Patient characteristics

In the phase Ib portion of this study, 27 patients were enrolled between February 2015 and July 2017. Nineteen patients were evaluable having had at least one follow-up imaging study; three patients withdrew from the study in cycle 1 and 5 patients with a dose-limited toxicity in cycle 1 did not have an imaging assessment. All phase II patients who received at least one dose of study drugs were included in the analysis. Twenty-two patients were evaluable for toxicity and 20 were evaluated for efficacy with at least one scan performed after their third cycle. Two non-evaluable patients who withdrew from the study did not have their first scan but are included as PD. Median age was 55.5 years (range 32–70). Hormone status (ER and/or PR) was positive in 13 patients and negative in 9. All patients were HER2-positive at baseline by local determination (Additional file 1: Table S3).

Safety assessment

Similar to phase Ib patients, diarrhea was the most frequent toxicity in phase II: grade 2, 6 patients (27%); grade 3, 8 (36%). Other grade 3/4 toxicities included: thrombocytopenia, 2 patients (10%); transaminase elevation, 3 patients (15%); and pneumonitis, 1 patient (5%). There were no unanticipated toxicities in the phase II expansion.

Efficacy

Among 19 evaluable patients in phase Ib, there were 3 CRs and 9 PRs for an ORR of 63% (12/19) [10]. In phase II, including all patients who received at least one dose of therapy, there were 2 CRs, 5 PRs for an ORR of 32% (7/22), and 3 SDs of 180 days or longer making the clinical benefit rate (CBR) 45% (10/22). In phase Ib and II, nine patients had sustained objective responses lasting approximately 1 year or longer (range 343–1453 + days, Additional file 1: Table S4; Additional file 2: Table S5). Treatment was discontinued at or before the first scan in 15 patients for a variety of reasons, including 5 DLTs (all in phase I) and 10 with clinical progression in phase I and II.

ctDNA clearance and treatment response

Because clearance of ctDNA has been associated with treatment response, we compared the outcomes of patients who were positive or negative for ctDNA after study treatment. The response rate among the ctDNA-positive patients who were still ctDNA-positive after study therapy was 9/19 (47%), but the ctHER2 DNA-positive patients who became ctDNA-undetectable at C2D1, the response rate was 6/6 (100%), demonstrating that the loss of ctDNA was associated with a very good response.

HER2 amplification in tissues and blood

We assessed the HER2 amplification status of TP0 and TP1 with a CLIA HER2 IHC/FISH assay, and with an Ampli-Seq NGS assay. These tissue samples were also compared to the HER2 amplification status in blood samples collected at C1D1 and C2D1 (Fig. 2B). There was good concordance between the CLIA HER2/FISH and Ampli-Seq assays (100% in TP1 tissues and 85% in all tissues), which demonstrated the technical accuracy of the Ampli-Seq. Concordance between TP0 tissue with IHC/FISH and C1D1 ctDNA was 71% (20/28). Concordance between C1D1 and C2D1 was 54% (14/26). Anti-HER2 treatment is potentially the causal reason for the discordance between C1D1 and C2D1, which showed a total loss of ctDNA in some samples and a loss of HER2 amplification in others.

Fig. 2

Amplification Status of Tissues and Blood: NSABP FB-10. A Response rates (CR/PR, CBR, and SD) for FB-10 HER2-amplified and non-amplified patients based on ctDNA results. B HER2-amplification status of FB-10 tumor tissues based on CLIA tests (IHC/FISH) and Ampli-Seq (Tissue) in baseline (TP0) and study entry (TP1) samples are shown. HER2-amplification status was determined in ctDNA at C1D1 and at C2D1 with the Guardant360 assays. Responses, amplification status, and changes in copy number in ctDNA between C1D1 and C2D1 are indicated as shown in the legend

Using the Guardant360 assay cut point of 2.14 for amplification among 43 C1D1 samples (22 in phase Ib, 21 in phase II), 6 patient samples were indeterminate (including 4 for which somatic mutations were not detected) (Fig. 2B, dark green), 1 was not evaluable (NE), and 1 other failed quality control. Among the remaining 37 samples, there were 21/37 (57%) patients with amplification and 17/37 (46%) without. The objective response (CR, PR) rate was 55% (11/20) in amplified patients and 41% (7/17) in non-amplified patients. The CBR in patients with ctHER2-amplification was 12/21 (57%) and in non-amplified patients it was 8/17 (47%). There was one patient with SD who was ctHER2-amp indeterminate. Mean duration of response (CBR) was substantially longer in amplified patients, 457 days compared to 131 days in non-amplified patients (P = 0.008) (Fig. 2A; Additional file 1: Table S4).

We compared progression-free survival (PFS) of patients whose ctDNA or tumor tissues had HER2 amplification to patients with no HER2 amplification. Patients with ctHER2-amp at C1D1 or in their TP1 tumor tissue had a significantly longer PFS than patients with no HER2 amplification (Fig. 3A–E).

Fig. 3

Association of HER2-amplification Status with Patient Outcomes: NSABP FB-10. A Kaplan-Meier plots of patients with or without HER2 amplification in ctDNA or in TP0 tissue (B & D), or in TP1 tissue (C & E) based on IHC/FISH (B & C) and on Ampli-Seq (D & E)

In phase I and II there were 26 C1D1 and C2D1 pairs, 15 with and 11 without ctHER2-amp at C1D1. Among the 15 with ctHER2-amp at C1D1, 14 showed HER2 loss at C2D1 as defined by a loss ≥ 28% of HER2 copy number or no ctDNA detected. The ORR among these 14 patients was 71% (10/14). Of the 10 responders, 5 cleared ctDNA completely by C2D1, 3 had detectable ctDNA but no ctHER2-amp, and 2 were HER2-positive but the amplification level in C2D1 had decreased dramatically (Fig. 2B; Additional file 2: Table S5). The 2 remaining patients with detectable ctHER2-amp with no loss of HER2 amplification had PD, suggesting that a loss of ctDNA and/or a loss of HER2 ctDNA amplification was a marker for a good response to study therapy. However, in 11 patients with no HER2 ctDNA amplification at C1D1, the ORR was 45% (5/11), indicating that some non-amplified tumors were responsive to study treatment.

Molecular response by ctDNA

We assessed the association between molecular response and objective radiologic response (Fig. 4A). A total of 21 patients (9 phase Ib and 12 phase II) had paired samples that met criteria for assessment of molecular response. Criteria included ≥ 1 alteration present in one of the paired samples plus a mutant molecule count of ≥ 15 in either sample. Molecular responders demonstrated a longer PFS compared to non-responders (median PFS 7.4 vs. 2.8, HR 0.28, 95%CI 0.09-0.90, P=0.033 using Wilcox test). [0.09–0.90, P = 0.033 using Wilcox test]. We also examined the association between molecular response and best RECIST response. Patients with CR/PR/SD had significantly lower Molecular Response values compared to patients with PD (P = 0.037; Fig. 4B).

Fig. 4

Molecular Response and Patient Outcomes: NSABP FB-10. A Kaplan–Meier curves showing association of MR with PFS using a molecular response cutoff of 50%. B Association between molecular response and best RECIST response

Mutations/variants in tissues and ctDNA

Because ERBB2 is the target of the study therapies, we have examined both tissue and ctDNA for mutations in the ERBB2 gene. No ERBB2 variants in tissue at a VAF of ≥ 10% were observed, however, in ctDNA 3 nonsynonymous, ERBB2 variants (I767M, V777L, and S310Y) were detected in the C1D1 samples from 3 patients. These variants have been associated with sensitivity to neratinib in breast cancer patients [18]. In this study, the patients whose tumors had a V777L or a S310Y mutation had a PR, but the one patient with a I767M mutation had PD with brain metastasis. The tumor with the I767M mutation also had a P1233L mutation [19]. Interestingly, in an exhaustive meta-analysis of 37,218 patients, including 11,906 primary tumor samples, 5,541 extracerebral metastasis samples, and with 1485 brain metastasis samples found that a nearby ERBB2 mutation (P1227S) was the only mutation restricted to brain metastasis. It is unknown whether any of these mutations played a role in the patient responses or the course of disease, but it is of interest to note them [20].

We examined DNA variants in all available TP0 and TP1 tissues using our NAR Ampli-Seq panel, which included ESR1, HER2, and 115 other genes in HER2-activated pathways. Based on our stringent criteria for variant detection, i.e., VAF ≥ 10%, plus other criteria as described in Additional file 1, we identified 27 variants among 28 samples, representing 21 patients (Fig. 5A).

Fig. 5

Variant Alleles in Patients and their Responses: NSABP FB-10. A Variants detected with a VAF of ≥ 10% in patients with PD, SD, PR or CR. *indicates a stop codon. B Variant alleles with a VAF of ≥ 5% in patients with PD, SD, PR or CR

The frequency of PIK3CA mutations among all sequenced patients was 34% (12/35), similar to that seen in other studies of unselected metastatic and early-stage breast cancer patients (cBioPortal). All of the mutations were in exons 9 and 20 at amino acid 545 and 1,047, respectively. These PIK3CA variants also have the highest VAFs (ranging from 10 to 72% across samples), however, PIK3CA mutations do not appear to influence patient outcomes, because response rates between PIK3CA mutant and WT tumors were similar: 42% (4/12) versus 45% (14/31), respectively. In one case, a PIK3CA mutation was detected only in TP1 but this patient had a PR, again indicating that PIK3CA is not a resistance marker for study therapy. Variants detected only in PD or CR patients represent potential resistance or sensitivity markers, respectively, to study therapy. Mutations detected only in TP1 samples among the 12 paired TP0/TP1 cases, included ADAM17_S770L, ERBB4_E1010K, ERBB4_R1040T, and IL6ST_S834* in one sample and an ESR1_EY537S mutation in another (Fig. 5A). Both patients had PD, perhaps indicating that these mutations may have emerged in response to prior therapies. Details of variants are presented in Additional file 1.

Whole transcriptomic profiling

We examined the PAM50 subtypes and the 8-gene trastuzumab benefit signature in all available tissues [16]. Among the 29 patients with response and gene expression data for TP0 tissue, we found that 19/34 (56%) were HER2E, 8 (24%) were luminal B, 4 (12%) were basal, 2 (5.9%) were normal, and 1 (2.9%) was luminal A. Patients with luminal subtype tumors had a lower CR/PR response rate (1/8 [12.5%]) than patients with a non-luminal subtype (12/23 [52%]) (Additional file 2: Table S5). Among the TP1 samples with gene expression data, the frequency of CR/PR was 1/4 in luminal patients and 5/9 in the non-luminal patients. Intrinsic subtypes differed between TP0 and TP1 tissues in some cases (Additional file 1: Table S4). Although numbers are small, the frequency of CR/PR rates were consistently lower among the luminal patients than non-luminal patients.

The 8-gene trastuzumab signature is a validated signature for identifying patients with large-, moderate- or no-benefit from trastuzumab when added to chemotherapy in the adjuvant setting [16, 17] and has been shown to associate with pCR rates in the neoadjuvant setting [21, 22]. We questioned whether this signature may also show an association with response in the metastatic setting. Among the large-, moderate- and no- benefit groups the percent of CR/PR patients was 67% (4/6), 50% (9/18), and 29% (2/7), respectively (data in Additional file 2: Table S5).

As expected, the level of HER2 RNA increased as the IHC status increased (i.e., 0, 1 + , 2 + , to 3 +) (Additional file 1: Fig. S2). In TP1 samples they were concordant with patient responses suggesting that HER2 RNA expression in study entry is associated with response to T-DM1 + neratinib (Fig. 6).

Fig. 6

RNA Expression Levels and Response to Therapy. RNA expression levels in TP0 tissues (A) and in TP1 tissues (B) from patients with PD, SD or CR/PR. The units for RNA expression were log 2 expression values

Significant differences were detected in RNA levels between IHC 1 + and 3 + and between 2 + and 3 + but not between 0 and 1 + nor between 1 + and 2 + (Additional file 1: Fig. S3). Although numbers are limited, these data show that the RNA levels are not different between 0 and 1 + . These patients may benefit from treatment with other ADCs. The DAISY and DESTINY-Breast04 trials signal that T-DXd may have significant activity in HER2-low patients [8, 23].