Proteomic identification of prognostic tumour biomarkers, using chemotherapy-induced cancer-associated fibroblasts

1 Metabolism

One of the major contributors to the aging process is mitochondrial dysfunction, which involves a decrease in the oxidative phosphorylation efficacy and an increase in the electron leakage resulting in reduced ATP generation [17]. Similarly, one of the hallmarks of the catabolic tumour stroma is the induction of metabolic stress that favours glycolysis to the detriment of mitochondrial metabolism. We have showed in a previous study that several chemotherapeutic agents, including azathioprine and taxol were able to stimulate stromal fibroblasts to consume more glucose and produce more lactate, which was released via enhanced MCT4 expression, hence increasing extracellular acidification (Figure 1). The cellular ATP content was also minor upon treatment suggesting a decrease in mitochondrial respiration [11].

To validate that chemotherapy-induced glycolytic phenotype and to identify other cellular metabolic pathways potentially altered by chemotherapy, we searched the proteomic data for changes in the expression of metabolic enzymes. Of all differentially regulated proteins, 22.04% and 25.40% were actually enzymes in azathioprine-treated and taxol-treated fibroblasts, respectively (Figure 3C).

The expression of most glycolytic enzymes was significantly altered (Table 1). Interestingly, most of the glycolytic enzymes that were found to be up-regulated are enzymes that perform an irreversible reaction in the glycolytic pathway and all down-regulated enzymes are reversible enzymes, able to perform the opposite reaction, hence also involved in gluconeogenesis. Curiously, LDHA, the enzyme involved in transforming pyruvate into lactate was found to be down-regulated in both treatments suggesting that pyruvate was further processed into the TCA cycle instead of being transformed into lactate.

Nevertheless, LDH is an enzyme that exhibits feedback inhibition, by which high lactate concentrations can suppress it. In fact, MCT4, the monocarboxylate transporter responsible for the secretion of lactate, turned out to be up-regulated in taxol-treated hTERT-BJ1 cells, and regarding the further processing of pyruvate into acetyl-CoA and the citric acid (TCA) cycle, the expression of practically all enzymes were found to be down-regulated in taxol-treated fibroblasts compared to vehicle indicating a dramatic down-regulation of mitochondrial metabolism. A similar down-regulation trend was observed in numerous proteins involved in all oxidative phosphorylation complexes (Table 2) and mitochondrial function proteins (Table S1). Finally, several enzymes involved in mitochondrial fatty acid β-oxidation were also identified as down-regulated, and a few enzymes involved in fatty acid biosynthesis, up-regulated (Table S2). Other metabolic changes included enzymes of the oxidative pentose phosphate pathway, responsible for the generation of antioxidant power (NADPH), the hexosamine synthesis pathway, accountable for the production of amino sugars used for the synthesis of glycoproteins, glycolipids and proteoglycans, and a few enzymes involved in the generation of ketone bodies (Table 1 and 5). Figure 4A summarises all changes observed in the expression of metabolic enzymes in taxol-treated fibroblasts and their contribution to different cellular metabolic pathways.

IPA revealed glycolysis, gluconeogenesis and pentose phosphate pathway as altered canonical pathways in both azathioprine and taxol treatments. Similarly, IPA revealed mitochondrial dysfunction, TCA cycle, and PPARα/RXRα activation, responsible for ketone body production and fatty acid metabolism, as two of the top canonical pathways affected by both treatments, and oxidative phosphorylation and fatty acid β-oxidation, as altered pathways also in taxol-treated cells (Figure 5 and Table S5). The toxicity impact of both drugs extensively involved mitochondrial dysfunction and damage, and also fatty acid metabolism and PPARα/RXRα activation (Figure 6 and Table S6).

To recapitulate, the previously identified chemotherapy-induced metabolic stress in stromal fibroblasts is clearly detected by quantitative proteomics and it represents not only an increase in glycolysis and a reduction in mitochondrial function, as observed in our previous study, but also affects other metabolic pathways, including hexosamine synthesis and pentose phosphate pathways, fatty acid metabolism and ketogenesis.

2 Antioxidant response and stress-related pathways

Another feature of the catabolic tumour stroma is the induction of oxidative stress, which can be induced by chemotherapy in cancer cells and also healthy tissues (Figure 1) [18, 19]. Particularly, we previously showed increased ROS production and antioxidant response in hTERT-BJ1 fibroblasts after treatment with taxol [11]. We have seen in this study that several enzymes of the oxidative branch of the pentose phosphate pathway, which is responsible for the generation of antioxidant power (NADPH), were up-regulated in chemotherapy-treated stromal fibroblasts. Thus, we sought for other antioxidant response proteins in the proteomics datasets. The expression of numerous proteins involved in Nrf2-mediated antioxidant response and other antioxidant proteins was significantly altered, in most cases up-regulated (Table 3), suggesting an activation of the pathway in both azathioprine and taxol-treated hTERT-BJ1 fibroblasts, although it was found to be notably higher in taxol-treated cells.

A higher presence of antioxidant response proteins was further confirmed using IPA, which revealed Nrf2-mediated antioxidant response to be one of the top canonical pathways affected by azathioprine and taxol treatments, although z score values demonstrated a significant activation of this pathway exclusively in taxol-treated fibroblasts. Similarly, taxol-treated cells showed altered glutathione and thioredoxin antioxidant pathways (Figure 3E and 5, and Table S5). The toxicity of these chemotherapeutic drugs also involved oxidative stress. HIF signalling, which is activated in response to stress and a central player in the regulation of cellular metabolism, was also identified as a toxic effect of azathioprine treatment (Figure 6 and Table S6), and had been previously identified as activated in hTERT-BJ1 fibroblasts treated with azathioprine and taxol [11]. Thus, proteomics analysis clearly detects an activation of the antioxidant response after treatment with chemotherapy.

3 Myofibroblastic differentiation

Cancer-associated fibroblasts are commonly identified by their expression of alpha smooth muscle actin (αSMA) [20] (Figure 1). Indeed, we previously showed up-regulation of αSMA by immunoblotting in taxol-treated hTERT-BJ1 [11]. Quantitative proteomics profiling of azathioprine and taxol-treated hTERT-BJ1 fibroblasts also revealed a significantly higher presence of αSMA and many other myofibroblastic markers, such as fibroblast activation protein (FAP) or vimentin, as well as muscle-related proteins compared to the vehicle-treated control (Table 4).

Myofibroblasts are mostly responsible for the presence of fibrosis [21], and the toxicity functions of these chemotherapeutic drugs involved tissue fibrosis, as analysed by IPA (Figure 6 and Table S6). Therefore, chemotherapy can independently induce the differentiation of hTERT-BJ1 fibroblasts into cancer-associated fibroblasts, which can be detected by a numerous increase of myofibroblastic markers in quantitative proteomics analysis.

4 Autophagy and senescence

Autophagy and senescence represent a common response to stresses such as exposure to DNA-damaging exogenous cytotoxic agents, including chemotherapy or radiation [22, 23]. Indeed our previous study indicated that azathioprine and taxol induce autophagic vesicle formation and increase β-galactosidase activity in stromal fibroblasts [11]. Thus, to further examine changes in autophagy and senescence upon chemotherapeutic exposure, the differential expression of several autophagy, mitophagy and lysosomal markers was analysed using the proteomics datasets. Numerous autophagy and senescence markers such as sequestrosome 1, also known as p62, cathepsin B or the lysosomal enzyme β-galactosidase were found to be up-regulated in either azathioprine or taxol treatments (Table 5). In addition, DNM1L, a protein involved in mitochondrial fission, was up-regulated in taxol-treated hTERT-BJ1 fibroblasts, and OPA1, a protein involved in mitochondrial fusion, was down-regulated in azathioprine-treated cells. Likewise, VAT1, a vesicle membrane protein that inhibits mitochondrial fusion, was up-regulated in both treatments, suggesting that mitophagy might be also activated in response to chemotherapy (Table 5 and Table S3).

During autophagy, protein and lipid degradation occur, the latter leading to the generation of ketone bodies. A long list of up-regulated proteins involved in vesicle formation and trafficking, in protein ubiquitination pathway and proteasomal degradation, and a few enzymes involved in ketogenesis were also identified (Table 5, S3 and S4). The boost in vesicle formation and trafficking proteins could explain the greater presence of enzymes involved in fatty acid synthesis and hexosamine biosynthesis (Figure 4A and Table 1 and S2). The observed effects of taxol treatment on autophagy, mitophagy and senescence are summarised in Figure 4B.

IPA analysis confirmed an alteration of ketogenesis in fibroblasts exposed to azathioprine (Figure 5 and Table S5). IPA also showed a robust inhibitory effect of azathioprine and taxol on EIF2 signalling, responsible for protein synthesis, as measured by z score, and also revealed an alteration in the regulation of eIF4 and p70S6K signaling and in the protein ubiquitination pathway. Finally, PPARα/RXRα activation, Rho, caveolar and clathrin-mediated signalling pathways, all involved in vesicle trafficking and motility, and the mTOR signalling, known for its role in autophagy, mitochondrial metabolism and lipid metabolism, as well as cytoskeleton dynamics, were some of the top canonical pathways altered by chemotherapy, (Figure 3E, 5, and Table S5). Interestingly, p53, known mediator of senescence, was predicted to be activated in both azathioprine-treated and taxol-treated hTERT-BJ1 cells according to IPA, and TFEB, a transcriptor factor that coordinates the expression of lysosomal hydrolases, membrane proteins and genes involved in autophagy, was also predicted to be activated by taxol treatment (Figure 3F and 4C). Our previous study also reported an increased expression of p53 in hTERT-BJ1 fibroblasts treated with azatioprine and taxol by immunoblotting [11].

Thus, quantitative proteomics analysis reveals a higher presence of markers of autophagy and senescence and proteins involved in protein degradation and vesicle trafficking in hTERT-BJ1 fibroblasts, upon treatment.

5 Inflammation

Senescent cells dramatically alter their secretome, enriching it with pro-inflammatory cytokines and matrix metalloproteinases. This senescence-associated secretory phenotype (SASP) can lead to chronic inflammation, which is a hallmark of aging [17]. Stromal fibroblasts secrete inflammatory cytokines when in contact with cancer cells [24]. Chemotherapy is also able to induce cytokine production in healthy tissues [12], and in particular taxol treatment induces IL6 secretion in stromal fibroblasts (Figure 1) [11]. STAT3, a known inducer of inflammation in response to stress [25, 26] was found significantly up-regulated in both azathioprine (1.69 fold increase) and taxol treatments (2.35 fold increase) relative to vehicle. STAT3 signalling was found to be activated in response to taxol also in our previous study [11].

Interestingly, several pathways involved in the inflammatory process such as IL8, acute phase response, leukocyte extravasation or Fcγ receptor-mediated phagocytosis signalling were amongst altered canonical pathways in both azathioprine-treated and taxol-treated hTERT-BJ1 cells relative to vehicle, according to IPA. N-formyl-Met-Leu-Phe (fMLP) signalling was also found to be altered in taxol-treated fibroblasts (Figure 3E and 5 and Table S5). Most of these pathways were clearly activated in taxol-treated fibroblasts as indicated by z score values, suggesting an induction of the inflammatory response in chemotherapy-treated stromal fibroblasts. Likewise, the chemokine (C-C motif) ligand 5 (CCL5 or RANTES), which plays a role in recruiting leukocytes into inflammatory sites, was identified as an upstream regulator in both treatments, particularly activated in taxol-treated fibroblasts (Figure 3F and 4C). Finally, acute phase response, which occurs soon after the onset of an inflammatory process, was identified as one of the toxic effects of both drugs (Figure 6 and Table S6). Therefore, pathway analysis of the proteomics results indicates an induction of the inflammatory response in stromal fibroblasts after exposure to chemotherapy.

Chemicals and sample preparation

Formic acid, trifluoroacetic acid, ammonium formate (10 M), ammonium bicarbonate TCEP (Tris (2-carboxyethyl)phosphine hydrochloride), MMTS (Methyl methanethiosulfonate) and trypsin were all obtained from Sigma. HPLC gradient grade acetonitrile was obtained from Fisher Scientific. Briefly, 2 × 10⁶ hTERT-BJ1 fibroblasts were seeded in 150 cm plates until cells were attached. Cells were then treated with azathioprine or taxol at the concentrations indicated. As control, vehicle-treated cells were processed in parallel. After 48 hours of treatment, cells were lysed in RIPA buffer (R0278, Sigma) and kept at 4°C for 20 minutes with rotation. Lysates were cleared by centrifugation for 10 minutes at 10,000 × g and supernatants were collected and kept frozen at −80°C.

Protein digestion

Lysate samples were thawed to room temperature and their concentrations equalised to 1 μg/μL (50 μL volume) with RIPA buffer, and further processed for trypsin digestion by sequential reduction of disulphide bonds with TCEP and alkylation with MMTS. Briefly, 1 μL benzonase (Novagen) was added to the 50 μL aliquot and placed on ice for 15 minutes. The sample was then taken to dryness using a SpeedVac, and resuspended in 22.5 μL trypsin reaction buffer (40 mM ammonium bicarbonate and 9% acetonitrile). One μL of 50 mM TCEP solution was added to each sample, mixed briefly and placed on a heater block at 60°C for 60 minutes. After cooling to room temperature, 0.5 μL of 200 mM MMTS solution was added to each sample and allowed to react for 15 minutes. Trypsin was added in two waves to ensure efficient digestion of the sample. Firstly, 20 μg of sequencing grade trypsin was resuspended in 1800 μL of trypsin reaction buffer; 225 μL of this solution were added to each sample for digestion, and the reactions were left at 37°C overnight with shaking (600 rpm). The following morning, a further aliquot of trypsin was added. Two ml of trypsin reaction buffer was added to 20 μL of sequencing grade trypsin; 250 μL of this solution were added to each of the digest samples from overnight, and the reactions were left at 37°C for 4 hours with shaking (600 rpm). Thirty-five μL 10% formic acid were added to the 500 μL digest sample (0.7% final concentration of formic acid) to stop the digestion. The digested solution was diluted in 7.5 mL of acetonitrile containing 0.3% formic acid.

HILIC solid phase extraction (SPE) of peptides

PolyhydroxyethylA SPE 12 μm, 300A, 300mg cartridges (obtained from PolyLC) were used for the HILIC procedure. Prior to use, cartridges required an overnight soak in 50 mM formic acid followed by rinsing with water the following day. Cartridges were preconditioned with 2 mL of Buffer A (90% acetonitrile, 5 mM ammonium formate, pH 2.7) followed by 2 mL of Buffer B (5 mM ammonium formate, pH 2.7) and finally re-equilibrated with 10 mL Buffer A. The diluted samples were loaded onto the cartridges and washed with a further 10 mL Buffer A. Finally, peptides were eluted in 1 mL Buffer C (9 parts Buffer B plus 1 part Buffer A) and the samples dried on a Speedvac to remove organic solvent prior to LCMS/MS analysis.

LC-MS/MS analysis

Lyophilised digests were resuspended in 50 μL of 0.1% TFA to give an approximate concentration of 1 μg/μL. One μL injection volumes were used throughout resulting in an on-column peptide loading of approximately 1 μg per injection. Analysis was performed in quintuplicate for each sample. All LC-MS/MS analyses were performed on an LTQ Orbitrap XL mass spectrometer coupled to an Ultimate 3000 RSLCnano system (Thermo Scientific). One μL injection volumes were used throughout and samples loaded directly onto the analytical column, PepMap RSLC C18, 2 μm × 75 μm id × 50 cm (Thermo Scientific). The composition (v/v) of LC buffers were as follows; Buffer A - 99.9% water plus 0.1% formic acid and Buffer B - 80% acetonitrile, 19.9% water and 0.1% formic acid. Peptides were loaded directly onto the column at a flow rate of 400 nl/min with an initial mobile phase composition of 1% B. The organic strength was increased linearly from 1% to 22.5% B over 22.5 minutes again at 400 nl/min, followed by an increase to 24.8% B over the next 2.6 minutes with a concomitant reduction in flow rate to 300 nl/min, and to 39% B over a further 14 minutes. A further increase to 60% B over the next 5 minutes was followed by a ramp to 95% B over 2.5 minutes where it was held for a further 2 minutes. The column was then allowed to re-equilibrate to 1% B for a total analysis time of 74 minutes. The mass spectrometer was instructed to perform data dependent acquisition on the top six precursor ions, which were measured in the Orbitrap FTMS detector over the mass range 370–1200 m/z, at a nominal resolution of 60,000. MS/MS spectra were acquired in the ion trap under CID conditions with normalized collision energy of 35, isolation width of 3 Th, Q value of 0.25 and 30 ms activation time. Gasphase fractionation was performed on the five replicate injections such that MS/MS data was collected for precursor ion range 370–494 m/z Injection 1, 494–595 m/z Injection 2, 595–685 m/z Injection 3, 685–817 m/z Injection 4 and 817–1200 m/z Injection 5.

Statistical analysis

Xcalibur raw data files acquired on the LTQ-Orbitrap XL were directly imported into Progenesis LCMS software (Waters Corp) for peak detection and alignment. Data were analysed using the Mascot search engine. Five replicates were analysed for each sample type (N = 5). Statistical analyses were performed using ANOVA and only fold-changes in proteins with a p-value less than 0.05 were considered significant.