Abstract

   As sorafenib is a first-line drug for treating advanced hepatocellular
   carcinoma, sorafenib resistance has historically attracted attention.
   However, most of this attention has been focused on a series of
   mechanisms related to drug resistance arising after sorafenib
   treatment. In this study, we used proteomic techniques to explore the
   potential mechanisms by which pretreatment factors affect sorafenib
   resistance. The degree of redundant pathway PI3K/AKT activation,
   biotransformation capacity, and autophagy level in hepatocellular
   carcinoma patients prior to sorafenib treatment might affect their
   sensitivity to sorafenib, in which ADH1A and STING1 are key molecules.
   These three factors could interact mechanistically to promote tumor
   cell survival, might be malignant features of tumor cells, and are
   associated with hepatocellular carcinoma prognosis. Our study suggests
   possible avenues of therapeutic intervention for patients with
   sorafenib-resistance and the potential application of immunotherapy
   with the aim of improving the survival of such patients.

   Keywords: Sorafenib resistance, Autophagy, Biotransformation,
   Proteomics, HCC, PI3K/AKT/mTOR

Graphical Abstract

   ga1
   [45]Open in a new tab

1. Introduction

   According to Global Cancer Statistics 2020, primary liver cancer is the
   sixth most common cancer and the third leading cause of cancer death
   worldwide, with a particularly high incidence in East Asia, where
   75–85% of the pathological types are Hepatocellular carcinoma (HCC)
   [46][1]. HCC has an insidious onset and low rate of early diagnosis and
   is usually seen in patients with chronic hepatitis associated with
   viral infections, alcohol abuse, or metabolic syndrome. Advanced HCC is
   not sensitive to radiotherapy or chemotherapy [47][2], [48][3].
   Sorafenib, a first-line drug for treating advanced HCC, is an oral
   multikinase inhibitor. It targets RAF, c-Kit, Flt-3, and RET in tumor
   cells as well as the vascular endothelial growth factor receptor
   (VEGFR) and platelet-derived growth factor receptor (PDGFR) in vascular
   endothelial cells [49][4], [50][5], [51][6]. However, pathways such as
   the epidermal growth factor receptor (EGFR), PI3K/AKT/mTOR, hepatocyte
   growth factor (HGF)/c-Met, and insulin-like growth factor (IGF)/IGFR
   pathways are associated with the development of HCC [52][7].

   Due to the more limited target of sorafenib, only approximately 30% of
   patients benefit from sorafenib, and individuals in these populations
   usually acquire resistance within 6 months of treatment initiation
   [53][3]. The mechanisms associated with resistance following sorafenib
   administration have received sufficient attention. Sorafenib treatment
   is currently thought to activate resistance mechanisms such as those
   mediated by noncoding RNAs, upregulation of proangiogenic signaling
   pathways, activation of the PI3K/AKT/mTOR pathway, abnormal
   intracellular pharmacokinetics and the hypoxic tumor microenvironment
   [54][5], [55][8]. However, considering the low patient benefit rate of
   sorafenib, there is also clinical value in predicting patients'
   sensitivity to sorafenib through evaluation of pretreatment factors to
   adjust treatment regimens. In addition, sorafenib resistance is
   considered to be reversible [56][5]: in a mouse model study,
   sorafenib-resistant HCC tumor cells from one mice were isolated and
   transplanted into another mice, and the transplanted tumors were not
   found to be resistant to sorafenib in the new host, i.e., the
   sorafenib-resistant phenotype is not propagated [57][9]; another study
   found that inhibition of adenosine triphosphate citrate lyase (ACLY), a
   key enzyme in lipid metabolism, reversed the resistance phenotype in
   sorafenib-resistant hepatocellular carcinoma cell lines HepG2 and Huh7
   [58][10]; in renal cell carcinoma, Ubenimex was found to inhibit Akt
   expression and enhance autophagy, promote apoptosis in 786-O-R cell
   line and reverse resistance to sorafenib, silencing of the
   apoptosis-inhibiting mucin MUC13 also had this effect [59][11],
   [60][12]; in gastric cancer studies, sorafenib resistance in cell lines
   was overcome by silencing Sirtuins 6 (SIRT6) to promote ferroptosis
   [61][13]. Alterations in the sorafenib resistance phenotype have been
   observed in many of the above tumor types in many aspects of
   modulation, and we therefore believe that modulating the relevant
   influences prior to treatment may improve patient outcomes.

   Among the pretreatment factors, autophagy helps cells maintain
   homeostasis by recycling aged or damaged organelles and is an essential
   means for cells to cope with survival stress in a harsh environment
   [62][14]. Biotransformation is a vital function of the liver, and
   although biotransformation is not equivalent to detoxification, it can
   reduce the toxicity of most substrates and render them readily
   excreted. The phase II reactions in biotransformation include
   glucuronidation, sulfation, methylation, acetylation, glutathione
   binding, and amino acid binding [63][15]. Autophagy and
   biotransformation may profoundly affect patient sensitivity to
   sorafenib. In this study, we performed proteomic analysis of the
   tissues of 16 Chinese HCC patients after postoperative relapse prior to
   sorafenib treatment and partially explained how some of the
   pretreatment factors, represented by autophagy and biotransformation,
   interact and thus influence the efficacy of subsequent sorafenib
   treatment.

2. Materials and methods

2.1. Sample collection

   We obtained surgical resection samples from 16 HCC patients before
   sorafenib treatment for proteomic analysis. These patients were started
   on sorafenib because of recurrence after radical hepatectomy. According
   to the American Joint Committee on Cancer (AJCC) 8th, the starting dose
   of sorafenib is 400 mg bid, and the dosage is adjusted during the
   course of treatment according to the patient's side effects, until the
   course of treatment is stopped when the patient becomes resistant to
   sorafenib. Seven patients (X1–7) were deemed to respond to sorafenib
   treatment, and the remaining nine were deemed nonresponding (X8–16) by
   more than two clinical internists. The criteria for response
   classification were as follows: patients with disease progression
   within 6 months of sorafenib administration were classified as
   nonresponders, while those without disease progression and those with
   tumor shrinkage were classified as responders. All biological specimens
   were obtained from the Cancer Institute/Hospital, Peking Union Medical
   College, with the approval of the Research Ethics Committee. None of
   the patients received any preoperative targeted therapy. The resected
   specimens were fixed with formalin and preserved by paraffin embedding.
   Histological diagnosis of HCC was performed by pathologists.
   Hematoxylin and eosin (HE)-stained slides were used to assess tumor
   purity, and 16 HCC tissues with tumor purity> 50% were selected for
   proteomic profiling. The clinical information is shown in [64]Table S1.

2.2. Processing of LC-MS/MS

2.2.1. Protein extraction

   Samples were sequentially soaked in fresh xylene, absolute ethanol, 85%
   ethanol, 70% ethanol, and 50% ethanol with shaking at a constant speed.
   Tissues were removed from the dried glass slides by scraping with 1 ml
   of precooled PBS, left at room temperature for 5 min, and then
   centrifuged at 14000g for 15 min to separate and remove the
   supernatant. Then, 150 µl of freshly prepared lysis buffer (1% sodium
   dodecyl sulfate and 2 mM dithiothreitol (DTT) in 0.1 M Tris-HCl; 10 µl
   protease inhibitor per 1 ml (Roche, REF 04693116001)) was added to the
   pellet and incubated at 99 °C for 30 min. The lysate was naturally
   cooled to room temperature, and subjected to 99 cycles of sonication
   (power 180 W, 1- s on and 2- s off). The lysate was again incubated at
   99 °C for 30 min and naturally cooled to room temperature. Then, the
   lysate was centrifuged at 14000g for 15 min, and the supernatant was
   collected and stored at −80 °C before use.

2.2.2. Protein digestion

   Proteins were digested by filter-assisted sample preparation (FASP)
   [65][16], [66][17]. In brief, each sample was diluted to 1.5 ml with UA
   solution (8 M urea in 0.1 M Tris-HCl), and the supernatant was
   sequentially added to a 30 kD filter for centrifugation. Next, 200 µl
   of UA solution was added, and the sample was washed once by
   centrifugation. Then, 100 µl of DTT solution (20 mM DTT in UA solution)
   was added, and the sample was incubated at 37 °C for 4 h. After the DTT
   solution was removed by centrifugation, 100 µl of IAA solution (50 mM
   iodoacetamide dissolved in UA solution) was added, and the reaction was
   allowed to proceed at room temperature for 30 min in the dark. The
   ultrafractionation tube was washed twice with 200 µl of UA and three
   times with 200 µl of ABC solution (50 mM ammonium bicarbonate in water,
   pH 8.5) by centrifugation. Next, 100 µl of ABC solution containing 1 µg
   of trypsin was added to each filter tube, and the tubes were sealed and
   incubated at 37 °C for 16 h. Peptides were collected by centrifugation
   at 14000g and dried at 45 °C. The peptide mixture was resolved with
   buffer A (0.1% formic acid) and quantified using a Nano-drop OneC
   spectrophotometer.

2.2.3. Liquid chromatography tandem mass spectrometry analysis

   The liquid chromatography-tandem mass spectrometry (LC—MS/MS) detection
   system consisted of a nanoflow high-performance liquid chromatography
   (HPLC) instrument (Easy nLC1200, Thermo Fisher) coupled to an Orbitrap
   Q-Exactive HF mass spectrometer (Thermo Fisher). For acquisition of
   mass spectra, the optimal data-independent acquisition (DIA) parameters
   described in the article by Shuang Weng et al. [67][18] were used. In
   brief, 2 µg of the peptide mixture resolved with buffer A was loaded on
   a 30-cm self-packing column (150-μm inner diameter, ReproSil-Pur
   C18-AQ, 1.9 µm; Dr. Maisch) with a 120-min LC gradient time ([68]Table
   S2) at a flow rate of 600 nl/min. The chromatograph peak width was
   18 s. A standard amount of iRT spike-in was added to ensure
   calibration. The MS1 survey scan range was 400–1200 m/z, and the MS1
   scan was followed by 32 MS2 scans with overlapping sequential precursor
   isolation windows (25 m/z isolation width, 1 m/z overlap). For the MS2
   survey scan, the automatic gain control (AGC) target was set to 3e6,
   and the maximum injection time (IT) was set to 45 ms.

2.2.4. Protein identification by Spectronaut-based database searching

   The raw data files obtained from the mass spectrometry analysis were
   imported into SpectronautTM software for processing, and the spectral
   library was selected from the human liver cancer proteome spectral
   library established in the article by Shuang Weng et al. [69][18]. The
   false discovery rate (FDR) threshold at the protein and peptide levels
   was set to 0.01. Peak areas at the MS2 level were selected for protein
   and peptide quantitation. Subsequent in-depth processing was performed
   using data packages in R software.

2.3. Proteomic data analysis

2.3.1. Sample quality control and data normalization

   Boxplots were used to visualize the numbers of identified components at
   three levels: protein group, peptide, and precursor. No sample
   exhibited an extremely high or low number of identified components
   ([70]Fig. S1), and all samples were included in subsequent analysis.
   Protein expression matrices were quantile normalized using the limma
   package (v 3.42.2) in R/Bioconductor. We further processed the
   expression matrix using the removeBatchEffect() function of limma (v
   3.52.4), and subsequent analysis was limited to the 5209 genes
   expressed in at least 50% of the samples ([71]Table S3). For algorithms
   that could not handle missing values, we used mice (v 3.15.0) [72][19]
   to fill in the missing values by a multiple interpolation method.

2.3.2. Bioinformatics analysis of proteome data

   Analysis of differential protein expression between nonresponders and
   responders was performed using limma (v 3.52.4) [73][20], with the
   following criteria: p ≤ 0.05 and |logFC|≥ 1. We then used
   clusterProfiler (v 4.4.4) [74][21] to perform Kyoto Encyclopedia of
   Genes and Genomes (KEGG) pathway enrichment analysis and gene set
   enrichment analysis (GSEA) on the differentially expressed protein,
   GSVA (v 1.44.5) to perform single-sample gene set enrichment analysis
   (ssGSEA) [75][22], Xcell (v 1.1.0) for immune infiltration analysis
   [76][23], ConsensusClusterPlus (v 1.60.0) for consensus clustering
   [77][24], and surv (v 3.4–0) and survminer (v 0.4.9) for survival
   analysis. We also obtained data from related studies by Jia Fan et al.
   and Fuchu He et al. for further validation [78][17], [79][25].
   Statistical analyses were performed on R version 4.2.1. For analysis of
   paired group data, the Wilcoxon signed-rank test was used. Statistical
   significance was accepted at p < 0.05 or a Benjamini–Hochberg adjusted
   p value < 0.05.

2.4. Immunohistochemistry (IHC) and Scoring

   Sections were baked at 65 °C for 4 h, deparaffinized with xylene and
   ethanol, and then incubated with 3% H[2]O[2] for 10 min in the dark to
   quench endogenous peroxidase activity. After antigen retrieval by
   heating in citrate repair solution (pH= 6.0) or EDTA buffer (pH= 9.0)
   in a microwave oven for 10 min, the sections were blocked with goat
   nonimmune serum and incubated with primary antibodies against ADH1A
   (1:1000, Abcam), STING1 (1:10000, Proteintech), CD163 (1:400, Cell
   Signaling), CD8 (1:200, Cell Signaling) and HLA_DRB (1:2000, Abcam)
   overnight at 4 °C. After incubation with secondary antibodies, a
   chromogenic reaction was performed using the DAB kit. The sections were
   then counterstained with hematoxylin. Staining was analyzed and scored
   by two experienced pathologists. Samples were scored on a scale from 0
   to 3 (0, negative; 1, weakly positive; 2, moderately positive; 3,
   strongly positive) based on the intensity of staining and the
   percentage of positive staining. The staining intensity score was
   multiplied by the percentage of the slices exhibiting staining for the
   object of interest to calculate the “Hscore” of the object of interest.
   The details of the reagents used are shown in [80]Table S4.

3. Results

3.1. Global proteomic profiling of sorafenib-nonresponding and
sorafenib-responding HCC patients

   To study the characteristics of patients with HCC exhibiting different
   treatment effects of sorafenib, we performed proteomic mass
   spectrometry on formalin-fixed paraffin-embedded HCC tissue samples
   from 16 patients before sorafenib treatment ([81]Fig. S2A). Among the
   patients, 7 responded to sorafenib treatment, and the other 9 who did
   not. By using Spectronaut software for database searching, only the
   proteins that were reliably quantified (Q value<0.01) were output; a
   total of 7177 proteins were quantified, with an average of 5139
   proteins quantified per sample ([82]Fig. S2B). Across the samples, 3063
   proteins were quantified in all samples. We examined the distribution
   of all protein abundances, and found that it was unimodal for all 16
   samples ([83]Fig. S2C). After data normalization, each sample was
   clustered with the same distribution ([84]Fig. S2D).

3.2. Differential proteomic analysis of the sorafenib-nonresponding and
sorafenib-responding groups of HCC patients

   A standard spartial least squares regression analysis approach (PLS_DA)
   was used to classify samples from the sorafenib-related proteomic
   profile data and the drug-nonresponding and drug-responding groups were
   found to be distinguishable ([85]Fig. 1A), with a significant
   difference in protein expression characteristics between the two
   groups. Differential analysis identified a total of 29 differentially
   expressed proteins, 14 of which were upregulated in the
   drug-nonresponding group and 15 in the responding group ([86]Fig. 1D).
   Among these proteins, stimulator of interferon response cGAMP
   interactor 1 (STING1) ([87]Fig. 1B, P = 0.0310), which was upregulated
   in the nonresponding group, and alcohol dehydrogenase 1A (ADH1A)
   ([88]Fig. 1C, P = 0.1700), which was upregulated in the responding
   group were the core proteins selected for our exploration of
   pretreatment factors affecting sorafenib resistance. The STING1 protein
   is an essential regulator of autophagy [89][26], [90][27], while the
   ADH1A protein is an important member of an enzyme family involved in
   biotransformation [91][25].

Fig. 1.

   [92]Fig. 1
   [93]Open in a new tab

   Differential analysis of sorafenib protein profile data. (A) PLS-DA can
   distinguish between the drug-nonresponding and drug-responding groups.
   (B) and (C) Expression of STING1 and ADH1A, as measured by mass
   spectrometry, in the two groups. (D) Volcano plot showing the
   differentially expressed proteins identified.

3.3. Autophagy-related and biotransformation-related pathways differ between
the responding and nonresponding groups

   KEGG enrichment analysis of the screened identified expressed proteins
   revealed that the ascorbate and aldarate metabolism, retinol
   metabolism, pentose and glucuronate interconversions, and glutathione
   metabolism pathways which are associated with biotransformation
   [94][28] were enriched in the responding group, while the lysosome and
   cytosolic DNA sensing pathway, which are associated with autophagy,
   were enriched in the drug-nonresponding group ([95]Fig. 2A, [96]Table
   S5-6) [97][29]. The protein expression profile data were converted to
   single-sample pathway enrichment scores using ssGSEA, and the Wilcoxon
   signed-rank test was used to evaluate the significance of differences
   between the two groups. Annotation via c2.cp.kegg.v2022.1.
   Hs.symbols.gmt revealed that the glycolysis gluconeogenesis and
   nicotinate and nicotinamide metabolism pathways were enriched in the
   responding group. In contrast, the cytosolic DNA sensing, VEGF
   signaling, and mTOR signaling pathways were enriched in the
   nonresponding group, and there was no difference in MAPK signaling
   pathway enrichment between the two groups ([98]Table S7). Further
   annotation by c2.cp.reactome.v2022.1. Hs.symbols.gmt and
   c2.cp.wikipathways.v2022.1. Hs.symbols.gmt revealed that the
   PI3K/AKT/mTOR pathway was significantly enriched in the nonresponding
   group ([99]Fig. 2B, [100]Table S8-9). It has been reported that mTOR
   inhibits the transcription of ADH1A by phosphorylating histone
   deacetylase 1 (HDAC1) [101][30], and our analysis in mass spectrometry
   data revealed a trend of upregulation of HDAC1 in the drug-
   nonresponding group ([102]Fig. S4A, P = 0.3000). GSEA of the pentose
   and glucuronate interconversions pathway, representing
   biotransformation, and the lysosome pathway, representing autophagy,
   also demonstrated significant differences between the two groups in
   these two aspects ([103]Fig. 2C).

Fig. 2.

   [104]Fig. 2
   [105]Open in a new tab

   Pathway enrichment analysis. (A) KEGG enrichment analysis revealed the
   upregulation of autophagy-related pathways in the nonresponding group
   and the upregulation of biotransformation-related pathways in the
   responding group. (B) ssGSEA was performed to convert expression
   profiles into pathway enrichment scores, and the Wilcoxon signed-rank
   test was performed to identify differential pathways between the two
   groups. (C) GSEA of representative pathways for biotransformation and
   autophagy. The heatmap on the right shows the differentially expressed
   proteins in the two groups.

3.4. Immune infiltration analysis suggests immune potential in the
nonresponding group

   Analysis of differences in the infiltration of each cell type in the
   drug-nonresponding and drug-responding groups using Xcell revealed that
   CD8 + T cells ([106]Fig. 3A, P = 0.0638), natural killer T (NKT) cells
   ([107]Fig. 3A, P = 0.0128), M2 macrophages ([108]Fig. 3A, P = 0.0108),
   gamma delta T (Tgd) cells ([109]Fig. 3A, P = 0.0258), B cells
   ([110]Fig. 3A, P = 0.0164), endothelial cells ([111]Fig. 3A,
   P = 0.0311) and pericytes ([112]Fig. 3A, P = 0.0164) were highly
   infiltrated in the nonresponders, and that CD4 + T cells ([113]Fig. 3A,
   P = 0.0195) were highly infiltrated in the responders ([114]Table S10).
   The same is true for the content and proportional structure of the
   above cells in the heatmap and stacking plot between the two groups
   ([115]Fig. 3B-C). Innate immunocytes were exhibited significantly high
   infiltration in the nonresponding group. Interestingly, both
   immunosuppressive and immunoreactive cells were also highly infiltrated
   in the nonresponding group.

Fig. 3.

   [116]Fig. 3
   [117]Open in a new tab

   Immune infiltration analysis. (A) Box plot showing the differences
   between the drug-nonresponding and drug-responding groups for each
   infiltrating cell type. (B) and (C) The heatmap and stacked plot show
   the abundance of each infiltrated cell type in each sample.

3.5. Immunohistochemical analysis shows a trend consistent with the immune
infiltration analysis

   The Hscore was defined as the staining intensity multiplied by the
   percentage of the section exhibiting staining for the object of
   interest to obtain the Hscore of each molecule in each sample.
   Intergroup comparisons between the responding and nonresponding groups
   were performed by the Wilcoxon signed-rank test. There were three
   different expression patterns for immunihistochemical markers in
   enrolled 16 samples in our study. Firstly, high stainging rario and
   intensity of ADH1A in HCC tumor cells in the responders but not the
   nonresponders ([118]Fig. 4A, P = 0.13). Secondly, there is a marker
   with insignificant difference of expression levels in the two groups,
   CD163 ([119]Fig. 4C, P = 0.62). And thirdly, relatively high expression
   level of STING ([120]Fig. 4B, P = 0.11), human leukocyte antigen-DRB
   (HLA_DRB, on both lymphocytes and tumor cells) ([121]Fig. 4E, P = 0.22;
   [122]Fig. 4F, P = 0.19) and CD8 ([123]Fig. 4D, P = 0.15), were detected
   in the nonresponders, the first two molecules exhibit diffuse and
   strong positive signals.

Fig. 4.

   [124]Fig. 4
   [125]Open in a new tab

   IHC staining images and differences in the Hscore between the two
   groups. (A)-(F) Schematic representation of IHC staining results and
   the difference in the Hscore between the two groups, with positive
   staining shown by the tan color located in cell cytoplasm (ADH1A in the
   responding group, STING1), cell membrane and cytoplasm (CD163 and
   HLA_DRB), and cell membrane (CD8). The negative results of the above
   immunohistochemical detection were the only blue signals in cell nuclei
   stained by hematoxylin (eg. ADH1A in the nonresponding group).

   HLA_DRB is a crucial component of MHC-II proteins [126][31], [127][32],
   and its expression level represents, to some extent, the degree of
   immune activation [128][33]. In the mass spectrometry data, we also
   found that HLA_DRB1 showed an upregulation trend in the nonresponders
   (P = 0.1700, [129]Fig. S4A). Unfortunately, the sample size was
   limited, and the p-values for the intergroup comparisons of each
   molecule indicated that the differences were not significant. However,
   the trend was consistent with that found in the previous analysis. The
   Hscore for each molecule can be found in [130]Table S11.

3.6. Molecules and pathways associated with sorafenib resistance correlate
with the malignant features of HCC

   We obtained the dataset for hepatitis B virus (HBV)- associated HCC
   published by Jia Fan et al. [131][25] and the dataset for early-stage
   HCC published by Fuchu He et al. [132][17] to increase the clinical
   applications values of molecules and pathways screened based on the
   sorafenib response/nonresponse status. In the HBV-associated HCC
   dataset, we transformed the proteomic expression profile data into
   pathway enrichment scores by ssGSEA and selected the pathways that were
   consistent with the specific differential pathways screened based on
   sorafenib response/nonresponse as the criteria for consensus
   clustering. Visually, the consensus matrix with a k value of 2 seemed
   to show the cleanest separation between clusters ([133]Fig. S3A). The
   consensus cumulative distribution function (CDF) and delta area plots
   show that the area under the consensus CDF does not increase
   significantly as k increases from a value of 2 ([134]Fig. S3B). The
   enrichment score of each pathway in each sample is shown on a heatmap.
   The tumor samples were generally classified as cluster 1, and the
   normal paracancer samples were generally classified as cluster 2. In
   addition, the pathways related to the drug nonresponse were enriched in
   the tumor samples and cluster 1, while the pathways related to the drug
   response were enriched in the normal samples and cluster 2 ([135]Fig.
   5A, [136]Table S12). To determine cutoff values for survival analysis,
   clinical data from the early-stage HCC dataset were analyzed separately
   for STING1 protein expression and lysosome and cytosolic DNA sensing
   pathways, which represent autophagy, and for ADH1A protein expression
   and the pentose and glucuronate interconversions and ascorbate and
   aldarate metabolism pathways, which represent biotransformation
   ([137]Fig. 5B-G). According to ADH1A ([138]Fig. 5B, P = 0.014), pentose
   and glucuronate interconversions, ascorbate and aldarate metabolism and
   cytosolic DNA sensing pathways ([139]Fig. 5D. [140]Fig. 5D, P = 0.039;
   [141]Fig. 5F, P = 0.037; [142]Fig. 5G, P = 0.0037) showed significant
   differences in overall survival (OS) in patients with early-stage HCC
   between the high and low expression groups. After grouping according to
   the expression levels of STING1 ([143]Fig. 5C, P = 0.075) and lysosome
   pathway ([144]Fig. 5E, P = 0.059), the OS between the two groups also
   approached significant differences. High expression levels of the
   STING1 protein and activation levels of the lysosome and cytosolic DNA
   sensing pathways, which represent autophagy, were associated with poor
   prognosis. High expression levels of the ADH1A protein and activation
   levels of the pentose and glucuronate interconversions and ascorbate
   and aldarate metabolism pathways, which represent biotransformation,
   were associated with good prognosis.

Fig. 5.

   [145]Fig. 5
   [146]Open in a new tab

   Additional datasets were obtained for further validation. (A) Consensus
   clustering of data for HBV-associated HCC using differential pathways
   identified from our sorafenib data, the heatmap of the enrichment
   scores for each pathway in each sample, and annotation of clustering
   information, tissue origin, and corresponding pathways in the
   nonresponding/responding groups. (B)-(G) Survival analysis of
   differential molecules and pathways between the sorafenib
   nonresponding/responding groups in the data for early-stage HCC.

4. Discussion and conclusions

   As a first-line treatment for advanced HCC, sorafenib has a worse than
   expected patient benefit rate [147][34]. We explored how pretreatment
   protein expression patterns and activation of signaling pathways affect
   sorafenib resistance by proteomic characterization of pretreatment
   samples from sorafenib responders and nonresponders.

   Among the differentially expressed proteins, we identified two key
   molecules, ADH1A and STING1. In the pathway enrichment analysis,
   autophagy-related pathways were enriched in the nonresponding group,
   while biotransformation-related pathways were enriched in the
   responding group. STING1, a core molecule mediating autophagy, is
   activated by the cell membrane DNA sensor cyclic GMP-AMP synthase
   (CGAS), and its acute activation facilitates antitumor therapeutic
   efficacy, while its chronic activation may mediate inflammation and
   promote tumor survival [148][27]. On the other hand, lysosomes are the
   central organelles of autophagy, meeting the energy needs of cells by
   degrading damaged organelles and long-lived proteins and attenuating
   cell death caused by endoplasmic reticulum (ER) stress [149][14].

   The pathways enriched in the responding group are mainly related to the
   phase II reactions of biotransformation. Glucuronide in the pentose and
   glucuronate interconversions pathway plays an important role in
   conjugates detoxification of containing toxic substances [150][35];
   L-ascorbic acid, in the ascorbate and aldarate metabolism pathway, is a
   potent antioxidant [151][36]. Both pathways were found to be
   significantly abnormal in a study of oxidative damage in the pancreas,
   suggesting that both pathways play a crucial role in resistance to
   oxidative stress [152][28], [153][36]. Downregulation of specific
   biotransformation pathways can cause substrates to be shunted into the
   remaining pathways and result in damage. For example, ADH1A, a key
   enzyme in the metabolism of alcohol and retinol, is coupled to
   dehydrogenation of the cofactor NAD when it metabolizes its substrate
   and therefore does not produce oxygen radicals [154][37]. This
   observation explains the enrichment of nicotinate and nicotinamide
   metabolism in the responding group ([155]Fig. 2B). When ADH1A is
   downregulated retinol becomes a substrate for P450s and generates
   oxygen radicals, as confirmed by a study that constructed an ADH1A-/-
   mouse model after feeding vitamin A found increased liver damage
   [156][37], [157][38]; ethanol induces lipid peroxidation through the
   microsomal pathway, and its metabolic end products interfere with
   normal cellular metabolism, for example, Ciuclan L et al. found that
   lipid deposition, increased oxidative stress, and enhanced alcohol
   toxicity in cells treated with alcohol after knockdown of ADH1A in
   hepatocytes [158][39], [159][40]. Therefore, the reduced
   biotransformation capacity increases hepatic oxidative damage to some
   extent, and oxidatively damaged DNA (e.g., 8-OHG-modified DNA) is a
   potent autophagy activator [160][29]. Therefore, we suggest that the
   reduced biotransformation capacity in the drug-nonresponding group
   partially promotes active autophagy in this group.

   Pathway enrichment analysis also revealed activation of the redundant
   PI3K/AKT/mTOR pathway in the nonresponding group, with similar degrees
   of Ras/Raf/MAPK pathway activation in both groups. Therefore, we
   believe that in the nonresponding group, sorafenib's target in the
   pathway is successfully bypassed via promotion of angiogenesis through
   alternative pathways prior to treatment, a hypothesis consistent with
   the enrichment of VEGF signaling in the nonresponding group ([161]Fig.
   2B), may partially explain the abundance of endothelial and pericytes
   in the drug-nonresponding group observed by immuno-infiltration
   analysis. Thus, poor efficacy is to be expected. As previously
   described, mTOR represses ADH1A transcription by phosphorylating
   histone deacetylase 1 (HDAC1) ([162]Fig. S4B) [163][30] and that mTOR
   inhibits autophagy [164][27]. Interestingly, it has been reported in
   cell line studies of both hepatocellular carcinoma and renal cell
   carcinoma that after inhibition of AKT and thus mTOR, autophagy was
   strongly activated and caused autophagy-dependent cell death (ADCD) in
   sorafenib-resistant cells [165][11], [166][41]. Combining the results
   of redundant activation of the PI3K/AKT/mTOR pathway and downregulation
   of ADH1A in the nonresponding group and related mechanisms reported by
   other researchers, we propose a hypothetical model: a balance between
   STING1-mediated autophagy activation and PI3K/AKT/mTOR
   activation-induced autophagy inhibition is maintained in the
   nonresponding group to maintain the intensity of autophagic activity at
   a level that reduces ER stress and meets the energy requirements to
   promote tumor survival. Besides, activation of the redundant
   PI3K/AKT/mTOR pathway not only bypasses the target of sorafenib but
   also interferes with the decrease in the biotransformation capacity and
   the activation of autophagy in the nonresponding group. Collectively,
   the results of the immune infiltration analysis and immunohistochemical
   staining, indicated that innate immunocytes, immunosuppressive cells,
   and immunoreactive cells were abundant in the nonresponding group.
   Interestingly, the IHC staining intensity of CD163, a marker of M2
   macrophages, was particularly nonsignificantly different between the
   two groups, suggesting that the difference in the abundance of
   infiltrating immunosuppressive M2 macrophages between the two groups
   may have been exaggerated in the immune infiltration analysis. STING
   signaling can regulate the NOD-like receptor signaling pathway through
   NOD-like receptor thermal protein domain associated protein 3 (NLRP3)
   to activate innate immunity [167][27], [168][42]. Simultaneously, the
   resulting induction of chronic inflammation along with oxidative damage
   following the decrease in biotransformation capacity results in
   recruitment of tumor-associated macrophages (TAMs) [169][27],
   [170][43]. However, in addition, the products of autophagy may contact
   antigen-presenting cells (APCs) as antigens or even be presented on the
   surface of tumor cells as endogenous proteins by MHC Ⅱ molecules to
   activate adaptive immunity [171][32], [172][44] as partially confirmed
   by immunohistochemical staining of HLA_DRB in lymphocytes and tumor
   cells ([173]Fig. 4E-F). Combining the results of immune infiltration
   analysis and immunohistochemistry and the reported mechanisms related
   to autophagy affecting tumor immunity, we suggests that the benefit of
   immunotherapy in sorafenib-nonresponding patients with autophagy
   activation is worth exploring. Analysis of the extended dataset
   demonstrated that sorafenib response/nonresponse-related autophagy and
   biotransformation are also malignant features that distinguish HCC
   cells from normal hepatocytes and that the STING1 protein, the lysosome
   pathway, and the cytosolic DNA sensing pathway representing autophagy
   and the ADH1A protein, the pentose and glucuronate interconversions
   pathway, and the ascorbate and aldarate metabolism pathway representing
   biotransformation, have some prognostic value in HCC.

   In summary, the degree of redundant PI3K/AKT/mTOR pathway activation,
   the biotransformation capacity, and the autophagy level in HCC patients
   prior to sorafenib treatment are likely to affect their sensitivity to
   sorafenib. These three aspects might interact mechanistically to
   promote tumor cell survival ([174]Fig. 6). Patients with autophagy
   activation may benefit from immunotherapy in the setting of sorafenib
   resistance, while approaches that interfere with autophagic
   homeostasis, such as modulation of redundant pathways, may reverse
   resistance via multiple mechanisms, and enhancing biotransformation in
   HCC patients may contribute to prolonged survival. These molecules and
   pathways are also potential markers for predicting sorafenib efficacy.

Fig. 6.

   [175]Fig. 6
   [176]Open in a new tab

   Mechanisms by which some pretreatment factors influence sorafenib
   resistance.

Ethical approval

   Ethical approval was obtained from the Ethics Committee of Cancer
   Hospital, Chinese Academy of Medical Sciences (#20/455–2651).

Funding

   This work was supported by the National Key Research and Development
   Program of China (2022YFE0103600) and Research funding from Cancer
   Hospital, Chinese Academy of Medical Sciences (LC2020A12).

CRediT authorship contribution statement

   Ruiqi Zheng: Conceptualization, Software, Formal analysis,
   Investigation, Data curation, Writing – original draft, Writing –
   review & editing. Shuang Weng: Software, Formal analysis, Writing –
   original draft, Writing – review & editing. Jianping Xu: Methodology,
   Writing – review & editing. Zhuo Li: Writing – review & editing. Yaru
   Wang: Writing – review & editing. Zulihumaer Aizimuaji: Writing –
   review & editing. Sheng Ma: Writing – review & editing. Linlin Zheng:
   Writing – review & editing. Haiyang Li: Writing – review & editing.
   Wantao Ying: Resources, Writing – review & editing. Weiqi Rong:
   Resources, Writing – review & editing. Ting Xiao: Conceptualization,
   Resources, Writing – review & editing, Supervision, Project
   administration, Funding acquisition. All authors have read and approved
   the final manuscript.

Declaration of Competing Interest

   The authors declare no competing interests.

Acknowledgements