Integrated Molecular and Immune Phenotype of HER2-Positive Breast Cancer and Response to Neoadjuvant Therapy: A NeoALTTO Exploratory Analysis

Amplification and/or overexpression of HER2 defines 15% to 20% of breast cancer cases as HER2 positive. HER2 status, as assessed by IHC/in situ hybridization, represents a unique biomarker for therapeutic decision-making in clinical practice, despite being insufficient for recapitulating anti-HER2 sensitivity/resistance (1). Clinical and biological heterogeneity is a critical aspect, as HER2-positive breast cancer is not a unique disease, but is instead dispersed along the whole breast cancer spectrum from luminal to basal-like subtypes (2).

In addition, the development of biomarkers for tailoring HER2-targeted therapy cannot ignore that anti-HER2 agents exert their activity not only through direct inhibition of the HER pathway, but also indirectly by modulating immune response (3).

HER2-positive breast cancer cases are immunologically “hot” based on assessment of tumor-infiltrating lymphocytes (TIL) in the tumor microenvironment (TME). According to data from a recent meta-analysis, 84% (range, 76–89) of HER2-positive breast cancer cases contained some levels of lymphocytic infiltration; 16% (range, 11%–24%) were considered lymphocyte-predominant breast cancer because their lymphocitic infiltration fraction was ≥50%, and only 9%, all from a single study and characterized by hormone receptor expression, presented without any lymphocitic infiltration (4).

TILs have been associated with favorable prognosis in early as well as the metastatic setting, and with increased pathological complete response (pCR) following neoadjuvant chemotherapy with or without anti-HER2 therapy (5). However, the aspects of the immune response involved in TIL-associated improved outcomes, and tumor characteristics that can modulate this response, are still under investigation. For example, TILs may lose their favorable prognostic value depending on the primary tumor genomic profile (6).

Gene expression profiling (GEP) of bulk breast cancer specimens has led not only to the identification of breast cancer subtypes, but also to genes that are associated with specific tumor-infiltrating immune cells (TIIC; ref. 7). More recently, cell type deconvolution using leukocyte signatures has been proposed as a highly performing approach to estimate TIIC composition of TME (8). GEP affords a unique opportunity to supplement knowledge gaps regarding tumor and host interplay by evaluating the specific characteristics of both, thus overcoming single-variable measurements such as HER2 or TILs.

Our study addressed whether information from five immune system–related metagenes (i.e., immune metagenes), and the fraction of 22 TIICs by CIBERSORT, provided additional predictive and prognostic information beyond standard clinical variables and molecular classifiers. We therefore evaluated GEP data from primary tumor biopsies of patients enrolled in the NeoALTTO trial in relation to pCR and survival outcomes. Specifically, we focused on two different HER2-positive breast cancer subgroups: HER2-addicted and non–HER2-addicted on the basis of the 41-gene classifier TRAR expression, which includes genes related to both HER2 and estrogen receptor (ER) signaling, and whose low levels correlated with pCR to neoadjuvant anti-HER2–based therapy (9).

The results of the multicenter randomized phase III NeoALTTO trial (NCT00553358) have been published previously (10). Briefly, a total of 455 patients with HER2-positive breast cancer were randomized to preoperative lapatinib (L; n = 154), trastuzumab (T; n = 149), or their combination (L+T; n = 152) for 6 weeks followed by the addition of paclitaxel for 12 weeks. After surgery, patients continued with three cycles of fluorouracil, epirubicin, and cyclophosphamide followed by the same HER2-targeted therapy administered in the preoperative setting to complete 1 year of treatment. The primary endpoint of NeoALTTO, which was conducted in accordance with the Declaration of Helsinki, was pCR in the breast. The combination of L+T resulted in a significantly higher pCR rate [51.3%; 95% confidence interval (CI), 43.1–59.5] than did T (29.5%; 95% CI, 22.4–37.5) or L alone (24.7%; 95% CI, 18.1–32.3; ref. 10). A secondary endpoint was event-free survival (EFS) defined as the time from randomization to first event with 6-year EFS rates of 67%, 67%, and 74% for L, T, and L + T, respectively (L vs. T: HR = 0.98; 95% CI, 0.64–1.51; P = 0.93; L + T versus T: HR = 0.81; 95% CI, 0.52–1.26; P = 0.35; ref. 11). Written informed consent was obtained from all patients at study entry and covered future biomarker research. The study complied with the Declaration of Helsinki. The Internal Review and Ethics Boards of Fondazione IRCCS Istituto Nazionale dei Tumori approved the gene expression study protocol.

RNA was obtained from snap-frozen primary tumor core biopsies prior to the initiation of neoadjuvant therapy as reported (9). RNA concentration was determined by the ND-1000 spectrophotometer (NanoDrop), RNA quality checked using TapeStation 2200 (Agilent), and the RNA integrity number. GEP data were generated using HumanHT12_v4 beadchips (Illumina). Normalized gene expression data were used to compute five immune metagenes, that is, hemopoietic cell kinase (HCK), lymphocyte-specific kinase (LCK), IFN, MHCII, and STAT1 (7), and the absolute proportion of 22 types of TIICs according to the CIBERSORT (8). The LM22 signature matrix with 1,000 permutations was used to perform the algorithm in the absolute mode. These TIICs included seven types of T cells (resting memory CD4⁺ T cells, activated memory CD4⁺ T cells, CD8⁺ T cells, naive CD4⁺ T cells, γδ T cells, follicular helper T cells, and regulatory T cells), memory B cells, naive B cells, activated mast cells, resting mast cells, activated dendritic cells (DC), resting DCs, macrophages (M0, M1, and M2), eosinophils, monocytes, activated natural killer (NK) cells, resting NK cells, plasma cells, and neutrophils. Of all these 22 TIICs, those with more than 75% of values equal to 0, namely memory B cell, eosinophil, mast cell resting, myeloid DC, NK-cell resting, and T-cell CD4, were excluded from the statistical analysis. In addition, eight TIICs with more than 25% of values equal to 0 were considered as dichotomous variables, that is, TIIC values of 0 (absence) versus >0 (presence). The remaining eight TIICs were considered as continuous variables. The expression of ERBB2 and ESR1 genes, the research-based PAM50, and the TRAR computed as reported previously (9) were considered to better characterize the final evaluable putative biomarkers.

The association between variables was assessed by using the nonparametric Kruskal–Wallis, chi-squared, Fisher exact test, or Spearman correlation coefficient, according to the type of variables. A univariate logistic regression model was implemented for each considered variable to estimate the OR and its 95% CI. Furthermore, a logistic model with the first-order interaction term between treatment arm (single-agent L or T vs. combination), and the considered variables was implemented. The predictive performance of statistically significant variables from univariate analysis was evaluated in a multivariate logistic regression model. The discriminatory capability of the model was evaluated in terms of area under the ROC curve (AUC) and the corresponding 95% CI. Moreover, the prognostic role of each variable was investigated in terms of EFS using a Cox regression model in both uni- and multivariate fashion. A 30-week landmark analysis was performed when pCR was considered. Finally, univariate logistic and Cox regression model were implemented for each variable within TRAR-low and TRAR-high subgroups (9). In the overall study population, logistic and Cox regression models were also implemented for each of the considered variables by adjusting for treatment arm. All statistical analyses were carried out with SAS (Version 9.4.; SAS Institute, Inc.) and R software by adopting a significance level of α = 0.05. A Bonferroni adjusted α level was considered for the TIIC data.

A total of 445 patients were enrolled in the NeoALTTO trial, of which 226 (49.6%) had GEP data prior to treatment. The baseline characteristics of this substudy cohort were similar to those of the entire NeoALTTO trial, as reported previously (9). In univariate analysis, the main drivers of pCR were ER status, treatment arm, TRAR, and PAM50, whereas only pCR and nodal status were significantly associated with EFS (Supplementary Table S1). In multivariate analysis, only treatment arm and TRAR retained their statistical significance with respect to pCR as reported previously (9). Notably, pCR was the unique prognostic factor.

We assessed pair-wise associations between each of the 21 putative biomarkers, that is, 5 metagenes plus 16 out of 22 TIICs (as memory B cell, eosinophil, mast cell resting, myeloid DC, NK-cell resting, and CD4 were not evaluable due to a rate of values equal to 0 >75%), and each of the four available clinical variables, namely age, primary tumor size, nodal status, and ER status.

Five of the 84 pair-wise associations were statistically significant. Specifically, Interferon and activated mast cells were significantly associated with ER status, whereas MHCII, HCK, and M2 macrophages with age, as shown in the top part of Fig. 1. ER-positive tumors displayed higher expression of Interferon metagene and activated mast cells, and young patients (<50 years old) were more likely to have tumors with lower M2 macrophages, HCK, and MHCII metagenes.

The correlation analysis showed that all immune metagenes were associated with each other with Spearman coefficient values >0.60, except for Interferon that showed a Spearman coefficient value >0.60 only with STAT1 (Supplementary Fig. S1). There was also a tight relationship between STAT1 and M1 macrophages (Spearman correlation coefficient of 0.77), as well as between LCK and activated NK, CD8+ T cells, and M1 macrophages (Spearman correlation coefficient values ranging between 0.65 and 0.79). Notably, there was no pattern of relationship between immune metagenes or TIICs and TRAR, ESR1, or ERBB2 expression levels as continuous variables (Supplementary Fig. S1).

Nevertheless, when we assessed the association between the putative biomarkers and the molecular classifiers PAM50 and TRAR, we found that two putative biomarkers were associated with both molecular classifiers, and one biomarker was exquisitely associated with TRAR. Specifically, HER2-enriched cases according to PAM50 showed lower levels of IFN metagene accompanied by a decreased fraction of activated mast cells, whereas TRAR-low cases presented also decreased levels of MHCII, as shown in the lower part of Fig. 1.

In the overall study cohort, the only immune biomarker of the 21 tested significantly associated with pCR was the fraction of γδ T cells (OR = 2.70; 95% CI, 1.01–7.22; P_adj = 0.046; Table 1). Although no immune metagene was predictive of pCR, a statistically significant interaction between LCK and treatment arm (P = 0.01) was observed (Supplementary Fig. S2). Specifically, the expression of LCK increased the likelihood of pCR to L+T treatment (OR = 2.53; 95% CI, 1.12–5.69; P = 0.03), but not to single-agent L or T (OR = 0.74; 95% CI, 0.45–1.22; P = 0.24). No statistically significant interaction was observed between TIICs and treatment arm (Supplementary Fig. S3).

In the TRAR-low (HER2-addicted) subgroup, both STAT1 and HCK were significantly associated with increased pCR (Table 2), and γδ Tcells were associated with increased pCR with borderline significance (OR = 3.57; 95% CI, 0.95–13.42; P_adj = 0.074; Supplementary Table S2). In TRAR-high (non–HER2-addicted) subgroup, no significant results were observed for either immune metagenes (Table 2) or TIICs, as the association between M2 macrophages with decreased pCR disappeared after correction for multiple comparisons (Supplementary Table S2).

Finally, from multivariate analysis γδ T cells and the interaction term between LCK and treatment arm remained significant even after adjusting for age, ER status, tumor size, nodal status, and TRAR (Supplementary Table S3). Notably, this model provided additional predictive capability (AUC = 0.83; 95% CI, 0.78–0.89) over that obtained by clinical variables and TRAR (AUC = 0.80; 95% CI, 0.74–0.86; Fig. 2).

STAT1 was the unique immune metagene related to improved EFS in the overall study cohort (HR = 0.66; 95% CI, 0.49–0.91; P = 0.01; Table 1) even when included in a multivariate model containing pCR, and nodal status (STAT1 HR = 0.68; 95% CI, 0.49–0.94; Supplementary Fig. S4). Remarkably, STAT1 retained its prognostic value also in patients with TRAR-low tumors (HR = 0.50; 95% CI, 0.30–0.86; P = 0.01; Table 2). No statistical significances were observed for TIICs overall (Table 1), or within TRAR groups (Supplementary Table S2).

Invasive breast cancer is considered a heterogeneous disease in terms of biological features, clinical course, and response to treatment, which is recapitulated into different molecular subtypes (12). It is becoming increasingly clear that the interplay between primary breast cancer and TILs depends on this heterogeneity, and that the prognostic and predictive relevance of TILs varies in accordance with the different intrinsic breast cancer subtypes (13). In addition, there is growing evidence that the prognostic and predictive relevance of TILs varies in breast cancer of different intrinsic subtypes (13). Although the prognostic and predictive value of HER2-positive breast cancer molecular features (14) and TILs (15) have been explored separately, the relationships between the composition of the immune cells milieu within the HER2-positive molecular subtype of breast cancer have not been fully elucidated to date.

In this study, we explored the pattern of five immune metagenes, and 22 TIICs in patients with HER2-positive breast cancer subgroups based on bulk GEP analysis from the NeoALTTO trial in relation to pCR, EFS, clinical characteristics, and molecular classifiers (Supplementary Fig. S5). Several key findings with clinical relevance were identified. Henceforth, the relationship of immune phenotype and other primary tumor features overall appears complex. Activated macrophages can be divided into M1, participating in inflammatory reactions and antitumor immunity, and M2 exerting protumorigenic properties. We observed a lower fraction of M2 macrophages in young patients, and HCK metagene displaying low expression in the same cases. These findings are consistent with the well-known function of the hck gene in monocyte and myeloid cell lineages (7). Mast cells stimulate tumor growth, neo-angiogenesis, and metastasis (16). Thus, finding their proportion up in ER-positive cases and down in TRAR-low cases was contrary to expectation. However, recent evidence suggests that mast cells influence the phenotype of breast cancer cells by stimulating a luminal phenotype (17). Preclinical data and in silico analysis of patients with breast cancer revealed a correlation between mast cell density and ESR1 expression (17). Mast cells may also mobilize T cells and DCs for the regulation of adaptive T-cell–mediated antitumor response (18), which is in line with the observed parallel increase of IFN in the same cases. Hence, TIICs may be differently represented depending on age, ER status, and HER2 dependency. The reasons for this are unclear given the number of cases analyzed and stochastic effects. However, data are consistent with the recently reported presence and function of myeloid cells in relation to tumor characteristics (19), and suggest that immune phenotype could be therapeutically exploitable in predefined patient populations.

Notably, immune metagenes were found to be associated with each other and this could suggest a parallel infiltration by both lymphoid and myeloid cells. None were associated with pCR in the overall study cohort. However, a significant increase in the likelihood of pCR was observed in patients with LCK metagene-expressing tumors treated with dual HER2-targeted therapy. Given the role of LCK in T-cell development and homeostasis, we can hypothesize that increased antitumor activity between HER2-targeting combination agents and chemotherapy occurs in tumors with high expression of immune-related markers. Notably, lapatinib is reported to enhance T-cell and IFNγ-based immunity (20), which are required for the cytotoxic activity of trastuzumab, as well as trastuzumab-mediated ADCC (21). Conversely, one cannot overlook that LCK is part of the Src family kinase and regulates cancer cell functions such as proliferation and survival (22). Whether the role played by LCK metagene in modulating response to anti-HER2 therapy is related to its immunomodulating properties or to direct effect on cell signaling is a critical question, as several drugs targeting Src kinases are already available.

Our study showed that the presence of an increased fraction of γδ T cells was an independent predictive factor of pCR in the overall study cohort. Consistently, a relationship between a higher fraction of γδ T cells and increased pCR has been reported in patients with breast cancer (23). This includes HER2-positive cases treated with neoadjuvant therapy and is possibly the result of γδ T-cell–triggered secretion of TNFα and IFNγ, inhibition of angiogenesis, and antigen presentation. Furthermore, like NK cells, γδ T cells exert an antitumor effect via direct ADCC by recognizing antibody-coated HER2-positive breast cancer cells. This mechanism could be predominant in the trastuzumab-containing study arms (17). Since γδ T-cell–triggered ADCC can be augmented by enhanced cancer cell coating, several studies are evaluating the use of bispecific (targeting HER2 and CD3), or even trispecific antibodies (two HER2-specific single-chain fragment variables fused to a fragment antigen bound to the CD16 expressed on γδ T; ref. 24).

Finally, high expression of the STAT1 metagene predicted for better EFS in the overall study cohort and outperformed all standard parameters in multivariate analysis. Notably, STAT1 correlated with IFN metagene and M1 macrophages. Therefore, its protective role could be ascribed to the cascade starting from INFγ-induced STAT1 through activation of CXCL9, CXCL10, and CXCL11 and recruitment of pro-inflammatory effectors (25, 26). Nevertheless, we cannot exclude other mechanisms such as cell-cycle arrest in response to INFγ, induction of cell death via the Fas and its ligand, or angiogenesis inhibition (26).

We are aware of several limitations to this study. First, its retrospective nature and the lack of a validation series do not allow us to draw any definitive conclusion regarding the effectiveness of the immune markers evaluated. In addition, it should be cautioned that there is some evidence of potential overlap of gene signatures using CIBERSORT between γδ T and other immune cells such as CD4 and CD8 lymphocytes and natural killer cells (27). Moreover, the small sample size, and the number of events precluded any meaningful analysis of immune markers according to TRAR status within each treatment arm. Nevertheless, our study has some strengths that include the precise characterization of the NeoALTTO trial cohort and our analytical approach. We utilized, for the first-time, immuno-genomics to study both immune-related genes and immune cell composition, integrating these with clinical and molecular data.

Altogether, our results show that there are narrow changes in immune metagene expression and TIIC composition according to age, hormone-receptor, and HER2-status. Increased expression of STAT1 and HCK paralleled by increased fraction of γδ T cells define the TRAR-low tumors with the highest chance of treatment response. No single immune marker is informative in TRAR-high tumors. The LCK metagene is associated with pCR in a treatment-dependent manner, because it is directly predictive of response to dual blockade, yet inversely predictive of response to single agent. γδ T cells and LCK add discriminatory capability to clinical and molecular variables. Patients with increased STAT1 levels have a favorable prognosis.

The clinical value of these data lies in the fact that anti-HER2 agents alone or in combination have different antitumor activity depending on primary tumor immune phenotype as inferred by genomic data; this immune phenotype varies according to patient and primary tumor intrinsic characteristics, and offers both predictive and prognostic additional information over known clinical and molecular variables. Furthermore, as new therapeutic approaches are sought aiming to induce immune response, our study urges to characterize the immune profile of patients with HER2-positive disease to identify those who might benefit from treatment with immunotherapy (28). In conclusion, although confirmatory studies on prospective independent case series are needed, incorporation of immune profile into a larger patient with HER2-positive breast cancer evaluation has the potential to optimally characterize primary tumor and host-tumor interplay, to overcome the limitations of predictive capability of molecular classifiers, and to finally assist in new therapeutic approaches including both anti-HER2 and immunomodulant agents.

E. de Azambuja reports grants from GSK/Novartis during the conduct of the study; E. de Azambuja also reports other support from Roche, Pierre Fabre, libbs, Lilly, Seattle Genetics, and Zodiac, as well as non-financial support from Roche outside the submitted work. D. Fumagalli reports grants from Novartis during the conduct of the study, as well as grants from Roche/Genentech, AstraZeneca, Novartis, Pfizer, Servier, Tesaro, and Sanofi outside the submitted work. N. Harbeck reports personal fees from AstraZeneca, Daiichi Sankyo, Lilly, Novartis, Pierre Fabre, Pfizer, Roche, Sandoz, and Seagen outside the submitted work. M. Izquierdo reports employment and ownership of stock with Novartis. P. Nuciforo reports grants from Novartis during the conduct of the study, as well as personal fees from Novartis, Genentech, MSD Oncology, Bayer, and Targos outside the submitted work. J. Huober reports personal fees from Roche, AbbVie, Seagen, Gilead, AstraZeneca, MSD, and Eisai. J. Huober also reports personal fees and other support from Pfizer; grants and personal fees from Lilly, Novartis, and Hexal; and other support from Daiichi Sankyo outside the submitted work. S. Cinieri reports employment with Associazione Italiana Oncologia Medica (AIOM). M. Piccart reports grants and personal fees from AstraZeneca, Immunomedics, Lilly, Menarini, MSD, Novartis, Pfizer, and Roche-Genentech; M. Piccart also reports personal fees from Camel-IDS, Debiopharm, Odonate, Radius, Servier, and Synthon, as well as grants from Seattle Genetics, Immutep, Seagen, and NBE Therapeutics outside the submitted work. S. Di Cosimo reports personal fees from Pierre-Fabre and Novartis Pharma, as well as grants from Fondazione AIRC and Associazione Italiana di Oncologia Medica (AIOM) during the conduct of the study. No disclosures were reported by the other authors.

S. Pizzamiglio: Conceptualization, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. C.M. Ciniselli: Conceptualization, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. T. Triulzi: Formal analysis, writing–review and editing. C. Gargiuli: Formal analysis, methodology, writing–review and editing. L. De Cecco: Formal analysis, investigation, methodology. E. de Azambuja: Investigation, writing–review and editing. D. Fumagalli: Investigation, writing–review and editing. C. Sotiriou: Investigation, writing–review and editing. N. Harbeck: Investigation, writing–review and editing. M. Izquierdo: Investigation, writing–review and editing. P. Nuciforo: Investigation, writing–review and editing. J. Huober: Investigation, writing–review and editing. V. Cappelletti: Investigation, writing–review and editing. S. Cinieri: Funding acquisition, writing–review and editing. M. Piccart: Funding acquisition, investigation, writing–original draft, writing-review and editing. M.G. Daidone: Investigation, writing–review and editing. G. Pruneri: Investigation, writing–review and editing. M.P. Colombo: Investigation, writing–original draft. E. Tagliabue: Investigation, writing–original draft. P. Verderio: Conceptualization, formal analysis, investigation, methodology, writing–original draft, project administration. S. Di Cosimo: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft.

The NeoALTTO study was sponsored by GlaxoSmithKline. Lapatinib is an asset of Novartis AG as of March 2, 2015. This substudy was sponsored by Fondazione Associazione Italiana Ricerca contro il Cancro (AIRC IG 20774 to S. Di Cosimo), and by Associazione Italiana di Oncologia Medica (AIOM).

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.