HER2 and EGFR amplification and expression in urothelial carcinoma occurs in distinct biological and molecular contexts

We analyzed a cohort of 599 cases of urothelial carcinoma for EGFR, ERBB2, and ERBB3 gene expression and genomic alterations. The cohort consisted of a reference set (n = 292) comprising all stages and grades and one set (n = 307) of advanced tumors. All cases were previously classified into urothelial carcinoma molecular subtypes. Genomic amplifications were established by array-CGH or in-situ hybridization, and gene expression both at mRNA and protein levels. Clinical HER2 status was independently evaluated using standard clinical procedures. EGFR amplifications were observed in 14% and ERBB2 amplifications in 23% of the reference cohort. EGFR gains were enriched in the Basal/SCC-like and ERBB2 gains in the Genomically Unstable subtypes. The expression data suggests that the Genomically Unstable show high ERBB2/ERBB3 but low EGFR expression and that Basal/SCC-like tumors show high EGFR but low ERBB2/ERBB3 expression. Whereas the frequency of ERBB2 genomic amplification were similar for cases of the Genomically Unstable subtype in the two cohorts, the Urothelial-like subtype acquires ERBB2 amplifications and expression during progression. Even though a good correlation between gene amplification and ERBB2 gene expression was observed in the Urothelial-like and Genomically Unstable subtypes less than half of the Basal/SCC-like cases with ERBB2 amplification showed concomitant ERBB2 mRNA and protein expression. We conclude that clinical trials using ERBB2 (HER2) or EGFR as targets have not fully appreciated the molecular heterogeneity in which activated ERBB2 and EGFR systems operate. Proper tumor classification is likely to be critical for arriving at thorough conclusions regarding new HER2 and EGFR based treatment regimes.


INTRODUCTION
Bladder cancer is the fifth most common malignancy in the Western world. It is associated with a high rate of mortality in patients with advanced disease. In spite of this, few significant improvements in survival has been achieved during the last three decades [1]. While a proportion of patients respond to conventional chemotherapy these responses are rarely long-lasting. Even though molecular targeted therapy has been successful for other cancers types, few advances have been made in bladder cancer. In analogy with other tumor types, members of the ERBB receptor family have been considered as potential targets also for urothelial carcinomas e.g., ERBB2 (HER2) [2] and EGFR [3]. In urothelial carcinomas, overexpression of ERBB2 protein has been reported to vary considerably between studies [4] as well as between geographically distinct cohorts [5]. Several studies have also reported a low concordance between ERBB2 protein level and gene amplification in urothelial carcinomas [5,6]. Hansel et al. detected ERBB2 overexpression (IHC) in 36% but genomic amplification in only 10% of tumors [7]. Similarly low levels of ERBB2 genomic amplifications have also been reported by other investigators (e.g., [8]). Hence no consensus regarding the status of HER2 alterations in urothelial cancer has Research Paper been arrived at. Furthermore, even though several trials aiming for HER2 as a target in bladder cancer have been initiated, no general conclusions have been reached. A factor contributing to the disparate results is most likely the underlying heterogeneity of bladder cancer and the lack of adequate molecular descriptions of bladder cancer molecular subtypes; the context in which the ERBB2 and EGFR targets operate. We have described three major molecular subtypes of bladder cancer, Urothelial-like (Uro) (previously termed Urobasal [9]), Genomically Unstable (GU) and Basal/SCC-like [9], that accommodates fundamental differences in their molecular biology [10][11][12][13][14]. By combining global mRNA gene expression and extensive immunohistochemistry (IHC) analyses we have shown that the Uro, GU, and SCC-like subtypes dominate also the muscle invasive tumors, albeit in more progressed and infiltrated versions [15]. Here we investigate EGFR, ERBB2 (HER2), and ERBB3 (HER3) genomic alterations and expression in relation urothelial carcinoma molecular subtypes. To be able to study changes during tumor progression we analyze two cohorts, one (n = 292) dominated by non-muscle invasive tumors and a second (n = 307) dominated by muscle invasive tumors. We show that ERBB2 amplifications and expression, as well as clinically "HER2 positive" cases, may be of two fundamentally different molecular subtypes, of Uro or the GU subtypes, and that EGFR expression is associated with the SCC-like subtype.
Co-occurring focal amplifications of EGFR, and ERBB2 or ERBB3 were not observed, while co-occurring gains of EGFR and ERBB2, and EGFR and ERBB3 were seen in eight and one cases, respectively. Five cases showed gains of all three genes. Examining the very high expressing cases (with Tumor Cell Score ≥ 2.5, indicating high protein expression throughout the tumor), EGFR+ERBB2 expression was only observed in one GU tumor, while two UroA tumors expressed both EGFR and ERBB3 at high levels. None of the tumors had strong simultaneous expression of all three proteins. Very high simultaneous expression of both ERBB2 and ERBB3 was seen in 11 tumors, where 6 were GU and 5 were UroA.

EGFR, ERBB2 and ERBB3 in advanced urothelial carcinoma cohort
The 307 samples in the advanced cohort were all from patients that underwent radical cystectomy. The tumors in this cohort have been classified into molecular subtypes through the combination of global mRNA expression and extensive immunohistochemistry [15]. The combined analyses produced robust subtype assignments into Urothelial-like (Uro), Urothelial-like B (UroB), Genomically Unstable (GU), Basal/SCClike, Mesenchymal Infiltrated (Mes-Inf), and Small Cell/ Neuroendocrine (Sc/NE) ( Figure 4). In addition, a small group (n = 6) of highly infiltrated, and thus unclassified, tumors were detected (Inf in Figure 4). The expression patterns defining the Basal/SCC-like subtype is illustrated in Figure 4A, with high KRT5 and KRT14, and low GATA3 and FOXA1 expression [9]. The majority of the non-Basal/SCC-like tumors belong to the Urotheliallike or Genomically Unstable category of tumors, which can be distinguished from each other by the expression patterns of FGFR3, CCND1, CDKN2A (p16), and RB1 ( Figure 4B). The combination of mRNA and IHC data also enabled us to resolve the remaining tumor cell phenotypes where examination of mRNA expression patterns alone is complicated by infiltrating immune cells or high stromal cell content [15]. The final subtype distribution was 109 Uro (35.5%), 24 UroB (7.8%), 66 GU (21.5%), 62 Basal/ SCC-like (20.2%), 16 Mes-Inf (5.2%), and 24 Small Cell/Neuroendocrine (Sc/NE) (7.8%) and 6 remaining Infiltrated (Inf) (2%).
Increased EGFR mRNA and protein levels were largely confined to the Basal/SCC-like subtype (p = 10 -12 and p = 10 -13 , respectively) ( Figure 4C). ERBB2 mRNA expression was significantly higher within the Uro and GU subtypes (p = 10 -9 and p = 10 -13 , respectively), an enrichment also seen at the protein level (p = 10 -7 and p = 10 -10 ). The ERBB3 mRNA expression was highest in the Uro and GU tumors (p = 10 -16 and p = 10 -9 ), and was strongly correlated with the mRNA levels of ERBB2 (r = 0.69, p = 10 -16 ). The Uro and GU tumors also had the strongest ERBB3 protein expression. The EGFR, ERBB2, and ERBB3 protein expression showed a strong correlation with the mRNA levels (r = 0.65, 0.75, and 0.69, respectively, all p < 10 -16 ) ( Figure 4C). The ratio of ERBB2+ERRB3 to EGFR mRNA expression summarizes the overall expression patterns and identifies Uro and GU tumors as ERBB2, ERBB3 high and EGFR low, and the Basal/SCC-like tumors as EGFR high and ERBB2, ERBB3 low ( Figure 4C). No tumors had very high simultaneous EGFR and ERBB2 protein expression (IHC Tumor Cell Score ≥ 2.5), while one Uro tumor had high EGFR and ERBB3 protein expression. Similarly, no tumor G2, blue; G3, red. White indicates missing information. Panel 2: Gene expression of EGFR, ERBB2, and ERBB3 for 292 samples ordered by subtype. Red, high expression; green, low expression. Panel 3: The ratio of EGFR/ ERBB2 and ERBB3 mRNA expression emphasizes the subtype specific pattern of expression. Red, high ERBB2 and ERBB3 expression, and low EGFR expression; blue, low ERBB2 and ERBB3 expression, and high EGFR expression. Panel 4: EGFR, ERBB2, and ERBB3 immunohistochemistry scores. The score represents a tumor cell score (TCS), where the staining intensity is multiplied by the fraction of the tumor cells that show staining. Dark brown, high expression; white, no expression; gray, missing data. (B) Copy number alterations as determined by array-CGH. Red, focal amplification; orange, gain; blue, loss; gray, missing data. (C) Clinical HER2-status evaluated using clinical guidelines. Panel 1: ERBB2 (HER2) IHC intensity scores (0-3) in > 10% of tumor cells. White, score = 0; light brown, score = 1, brown, score = 2; dark brown, score = 3; gray, missing data. Panel 2: HER2 amplifications as determined by in-situ hybridization (ISH) (ERBB2/CEN17 ISH ratio ≥ 2 or ≥ 4 ERBB2 copies). Red, ERBB2 amplification; white, no amplification; gray, missing data. Panel 3: HER2-positive samples (IHC2+ with HER2 amplification). Blue, HER2 positive; white, HER2 negative; gray, missing data. All panels show the 292 samples. had simultaneous high expression of all three proteins. There were 22 tumors with very strong ERBB2 and ERBB3 protein expression (9 Uro, 12 GU, and 1 Sc/Ne).
The advanced UC tumors were evaluated for clinical HER2 status ( Figure 5A). In total, 88 tumors (29%) were deemed HER2 positive. The proportion of HER2 positive tumors was 38% (41/109) in Uro, 21% (5/24) in UroB, 47% (31/66) in GU, 10% (6/62) Basal/SCC-like, 6% (1/16) in Mes-Inf, 17% (4/24) in Sc/NE, and none in the remaining six infiltrated tumors. Notably, HER2 positive Uro and GU patients showed higher ERBB3 mRNA and protein expression compared to the other subtypes, indicating a different context of ERBB2 amplification and overexpression ( Figure 5B). The Uro tumors may be divided into one group that contain the majority of the < T2 and < G3 tumors found in the dataset (indicated by green in the lower subtype bar in Figure 5), and in a more progressed UroC group (indicated by blue in the lower subtype bar in Figure 5) defined in Sjödahl et al. 2017 [15] with almost exclusively ≥ T2 G3 tumors. The latter group represents a progressed form of Uro distinct from UroB. The advanced Uro tumors exhibited a 34% (14/41) HER2 positive frequency while the UroC group showed a frequency of 54% (19/35), in stark contrast to the 11% seen in non-muscle invasive UroA tumors in the reference cohort ( Figure 1). In contrast, GU tumors were HER2 positive in 45% the reference cohort and in 47% in the advanced cohort.

DISCUSSION
To clarify the molecular context in which ERBB2 and EGFR gene amplifications and overexpression occur we performed an in-depth study of EGFR, ERBB2, and ERBB3 alterations in relation to existing urothelial carcinoma molecular subtypes. In essence, urothelial carcinomas could be divided into two distinct categories based on EGFR, ERBB2, and ERBB3 expression and genomic alterations. One category showed high ERBB2/ ERBB3 expression with concomitant high frequencies of clinically HER2 positive patients. This category coincided very well with the molecular subtypes Uro and GU (Summarized in Figure 6). The Uro and GU subtypes combined correspond to the "Luminal" subtypes of urothelial carcinoma as described by Choi et al. [16], Damrauer et al. [17], and the TCGA Clusters I and II [18].
The increased ERBB2 amplification and expression events seen in the Uro subtype seems to be a part of the Uro progression process from non-muscle invasive to muscle invasive growth as the frequency of HER2 positive cases increased from 11% in the non-muscle invasive UroA to 54% in the more advanced and muscle invasive UroC. In contrast ERBB2 amplification and expression seems to be a founding feature of the GU phenotype [10,19], even though no separate and distinct "HER2 subtype" of UC analogous the breast cancer "HER2 enriched" have emerged. Uro tumors differ from GU tumors by exhibiting high frequencies of FGFR3 and PIK3CA mutations [10] and in muscle invasive urothelial carcinoma FGFR3 mutations are heavily skewed towards the Uro subtype [19]. Furthermore, bi-allelic loss of the CDKN2A locus and overexpression of CCND1 are frequent events in the Uro subtype, indicating a dependency on mitogenic MAPK/PI3-K signaling for cell proliferation [19]. In contrast, genomic events associated with the GU subtype, i.e., RB1 loss, TP53 mutation, and E2F3 gene amplification, allow for a more uncontrolled proliferation [19]. In addition, GU tumors, but not UroA tumors, strongly express FOXM1, a direct downstream target of ERBB2 (HER2) [20], and a FOXM1 related gene signature [13]. Furthermore, in breast cancer FOXM1 and HER2 expression is tightly correlated [21]. Consequently, it may be expected that "clinical HER2 positivity" may have different implications in Uro and GU due to context differences. Additionally, the Urothelial-like subtype is to some extent analogous to the ESR1 expressing breast cancer "Luminal A" subtype [13] that responds less well to HER2 targeted therapy [22,23]. We also noted that less than half of the ERRB2 amplifications found in tumors Red, high FOXA1 and GATA3 expression, and low KRT5 and KRT14 expression; blue, low FOXA1 and GATA3 expression, and high KRT5 and KRT14 expression. Panel 5: Immunohistochemistry scores (TCS) of KRT5, KRT14, GATA3 and FOXA1. Dark brown, high expression; white, no expression; gray, missing data. (B) Panel 1: mRNA heat map of key genes FGFR3, CCND1, CDKN2A, and RB1 that distinguish Uro from the GU subtype. Panel 2: Ratio between the mRNA levels of FGFR3, CCND1, and RB1, and CDKN2A (p16). Red, high FGFR3, CCND1, and RB1 expression, and low CDKN2A expression, blue, low FGFR3, CCND1, and RB1 expression, and high CDKN2A expression. Panel 3: Immunohistochemistry scores (TCS) of FGFR3, CCND1, CDKN2A (p16), and RB1. Dark brown, high expression; white, no expression; gray, missing data. (C) Panel 1: mRNA expression of EGFR, ERBB2, and ERBB3. Panel 2: Ratio between the mRNA expression of ERBB2 + ERBB3, and EGFR. Red, high ERBB2 and ERBB3 expression, and low EGFR expression; blue, low ERBB2 and ERBB3 expression, and high EGFR expression. Panel 3: IHC scores (TCS) for EGFR, ERBB2, and ERBB3. Dark brown, high expression; white, no expression; gray, missing data. Panel 4: ERBB2 amplifications. Orange, ERBB2/CEN17 ISH ratio ≥ 2 or ≥ 4 ERBB2 copies; red, ERBB2/CEN17 ISH ratios ≥ 3 or ≥ 6 ERBB2 copies. Panel 5: ERBB2 S310F/Y mutations indicated in black, ERBB3 V104M/L mutation indicated in black and M91I mutation in gray. of the Basal/SCC-like subtype led to increased mRNA and protein levels, indicating that genomic amplification of 17q12 in Basal/SCC-like, as detected by HER2 ISH analyses, does not have the same consequence as in Uro and GU tumors. The strong association between ERBB2 and ERBB3 overexpression in Uro and GU HER-positive cases, but not in the Basal/SCC-like HER2-positive cases, should also be considered, as ERBB3 has been shown to be essential for ERBB2 driven tumor formation and maintenance [24][25][26]. Hence, recruiting urothelial carcinoma patients for HER2 directed therapy almost certainly have to take the molecular context i.e., molecular subtypes, into consideration.
Even though clinically positive HER2 tumors showed heterogeneity in respect to molecular context, the EGFR positive samples represents a subset of welldefined urothelial carcinoma tumors. EGFR expressing cases coincided well with tumors defined as Basal/SCClike using the consensus definition i.e., KRT5/KRT14 high and FOXA1/GATA3 low [9]. In line with this, urothelial tumors of the Basal/SCC-like subtype respond well to erlotinib in experimental systems [3]. The Basal/ SCC-like subtype show extensive molecular similarities with the breast cancer basal-like subtype as well as the SCC subtype of lung carcinomas [13,27]. Consequently, lessons learned from EGFR targeted treatment in these tumor types may be translated into treatment of Basal/ SCC-like urothelial carcinomas. The fact that the EGFR/ ERBB2 and ERBB3 expression ratio readily identified two distinct classes of urothelial carcinomas strongly associated with the independently determined molecular subtypes shows that the status of the EGFR, ERBB2, and ERBB3 receptors may represent fundamentally different tumor cell phenotypes ( Figure 6).
In conclusion, we argue that future clinical trials using targeted therapies against ERBB2 (HER2) may have to take the molecular context i.e., molecular subtypes, into consideration. It may be that clinical trials up to now have not fully appreciated the molecular background, and thus the molecular heterogeneity, in which activated ERBB2 and EGFR signaling systems are operating in. We believe that a proper tumor classification is critical for arriving at thorough conclusions during the evaluation process of new treatment regimes.

Tumor cohort and data availability
The present investigation makes use of two cohorts, one reference cohort containing all tumor stages and one cohort of advanced tumors. The reference cohort consists of 292 cases established from previously published data [10,12] (GSE32894 and GSE32549) and includes all tumor stages (123 Ta,   like, n = 32). Among the 292 samples, 249 had copy number array data derived from one or more platforms [12,14,28,29]. Tumors showing narrow amplifications peaks on the lower resolution array platforms were hybridized to a custom made high density zoom-array with an average genome-wide probe spacing of 17 500 bp and between 7 000 -12 000 bp in select target regions [30]. All copy number profiles and copy number calls for EGFR, ERBB2, and ERBB3 were manually evaluated. For 217 samples both copy number data and IHC data were available.
The advanced cohort consists of 307 formalin-fixed paraffin-embedded TUR-B samples from patients that underwent radical cystectomy between 2006 and 2011. Of these, 243 were muscle invasive in the pathological TUR-B evaluation. Generation of gene expression and immunohistochemistry data is described in detail by Sjödahl et al. 2017 [15]. Briefly, tissue-microarray cores were positioned to yield high tumor cell content and to be morphologically and histologically representative of the tumor. RNA extraction was performed on macrodissected tissue areas located close to the TMA core sites, using HighPure isolation kits (Roche). RNA was amplified and labeled using SensationPlus kits (Affymetrix) and hybridized to the Gene ST 1.0 platform (Affymetrix). Raw-and normalized data is deposited in Gene Expression Omnibus under GSE83586. The tumors were classified using a combination of global mRNA gene expression analysis and IHC analysis into Urothelial-like, Urothelial-like B, Genomically Unstable, Basal/SCC-like, Mesenchymal Infiltrated, and Small cell/Neuroendocrine subtypes as detailed in Sjödahl et al. 2017 [15]. Six heavily infiltrated cases could not be assigned to any of the subtypes. These were categorized as "infiltrated", and referred to as "Inf" in the text. Array CGH data was not available for this cohort. Informed consent was obtained from all patients and the study was approved by the Local Ethical Committee of Lund University in accordance with the Helsinki declaration.

Immunohistochemistry
Tissue microarrays (two 1.0 mm cores per tumor, 4 µm sections) were analyzed with antibodies against EGFR (3C6, Ventana), ERBB2 (4B5, Ventana), ERBB3 (Dako, DAK-H3-IC). TMA cores were evaluated as blinded digitalized image files. Each TMA slide contained 120 randomly selected tumor cores to obtain an efficient distribution between low (blank) and high expressing cases. Antibody dilutions were tuned to obtain a staining ranging from blank to intensive staining. IHC stainings for EGFR, ERBB2, and ERBB3 were quantitatively scored from 0 to 3. When hetrogeneous staining was observed within a single core, the fraction of tumor cells with each score was recorded and multiplied with the intensity for a tumor cell score (TCS). The mean tumor cell score of core pairs from the same sample was calculated. includes GU Ta and T1 tumors; Uro MI, includes Uro, UroB, and UroC cases from the series of advanced tumors; GU MI, includes GU cases from the series of advanced tumors; Basal/SCCL MI, includes Basal/SCC-like cases from the series of advanced tumors. Non muscle invasive cases of Basal/SCC-like were too few to be included. Panel 1: Summary of EGFR, ERBB2, and ERBB3 mRNA expression. Red, high expression; green, low expression. Panel 2: Summary of EGFR, ERBB2, and ERBB3 protein expression. Dark brown, high protein expression; white, low protein expression. Panel 3: Summary of HER2 positive cases. Dark blue, high frequency of HER2 positive cases; white, low frequency of HER2 positive cases. Immunohistochemistry analyses of KRT5, KRT14, GATA3, FOXA1, FGFR3, CCND1, CDKN2A (p16), and RB1 was as reported by Sjödahl et al. 2017 [15].

Clinical HER2 status
Clinical HER2 status was assessed using dual in-situ hybridization (ISH) and IHC according to Swedish guidelines for HER2 testing in breast cancer [31]. The clinical scoring was performed on a separate section, independently evaluated by an experienced bladder and breast cancer pathologist (GC) using Ventana Pathway HER2 4B5 and Ventana Dual ISH assay HER2 DNP and CHR17 DIG probes on the BenchMark Ultra platform. The IHC score (0, 1+, 2+, 3+) was determined by the highest staining observed in >10% of the tumor cells. An ERBB2/ CEN17 ISH ratio ≥ 2, or ≥ 4 ERBB2 copies when CEN17 was not determinable, was used for amplification calls. In samples with two TMA cores the highest ratio was used for calling amplifications. HER2 amplified tumors with IHC 2+ scores in > 10% of cells were considered HER2positive. The inter-observer agreement between the IHC core intensity scores by GS and the clinical re-evaluation by GC was excellent (kappa 0.807 with equal weights).