Adrenocortical carcinoma (ACC) is a rare highly aggressive malignancy ⁠of which >80% are steroidogenic (1–4)⁠⁠. Overt hypercortisolism is observed in 37.6% (197/524) of adults with ACC, excluding cases with elevated cortisol levels without clinical manifestations, which may also have some molecular impact (4)⁠. The phenotypes seen with molecular changes in ACC may be largely derived from the dominant co-secretion of cortisol and androgens. These differences in combinations of steroid underlie the importance of defining steroid phenotypes and identifying the relevant genetic alterations that characterize the various immune system profiles observed.

Pan-genomic and molecular characterizations of ACC have been the focus of diverse studies in an attempt to establish useful classifications for prognosis or therapy. Zheng et al. (5)⁠ suggested the differentiation of patients with ACC based on three CpG island hypermethylated phenotype (CIMP) statuses: high, intermediate, and low. To date, CIMP is the most robust biomarker for ACC pan-molecular subgroups (6)⁠. Despite providing meaningful prognosis, this classification cannot be translated into determining optimal target treatment options (6, 7)⁠. Besides DNA methylation status, other molecular strategies have been used to characterize ACC subtypes (5, 8, 9)⁠. Based on transcriptome profiling, de Reyniès et al. (8)⁠ and Giordano et al. (9)⁠ defined two ACC subgroups, C1A and C1B, while Zheng et al. (5)⁠ clustered ACC mRNA-seq data in two subgroups with a high steroid phenotype (HSP) and low steroid phenotype (LSP), separating the expression pattern of genes in the steroid synthesis pathway. This was further refined into four subgroups: HSP (n = 25), HSP + proliferation (n = 22), LSP (n = 27), and LSP + proliferation (n = 4). The proliferation profile was mainly associated with HSP, whereas the LSP + proliferation was restricted to only four cases in The Cancer Genome Atlas (TCGA) cohort (n = 78). These four subgroups demonstrated a high overlap of C1A and C1B subtypes, with most HSP patients classified as C1A and LSP patients classified as C1B (5)⁠. Zheng and colleagues found an immune signature associated with the LSP profile upon comparison of immune-suppressed ACC with strong steroidogenesis (5)⁠.

Landmark studies have extensively focused on profiling biomarkers and cells associated with cancer immunity with extensive immunogenomic analysis of tumors (10, 11)⁠. Thorsson et al. (11)⁠ provided new insights into the immune landscape of cancers using TCGA data. Assembling all ACC cases without consideration of more specific clinical or steroidal phenotypes, ACC was found to exhibit the second lowest leukocyte fraction among the cancer types studied (11). These investigators analyzed over 10,000 TCGA tumors and using cluster analysis of a robust list of immune expression signatures reported by other studies proposed six immune subtypes with specific features and prognoses: C1, wound healing; C2, IFN-γ dominant; C3, inflammatory; C4, lymphocyte depleted; C5, immunologically quiet; and C6, TGF-ß dominant (11). Based on the dominant sample characteristics, they proposed the ACCs in the TCGA database (patient ages ranging from 14 to 77 years) fit better within the C4 immune subtype, with a more prominent macrophage signature, suppression of T helper type 1 (Th1) cells, and a high M2 response (11). These cancers are among those with the least favorable outcome. This immunosuppressed profile is associated with hypercortisolism and low survival in the majority of adult patients with ACC (12)⁠, while younger children with a dominant androgen steroid phenotype have a better prognosis (13, 14)⁠.

As tumors grow, they acquire mutations. These mutations may impair the function of the immune system in ACC subtypes (with both low and high steroidogenesis). These genetic changes include mutations in driver genes, DNA damage, copy number variation, aneuploidy, loss of heterozygosity, intratumor heterogeneity, and decreased telomere length (5, 15)⁠. Consequently, overall survival (OS) analyses are frequently less accurate in ACC cases because of the complex interactions between these changes in the tumor and the immune system (16)⁠. The worse prognosis for adults with ACC is explained by molecular differences and the higher frequency of elevated glucocorticoids (17–19)⁠. After C4 (49 of the 78 ACC cases), the second most common immune subtype was C3, found in 23 of the 78 ACC cases, predominantly in LSP (16/31). The C3 immune subtype is represented by an inflammatory response with higher presence of Th1 and Th17 responses and a balanced presence of macrophages and lymphocytes. In a pan-cancer analysis, the C3 immune subtype presented the best prognosis. The other six ACC cases were C5 (n = 3) and one each of subtypes C1, C2, and C6.

Lymphocyte expression signature, high number of unique T-cell receptor clonotypes, cytokines produced by activated CD4+ Th1 and Th17 lymphocytes, and M1 macrophages, which secrete pro-inflammatory cytokines and chemokines, are associated with improved OS in some cancer subtypes (11). The ability to kill transformed cells is essential for an effective anti-tumor response by the immune system. This ability relies on the cytotoxic activity of natural killer (NK) cells and CD8+ T lymphocytes (CD8+TL) (20)⁠. Activated NK cells are able to recruit dendritic cells and elicit an antitumor T-cell response (21)⁠. The presence of activated tumor-infiltrating CD8+TL in human cancers is associated with improved prognosis, expression of specific antigens, and increased immune-cancer interactions (10)⁠. For example, there is a positive correlation between melanoma-associated antigen 3 (MAGEA3) expression and CD8+TL infiltration in melanoma, explaining the success of the MAGEA3-based vaccine (22)⁠. Another successful example of immunomodulation is the stimulation of blood cells with bacille Calmette-Guérin (BCG) or recombinant BCG to enhance the expression of CD25 and CD69 on human CD4+ T cells, a highly effective immunotherapeutic agent against bladder cancer (23, 24)⁠. Interestingly, most patients with bladder cancer are diagnosed in their seventies when the immune system may be impaired (25)⁠. Similarly, more than 50% of adults with ACC are diagnosed in patients 50 years or older (26)⁠. As CD8+TL are crucial for protective immunity, it is vital to identify candidate activators of the host immune response to cancer. Specific driver mutations (e.g., TP53) are correlated with higher leukocyte levels across all cancers (11). This is not the case in pediatric ACC with the germline TP53 R337H variant, where higher CD8+TL infiltration is associated with other factors, such as stage 1 and diagnosis at a younger age (14)⁠. This is in line with the observed changes in the immune infiltrate composition at each tumor stage, where there is a decrease in T-cell density and an increase in T-follicular helper cells and innate immune cells with tumor progression (27)⁠.

T-cell activation requires two distinct signaling pathways, one mediated by the binding of T-cell receptor (TCR) with an antigen bound major histocompatibility complex (MHC) and the second by the binding of CD28 to T cells with B7-1 (CD80) or B7-2 (CD86) of antigen-presenting cells (28)⁠. In addition to B7-1 and B7-2, other B7 family proteins interact with other T-cell membrane molecules and may act as co-inhibitors, such as B7-H1, which is well known as programmed cell death ligand 1 (PD-L1), B7-DC, which is known as programmed cell death ligand 2 (PD-L2), and B7-H3 (CD276), among others (29)⁠. Blockade of immune checkpoint evasion was expected to provide a more effective treatment for patients with ACC. However, the use of antibodies targeting programmed cell death 1 (PD-1) expressed by T cells (30–32)⁠ or PD-L1 (33)⁠ is associated with poor efficacy. In the latter study (33)⁠, 50% of patients were under concomitant treatment with mitotane to block immunosuppressive glucocorticoids. However, this study did not consider the possibility of evasion through PD-1/PD-L2. Despite these setbacks, other approaches have been suggested to overcome the reported modest responses (7, 34)⁠.

The impact of excess steroids on the immune system interaction with ACC tumors, including a complete understanding of the related molecular profiles, has not been fully evaluated. Our aim was to perform a comprehensive and translational analysis of ACC immune biomarkers to identify those that could predict a favorable outcome with and without steroid interference. We evaluated possible modifiers of immune evasion (CD274 that encodes PD-L1 and PDCD1LG2 that encodes PD-L2), and investigated how these ligands correlate with CD8B expression in ACC, revisiting TCGA cohort data and the molecular steroidogenic classification established by Zheng et al. (5). Considering the hypothesis that HSP has a worse prognosis because of its effect on the immune system, we discuss strategies, supported by prior studies, for downregulating steroid biosynthesis before using possible immune activators.

We present a complementary approach to the analyses of the pioneering studies by Zheng et al. (5) and Thorsson et al. (11) and revisit the potential biomarkers underlying the immunological processes in ACC, considering important contributions from previous studies (7–9, 12)⁠. The interplay between steroid production and the immune response in ACC has been a topic of recent studies (5, 12, 14, 18, 19)⁠. We have uncovered further interesting findings from the ACC datasets, deepened the investigation of immune pathways, and revealed possible druggable targets that may guide the selection criteria for ACC-targeted immunotherapy.

We demonstrated the immune system capabilities were dramatically better in patients with LSP ( Figures 1 and 2 ), which is in agreement with studies describing the immune suppression caused by excess glucocorticoids (12)⁠. This higher immune activation seen in a subset of patients with ACC was in contrast with worst outcomes previously demonstrated for ACC in a pan-cancer analysis of solid tumors, which considered all patients with ACC without distinction of subtypes (5, 11). Our analyses, which considered steroid phenotype, improved our understanding of how the immune system reacts to ACC.

The immune cell composition in the TME is associated with important prognostic and therapeutic characteristics (14, 39)⁠. Our analysis of TME was based on scores inferred by previous studies (11, 39)⁠ and used two different methods, xCell and Cibersort, to detect immune cells according to bulk RNA-seq data. Some observed divergences (e.g., native CD4+ T cells) may be explained by limitation of the methods, such as background prediction or the spillover effect, as discussed by Sturm et al. (39)⁠. However, a higher immune cell infiltrate was observed in general for patients with LSP compared with that for patients with HSP ( Figures 5A, B ). Furthermore, both methods showed a significantly higher presence of cells associated with cytotoxic activity ( Figure 5B ), such as CD8+ T cells (p = 4.3 × 10⁻⁸ for xCell and p = 7 × 10⁻⁸ for Cibersort scores), T NK cells (p = 1.1 × 10⁻⁵ for xCell scores), and activated NK cells (p = 3.8 × 10⁻⁶ for Cibersort scores).

Notably, we observed an activation of immune system-related genes ( Figure 6A ), as well as expression of immune evasion-related genes in patients with LSP ( Figures 6A and 8 ). We also demonstrated a positive correlation between CD8B and CD274 (PD-L1) expression (R = 0.61, p = 1 × 10⁻¹¹) and CD8B and PDCD1LG2 (PD-L2) (R = 0.77, p = 8.6 × 10⁻¹⁸) ( Figure 8 ). This may indicate that selective pressure induced enhanced immune evasion by tumor cells, which is in agreement with recent studies (51)⁠. PD-1/PD-L1-related and PD-1/PD-L2-related immune escape mechanisms are known to downregulate the CD8+ T-cell activation process. CD274 is expressed by other immune cells in greater a proportion than that of PDCD1LG2 expression (58)⁠, which may help explain why the dispersion seen in scatter plots was more evident for CD274. In addition, patients with HSP ACC overall presented lower levels of CD8B and CD274/PDCD1LG2 expression compared to those of patients with LSP ACC. However, the positive correlation observed between CD8B and CD274/PDCD1LG2 expression may imply that following glucocorticoid suppression, mechanisms of immune evasion may be induced in patients with HSP ACC. The correlation between CD8+TL and PD-L1 expression levels cannot be determined using immunohistochemistry (14, 59)⁠. However, findings from a previous study (59)⁠ and our current study suggest that PD-L2 plays a greater role than that of PD-L1 in ACC cells ( Figure 6A ). Nevertheless, this hypothesis has not been assessed in a clinical trial. Anti-PD-1 clinical trials conducted without inhibiting steroidogenesis (30–32)⁠ may have had modest results due to the immune suppression caused by hormone production in the TME. On the other hand, Le Tourneau et al. (33)⁠ evaluated the concomitant suppression of steroidogenesis with and without mitotane administration. However, only anti-PD-L1 was used in their study and they did not consider the possibility of PD-L2 being an important immune checkpoint in ACC.

In agreement with recent findings showing that 90% of ACC samples (n = 48) were positive for B7-H3 (CD276) in immunohistochemical assays (60)⁠, our analysis demonstrated that CD276 overexpression was detected in both LSP and HSP. Remarkably, its expression was considered high in a pan-cancer comparison of most ACC patients ( Figure 7B ). Moreover, CD276 overexpression has been implicated in resistance to anti-PD1 therapies (61–63)⁠. B7-H3 is a member of the B7 ligand family and is a recent and attractive target for antibody-based immunotherapy. B7-H3 is differentially overexpressed in malignant cells at high frequency (60% of 25,000 tumor samples), is detected at low levels in normal tissues, and probably has an inhibitory role in adaptive immunity, suppressing T-cell activation and proliferation (60–62)⁠. B7-H3 modulates T-cell responses, but unlike PD-1 and CTLA-4, may inhibit T-cell activation, possibly by interacting with tumor necrosis family receptors (64)⁠. However, other investigators believe the B7-H3 receptor remains unknown (61). Recent studies have shown that blocking B7-H3 results in a response of tumor resistance to treatments with anti-PD1/PD-L1 antibodies (61)⁠. Yonesaka et al. (63)⁠ demonstrated the dual blockade of B7-H3 and PD-L1 enhances the antitumor reaction compared to that of PD-L1 blockade alone, which may indicate a role of B7-H3 overexpression in immune evasion. These findings suggest that anti-B7-H3 combined with anti-PD-1/PD-L1 antibody therapy is a promising approach for B7-H3-expressing cancers, such as ACC. In addition to CD276 (B7-H3), FLT1 (VEGFR1) appeared in both C3 and C4 networks ( Figure 7A ) and was highly expressed in ACC compared with that in other cancers ( Figure 7B ). This may imply a role in angiogenesis, which has been described as an important pathway in ACC (65, 66)⁠. However, clinical trials to date using VEGFR inhibitors for ACC have demonstrated limited results (67, 68)⁠.

In contrast to CD276 (B7-H3) and FLT1 (VEGFR1), BTLA, interleukin 10 (IL10), IL10 receptor subunit alpha (IL10RA), integrin beta chain-2 (ITGB2), CD274 (PD-L1), and PDCD1LG2 (PD-L2) are significantly downregulated in HSP ( Supplementary Table 1 , Figures 6A and 7B ). Similar to PD-1 and CTLA-4, BTLA is an immune checkpoint associated with suppression of the immune response and its blockade has been related to a reduction of ovarian carcinoma via the regulation of IL6 and IL10 (69)⁠. IL10 mediates immunosuppressive signals through its interaction with IL10 receptors, leading to inhibition of synthesis of proinflammatory cytokines (70)⁠. Integrins, such as ITGB2, play significant roles in cellular adhesion and cell surface signaling, and also participate in leukocyte adhesion and extravasation to inflame tissues, and mediate the formation of immunological synapses (71)⁠.

Curiously, TNFSF9 (4-1BB-L) was present at a surprisingly high abundance in ACC when compared to that of other TCGA tumor types, but not its receptor TNFRSF9 (4-1BB), which was present mainly on CD8+ T cells, ( Figure 7B ). In the C4 network, TNFSF9 appeared in concordance with dendritic and NK cells ( Figure 7A ). TNFSF9 is commonly expressed by antigen-presenting cells and its interaction with its receptor TNFRSF9 tends to stimulate T cells and cytotoxic activity, demonstrating strong anti-tumor activity (72)⁠. Although the use of TNFSF9 analogues as an immunotherapy strategy has been considered (73)⁠, the relatively low level expression of its receptor TNFRSF9 on T cells may limit the efficacy of this type of treatment. Understanding the role of TNFSF9 overexpression in ACC may be a subject of future studies.

Classification of ACC into subgroups based on methylation profiles, which was also proposed by Zheng et al. (5), appears to be robust as reviewed by Crona and Beuschlein (6). However, despite providing good clinical prognostic indications, this classification has not resulted in novel treatment possibilities for each of the groups (6, 7, 74)⁠. One strong point of our current study using the molecular classification of steroidal profiles was the possibility of separating the immune activation profile of tumor subtypes. This is in agreement with the suggestion by Fiorentini et al. (7)⁠ that patients who may benefit from immunotherapies should be selected based on some type of molecular classification. We believe that separation based on steroid profiles may provide useful information regarding this criterion. Furthermore, the LSP classification successfully grouped all patients with higher expression of immune modulator genes, in addition to higher leukocyte fractions. Remarkably, the immune activation distinction was not clear when using the classification of cases based solely on clinical parameters, that is, hypercortisolism versus non-hypercortisolism (data not shown). This was despite its good correlation with HSP (Fisher’s exact test, p = 2.55 × 10⁻⁴). In addition, the available HSP data are associated with downregulation of the immune system, higher tumor proliferation, and worst prognosis (75)⁠.

Genomic and epigenetic changes frequently occur in ACC, in both childhood (15, 76)⁠ and adulthood (5, 75, 77) cases. However, the TCGA ACC cohort is mainly limited to adults (median age, 46 years). The accumulation of these alterations may alter lymphocyte recruitment (11) and drive a higher proliferation profile, which would consequently result in a worse prognosis. It is currently not clear if the anti-tumor immune response is not strong enough to surpass tumor proliferation or if there is a surge in proliferation due to the immunosuppressed TME.

Interestingly, it was not possible to determine a clear correlation between leukocyte fraction and mutations in driver gene known to impact the immune response. This could have been due to the small number of patients with these mutations, as well as the fact that most of these patients were from the immunosuppressed HSP group. In analysis of the whole cohort (LSP + HSP patients) CTNNB1 mutations were noted only in patients of the HSP group and they demonstrated a weak negative correlation with leukocyte fraction (p = 3.8 × 10⁻² on Wilcoxon test). However, when analyzing only the HSP group, the presence of CTNNB1 mutations did not demonstrate any significant impact on leukocyte fraction (p = 9.9 × 10⁻¹ on Wilcoxon test), indicating the weak correlation observed was mainly due to the steroid phenotype. This indicates the steroid phenotype should be taken into account during immune response analysis of ACC. Failing to do so may generate confounding effects.

When comparing the inflammatory C3 and lymphocyte-depleted C4 immune signatures, we found a slightly better survival for C3 (p = 3.2 × 10⁻² for OS and p = 4.7 × 10⁻² for PFI; log-rank test) ( Figure 9B ). In the pan-cancer analysis performed by Thorsson and colleagues, C3 had a significantly better outcome than that of C4; however, the steroid phenotype provides more characteristics regarding tumor aggressiveness and outcome prognosis in ACC than that of the C3-C4 signatures. For example, there were no death events among the 16 patients with C3 LSP, whereas there were 3 deaths among the 7 patients with C3 HSP (42.9%). Furthermore, among the 12 patients with C4 LSP, only 1 died (8.33%) compared with 19 deaths in the group of 37 patients with C4 HSP (51,35%). When distinguishing C3 and C4 patients within the steroid phenotypes (density plots in Figures 5B , 6B ), differences were observed in LSP C3 versus C4, especially in the leukocyte fraction and lymphocyte infiltration scores; however, HSP C3 versus C4 does not in general present great distinction. Considering there were only seven patients with HSP C3, the steroid phenotype distinguished some immune features better than that of the C1–C6 subtype classification alone.

In addition to differences in the targeted immune checkpoints, a crucial difference affecting the success of immunotherapies for other tumors (22–24, 78)⁠ and the modest results of immunotherapy in ACC, such as less than 15% of patients with ACC benefiting from PD-1 or PD-L1 inhibition (30–33)⁠, seems to be the immunosuppressive and anti-inflammatory effects of intratumoral glucocorticoids (12)⁠. Among potential prognostic factors, CD3+ and CD8+ cytotoxic T lymphocytes were found in 84% of 146 patients with ACC and were associated with overall and recurrence-free survival (12)⁠. This parameter could be further improved by blocking steroidogenesis in combination with different immune checkpoint inhibitors. Patients with tumor progression and mitotane-resistance may still benefit from a reduction in the dose aiming for steroidogenesis blockade. One important feature of mitotane treatment is the ability to block steroid biosynthesis at lower daily doses and with less toxic side effects (approximately <10 µg/mL of plasma) (79)⁠. Future studies may investigate if this blockade is adequate to induce an efficient immune response.

Despite the reduced number of cases, we found evidence of considerable immune cell infiltrates within LSP and sought to characterize the interactions and networks in order to identify transcriptomic biomarkers associated with significant immune activation. Our analysis also exploited target immune modulators that may regulate potential pathways associated with the anti-tumor response. Our analysis framework had limitations as it relied on a small number of accessible profiles and a simple binary change (LSP and HSP) where HSP combined different steroid subtypes that could not be identified using only the clinical manifestations of hypercortisolism. It is important to note that the activation of steroid biosynthesis pathways seen in HSP was based on a transcriptome profile, which may imply a possible hormone excess in the TME had not been related to clinical manifestations. Our experimental design using a probabilistic approach to identify potential targets for immunotherapy was restricted to in silico findings. Despite the rationale, this may not be translatable to in vivo assays. Further work is needed to fully understand the possibility of changing the outcome of ACC immunotherapy, which may start with the selection of patients with LSP ACC and the inhibition of steroidogenesis in patients with HSP ACC. Despite significant progress in identifying immunogenic biomarkers in ACC through evaluation of the TCGA cohort (5), there are still many limitations and puzzles to be solved. Additional efforts beyond RNAseq analyses are needed to validate functional studies to define the key pathways in a larger number of patients with LSP ACC. Considering we are still in the early stage of understanding the B7-H3 network, it is important to define the actual B7-H3 receptor in order to fully understand this pathway in ACC.

An important frontier of ACC therapeutics to improve the anti-tumor activity of the immune system starts explicitly by modeling steroidogenic differences. The analytical approach we describe herein requires further studies to ascertain the possible druggable role of B7-H3 in relation to PD1-PDL1/PDL2 in ACC. Our present approach was unable to predict improved outcomes of immunotherapy; however, it points to possible selection criteria and potential targets. Future work should confirm these findings using a larger cohort, in addition to including pre-clinical or clinical trials.

All scripts, datasets, software and algorithms used to generate results, figures, and tables for this study are available on the GitHub repository (https://github.com/sysbiolab/Sup_Material_Muzzi2021) and Supplementary Tables 3 and 4 .

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

The studies involving human participants were reviewed and approved by Pequeno Príncipe Ethics Committee. Written informed consent from the participants’ legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements.

All authors made substantial contributions to the conception of the work. JCDM was responsible for data acquisition and analysis. MAAC contributed to the acquisition of data and method analyses. All authors contributed to data interpretation. JCDM and BF wrote the first draft of the manuscript. JMM and MAC helped in code review and creating the public repository. All authors contributed to the article and approved the submitted version.

JCDM, JMM, and MAC receive scholarships from the BIG DATA innovation program from the Associação Hospitalar de Proteção à Infância Raul Carneiro-AHPIRAC (2020). MAAC and JM are funded by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq).

The authors declare the current research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.