Spatial Immunoprofiling of Adenoid Cystic Carcinoma Reveals B7-H4 Is a Therapeutic Target for Aggressive Tumors

Adenoid cystic carcinoma (ACC) commonly arises in the salivary glands and is characterized by high rates of recurrence despite aggressive local therapy (1). Histologically, ACC is composed of cells with epithelial and myoepithelial differentiation that are arranged in three histologic patterns: cribriform, tubular, and solid (2). The ratio of myoepithelial and epithelial cells affects the tumor growth rate (3); for instance, the loss of myoepithelial cells is associated with solid histology, aggressive clinical behavior, and poor prognosis (1, 4). Indeed, DNA sequencing of ACC samples has shown differentially expressed transcripts according to epithelial or myoepithelial-dominant tumors, which supports the important role of tumor composition in ACC biology (3, 5).

Recently, our research group reported a study that integrated proteogenomic and clinical data from 54 ACC patients and revealed two major molecular ACC subtypes, ACC-I and ACC-II. Notably, each subtype was associated with unique therapeutic vulnerabilities and a dramatically distinct prognosis (6). ACC-I was associated with the enrichment of NOTCH1 activating mutations, upregulation of MYC, MYC target genes and mRNA splicing, and an overall poor prognosis [median overall survival (OS) duration = 3.4 years]. On the other hand, ACC-II exhibited upregulation of TP63 and receptor tyrosine kinases (e.g., AXL, MET, and EGFR), and less aggressive clinical behavior (median OS duration = 23 years). These findings highlight ACC's biological and clinical heterogeneity and the importance of developing biomarkers to optimize treatment decisions and personalize care in this diverse patient population.

The distinct transcriptome that characterizes each ACC subtype is translated into a specific protein profile that may tailor the tumor microenvironment (TME) composition and specific phenotypic patterns that influence the disease's clinical behavior. Our research group previously analyzed the expression of 174 cancer-relevant proteins in 38 ACC tumors using reverse phase protein array (RPPA) and found differentially expressed proteins in ACC-I than in ACC-II. Specifically, c-MYC, cleaved NOTCH1 (NICD1), Bcl-2, and the immune-suppressor marker B7-H4 were more abundant in ACC-I than in ACC-II (6). Some of the limitations of RPPA are the absence of information regarding protein expression by cell subpopulations and protein localization in the tumor and its microenvironment. Thus far, studies focusing on the in-depth profile of the ACC TME are lacking.

The goals of the current study were to comprehensively explore the ACC TME contexture using RNA-sequencing (RNA-seq) and imaging mass cytometry (IMC; a high-dimensional tissue imaging system); assess the association between the TME and ACC molecular subtypes, histologies, and clinical outcomes; and identify potential biomarkers for future drug development.

The RNA-seq cohort comprised three cohorts of ACC tumors with available RNA-seq data that had been collected as part of previous studies. After the removal of duplicate samples, there were 54 samples from our group (Ferrarotto and colleagues; ref. 6), 37 from Bell and colleagues (3), and 37 from Frerich and colleagues (7), for a total of 128 unique patient tumor samples.

The IMC cohort comprised a subset of 62 patients out of the 128 RNA-seq ACC cohort [n = 35 from our previous study (6) and n = 27 from Bell's cohort (3)] with available clinicopathologic data. The ACC histologic subtype was defined by experienced head and neck pathologists (DB and AEN). The solid subtype was defined as the presence of any solid growth in the tumor tissue.

This study was conducted in accordance with the Declaration of Helsinki. Samples were obtained under The University of Texas MD Anderson Cancer Center (MDACC) Institutional Review Board (IRB)–approved waiver of informed consent (for deceased patients) or informed consent for molecular and clinicopathologic analyses with approved IRB protocols (LAB10-0918 and 2021-0355).

Classification of the 128 ACC samples as ACC-I and ACC-II was performed as previously described (6). Immune signatures were curated from Danaher and colleagues (8), Bindea and colleagues (9), and Rooney and colleagues (10), along with specific immunologic genes of interest. The scores for each cell type or pathway were determined as the average value of log-transformed and z-normalized expression values for each gene in the signature. Final statistical significance was determined by meta-analysis across the three cohorts using a fixed effects model.

IMC was performed at the Flow Cytometry and Cellular Imaging Facility at MDACC with the Fluidigm Helios CyTOF instrument and Hyperion Imaging System Laser ablation, as previously described (11).

Two tissue microarrays were constructed from FFPE blocks, with two representative regions for each sample (tumor center and tumor–stroma interface), totaling 124 representative 1-mm³ samples. Ten 1-mm³ tonsil samples were added to the TMAs as controls for the antibodies being tested. Detailed information regarding IMC procedures is available in the Supplementary Materials. The custom antibody target panel is available in Supplementary Table S1.

Tissue and cell segmentation and a multiplex imaging analysis were performed with Visiopharm software using pretrained artificial intelligence algorithms. The marker combinations used to define cell populations are specified in Supplementary Table S2.

An IHC analysis was performed as previously described (6). In brief, cleaved NOTCH1 monoclonal antibody (mAb; D3B8; #4147; Cell Signaling Technology) was used for NICD1 IHC staining. NICD1 expression was considered positive if it stained in at least 70% of tumor cells. IHC detection of B7-H4 was performed in a subset of TMA samples (n = 37) using the B7-H4 (D1M8I) XP rabbit mAb (Cell Signaling Technology, #14572) and on tumor tissue sections derived from patient-derived xenografts (PDX) using an antihuman B7-H4 rabbit IgG1 mAb (CAL63; Abcam) on an automated Ventana Benchmark staining platform (Ventana Medical Systems; ref. 12). B7-H4 expression was considered high if it stained in at least 70% of tumor cells.

All animal procedures were performed at XenoSTART (South Texas Accelerate Research Therapeutics) following Institutional Animal Care and Use Committee protocols. Tumor fragments (∼70 mg) from previously established ACC-I (ACCX9 and ACCX11) and ACC-II (ACCX5M1) PDXs (ref. 13; Supplementary Fig. S1) were implanted subcutaneously into the flanks of athymic nude mice [Ctrl:NU(NCr)-Foxn1^nu; RRID:IMSR_CRL:490; The Jackson Laboratory]. Once tumors reached 150–300 mm³, the animals were randomly assigned to groups receiving either one intravenous dose of AZD8205, B7-H4 antibody–drug conjugate (ADC) bearing a topoisomerase I inhibitor (1.25 or 3.5 mg/kg), a nontargeting isotype control ADC (Iso-ADC) using an isotype-matched antibody with a comparable drug–antibody ratio (1.25 and/or 3.5 mg/kg), or PBS (n = 10 PDXs per each group; ref. 12). Animals were observed daily; tumor dimension data were collected twice weekly using digital calipers. The percentage of tumor growth inhibition (%TGI) was calculated for each treatment group (T) versus control (C) using initial (i) and final (f) tumor measurements and the following formula: %TGI = 1 − T_f − T_i/C_f − C_i. Individual mice with a tumor volume ≤50% of the day 0 measurement for two consecutive measurements over a 7-day period were considered partial responders (PR). Individual animals lacking palpable tumors (<4×4 mm²) for two consecutive measurements over a 7-day period were considered complete responders (CR).

Comparisons of cell types and marker densities between molecular subtypes (ACC-I vs. ACC-II) and histology subgroups (solid vs. nonsolid) were analyzed with the Wilcoxon signed-rank test. Associations with OS duration were assessed with log-rank and Cox proportional hazards. The OS duration was calculated from the date of diagnosis to the date of the last follow-up or death. Multiple comparisons were accounted for by the Benjamini–Hochberg procedure. An FDR of ≤ 0.2 and a P value of < 0.05 were considered statistically significant.

The data generated in this study are available upon request from the corresponding author. The RNA-seq data from the studies previously discussed (refs. 3, 6, 7) are in NCBI BioProject PRJNA306160 (3), Mendeley Data doi:10.17632/6sbv7bpj5n.1 (6), and NCBI BioProject PRJNA287156 (7).

Leveraging our well-characterized ACC cohort transcriptomic and proteomic data (6), we explored the ACC immune microenvironment and determined how it varies between ACC subtypes (ACC-I vs. ACC-II).

We began by evaluating the ACC immune landscape from the bulk RNA-seq transcriptional profile in our 54-patient cohort and found that most ACC tumors had low tumor-infiltrating leukocytes, with only 33% being “hot” on the basis of previously described immune signatures (Fig. 1A) (8–10). This result was validated using two other independent ACC cohorts (refs. 3, 7; Fig. 1B and C), with an overall average of 28.3% ± 4.3% of tumors hot.

We then compared the immune landscape derived from RNA-seq deconvolution between ACC-I and ACC-II in the merged three ACC cohorts (n = 128). Immune cell populations and immune checkpoints that were differentially expressed in ACC-I versus ACC-II (FDR < 0.2) are shown in Fig. 1D. Consistent with our prior targeted proteomics analysis using RPPA, we found that the immunosuppressive marker B7-H4 was overexpressed in ACC-I.

We hypothesized that increased expression of immunosuppressive B7-H4 and corresponding worsened prognosis would be associated with a “colder” TME. Unexpectedly, we found that ACC-I exhibited a higher presence of tumor-infiltrating leukocytes, including CD8 T cells, T follicular, Th2 helper T cells, and B cells, than ACC-II (FDR < 0.2; Fig. 1D). As tumor immune infiltration, particularly of CD8 T cells, is generally associated with improved prognosis, we sought to elucidate the apparent paradox between the association of immune infiltration with ACC outcomes using spatial analysis at the protein level.

Patient and tumor characteristics of the IMC cohort (n = 62) are depicted in Fig. 2A. Most samples [49 (79%)] were from the primary tumor, 52% of patients were male, and 37% had a solid histology component. Among patients with nonmetastatic disease at diagnosis (82%), 58% experienced disease recurrence.

After a median follow-up of 7.9 years, the median OS duration was 9.7 years (95% CI, 7.8–not reached). Patients whose tumors were classified as ACC-I on the basis of RNA-seq data had a median OS duration of 2.1 years versus 23 years for those classified as ACC-II [HR = 5.68 (95% CI, 2.13–15.10); P < 0.0001; Fig. 2B]. Similarly, patients with a solid histology had a shorter OS duration than did those with a nonsolid histology [5.2 vs. 14.7 years, HR = 2.79 (95% CI, 1.38–5.65); P = 0.0043; Supplementary Fig. S2].

The IMC final data set comprised 507,524 single cells. Figure 3A depicts the cell populations that were explored (for further details refer to Supplementary Table S2). The most prevalent cell populations in the ACC stroma were fibroblasts, followed by myeloid cells and T cells (Fig. 3B). The myeloid population was mostly represented by macrophages. Inside the tumor core, the densest population other than tumor cells was fibroblasts, followed by myeloid cells and endothelial cells (Fig. 3C). Dendritic cells, regulatory T cells, CD4 T cells, CD8 memory T cells, and B cells were overall scarce in the ACC TME. There were no significant differences in TME composition by patients’ age, gender, or primary tumor site.

We then compared RNA-seq deconvolution and IMC analysis in regard to the TME composition in 27 matched tumors. We found that the abundance of T and B cells was significantly correlated with RNA deconvolution and IMC (Fig. 3D). On the other hand, we did not find a significant correlation between macrophages and NK cell markers and techniques; the poor correlation observed in these cell types may be attributed to our IMC panel, which only contained markers for identifying M2 macrophages (CD206 and CD163) and CD56-high NK cells (Supplementary Table S2).

A differential analysis of TME based on ACC-I versus ACC-II subtypes revealed that ACC-I had significantly more epithelial tumor cells (FDR = 0.01), NK cells (FDR = 0.07), endothelial cells (FDR ≤ 0.01), proliferating cells (FDR < 0.01), and B7-H4-expressing tumor cells (FDR < 0.01) than did ACC-II (Fig. 4A). On the other hand, ACC-II had a significantly higher density of myoepithelial tumor cells (FDR < 0.01) and fibroblasts (FDR = 0.02). There were no significant differences in the expression of other immunomodulatory markers such as PD-L1, LAG-3, TIM-3, and TIGIT between ACC molecular subtypes (Supplementary Fig. S3).

Notably, our RNA-seq deconvolution analysis indicated a higher abundance of immune cells [e.g., T cells and cytotoxic (CD8⁺) T cells] in ACC-I. However, a spatial analysis demonstrated that these immune cell populations were significantly more restricted to the stroma and not the infiltrating tumor core in ACC-I than in ACC-II. The tumor–stroma ratio of cytotoxic cells, which represents the average density of cytotoxic T cells inside the tumor core in comparison with the stroma, was significantly lower in the ACC-I versus ACC-II. For instance, there were 11.6-fold more cytotoxic T cells in the stroma than tumor of ACC-I, and 2.2-fold more cytotoxic T cells in the stroma than in the tumor of ACC-II; thus, characterizing an immune-excluded TME in ACC-I (Fig. 4B and C). To further explore TME composition, we analyzed the interaction between B7-H4 expression and specific cell populations. Interestingly, B7-H4 expression was associated with T-cell restriction to the stroma [P = 0.003; coefficient of correlation (R) = −0.37; Fig. 4D].

Because most ACC-I tumors are solid histology, a poor prognostic factor that is routinely reported in ACC pathologic specimens, we compared the TME composition and expression level of immunomodulators in solid versus nonsolid ACC. Our findings revealed that solid ACC had a significantly higher density of epithelial tumor cells, endothelial cells, proliferating cells, and B7-H4 and Bcl-2–expressing tumor cells than nonsolid histology (P < 0.05 and FDR < 0.2; Supplementary Fig. S4). We did not find significant differences in the density of fibroblasts, macrophages, and T cells between solid and nonsolid ACC. Likewise, NOTCH1-activating mutations and pathway activation are known to be enriched in ACC-I and solid histology. We used NICD1 IHC to detect Notch1 pathway activation and observed that tumors with high NICD1 expression also had higher expression of Ki67, B7-H4, and Bcl-2 than did those with low-NICD1 levels (FDR < 0.2; Supplementary Fig. S5).

In terms of the prognostic relevance of TME cell composition, a high density of endothelial cells and immune cells, including T cells, cytotoxic T cells, dendritic cells, and myeloid cells (nonmacrophage or macrophage), was significantly associated with worse OS (FDR < 0.1; Supplementary Fig. S6). In an analysis of cell composition by compartment (intratumoral vs. stroma), the intratumoral density of endothelial cells (FDR = 0.04) and immune cells (FDR = 0.08), specifically regulatory T cells (FDR = 0.02) and myeloid cells (FDR = 0.02), was predictive of a worse OS, while differences in the stromal cell composition and density were not significantly associated with OS. Regarding tumor markers, a high density of myoepithelial tumor cells and fibroblasts was associated with an improved OS, while a higher density of epithelial tumor cells was associated with a poor prognosis (FDR < 0.05). Moreover, higher expression levels of B7-H4, Ki67, and Bcl2 were significantly associated with worse OS (Supplementary Fig. S7A, FDR < 0.01), and the prognostic value of each individual marker remained statistically significant after adjusting for histology (solid vs. nonsolid) and stage (I–II vs. III–IV), as well as a sensitivity analysis with grouped variables (high vs. low; Supplementary Fig. S7B–S7D). The expression of other inhibitory checkpoints such as PD-L1, LAG-3, TIM-3, and TIGIT did not correlate with survival (Supplementary Figs. S6 and S8).

We previously found that B7-H4 was overexpressed at the protein level, as assessed by RPPA in the aggressive ACC-I subtype (6). Consistently, our IMC analysis showed that ACC-I had significantly higher expression of B7-H4 than did ACC-II (P value and FDR < 0.001; Fig. 4A), and a spatial analysis showed that its expression was mainly in the tumor core (99.1%; Supplementary Fig. S9).

In order to implement screening for B7-H4 as a biomarker, it must be assessed through standard clinical assays such as IHC. To assess the feasibility of quantifying B7-H4 by IHC and validate our IMC results, we performed standard IHC staining for B7-H4 (Fig. 5A) in one of the TMAs used for IMC analysis (71 overlapping samples from 37 of 62 patients). IHC staining for B7-H4 revealed a high positive correlation between the percentage of area staining positive by IHC and positive cells determined by IMC (P = 0.01; Fig. 5B). Likewise, ACC-I tumors showed a significantly higher percentage of the tumor staining positive for B7-H4 (P = 4.3 × 10⁵; Fig. 5C), and high staining for B7-H4 was also associated with poor prognosis (P = 0.01, HR = 2.54; Fig. 5D).

B7-H4 has an inhibitory function in T-cell response and has recently emerged as a promising therapeutic target in cancer (14, 15). To determine whether high B7-H4 expression was associated with poor prognosis solely as a result of the exclusion of intratumoral T cells or whether there were additional potential immunosuppressive functions, we analyzed the survival duration of patients, stratified by both intratumoral T-cell and B7-H4 levels. The analysis revealed that B7-H4 was the strongest predictor of poor outcome, even after controlling for stage and histology, and that after controlling for B7-H4 levels, the abundance of cytotoxic T cells was not associated with poor prognosis (Supplementary Fig. S10). Thus, B7-H4 expression remained the strongest predictor of poor survival in our analysis. We also explored the expression of other immune checkpoints according to B7-H4 expression and found that TIM-3 directly correlated with B7-H4 expression; however, TIM3 is expressed at significantly lower levels than B7-H4 and has no prognostic value (Supplementary Fig. S11). Given its strong expression in the poor-prognosis ACC-I subgroup, we explored B7-H4 therapeutic potential preclinically using PDX models.

The antitumor activity of AZD8205, a novel B7-H4 targeting ADC that uses a topoisomerase 1 inhibitor as payload (12), was assessed in three ACC PDX models: two ACC-I (ACCX9 and ACCX11) and one ACC-II (ACCX5M1) (Supplementary Fig. S1) (16). B7-H4 expression, as assessed by IHC (Supplementary Table S3), was high in both ACC-I models (ACCX9 and ACCX11) and low in the ACC-II ACCX5M1 PDX (Fig. 6A–C). A single dose of AZD8205 significantly inhibited tumor growth and led to a regression in 100% (n = 30) of ACCX9 and ACCX11 tumors by day 60. Complete responses were obtained in 90% of mice for both ACCX9 and ACCX11 at 3.5 mg/kg, and 70% of mice exhibited complete responses at 1.25 mg/kg in the ACCX9 model (Fig. 6D and E). No effect was observed following treatment with a nontargeted isotype control bearing the same topoisomerase 1 inhibitor payload. Moreover, AZD8205 failed to inhibit tumor growth in the ACC-II/B7-H4 low ACCX5M1 PDX model (Fig. 6F). These results support B7-H4 as a therapeutic target for ACCs with high expression of B7-H4.

This study is the first to comprehensively assess the composition and spatial distribution of the ACC TME through IMC, a highly multiplex imaging method, and integrate the TME findings with genomic and clinical data. Our results elucidate the attributes of the TME that are associated with distinct ACC molecular subtypes (ACC-I vs. ACC-II) and reveal the localization of independent prognostic markers and a potential therapeutic target for the most aggressive ACC.

The composition and activation status of the TME has crucial biological and therapeutic implications in solid tumors (17). The immune microenvironment of ACC was explored using bulk RNA-seq analysis in a Chinese cohort (n = 75; ref. 18) and an additional smaller cohort (n = 20; ref. 19) and is reported to be overall “cold,” characterized by rare T-cell infiltration and low levels of PD-L1. The results of our RNA-seq deconvolution analysis of the ACC immune environment corroborated these findings. We found a scarce density of antigen-presenting cells and T cells in most ACC tumor samples; however, in our cohort, 25% to 30% of ACC samples harbored high levels of immune-inflammatory cell populations, and most were categorized as ACC-I subtype.

Notably, consistent with most solid tumors, our IMC analysis revealed that the most common cell population in the ACC TME is fibroblasts. Fibroblasts comprise a highly heterogeneous cell population that has a wide range of functional roles in cancer, with cancer-promoting or tumor-restraining properties (20–27). In our study, a high density of fibroblasts was found in the less aggressive molecular subtype (ACC-II), suggesting that fibroblasts have a protective role in ACC.

The second most common cell population in the ACC TME was myeloid cells, mainly macrophages. Macrophages play key roles in immunoregulation, neoangiogenesis, and tumor migration and are considered highly plastic; interactions between tumor cells and the stroma directly modulate the macrophage phenotype, which is commonly categorized into two major poles: proinflammatory (M1) and protumoral (M2), with a broad spectrum between them (28–30). We did not find significant differences in macrophage density between the two ACC molecular subtypes; nevertheless, a higher macrophage density was overall associated with poor survival duration. Our study was limited to explore macrophage subpopulations, given the lack of an M1-polarized marker. Mostly, the macrophages reported in our study were classified as M2 (CD163⁺/CD206⁺).

Interestingly, our RNA-seq deconvolution analysis revealed that ACC-I had significantly more immune-inflammatory cells than ACC-II, which was subsequently confirmed by an IMC analysis. However, IMC showed that most immune cells in ACC-I, including cytotoxic T cells, were significantly restricted to the stroma and rarely infiltrated the tumor core. Tumor biology is inherent to the spatial distribution of cell and functional markers, which cannot be inferred by bulk RNA-seq. Our IMC analysis was crucial to characterize the immune-excluded TME in ACC-I compared with that in ACC-II and poses the hypothesis that in addition to intrinsic tumor characteristics, immune evasion may contribute to cancer progression in the aggressive ACC-I subtype.

Several mechanisms can be leveraged by tumor cells to overcome antitumor immune responses, including expressing inhibitory immunologic checkpoints (31). Although the most studied immune checkpoints, such as PD-L1 and CTLA-4, have rarely been found to be expressed in ACC (32, 33), we found high levels of B7-H4 in tumor cells, particularly those with the poor-prognosis ACC-I subtype. B7-H4, a member of the B7 family for which a ligand has not yet been identified, is abnormally expressed in cancer cells from several solid tumors and has been identified as a potential mechanism of immune evasion through a molecular shield model, as proposed for PD-L1 (14, 34–38). B7-H4 may also inhibit immune infiltration into the tumor by suppressing antitumor T-cell immunity and increasing the proliferation of regulatory T cells (39–41). Previous studies have linked high expression of B7-H4 to poor clinical outcomes (42–44). Consistently, we found a strong correlation between high expression of B7-H4 and worse overall survival, which remained statistically significant even after histologic and stage adjustments.

Given its role in promoting tumor immune evasion and its overexpression in tumors compared with in most normal tissues, B7-H4 is being explored as a therapeutic target for the treatment of cancer. In vivo and in vitro studies have shown that anti–B7-H4 antibody treatment can stimulate IFNy secretion and promote T-cell proliferation, which culminates in the rescue of T-cell antitumor responses (45–47). In this study, we choose B7-H4 as a potential therapeutic target because it was highly expressed in ACC-I, had a strong association with poor prognosis, and is a drug target in active clinical development. We treated two ACC-I subtype PDXs (ACCX9 and ACCX11) that express high levels of B7-H4 with AZD8205, a B7-H4 ADC that is currently in clinical development (NCT05123482). Consistent with its mechanism of action, intracellular delivery of a TOP1i payload to B7-H4–positive cells, leading to DNA damage and apoptotic cell death, a single dose of AZD8205 led to 100% tumor response in both B7-H4 high ACC-I models, but it did not lead to tumor growth inhibition in the B7-H4 low ACC-II model. Notably, ACC with higher levels of B7-H4 expression also had a higher density of proliferating tumor cells (Ki67-positive), and the topoisomerase 1 inhibitor payload is more likely to affect dividing cells. Of note, potential immune effects induced by the B7-H4 antibody that may further improve its efficacy in patients were likely missed in our PDX experiments because we used immunocompromised models. Future studies exploring the role of B7-H4 as an immune checkpoint using specific anti–B7-H4 antibody in immune-competent murine models are needed (47).

The limitations of this study are the preselected antibody panel that we used to explore the TME by IMC and the lack of markers for phenotyping subpopulations of macrophages and fibroblasts. Our IMC panel was selected to represent all of the main cellular components of the TME, particularly the immunocytes, and the functional markers were chosen on the basis of our preliminary RNA-seq deconvolution analysis and previously published RPPA findings (6). IMC allows for up to 37 heavy metal–tagged antibodies to be used during the same stain; however, the metal signal and pH can limit the performance of antibodies. In our panel, 9 of 37 antibodies were not successfully validated, including NICD1, which was assessed by IHC staining in a subset of tumors. Another potential limitation of our study is the intratumoral heterogeneity, which could lead to distinct TMEs in different parts of the tumor, although we tried to account for this by selecting two areas of each sample. Lastly, patients’ cohorts were retrospectively collected, which can be associated with selective bias.

In summary, this study provides insight into ACC TME composition and spatial distribution using multiplex spatial proteomic profiling. Integration of the TME with proteogenomic and clinical data revealed distinct TME attributes according to ACC molecular subtype and histologic type. Our study revealed that B7-H4 expression is related to an immune-excluded TME and is an independent poor prognostic factor in ACC. Notably, our preclinical findings support B7-H4 as a promising target for the most aggressive ACC subtype, which represents a major unmet clinical need and provides a scientific rationale to conduct clinical trials of B7-H4–targeting agents in biomarker-selected ACCs.

Z.A. Cooper reports other support from AstraZeneca during the conduct of the study; and is an employee of AstraZeneca and may hold stock ownership or stock options or interest in AstraZeneca. K. Kinneer reports personal fees from AstraZeneca outside the submitted work; and has a patent for WO2022/053650 A1 pending. R. Ferrarotto reports personal fees and other support from Prelude Therapeutics and Merck Serono; personal fees from Bicara Therapeutic, Regeneron, Sanofi, Elevar Therapeutics, Guidepoint global, Remix, Intellisphere, and G1 Therapeutics; personal fees and nonfinancial support from Ayala Pharmaceuticals; and other support from EMD Serono, ISA, Genentech/Roche, Merck Serono, Pfizer, and Viracta outside the submitted work. No disclosures were reported by the other authors.

L.G. Sousa: Data curation, formal analysis, investigation, writing–original draft, writing–review and editing. D.J. McGrail: Data curation, formal analysis, writing–original draft, writing–review and editing. F. Lazar Neto: Data curation, formal analysis, investigation, writing–review and editing. K. Li: Methodology, writing–review and editing. M.L. Marques-Piubelli: Data curation, writing–review and editing. S. Ferri-Borgogno: Methodology, writing–review and editing. H. Dai: Methodology, writing–review and editing. Y. Mitani: Methodology, writing–review and editing. N. Spardy Burr: Investigation, writing–original draft, writing–review and editing. Z.A. Cooper: Investigation, writing–original draft, writing–review and editing. K. Kinneer: Investigation, writing–original draft, writing–review and editing. M.A. Cortez: Writing–review and editing. S.-Y. Lin: Writing–review and editing. D. Bell: Conceptualization, writing–review and editing. A. El-Naggar: Writing–review and editing. J. Burks: Methodology, writing–review and editing. R. Ferrarotto: Conceptualization, resources, supervision, writing–original draft, writing–review and editing.

Daniel J. McGrail, Kaiyi Li, Yoshitsugu Mitani, Adel K El-Naggar, and Renata Ferrarotto were supported by DOD grant number W81XWH-21-0409; R. Ferrarotto was supported by Adenoid Cystic Carcinoma Research Foundation (ACCRF); R. Ferrarotto was supported by Wold Foundation and DeWayne Everage Adenoid Cystic Carcinoma Fund; and Daniel J. McGrail was supported by NCI R00 CA2406892. We thank Editing Services, Research Medical Library, for editing this article and Kathryn Brayer (University of New Mexico) for biostatistical analysis.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.