Tradeoff between metabolic i-proteasome addiction and immune evasion in triple-negative breast cancer

Abstract

In vitro studies have suggested proteasome inhibitors could be effective in triple-negative breast cancer (TNBC). We found that bortezomib and carfilzomib induce proteotoxic stress and apoptosis via the unfolded protein response (UPR) in TNBC cell lines, with sensitivity correlated with expression of immuno-(PSMB8/9/10) but not constitutive-(PSMB5/6/7) proteasome subunits. Equally, the transcriptomes of i-proteasome-high human TNBCs are enriched with UPR gene sets, and the genomic copy number landscape reflects positive selection pressure favoring i-proteasome activity, but in the setting of adjuvant treatment, this is actually associated with favorable prognosis. Tumor expression of PSMB8 protein (β5i) is associated with levels of MHC-I, interferon-γ-inducible proteasome activator PA28β, and the densities of stromal antigen-presenting cells and lymphocytes (TILs). Crucially, TILs were protective among TNBCs that maintain high β5i but did not stratify survival amongst β5i-low TNBCs. Moreover, β5i expression was lower in brain metastases than in patient-matched primary breast tumors (n = 34; P = 0.007), suggesting that suppression contributes to immune evasion and metastatic progression. Hence, inhibiting proteasome activity could be counterproductive in the adjuvant treatment setting because it potentiates anti-TNBC immunity.

Figure 1.. In vitro analysis of proteasome subunit expression and sensitivity to proteasome inhibitors.

(A) Proteasome subunit expression in breast cancer cell lines (Neve et al, 2006). (B) Inducible-to-constitutive subunit expression ratios (i:c) across molecular breast cancer subtypes (stats: two-way ANOVA). (C) Linear regression and Pearson correlation analysis of the relationships between i:c and sensitivity to bortezomib or carfilzomib (LD50, lethal dose 50%). Correlation coefficients (r) and regression fit values (r²) indicated. (D) Western analysis of inducible subunits and the PA28 cap in lines with a range of bortezomib sensitivities. ^#MCF10A is a spontaneously immortalized breast-derived line included for comparison. (E) Native in-gel proteasome activity assay with lysates from MDA-MB-468 and MCF7 with/without 2-h bortezomib treatment. (F) Light microscope images of MDA-MB-468 and MCF7 48 h after bortezomib treatment (captured at 20× magnification, scale bar 50 μm). (G) Cell viability after siRNA-mediated depletion of PSMB8 (i) or PSMB5 (ii). P-values in this figure were from unpaired, two-tailed t tests (pair-wise comparisons) or one-way ANOVA tests (comparison across multiple groups): *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Source data are available for this figure.

Figure S1.. Data supporting Fig 1.

(A) Proteasome subunit expression in breast cancer cell lines. Gene expression array data from (Neve et al, 2006). (B, C) Representative RT-qPCR (B) and Western analysis (C) performed 72 h after siRNA transfection of breast cancer cell lines. NTNC, non-targeted negative control.

Figure 2.. In vitro activation of the unfolded protein response (UPR) by bortezomib in relation to proteasome subunit expression.

(A) Working model of UPR induction by three major signaling axes. BiP, binding immunoglobulin protein (GRP78). (B) qRT-PCR for ATF4 following bortezomib treatment. (B, C) Log₂ fold-change (FC) in XBP1s, apoptosis markers, and NFκB at multiple time points after bortezomib treatment (B, C: qRT-PCR). (D) Inverse association between PSMB8/UPR induction and bortezomib sensitivity. (E) PSMB8 qRT-PCR (i) and bortezomib-induced cell death (ii) after pretreatment with IFN-γ. (F) IFN-γ–mediated induction of IRF1 and antigen processing genes in MDA-MB-468/MCF7 cells transfected with scrambled (SCR) or IRF1-specific siRNAs. P-values in this figure were from unpaired, two-tailed t tests: **P < 0.01; ***P < 0.001; ****P < 0.0001.

Figure 3.. i-proteasome genes are induced in claudin-low TNBCs and coordinately expressed with unfolded protein response genes.

(A, B) i:c Subunit expression ratios in breast cancer subtypes (METABRIC data, ANOVA test. *P < 0.05; ****P < 0.0001). (C) Gene set enrichment analysis plots showing running enrichment scores (ESs) skewed toward the PSMB8-correlated transcriptome. NES, normalized ES.

Figure 4.. Proteasome gene copy number aberrations related to clinical outcomes in breast cancer.

(A) TCGA 450k methylation array data for chromosome 6. (i) GISTIC calls for each probe shown as fractions of all triple-negative breast cancers (TNBCs). (ii) Zoomed region encoding antigen-processing genes. Fractions of HER2+ and ER+ cases shown for comparison. (B) CN status of subunit genes in major disease subtypes. #, instances of inducible subunit gain and constitutive subunit loss in a large percentage of TNBCs. (C) Violin plots showing i:c subunit CN switching (stats: pairwise Kruskal–Wallis tests with Dunn’s correction for multiple comparisons; ****P < 0.0001). (D) Kaplan–Meier analysis of i:c subunit expression in TNBC patients treated with (i) or without (ii) chemotherapy (CT) and/or radiotherapy (XRT). Q4/2-3/1, upper/mid/lower quartiles. Log-rank P-values shown. (E) i:c subunit expression ratios in TNBC subtypes (Burstein et al, 2015): BLIA, basal-like immune-activated; BLIS, basal-like immune-suppressed; LAR, luminal androgen receptor-like; MES, mesenchymal. Kruskal–Wallis test: ***P < 0.001, ****P < 0.0001. (F) Kaplan–Meier analysis of (i) METABRIC and (ii) TCGA TNBCs classified by whether i-subunit gene copy number outnumbers that of constitutive subunit counterparts. Log-rank P-values shown.

Figure S2.. Data supporting Fig 2. Relationships between proteasome subunit RNA expression and probability of survival in various subgroups using KM-Plotter datasets.

(A, B, C) i-Proteasome subunits: PSMB8, PSMB9, and PSMB10. (D, E, F) c-Proteasome subunits: PSMB5, PSMB6, and PSMB7. (i) Tumors classified as TNBC by IHC (ii, iii, and iv); basal-like cases according to the PAM50 molecular classifier. (ii) All basal-like cases, excluding patients treated with hormone therapy (HT), which are clinically ER+. (iii) Untreated basal-like cases. (iv) Basal-like breast cancers treated with adjuvant and/or neoadjuvant (NA) chemotherapy.

Figure S3.. Data supporting Fig 4.

(A) Relationships between subunit CNAs and expression (TCGA data; amp, amplification; del, deletion; dip, diploid; and TPM, transcripts per million). (B) Spearman correlation of tumor purity and proteasome subunit expression (from the TGCA BRCA provisional RNAseq dataset [Cancer Genome Atlas Network, 2012; Aran et al, 2015]).

Figure 5.. Tumor compartment–specific expression of PSMB8 relates to clinicopathologic variables.

(A) Representative β5i, MHC-I, and PA28β IHC analysis of normal breast ducts and lobules (i) and invasive breast tumors (ii, iii), illustrating cases that exemplify maintenance (ii) or selective loss (iii) of these class-I antigen presentation components. Cores are 1.0 mm i.d. and insets 140 μm². (B) Contingency analysis of the relationships between β5i, MHC-I, and PA28β in triple-negative breast cancer (TNBC). Fisher’s exact test P-values indicated *P < 0.05; **P < 0.01; ***P < 0.001. (C) Contingency analysis of relationships between β5i, MHC-I, or PA28β, with TILs density (i), TILs PD-L1 positivity (ii), tumor cell PD-L1 positivity (iii), and the density of stromal APCs (sAPCs; iv). Fisher’s exact test P-values indicated *P < 0.05; **P < 0.01; ***P < 0.001. (D, E, F) Kaplan–Meier analysis of TNBCs stratified by β5i, MHC-I, or PA28β (i) or by TIL density after classifying TNBCs by their maintenance or loss of β5i, MHC-I, or PA28β (ii). HR, hazard ratio (95% confidence interval); log-rank P-values shown. (G) Change in β5i IHC scores in brain metastases (BrM) compared with matching primary breast cancers (BC). Overall numbers of cases exhibiting increases (↑), decreases (↓), or no change (–) are indicated. Paired, two-tailed t test P-value shown.

Figure S4.. Data supporting Fig 5.

(A, B, C, D, E, F) Representative IHC images (A, C, E) and chi-squared analysis in key breast cancer subtypes (B, D, F) for β5i (A, B), MHC-I (C, D) and PA28β (E, F). Cores are 1.0 mm i.d. and insets 180 μm². (G) Relative proportions of all genes exhibiting copy number change, according to the PSMB8 copy number status. (H) Average number of alterations (single-nucleotide variants [SNVs], small deletions, and insertions) per case, according to the PSMB8 copy number status.

Figure 6.. Model contrasting the potential consequences of i-proteasome expression before and after treatment.