Copy-Number Alteration Burden Differentially Impacts Immune Profiles and Molecular Features of Hepatocellular Carcinoma

Liver cancer is the fourth leading cause of cancer-related death and a major health problem globally (1). Among liver cancers, hepatocellular carcinoma (HCC) is the most common form (∼90% of primary liver cancers), with more than 850,000 new cases per year worldwide, and is a type of tumor that occurs more commonly among men (1). Despite recent major advances in the understanding of the molecular pathogenesis of HCC, current therapeutic options remain limited, with curative approaches only available for a minority of patients diagnosed at early stages (2, 3). Of note, around 40% of patients with HCC are diagnosed at advanced stages, where the kinase inhibitors sorafenib and lenvatinib are first-line standard systemic therapies, while regorafenib, cabozantinib, and ramucirumab are administered in second line (2). Only recently, the first-line combination therapy with the immune checkpoint inhibitor (ICI) atezolizumab and the VEGF inhibitor bevacizumab [mAbs against programmed cell death 1 ligand 1 (PD-L1) and VEGF-A, respectively], showed significantly improved survival rates in patients with unresectable HCC compared with the standard-of-care sorafenib in the IMbrave 150 phase III clinical trial (4). Response rates in this trial and in phase I/II trials testing nivolumab and pembrolizumab (anti–PD-1) or combinations of nivolumab and ipilimumab (anti–CTLA-4; refs. 5–7) ranged between 15% and 30% of patients, highlighting the need for biomarkers of response/resistance in HCC and an overall better comprehension of the tumor immune responses (8).

In this regard, we have recently characterized the immune landscape of HCC tumors and identified the HCC immune class, a subgroup of approximately one-third of HCCs whose immune profile is characterized by high tumor-infiltrating lymphocytes (TIL), high expression levels of PD-L1 and PD1, and markers of cytolytic activity (9). In addition, we have described the HCC exclusion class, with exclusion of TILs and other “cold” tumors traits, which is prevalent in approximately 30% of HCCs (10).

Copy-number changes of chromosomal segments, also known as aneuploidy, are a common feature of human cancers and have been proposed as a driving force of tumorigenesis (11). Aneuploidy encompasses broad somatic copy-number alterations (CNAs), involving large portions of chromosomes, and smaller focal CNAs. Recently, it has been uncovered that the loads of broad and focal CNAs differentially correlate with gene expression markers related to hallmarks of cancer, cell proliferation, and immune evasion (12, 13), suggesting that they are involved in the carcinogenic process through distinct mechanisms. While high levels of focal events correlate with proliferation markers (12) and can have increased prognostic value compared with broad CNAs (14), high levels of broad events are strongly associated with markers of immune evasion (in several tumor types) and with a reduced response to immunotherapy [e.g., in non–small cell lung cancer (NSCLC) and melanoma; refs. 12, 13, 15]. Thus, the interaction between the cancer genome and the immune system might be regulated through a general gene dosage imbalance governing over specific gene alterations. Nevertheless, none of the above-mentioned studies, or any other published, have specifically deepened on such analysis in HCC. In addition, recent reports have shown that the intensity and direction of the association between CNAs and tumor immunity may not be universal in all cancer types (15–18), underlining the need to investigate the impact of aneuploidy considering the tumor tissue–specific context.

In a previous work, we assessed whether the number of statistically significant recurrent amplifications or deletions in HCC correlated with the immune classes (9), but a thorough characterization of the CNA burden was unaddressed. Thus, we analyzed the effect of focal and broad CNA loads using a newly described scoring strategy (19) and correlated the scores with molecular and immune features. Our findings suggest that HCCs with high burden of broad CNAs correlated with immune exclusion profiles and proliferation signatures; on the other hand, tumors with lower burdens of broad CNAs were enriched in the HCC immune class (9) and proinflammatory pathways, presented higher cytolytic activity, and exhibited absence of CTNNB1 mutations. Consequently, we propose that the “hot” tumor traits of HCCs with low CNA burdens may make them potential candidates to a sustained favorable response to ICIs. In contrast, the “noninflamed” traits and reduced cytotoxic immunophenotype of HCCs with higher levels of broad CNAs are consistent with the notion that broad events may be implicated in immune escape mechanisms and could be indicators of resistance to immunotherapies, as observed in other cancer types (12).

A total of 452 samples were analyzed (Supplementary Fig. S1A), including a discovery cohort (n = 107; ref. 20) and a validation cohort [n = 345; The Cancer Genome Atlas (TCGA); ref. 21]. The main clinicopathologic features of these two cohorts are summarized in Supplementary Table S1. The HEPTROMIC cohort is characterized by mostly Caucasian patients, median age 66 years, males (78%), with Hepatitis C virus (HCV)-related HCC (46%), at early stages of the disease (86% BCLC 0–A), undergoing resection for single tumors (77%), and with α-fetoprotein < 400ng/mL (85%) in patients with well-preserved liver function (Child-Pugh A, 98% and platelet count > 100,000, 82%). At pathologic examination, 36% of tumors presented microvascular invasion and 24% satellites. Patients from TCGA cohort were characterized by a median age of 61 years, 68% males, with 12% of cases related to HCV infection, at early stages of the disease (75% AJCC-T1 and AJCC-T2), and the majority of them presented a well-preserved liver function (Child-Pugh A, 91% and platelet count >100,000, 94%). At pathologic examination, 35% of tumors presented microvascular invasion. In addition to these two cohorts, previously published premalignant samples (n = 25) and very early HCCs (veHCC, n = 18) were also included in the study (20, 22). Finally, a publicly available transcriptome dataset corresponding to 65 pretreatment samples of patients with cancer treated with anti–PD-1 (23) was used to run the subclass mapping. More details on the cohorts have been included in Supplementary Data.

The HEPTROMIC and TCGA-Liver HCC (LIHC) SNP array data, and the CNApp web tool (19), were used to quantify the individual CNA burdens of each sample, and to generate the broad and focal CNA scores (BS and FS, respectively). Details are provided in the Supplementary Data.

Additional information related to Material and Methods is provided in the Supplementary Data.

To determine the impact of the genomic imbalances in the molecular and immune features of HCC, we analyzed 452 samples profiled by SNP array: 107 HCC samples (HEPTROMIC, discovery cohort) and 345 HCC samples (TCGA-LIHC, validation cohort; Supplementary Fig. S1). Using the validated CNApp resegmentation approach (see Materials and Methods), we first generated each sample's genome-wide CNA landscape, and then categorized the detected chromosomal amplifications and deletions into broad and focal events (12). HCC tumors displayed a median of 6.8 and 8.3 broad CNAs per sample in the HEPTROMIC and TCGA-LIHC cohorts, respectively; and a median of 21.0 and 30.3 focal CNAs per sample. In terms of lengths, broad chromosomal fragments had, respectively, a median size of 69.7 and 62.4 Mb (HEPTROMIC, range of 9.5–243 Mb and LIHC-TCGA, range of 8.8–245 Mb). Focal CNAs had a median length of 1.7 Mb in HEPTROMIC and 0.6 Mb in LIHC-TCGA, and 84% of all focal CNAs were below 12 Mb [ref. 24; 81% in HEPTROMIC (n = 107) and 85% in LIHC-TCGA (n = 345); Supplementary Fig. S1]. Overall, no major differences were identified between the CNA landscapes of the two cohorts (Supplementary Fig. S1), and the profiled landscapes of the most frequent events were consistent with previously published analysis applying the GISTIC algorithm on other HCC cohorts (24–26). Briefly, the most recurrent broad gains in the two cohorts (n = 452) were in 1q (52%) and 8q (47%), and the most frequent broad losses were in 8p (52%) and 17p (45%). We also detected slightly less prevalent broad losses in 4q (30%), 6q (24%), and 16q (28%). CNApp also identified frequent high focal gains, characteristic of HCC, such as the ones located in 5p15.33 [in 63/452 HCCs (14%); involving TERT], in 11q13.3 (>6%; affecting CCND1 and FGF19), or in 6p21.1 (10%; including VEGFA), among others. Despite the similar genomic landscapes, the two analyzed cohorts presented some differences at the clinicopathologic level (Supplementary Table S1). TCGA-LIHC patients were at more advanced stage and presented poorer liver function. Median age was also slightly different [66 in HEPTROMIC vs. 61 in TCGA (P < 2.6E-5)] and etiology distribution differed in the two cohorts, HCV-related HCC in HEPTROMIC accounted for 45% of cases and alcohol-related HCC in TCGA-LIHC accounted for 32% of cases. These differences, however, did not impact the CNA profiles as no significant associations were found between CNAs and the analyzed clinicopathologic variables. Next, we quantified the load of broad and focal CNAs in each sample using CNApp. The power of CNApp at capturing the actual fraction of altered genome through scores has been previously demonstrated using TCGA pan-cancer dataset of more than 10,000 samples (19). To summarize, for each tumor, CNApp provided a BS and a FS accounting for the number, amplitude, and length of broad and focal segments.

To rank the aneuploidy levels of HCC among the cancer spectrum, we conducted a pan-cancer analysis using the CNApp-derived BSs and FSs of 10,635 TCGA samples representing 33 tumor types (ref. 19; Supplementary Table S2). This analysis revealed that in HCC, the BS and FS distributions were intermediate when compared with the other 32 tumor types (Fig. 1A and C).

To explore the impact of the CNA profiles on the molecular and immune characteristics of the tumors, we categorized each sample into high/low-BS and high/low-FS according to the BS and FS quartiles. The threshold for high-BS tumors was set at ≥11 (the top quartile of both TCGA pan-cancer cohort and the discovery cohort) and for low-BS tumors at ≤4 (coinciding with the bottom quartile in the discovery cohort). In terms of FS, low-FS samples were defined as those with FS ≤ 13.5, while high FS was defined as those with FS ≥ 47, considering the quartile values in the discovery cohort. Overall, 17% of the analyzed tumors presented low broad CNA burdens (75/452) and 40% of the tumors presented high loads of broad CNAs (183/452). In terms of focal events, 20% of cases were low-FS (92/452) and 43% were high FS (196/452).

Stratifying HCC tumors according to their BS and assessing the correlation with molecular and immune features revealed that HCCs with fewer broad genomic imbalances (i.e., those with low BS) were characterized by high inflammation (increase of hallmarks of inflammatory response, TNFα, and IL2-JAK-STAT signaling) and IFN signaling (Fig. 2A). Interestingly, these tumors presented enrichment of the HCC immune class and active immune subclass (9) when compared with tumors with higher BS (P = 0.02 and P = 0.04, respectively; Fig. 2A). In addition, low-BS tumors were deficient in S2 (27) proliferative traits (P = 0.006), presented significant activation of different immune-related pathways, and had LXR/FXR/RXR signaling as one of the top canonical pathways (P < 0.001; Supplementary Table S3).

In terms of ploidy, we found 40% of HCCs being polyploid (3.9 ± 0.7 sets of chromosomes on average), and the remaining 60% being diploid (2 ± 0.2 sets of chromosomes). Of note, polyploid tumors were significantly enriched in high-BS HCCs [17/25 (68%) in the discovery cohort, P < 0.001 and 69/158 (44%) in the validation cohort; P < 0.0001; Supplementary Tables S5 and S6]. Low-BS HCCs were mainly diploids (P = 0.002; Supplementary Tables S5 and S6). At the clinicopathologic level, none of the variables assessed (including etiology and tumor stage) were linked to low BS (Supplementary Table S4).

On the other hand, HCC tumors with high burden of broad CNAs presented a completely different molecular profile. High-BS tumors were characterized by hallmarks of mismatch repair and proliferation and presented low immune infiltration, inferred from the tumor transcriptomic profile (Fig. 2A; Supplementary Table S5; ref. 28). Consistently, they were significantly excluded from the HCC immune class (P = 0.02). Pathway analysis further supported these findings, providing cell-cycle control of chromosomal replication and DNA damage as top canonical pathways (Supplementary Table S3). In addition, integrative pathway analysis (IPA) revealed a role for the arginase pathway in these tumors. The parallel evaluation of TCGA cohort confirmed these observations (Supplementary Fig. S2; Supplementary Table S6).

When classifying the tumors according to their FS, we observed that those with high levels of focal events were characterized by classical proliferative features and were associated with poor survival signatures both in the discovery and in the validation cohorts (Fig. 3A; Supplementary Fig. S3A; Supplementary Tables S7 and S8). Consistently, transcriptome-based pathway analysis revealed cell-cycle–related pathways as the top canonical pathways for high-FS tumors. In addition, these tumors were enriched in the HCC proliferation subclass (ref. 24; P < 0.004) and were excluded from the CTNNB1 subclass (ref. 24; P = 0.032). This molecular profile was aligned with CTNNB1 mutations being significantly excluded from high-FS tumors (P < 0.001), while TP53 mutations were enriched among them (P < 0.001; Supplementary Fig. S3A; Supplementary Table S8). In addition, high-FS tumors were positive for the signatures of hepatic progenitors (ref. 29; P = 0.004), CK19 (refs. 30, 31; P = 0.01), vascular invasion (ref. 32; P = 0.02), MET (ref. 33; P = 0.029), and late TGFβ signaling (ref. 34; P = 0.005). At the clinicopathologic level, these tumors were associated with poor cell differentiation (P < 0.001; Supplementary Tables S7 and S8). Interestingly, we did not find a significant association between FS and tumor immune profiles (Fig. 3A; Supplementary Fig. S3; Supplementary Tables S7 and S8). Despite displaying traits of poor prognosis, FS status did not correlate with patient's overall survival. Levels of broad CNAs were also not associated with patient's outcome.

Overall, and as in other cancer types, burdens of broad CNAs, but not focal, have an impact on the HCC immune profile, and this also supports a differential role for focal and broad events as determinants of HCC antitumor immunity.

As mentioned above, low-BS tumors presented hallmarks of inflammatory and IFN signaling and were characterized by immune traits as seen by an enrichment of the HCC immune class (9). Chemokine and cytokine signaling as well as signatures of innate (cytotoxicity mediated by natural killer cells, monocytes, and dendritic cells) and adaptive immunity (cytotoxic T cells, Th lymphocytes, and B- and T-cell receptor signaling) were also enriched in low-BS tumors (Supplementary Fig. S4A). Interestingly, expression levels of immune checkpoints such as programmed cell death 1 (PD-1) and CTLA-4 were higher in these tumors compared with HCCs with intermediate or high BS (Supplementary Fig. S4A). The presence of higher immune infiltrate in low-BS HCCs was confirmed at the pathologic level and by using transcriptome-based independent algorithms (Fig. 2; refs. 28, 35, 36). First, pathologic assessment of tumor immune infiltrates in our cohort revealed a trend toward higher immune cell infiltration (P = 0.054) and higher count of tertiary lymphoid structures (TLS; P = 0.06; Fig. 2A and E; Supplementary Table S5) in low-BS tumors. TLS density has been reported to correlate with CD8⁺ and CD4⁺ T-cell density in tumors and its presence is associated with favorable prognosis (37, 38). Second, levels of immune cell infiltrate were inferred from the tumor whole-genome expression profiles (28), and were found significantly increased among HCCs with low CNA burden (low BS; P = 0.0001; Fig. 2B). Third, we adapted the immunophenoscore (ref. 35; Supplementary Table S9) to further dissect the transcriptome of our discovery and validation cohorts and confirmed the above-mentioned immune traits of low-BS HCCs (Supplementary Figs. S2C and S4B). We also inferred the cytolytic activity of each tumor (36) and observed that it was higher in tumors with reduced broad CNA burdens (P < 0.001; Fig. 2C). In addition, we performed a submap analysis comparing the molecular profile of our HCC tumors grouped according to BS with a publicly available molecular print of treated tumors (23), classified according to their response to anti–PD-1 treatment. Remarkably, we only observed a significant association between low-BS tumors and tumors responding to anti–PD-1 therapy (P = 0.01; Fig. 2D). Furthermore, we identified significantly elevated expression of HLA-A molecules in BS-low tumors when compared with the rest (P < 0.001; Supplementary Fig. S4). Finally, pathway analysis revealed several ILs and IFNγ among the top upstream regulators in low-BS HCC tumors (Supplementary Table S3), as well as a consistent association with different immune-related pathways [antigen presentation (P < 0.001); communication between innate and adaptive immune cells (P < 0.001); differential regulation of cytokines (P < 0.001); IL17-mediated production of immune cells (P < 0.001); and B-cell development (P < 0.001)].

Regarding HCC tumors with a high burden of broad CNAs (high BS), in addition to low immune infiltration, they were significantly excluded from the HCC immune class (n = 2/25; P = 0.02) and the HCC active immune subclass (n = 0/25; P = 0.01; Supplementary Table S5). This was mirrored in TCGA cohort, where high-BS tumors had significantly less HCC immune class patients compared with low-BS tumors (17% vs. 37%; P = 0.006). In addition, in both cohorts, high-BS HCCs were not related to immunosuppressive features (Supplementary Figs. S2C and S4B). This is consistent with the notion that aneuploidy may play a role as a mechanism of tumor immune evasion (12).

In this regard, a previous study analyzing the aneuploidy profile of 12 cancer types (not including HCC) proposed that high levels of CNAs were drivers of immune exclusion (12). Thus, we assessed whether our scores were associated with immunity in a large spectrum of cancer types. To this end, we correlated the BSs and FSs from 10,635 TCGA samples representing 27 tumor types with a transcriptomic surrogate of the immune infiltrate level: the immune score from the ESTIMATE tool (28). Strikingly, in terms of broad alterations, we observed a significant negative correlation between immune score and BS in 14 of the 27 tumor types analyzed, including HCC (Fig. 1B and D). In terms of focal events, our FS correlated significantly with the immune score in 12 of 27 tumor types, but HCC was not among them (Fig. 1B and D). This places HCC in the subset of tumors were broad aneuploidy events adversely affect immune cell action against tumors (12), but also corroborates that the relation between immunity and CNAs is not universal across cancer types, highlighting the importance of considering the tumor-specific context (15–18).

Following the identification of immune features associated to low-BS HCCs, we sought to explore whether the burden of broad copy-number changes (BS) could also influence the modulation of the immune system at the early stages of hepatocarcinogenesis. To this end, we analyzed 25 premalignant dysplastic nodules and 18 veHCCs (20, 22). Because the preneoplastic lesions presented very few broad chromosomal rearrangements, we categorized them according to the presence or absence of broad CNAs, rather than computing their BSs. Overall, we found the presence of broad alterations in four of the 25 (16%) analyzed dysplastic nodules (“presence” being equivalent to “BS-high”) and it was linked to reduced molecular features of active antitumor immunity (Supplementary Fig. S5). More precisely, BS-high dysplastic nodules displayed trends toward reduced cytolytic activity values (Supplementary Fig. S5A) and toward reduced immune cell infiltration (Supplementary Fig. S5B). On the other hand, lesions without broad CNAs presented enrichment of effector cell features, immune checkpoints, MHC-related components, and suppressor cells (Supplementary Fig. S5C and S5D). Similarly, analysis of veHCCs also revealed a significant negative correlation between BS and cytolytic activity, as well as the expression-derived immune score reflecting immune infiltration (Supplementary Fig. S6).

Overall, our results suggest that there is a continued association between BS and immunity throughout the hepatocarcinogenic process, from dysplastic nodules to veHCCS to more advanced tumors.

The fact that differences in broad CNA burdens were associated with distinct immune profiles suggests that mechanisms related to overall gene imbalance may contribute to determining the activation of molecular and immune pathways in HCC. Here, we assessed the potential contribution of the mutational load (both silent and nonsilent mutations), the TP53 status, the total predicted neoantigens, and the ratio of observed/expected neoantigens in determining the different HCC immune characteristics of specific CNA profiles. For this purpose, we used the genomic profiles of 179 patients available from TCGA HCC cohort (36), that presented an average of reported nonsilent mutations per megabase of 2.5 (tumor mutational burden, TMB), with only 2% of the samples (n = 6/302) with TMB > 10 mutations/Mb (39). Our analysis on the total number of mutations per sample revealed that this was significantly lower among low-BS tumors as compared with the rest (P = 0.023; Fig. 4A). Consistently, pan-cancer studies have described increased mutational burdens among tumors with increased aneuploidy (15). Because the ability to trigger a tumor immune response does not only depend on the number of mutations, but also on their antigenicity, we also calculated the ratio of observed/expected neoantigens. This ratio measures how much the number of observed neoepitopes (defined on the basis of the mutation's potential to bind HLA with high affinity, and taking into account the gene expression) deviates from the number of expected neoantigens (inferred using a pan-cancer empirical model) in each sample (36). Interestingly, we observed that low-BS tumors displayed higher ratios of observed/expected neoantigens when compared with high-BS tumors (P = 0.015) and that BS negatively correlated with this ratio (P = 0.028; Fig. 4B). In terms of mutations, low-BS tumors presented a positive correlation between total number of mutations and ratio of observed/expected neoantigens (P = 0.028), whereas such correlation was not observed in the rest of the cohort (Fig. 4C). Overall, this suggests that, despite presenting lower loads of mutations, low-BS tumors rendered more neoantigens than expected. In contrast, the higher mutational burdens together with the lower ratios of observed/expected neoantigens among high-BS tumors point toward a negative selection of neoantigens that could be rendering higher rates of less immunogenic mutations among tumors accumulating broad CNAs (Fig. 4A–D).

We next assessed whether our observed association between CNA burden and immune infiltration further depended on the mutational status of TP53, given its role in preventing genome mutations. We observed that the total number of mutations (silent and nonsilent) was independent from the TP53 mutational status among high-BS tumors, whereas for the rest of tumors, the number of mutations was higher when TP53 was nonfunctional (P = 0.018; Fig. 4E). A similar observation was made regarding neoantigen burdens, predicted neoantigens were independent from the TP53 status only in high-BS tumors. In the rest of tumors, those with mutated TP53 displayed significantly higher numbers of predicted neoantigens as compared with those with wild-type TP53 (P = 0.0082; Supplementary Fig. S7A). Thus, our analysis suggests that TP53 could have a role in shaping the immunogenicity of tumors that do not accumulate high loads of broad genomic imbalances.

In addition, we assessed whether amplification or deletion of specific immune-related genes could contribute to explain the link between immunity and BS levels in HCC. Screening the CNApp segmentation output, we identified copy-number losses significantly enriched in high-BS tumors from the discovery and validation cohorts, potentially contributing to the observed immune phenotype. These included losses in genes from the antigen presentation machinery, T-cell receptor components, and immune checkpoints (Supplementary Fig. S7). Remarkably, few of these alterations were detected in low-BS tumors. Among others, we found significant losses of HLA-DQB1, coding for the type MHC class II heterodimer, in high-BS tumors [65/158 high-BS HCCs (35%) vs. 6/51 low BS (12%); P < 0.0001]. This was accompanied by a significant reduction of HLA-DQB1 expression in high-BS tumors compared with low-BS tumors in the discovery cohort (P = 0.002) and a trend in the validation cohort. Of note, the loss of HLA-DQB1 could mediate a deficit in cytotoxic cell responses in high-BS HCCs. In this regard, in lung cancer, HLA-DQB1 expression on tumor cells has been positively correlated with CD4- and CD8-positive lymphocyte infiltration (40).

In addition, considering recent evidence revealing the correlation between global demethylation and CNAs across different tumor types (41), and given that this may promote immune evasion in tumors with high copy-number load, we analyzed the methylation data available for our cohorts. In the discovery cohort, we observed that among the 17,090 probes differentially methylated between low-BS tumors and the rest, 13,252 (77%) presented higher methylation levels (Supplementary Fig. S8). Similarly, in the validation cohort, 65,361 (88%) of 74,195 differentially methylated probes displayed higher methylation levels among low-BS samples as compared with the rest (Supplementary Fig. S8). Therefore, our data align with the fact that the accumulation of broad CNAs leads to a widespread hypomethylation (41), and suggest an impact of epigenetic regulation in HCC immune profiles. The fact that low-BS tumors are characterized by high immune infiltration and higher methylation levels aligns with our previous observation that tumors within the HCC immune class present higher overall levels of methylation (9).

Here, we questioned whether CNA gains and losses contributed equally in determining the tumor immune profile. This was motivated by a report in melanoma suggesting that the total number of genes with copy-number losses was a marker of resistance to immune checkpoint–related treatments (42). Similar to melanoma, we observed a significant negative correlation between broad losses and cytolytic activity in HCC, both in the discovery and validation cohorts (P = 0.0004 in HEPTROMIC and P = 0.01 in LICH-TCGA; Supplementary Fig. S9). Thus, the more broad losses the tumors have, the less cytolytic activity they present; and therefore, less chances to respond to ICIs. This was also consistent with our finding that high-BS HCCs are depleted from tumor immune features and suggest that CNA losses may contribute more to this phenotype. No consistent correlations were found in terms of gains (neither broad nor focal) and in terms of focal losses.

CNAs are considered a hallmark of cancer. Recently, pan-cancer studies have described broad and focal CNA burdens as genomic features able to determine tumor immune infiltration and immune exclusion in several cancer types (12, 15). Notably, the strength and direction of this immunogenomic correlations may not be common in all cancer types (16–18), mainly due to the different context of each tumor (43, 44). The impact of CNAs in the setting of HCC has not been comprehensively addressed. Here, we used data from a high-resolution whole-genome SNP array to understand the immune and molecular impact of genomic CNA burdens in HCC, taking into account each tumor's landscape determined using CNApp (19). Similar to other tumor types, in HCC we observed that different broad copy-number landscapes presented distinct immune profiles, while focal CNAs did not impact tumor immunity. Specifically, we observed that HCC tumors with a high burden of broad chromosomal alterations were linked to DNA repair and proliferation signatures and presented immune exclusion. On the other hand, tumors with low levels of broad CNAs displayed features of immunologically “hot” HCCs, higher immune infiltration and cytolytic activity, as well as upregulation of proinflammatory cues. Thus, like it has been described in melanoma (12), we hypothesized that HCCs with low BS might represent potential responders to immunotherapies, whereas high-BS HCCs, potential nonresponders. Our hypothesis is sustained by the fact that (i) tumors responding to anti–PD-1 treatment have a molecular profile similar to low-BS HCCs and (ii) patients with tumors responding to anti–CTLA-4 or anti–PD-L1 correlate with the expression of IFNγ and GZMB, both overexpressed among low-BS tumors (45–47). With regard to focal events, HCCs with a high burden of focal CNAs were characterized by pro-proliferative and more aggressive features, and exhibited no CTNNB1 mutations. Overall, in our study we propose a strategy to classify HCCs (which usually do not present high TMB) into immune/inflamed and excluded/noninflamed tumors based on CNAs, regardless of the mutational load. This classification expands on our proposed HCC immune classification (9, 10). In particular, we propose a BS ≤ 4 (low BS) as a putative marker to identify patients with HCC with inflamed tumors, and BSs ≥ 11 (high BS) for HCCs lacking immune-related features (Fig. 5). The fact that immune traits may be predicted by broad, but not focal CNA loads, is consistent with the above-mentioned pan-cancer study (12). According to this study, tumors with fewer broad copy-number events may sustain higher infiltration of cytotoxic immune cells, and the genetic variability induced in high-BS tumors may provide tumor cells with an advantage to avoid immunologic recognition. However, because no molecular data of patients with HCC treated with immunotherapies are publicly available to date, the predictive value of the scores warrants validation.

We also analyzed the potential mechanisms by which an increase in broad CNA burdens might be associated with exclusion of immune infiltration. It has been described that neoepitopes drive cytolytic activity in tumors (36), and therefore, high mutational loads could render higher neoantigen burdens and greater immune infiltration. In this study, we observed that low-BS HCCs displayed lower burdens of mutations in comparison with high-BS tumors. However, the ratio of observed/expected neoepitopes was lower among high-BS tumors, suggesting that the accumulation of broad CNAs would be linked to an enhanced pruning of neoantigenic mutations, regardless of overall mutation burden. Because the antitumor immune responses are conditioned by the functional presentation of tumor antigens (48), such pruning could contribute to the decreased antitumor immunity features of high-BS HCCs. Of note, an additional reason could be the reduced antigenicity of neoantigenic mutations among high-BS tumors. In other cancer types, it has been reported that not only the number of neoantigens was a determinant of the tumor immune status, but also the ability of neoantigens to induce an immune response (49, 50).

In addition, specific CNAs involving immune-related genes could impact the tumor immune features. In particular, gene losses involved in antigen presentation, found enriched among high-BS HCCs (i.e., HLA-DQB1), could drive the failure to attract effector T cells in these tumors. This, and the fact that high-BS HCCs were enriched in arginine degradation pathways, suggests that these tumors may evade immune destruction by impairing the MHC-mediated antigen presentation, as in lung cancer (48), rather than developing an immunosuppressive microenvironment. Overall, our findings are in concordance with recent studies on NSCLC (TRACERx; ref. 51). Tumors sparsely infiltrated exhibited decreased neoantigen editing indicative of copy-number losses of clonal neoantigens, and dysfunction in the neoantigen-presenting machinery, among the mechanisms of immune evasion (48, 51).

In parallel, the demethylation observed in high-BS tumors could also promote lack of immune infiltration (41). Even though our data suggest potential explanations, further mechanistic studies will be required to determine the causality between tumor CNA burden and antitumor immunity.

On the other hand, the observation that HCCs with high levels of broad CNAs were markedly polyploid, with proliferative features, and associated with poor prognosis signatures, is consistent with a recent publication (52) that proposes ploidy as a prognosis marker.

With regard to focal copy-number events, the fact that high-FS HCCs were more proliferative is consistent with data from other tumor types (12), and has even been demonstrated experimentally through the depletion of CDKN2A in human cell lines (53).

Finally, we observed that our findings in HCC were extensible to preneoplastic lesions and very early HCCs. Although further confirmatory studies using larger cohorts would be required, these data align with the idea that CNAs may contribute to the escape from antitumor immune pressure (12, 54).

Overall, our study revealed differential associations between broad and focal genomic CNA burdens and several molecular features of HCC. Most importantly, our data reinforce the evidence supporting a differential role for broad CNAs in determining antitumor immune profiles in HCC and provide evidence in favor of chromosomal stability as a hallmark of tumor immunogenicity, and consequently, patient response to immunotherapies.

C.E. Willoughby reports grants from Instituto de Salud Carlos III (ISCIII) and the European Social Fund [Sara Borrell Fellowship (CD19/00109)] during the conduct of the study. R. Montal reports personal fees from Servier and Roche outside the submitted work. J.M. Llovet reports grants and personal fees from Bayer HealthCare Pharmaceuticals, Eisai Inc, and Bristol Myers-Squibb, grants from Boehringer-Ingelheim and Ipsen, and personal fees from Celsion Corporation, Exelixis, Eli Lilly, Roche, Genentech, Glycotest, Nucleix, Can-Fite Biopharma, and AstraZeneca outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

L. Bassaganyas: Conceptualization, resources, data curation, formal analysis, supervision, validation, methodology, writing-review and editing. R. Pinyol: Conceptualization, data curation, formal analysis, validation, investigation, writing-original draft, writing-review and editing. R. Esteban-Fabró: Conceptualization, resources, data curation, formal analysis, validation, investigation, writing-original draft, writing-review and editing. L. Torrens: Formal analysis, writing-review and editing. S. Torrecilla: Formal analysis, writing-review and editing. C.E. Willoughby: Writing-review and editing. S. Franch-Expósito: Conceptualization, resources, software, writing-review and editing. M. Vila-Casadesús: Data curation, software. I. Salaverria: Conceptualization, formal analysis, writing-review and editing. R. Montal: Investigation, writing-review and editing. V. Mazzaferro: Resources, writing-review and editing. J. Camps: Conceptualization, investigation, writing-review and editing. D. Sia: Writing-review and editing. J.M. Llovet: Conceptualization, resources, supervision, funding acquisition, investigation, writing-review and editing.

We would like to thank Dr. Carla Montironi for her contribution in the article. This study has been developed, in part, in the Centre Esther Koplowitz from IDIBAPS/CERCA Programme/Generalitat de Catalunya. The results shown here are, in part, based on data generated by TCGA Research Network: http://cancergenome.nih.gov/. J.M. Llovet was supported by grants from the European Commission Horizon 2020 Program (HEPCAR, proposal number 667273-2), the U.S. Department of Defense (CA150272P3), the NCI (P30 CA196521), the Samuel Waxman Cancer Research Foundation, the Spanish National Health Institute (MICINN, SAF-2016-76390 and PID2019-105378RB-I00), through a partnership between Cancer Research UK, Fondazione AIRC, and Fundación Científica de la Asociacion Española Contra el Cáncer (HUNTER, ref. C9380/A26813), and by the Generalitat de Catalunya (AGAUR, SGR-1358). L. Bassaganyas, R. Esteban-Fabró, and R. Montal were supported by a Beatriu de Pinos grant (AGAUR, 2016BP00161), a MICINN fellowship (BES-2017-081286), and a FSEOM-Boehringer Ingelheim grant, respectively. D. Sia was supported by the Klion Young Scientist award and the PhD Scientist Innovative Research award. C.E. Willoughby was supported by a Sara Borrell fellowship (CD19/00109) from the ISCIII and the European Social Fund. J. Camps was supported by grants from the Instituto de Salud Carlos III and cofunded by the European Regional Development Fund (CPII18/00026 and PI17/01304), the CIBEREHD program, and the Agència de Gestió d'Ajuts Universitaris i de Recerca, Generalitat de Catalunya (2017 SGR 1035). S. Franch-Expósito was supported by a CIBEREHD contract.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.