstream p21 [10]. The targeted inhibition also evoked poly-ADP ribose polymerase (PARP)-mediated apoptosis involving activation of caspase-3 or caspase-8 or aggregation of Fas death receptor upon ultraviolet irradiation [10]. Several attempts have been made to develop small-molecule inhibitors (e.g., pentamidine) that inhibit molecular interactions between the Ca2+-S100B apoprotein and its binding proteins [12]; however, the small-molecule inhibitors, as well as siRNA-mediated knockdown, of S100B resulted in only transient restriction of melanoma cell growth. We aimed to develop a more stable and effective targeted approach that uses a CRISPR (clustered regularly interspaced short palindromic repeats) platform for stable suppression of S100B protein. Targeted epigenome editing previously was attempted by using a CRISPR and associated protein 9 (Cas9) endonuclease system that contained deactivated Cas9 (dCas9) fused to a transcriptional suppressor, Krüppel-associated box repressor (KRAB) [13]. The KRAB domain exerts widespread transcriptional suppression through enrichment of H3K9me3 and condensation of chromatin [14]. Herein, we report development of a dCas9-KRAB (DK) system for targeted perturbation of S100B expression to rescue WT-p53 and its associated tumor-suppression properties in melanoma cells.
Data from 28 retrospective studies available through the Cancer Genome Atlas (TCGA) were used to examine S100B expression in samples from patients with cutaneous (n = 443) or ocular (n = 80) melanoma, as well as patients with different cancer types (n = 10,071; comparison group) (Table S1). From the TCGA Firehose Legacy study, we selected a cohort of patients with skin melanoma (n = 287) on the basis of available data for DNA methylation, copy number (CN) alterations, and gene expression from the matched patient samples (Table S2) [15]. Genetic and transcriptomic data from TCGA were extracted through the cBioPortal (www.cbioportal.org; accessed on 17 January 2022). GEP was analyzed from mRNA expression (RNA sequencing [RNA-seq]) data; we represent GEP data expressed in RNA-seq by expectation-maximization (RSEM), log2 scale. CN alterations are presented as the capped relative linear copy-number values, where we consider patients with diploid genomes for S100B, compared to those with gain (+1) or loss (–1) in CN. There were single cases for amplification (+2) or deep deletion (–2); these were excluded from the analysis. DNA methylation was assayed on an Illumina HumanMethylation450 (HM450) BeadChip platform (San Diego, CA, USA), and the average β-value from all probe sets against S100B was considered. Raw values for GEP, CN, and methylation from matched patients are summarized in Table S3.
A proteomics data set was analyzed from the consented (study #204543) and deidentified tissue biopsies from a cohort of patients with melanoma (n = 23) that was diagnosed at stage IV and treated with first-line ICIs (i.e., anti-CTLA4, PD-1, or a combination of monotherapies) at the University of Arkansas for Medical Sciences. The proteomics sample was prepared, and data were analyzed as reported earlier [16]. Clinicopathological information of the patients, including response to ICIs, is summarized in Table S4.
The chromatin state at the S100B promoter, including transcriptional start sites (TSSs) and intragenic regions, was examined by comparing DNase-sequencing data from SK-MEL-5 melanoma cell lines (accession ID: ENCFF627WEJ) to that from primary keratinocytes derived of newborn foreskin (accession ID: ENCFF910KFI); both data sets were acquired from the ENCODE database. The DHS signals were represented as read-depth normalized signal. S100B mRNA levels in the same cell line (accession ID: ENCFF249LKO) and primary keratinocytes (accession ID: ENCFF249BFY) were also analyzed. The bigwig (hg38) files for both DHSs and RNA-seq data were loaded on the integrative genomics viewer (v 2.14.0, Broad Institute, Cambridge, MA, USA) for visualization.
The amino acid sequences of human (S100B) and murine (S100b) homologs of the protein were compared by using Clustal Omega (v 1.2.4) (EMBL-EBI, Hinxton, UK) function and were presented by using Jalview (v 2.11.2.0) [17].
The putative TFs that interact with the murine single-guide RNA (m-sgRNA)-binding sites on S100b were predicted by using the target regions as input for the PROMO database (http://alggen.lsi.upc.es/; accessed on 21 May 2020) that uses TRASFAC for the prediction analyses [18,19]. The output was specific to Mus musculus.
A plasmid containing the KRAB domain fused to dCas9 and the programmable sgRNA vector were obtained from Addgene, Watertown, MA, USA (#99372, and #44248, respectively). The plasmid #44248 contains an EGFP (enhancer green fluorescent protein)-specific sgRNA and has been used as an off-target control for the present study. Three murine S100b-specific sgRNAs were selected for the present study, based on their predicted efficacy of inhibition at the target site. m-sgRNA-1 and m-SgRNA-2 were selected from the sgRNA library that was recommended for the mouse genome [20]; m-SgRNA-3 was custom designed by using the CHOPCHOP (https://chopchop.cbu.uib.no/; accessed on 2 July 2020) algorithm design tool (Table S5). All sgRNA-binding sites were mapped with Mouse Genome Informatics (http://www.informatics.jax.org/; accessed on 12 June 2020) and annotated against the GRCm38/mm10 genome. The selected sgRNAs in combination with the scaffolds were obtained as gene blocks (gblocks) from Integrated DNA Technologies, Coralville, IA, USA. The gblocks were PCR amplified (Table S6) with CloneAmpHiFi PCR Premix (Takara, Kusatsu, Shiga, Japan) and were integrated into the sgRNA plasmid, flanked by BstXI and XhoI (New England Biolabs, Ipswich, MA, USA) sites. Correct clones were confirmed with low-throughput sequencing (Table S6, sequencing primers).
Murine B16-F1 (CRL-6323) melanoma and B16-F10 (CRL-6475) metastatic melanoma cells were purchased from ATCC (American Type Culture Collection, Manassas, VA, USA). Cell lines were cultured in Dulbecco’s Modified Eagle Medium supplemented with 10% FBS, 1% penicillin–streptomycin (Thermo–Fisher, Waltham, MA, USA), and 1% glutamine and maintained at 37 °C and 5% CO2. Cells were transduced with plasmid encoding DK alone, or in combination with the plasmids encoding sgRNAs specific to S100b or EGFP (sgRNAEGFP). Transduction was accomplished with a one-step lentivirus packaging system (Takara, Kusatsu, Shiga, Japan) per the manufacturer’s instructions. Briefly, 4–5 × 106 LentiX-293T cells (Takara) were transfected with the plasmids in Opti-MEM reduced serum medium (Thermo–Fisher, Waltham, MA, USA) and were incubated for 4–6 h. Next, each culture was supplemented with 14 mL complete medium and then incubated for 48 h. Supernatants were collected, passed through 0.45 µM syringe filters, and added dropwise to polybrene (8 µg/mL) charged F1 or F10 cells. Cells co-transduced with DK and sgRNA constructs were selected with puromycin (1 µg/mL), followed by FACS sorting (FACSARIA III, BD, San Jose, CA, USA) for mCherry+ cells (>95%).
Expression of target genes was determined in reference to endogenous GAPDH, out of total RNA extracted from the cells. From cells that had been treated for 24 h, total RNA was extracted (RNeasy Mini Kit, QIAGEN, Hilden, Germany) and converted to c-DNA (Iscript Advanced cDNA synthesis kit, Bio-Rad, Hercules, CA, USA). The fold change in S100b expression was then determined, in triplicate, with qPCR (StepOnePlus Real-Time PCR Systems; v 2.0 Applied Biosystems, Waltham, MA, USA), compatible with SYBR green master mix (Thermo–Fisher). Amplification reactions were carried out at 95 °C for 1 min, followed by 40 cycles at 95 °C for 15 s and 60 °C for 1 min. Primers used to amplify murine S100b (qMmuCID0015305), TP53 (qMmuCIP0032520), and GAPDH (qMmuCED0027497) were purchased from commercial sources (PrimePCR SYBR Green Assay, Bio-Rad, Hercules, CA, USA). The primers for DIP2A, PRMT2, CDKN1A, AIFM1, CASP3, PARP-1, and GAPDH were custom synthesized with IDT (Integrated DNA Technologies, Coralville, IA, USA), and primer sequences are summarized in Table S7.
We prepared whole-cell lysates by using RIPA buffer (Thermo–Fisher, Waltham, MA, USA) per the manufacturer’s instructions. To determine levels of S100b protein, an aliquot of lysate (40 µg total protein, determined with the BCA assay; Pierce) was loaded onto 4–20% Bis–Tris gels (Thermo–Fisher, Waltham, MA, USA); to determine levels of p21, p53, AIFM1, cleaved Caspase-3, or PARP-1 proteins, aliquots of lysate (approximately 20 µg total protein) were loaded onto 4–12% Bis-Tris gels (Thermo–Fisher, Waltham, MA, USA), and proteins were electrotransferred to PVDF membranes (Bio-Rad, Hercules, CA, USA). The blots were blocked in 5% milk in TBST for 1 h, followed by overnight incubation at 4 °C with primary antibodies in 1% milk. Blots were probed with primary antibodies (Table S8) per the manufacturer’s instructions. Blots were then washed 3 times with 1% milk and incubated with HRP-tagged secondary antibody for 1 hour. Finally, blots were washed 3 times in TBST and were developed with enhanced chemiluminescence reagents (Perkin Elmer, Waltham, MA). Densitometries of the bands from the blots (wherever applicable) were prepared with ImageJ (https://imagej.nih.gov/ij/; accessed on 23 February 2023), v 1.53t, National Institute of Health, Bethesda, MD, USA).
F1 and F10 cells were seeded in 6-well plates (5 × 104 cells per well). After 24 h, they were treated with staurosporine (Seleckchem, Houston, TX, USA) at indicated doses for 24 h. Apoptosis and cellular mortality then were determined with flow cytometry (FACS Verse, BD Biosciences, San Jose, CA, USA); staining with annexin V-FITC (Biolegend, San Diego, CA, USA) and DAPI (Sigma, St. Louis, MO, USA) was used to determine the percent viable (i.e., Annexin V–, DAPI–) cells.
Both B16 melanoma cell lines were seeded (5 × 104 cells per well) in 12-well plates and imaged (from at least 10 different fields) after 24 and 48 h to examine differences in cell proliferation between the treatment and control groups. Additionally, B16 melanoma cells were seeded in 96-well plates (5 × 103 cells per well), incubated for 24 h, and treated as indicated. Viability of co-transduced (DK + m-sgRNA-1) cells and untreated control cells was determined with a CellTiter-Glo Luminescent Cell Viability Assay kit (G7570, Promega, Madison, WI, USA) after 24 and 48 h in the presence or absence of two chemotherapeutic agents, cisplatin (13119, Cayman Chemicals, Ann Arbor, MI, USA) and tunicamycin (11445, Cayman Chemicals, Ann Arbor, MI, USA). Drugs were serially diluted from the stock solution (2 mM) and added to cell cultures in the range of 5 µM to 10 nM. At the end of 24 and 48 h of incubation, equal volumes (100 µL each) of assay solution and cell suspension were mixed and incubated at room temperature for 10 min. Relative luminescence was recorded with a spectrometer (Agilent BioTek, Winooski, VT, USA Gen5 Microplate reader), and the half-maximal inhibitory concentration (IC50) value for each drug was plotted with GraphPad Prism, v7.0, Boston, MA, USA.
We used the non-parametric Mann–Whitney U test or Dunn’s multiple comparison test to determine the significance of differences between the groups being compared. Significance was defined as p < 0.05. Statistical analyses and associated graphs were generated with GraphPad Prism, Boston, MA, USA.
We evaluated S100B expression (RNA-seq V2, RSEM, log2) in 10,071 patient-derived samples of 28 different cancers from the TCGA repository (Table S1). Expression of S100B was highest in glioblastoma, followed by skin cutaneous melanoma (SKCM). S100B expression in SKCM samples (n = 443) ranged (25–75% quartile) from 4287.6 to 16,048.4 with a median of 8959.4 ± 14,212.4 (standard deviation). In samples of ocular melanoma (UVM) (n = 80), the gene was expressed at a similar fashion to SKCM or was overexpressed (646.7 ± 1474.9), relative to other cancer types (Figure 1A). A separate in-house proteomics study of a cohort of patient-derived SKCM samples demonstrated that, in the subgroup (n = 12) that was non-responsive to ICIs anti-CTLA-4, anti-PD1, and anti-CTLA4 + anti-PD1 combined therapy, the S100B protein was upregulated (p < 0.05, >2-fold change), compared to the subgroup (n = 10) that was responsive (Figure 1B). These findings strongly suggest that ectopic upregulation of S100B might play a critical role in development of refractory SKCM [10,21] and could contribute to therapeutic resistance in the disease. Next, we combined the data on S100B expression with data on CN alterations and DNA-methylation profiles in matched SKCM samples (n = 285), available from TCGA. We observed that 17.5% (50/285) of samples had gains in S100B CN, and 21% (60/285) of samples had heterozygous deletions in CN of S100B, while 61.4% (175/285) of samples were diploid for S100B. A significant (p < 0.01) decrease in median expression of S100B (RSEM = 5624.3) was observed in samples with CN losses, compared to those with gains (RSEM = 14,119.5) (Figure 1C) but not to those with diploidy. However, weak correlations (p > 0.05; R2 < 0.2) were observed when the relative linear values of CN for individual subgroups (gain, loss, or diploid) and their expression profiles were compared (Figure 1D). We also investigated whether S100B expression correlates to DNA methylation (β-values, probe: cg12092309) of S100B. Methylation ranged (25–75% quartile) from 0.54 to 0.87, with a median value 0.74. We observed a moderate (R2 = 0.216; p < 0.01) negative correlation between S100B methylation level and expression in SKCM samples, such that samples with lower methylation levels have relatively higher expression and vice versa (Figure 1E). Additionally, we analyzed signal intensities based onDHS (DNAse hypersensitive sites) to assess the chromatin state of S100B in the SK-MEL-5 cell line, which is representative of human SKCM, and compared it to that in primary keratinocytes derived from foreskin of human newborns. We observed epigenetic priming in the melanoma cells, such that DHS enrichment was observed in SK-MEL-5 cells at the S100B TSS and upstream promoter, which spans 578 bp (chr21:46,604,961-46,605,537) (Figure 1F). The conformational change in chromatin state may be best linked to S100B upregulation in melanoma. This is supported by RNA-seq analysis of the same set of samples, which revealed no S100B transcript in primary keratinocytes but marked S100B transcript from exon (E)2 and E3 in SK-MEL-5. Next, we evaluated amino acid sequence homology between human S100B and its mouse ortholog S100b. The analysis showed that the proteins are almost identical. Each contains 92 amino acids with a single base substitution (asparagine to glutamic acid) at position 63 (Figure 1G), which is predicted not to significantly alter the protein’s secondary structure (Jalview 2.11.1.0). Mouse B16 melanoma cells have been used for expressing differential levels of S100b under different pathological conditions in conjunction with tumor development [22,23,24]. We used the F1 and metastatic F10 melanoma cell lines to determine whether suppression of the gene can evoke anticancer phenotypes in treated cells. S100b expression (signal intensity relative to endogenous GAPDH) in F10 cells (0.58) was significantly (p < 0.05) lower than in F1 cells (1.72) (Figure 1H).
We aimed to suppress the S100b expression by using CRISPR-KRAB-mediated interference at selected loci of the gene. S100b-specific m-sgRNAs were selected from the previously reported CRISPR inhibition library specific for the mouse genome or were designed with an online algorithm tool [20]. Ref. [25] m-SgRNA-1 and m-SgRNA-2 were targeted at 10 bp and 55 bp, respectively, downstream from the S100b TSS (chr10:76253899-76253915); m-SgRNA-3 was targeted at 49 bp upstream of the TSS (chr10:76253817-76253822) (Figure 2A). The targeted S100b regions also contain putative binding sites (predicted by the TRANSFAC tool) for multiple TFs. In particular, we observed the prevalence of c-Fos, C/EBP-β, and NF-1 TFs within 20 bp of the target sites of the 3 m-sgRNAs, suggesting their putative regulatory roles in determining upregulation of S100b (Figure 2B) [26,27,28]. The DK repressor protein was stably expressed in F1 and F10 cells via lentiviral transduction and was maintained under puromycin selection (Figure 2C). Additionally, the m-sgRNA vectors harboring an mCherry fluorophore allowed the use of FACS sorting to enrich (>95%) co-transduced cells (Figure 2D). The cells were transduced with DK module with or without three individual S100B-specific sgRNAs. Cells transduced with DK in combination with sgRNAEGFP served as an off-target control for the current study. The inherent transcription-repressive nature of the KRAB protein resulted in reduced S100b mRNA levels in both F1 and F10 cell lines, whereas combinatorial treatment with DK plus selective m-sgRNAs resulted in different degrees of reduction in the gene expression, compared to the parental lines (except for m-sgRNA-3 in F10). Therefore, to control for the effects of KRAB, we used DK-transduced cells as the baseline to which we compared the cells that were also transduced with the m-sgRNAs, for determining their effects on S100b expression. S100b expression was significantly (p < 0.05) reduced in the B16-F1 cells treated with the combination of DK plus m-sg-RNA-1, compared to the cells treated with DK alone (Figure 2E). In contrast, S100b expression was increased in F1 cells when treated with the combinations of DK plus remaining S100b-specific sgRNAs (m-sgRNA-2/3) or sgRNAEGFP (Figure 2E). A similar pattern was observed in the change in S100b mRNA expression in the B16-F10 cells, such that the gene expression was further reduced (p > 0.05) in cells treated with the combination of DK and m-sg-RNA-1, compared to cells transduced with DK alone. The combination of DK with m-SgRNA-2/3 or with sgRNAEGFP in either type of B16 cells remained ineffective in suppressing S100b expression. Additionally, S100b expression remained unaltered or insignificantly changed in both cell lines when transduced with vectors containing individual m-sgRNAs specific to S100b or EGFP without the KRAB module, compared to non-transduced parental lines (Figure S1). We observed a significant decrease in S100b mRNA expression in both the B16 cells, transduced with DK fusion alone, whereas the protein expression completely disappeared in cells transduced with the combination of DK plus m-SgRNA-1. Therefore, the S100b protein-expression changes in the similar direction to the changes in mRNA level in both F1 and F10 cells. Western blot results also corroborated the fact that S100b is much less expressed, both at the level of mRNA (Figure 1H) and protein (Figure 2G) in the F10 cells compared to the F1 cells. However, the S100b protein-expression band was totally resolved in the F10 cells treated in combination with DK plus m-SgRNA-1. Next, to examine the specificity of the CRISPR-KRAB toolbox developed here, we looked into the possible changes in expression of PRMT2 and DIP2A, two adjacent genes to S100b on Chr.10 (mm10) (Figure 3A). We did not observe any significant (p > 0.05) reduction in the expression of PRMT2 and DIP2A, indicating that the action of CRISPR-KRAB interference was target-specific (Figure 3A). In summary, the combination of DK plus m-SgRNA-1 imparted the greatest repressive effects on S100b expression, and CRISPR-KRAB interference was more profound in F1 cells than in F10 cells. Therefore, we continued with this combination for the downstream biological analyses.
Based on the previous literature [2,10], we hypothesized that DK-mediated suppression of S100b might increase the availability of intracellular free WT-p53 and reactivate the tumor suppressor function of TP53 (Figure 3B). DK plus m-SgRNA-1 resulted in no observed increases in mRNA levels of TP53 in F1 or F10 cells (Figure S2). However, in both cell types, co-transduction with DK plus m-SgRNA-1 resulted in significantly increased (p < 0.05) p53 protein levels (Figure 3C,D), while transduction with only DK resulted in no marked changes in p53 protein levels in both cell types. As previously mentioned, targeted inhibition of S100b may release intracellular S100b-bound p53, potentially increasing levels of WT-p53 and its downstream product p21 or apoptogenic proteins [10,29,30]. Therefore, to investigate the consequences of S100b suppression, we assessed the effects of DK on induction of apoptosis in the target cells. In both F1 and F10 cells, transduction with DK plus m-SgRNA-1 resulted in significantly increased (p < 0.05) expression of CDKN1A (p21) in F1 cells both at the level of mRNA (14% increase) (Figure S2) and protein (Figure 3C), compared to F1-control. In contrast, a marginal (p = 0.05) increase in CDKN1A (11%) expression was observed in F10 cells (Figure S2). Nonetheless, the p21 protein level was significantly increased (p < 0.05) in the F10 line, compared to the control (Figure 3D). To test the effects of elevated p53 and p21 levels on apoptosis, we treated the control (F1/F10) and DK-m-SgRNA-1 transduced cells with staurosporine (STS), a generic inducer of apoptosis. S100b-suppressed F1 cells, compared to the parental line, had increased cell death and susceptibility to serial concentrations (0.1 nM, 0.05 nM, 0.025 nM, 0.01 nM) of STS. An untreated control and exceedingly high dose (10 µM) of STS were used as negative and positive controls for cell viability assays. For instance, at 0.1 µM STS, we observed 11% less cell viability in F1 cells transduced with DK plus m-SgRNA-1 (26.1%) than in the parental cells (37.5%) (Figure 4A); however, we did not observe similar significant changes in cell death in studies with the F10 cells (data not shown). Next, to further investigate the observed cell death in the S100b-suppressed cells, we examined expression of key caspase and non-caspase proteases that are involved in apoptotic signaling pathways. We started by determining expression of mitochondrial apoptosis-inducing factor (AIFM1) [31,32], which was significantly increased (p < 0.05) in both mRNA and protein expression in the DK+m-SgRNA-1-transduced F1 cells, compared to non-transduced and DK+/-sgRNAEGFP control cells (Figure 4B,C). The impact of differential upregulation of AIFM1 then was evaluated in conjunction with caspase-3 activation. We did not observe any significant changes in the CASP3 mRNA expression in DK+m-SgRNA-1 transduced cells (Figure 4B). In contrast, a capase-3 protein cleavage (~17 kDa) was evident in cells transduced with DK+m-SgRNA-1 (Figure 4C); however, we did not observe noticeable changes in caspase-8 mRNA or protein levels (data not shown). Finally, we examined whether DK-m-SgRNA-1 affects expression of PARP-1, the substrate of caspase-3 [33]. We detected a significant (p > 0.05) increase in PARP-1 mRNA level (Figure 4B), while the cleaved subunit (~95 kDa) of PARP-1 protein was observed in DK+m-sgRNA-1 transduced cells but not in control cells (Figure 4C). In contrast, F10 cells transduced with DK+/-m-sgRNA-1 or sgRNAEGFP failed to evoke any apoptogenic changes, compared to the control cells. In F10 cells, S100b suppression also did not result in altered protein levels of caspase-3/caspase-8 or in cleaved PARP-1 (Figure S3), indicating that an apoptotic response was not evoked. In summary, DK suppression of S100b efficiently evoked p53-mediated mitochondrial apoptotic machinery in F1 melanoma cells (Figure 4C) but not in F10 cells, despite elevation of WT-p53 and p21 proteins in these cells. This suggests a mechanistic difference in S100b regulation between the B16 lines, which needs further investigation.
We observed reduced cell viability in both the F1 and F10 cells co-transduced with DK in combination with m-SgRNA-1. We determined the percentage of viability by determining the amount of ATP produced by live cells 24 h and 48 h after initial seeding of cells in a 96-well plate (1 × 104 cells per well). Compared to the control cells, co-transduced F1 cells had a 65% reduction and F10 cells had a 40% reduction (p < 0.05) in cell viability (Figure 5A,B) after 24 h. In contrast, after 48 h, F10 cells co-transduced with DK plus m-SgRNA-1 had only 13% less viability (p = 0.14) than control cells, but co-transduced F1 cells continued to have at least 32% less viability (p < 0.05) than control cells. Next, we treated S100b-suppressed F1 and F10 cells with chemotherapeutic agents against metastatic melanoma: cisplatin, a platinum analog, and tunicamycin, an inducer of endoplasmic reticulum stress [34,35]. We determined the IC50 after incubating cells with each drug for 24 h; as control comparison groups, we used F1 and F10 cells that were not CRISPR-transformed. For controlling melanoma cell growth, tunicamycin was found to be more efficacious than cisplatin. The IC50 against cisplatin was achieved at 77.5 nM (R2 = 0.83) in the S100b-suppressed F1 cells, compared to the F1 control cells (IC50 = 851.3 nM; R2 = 0.93); the IC50 against cisplatin was achieved at 138.4 nM (R2 = 0.97) in the S100b-suppressed F10 cells, compared to the F10 control cells (IC50 = 611.8 nM; R2 = 0.92) (Figure 5C). The IC50 against tunicamycin, however, was achieved at a concentration as low as 9 nM (R2 = 0.93) in the S100b-suppressed F1 cells, compared to 185.3 nM (R2 = 0.96) in the F1 control cells. In contrast, in the S100b-suppressed F10 cells, the IC50 of tunicamycin was attained at 50 nM (R2 = 0.91), compared to 165 nM (R2 = 0.96) in the F10 control cells (Figure 5D). Overall, our data suggest that S100B suppression cooperatively works with the chemotherapeutic agents that trigger cell death by evoking apoptotic responses, thus potentially sensitizing melanoma cells to these inhibitors.
Overexpression of S100B and its clinicopathological relevance in melanoma has been well studied. The study reported here elucidated the role of genetic and epigenetic modifications underlying oncogenic overexpression of S100B. For instance, we demonstrated that S100B expression is significantly (p < 0.05) higher in patient-samples with gains in CN of the gene, compared to those with losses in CN of the gene (Figure 1C). However, this observation needs further validation and inclusion of more study cohorts, because we did not observe significant alterations in gene expression levels when comparing samples with losses in S100B CN and those diploid for S100B. At the epigenetic level, we investigated alterations in DNA methylation and chromatin state of the gene. We observed a moderate negative correlation between DNA methylation and expression, suggesting that the HM450 probe possibly is targeted to the promoter of the gene, where DNA methylation is inversely proportional to gene expression [36]. At the chromatin level, we observed an event of epigenetic priming at the S100B promoter, such that the TSS at the upstream promoter showed enrichment for DHS in SK-MEL-5 melanoma cells. This seems reliable because, in primary keratinocytes, the same region remains devoid of such DHS marks. The chromatin state alterations of S100B in melanoma cells could be logically ascribed to overexpression of the gene because it does not seem to be expressed in primary keratinocytes at levels comparable to those in melanoma cells. Enhanced molecular interactions between S100B and WT-p53 proteins in melanoma patient samples were demonstrated to impair WT-p53 function for tumor suppression through restricted cell cycle arrest [29], which leads to increased resistance to chemotherapeutics [11,12,37,38,39]. In the present study, we customized inactivation of S100b expression in two murine melanoma cell lines by using a CRISPR/dCas9 system that allowed for restoring intercellular levels of WT-p53 that might otherwise be bound to the S100b protein, thereby salvaging p53-mediated cell death and apoptosis. Our approach could be broadly applicable to most melanoma cases, irrespective of S100B CN gain or loss because the CRISPR/Cas9 system can intrinsically suppress multiple copies of the same gene [40]. We observed varying degrees of DK efficacy among the tested sgRNAs, specific to S100b or specific to EGFP and cell lines, which may be attributed to several factors. For instance, we observed little to no S100b-inhibition in either cell types transduced with a combination of DK plus m-SgRNA-2 or m-SgRNA-3. This could be partly due to the fact that the target region is already masked or occupied with endogenous TFs, leaving a relatively narrow window of DNA sequence available for binding sgRNAs or dCas9. We observed that the combination of m-SgRNA-1 and DK imparted the strongest suppression of S100b expression in both cell lines that were studied. However, combining DK with all three m-sgRNAs did not result in better suppression of S100b (data not shown), which could be partly related to the inefficacy of m-SgRNA-2 and m-SgRNA-3. For the current study, cells transduced with DK plus sgRNAEGFP were used as an off-target control. It was observed that induction of cells with DK alone or DK in combination with S100b-specific sgRNA-2/3 or gRNAEGFP reduced the gene expression, compared to the parental lines (Figure 2E,F), which, however, did not correlate to the changes in downstream apoptotic responses. Therefore, the generic reduction in S100b expression may be ascribed to the lentiviral effect on cells, which was not consistent with the downstream apoptotic signaling. In contrast, reduction of S100b expression was consistent with the expression of apoptosis-responsive proteins. This makes the combination of DK plus m-sgRNA-1 a unique and selective candidate for the intended purpose. S100b suppression significantly (p < 0.05) increased levels of p53 and its downstream p21 protein. However, CRISPR interference was not evident at the mRNA level for TP53 in either cell line (Figure S2). This could be explained by the fact that S100B does not act upstream of the TP53 signaling cascade and, therefore, may not have significant effects on expression of the gene. The disconnect between p53 mRNA and protein levels also aligned with results of a previous study in which siRNA-based perturbation of S100b did not alter TP53 mRNA levels but significantly increased levels of total p53 and of phosphorylated p53 [10]. Overall, our findings support the hypothesis that suppressed S100b possibly lost its affinity for intracellular p53, thereby facilitating elevated levels of WT-p53 protein. We also observed that the degree of S100b inhibition correlated with p53-mediated activation and apoptotic changes, such that S100b inactivation and apoptotic changes in the F1 cells were profound, compared to those in the F10 cells. The decreased efficacy of DK against the F10 cells may be due to intrinsically lower expression of S100b in F10, relative to F1 cells, or due to ineffective binding of m-SgRNA-1 to the target region, presumably due to occupancy by other endogenous TFs. Disparities in the degree of functional outcomes between the two cell lines also could be explained by findings from previous studies reporting differential cytogenetic properties [41]. In this case, optimization of the KRAB suppressor system by integrating additional repressors, such as MeCP2 or EZH2, may impart superior gene inactivation effects against the F10 cells [42,43]. Finally, S100b inhibition decreased cell viability (Figure 5A,B) and increased susceptibility (Figure 5C,D) of melanoma cells to the chemotherapeutics. A previous study found cisplatin to be moderately effective against melanoma patients [35]. Similar findings were observed for the tunicamycin treatment. Similar to the S100b inhibition and apoptotic responses, chemotherapeutic susceptibility was also achieved at relatively lower concentrations in F1, compared to the F10 cells. A rational follow-up of the present study is aimed at determining the efficacy and persistence of the DK/CRISPR toolbox in other human and murine melanoma lines having high levels of S100b. Another limitation of the current study is that the S100b-specific DK/CRISPR tool needs further functional validations in vivo. Accordingly, the toolbox might need further modifications such that we may need to introduce additional suppressor domains, such as MeCP2, to the KRAB domain for better suppression efficacy [43]. Nonetheless, based on the outcomes from the reported system, we can flexibly modify the current toolbox by introducing additional transcriptional repressors or epigenetic modulators for robust suppression of any gene or gene network. The proof of concept generated here strongly encourages studies of different signaling pathways involved in sustaining malignant growth in melanoma cells. | true | true | true |
|
PMC10000388 | Fei-Yuan Yu,Qian Xu,Xiao-Yun Zhao,Hai-Ying Mo,Qiu-Hua Zhong,Li Luo,Andy T. Y. Lau,Yan-Ming Xu | The Atypical MAP Kinase MAPK15 Is Required for Lung Adenocarcinoma Metastasis via Its Interaction with NF-κB p50 Subunit and Transcriptional Regulation of Prostaglandin E2 Receptor EP3 Subtype | 22-02-2023 | MAPK15,EP3,p50,LUAD,metastasis | Simple Summary Due to the lack of effective early diagnostic markers for lung cancer and the rich blood circulation in the lungs, it is very easy to cause lymph node metastasis and distant metastasis of lung cancer, making lung cancer as one of the top ten cancer types with the highest mortality rate in the world. This study found that MAPK15 is highly expressed in the tissues of patients with lung adenocarcinoma lymph node metastasis, and MAPK15 interacts with p50 to regulate the expression of EP3 at the transcriptional level, thereby promoting cancer cell migration. This suggests that MAPK15 plays a key role in the metastasis of lung cancer cells, and MAPK15 can be used as a molecular marker for the early diagnosis or prognosis assessment of lung cancer. Its molecular mechanism for regulating lung cancer metastasis can provide valuable information and insights on novel therapeutic options at molecular levels. Abstract Studying the relatively underexplored atypical MAP Kinase MAPK15 on cancer progression/patient outcomes and its potential transcriptional regulation of downstream genes would be highly valuable for the diagnosis, prognosis, and potential oncotherapy of malignant tumors such as lung adenocarcinoma (LUAD). Here, the expression of MAPK15 in LUAD was detected by immunohistochemistry and its correlation with clinical parameters such as lymph node metastasis and clinical stage was analyzed. The correlation between the prostaglandin E2 receptor EP3 subtype (EP3) and MAPK15 expression in LUAD tissues was examined, and the transcriptional regulation of EP3 and cell migration by MAPK15 in LUAD cell lines were studied using the luciferase reporter assay, immunoblot analysis, qRT-PCR, and transwell assay. We found that MAPK15 is highly expressed in LUAD with lymph node metastasis. In addition, EP3 is positively correlated with the expression of MAPK15 in LUAD tissues, and we confirmed that MAPK15 transcriptionally regulates the expression of EP3. Upon the knockdown of MAPK15, the expression of EP3 was down-regulated and the cell migration ability was decreased in vitro; similarly, the mesenteric metastasis ability of the MAPK15 knockdown cells was inhibited in in vivo animal experiments. Mechanistically, we demonstrate for the first time that MAPK15 interacts with NF-κB p50 and enters the nucleus, and NF-κB p50 binds to the EP3 promoter and transcriptionally regulates the expression of EP3. Taken together, we show that a novel atypical MAPK and NF-κB subunit interaction promotes LUAD cell migration through transcriptional regulation of EP3, and higher MAPK15 level is associated with lymph node metastasis in patients with LUAD. | The Atypical MAP Kinase MAPK15 Is Required for Lung Adenocarcinoma Metastasis via Its Interaction with NF-κB p50 Subunit and Transcriptional Regulation of Prostaglandin E2 Receptor EP3 Subtype
Due to the lack of effective early diagnostic markers for lung cancer and the rich blood circulation in the lungs, it is very easy to cause lymph node metastasis and distant metastasis of lung cancer, making lung cancer as one of the top ten cancer types with the highest mortality rate in the world. This study found that MAPK15 is highly expressed in the tissues of patients with lung adenocarcinoma lymph node metastasis, and MAPK15 interacts with p50 to regulate the expression of EP3 at the transcriptional level, thereby promoting cancer cell migration. This suggests that MAPK15 plays a key role in the metastasis of lung cancer cells, and MAPK15 can be used as a molecular marker for the early diagnosis or prognosis assessment of lung cancer. Its molecular mechanism for regulating lung cancer metastasis can provide valuable information and insights on novel therapeutic options at molecular levels.
Studying the relatively underexplored atypical MAP Kinase MAPK15 on cancer progression/patient outcomes and its potential transcriptional regulation of downstream genes would be highly valuable for the diagnosis, prognosis, and potential oncotherapy of malignant tumors such as lung adenocarcinoma (LUAD). Here, the expression of MAPK15 in LUAD was detected by immunohistochemistry and its correlation with clinical parameters such as lymph node metastasis and clinical stage was analyzed. The correlation between the prostaglandin E2 receptor EP3 subtype (EP3) and MAPK15 expression in LUAD tissues was examined, and the transcriptional regulation of EP3 and cell migration by MAPK15 in LUAD cell lines were studied using the luciferase reporter assay, immunoblot analysis, qRT-PCR, and transwell assay. We found that MAPK15 is highly expressed in LUAD with lymph node metastasis. In addition, EP3 is positively correlated with the expression of MAPK15 in LUAD tissues, and we confirmed that MAPK15 transcriptionally regulates the expression of EP3. Upon the knockdown of MAPK15, the expression of EP3 was down-regulated and the cell migration ability was decreased in vitro; similarly, the mesenteric metastasis ability of the MAPK15 knockdown cells was inhibited in in vivo animal experiments. Mechanistically, we demonstrate for the first time that MAPK15 interacts with NF-κB p50 and enters the nucleus, and NF-κB p50 binds to the EP3 promoter and transcriptionally regulates the expression of EP3. Taken together, we show that a novel atypical MAPK and NF-κB subunit interaction promotes LUAD cell migration through transcriptional regulation of EP3, and higher MAPK15 level is associated with lymph node metastasis in patients with LUAD.
The incidence of lung cancer is high among malignant tumors, which seriously affects human health. The mortality rate of lung cancer patients is high because lung cancer is usually in an advanced stage when diagnosed, with lymph node metastasis or even distant metastasis. Radiotherapy and chemotherapy have very limited therapeutic effects on advanced lung cancer. Targeted therapies, such as the use of targeted drugs EGFR tyrosine kinase inhibitors [1,2], can improve the survival of lung cancer patients to a certain extent, but they still face the problem of chemotherapy resistance, recurrence of targeted therapy, etc. Therefore, it is still not possible to effectively control the malignant development of lung cancer [2]. The study of molecular markers related to lung cancer metastasis and their corresponding molecular mechanisms still needs to be further explored. The classical mitogen-activated protein kinases (MAPKs, e.g., ERK1/2, p38, and JNK/SAPK) play important roles in regulating gene expression, cell growth, proliferation, etc. Atypical MAPKs such as ERK3, ERK4, and NLK (nemo-like kinase) also play critical roles in many cellular responses [3,4]. MAPK15, alias extracellular signal-regulated kinase 7/8 (ERK7/8), is the most recently discovered atypical MAPK. Current research indicates that MAPK15 can promote the transformation of colon cancer by mediating the activation of the transcription factor c-Jun [5,6] or promoting the growth of gastric cancer cells [7]. MAPK15 has also been found to interact with autophagy-related proteins such as GABARAP and LC3 to control tumor development [8]. In addition, MAPK15 can be activated by carcinogenic factors such as RET/PTC3 [9] or involved in the regulation of telomerase activity [10] to participate in the development of tumors. Recently, our group has reported that MAPK15 can promote arsenic trioxide-induced apoptosis, as well as boosting the efficacy of combination therapy with cisplatin and TNF-α, in lung cancer cells [11,12]. At present, research about the function of MAPK15 is still limited, and its role in lung cancer metastasis remains unclear. EP3 is one of the four G protein-coupled receptors of prostaglandin E2 (PGE2), which plays an important role in cell proliferation, differentiation, apoptosis, cardiovascular system regulation, and inflammation. It has been reported that tumor angiogenesis and tumor cell growth were significantly inhibited in a mouse lung cancer model with EP3 knocked out [13]. Yamaki et al. found that PGE2 promotes the growth of lung adenocarcinoma (LUAD) cell line A549 via the EP3 receptor-activated Src signaling pathway [14]. However, the molecular mechanism of EP3 in regulating lung cancer progression is still not fully clarified. In this study, we detected the expression of MAPK15 in lung cancer tissues, and found that the expression of MAPK15 is positively correlated with lymph node metastasis in LUAD patients; remarkably, our results showed that the expression of EP3 was transcriptionally regulated by MAPK15, and the expression of EP3 was positively correlated with the expression of MAPK15 in LUAD tissues. Furthermore, we revealed the first time that MAPK15 promotes the expression of EP3 by interacting with p50, thereby enhancing the migration of lung cancer cells.
Lung cancer tissue microarray (BC041115c, US Biomax, Rockville, MD, USA) was purchased and all human tissues were collected according to HIPPA-approved protocols as described by US Biomax (https://www.biomax.us/FAQs, accessed on 14 June 2022). Immunohistochemistry was performed to detect the expression of MAPK15 and EP3. Briefly, tissue microarray was deparaffinized thrice in xylene (10 min for each) and rehydrated in gradient series ethanol (100%, 95%, 90%, 90%, 5 min for each), respectively. After being rinsed with water, tissue slides were incubated with 3% hydrogen peroxide for 40 min to block endogenous peroxidase. Tissue slides were then rinsed with PBS and immersed in 0.01 M citrate acid antigen retrieval solution and heated at 98 °C for 20 min using a water bath. After natural cooling, tissue slides were washed with PBS and incubated with 5% BSA for 30 min. Tissue slides were then incubated with MAPK15 [15] or EP3 (Cat. 101760, Cayman Chemical, Ann Arbor, USA) antibody at 4 °C overnight. After being rinsed with PBS, tissue slides were incubated with secondary antibody for 45 min at RT. Subsequently, tissue slides were washed with PBS and reacted with 3,3′-diaminobenzidine (DAB, Zhongshan Golden Bridge Inc. Beijing, China) and counterstained with hematoxylin. Then, tissue slides were mounted with glycerogelatin and photographed with a light microscope. Immunostaining of tissue microarray were scored according to immunoreactive score (IRS) [16,17]. Each tissue in the microarray was semiquantitatively scored for intensity (0, absent; 1, weak; 2, moderate; 3, strong) and extent of staining (percentage of the positive tumor cells: 0, ≤5%; 1, 6–25%; 2, 26–50%; 3, 51–75%; 4, >75%). Intensity and extent of each tissue were multiplied to give a composite score: 0–3, deemed as low expression, “−”); 4–12, deemed as high expression (4–6, “+”; 7–9, “++”; 10–12, “+++”).
All cells were grown at 37 °C in a 5% CO2 incubator. HEK293T, H1299, and A549 cells were purchased from ATCC Cell Bank of the Chinese Academy of Sciences (Shanghai, China) and maintained in MEM, RPMI-1640, or F12-K medium supplemented with 10% FBS and 1% PS, respectively. MAPK15 stable knockdown LUAD H1299 cells (H1299-shMAPK15) and control cells (H1299-shCtrl) were established previously [12]. For transfection, cells were mixed with siRNA/plasmids-polyethylenimine mixture and cultured for the indicated time point. Negative control siRNA (siN05815122147) and siRNA duplexes against EP3 were purchased from IGE Biotechnology LTD (Guangzhou, China) and listed in Supplementary Table S1.
RNA was extracted using RNAiso Plus (Takara, Dalian, China) from cells. Then, cDNA was synthesized using GoScript™ Reverse Transcription Mix (Promega, Madison, WI, USA) by following the manufacturer’s instructions. Specific primers were used, and real-time PCR was performed using GoTaq qPCR Master Mix (Promega, Madison, WI, USA) on Applied Biosystems 7500 Real-Time PCR System. The 2ΔΔCT method was used to calculate the relative expression of target genes compared to internal control (β-Actin) as described previously [18]. Primers were synthesized by IGE Biotechnology LTD (Guangzhou, China) and listed in supplementary Table S1.
Equivalent amounts of extracted protein were resolved by 10% SDS-PAGE and transferred onto polyvinylidene fluoride membranes. The membranes were blocked with 5% nonfat milk in PBS containing 0.05% Tween 20 followed by incubation with primary antibody overnight at 4 °C. After reacting with primary antibody, membranes were incubated with secondary antibody and proteins were visualized with ECL reagent using Tanon 5200 system (Tanon, Shanghai, China). The optical density of each protein band was quantified by Gel-Pro Analyzer 4 (Toyobo, Osaka, Japan) software. Original blots and blot quantification are shown in Figures S3–S5, S7 and S8.
Transwell assay was performed as described previously [16]. Briefly, 3.0 × 104 cells were seeded in the upper compartment of transwell inserts with 8 μm microporous membrane (cat no. 3422, Corning Inc., Corning, NY, USA). After being incubated for 24 h, unmigrated cells on the upper surface of the microporous membrane were wiped using a cotton swab. Cells on the lower surface of the microporous membrane were fixed with 4% PFA for 20 min and subsequently stained with 0.1% crystal violet for 15 min. The transwell chamber was rinsed with PBS to remove excess crystal violet, and images of migrated cells were captured using an Axiovert 40 CFL microscope (Carl Zeiss AG, Oberkochen, Germany) with CCD camera (magnified 100×). Finally, the crystal violet in the migrated cells was dissolved with 33% acetic acid, and absorbance was measured at OD595.
Cells seeded on coverslip in 6-well plate were incubated for the indicated time point and fixed with 4% PFA for 15 min. After being rinsed with PBS, cells were permeabilized for 10 min with PBS containing 0.25% Triton X-100. Subsequently, cells were incubated with 5% BSA for 30 min to block unspecific binding of antibodies. Then, cells were incubated with primary antibody in a humidified chamber at 4 °C overnight. After decanting of primary antibody solution, cells were washed with PBS and incubated with secondary antibody for 1.5 h at room temperature in the dark. Coverslips were counterstained with 1 μg/mL Hoechst 33342 and mounted with mounting medium. Images were captured with Axiovert 40 CFL Microscope (Carl Zeiss AG, Germany) or Zeiss lsm 800 confocal microscope (Carl Zeiss AG, Germany).
In vivo peritoneal metastasis assay was performed as described previously [19]. Briefly, 5 × 106 MAPK15 stable knockdown H1299 or control cells in 200 μL of phosphate-buffered saline were injected intraperitoneally into BALB/c nude mice (Beijing Vital River Animal Technology Co., Ltd., Beijing, China, licensed by Charles River). After 7 weeks, the mice were sacrificed, and tumor nodules were quantified.
HEK293T cells cultured in 10 cm dish were transfected with pcDNA4/Xpress-MAPK15 plasmids [15] and incubated for 24 h. Prior to immunoprecipitation, 1 μg of Xpress antibody or normal IgG was pre-adsorbed with 20 μL Protein A/G Sepharose slurry for 2 h at 4 °C with rotation. After transfection, cells were harvested and lysed with NP-40 lysis buffer using repetitive freeze-thawing method. An amount of 300 μg of lysates to be used for immunoprecipitation was precleared with 20 μL Protein A/G Sepharose at 4 °C for 1 h with rotation. The supernatant was then incubated with the Xpress antibody–Protein A/G Sepharose complexes overnight at 4 °C with rotation (anti-mouse IgG was used as negative control). In total, 10% of the supernatant was used as input. The Sepharose beads were collected by centrifugation and washed extensively in 500 μL of lysis buffer, and eluted in 20 μL of SDS sample buffer by heating to 98 °C for 5 min. After centrifugation at 10,000× g, the supernatant was collected for immunoblot analysis.
Chromatin immunoprecipitation assay was performed using SimpleChIP® Enzymatic Chromatin IP Kit (Cell Signaling Technology, Danvers, MA, USA). Briefly, formaldehyde cross-linked H1299 cells were lysed, and chromatin was digested with micrococcal nuclease into DNA/protein fragments. Then, p50 antibody (Santa Cruz Biotechnology, Dallas, TX, USA) was added and the complex is captured by protein G magnetic beads. Seven p50 binding sites (site1–site7) in the EP3 promoter region (−2000 bp) were predicted by JASPAR databases and PCR was used to detect p50 binding.
Five repeats of p50 binding sequence (site5, sequence: GGGGCTTCCC) and 12 bp linker sequences with AflII and NsiI sites were synthesized by IGE Biotechnology LTD (Guangzhou, China) and ligated to a modified pJC6-GL3 plasmid [11] to construct luciferase reporter plasmid (5 × p50-Luc). Then, the 5 × p50-Luc plasmid was co-transfected with/without pCMV-p50 plasmid into equal amount of H1299 cells in 12-well plate. Afterward, cells were lysed for luciferase assay following manufacturer’s instructions (dual-luciferase reporter assay system, Promega, Madison, WI, USA).
Mean comparisons were performed using the GraphPad Prism 8 for unpaired t-test. Fisher’s exact test was used to study the correlation between MAPK15 expression and clinical parameters. Spearman rank correlation analysis was used to compare the correlation between the expression of MAPK15 and EP3 in lung cancer tissues using SPSS 19 software. The above statistical analysis was two-tailed; p < 0.05 suggested that the difference was statistically significant.
To study the role of MAPK15 in lung cancer, we analyzed the relationship between MAPK15 and clinical–pathological parameters such as age, gender, depth of tumor invasion, lymph node metastasis, distant metastasis, tumor differentiation, clinical stage, etc. We found that there was a positive correlation between MAPK15 expression and lymph node metastasis (p = 0.012) as well as clinical stage (p = 0.033) (Supplementary Table S2). The expression of MAPK15 is higher in patients with lymph node metastasis (N1 + N2) as compared to patients without lymph node metastasis (N0) (Supplementary Table S2). Other clinical–pathological parameters such as age, gender, depth of tumor invasion, distant metastasis, and tumor differentiation were not significantly correlated with the expression of MAPK15 (Supplementary Table S2). Adenocarcinoma and squamous cell carcinoma are major types of non-small-cell lung cancer (NSCLC). As compared to squamous cell carcinoma, we revealed that the expression of MAPK15 is relatively higher in adenocarcinoma (Figure 1A, Supplementary Table S3) and is associated with lymph node metastasis (p = 0.013) (Table 1, Figure 1B).
The above results indicate that MAPK15 is expressed more highly in lymphatic metastatic LUAD. In MAPK15 stable knockdown LUAD H1299 cells (Figure 1C,D), cell migration was significantly inhibited (Figure 1E,F). The expression of Snail1 was decreased in MAPK15 knockdown cells (Figure 1G), which can down-regulate the expression of E-cadherin by post-translational modifications such as deacetylation and methylation during EMT [20]. Consequently, the expression of epithelial marker E-cadherin was increased, while mesenchymal marker integrin β1 is decreased after MAPK15 knockdown (Figure 1G). Then, we performed an in vivo peritoneal metastasis assay using H1299-shMAPK15 cells and found that loss of MAPK15 significantly reduces metastasis to mesentery in vivo (Figure 1H,I). The above results indicate that H1299 cells undergo mesenchymal–epithelial transition after MAPK15 knockdown, thereby decreasing migration and metastasis.
It has been reported that the expression of MMP2 was depressed in EP3 knock-out mice under hypoxic stress [21], which indicates a correlation between the expression of MMP2 and EP3. Our results showed that MMP2 was down-regulated in MAPK15 knockdown H1299 cells (Figure S1). To investigate whether EP3 is involved, we detect the expression of EP3 in H1299-shCtrl and H1299-shMAPK15 cells. We found that the mRNA and protein level of EP3 was significantly decreased in MAPK15-deficient cells (Figure 2A,B) and the protein level of EP3 was not affected by proteasome inhibitor MG132 (Figure 2B), suggesting that the decreased EP3 in H1299-shMAPK15 cells was transcriptionally regulated. Moreover, the migration of H1299 cells was inhibited (Figure 2F,G) after EP3 was knocked down (Figure 2C–E). The decreased EP3 in MAPK15 knockdown cells suggested that there might be a correlation between the expression patterns of these two molecules. Then, we detected the expression of EP3 in a serial section from the same tissue that we stained with MAPK15 antibody and found that the expression of EP3 is positively correlated with MAPK15 (r = 0.589, p < 0.001, Figure 2H and Table 2). Taken together, the above results show that MAPK15 affects cell migration through the regulation of EP3.
The molecular mechanism of how EP3 is transcriptionally regulated by MAPK15 is unknown. It has been reported that the expression of MAPK15, NF-κB1 (p50), and NF-κB2 (p52) were obviously decreased in ovarian cancer cell lines [22], which indicate there are correlations between MAPK15 and the NF-κB family. To investigate the relationship between MAPK15 and NF-κB family members, we transfected the pcDNA4/Xpress-MAPK15 plasmid into 293T cells and the immunoprecipitation assay revealed that MAPK15 interacts with p50 but not p65 and c-rel (Figure 3A). We also detected the localization of MAPK15 and p50 in H1299 cells by confocal microscopy and found that MAPK15 is distributed both in the cytoplasm and nucleus, and colocalizes with p50 (Figure 3B), indicating that there is an interaction between these two proteins in LUAD cells which might contribute to the expression of EP3. To study the relationship between MAPK15/p50 and EP3, we overexpressed MAPK15 (Figure 3C) and p50 (Figure 3D) in H1299 cells and found that the expression of EP3 was increased (Figure 3E), which indicates that MAPK15 and p50 positively regulate the transcription of EP3. The chromatin immunoprecipitation assay found that p50 binds to two p50 binding motifs in the EP3 promoter (Figure 3F, site2 and site5). Subsequently, we chose site 5 (Figure 3F) to construct the luciferase reporter plasmid and co-transfect with/without pCMV-p50 in H1299 cells for the luciferase reporter assay. Our results indicate that the luciferase activity is significantly increased in cells overexpressed with p50 (Figure 3G), which revealed that p50 can transcriptionally regulate EP3 by binding to the EP3 promoter.
MAPK15 interacts with p50 intracellularly, indicating potential gene regulation and cellular phenotypic change. Beinke et al. reported that the p105 pathway can positively regulate gene transcription under TNF-α stimulation [23]. We hypothesize that TNF-α might promote the expression of EP3 through the p50 pathway, thereby contributing to cell migration. In TNF-α-treated H1299 cells, we found that TNF-α promoted EP3 expression in a dose- (Figure 4A) and time-dependent manner (Figure 4B). Furthermore, TNF-α promoted the migration of H1299 cells but had no significant effect on the migration of MAPK15 knockdown cells (Figure 4C,D). This result suggests that TNF-α promotes cell migration through MAPK15. In TNF-α-treated A549 cells, we found that TNF-α promotes nuclear localization of MAPK15 and p50 (Figure S2). In H1299 cells, we found that p50 is distributed in both cytoplasm and nucleus, whereas in MAPK15 knockdown cells, p50 is mainly located in the cytoplasm (Figure 4E), indicating that nuclear localization of p50 is dependent on MAPK15. At the same time, we treated H1299 cells with TNF-α and found that p50 is mainly located in the nucleus, whereas in MAPK15 knockdown cells, p50 is distributed in both the cytoplasm (white arrows) and the nucleus (Figure 4E). The above results indicate that TNF-α-induced nuclear translocation of p50 is dependent on MAPK15. In addition, we found that the expression of EP3 in TNF-α-treated H1299 cells was increased, while the expression changes of EP3 in MAPK15 knockdown H1299 cells were not significant (Figure 4F), and TNF-α could not promote H1299 cell migration while EP3 was knocked down (Figure 4G,H). Taken together, these results reveal that TNF-α promotes H1299 cell migration through induction of MAPK15-p50 nuclear localization and EP3 expression in cells with MAPK15 expression.
JSH-23 is an NF-κB inhibitor. When using JSH-23 to treat H1299 cells, we found that JSH-23 inhibited the expression of EP3 in a dose- (Figure 5A) and time-dependent manner (Figure 5B). Furthermore, JSH-23 inhibited the migration of H1299 cells but had no significant effect on the migration of knockdown MAPK15 cells (Figure 5C,D). This result suggests that JSH-23 inhibits cell migration through MAPK15. In addition, we found that the expression of EP3 in H1299 cells treated with JSH-23 was decreased, while the expression of EP3 in MAPK15 knockdown H1299 cells did not change significantly (Figure 5E), and JSH-23 could not inhibit H1299 cell migration when EP3 was knocked down (Figure 5F,G). The above results indicate that JSH-23 inhibits cell migration by inhibiting MAPK15-induced EP3 expression.
Lung cancer is usually at an advanced stage with lymph node or distant metastasis when diagnosed, which leads to high mortality. Medical knowledge still lacks effective diagnostic molecular markers for metastatic lung cancer. In the present study, we revealed that MAPK15 is more highly expressed in the tissues of LUAD patients with lymph node metastasis (Figure 1B), and MAPK15 interacts with p50 to promote EP3 expression at the transcriptional level (Figure 6), thereby enhancing cancer cell migration and metastasis. MAPK15 is a member of the ERK subfamily, which is involved in the regulation of cell growth and differentiation like other well-known ERKs. Previous research indicates that MAPK15 is involved in the transformation of colon cancer [6], promotes gastric cancer cell proliferation [7], and is associated with autophagy [8]. However, its clinical pathological role has, until now, not been examined in lung cancer. The correlation between MAPK15 and lymph node metastasis in LUAD described here suggests that MAPK15 plays an important role in lung cancer development, which may lead to poor clinical outcomes. Since we used a commercialized lung cancer tissue array in this study, there is a lack of relevant information on disease progression, so it is impossible to conduct a longitudinal assessment of the relationship between MAPK15 expression and patients’ disease-free survival/overall survival, recurrence, metastasis, etc. However, with the in-depth study of MAPK15, we gradually realized its important role in LUAD. In future studies, multicenter, larger-sample-size studies should be conducted through longitudinal assessment of the patients’ critical long-term clinical outcome to further clarify MAPK15 expression and the significance of clinical parameters. Due to the significant correlation between MAPK15 and the clinical features of the LUAD patients we observed, MAPK15 and its signaling pathway in LUAD may be a potential therapeutic target for metastatic LUAD. As a kinase, MAPK15 carries out different functions in various cancers, indicating the deregulation of key pathways. Studies have indicated a pivotal role of MAPK15 in mediating the effect of gene transcription. We have previously shown that MAPK15 promotes the transformation of colon cancer by mediating the activation of c-Jun [6]. Here, the identification of MAPK15 as an upstream regulator for EP3 unveiled a previously unknown mechanism for the MAPK15 or EP3 signaling pathway and their roles in the regulation of cell migration in LUAD. The role of EP3 in tumor progression is still controversial. It has been reported that EP3 coupled with G proteins can effectively inhibit tumor growth. Shoji et al. found that EP3 can significantly inhibit the proliferation of tumor cells in advanced-stage colon cancer [24]. Sanchez et al. found that EP3 can promote the expression of p21 by reducing cAMP, thereby arresting the cell cycle in the S phase, and ultimately inhibiting the proliferation of 3T6 fibroblasts [25]. On the other hand, there are more and more studies showing that EP3 can promote the development of tumors. Finetti et al. found that EP3 is involved in regulating the formation of tumor blood vessels [26]. Amano et al. found that in an EP3-deficient mouse tumor model, tumor angiogenesis and tumor cell growth were effectively inhibited [13]. Yamaki et al. found that EP3 participates in the Src signaling pathway to promote the growth of LUAD A549 cells [14]. In this study, we reveal that knocking down EP3 can inhibit the migration of LUAD cells and that the expression of EP3 was positively regulated by MAPK15, which expands our understanding of EP3 and its regulation in lung cancer. NF-κB is a type of transcription factor that plays an important role in the occurrence and development of tumors. The ERK family was linked to the NF-κB pathway [27,28]. As the most recently discovered MAPK family member, the relation between MAPK15 and NF-κB is mainly uncharacterized. Previous studies on the NF-κB protein family mainly focused on the activity of IκB or p65 in the p50/p65 complex to promote gene transcription. However, more and more studies have shown that p50 can bind to the promoter of the gene and activate gene transcription. The study of Hong et al. showed that overexpression of p50 in BAR-T cells significantly enhanced the activity of the DNMT1 gene promoter [29]. Karst et al. showed that overexpression of NF-κB p50 in melanoma cells MMRU can promote angiogenesis and up-regulate IL6 expression. They confirmed by Chip assay that p50 can bind to the promoter region of IL6 gene and activate its transcription [30]. Similarly, Southern et al. found that the BAG-1 protein can interact with the p50-p50 homodimer and bind to the promoter region of downstream genes to play a positive role in regulating gene transcription [31]. Beinke et al. reviewed that TNF-α/IL-1/LPS can activate the classic p50/p65 dimer NF-κB signaling pathway and the p100/RelB non-canonical signaling pathway, as well as the p105/p50 signaling pathway [23]. In this study, we found that MAPK15 interacts with p50 in LUAD cells, and the nuclear translocation of p50 may require the assistance of MAPK15. In addition, we also found that the mRNA expression level of EP3 increased when p50 was overexpressed in H1299 cells, indicating that p50 can regulate the expression of EP3 at the transcriptional level, and CHIP assay and luciferase reporter assay confirmed that p50 can bind to the promoter region of EP3 and promote the transcription of EP3. Our results revealed that MAPK15 interacts with p50 to promote the transcription of EP3, thereby affecting biological functions such as the migration of LUAD cells.
In conclusion, this study demonstrates the role of MAPK15 in the metastasis of LUAD. We revealed that MAPK15 promotes LUAD cell migration via p50 and EP3 signaling and is associated with lymph node metastasis in LUAD patients, which indicates that MAPK15 might be a potential prognostic biomarker for LUAD and a therapeutic target to inhibit metastasis in metastatic LUAD patients. The insights provided by this study could facilitate understanding the role of MAPK15 in lung cancer progression and its potential modulatory role in cancer metastasis. | true | true | true |
|
PMC10000786 | Shayan Shafiee,Jaidip Jagtap,Mykhaylo Zayats,Jonathan Epperlein,Anjishnu Banerjee,Aron Geurts,Michael Flister,Sergiy Zhuk,Amit Joshi | Dynamic NIR Fluorescence Imaging and Machine Learning Framework for Stratifying High vs. Low Notch-Dll4 Expressing Host Microenvironment in Triple-Negative Breast Cancer | 25-02-2023 | cancer,tumor microenvironment modifier,notch-DLL4,consomic xenograft model,machine learning,binary classification,dynamic enhanced NIR imaging,indocyanine green,time series,tumor detection | Simple Summary Breast cancer is a disease that is affected by both the tumor cells and the host environment. It is well known that the tumor blood vessels are aberrant in structure and function due to rapid angiogenesis, and this aberrant vasculature plays a major role in drug delivery and therapy response of breast cancer. Dll4 is a protein that helps control the growth of blood vessels in tumors. This study used near-infrared optical imaging and a novel machine learning framework to determine if Dll4 levels can be predicted from simple noninvasive imaging assays. The eventual results of this study may help physicians decide if a given triple-negative breast cancer patient will benefit from a Dll4 targeted therapy. Abstract Delta like canonical notch ligand 4 (Dll4) expression levels in tumors are known to affect the efficacy of cancer therapies. This study aimed to develop a model to predict Dll4 expression levels in tumors using dynamic enhanced near-infrared (NIR) imaging with indocyanine green (ICG). Two rat-based consomic xenograft (CXM) strains of breast cancer with different Dll4 expression levels and eight congenic xenograft strains were studied. Principal component analysis (PCA) was used to visualize and segment tumors, and modified PCA techniques identified and analyzed tumor and normal regions of interest (ROIs). The average NIR intensity for each ROI was calculated from pixel brightness at each time interval, yielding easily interpretable features including the slope of initial ICG uptake, time to peak perfusion, and rate of ICG intensity change after reaching half-maximum intensity. Machine learning algorithms were applied to select discriminative features for classification, and model performance was evaluated with a confusion matrix, receiver operating characteristic curve, and area under the curve. The selected machine learning methods accurately identified host Dll4 expression alterations with sensitivity and specificity above 90%. This may enable stratification of patients for Dll4 targeted therapies. NIR imaging with ICG can noninvasively assess Dll4 expression levels in tumors and aid in effective decision making for cancer therapy. | Dynamic NIR Fluorescence Imaging and Machine Learning Framework for Stratifying High vs. Low Notch-Dll4 Expressing Host Microenvironment in Triple-Negative Breast Cancer
Breast cancer is a disease that is affected by both the tumor cells and the host environment. It is well known that the tumor blood vessels are aberrant in structure and function due to rapid angiogenesis, and this aberrant vasculature plays a major role in drug delivery and therapy response of breast cancer. Dll4 is a protein that helps control the growth of blood vessels in tumors. This study used near-infrared optical imaging and a novel machine learning framework to determine if Dll4 levels can be predicted from simple noninvasive imaging assays. The eventual results of this study may help physicians decide if a given triple-negative breast cancer patient will benefit from a Dll4 targeted therapy.
Delta like canonical notch ligand 4 (Dll4) expression levels in tumors are known to affect the efficacy of cancer therapies. This study aimed to develop a model to predict Dll4 expression levels in tumors using dynamic enhanced near-infrared (NIR) imaging with indocyanine green (ICG). Two rat-based consomic xenograft (CXM) strains of breast cancer with different Dll4 expression levels and eight congenic xenograft strains were studied. Principal component analysis (PCA) was used to visualize and segment tumors, and modified PCA techniques identified and analyzed tumor and normal regions of interest (ROIs). The average NIR intensity for each ROI was calculated from pixel brightness at each time interval, yielding easily interpretable features including the slope of initial ICG uptake, time to peak perfusion, and rate of ICG intensity change after reaching half-maximum intensity. Machine learning algorithms were applied to select discriminative features for classification, and model performance was evaluated with a confusion matrix, receiver operating characteristic curve, and area under the curve. The selected machine learning methods accurately identified host Dll4 expression alterations with sensitivity and specificity above 90%. This may enable stratification of patients for Dll4 targeted therapies. NIR imaging with ICG can noninvasively assess Dll4 expression levels in tumors and aid in effective decision making for cancer therapy.
Breast cancer heterogeneity has been extensively studied and it has enabled the classification and categorization of tumors into molecular subtypes depending on the overexpression of antigens or hormone receptors on tumor cells [1]. Identification of tumor subtypes improves cancer patients’ care and prognosis by tailoring therapies to the subtypes [2,3,4]. Breast and many other cancers are highly heritable, yet most causative variants are unknown, and most of the known risk variants are considered tumor-cell-autonomous, with far less emphasis placed on identifying the role of germline variants impacting the tumor microenvironment (TME). The TME is a complex and dynamic system that includes cancer cells, stromal cells, blood vessels, and extracellular matrix [5] and plays a significant role both in tumor cell proliferation and in chemo- or radiotherapy delivery and response [5,6,7]. There is growing evidence that heritable modifiers of the tumor microenvironment can profoundly impact tumor behavior and response to diagnostic and therapeutic interventions [8,9,10,11,12]. Tumor blood vessels have abnormal structure and function, which leads to heterogeneity in blood perfusion both temporally and spatially [13]. This heterogeneity has multiple adverse consequences, including limiting the access of blood-borne drugs and effector immune cells to poorly perfused regions of tumors [14]. As a result, these areas become hypoxic and have low extracellular pH [15]. Hypoxia has been shown to play a significant role in tumor progression and metastasis by inducing genetic instability, angiogenesis, immunosuppression, inflammation, and resistance to cell death by apoptosis and autophagy [16,17]. Anti-angiogenic drugs are designed to target the vasculature in order to starve tumors and prevent them from growing. However, recent studies have shown that the efficacy of these drugs may be limited by specific biomarkers and pathways associated with resistance [15]. For example, it has been shown that some patients may not benefit from anti-VEGF therapies if they have elevated levels of plasma sVEGFR1 [18]. Similar outcomes have been observed with increased levels of SDF1α and anti-VEGF therapies [19]. Further the vascular TME and therapy response differs in primary tumor and metastasis sites [20] and the anatomic location [21,22,23]. Thus, characterizing angiogenesis in tumors holistically may have therapeutic implications. The process of tumor angiogenesis is closely regulated by a balance between promoting and suppressing angiogenic factors [24,25]. Delta like canonical notch ligand 4 (Dll4) is a protein-coding gene that provides instructions for making a protein part of a signaling pathway known as the notch pathway, which is essential for the normal development of many tissues throughout the body, affecting cell functions [26,27], modulating tumor angiogenesis [28], promoting vessel maturation, and inhibiting vessel sprouting by inducing apoptosis of tip endothelial cells (TECs) [28,29,30]. Dll4 is overexpressed in various types of cancer, including breast, ovarian, and colorectal cancer, and has been shown to promote tumor angiogenesis, growth, and metastasis by interacting with receptors on endothelial cells (ECs) [31,32,33,34]. Blockade of Dll4 activity results in enhanced vessel sprouting and increased vascular permeability [29,30,35], but anti-Dll4 therapy has not been universally successful, as Dll4 has been shown to have both pro-tumorigenic and anti-tumorigenic effects depending on the context of its expression [34,36,37]. We recently reported that Dll4 expression on the host TME rather than on tumor cells determines the EPR or enhanced permeation and retention effect in breast tumor xenografts and thus governs nanomedicine delivery and therapy response [38]. Despite the increasing evidence about the function of germline genetic modifiers, such as Dll4, in TME heterogeneity and enhanced permeability and retention (EPR) effects, the underlying influencers have mainly remained unexplored because of the lack of a systematic approach to studying them. Therefore, we developed the Consomic Xenograft Model (CXM) as a strategy for mapping heritable modifiers of TME heterogeneity. In the CXM, human breast cancer cells are orthotopically implanted into consomic xenograft host strains. These strains are derived from two parental strains with different susceptibilities to breast cancer. Salt-sensitive (SS) rats were employed as a tumor promoting strain, while Brown Norway (BN) rats were used as a tumor suppressing strain. A sequence of consomic strains were generated with chromosomes of SS rats replaced by those of BN rats one at a time and used for breast tumor xenograft studies [7,38,39]. Because the host backgrounds genetically differ by one chromosome, whereas the tumor cells are unvaried, any observed phenotypic differences are due to TME modifier(s) and can be linked to a single chromosome. These modifiers can be further localized by congenic mapping (inbred strains containing a given sub-chromosomal region in their genome). By combining CXM with dynamic epifluorescence near-infrared (DE-NIR) imaging, systemic injections of indocyanine green (ICG) through a tail vein in tumor-bearing rats, and multiparametric analysis of pharmacokinetic modeling, we localized and identified the function of the vascular-specific Dll4 allele on rat chromosome 3 (RNO3) as a heritable host TME modifier of EPR [38]. The SS.BN3IL2Rγ− CXM strain with low-level expression of Dll4 (referred to as Dll4−) had significant tumor growth inhibition compared with the parental SSIL2Rγ− strain with higher expression of Dll4 (Dll4+), despite a paradoxical increase in tumor blood vessel density in Dll4+. Further analysis revealed that the changes in the Dll4+ tumors were accompanied by altered expression of Dll4, which was previously linked with nonproductive angiogenesis. Additionally, Dll4 was found to be co-localized within a host TME modifier locus (Chr3: 95–131 Mb) identified by congenic mapping and correlated with the phenotypic differences observed at the consomic level [7,39,40]. The inheritance of functionally different Dll4 alleles can influence the efficacy of nanoparticle (NP) therapy, and previous results indicate that inherited microvascular distribution patterns, rather than overall NP uptake, ultimately determine the effectiveness of NP-mediated photothermal therapy (PTT). Consequently, patients with high endothelial Dll4 expression can be selected for treatment with anti-Dll4 targeted nanoparticles as opposed to patients with low Dll4 expression, where PEGylated nanoparticles will provide sufficient therapy response [38]. Recent advances in dynamic vascular imaging techniques, such as DCE-MRI and perfusion computed tomography, have facilitated the investigation of the time kinetics of a contrast agent to extract multiple vascular parameters and have been successfully applied in clinical trials of anti-angiogenic drugs [41,42]. However, these techniques have certain drawbacks, including a lack of high temporal resolution and the need for a heavy hardware system with sophisticated analysis software. Dynamic NIR fluorescence imaging, on the other hand, offers a sufficient and effective alternative to other dynamic vascular imaging techniques for characterizing germline-dependent vascular phenotypes [7,43]. This has led to the combination of these modalities, such as in the paired agent MRI-coupled fluorescence tomography approach for noninvasive quantification of paired-agent uptake in response to anti-angiogenesis therapy in vivo [44]. As the field of artificial intelligence continues to advance, researchers are increasingly utilizing AI techniques, particularly machine learning, to develop predictive models that can support effective decision making in various domains including cancer therapy selection [45,46]. Previous research has investigated the use of machine learning algorithms to analyze near-infrared (NIR) signal intensity and perfusion patterns to differentiate between healthy, benign, and malignant tissue [47]. This work demonstrated that the signal intensity time course of an FDA-cleared near-infrared dye ICG inflow during the wash-in phase and ICG outflow during the wash-out phase could serve as significant markers for tissue distinction. This finding offers a new method for noninvasive tissue distinction and has prognostic potential [48,49]. However, there remains a need for further exploration of the use of machine learning for classifying host genetic tumor microenvironment (TME) modifiers and predicting therapy responses based on dynamic contrast-enhanced imaging of tumors, particularly DE-NIR fluorescence imaging data [47].
We hypothesize that the observation of subtle differences in vasculature structure and perfusion patterns characterized by ICG inflow and outflow using DE-NIR imaging could be used to differentiate between inherited tumor vascular microenvironment differences, such as Dll4 expression levels. We propose an experimental framework to noninvasively assess Dll4 expression levels in tumors based on the NIR signal intensity time course of perfusion patterns characterized by ICG time kinetics to develop a predictive model to support effective decision making in cancer therapy. Herein, we used two rat-based CXM strains of breast cancer, SSIL2Rγ−(Dll4+) and SS.BN3IL2Rγ− (Dll4−) [7,38,50], as well as eight congenic xenograft strains, CG1–CG8 (Figure 1a,b), to assess the impact of germline TME vascular heterogeneity on the signal intensity of DE-NIR imaging with systemically delivered ICG. Principal component analysis (PCA)-based decomposition of time-dependent epifluorescence videos (image stacks) was used for visualization and anatomical segmentation of tumors, liver, lungs, and fat pads [7]. In addition, we utilized modified principal component analysis (PCA)-based anatomical segmentation techniques to identify and analyze regions of interest (ROIs) representing potential tumors within the current dataset. To gather further information, we calculated the average NIR intensity for each ROI by analyzing the brightness of individual pixels at each time interval, resulting in a series of intensity measurements for each ROI. From this analysis, several easily interpretable features were extracted, including the slope of the initial uptake of ICG, the time it takes to reach peak perfusion, and the rate of ICG intensity changes once the half-maximum intensity is reached (which, to the best of our knowledge, has not been previously reported in the literature). We then applied a subset of machine learning algorithms, including Support Vector Machines (SVMs), Naive Bayesian Classifiers (NBCs), Generalized Additive Models (GAMs), Decision Trees (DTs), Nearest Neighbors (NN), and Logistic Regression (LR) to select the most discriminative features for classification. The performance of the model was evaluated using confusion matrix, receiver operating characteristic curve (ROC), and the area under the curve. To further evaluate our hypothesis of detecting Dll4 expression levels from DE-NIR imaging and test the generalizability of our framework, we conducted a secondary performance evaluation method using congenic groups with high and low Dll4 expression levels. The classification models were trained based on the selected features, and the performance of the model was tested on the remaining congenic groups. We demonstrate that robust ML methods can identify the alterations in host Dll4 expression from the tumor dynamic imaging datasets, and thus these methods can potentially stratify patients for Dll4 targeted therapies.
All methods have been carried out in accordance with relevant guidelines and regulations. Approved protocols by the Medical College of Wisconsin Institutional Biosafety Committee (IBC) and Institutional Animal Care and Use Committee (IACUC) were followed. All live animal experiments are reported per the ARRIVE guidelines’ recommendations [51]. All results were rigorously adjusted for multiple comparisons.
All animal protocols employed in this study were approved by the Institutional Animal Care and Use Committee (IACUC), Medical College of Wisconsin (MCW). The MCW has an Animal Welfare Assurance (Assurance number D16-00064 (A3102-01)) on file with the Office of Laboratory Animal Welfare, National Institutes of Health (NIH). Animal experiments were performed according to the relevant guidelines and regulations and in compliance with the Guide for the Care and Use of Laboratory Animals published by the US National Institutes of Health (NIH publication no.85–23, revised 1996). SS.BN3 rats were developed as part of the Consomic Xenograft Model at the Medical College of Wisconsin (MCW) [40]. SS and SS.BN3 rats were purchased from the Rat Research Models Service Center at the Medical College of Wisconsin [52]. All rats were provided reverse osmosis hyper-chlorinated water ad libitum. All animal experiments were performed on anesthetized animals. The animal was placed in a transparent induction chamber to induce anesthesia. Isoflurane was delivered through a precision vaporizer and compressed O2 to the chamber. For induction, the percentage of isoflurane was up to 5%. Once the animal was unconscious, it was removed from the chamber. The unconscious animal was then placed on a warm surface and fitted with a nose cone attached to the vaporizer in the presence of a scavenging system and oxygen source. At this point, the concentration of isoflurane was reduced to this level that maintained the correct anesthesia plane, usually between 0.5 and 3%. After the end of the experiment, or when other criteria for animal protocols were justified, rats were euthanized. Rats were placed in an approved euthanasia chamber and exposed to CO2 from a compressed gas cylinder until the animal was no longer breathing. To ensure death in rats, a pneumothorax was created via thoracotomy for rats weighing more than 200 g. For rats weighing less than 200 g, a pneumothorax was created, or a cervical dislocation was performed. As previously described [39,40], consomic strains (SS and SS.BN3 rats) were generated by sequentially replacing SS chromosomes with the outbred wild-type and tumor-resistant strain of Brown Norway (BN) rats referred to as SS.BN#, which are reported for their tumorigenic potential, where # refers to the chromosome number. These parental SS and consomic SS.BN# strains were genetically ablated by knocking down the IL2Rγ gene to allow the grafting and growth of human cancer cell lines. Such immunocompromised strains are labeled as SSIL2Rγ−(Dll4+) and SS.BN#IL2Rγ−(Dll4−). Previous research has localized inherited modifier(s) of TME vascular heterogeneity to RNO3 by CXM mapping and further narrowed by congenic mapping to a 36 Mb locus containing Dll4 alleles with distinct vascular expression patterns in the SS.BN3IL2Rγ− consomic (Dll4−) and SSIL2Rγ− (Dll4+) rat strains, and verified via species-specific RNA sequencing and immunohistochemistry that strains inheriting the SS Dll4 allele on chromosome 3 have higher Dll4 expression on tumor-associated endothelium and that the blood vessel tortuosity and dysfunction increased in Dll4− strains [38,39]. Since there are many other candidate alleles on chromosome 3 that could also potentially account for the observed differences in therapeutic efficacy between the SS.BN3IL2Rγ− and SSIL2Rγ− strains [38], to further investigate the potential contribution of Dll4 to inherited tumor vascular heterogeneity, eight novel SS.BN3IL2Rγ− congenic xenograft host strains (CG1 to CG8) were constructed by introgression of the F1 progeny and F2 generation to capture different regions of RNO3 by marker-assisted selection, as described previously [50,53]. The exclusion congenic mapping localized a 7.9 Mb candidate region (marked by the SSLP marker D3Mgh11) that was associated with inherited tumor vascular heterogeneity and contained the Dll4 locus. As a result, eight new congenic strains CGN(s-e) were generated, where N (1 to 8) refers to the congenic group, while s and e refer to the starting and ending of Simple Sequence Length Polymorphism (SSLP) marker regions, respectively. This resulted in generating CG1(D3Rat26-D3Mgh30), CG2(D3Rat222-D3Got42), CG3(D3Rat222-D3Mco33), CG4(D3Rat164-D3Rat218), CG5(D3Rat26-D3Mco218), CG6(D3Rat86-D3Rat218), CG7(D3Mgh13-D3Rat218), and CG8(D3Rat160-D3Rat218) congenic groups (Figure 1a,b).
As previously described [38], triple-negative breast cancer MDA-MB-231 cells were maintained in DMEM media (Sigma, Burlington, MA, USA) supplemented with 10% FBS (Gibco, New York, NY, USA) and 1% penicillin and streptomycin (Lonza, Cohasset, MN, USA) and incubated in 5% CO2 at 37 °C. These cells (6 × 106) in 50% Matrigel were orthotopically implanted into the mammary fat pads (MFP) of 4- to 6-week-old female Dll4+ (n = 8) and Dll4– (n = 17) rats and eight congenic strains CG1 (n = 19), CG2 (n = 2), CG3 (n = 26), CG4 (n = 12), CG5 (n = 28), CG6 (n = 12), CG7 (n = 5), and CG8 (n = 4) (Figure 1c) [54]. Tumors were treated after 10 days of implantation at an approximate size of 600 mm3, consistent across all rat strains.
A customized NIR imaging system was assembled for imaging the rats. A bifurcated optical fiber bundle was used to deliver 785 nm excitation light (~5 mW/cm2 power at the surface, diode laser, ThorLabs Inc., Newton, NJ, USA) from two positions for uniform illumination of the entire rat body surface. A 16-bit deep-cooled intensified charge-coupled device camera (PIMAX4 ICCD, Princeton Instruments, Trenton, NJ, USA) equipped with 830 nm long-pass filter positioned following a holographic notch rejection filter in the optical path (ThorLabs Inc.) was used to image the rats through computer-controlled LightField® software (Teledyne Princeton Instruments, Trenton, NJ, USA) (Figure 1d). Dynamic contrast-enhanced NIR fluorescence imaging was performed on anesthetized rats, as reported previously for 800 nm NIR imaging [7]. In this study, the setup was used for imaging the whole body. A total of 133 rats (Dll4+ (n = 8), Dll4– (n = 17)) and eight congenic strains CG1 (n = 19), CG2 (n = 2), CG3 (n = 26), CG4 (n = 12), CG5 (n = 28), CG6 (n = 12), CG7 (n = 5), and CG8 (n = 4) were imaged. NIR imaging was performed for approximately 6 min following ICG injection with the CCD array hardware binned to 256 × 256 with a frame rate of 10.6 fps. A total of 3000 frames were captured for each imaging case, including about 50 frames for background correction. ICG (MP Biomedicals) was delivered in an intravenous bolus of 1200 µM ICG/200 g body weight into the tail vein via a catheter with a 32-gauge needle tip connected to a syringe pump (Harvard Apparatus PHD 2000 syringe pump, Holliston, MA, USA) operated at a speed of 0.2 mL/s. The injected volume was calibrated to provide a body-weight-equilibrated dose to each rat.
Image processing and data analysis were performed in MATLAB (R2021b MATHWORKS Inc.) software. The time-dependent image frames were assembled as 3D arrays (two spatial and one temporal dimensions) for all animals. A custom-designed breathing correction method with a low-pass temporal filter combined with a 1D wavelet-based denoising was used to filter the high-frequency jitter generated by the animal’s respiratory motion from the fluorescence kinetic sequences of each pixel, as described previously [7]. An average of pre-ICG injection frames (acquired in the ~5 s before ICG injection) was used as background, incorporating contributions from CCD noise and excitation light leakage from emission filters and subtracted from all the frames. (Refer to Video S1 for respiratory motion corrected time course images of ICG biodistribution).
First, motion correction and background subtraction were performed on the imaging data described in the previous steps. This was carried out to remove any potential artifacts or noise that could affect the accuracy of the principal component analysis. Next, the data were decomposed by PCA along the time dimension using MATLAB software following the previously published methods [55,56]. This resulted in converting the imaging data to a k-component vector for each pixel, where k is the number of time-frames in the original dataset. The PCA on the dynamic fluorescence image was used to extract spatial patterns of internal organs linked to statistically similar kinetic behaviors. This was carried out by comparing the k-component vectors for each pixel and identifying those that displayed similar patterns over time. The contribution of the first six principal components on a time basis is illustrated in Figure S1.
The ROI detection module employed three steps in order to identify the tumor area in images (Figure 2a): (1) spatial alignment, (2) PCA ranking and selection, and (3) ROI selection and masking. First, images were registered to a reference image using a rigid body transformation in order to ensure consistent spatial alignment for the detection of the tumor area in subsequent steps. The variability in the visual appearance of internal organs or tumors within certain principal components necessitated the adaptation of existing methods for ranking normalized components based on their two-dimensional cross-correlation (2DCC) with a reference image containing the tumor. This was achieved by generating a stack of the first ten normalized principal components and applying a two-dimensional cross-correlation function (xcorr2 in MATLAB) to each component. The principal components were ranked according to the correlation scores obtained and a semi-automatic algorithm was then utilized to carry out the subsequent two steps for tumor ROI generation. To select the appropriate principal component for tumor identification, the algorithm prioritized the PC with the highest likelihood of containing the tumor tissue based on the ranking from the previous step. Once the proper PC was selected, we used 2DCC to estimate the tumor’s location in the frame. This was performed by creating a stack of four reference images using PCs that clearly visualized the tumor and generated a bounding box around the tumor. Next, we applied basic morphological dilation and image arithmetic operations to emphasize the tumor’s boundaries. A threshold was then applied to separate the tumor from the background. Finally, we used Blob Analysis (Computer Vision Toolbox™, MATLAB, MathWorks, MA, USA) to extract the centroid and exact location of the tumor in the frame, generating a region of interest (ROI). The use of a graphic user-interface-enabled semi-automatic platform allowed for real-time evaluation and adjustment of the algorithm’s performance if necessary.
The underlying assumption for the NIR fluorescence intensity video analysis is that intensity was proportional to the ICG concentration in the tissue, which has been shown in the literature [57]. From each available video, , between 4 and 5 fluorescence time series are extracted, one based on the tumor ROI generated from earlier steps, and between three and four from mammary fat pads (MFPs) (Figure 2b). For each ROI, , and each included time step, the mean brightness of the pixels inside that ROI in the NIR video is stored as . The result is a collection of time series , where p ranges from 1 to (the number of animals in each group in our dataset), ranges from 4 to the number of ROIs generated in the specific animal (up to 5, 1 tumor and 3 or 4 fat pads), and t from 0 to 3000 (Figure 2b).
First, we started with smoothing the data to exclude any potential noise from motion artifacts. The smoothing was conducted using a Savitzky–Golay filter of order 3 with window length 31 [58]. The parameters for this filter were determined through manual annotation of peak and latency in a subset of the data, with the goal of minimizing average estimation error. Subsequently, we used a MATLAB function to identify the time point of maximum intensity, which corresponded to the peak of the fluorescence signal. We employed a custom latency detector script that analyzed the smoothed derivative of the curves and identified the first “robust” zero crossing as the latency point () [48].
The features were chosen in two steps as previously described (Figure 3) [47,48]: first, the following characteristic numbers were chosen for each normalized time series individually. The time to peak () is simply the time difference between the peak intensity time point and the latency . The upslope () is computed as which is the average slope between initial ICG arrival and the peak. The downslopes () are the downslopes between the peak and seconds further: The time ratio () is the ratio between and when reaches half the peak values. The half intensity forward () is the average slopes between when reaches half the peak values () and seconds further, so To increase the robustness of estimated downslope-based and -based features, we introduced a window around and time steps taking the median of those downslopes: These features provide insight into the tumor’s ICG uptake and decay. and relate to the initial uptake of ICG, while relates to the decay of ICG fluorescence [48]. is a measure of the temporal inhomogeneity of the initial uptake, and is a measure of the temporal inhomogeneity of both the initial uptake and decay of ICG fluorescence. To address inter-animal variation, we propose a feature design that relates the features of tumor intensity () to the features of healthy tissue intensity () in the same animal. The median value of a feature across the healthy tissue (fat pad) is chosen as a reference value, and each feature is defined as its percentage difference to that reference value. This results in a single normalized feature for each animal. By using the median as an average that is robust to outliers, our feature design allows for a more accurate comparison of tumor intensity across different animals.
The feature extraction and design required specialized knowledge of the tumor microenvironment (TME) and its impact on near-infrared intensity. However, the classification based on the inheritances of Dll4 can be considered a standard binary classification problem with feature selection as a sub-problem: given the features of an animal for which the Dll4 inheritance is unknown, we want to assign the label “Dll4high” or “Dll4low” to it. We also want to investigate which small subset of features performs best [44]. We restricted ourselves to a subset of available ML algorithms which were reported to perform best with intensity time series classification. [48], We excluded neural networks from consideration and evaluated Support Vector Machines (SVMs), Naive Bayesian Classifiers (NBs), Generalized Additive Models (GAMs), Decision Trees (DTs), Nearest Neighbors (NN), and Logistic Regression (LR).
DTs and SVMs using the full set of 86 features were trained. Each ML model was tuned to the training set in an internal cross-validation procedure of 10-fold. This process was repeated 20 times, and the performance metrics are reported as classifier performance. One of the key metrics used to evaluate the performance of a classification algorithm is accuracy, which measures the proportion of correctly classified instances in the dataset. Other important metrics include sensitivity, which measures the proportion of positive instances that were correctly classified, and specificity, which measures the proportion of negative instances that were correctly classified. In our case, we are interested in classifying animals as either Dll4high or Dll4low, and we use the average metric (Ascore) for each pair of groups to evaluate the performance of the different classification algorithms. The Ascore for a given pair of groups is calculated as the average of accuracy, sensitivity, and specificity, as follows:
In this study, we describe a two-step method for selecting congenic pairs with high and low Dll4 expression levels for feature selection and hypothesis testing. This method was used to identify well-behaved pairs for binary classification and to identify the pair with the highest classification performance: The primary selection of congenic pairs is based on their primary classification scores () and their performance against parental strains using all available features. The secondary selection is based on the classification performance using the 2 best-performing features. The congenic pair selection process involved the selection of all possible combinations of congenic pairs with high and low Dll4 expression levels with n > 5, resulting in a total of 12 pairs: Dll4+|Dll4−, Dll4+|CG3−, Dll4+|CG4−, Dll4−|CG1+, Dll4−|CG5+, Dll4−|CG6−, CG1+|CG3−, CG1+|CG4−, CG3−|CG5+, CG3−|CG6+, CG4−|CG5+, CG4−|CG6+. It should be noted that we only included subgroups with n > 5 in our analysis, to avoid introducing noise into the feature selection process. This is because smaller sample sizes can be more prone to variability and may not represent the larger population [59]. Therefore, we focused on more significant subgroups to ensure our feature selection process was robust and reliable. These 12 pairs have gone through our primary classification algorithm with a split of 75%–25% for testing and training with seeded randomization and proportional distribution of each group in training and testing datasets. Each ML model was tuned to the training set in an internal cross-validation procedure of 10-fold and evaluated by its performance on the test set. This process was repeated 20 times, and the best Ascore was reported as the metric of classifier performance. The result of this step was used to identify well-behaved congenic pairs for binary classification by calculating the separation score (Sscore) for each pair of congenic groups with CG#+ and CG#− as below, which is bound between 0 and 1 and reported in Table 1: The congenic pairs with Sscore above 80% were selected for secondary congenic pair selection. For the secondary selection, we evaluated the classification performance () of each pair determined this time by using the two most effective features as described in Section 3.13. We used a 75%–25% split for testing and training, with randomization to ensure a proportionate representation of each group in both datasets. The pair with the highest classification performance, Ascore, based on the two most effective features was selected for the final classification model training. To ensure a high-quality classification model, we set a minimum threshold for the Ascore of 0.70 for inclusion in the feature selection process. Congenic groups with an Ascore below 0.70 were considered to have an insignificant contribution to the classification model and were excluded from further consideration in the feature selection process. This approach allowed us to focus on the most informative feature pairs, improving the overall classification performance of our machine learning model. Each machine learning model was tuned to the training set using a 10-fold cross-validation procedure and evaluated based on its performance on the test set. We repeated this process 20 times and reported the best Ascore as the classifier’s performance.
We selected the best pair of features in terms of achieved sensitivity, specificity, and accuracy by a two-step procedure as previously reported: DTs and SVMs using the full set of 86 features were trained, and recursive feature elimination (RFE) was performed to refine a much smaller set of best-performing features [60]. RFE is a widely used machine learning classification algorithm that helps in reducing the dimensionality of feature space and selecting a small subset of features that yield the best classification performance. This was achieved through an iterative procedure that uses a ranking criterion to eliminate features one or more at a time. The RFE algorithm started by selecting a subset of features and training a model on this subset. The features were then ranked based on their contribution to the model’s performance, and the least important feature was eliminated. The process was then repeated with the remaining features, and the best subset of features was selected based on a model selection criterion [61]. One of the main advantages of RFE is that it helps to reduce the risk of overfitting when the number of features is large, and the number of training patterns is comparatively small [62]. This is because the algorithm selects only a subset of features that are relevant to the classification task, and this helps to avoid the inclusion of irrelevant and redundant features. RFE can be used in conjunction with other techniques such as regularization and support vector machines (SVMs) to further improve the performance of the classification model. In addition, projection methods such as principal component analysis can reduce the feature space’s dimensionality before applying RFE [55]. We used a k-fold cross-validation strategy to assess the performance of our model. We also reserved a portion of the training data for primary testing of the model after hyperparameter optimization. Our experiments were conducted with 20 random splits of the training and test datasets, and the mean performance metrics were reported for sensitivity, specificity, and accuracy as Ascore. To facilitate the interpretation of our results, we limited the number of final features to two. Furthermore, given the small size of our dataset, there was no justification for using high-dimensional feature spaces.
The use of data augmentation has become a popular technique in machine learning and deep learning, especially in the field of computer vision. Data augmentation involves applying random transformations to the training dataset to increase its diversity and improve the performance of a model. In this study, we used data augmentation on raw near-infrared (NIR) image stacks to evaluate the robustness of a classification model. We used a dataset of 3000 frames of the original raw 256 × 256 NIR images for this part of our study. These images were augmented using TensorFlow and the Keras API, which allowed us to apply random transformations to the dataset. The transformations included random rotation followed by a horizontal flipping, and up to 2% rescaling.
The final training and testing dataset for the machine learning models was determined by the outcome of the congenic pair selection and feature selection steps. This dataset included all congenic groups except for Dll4+ and Dll4−, as well as the selected CG#+ and CG#− groups in the previous step. This step was conducted separately for the original dataset and augmented dataset. The models were trained using 10-fold cross-validation and a portion of the training data was reserved for testing after hyperparameter optimization, with 25% for the original dataset and 20% for the augmented dataset. The performance of the models was assessed using a confusion matrix, receiver operating characteristic curve (ROC), and the area under the curve.
Repeated measures models are a powerful tool in statistical analysis that allow researchers to study the effects of different factors on a given outcome while accounting for the inherent dependence of multiple measurements taken on the same subject. In this study, a mixed effects model with appropriate time varying covariates was used to analyze the average fluorescence intensity of indocyanine green (ICG) in the tumor with multiple measurements per subject, with the subject number serving as the repeated measure indicator and the rat strain serving as a covariate. This allows for flexible time-based modeling when using multiple measures, likely dependent from the same animal [63,64]. Customized scripts in MATLAB were used to generate the fitted coefficients, covariance parameters, design matrix, error degrees of freedom, and between- and within-subjects factor names for the repeated measures model. The output was then analyzed with a multiple comparison of the estimated marginal means based on the variable strain, using the Tukey–Kramer test statistic [65]. This allowed estimation of multiplicity-adjusted p-values for the post hoc comparisons, which indicate whether the groups significantly differed with respect to strain. The data were then visualized as a p-value matrix, providing a clear illustration of the significant differences between groups.
This study employed the established consomic rat models SS and SS.BN3 as well as our congenic strains CG1 to CG8. The publicly accessible and NIH-supported Rat Genome Database (rgd.mcw.edu) catalogs has tools to explore the genotype and phenotype information for the SS (Dll4+) and SS.BN3 (Dll4−) and congenic strains under strain records: Dll4+ (RGDID:61499), Dll4− (RGDID:1358154), CG1 (RGDID:155782881), CG2 (RGDID:155782883), CG3 (RGDID:155782884), CG4 (RGDID:155791428), CG5 (RGDID:155791426), CG6 (RGDID:155791430), CG7 (RGDID:155791429), and CG8 (RGDID:155791427).
Dynamic contrast-enhanced NIR fluorescence imaging has been widely used for tumor detection in various studies [66,67,68]. The use of NIR imaging allows for the visualization of internal organs and tissues without the need for invasive procedures, which can be particularly useful in detecting tumors due to their vascular heterogeneity compared to surrounding healthy tissues. In previous studies, the use of principal component analysis (PCA) on the time domain of dynamic fluorescence images was utilized to extract spatial patterns of internal organs linked to statistically similar kinetic behaviors, such as liver, kidneys, lungs, and various tumors [7,56]. However, this technique required manual inspection and selection of proper principal components, which was time consuming and prone to human error and bias. In order to overcome the limitations present in the current dataset, we implemented a modified method that utilizes near-infrared imaging and principal component analysis to detect tumors with high accuracy and without the need for manual correction (Figure 2 and Figure S1). The use of principal component analysis in this context not only allows for dimensionality reduction and noise removal but also enhances the robustness and efficiency of the method. Our study also implemented a novel method of ranking PCA components based on the 2D cross-correlation of a reference image containing the tumor. This added to the simplicity and computational efficiency of the framework. However, it should be noted that this method may not be effective for detecting tumors with random locations. On the other hand, it could be useful for detecting tumors or tissues of interest with high localization, such as the lungs, liver, and kidney, and lesions in breast tissue. Overall, our method shows potential for improving the accuracy and efficiency of tumor detection using NIR imaging and PCA (Figure 4a). However, further experimentation is needed to expand the framework to a general tumor detection algorithm.
The analysis of the average fluorescence intensity of indocyanine green (ICG) in the tumor tissue of Dll4+ and Dll4− rats bearing triple-negative breast cancer (TNBC) tumors revealed that ICG uptake occurred more rapidly in Dll4− tissues and was retained for longer periods of time compared to Dll4+ hosts (Figure 4b). This indicates systemic differences in vascular function between the two rat strains. Our previous histological data showed that Dll4+ tumors have a higher vascular density and tortuosity, indicating a genetic microenvironment that promotes nonproductive angiogenesis [38]. This is further supported by the slower ICG wash-out observed in the Dll4+ tumors. These findings provide insight into the effects of host genetics on tumor angiogenesis and suggest potential therapeutic targets for TNBC. In order to further investigate the role of Dll4 in vascular function in tumors, we divided chromosome 3 into regions with and without the Dll4 gene in congenic rat strains (Figure 1b) and then examined the ICG fluorescence intensity of tumors in Dll4-high and Dll4-low rats (Video S2) (Figure 4c). Our findings reveal significant systemic differences in vascular function between tumors in Dll4+ and Dll4− rats (parental strains), indicating the critical role of the Dll4 gene in tumor angiogenic response [38]. However, analysis of the ICG fluorescence intensity of tumors for individual strains (Figure 4d) reveals more complex behavior than the obvious differences in wash-in and wash-out patterns observed between Dll4+ and Dll4−. This supports the need for further investigation into the impact of Dll4 on NIR time series signatures and the potential use of Dll4-directed therapies for cancer treatment. It is worth noting that although there are significant differences in Dll4-low vs. Dll4-high rat strains (when all the strains of Dll4 expression levels are combined), they are inconsistent with the observations made in Dll4+ and Dll4− rats. These results have significant implications for developing novel therapies that target Dll4 and other host TME modifiers involved in angiogenesis, as they demonstrate the critical role of these genes in tumor vascular function and angiogenic response. Additionally, our research further highlights the capricious nature of the NIR signal, which is influenced by various heritable tumor microenvironments across different groups, as shown in Figure 4b,c. We aim to illustrate and categorize the impact of the Dll4 expression level on the NIR signal through this erratic behavior. We used a repeated measures model to analyze the average fluorescence intensity of ICG in the tumor over time, with the rat strain serving as a covariate. Figure 5a,b show the estimated response covariances matrix, which is the covariance of the repeated measures. The higher values in this matrix indicate the time points at which groups experience the greatest differences. By projecting the diagonal of the covariance matrix onto the time axis (Figure 5c), we were able to visualize the amount of difference between groups over time. This projection, when compared to the average fluorescence intensity of ICG in the tumor (Figure 4b,d), showed the strongest differences between groups at the points where the NIR signal regions from half of its peak value to the peak value and at the tail of the curve, which are measures of the temporal inhomogeneity of the initial uptake and the decay of ICG fluorescence, were found to be particularly useful in discriminating between groups with different levels of Dll4 expression. This projection of the diagonal of the estimated response covariances matrix on the time curve can be used in feature design to focus on regions with the maximum amount of useful information for discriminating between groups and, subsequently, between classes with different levels of Dll4 expression. This could potentially improve the accuracy of tumor classification and ultimately improve therapy outcomes. Our repeated measures model, which included responses as measurements and strains as predictor variables, allowed us to conduct multiple comparisons of estimated marginal means between groups. The resulting p-value matrix (Figure 5d) revealed significant differences in estimated marginal means between the Dll4+ and Dll4− groups, with a p-value of 4.71 × 10−7. In addition, we observed significant differences between Dll4+ and CG3, CG4, and CG8, with p-values of 1.67 × 10−5, 7.03 × 10−7, and 2.18 × 10−3, respectively. For each group pair with high and low Dll4 expression levels, the separation score was calculated. First each of Dll4+|CG#−, CG#+|Dll4−, and CG#+|CG#− went through our classification algorithm with 10-fold cross-validation using Nearest Neighborhood, Linear SVM, RBF SVM, Decision Tree, Naive Bayes, and Logistic Regression models. The highest average classification metrics (Accuracy + Specificity + Sensitivity)/3 for Dll4+|CG#−, CG#+|Dll4− and CG#+|CG#−) was used to calculate the separation score (Score Dll4+|CG#− + Score CG#+|Dll4− + 2 × Score CG#+|CG#−)/4. Furthermore, our analysis showed significant differences between Dll4− and CG5 and CG6, with p-values of 8.70 × 10−3 and 2.58 × 10−4, respectively. This supports the hypothesis that Dll4 expression levels can act as a heritable TME modifier on NIR time series intensity. However, the smallest p-value between Dll4+ and Dll4− suggests that there are other factors on chromosome 3, in addition to Dll4, that contribute to the observed differences in the NIR time series signature between these groups. In contrast, no significant differences were found between Dll4− and the congenic strains with low levels of Dll4 expression (CG2, CG3, CG4, CG7, and CG8). This further supports the notion that Dll4 plays a crucial role in determining tumor vascular function and NIR time series intensity. Among the congenic groups, the most significant differences were observed between CG5, CG6, and CG3, CG4 from the Dll4-high and Dll4-low groups, respectively. Notably, the differences were most significant between CG4 and CG6, with a p-value of 0.0003. This suggests that very narrow regions of differences on chromosome 3 between these two groups, one containing Dll4 and the other lacking it, have a significant effect on the NIR time series signature.
The relationship between Dll4 expression and classification performance was analyzed using a total of 12 congenic pairs with n > 5 based on their levels of Dll4 expression (Dll4+|Dll4−, Dll4+|CG3, Dll4+|CG4, Dll4−|CG1, Dll4−|CG5, Dll4−|CG6, CG1|CG3, CG1|CG4, CG3|CG5, CG3|CG6, CG4|CG5, and CG4|CG6). The pairs were then subjected to a primary classification algorithm and their mean performance metrics, the Ascore, were calculated and reported in Table 1. The congenic pairs with low levels of Dll4 expression showed a mean Ascore of 0.91 +/− 0.01, indicating a high level of classification performance when compared to the Dll4+ parental strain. In contrast, the congenic pairs with high levels of Dll4 expression showed a mean Ascore of 0.8 +/− 0.05 when classified against the Dll4− consomic strain. Among the congenic pairs, the CG5|CG4 pair demonstrated the highest Ascore of 0.8, followed by the CG6|CG4 and CG6|CG3 pairs with Ascore values of 0.78 and 0.77, respectively. The results of the Ascore calculation are visualized in Figure 5e through a Sankey diagram. To account for potential differences between the congenic pairs and the parental pairs, the Sscore was calculated. The CG5|CG4, CG6|CG3, and CG6|CG4 pairs showed the highest Sscore values of 0.84, 0.84, and 0.85, respectively, and were selected for the feature selection step. These results align with the multiple comparison of estimated marginal means between groups, indicating that CG5|CG4, CG6|CG3, and CG6|CG4 show the strongest differences in classification performance.
RFE is a wrapper method that evaluates the entire classification algorithm and has shown improved classification accuracy and reduced overfitting compared to other feature selection methods [69]. However, RFE can be sensitive to noise and irrelevant features, leading to suboptimal feature subsets and reduced classification performance. Additionally, RFE is computationally intensive, which can pose a challenge for large datasets with a high number of features. Despite these limitations, RFE remains a valuable tool for selecting an optimal subset of features that maximizes classification performance [70,71]. To address these limitations, we performed feature selection in two steps to optimize the selection process and improve the performance of the classifier. First, we used RFE to select only two features out of the 86 available features for congenic pairs Dll4+|Dll4−, Dll4+|CG4, CG5|Dll4−, CG5|CG4, CG6|Dll4−, and CG6|CG4. The CG3 and its combinations (CG6|CG3, CG5|CG3, and Dll4+|CG3) were dropped from the feature selection process as the performances of the classifiers, the Ascore, using only two features were below 0.70, and lower than the other strains. The congenic pairs CG5|CG4 and CG6|CG4 as well as the parental and consomic group Dll4+|Dll4− went through our feature selection algorithm, and for each pair the two best-performing classification algorithms based on Ascore and associated feature pair were reported (Table 2). The highest Ascore for the two best-performing models for CG5|CG4 was 0.78 ± 0.04 compared to CG6|CG4 with an Ascore of 0.72 ± 0.19 and 0.72 ± 0.22, resulting in the selection of CG5|CG4 for final congenic pair selection. Finally, from each pair of Dll4+|Dll4−, Dll4+|CG4, CG5|Dll4, and CG4|CG5, four of the best performing features regardless of the ML model were chosen and were used as a collection of features for the final feature selection (Table 2). A combination of parental and consomic groups and the final selected congenic pair (Dll4+, Dll4−, CG5, and CG4) was used to select the final feature pair out of the 16 selected features, resulting in the selection of HIF5_avg and HIF50_avg as the best-performing features.
To evaluate the performance of the selected features, we trained datasets consisting of all the remaining congenic groups excluding the Dll4+, Dll4−, CG4, and CG5 (CG1 to CG3 and CG5 to CG8) using 10-fold cross-validation and keeping 25% of the dataset for testing the trained models. This allowed us to assess the generalizability of our model and test it on previously unseen datasets. The results of this step are reported as a confusion matrix, ROC curve, and AUC (Figure 6), as well as general classification metrics (Table 3). The best-performing models based on the selected features were SVM and KNN, with sensitivity and specificity of 1.00 and 0.81 and 1.00 and 0.75, respectively. In order to further assess the effectiveness of our model, the selected features, and the generated congenic pair, we generated an augmented dataset consisting of all remaining congenic pairs excluding Dll4+, Dll4−, CG4 and CG5 (CG1 to CG3 and CG5 to CG8, with random variations in rotation, horizontal flip, and limited scaling (up to ±2%) to increase the diversity of the dataset. This resulted in a total of 606 data points. The performance of the models was evaluated using 10-fold cross-validation and a 20% hold out. The results of this step were reported as a confusion matrix, ROC curve, AUC (Figure 7), and overall classification metrics (Table 4). The best-performing models based on the selected features were SVM and KNN, with sensitivity and specificity of 0.97 and 0.91, and 0.97 and 0.92, respectively. These results align closely with the performance of the models over the original dataset, indicating the generalizability of our framework. It is noteworthy that of the 16 most contributing features used to select the final feature pair, 12 were the newly proposed HIF features, and the other 4 were DS features, which we previously reported [48]. Additionally, the HIF features were amongst the best features for identifying genetic TME modifiers. The relationship between covariance of the repeated measures and optimal feature design in machine learning classification algorithms is an essential factor in developing effective classification algorithms. Combined with our recent report [47], our analysis found that the DS and HIF features, which are generated in regions where the NIR signal varies from half of its peak value to the peak value and at the tail of the curve, were particularly effective in discriminating between benign/malignant tumors (DS features) and groups with different levels of Dll4 expression (HIF features). Furthermore, the projection of the covariance matrix onto the time axis revealed similar regions, indicating a relationship between this projection and optimal feature design. These findings have significant implications for feature design in machine learning classification algorithms. By focusing on the regions with the greatest amount of useful information for discrimination, we can design features specifically to capture these differences and improve the accuracy of tumor classification. This can ultimately lead to better therapy outcomes for patients. It is worth noting that this relationship between the covariance of the repeated measures and optimal feature design is not limited to HIF features and the specific context of our analysis. In general, considering the covariance of repeated measures can provide valuable information for identifying key regions and designing effective features for machine learning classification algorithms.
Dynamic vascular imaging techniques such as DCE-MRI and perfusion CT are used to extract multiple vascular parameters and have been used in clinical trials of anti-angiogenic drugs. However, these techniques have limitations, such as low temporal resolution and the need for specialized hardware and software [41,42]. To overcome these limitations, dynamic near-infrared (NIR) fluorescence imaging can serve as an effective alternative for characterizing germline-dependent vascular phenotypes. It can be combined with other modalities, such as in a paired-agent or multimodal MRI and fluorescence tomography approaches for noninvasive quantification of response to anti-angiogenesis therapy and classifying in vivo vascular phenotypes [7,43,44,67]. Furthermore, DE-NIR imaging, as a potential alternative for characterizing germline-dependent vascular phenotypes in preclinical models, can be extended to clinical modalities upon validation with cross-sectional dynamic contrast enhanced imaging. The present study proposes that by combining DE-NIR imaging and machine learning algorithms with consomic xenograft models with human tumors, the role of inherited notch protein Dll4 (rat variant of delta like canonical ligand 4) expression specifically in the host vascular microenvironment can be studied. Specifically, in the context of breast cancer, where different genetic subtypes can impact treatment outcomes, identifying patients with high or low DLL4 (human variant of delta like canonical ligand 4) expression levels through noninvasive imaging could assist in selecting personalized treatment options. Nonetheless, the study authors acknowledge notable differences between the rat model utilized in the study and the human system, which could affect the generalization of the findings to human cases, as in human tumors DLL4 expression maybe be both on tumor cells and host vasculature, whereas in our CXM model, we focused specifically on the inherited variation in rat-derived host vasculature Dll4 expression in human xenograft tumors. Such differences may result in amplification or suppression of vascular phenotype responses if both tumor cells and the host microenvironment express high levels or contrasting levels of DLL4. However, even in that case, dynamic imaging will be useful in identifying patients likely to respond better to DLL4 targeted therapies. Future studies will be necessary to validate this study’s findings, to assess the reliability and validity of the developed imaging and machine learning algorithms in a large and diverse patient population to determine if contrast agent kinetic profiles observed in human DCE-MRI or dyna-CT imaging datasets for primary and/or metastatic disease differ in human patients with high vs. low DLL4 expression. In the metastatic setting, where surgery is no longer an option, a machine-learning-enabled analysis of dynamic contrast-enhanced imaging will be valuable to assess the expression levels of DLL4 and guide therapy selection, especially in cases where a biopsy is not taken or if biopsy results are inconclusive [72,73]. Human anti-DLL4 antibodies have been reported for cancer treatment [54,74,75,76]. In one study, immunotoxin DLL4Nb-PE was developed, potentially as a cell cytotoxic agent and angiogenesis maturation inhibitor [76]. Another study successfully developed a bispecific monoclonal antibody that targets both human DLL4 and VEGF and showed efficacy in inhibiting proliferation, migration, and tube formation of human umbilical vein endothelial cells (HUVEC) [74]. In a phase 1a trial, navicixizumab, a bispecific antibody that inhibits DLL4 and VEGF, was tested in refractory solid tumor patients and showed the potential to inhibit tumor growth [75]. While DLL4 blockade is an attractive therapy, long-term extended use of DLL4 mAbs has demonstrated concerning off-target effects [77,78]. Pharmacokinetic modulation of DLL4 mAbs may reduce off-target effects [77], such as via short-term administration or by focusing on patients where dynamic contrast imaging indicates a high DLL4 vascular phenotype. As we have shown in prior work, high vascular DLL4 expressing tumors may also be susceptible to DLL4 targeted nanomedicine [38] or as combination therapy with anti-DLL4 monoclonal antibodies with nanomedicine drugs such as nab-paclitaxel (Abraxane) or Liposomal Doxorubicin (DoxilTM). The use of noninvasive DE-NIR imaging to detect heritable TME modifiers is significant for several reasons. First, this method allows for the identification of potential modifiers without the need for invasive procedures, reducing the potential for discomfort and complications for patients. Second, the use of machine learning and DE-NIR imaging to develop a predictive model for cancer nanomedicine therapy can support effective decision making in the treatment process. While data processing and preparation and algorithm training can be complex, the resulting algorithms are simple and allow for the prediction of heterogeneity in a single step using ROI brightness measurements. Interestingly, traditional features such as time-to-peak and upslope do not appear in our selection of the most discriminative features. However, two novel features derived from (HIF5_avg and HIF50_avg), which is a measure of the temporal inhomogeneity of both the initial uptake and decay of ICG fluorescence, were identified. It is important to note that the training and testing sets used in this study are minimal, and therefore the high accuracy rates obtained should be interpreted with caution. Further research with larger datasets will be necessary to assess the reliability and validity of these findings with confidence. We have reported novel dynamic enhanced near-infrared (NIR) fluorescence imaging and machine learning algorithms to noninvasively assess Dll4 expression levels in tumors. Our results showed that observation of subtle differences in vasculature structure and perfusion patterns characterized by ICG time kinetics could be used to differentiate between inherited tumor vascular microenvironment differences, such as Dll4 expression levels. Additionally, our analysis demonstrated the importance of considering the covariance of the repeated measures in the design of features for machine learning classification algorithms. By utilizing this information, we can improve the accuracy of tumor classification and ultimately improve therapy outcomes for patients. To summarize, based on our recent study, we investigated the impact of genetically heterogeneous notch-Dll4 inheritance on the contrast agent uptake and clearance in triple-negative breast cancer xenografts. The differences in Dll4 inheritance have been shown to impact nanomedicine biodistribution, pharmacokinetics, and therapy response in our prior work. Thus, our results indicated that imaging can be potentially employed for selecting patients for Dll4-directed therapies by identify host microenvironments with high- or low-expressing Dll4 inheritance. This further suggests that the success of nanomedicine might depend on hereditary tumor microenvironment genes, regardless of tumor type. Additionally, host genes such as Dll4 can affect individual differences in NP uptake and response to NP-mediated therapies, providing the potential for more effective personalized Dll4 targeted nanomedicine for therapy-resistant hosts. Further studies are needed to validate these findings and explore the potential clinical applications of this approach. | true | true | true |
|
PMC10001392 | Li Wang,Tingting Wei,Li Zheng,Fangfang Jiang,Wentao Ma,Min Lu,Xiaomao Wu,Huaming An | "Recent Advances on Main Active Ingredients, Pharmacological Activities of Rosa roxbughii and Its De(...TRUNCATED) | 01-03-2023 | "R. roxburghii,nutritional composition,active ingredient,pharmacological activity,development,utiliz(...TRUNCATED) | "Rosa roxburghii tratt (R. roxburghii) is an important plant resource that is widely distributed in (...TRUNCATED) | "Recent Advances on Main Active Ingredients, Pharmacological Activities of Rosa roxbughii and Its De(...TRUNCATED) | true | true | true |
|
PMC10001635 | "Jose M. Romero-Márquez,Tamara Y. Forbes-Hernández,María D. Navarro-Hortal,Rosa Quirantes-Piné,G(...TRUNCATED) | "Molecular Mechanisms of the Protective Effects of Olive Leaf Polyphenols against Alzheimer’s Dise(...TRUNCATED) | 22-02-2023 | olive leaves,bioactive compounds,Alzheimer’s Disease,oleuropein,hydroxytyrosol | "Alzheimer’s Disease (AD) is the cause of around 60–70% of global cases of dementia and approxim(...TRUNCATED) | "Molecular Mechanisms of the Protective Effects of Olive Leaf Polyphenols against Alzheimer’s Dise(...TRUNCATED) | true | true | true |
|
PMC10001743 | "Stana Tokić,Maja Jirouš,Vera Plužarić,Martina Mihalj,Marija Šola,Maja Tolušić Levak,Kristina(...TRUNCATED) | "The miR-20a/miR-92b Profile Is Associated with Circulating γδ T-Cell Perturbations in Mild Psoria(...TRUNCATED) | 21-02-2023 | hsa-mir-20a,hsa-mir-92b,hsa-mir-29a,hsa-let-7c,psoriasis vulgaris,γδ T cells | "Psoriasis vulgaris (PV) is an autoinflammatory dermatosis of unknown etiology. Current evidence sug(...TRUNCATED) | "The miR-20a/miR-92b Profile Is Associated with Circulating γδ T-Cell Perturbations in Mild Psoria(...TRUNCATED) | true | true | true |
|
PMC10002134 | "Viviana Costa,Marcello De Fine,Lavinia Raimondi,Daniele Bellavia,Aurora Cordaro,Valeria Carina,Ricc(...TRUNCATED) | Timing Expression of miR203a-3p during OA Disease: Preliminary In Vitro Evidence | 21-02-2023 | osteoarthritis,microRNAs,osteoblasts,interleukines,CX-43,SP-1,TAZ | "Osteoarthritis (OA) is a degenerative bone disease that involves the microenvironment and macroenvi(...TRUNCATED) | "Timing Expression of miR203a-3p during OA Disease: Preliminary In Vitro Evidence\n\nOsteoarthritis (...TRUNCATED) | true | true | true |
|
PMC10002166 | Molin Li,Weimin Gong,Jie Chen,Yining Zhang,Yufei Ma,Xiaolin Tu | "PPP3R1 Promotes MSCs Senescence by Inducing Plasma Membrane Depolarization and Increasing Ca2+ Infl(...TRUNCATED) | 23-02-2023 | BMSC,aging,PPP3R1,Ca2+,membrane potential | "Aging of mesenchymal stem cells(MSCs) has been widely reported to be strongly associated with aging(...TRUNCATED) | "PPP3R1 Promotes MSCs Senescence by Inducing Plasma Membrane Depolarization and Increasing Ca2+ Infl(...TRUNCATED) | true | true | true |
|
PMC10002460 | "Prakash P. Pillai,Muthukumar Kannan,Susmita Sil,Seema Singh,Annadurai Thangaraj,Ernest T. Chivero,R(...TRUNCATED) | Involvement of lncRNA TUG1 in HIV-1 Tat-Induced Astrocyte Senescence | 22-02-2023 | aging,astrocyte senescence,HIV-1 Tat,lncRNA TUG1,proinflammatory cytokines | "HIV-1 infection in the era of combined antiretroviral therapy has been associated with premature ag(...TRUNCATED) | "Involvement of lncRNA TUG1 in HIV-1 Tat-Induced Astrocyte Senescence\n\nHIV-1 infection in the era (...TRUNCATED) | true | true | true |
End of preview. Expand
in Dataset Viewer.
LitScan high quality subset
This is a subset of the ePMC articles located by LitScan that mention the RNA ID in their title.
We're calling this high quality because the hope is that if the RNA is mentioned in the title, then the paper is definitely about that RNA and doesn't just mention it in passing.
There are approx 50k articles in this subset, which isn't loads.
- Downloads last month
- 4