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

Engaging the immunoproteasome helps maintain proteostasis during TNBC development, but ultimately leads to increased immune visibility and favorable prognosis. Assessing i-proteasome expression could augment the prognostic and/or predictive value of TILs assessment in diagnostic practice. 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.


Introduction
The ubiquitin-proteasome system (UPS) is vital for cellular homeostasis, acting to regulate protein expression and eradicate oxidized, misfolded proteins. The proteasome is its 2.5-MD catalytic engine-the major cellular protein recycling complex that degrades poly-ubiquitinated substrates (Sorokin et al, 2009). Its 20S core comprises 28 subunits arranged in four heptameric rings-two outer "alpha" and two inner "beta" rings surround a central pore through which unfolded proteins pass as they are cleaved. Three subunits (six per core) provide a complete spectrum of proteolytic activity after acidic, basic, and hydrophobic amino acids: β1 caspase-, β2 trypsin-, and β5 chymotrypsin-like (encoded by PSMB6, PSMB7, and PSMB5). Proteasome substrate specificity, throughput, and subcellular localization are regulated by activator complexes that dock with the core, forming single-and double-capped holoenzymes. A 19S cap is essential for de-ubiquitination, translocation, and ATP-dependent substrate unfolding. Under normal conditions, 19S-capped 20S core complexes the predominant proteasome species present in the cell ("constitutive" proteasomes). However, in certain conditions, the core can also dock with PA28α/β, PA28γ, and PA200 caps, which lack catalytic activity but widen the 20S pore and increase throughput (reviewed in Kors et al [2019]).
In addition to posttranslational regulation and buffering against proteotoxic stress, the UPS is also the major source of peptides presented to the immune system via MHC class-I molecules. Accordingly, IFN-γ is a potent stimulus for proteasome compositional change. IFN-γ down-regulates constitutive proteasome subunit genes and induces the genes that encode PA28α and PA28β (PSME1 and PSME2) and alternative 20S core subunits β1i, β2i, and β5i (PSMB9, PSMB10, and PSMB8), which increases proteasome throughput and produces a varied repertoire of MHC-I epitopes via different cleavage site preferences (Aki et al, 1994). Hence, in the context of an immunologic challenge, IFN-γ-stimulated cells express a variety of hybrid "immuno(i)-proteasome" complexes.
Proteasome inhibitors (PIs) such as bortezomib and carfilzomib are effective for treatment of multiple myeloma and mantle cell lymphoma. Their clinical activity is partly attributed to disruption of the UPS, which stabilizes apoptotic proteins (e.g., p53 and Bcl2) and IκBα, inhibitor of the NFκB pro-survival pathway. Both inhibitors bind to β5 and β5i subunits, but carfilzomib binds irreversibly and has a favorable side effect profile (Cromm & Crews, 2017). PIs also exploit metabolic addiction to the proteasome. By promoting accumulation of misfolded and oxidized protein aggregates in the ER, they trigger the unfolded protein response (UPR), a sensing mechanism that promotes apoptosis in the event of unsustainably high rates of cellular metabolism (Sorokin et al, 2009). In cancer, rapid division of cells with mutated genomes should incite the UPR, but strategies evolve to cope with proteotoxicity, including proteasome overexpression (Adams, 2004). The UPR-inducing activity of PIs is attributed mainly to inhibition of constitutive proteasomes, based on a persisting assumption that this complex is the most abundant in non-hematologic tissues (Roeten et al, 2018). However, solid tumors also express the i-proteasome (Altun et al, 2005;Ho et al, 2007), and this could represent an efficient means for maintaining a concentration of toxic protein aggregates below the UPR induction threshold (Shringarpure et al, 2003).
Triple-negative breast cancers (TNBCs) comprise 15-20% of invasive breast carcinomas. They lack clinically significant expression of first-line systemic drug therapy targets (estrogen receptor, human epidermal growth factor receptor-2 [HER2]), are typically high-grade, are metabolically active, and often exhibit basal-like and/or mesenchymal phenotypes (Perou et al, 2000;Prat et al, 2010). Cytotoxic chemotherapy is a mainstay of clinical management, but 40-80% of patients still experience distant relapse and premature death, often involving visceral and brain metastases (Fulford et al, 2007;Pusztai et al, 2019). TNBC exhibits considerable molecular and clinical heterogeneity, unsurprising given theirs is a diagnosis of exclusion. Identifying biomarkers and therapeutic targets remain top research priorities. Promising recent developments have centered on exploiting the predictive and therapeutic significance of tumor-infiltrating lymphocytes (TILs). There is now international consensus that the presence and density of TILs in pretreatment diagnostic specimens predicts the response to neoadjuvant chemotherapy and also to second-line treatment where a low-moderate residual disease burden remains (Loi et al, 2019;Luen et al, 2019). Thus, despite the systemic immunosuppression experienced by patients undergoing active treatment, the presence of TILs before this indicates the capacity of the host immune system to synergize with chemotherapy.
In vitro and in vivo studies suggest that the proteasome is a potential therapeutic target in TNBC (Cardoso et al, 2004;Petrocca et al, 2013;Weyburne et al, 2017); however, preliminary clinical data from metastatic patients have been mixed (Yang et al, 2006;Engel et al, 2007;Schmid et al, 2008), and these trials pre-dated our understanding of the role of antitumor immunity in TNBC. This study was undertaken to establish the molecular basis of PI activity and relevance to the treatment of breast cancer, incorporating in vitro mechanistic studies with genomic and protein-level data from large human clinical sample cohorts.

Results and Discussion
Basal-like breast cancer cell lines are dependent on i-proteasome activity We initially characterized the relationship between proteasome subunit expression and sensitivity to chemical or genetic inhibition of the proteasome in vitro. Analysis of breast cancer cell line gene expression data (Neve et al, 2006) indicated significantly higher expression of i-relative to constitutive subunit in basal-A and claudin-low (basal-B) lines than luminal and HER2+ cell lines (Fig 1A and B). We then performed bortezomib and carfilzomib dose-response assays with a panel of lines that model a range of proteasome subunit levels. Expression of inducible relative to constitutive subunit RNA was anticorrelated with response to both inhibitors (Figs 1C and S1A). The relationship to PI sensitivity was most noticeable for PSMB8 (bortezomib LD 50 correlation r − 0.76, P = 0.01), which was also evident at the protein level ( Fig 1D). There was no obvious relationship between bortezomib sensitivity and baseline expression of the proteasome activator cap subunit PA28γ (Fig 1D). This was not a consequence of preferential i-proteasome inhibition, as the total proteasome activity was suppressed in both MDA-MB-468 (i-prot high , low bortezomib LD 50 ); and MCF-7 (i-prot low , low bortezomib LD 50 ) (Fig 1E and F).
To rule out off-target effects, we transfected six lines with PSMB8-targeted siRNAs and measured cell viability after 48 h using flow cytometry. With the exception of MDA-MB-231, i-prot high basal lines were significantly more sensitive to PSMB8 depletion than luminal lines (Fig 1Gi). MDA-MB-468 (basal/i-prot high ) and MCF7 (luminal/i-prot low ) transfected with PSMB5-targeted siRNAs showed no significant differences (Fig 1Gii, with confirmation of RNA and protein knockdown in Fig S1B and C).

Bortezomib sensitivity correlates with the UPR in vitro
We reasoned that in cell lines addicted to the i-proteasome, pharmacologic suppression of proteasome activity would cause ER stress and trigger the UPR. Three parallel signaling axes mediate the UPR, initiated by ER-membrane sensors: PRKR-like ER kinase (PERK), activating transcription factor 6α (ATF6α) and inositol-requiring protein 1α (IRE1α). Induction of ATF4 marks PERK activation, whereas the IRE1α branch leads to expression of a specific X-box transcription factor splice isoform, XBP1s, which is associated with UPR-related effects in the nucleus. Collectively, these pathways induce genes required for protein folding, secretion, and clearance, or apoptosis if misfolding cannot be resolved (Fig 2A [Wang & Kaufman, 2014]).
In line with our hypothesis, bortezomib treatment induced ATF4 in the four lines assayed ( Fig 2B). Further analysis of i-proteasomedependent MDA-MB-468 cells revealed induction of XBP1s within 24 h of treatment, concomitant with apoptosis markers and reduction of pro-survival marker NFκB ( Fig 2C). Thus, PSMB8-high, i-proteasomeaddicted BC cell lines exhibit UPR-driven apoptosis in response to proteasome inhibition ( Fig 2D). In terms of extrinsic modulation, IFNγ-induced PSMB8 in both MDA-MB-468 and MCF7 (Fig 2Ei), but there was no effect on bortezomib sensitivity (Fig 2Eii). Assaying key IFN-γ targets after treatment, including the IFN-stimulated response element transactivator IRF1, confirmed the IFN-γ-IRF1-STAT1-PSMB8 axis was intact ( Fig 2F). Moreover, siRNA-mediated IRF1 depletion suppressed target induction ( Fig 2F). Thus, although i-proteasome levels are associated with bortezomib sensitivity, this is not an exclusive determinant. Sensitivity is likely to be linked to functional dependence on the i-proteasome rather than the expression of its components.

i-proteasome induction is associated with the UPR in TNBC
To explore whether PI therapy is an appropriate strategy for treatment of breast cancer, we mined several large genomic datasets: The Cancer Genome Atlas (TCGA; RNA-sequencing data from 1,092 cases including 176 TNBCs) (Cancer Genome Atlas Network, 2012), the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC; expression array data from 1,980 cases incl. 333 TNBCs) (Curtis et al, 2012), and KM-Plotter, a composite of expression array studies with comprehensive clinical annotation (Gyorffy et al, 2010). We interrogated these datasets to identify patient subgroups that may benefit from PIs and to test the hypothesis that i-proteasome activation is associated with the UPR in human tumors. Meta-analysis of expression data indicated that compared with estrogen receptor-positive (ER+) and HER2+ tumors, TNBCs preferentially express i-proteasome subunit genes over constitutive counterparts (β1i:β1, β2i:β2, and β5i:β5; Fig 3A), and this is most evident in tumors with a claudin-low phenotype ( Fig 3B).
We then performed UPR gene set enrichment analysis (The Gene Ontology Consortium, 2019) after ranking the transcriptome according to correlations with PSMB8. The distributions of multiple UPR gene sets were skewed toward genes co-expressed with PSMB8 ( Fig 3C). This was evident in both datasets, specifically in TN, basal-like tumors. PSMB9 and PSMB10 were among the top 20 PSMB8-correlated genes (r 0.89/0.83 and 0.68/0.54 in TCGA/META TNBCs), reflecting coordinated co-expression of i-proteasome subunits. These findings support the idea that i-proteasome addiction could be a therapeutically targetable vulnerability in breast cancer.  (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 (r 2 ) 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.
i-proteasome switching is driven by gene copy number alterations (CNAs) and stratifies survival in TNBC PSMB8 and PSMB9 genes are located within a 25-Mbp region on chromosome 6p that is frequently affected by CNAs in cancer, specifically, in the class-II locus of the HLA complex, which spans 4 Mbp on chromosome 6p.21 (Fig 4A). 6p gains are common in highgrade malignancies (Santos et al, 2007), suggesting proteotoxic stress resistance may be a fundamental requirement for cancers with high metabolic activity. TCGA CNA data indicated that 6p is affected by complex genomic instability in TNBC, marked by focal gains and losses (Fig 4Ai). PSMB8/9 are gained in 46% and amplified in 5.2% of TNBCs. The locus is also gained in ER+ and HER2+ cases, but at lower frequency (17% and 27%; Fig 4Aii). Flanking PSMB8/9 are co-altered genes encoding TAP1 and TAP2-ER membrane channels that internalize proteasomal peptides and facilitate MHC class-I antigen presentation (Fig 4Aii).
Quantifying proteasome subunit gene CNAs across BC subtypes revealed that in addition to gains at 6p21.32 (PSMB8/9) in TNBC, other events frequently affecting proteasome subunit CN are losses at 17p13.2 and 16q22.1 (PSMB6, PSMB10; Fig 4B). However, the genomic landscape of TNBC is unique in that i-subunit gains occur concomitantly with loss of constitutive subunit genes in more than half of cases ( Fig 4B). This was also evident within individual cases, with 65.3%, 31.2%, and 58.4% of TNBCs exhibiting a predominance of inducible β1, β2, and β5 subunits, respectively ( Fig 4C). Considering that subunit gene copy number is a determinant of overall expression (Fig S3A), these findings suggest that high i-proteasome expression in TNBC is at least partly a consequence of selection pressure during tumorigenesis.
Next, we used Kaplan-Meier analysis to investigate associations between proteasome subunit expression and patient outcomes. In independent TNBC datasets, i:c subunit expression ratios stratified 10-yr breast cancer-specific survival, but cases with the highest i:c ratios had longer survival (Figs 4D and S2). Stratification was more prominent among patients treated with chemotherapy and/or radiotherapy (Figs 4D and S2). Therefore, althoguh i-proteasome switching subverts metabolic stress in vitro and may confer a selective advantage during TNBC development, in the clinical setting, this seemed to be associated with favorable responses to treatment.
Because (1) the i-proteasome is associated with both metabolic homeostasis and antigen processing (2), multiple antigen-processing machinery components are co-altered by genomic alterations in TNBC, and (3) efficient antigen processing is conducive to antitumor immunity in cancer generally; we reasoned that i-proteasome overexpression could be both a vestige of metabolic addiction and indicator of effective antigen presentation. The fact that the link between i-proteasome and clinical outcome is specific to patients treated with chemotherapy, and radiotherapy is consistent with the evidence that these treatments promote immunogenic tumor cell death by reactivating immune surveillance (Dushyanthen et al, 2015). Indeed, i-subunit but not c-subunit expression is inversely proportional to tumor purity (Fig S3B), and the i:c expression ratio is highest in "basal-like, immune-activated" TNBCs ( Fig 4E) characterized by dense lymphocytic infiltrates (Burstein et al, 2015).
This goes some way to reconciling the i-proteasome's involvement with both tumor development and responses to treatment, but also raises the question of how much of the i-proteasome RNA measured in tumor homogenate is attributable to tumor versus stromal components. Unlike RNA expression, CNAs are intrinsic to the tumor cell compartment, so we performed Kaplan-Meier analysis of TNBCs after classifying them according to shifts in inducible and constitutive proteasome subunit gene copy number. In both METABRIC and TCGA datasets, TNBCs with higher overall i-subunit copy number survived longer after the treatment (Fig 4F), supporting the idea that higher levels of inducible proteasome subunits are protective.
β5i is a favorable prognostic indicator associated with tumorspecific immunity in TNBC To validate these findings at the protein level, we performed immunohistochemistry (IHC) analysis of β5i (PSMB8) in relation to clinicopathologic variables and disease-specific survival using a third cohort of 424 invasive breast tumors. We also analyzed the MHC-I complex and proteasome activator subunit PA28β (PSME2), which is similarly induced by IFN-γ but located outside the HLA complex on chromosome 14q12 and less frequently affected by CNAs (14.5% and 1.7% of TNBCs with PSME2 gain or amplification; 46% and 5.2% for PSMB8). β5i and PA28β were detected in the nuclei and cytoplasm of normal mammary epithelial structures and breast tumor cells, and MHC-I in cytoplasm and cell membranes (Fig 5A and Tables S1-S3). On average, 83% of cases were strongly positive for β5i, and 17% showed selective loss in the tumor compartment (Figs 5A and S4A). Luminal-A-like tumors had the most frequent expression overall, followed by basal-like TNBCs (Fig S4B). Around one-third of all tumors exhibited loss of MHC-I expression, but again expression was frequently maintained in basal-like TNBCs (Fig S4C and D and Table S2). For PA28β, 92% of cases were positive, with minimal differences between breast cancer subtypes ( Fig S4E and F and Table S3). Among TNBCs, there was a direct relationship between levels of β5i and PA28β, and they were both strongly associated with expression of MHC-I (Fig 5B), indicating that overall, there is coordinated expression of MHC-I antigen processing and presentation pathway components in TNBC.
We then reviewed matching hematoxylin and eosin-stained whole sections to quantify immunologic correlates of β5i, PA28β, and MHC-I expression: (1) the density of stromal TILs (Salgado et al, 2015), an established marker of treatment response in TNBC (Fig  5Ci), (2) the frequency of PD-L1 expression by TILs (Fig 5Cii), and (3) tumor cells (Fig 5Ciii), a marker of effector T-cell responses in breast cancer (Dushyanthen et al, 2015), and (4) the density of stromal cells expressing high levels of MHC-II, a marker of professional APCs (macrophages, dendritic cells, and B-cells; Fig 5Civ). Contingency analysis showed that TNBCs with sustained expression of β5i and MHC-I were more likely to be infiltrated by TILs and APCs and to exhibit signs of antigen-specific T-cell engagement. These findings are consistent with another report showing a direct relationship To investigate the potential consequences of i-proteasome components and/or MHC-I being dysregulated or lost in TNBC, we performed the Kaplan-Meier analysis. Expression of β5i or MHC-I modestly stratified survival (Fig 5Di, Ei, and Fi), but more striking differences began to emerge when we considered the prognostic significance of TILs in the tumor microenvironment of TNBCs that maintained or lost antigen presentation components. TIL density remains the most reliable prognostic indicator in TNBC (Loi et al, 2019;Luen et al, 2019), yet we found that this was specific to tumors that retain expression of β5i, with little difference in outcome over 20 yr among TNBCs that had lost β5i expression (Fig  5Dii). MHC-I and PA28β showed similar trends, although their interactions with TILs density were not as significant (Fig 5Eii and Fii). We reasoned that if β5i loss enabled immune escape, expression would be suppressed during metastatic progression. Indeed, IHC analysis of β5i in brain metastases compared with patient-matched primary breast tumor samples revealed lower β5i expression in brain metastases in 16/34 of cases (Fig 5G; paired t test P = 0.007).
Taken together, these findings suggest that antigen processing via i-proteasome subunit β5i is associated with antitumor immune responses. Given that we saw such a striking difference in TIL-based prognostication depending on the β5i status, further cohort studies are warranted to investigate predictive testing of i-proteasome expression in conjunction with TILs assessment. We explored the possibility that β5i-high TNBCs might be intrinsically more immunogenic by comparing the mutation burden of PSMB8 gained versus copy number-neutral TNBCs in the TCGA cohort but found no evidence suggesting that their neoantigen load is higher than other TNBCs ( Fig  S4G and H). Hence, β5i expression is unlikely to be causally associated with a dense TIL phenotype, but probably potentiates tumor-specific immune responses providing there is sufficient capacity.
We propose that the i-proteasome is exploited during TNBC development to cope with proteotoxicity, but in immunologically "hot" tumors (defined here as TILs occupying ≥20% of tumorassociated stroma), this ultimately becomes a liability because i-proteasome activity is linked to the efficacy of radiotherapy and chemotherapy (Fig 6). Hence, adjuvant PI therapy could be counterproductive in TNBC. On the other hand, increasing i-proteasome activity could potentiate other first-line therapies.
Supporting this idea in principle, Tripathi and colleagues showed that PSMB8/9 are suppressed in mesenchymal-like non-small cell lung cancers, and that reactivating these genes with exogenous IFN-γ restored a repertoire of MHC-I-bound tumor antigens that could efficiently prime cytotoxic responses from patient-derived peripheral blood mononuclear cells (Tripathi et al, 2016). 5-Aza-29-deoxycytidine (decitabine), a cytosine analog and DNA methylation inhibitor, had the same effect by suppressing methylation at PSMB8/9 enhancers (Tripathi et al, 2016).
Moreover, we recommend that additional cohort studies are warranted to quantify the prognostic and predictive value of adding β5i IHC to diagnostic assessment of TILs density in TNBC.

Cell lines and qRT-PCR
Cell lines were purchased from the American Type Culture Collection, maintained in recommended culture conditions and regularly screened for Mycoplasma (MycoAlert; Lonza). Cell line working stocks used in this study were authenticated by STR profiling (Promega GenePrint 10 System). For RNA inhibition studies, the cells were transfected in six-well plates (3 × 10 5 cells/well) with 50 nM siRNAs (GenePharma) using Lipofectamine RNAiMAX (Life Technologies). An IncuCyte live-cell imaging system (Essen BioScience) was routinely used to monitor cell attachment and growth in real time. Expression analyses were performed 72 h post-transfection. For qRT-PCR, total RNA was extracted from cultured cells using the RNeasy Mini Kit (QIAGEN), cDNA was random-primed from 1 μg total RNA (MMLV RT; Life Technologies), and qPCR was performed using SYBR Green (Life Technologies; Table 1) and a CFX96 instrument (Bio-Rad). Experiments were repeated three times.
Gene set enrichment analysis was performed using the GSEAPreranked module of GenePattern (weighted scoring scheme, 1,000 permutations) (Reich et al, 2006). Gene sets were extracted from the Gene Ontology Consortium database (The Gene Ontology Consortium, 2019) using the search term "UPR." Preparation of graphs and all other statistical tests were performed using GraphPad Prism software (v8.4). Statistical tests are described in figure legends.

Clinical sample cohorts and IHC analyses
IHC analysis of archival specimens and clinicopathologic data were approved by human research ethics committees at The Royal Brisbane and Women's Hospital  and The University of Queensland (2005000785). IHC analyses were performed on two separate cohorts: 1. The Queensland follow-up cohort. This resource was built from archival tissue specimens of breast cancer patients treated in Queensland between 1987 and 1992, sourced from the statewide Pathology Service, Pathology Queensland, and sampled in duplicate on tissue microarrays (TMAs). Median follow-up is 13.9 yr (range Clinical and pathology data were extracted from diagnostic reports, our internal diagnostic histopathology review (SRL) and the Queensland Cancer Registry. Analysis included cross-referencing to clinicopathologic parameters that were assessed and published previously (e.g., expression of ER, PR, and HER2; histological grade and subtype [Junankar et al, 2015;Zhang et al, 2015;Burgess et al, 2016;Hernandez-Perez et al, 2017;McCart Reed et al, 2018;Raghavendra et al, 2018;Wiegmans et al, 2019]). For IHC, 4-μm TMA sections were heat-retrieved in using a Decloaking Chamber and blocked for 15 min at room temperature before staining ( Table  2). The MACH 1 Universal HRP-Polymer Detection Kit was used for detection. Reagents and equipment were from Biocare Medical unless otherwise specified. For image analysis, hematoxylin-counterstained sections were scanned at 40× magnification on an Aperio AT Turbo slide scanner (Leica Biosystems). De-identified digital TMA core images were scored by one assessor and reviewed by a second. The maximum scores of duplicate TMA cores were calculated for each case, except for β5i, MHC-I, and PA28β, where minimum scores were used because the disease-associated phenotypes were a loss of expression. Associations between biomarkers and clinicopathologic variables were investigated using chi-squared, Fisher's exact, and logrank tests (GraphPad Prism v8.4).  (Salgado et al, 2015). Intensity and percentage of TILs stained were recorded as a Histo-score. A cut-off of ≥1% was considered positive. BSA, bovine serum albumin; ON, overnight; RT, room temperature.