HER2-Low Breast Cancer: Molecular Characteristics and Prognosis

Simple Summary The dichotomous classification of HER2 status is experiencing a paradigm change, leading to the identification of the so-called “HER2-low” category. The aim of our retrospective, observational study was to characterize intrinsic PAM50 subtypes within HER2-low primary breast tumors extracted from The Cancer Genome Atlas (TCGA) dataset and to describe the prognostic impact of HER2-low on survival outcomes. We evaluated 804 breast cancers and we identified 410 HER2-low tumors (336 with positive hormonal receptor status (HR+) and 74 with negative HR status (HR−)). HER2-enriched tumors were more frequent in HER2-low/HR− and HER2-low/HR+ subtypes, compared to HER2-negative/HR− and HER2-negative/HR+ subtypes, respectively (13.7% versus 1.6% and 1.2% versus 0.5%, respectively). We observed no significant differences in prognosis between HER2-low subtypes and each non-HER2-low subtype when paired by HR status. Our characterization of PAM50 intrinsic subtypes within HER2-low breast cancer ultimately supports further investigation of new treatment strategies in the HER2-low category, with new promising drugs being tested in the context. Abstract Background: We aimed to determine the distribution of intrinsic subtypes within HER2-low breast cancer (BC), and to describe the prognostic impact of HER2-low status on survival outcomes. Methods: This is a retrospective, observational study of primary BC extracted from The Cancer Genome Atlas dataset. We described the distribution of PAM50 intrinsic subtypes within HER2-low BC subtype according to hormonal receptor status (positive (HR+) and negative (HR−)). Secondly, we assessed the impact of HER2-low on survival outcomes (progression-free interval (PFI), disease-free interval (DFI), and overall survival (OS)). Results: We analyzed 804 primary BCs, including 410 (51%) HER2-low BCs (336 HR+ and 74 HR−). The proportion of HER2-enriched tumors was higher in the HER2-low/HR− group compared to HER2-low/HR+ (13.7% versus 1.2%, respectively). HER2-enriched tumors were more frequent in HER2-low/HR− and HER2-low/HR+ subtypes, compared to HER2-negative/HR− and HER2-negative/HR+ subtypes, respectively (13.7% versus 1.6% and 1.2% versus 0.5%, respectively). We observed no significant differences in PFI, DFI, and OS between HER2-low subtypes and each non-HER2-low subtype paired by HR status. Conclusions: Our characterization of PAM50 intrinsic subtypes within HER2-low breast cancer may explain the different clinical behaviors and responses to treatment, and ultimately support further investigation of new treatment strategies in the HER2-low category. Moreover, it highlights the importance of considering HR status in the HER2-low category.


Introduction
The characterization of the Human Epidermal Growth Factor Receptor-2 (HER2) in breast cancer (BC) has dramatically evolved over the last three decades, from a biomarker of poor prognosis to one of clinical benefit for trastuzumab and other anti-HER2 agents, when overexpressed/amplified in tumor cells [1][2][3][4][5]. In clinical practice, international guidelines recommend a binary distinction between HER2-positive (HER2+) and HER2-negative (HER2−) BCs, to guide physicians' treatment decisions [6]. However, a relevant proportion (≈45-55%) of tumors that are classified as HER2− are in fact what is now called HER2low [7][8][9]. The HER2-low BC subtype represents a new recently proposed nomenclature for those tumors with an immunohistochemistry (IHC) assay score of 1+ or 2+ but with negative in situ hybridization (ISH) assay [7,10]. In clinical practice, these tumors are classified as triple-negative BC (TNBC) or, if hormone receptors (HRs) are expressed, as luminal BC, thus being considered to not benefit from anti-HER2 treatments. HER2-low BCs comprise a majority of HR-positive (HR+) tumors (≈83%) [10,11]; however, little is known about their molecular characterization. Recently, Schettini et al. have presented the distribution of PAM50 intrinsic subtypes of a large retrospective cohort (n = 1576) of HER2-negative BCs extracted from databases pertaining to different studies [12]. Although this was the first molecular description of HER2-low BC in a large BC cohort, the authors pointed out several limitations of their work, including the combination of data from several different datasets, as well as the heterogeneity of the study population, which included a large proportion of metastatic patients. Moreover, in almost 30% of cases, the analyzed biospecimen was not the primary tumor. In order to provide complementary information and to shed further light on the molecular characteristics of HER2-low BC, we performed a retrospective analysis of primary BCs extracted from The Cancer Genome Atlas (TCGA) dataset. Moreover, we aimed to provide a comprehensive evaluation of HER2-low BCs' prognosis by assessing survival outcomes of the above-mentioned cohort.
Indeed, despite the paucity of data available in HER2-low breast disease, being a rather new and quite unexplored field, there is a strong rationale that, also in HER2-low tumors, different intrinsic subtypes may explain different clinical behavior and response to treatments, as this has been already observed for other breast cancer subtypes (i.e., HER2-positive, HR-positive, HER2-negative and triple-negative breast cancers) [13][14][15].

Study Population and Data Extraction
We analyzed a retrospective cohort of primary BCs extracted from the publicly available breast cancer dataset generated by the TCGA Research Network (http://cancergenome. nih.gov/ (accessed on 15 October 2020)). Gene expression data (FPKM-UQ), PAM50 subtypes of primary tumor samples, and clinicopathological data were downloaded from the GDC Data Portal using TCGAbiolinks [16].
Only tumors with available HR and HER2 status by IHC were included in the present study.
HER2 status was derived by IHC score/status and FISH status available in TCGA. In particular, information on HER2 status in TCGA dataset was evaluated according to three different forms provided: "HER2 IHC score", "HER2 IHC status", and "HER2 FISH status".
Tumors with a HER2 IHC assay score of 2+ were included only provided that HER2 status by FISH was available and not defined as equivocal. Tumors with an IHC assay score of 0, 1+ or 3+ could be included in the analysis even in absence of HER2 status by FISH. Tumors with HER2 IHC score "not available" could be included in the analysis, provided that HER2 IHC status was available. HER2-low status was defined as an IHC assay score of 1+ or 2+ without amplification at FISH assay, as defined in the clinical TCGA database.
Estrogen receptor (ER) and progesterone receptor (PgR) status were characterized according to the IHC data available. HR status was considered positive (HR+) when ER and/or PgR status was positive.
We included in the HER2−/HR− and HER2−/HR+ groups all tumors with HER2 IHC score of 0 as well as tumors with not available IHC score but defined as "negative" by IHC status and with FISH negative/not evaluated.
The "non-HER2-low" group was constituted by all breast tumors not matching the "HER2-low" definition, namely HER2-positive tumors as per ASCO-CAP definition (i.e., HER2 score 3+ at IHC and/or score 2+ with FISH amplification), as well as HER2-negative tumors, as defined above [17].
Intrinsic subtypes were assessed according to the PAM50 Breast Cancer Gene Signature Assay, which classifies BC subtypes in Luminal A, Luminal B, HER2-enriched, Basal-like, and Normal-like [18].
Curated survival data including overall survival (OS), progression-free interval (PFI), and disease-free interval (DFI) were obtained from the TCGA Pan-Cancer Clinical Data Resource (TCGA-CDR), developed by Liu et al. [19].

Study Objectives
Our primary objective was to characterize the molecular profile of HER2-low tumors included in our cohort of primary BCs extracted from TCGA dataset. Therefore, we characterized the distribution of PAM50 intrinsic subtypes within the HER2-low BC cohort, and we assessed the level of mRNA expression of ERBB2 and ESR1 genes in the different BC subtypes, including HER2-low tumors.
Our secondary objective was to describe the prognostic impact of HER2-low on patients' outcomes. Therefore, we evaluated OS, PFI and DFI of HER2-low tumors compared to non-HER2-low tumors, and of HER2-low/HR+ tumors compared to HER2-low/HR− ones. OS was defined as the time from diagnosis until the date of death from any cause. PFI was defined as the time from diagnosis until the date of first occurrence of a new tumor event, which included disease progression, locoregional recurrence, distant metastasis, new primary tumor, or death with tumor. DFI was defined as the time from diagnosis until the date of the first new tumor event subsequent to the determination of a patient's disease-free status after their initial diagnosis and treatment. Such event could be either a locoregional recurrence or a distant metastasis, the development of a new primary tumor in the same organ, or death from advancing of the same tumor. For PFI, unlike DFI, it was not necessary to first know whether a patient ever achieved disease-free status following their initial diagnosis and treatment.

Statistical Analysis
Statistical hypothesis tests involving the comparison of two or more groups were carried out using the Mann-Whitney U-test. False discovery rates (FDRs) were obtained using the Benjamini and Hochberg method. Survival data are shown using the Kaplan-Meier estimator, while the log-rank test was used to calculate the associated p-value. Fisher's exact tests were used to evaluate associations between categorical data. Spearman correlation was used to determine the correlation between ESR1 and ERBB2 gene expression levels. All analyses were performed in R (version 4.0.2).
We further explored the correlation between ERBB2 and ESR1 gene expression levels according to HER2 status (i.e., negative, low, positive) ( Figure S3). In an analysis including both HR+ and HR− tumors, a negative correlation was found in HER2+ tumors (Spearman = −0.46, p < 0.001), while both the HER2-low and HER2-groups presented a positive correlation between ESR1 and ERBB2 expression levels (Spearman 0.34, p < 0.001 and Spearman 0.38, p < 0.001 for HER2-low and HER2-tumors, respectively). The correlations were still significant although less pronounced when limiting the analysis to HR+ tumors (Spearman correlations of −0.29, 0.18, and 0.14 in HER2+, HER2-low and HER2-tumors, respectively), while no significant correlation (p > 0.05) was noted in HR− tumors.
We further explored the correlation between ERBB2 and ESR1 gene expression levels according to HER2 status (i.e., negative, low, positive) ( Figure S3). In an analysis including both HR+ and HR− tumors, a negative correlation was found in HER2+ tumors (Spearman = −0.46, p < 0.001), while both the HER2-low and HER2-groups presented a positive correlation between ESR1 and ERBB2 expression levels (Spearman 0.34, p < 0.001 and Spearman 0.38, p < 0.001 for HER2-low and HER2-tumors, respectively). The correlations were still significant although less pronounced when limiting the analysis to HR+ tumors (Spearman correlations of −0.29, 0.18, and 0.14 in HER2+, HER2-low and HER2-tumors, respectively), while no significant correlation (p > 0.05) was noted in HR− tumors.

Discussion
The dichotomous classification of HER2 status is experiencing a paradigm shift leading to the identification of the so-called "HER2-low" category, for which new treatment strategies are currently under evaluation [20][21][22][23][24][25]. The rising role of molecular profiling further helps in dissecting the heterogeneity within each BC subtype [13,[26][27][28], including HER2-low BC [12]. In this work, we presented a retrospective analysis of molecular characteristics of primary HER2-low BC and a comprehensive analysis of its prognosis. In our cohort, HER2-low BC represented 51% of all tumors, with the majority (82%) being HR+; these findings are in line with previous reports [7,8,11]. When focusing on the distribution of PAM50 intrinsic subtypes in HER2-low tumors, we noticed that the distribution of intrinsic subtypes within HER2-low/HR+ BC was more similar to the one of HER2−/HR+ BC, rather than to HER2-low/HR− BC, and this is reassuring for clinical practice, since, thus far, HER2-low/HR+ BC are ascribed to HER2−/HR+ BC for the purpose of treatment. On the other hand, HER2-low/HR− BC showed a higher proportion of HER2-enriched intrinsic subtype, compared to HER2−/HR−. Hitherto, HER2-low/HR− BC is considered as HER2−/HR− in clinical practice, but this difference might reflect a dissimilar clinical behavior, and ultimately further supports the clinical investigation of new HER2-targeting agents in HER2-low BC. Antibody-drug conjugates (ADCs), such as trastuzumab deruxtecan and trastuzumab duocarmazine, have shown promising efficacy data in patients with metastatic HER2-low BC, attributed to the retention of all trastuzumab antitumor properties, associated with a bystander killing effect, which allows targeting and killing cancer cells even in tumors with lower degrees and heterogeneity of HER2 expression, a once limiting step for the clinical activity of anti-HER2 agents [20,21].

Discussion
The dichotomous classification of HER2 status is experiencing a paradigm shift leading to the identification of the so-called "HER2-low" category, for which new treatment strategies are currently under evaluation [20][21][22][23][24][25]. The rising role of molecular profiling further helps in dissecting the heterogeneity within each BC subtype [13,[26][27][28], including HER2-low BC [12]. In this work, we presented a retrospective analysis of molecular characteristics of primary HER2-low BC and a comprehensive analysis of its prognosis. In our cohort, HER2-low BC represented 51% of all tumors, with the majority (82%) being HR+; these findings are in line with previous reports [7,8,11]. When focusing on the distribution of PAM50 intrinsic subtypes in HER2-low tumors, we noticed that the distribution of intrinsic subtypes within HER2-low/HR+ BC was more similar to the one of HER2−/HR+ BC, rather than to HER2-low/HR− BC, and this is reassuring for clinical practice, since, thus far, HER2-low/HR+ BC are ascribed to HER2−/HR+ BC for the purpose of treatment. On the other hand, HER2-low/HR− BC showed a higher proportion of HER2-enriched intrinsic subtype, compared to HER2−/HR−. Hitherto, HER2-low/HR− BC is considered as HER2−/HR− in clinical practice, but this difference might reflect a dissimilar clinical behavior, and ultimately further supports the clinical investigation of new HER2-targeting agents in HER2-low BC. Antibody-drug conjugates (ADCs), such as trastuzumab deruxtecan and trastuzumab duocarmazine, have shown promising efficacy data in patients with metastatic HER2-low BC, attributed to the retention of all trastuzumab anti-tumor properties, associated with a bystander killing effect, which allows targeting and killing cancer cells even in tumors with lower degrees and heterogeneity of HER2 expression, a once limiting step for the clinical activity of anti-HER2 agents [20,21].
Furthermore, we presented the distribution of BC subtypes (HER2−/HR+, HER2+/HR+, HER2+/HR−, HER2-low/HR+, HER2-low/HR−, HER2−/HR−) according to PAM50 intrinsic subtypes. Interestingly, more than 50% of both Luminal A and B intrinsic subtypes were HER2-low/HR+ subtype, suggesting that in HER2-low/HR+ tumors, HR and "luminal" genes expression, and not HER2, might be the principal oncological driver. On the contrary, the HER2-low/HR− subtype was mainly represented in Basal-like intrinsic subtype (40.3%). The discrepancies existing between different classifications of breast tumors, according to pathological and molecular characteristics, may underline the need of a major integration of molecular information when selecting patients to maximize treatment benefit. Recent data on a biomarker analysis of intrinsic subtypes and efficacy across the MONALEESA trials showed that the HER2-enriched intrinsic subtype had the greatest reduction in the risk of progression and death after treatment with the CDK4/6 inhibitor ribociclib, compared to other intrinsic subtypes, while the Basal-like was the only subtype showing no benefit [15]. Taken together, these data support the role of molecular profiling as a possible strategy to dissect biological differences beyond the simple IHC classification of BC subtypes, and to better select patients who may benefit most from specific treatments.
We then evaluated the level of mRNA expression of ERBB2 and ESR1 according to BC subtypes. ERBB2 expression was higher in HER2+ subtypes, followed by HER2-low subtypes, which showed significantly higher levels compared to HER2-negative subtypes. Moreover, ERBB2 expression was higher in HER2-low/HR+ compared to HER2-low/HR− tumors. This latter observation is in apparent contrast with the fact that we observed no significant or pronounced enrichment for the HER2-enriched subtype in HER2-low/HR+ as compared to HER2-low/HR− by PAM50. However, a possible explanation resides in the fact that PAM50 subtyping is not only defined by ERBB2 and ESR1 expression levels. Our findings are consistent with those presented by Schettini et al. [12], with the authors suggesting that the lack of enrichment of the HER2-enriched subtype within the HER2-low disease may be explained based on previous evidence showing that the HER2enriched phenotype is not defined by the sole expression of ERBB2. Indeed, it rather takes into account proliferation and ESR1 expression levels as well [14]. Furthermore, despite the different expression profiling techniques used in the analysis of Schettini et al. and ours (Nanostring nCounter versus RNA-sequencing data, respectively), obtaining highly similar results further corroborates our findings. In addition, the two variables (i.e., HER2-enriched subtype and ERBB2 levels) provide independent predictive and prognostic information [28,29].
In our analysis, ESR1 expression was higher in HR+ subtypes (namely, HER2−/HR+, HER2+/HR+, and HER2-low/HR+), as expected, with no significant differences between HER2-low/HR+ and HER2−/HR+. The HR+ tumors were clearly dominated by a "luminal" phenotype, and the high expression of ESR1 and other "luminal" genes may justify the lower proportion of HER2-enriched tumors observed in the HER2-low/HR+, despite the (marginally) higher ERBB2 expression levels compared to the HER2-low/HR−. Moreover, we observed an inverse correlation between ERBB2 and ESR1 gene expression levels in HER2-positive and -negative tumors (both when HR+ and HR− tumors were analyzed together and when limiting the analysis the the HR+ tumors), with results in line with previous reports [30]. In particular, HER2-low tumors behaved similarly to the HER2-negative ones, with a positive correlation between ESR1 and ERBB2 in the whole cohort including HR+ and HR− tumors, as well as in the HR+ subgroups, although less pronounced in the latter. Overall, these findings underline the need to separate the expression of single genes from the underlying tumor phenotype, and suggest that biological differences exist, beyond the simple distinction between HR+ and HR− tumors. Moreover, they can contribute to explain the different response rates observed for HER2-low/HR+ and HER2-low/HR− in clinical trials evaluating new ADCs in HER2-low metastatic BC [20,21]. For instance, in the phase I dose-escalation and dose-expansion study of trastuzumab duocarmazine, 9/32 (28%) patients with HER2-low/HR+ breast cancer and 6/15 (40%) patients with HER2-low/HR− breast cancer exhibited an objective response. Although not sufficient to postulate that novel anti-HER2 ADCs could be more appropriate in HR− versus HR+ HER2-low tumors, these data suggest the important role of HR status also within the HER2-low group. Taken together, our findings overall strengthen the ones previously published [12] and might be particularly useful for future therapeutic developments in patients with breast cancer.
In addition, in order to evaluate the impact of HER2-low status on prognosis, we presented survival analyses comparing HER2-low to the other BC subtypes. We did not observe any significant difference in terms of DFI, PFI, and OS when comparing HER2-low tumors with non-HER-low BCs, and this is reassuring for clinical practice, considering that patients with HER2-low BC likely did not receive any HER2-targeting therapy. However, these prognostic data are in apparent contrast with other retrospective data in the literature [31,32] and should be read with caution and considered only as exploratory, due to the retrospective design of our analysis and to the lack of adequate information about treatment received. Interestingly, HER2-low/HR+ tumors presented better DFI and PFI compared to HER2-low/HR− BC, and this further underlines the heterogeneity existing within the HER2-low group.
For the purpose of survival analyses, we used data from the integrated the TCGA Pan-Cancer Clinical Data Resource, a standardized dataset developed by Liu et al. [19], aiming to ensure proper use of clinical data in association with genomic features reported in TCGA dataset. Indeed, the relatively short follow-up and the incomplete annotation of patient outcomes as well as of treatment data for each sample have been reported as limitations of the TCGA dataset [33,34]. According to the recommendations provided by Liu et al. about the use of clinical outcome's endpoints in each type of disease reported in TCGA dataset, both PFI and DFI are recommended for use when analyzing breast cancer data. On the contrary, OS should be used with caution, due to the relatively short follow-up (median follow-up: 27.7 months). Nonetheless, OS represents an important endpoint, being free from any ambiguity in the definition of an OS event. Therefore, in the present study, we included OS analysis, which should be interpreted with caution, due to the above-mentioned limitation.
Our study has some other limitations: (i) the retrospective design, with associated potential biases; (ii) a high number of cases presented incomplete information regarding HER2, ER, and PgR status. This lack of data affected the definition of the luminal and TNBC groups we used in the present study, although we performed subsequent analysis to ensure the reliability of data included in our analysis, and further highlights the importance of annotating clinical data with standardized procedures in genomics studies; (iii) different criteria for ER, PgR, and HER2 evaluation could be used, with no central confirmation of pathological assessment performed. For instance, most of the tumors ER and PgR status considered to be negative had IHC expression scores <10%, although the most recent guidelines recommend considering tumor samples with 1% or more nuclei positive as ERpositive [35]. In TCGA HER2 status was not defined according to the current ASCO/CAP guidelines [6], and since the definition of the HER2 status had evolved in recent years, it will further have a role in the precise identification and characterization of HER2-low tumors; (iv) for the purpose of this paper, we limited the analyses to the evaluation of PAM50 subtypes, as well as ERBB2 and ESR1 expression levels, while a broader analysis of gene expression profiles could provide additional information on HER2-low tumors; (v) the incomplete data available in the TCGA dataset about treatments received, which means the prognostic data should be interpreted with caution.
On the contrary, our study has some strengths. First, we used a homogeneous cohort of primary BCs extracted from a single dataset: focusing on primary BC samples allowed us to limit the variability existing between primary tumor and metastases [36][37][38]. Second, we provided a comprehensive analysis of the impact of HER2-low status on three different clinical outcomes, based on the recommendations provided in the TCGA CDR, a standardized dataset with a transparent derivation of clinical outcome endpoints and with resolution of quality concerns.
Our exploratory data can be hypothesis-generating for further studies, as it will be interesting to verify how PAM50 subtypes or ERBB2 expression levels could impact treatment response (for instance in relation to the role of the HER2-targeted part of the ADCs in HER2-low tumors with higher ERBB2 expression, with a HER2-enriched PAM50 subtype or according to HR status) in these breast cancer subtypes [39].

Conclusions
This characterization of the PAM50 intrinsic subtypes within HER2-low primary BC underlines the heterogeneity of this subtype reflected on clinical outcome. HER2-low/HR+ and HER2-low/HR− BCs showed a higher proportion of HER2-enriched intrinsic subtype as well as higher ERBB2 expression levels compared to HER2−/HR+ and HER2−/HR− subtypes, respectively. This finding is consistent with previous data, and further supports the investigation of new treatment strategies in the HER2-low category. Finally, the heterogeneity in terms of PAM50 subtypes and prognosis highlights the importance of considering HR status in the HER2-low category.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/cancers13112824/s1, Figure S1: STROBE flow-chart resuming the patient selection process. Figure S2: ERBB2 gene expression levels according to HER2 IHC score in HER2-low/HR+ versus HER2−/HR+ tumors (left plot) and in HER2-low/HR− versus HER2−/HR− tumors (right plot). Figure S3: Correlation between ERBB2 and ESR1 gene expression levels according to HER2 status (i.e., positive, low and negative) in the whole population, in the HR+ and HR− subgroups. Table S1: Primary breast tumors included in the present study (n = 804). Table S2: Treatments received by all subjects and according to breast cancer subtype, as available in the TCGA dataset [13]. Data Availability Statement: The breast cancer TCGA dataset analyzed in this study is publicly available in the GDC Data Portal (https://portal.gdc.cancer.gov/ (accessed on 15 October 2020)). grants: Roche/GNE, GSK/Novartis. Research grant for his institute: Roche/GNE, AstraZeneca, Novartis, and Servier (outside the submitted work).