Elucidation of Mechanisms Underlying Proteasome Addiction in Sarcomas

1. Introduction

Sarcomas are a broad group of over 70 types of neoplasm originating in mesenchymal tissues derived from embryonic mesoderm layer cells.[1,2] Approximately 80% of sarcomas originate in soft tissues including connective tissues, blood vessels, fat, and smooth and skeletal muscle. The remaining 20% arise in bone or cartilage.[3] Age of onset varies with sarcoma type, but cases are most prevalent in children, adolescents, and young adults under 30 years of age.[4,5] Roughly 15% of all cancers in children are sarcomas.[6] Roughly 17,000 new cases of sarcoma were diagnosed in the United States in 2023, with approximately 13,460 of these being soft tissue sarcomas.[7] Approximately 7,280 deaths occurred in the United States in 2023 due to all types of sarcomas.[8] First line treatment for sarcomas is typically surgery with or without chemotherapy and/or radiation therapy.[9] In gastrointestinal stromal tumors, imatinib (which inhibits c-KIT) is the first line treatment option.[9] On a case-by-case basis, for locally advanced or metastatic sarcomas, targeted therapeutics or immunotherapies may be administered as first line treatment.[10,11] Five-year survival rates for soft tissue sarcoma are 81% for localized, and 56% and 16%, respectively, for regional and distant metastatic. Definitive diagnosis is often slow due to the rarity of these cancers, subtle slowly progressive symptoms, and symptoms being similar to those of less serious conditions^.[12] Delays in diagnosis result in low rates of treatment at localized stage, reflected in the 65% overall 5-year survival rate, making sarcomas among the deadliest of rare cancers and illustrating the need for effective new therapy options.

The proteasome inhibitor bortezomib received FDA approval for treatment of multiple myeloma in 2003 and for treatment of mantle cell lymphoma in 2006. Investigations, including clinical trials, are underway to characterize the efficacy of proteasome inhibition in non-small cell lung carcinoma, breast, prostate, and ovarian cancer, and various other solid tumor types. A recent Ewing sarcoma cell-based screen for small molecules that could inhibit proliferation revealed a class of benzyl-4-piperidone compounds that can selectively inhibit proliferation of multiple Ewing sarcoma cell lines. These compounds were shown to inhibit 19S proteasome function by inhibiting its integral deubiquitinating enzymes USP14 and UCHL5.[13] A genome-wide gene knockdown screen using shRNA also indicated specific proteasome dependence in Ewing sarcoma. Treatment of osteosarcoma cells with bortezomib inhibits cell proliferation and induces apoptosis via a mechanism involving upregulation of the osteoblastic differentiation factor Runx2.[14] Proteasome activator complex subunit 2 (PSME2) inhibits osteosarcoma cell proliferation, migration, and invasion, suggesting another mechanism of proteasome addiction.[15] Some evidence also exists for proteasome addiction in soft tissue sarcomas. The proteasome inhibitor MG132 dose dependently increases apoptosis and antagonizes proliferation in 3 uterine leiomyosarcoma cell lines and induces G2/M phase arrest in 2 of these cell lines,^.[16] and high nuclear expression of proteasome activator complex subunit 1 (PSME1) is associated with poor survival in soft tissue leiomyosarcoma patients.[17]

To better understand mechanism(s) of proteasome addiction in soft tissue sarcomas, we investigated the impact of each of the 377 ubiquitin ligases in the list curated by the Epithelial Systems Biology Laboratory of the National Heart Lung and Blood Institute [18] on overall survival among a cohort of soft tissue sarcoma patients (259 samples from 206 patients) using data from the cancer genome atlas (TCGA). This cohort includes 80 leiomyosarcoma (53 soft tissue and 27 uterine), 50 dedifferentiated liposarcoma, 44 undifferentiated pleomorphic sarcoma, 17 myxofibrosarcoma, 10 synovial sarcoma, and 5 malignant peripheral nerve sheath tumor patients.[19] For those E3 ligases whose expression level correlated significantly with overall survival and altered survival by >40 months, we identified sets of known and predicted substrates using Ubibrowser 2.0[20] and performed enrichment analyses to learn what cellular functions are potentially modulated by proteasomal activity guided by relevant E3 ligases. These analyses were limited to E3 ligases with both significant correlation with survival impact and altered survival by >40 months to focus narrowly on E3 ligases with both significant and robust effects.

2. Materials and Methods

2.1. Kaplan-Meier Overall Survival Analysis

To identify E3 ubiquitin ligases that influence patient survival, we used Kaplan-Meier Plotter[21,22] to produce survival curves for sarcoma patients with high or low expression of each E3 human ubiquitin ligase on the list curated by the epithelial systems biology laboratory (ESBL) in the National Heart Lung and Blood Institute (See supplemental document 1).[18] Kaplan Meier Plotter curates patient data from GEO, EGA, and TCGA databases, including TCGA-housed data from a study of 259 soft tissue sarcoma samples[19]. Cutoffs were empirically determined by assessing all cutoffs between top and bottom quartiles and selecting that cutoff with the highest significance. Patient data includes mRNA-seq based gene expression data, relapse free survival, and overall survival. As a validation study, we performed independent Kaplan Meier analyses of the impact of each E3 ubiquitin ligase found to significantly increase or decrease OS or RFS by >40 months using overall and relapse-free survival data for the 206 patients studied in Abehouse et. al[19]. Median transcripts per million (TPM) values for each assessed E3 ligase were set as cutoff to split tumor samples into “High” and “Low” expressing groups. The R package “survminer”[23] was used to generate Kaplan Meier plots. P-values were generated using log-rank tests. Validated E3 ligases are indicated with asterisks in column 1 of tables 1-4 (* P≤0.05, ** P≤0.01, *** P≤0.001).

2.2. Violin Plot Analyses

To determine whether high and low E3 ligase expression was segregating sarcoma types, we performed violin plot analyses to visualize expression of each E3 ligase in each patient with patients grouped in columns by sarcoma type (DDLPS = dedifferentiated liposarcoma; MFS = myxofibrosarcoma; MPNST = Malignant Peripheral Neural Sheath Tumor; SS = Synovial Sarcoma; STLMS = Soft tissue Leiomyosarcoma; ULMS = Uterine Leiomyosarcoma; UPS = Undifferentiated Pleomorphic Sarcoma).

2.3. Identification of Known and Predicted E3 Ligase Substrates

We employed UbiBrowser 2.0[20] to identify known and putative substrates of each E3 ubiquitin ligase whose up or downregulation was found to correlate significantly with an increased overall survival of greater than 40 months. We then manually filtered out duplicates and substrates found in both datasets.

2.4. Pathway Enrichment Analyses

Two sets of proteins were generated using UbiBrowser 2.0 that represent potential substrates uniquely associated with E3 ubiquitin ligases for which upregulated expression correlates with increased overall patient survival, and E3 ubiquitin ligases for which upregulated expression correlates with decreased overall patient survival. Each set of proteins was manually filtered to remove shared genes and subjected to enrichment analysis using Reactome.[25,26] Reactome curates 15,886 reactions that function within 2803 human pathways. The pathways are hierarchically organized under 29 broad pathway categories. These categories are autophagy, cell cycle, cell-cell communication, cellular responses to stimuli, chromatin organization, circadian clock, developmental biology, digestion and absorption, disease, DNA repair, DNA replication, drug ADME, extracellular matrix organization, gene expression, hemostasis, immune system, metabolism, metabolism of proteins, metabolism of RNA, muscle contraction, neuronal system, organelle biosynthesis and maintenance, programmed cell death, protein localization, reproduction, sensory perception, signal transduction, transport of small molecules, and vesicle mediated transport.

2.5. Statistical Analyses

Kaplan-Meier plotter uses a PostgreSQL server to integrate gene expression and clinical data. For any given gene of interest, the relevant set of patient records is split into high and low expression groups. A Kaplan-Meier survival plot is then created and a hazard ratio using 95% confidence intervals and logrank P value are calculated. In enrichment analyses using Reactome, p-values associated with enriched functions are calculated using Fisher’s exact test.

3. Results

3.1. Impacts of E3 Ubiquitin Ligases on Patient Overall Survival

We identified subsets of ubiquitin ligases for which high level expression was significantly associated with extended and shortened overall survival (OS). Representative Kaplan Meier plots for TTC3 and VPS18 (the ubiquitin ligases with the largest negative and positive respective impacts on OS) are presented in Figure 1 A and B. Among the 377 human E3 ubiquitin ligases analyzed, 102 (27%) ligases revealed significant association between high expression and reduced OS, with p values ranging from 0.05 to 5.2 X 10^-8, and impacts on survival ranging from 6.2 to 50.6 months (median 31.6 ± 9.2). Thirteen of these (see Table 1) shorten OS by >40 months. Eight of these 13 were validated as still significant when using median TPM as cutoff. Violin plot analysis of these 13 E3 ligases indicates that in most cases, each sarcoma histology type would contribute patients to both high and low expression groups (Figure 2). Two exceptions are 1) relatively high CCNB1IP1 expression in all synovial sarcoma patients, and 2) relatively low PCGF3 expression in all myxofibrosarcoma patients. Violin plotting also reveals very low expression of TRIM71 and MARCHF10 in all patients, with median TPM below 2 in both cases. Another 73 (19.4%) revealed significant association between high expression and extended OS, with p values ranging from 0.049 to 9.4 X 10^-6, and impacts on survival ranging from 3.1 to 50.8 months (median 32.4 ± 9.1). Eighteen of these (see Table 2) extend OS by >40 months. Only two of these (RMND5B and TRIM21) were validated as still significant when using median TPM as cutoff. Violin plot analysis of these 18 E3 ligases indicates that in most cases, each sarcoma histology type would contribute patients to both high and low expression groups (Figure 3). Eight of these 18 E3 ligases display uniformly low expression in synovial sarcoma, and SYVN1 displays uniformly high expression in MPNST.

3.2. Impacts of E3 Ubiquitin Ligases on Patient Recurrence Free Survival

In total, 102 E3 ligases displayed a positive correlation between elevated expression and diminished OS, and 73 E3 ligases displayed a positive correlation between elevated expression and enhanced OS. We assessed each of these E3 ligases for the impact of its expression level on recurrence free survival (RFS). Among the E3 ligases whose elevated expression correlates with diminished OS, 15 (15%) also significantly correlate with diminished RFS (see Table 3). Among these, nine reduce median RFS by 56-75 months, and 6 reduce median survival by 3-10 months. Three of the nine E3 ligases with large survival impacts (RNF26, TRAF4, and UHRF1) were validated as still significant when using median TPM as cutoff. Violin plots indicate that in most cases, each sarcoma histology type would contribute patients to both high and low expression groups (Figure 4). RAD18 is uniformly low in synovial sarcoma, and MEX3C and MKRN2 are uniformly high in MPNST. Among the E3 ligases whose elevated expression correlates with enhanced OS, 12 (16%) also significantly correlate with enhanced RFS (see Table 4). Among these, 11 extend median RFS by 48.8-72.1 months (TRIM8 extends RFS by 1.9 months). Only one of the eleven E3 ligases with large survival impacts (RMND5B) was validated as still significant when using median TPM as cutoff. Violin plots indicate that in most cases, each sarcoma histology type would contribute patients to both high and low expression groups (Figure 5). RNF130 expression is skewed high in MPNST, and TRIM8 expression is skewed low in synovial sarcoma. Representative Kaplan Meier plots for MNAT1 and TRIM10 (the ubiquitin ligases with the largest negative and positive respective impacts on RFS) are presented in Figure 1 C and D.

3.3. Putative Substrate Pools of E3 Ligases That Significantly Impact Overall Survival

For each of the 13 E3 ligases whose expression is significantly associated with shortened OS and that truncate survival by >40 months, known and predicted substrates were identified using Ubibrowser 2.0. Duplicate proteins were eliminated, resulting in a pool of 1636 plausible substrates (82 known and 1554 predicted). For each of the 18 E3 ligases whose elevated expression is significantly associated with extended OS and that extend survival by >40 months, known and predicted substrates were identified using Ubibrowser 2.0. Duplicate proteins were eliminated, resulting in a pool of 4339 plausible substrates (118 known and 4221 predicted). To see functions uniquely enriched in each pool of substrates, we filtered each pool to remove substrates present in common, resulting in a pool of 393 substrates unique to those E3 ligases whose elevated expression is associated with shortened OS, and 3238 substrates unique to the E3 ligases whose elevated expression is associated with extended OS (supplemental tables 1 and 2). Removal of substrates of TRIM71 and MARCHF10 due to their very low expression in all patients reduces the pool of substrates unique to those E3 ligases whose elevated expression is associated with shortened OS from 393 to 275.

3.4. Enriched Functions Associated with E3 Ligases Whose Elevated Expression Is Associated with Diminished OS

Elevated expression of the E3 ubiquitin ligases PCGF3, CCNB1IP1, MSL2, RNF2, MARCHF5, ZBTB12, MEX3D TRIM71, UBE4B, RNF8, MEX3A, MARCHF10, and TTC3 (see Table 1) is associated with significantly (p = 0.029 to 5.2 X 10^-8; HR = 1.64-2.78) shortened OS (40.1 to 50.6 months), suggesting that these E3 ligases make a proportionately large contribution to proteasome addiction in Ewing Sarcoma. UbiBrowser 2.0 identified or predicted a pool of 393 substrates of these E3 ligases that are not substrates of the 18 E3 ligases whose elevated expression is associated with increased OS. Excluding functions related to infectious disease, 118 enriched functions had p values ≤1 X 10^-8. Among the 118 enriched functions with p values ≤1 X 10^-8, the most highly represented pathway category at the top of the Reactome hierarchy is gene expression (31 enriched functions), followed by immune system (20 enriched functions), DNA repair (16 enriched functions), signal transduction (11 enriched functions), developmental biology (6 enriched functions), chromatin organization (5 enriched functions), cellular responses to stimuli, disease, DNA replication, and metabolism of proteins (4 enriched functions each), and reproduction and vesicle mediated transport (1 enriched function each). This data is represented in pie chart form in Figure 6A. The 25 most significant results of enrichment analysis using Reactome (from all hierarchical levels) are presented in Table 5. Enrichment analysis of the 275 putative E3 ligase substrates excluding those of TRIM71 and MARCHF10 yielded the same top 13 enriched functions in the same order. Another 10 of the top 25 enriched functions in the 393-substrate set remain present with slightly altered rankings in the 275-substrate set. The two functions unique to the 393-substrate set are “Activation of anterior HOX genes in hindbrain development during early embryogenesis” (#14), and “Activation of HOX genes during differentiation” (#15). The two functions unique to the 275-substrate set are “Chromatin modifying enzymes” (#14) and “Chromatin organization” (#22).

Enriched gene sets associated with E3 ligases whose elevated expression is associated with increased OS

Elevated expression of the E3 ubiquitin ligases RNF144B, DTX3L, RNF25, SYVN1, MARCHF1, MARCHF8, MARCHF2, ZNRF2, BIRC3, RNF135, PCGF5, TMEM129, RMND5B, TRIM69, TRIM22, HERC6, TRIM21, and VPS18 (see Table 2) is associated with significantly (p = 0.03 to 9.4 X 10^-6; HR = 0.65-0.43) lengthened OS (40.1 to 50.9 months), suggesting that these E3 ligases target substrates that support tumor progression. UbiBrowser 2.0 identified or predicted a pool of 3238 substrates of these E3 ligases that are not substrates of the 13 E3 ligases whose elevated expression is associated with diminished OS. Because this pool of substrates is 6.9 times larger than that of the E3 ligases associated with diminished OS, enriched functions don’t reach the same levels of significance. Only 3 enriched functions have p values ≤1 X 10^-8. The 20 most significantly enriched functions identified using Reactome are presented in Table 6, and false discovery rate (FDR) values are also presented to aid in assessment of their potential pathophysiological impact. The most highly represented pathway category at the top of the Reactome hierarchy is vesicle mediated transport (8 uniquely enriched functions), followed by extracellular matrix organization (6 enriched functions), and neuronal system (2 functions). This data is represented in pie chart form in Figure 6B.

4. Discussion

4.1. Substrates of E3 Ligases Whose Upregulation Shortens OS Are Enriched in Tumor Suppressor Functions.

Among the 80 most significantly enriched functions associated with substrates of the E3 ligases whose elevated expression diminished OS, suggesting that they play key roles in proteasome addiction, 30 fall under overarching categories (chromatin organization, cell cycle, DNA repair and DNA replication) known to harbor tumor suppressors. Twenty-five functions fall under gene expression. These include 10 functions related to rRNA transcription by RNA polymerase I. Functional ribosomes are crucial for rapid growth and cell proliferation in Ewing and osteosarcomas. RNA polymerase I is considered a potential therapeutic target in some sarcomas, with the RNA pol 1 inhibitor CX-5461 inducing an antiproliferative effect in SK-UT-1 leiomyosarcoma cells via G2 cell cycle arrest [27]. Five functions relate to epigenetic regulation of gene expression by mechanisms including DNA and histone methylation and MLL3 and MLL4 complex activity. Loss of function alterations in epigenetic modulators are common in mesenchymal cancers and have been found in sarcomas including synovial, epithelioid, and Ewing [28]. The methyltransferases MLL3 and MLL4 are known to play protective roles against sarcomas by maintaining methylation patterns in the promoter regions of various potential oncogenes [29,30]. Another two gene expression functions involve RUNX1, which can act as an oncogene or tumor suppressor depending on context. In the blood, it plays a role in proper differentiation of stem cells, and its loss or participation in a fusion protein is seen in various blood and gastrointestinal cancers [31,32,33]. Six functions fall under developmental biology. These functions can be grouped by their correlation with maternal to zygotic transition (MZT) or HOX gene activation. Proteins involved in MZT include epigenetic regulators involved in establishing totipotency in zygotes that can later function as tumor suppressors [34,35]. HOX genes can act as tumor suppressors in context specific cancers. For instance, HOXA5 has been shown to function as a tumor suppressor in osteosarcoma [36]. Finally, four functions fall under cellular responses to stimuli, all of which relate to cellular senescence. Senescence in cancer cells can halt tumor progression [37] although the senescence associated secretory phenotype can create a microenvironment that can foster cancer growth and progression [38]. Only 10-20% of soft tissue sarcoma patients exhibit an objective response. The presence of MARCHF5 among the E3 ligases whose upregulation shortens OS may be contributory to this limitation. MARCHF5 substrates identified by UbiBrowser 2.0 include 100 human leucocyte antigens and 18 immunoglobulins, and enrichment analysis of these substrates alone yields “immunoregulatory interactions between a lymphoid and non-lymphoid cell”, “adaptive immune system”, “immune system”, “cytokine signaling in immune system”, “interferon gamma signaling” and “interferon signaling” as the top 6 most significantly enriched functions. A recent report by Wang and Zou describe attenuation of antigen presentation via PD-L1 upregulating MARCHF5-mediated lysosomal degradation of IL2Rγ_c [39]. IL2 sensitivity is thus attenuated by reduced receptor levels. These observations provide a mechanism explaining the known synergistic efficacy of coadministration of PD-1/PD-L1 blockade with IL2.

4.2. Substrates of E3 Ligases Whose Upregulation Extends OS Are Enriched in Oncogenic Functions.

Because this pool of E3 ligase substrates is 6.9-fold larger (3171 vs 459) than that of the substrates associated with decreased OS, the significance of enriched functions is correspondingly reduced and only the most significantly enriched 20 functions are listed in Table 6. These functions fall under five overarching categories, two of which (immune system and signal transduction are also enriched among the substrates of E3 ligases associated with decreased OS. The remaining three are neuronal system, extracellular matrix (ECM) organization, and vesicle mediated transport. The two functions under neuronal system both relate to AMPA receptors. While AMPA receptor function has not to date been associated with sarcomas, it is associated with glutamate signaling dependent tumor growth, invasion, and migration in other cancers, notably including glioma, and lung, breast, and prostate carcinomas [40,41,42]. The functions under ECM organization include integrin and non-integrin ECM cell membrane interactions (including laminin interactions). Some sarcomas rely on integrins and discoidin domain receptors to sense mechanical changes in the surrounding ECM to drive proliferation, invasion, metastasis, and resistance to therapeutics [43,44]. ECM organization functions also include ECM organization and degradation, with degradation being an essential process for invasion and metastasis in solid tumors in general. In this light, matrix metalloprotein 14 (MMP14) has been shown to be highly expressed in most sarcomas and to promote invasion and metastasis [45]. Enriched functions under vesicle mediated transport include functions related to anterograde and retrograde protein transport and cargo concentration in the ER. Anterograde transport could be important for secretion of MMPs and transport of important membrane proteins, and retrograde transport is often hijacked by cancers to alter localization of proteins such as sortilin and glucose transporters, impacting cell metabolism and tumor growth [46].

4.3. Adverse Events Associated with Proteasome Inhibitor Therapy

Real world data associates proteasome inhibitor therapy with cardiovascular adverse events including heart failure, hypertension, arrhythmias, and ischemic heart disease including acute coronary syndrome and angina. It can also negatively impact blood cell counts and cause gastrointestinal symptoms including nausea, vomiting, diarrhea, and constipation.[47] Rates of serious adverse events can be substantial and vary with proteasome inhibitor. For instance, respective cardiovascular adverse event frequencies associated with carfilzomib, bortezomib, and ixazomib are 7-27%, 0.6-4.1%, and 1.3%.[48] A growing body of evidence indicates that a functional circadian clock exerts potent tumor suppressive activity, impacting immune functions, metabolism, cell cycle control, and DNA repair.[48] While proteasome function is not specific to the circadian clock, it is essential for circadian clock function. The proteasome degrades core clock proteins including period and cryptochrome proteins at specific times within the circadian period, allowing the clock to reset. Thus the proteasome is essential for persistence of the circadian clock over multiple cycles, and impacts the length of a circadian period. It is thus possible that proteasome inhibitor therapy could exert oncogenic effects that counterbalance its therapeutic effects in cancer patients.

5. Conclusions

The known and potential adverse events associated with proteasome inhibition suggest that a better understanding of proteasome addiction through upstream analysis of key ubiquitin ligases and their target proteins and processes could be of value in identifying novel targets with similar efficacy but fewer or milder side effects. Our results indicate that only 3.4% (13/377) of the human repertoire of E3 ubiquitin ligases, and 4.9% (393/8047) of known and predicted substrates of the 31 E3 ubiquitin ligases that significantly increase or decrease OS by >40 months impactfully contribute to proteasome addiction mechanisms. Among the 118 most significantly enriched functions among the substrates of the 13 proteasome addiction-associated E3 ligases, 57% are considered tumor suppressive, and 90% of the 20 most significantly enriched functions among substrates of the 18 E3 ligases whose elevated expression extends OS are considered cancer promoting, supporting the conclusion that proteasome addiction mechanisms are more fine-tuned and specific than proteotoxic stress.