A total of 58 patients were approached for consent. Out of these, 42 patients were ineligible, 3 declined and 13 consented. Of the 13, 8 were enrolled in the study while 5 failed to meet the inclusion criteria. Among the 8 FAS, 5 patients completed the treatment protocol while 3 discontinued due to either inability to comply, serious adverse events or investigator’s decision (Fig. 1).

Consort diagram. A total of 58 patients were approached, 13 consented and were screened with 8 patients enrolled. Five patients completed the treatment regime while 3 withdrew due to non-compliance, SAE or clinical decision

The baseline characteristics of the ALERT cohort (n = 8) and those who had at least one follow-up tumour assessment (performed at months 3, 6 and 9) (n = 6) are shown in Table 1 with full tumour assessment data per patient presented in Supp Table 1.

Among the 8 enrolled patients, 6 were given AI treatment (detailed in Table 2). The two patients (007 and 009) who did not receive any AI treatment withdrew from the study after 28 and 53 days. Patient 007 had inability to comply with the protocol (Supp Table 2) whereas patient 009 suffered SAEs (Supp Table 3). Both patients left the study before their month 3 follow-up. Furthermore, patient 010 left the study before the month 6 follow-up (day 78) due to poor clinical response and a decision to switch treatment. Finally, patient 002 had no more information for month 9 after developing progressive disease on month 12. Of the remaining patients, 004 and 006 completed two 9-week alternating cycles of eribulin and hormonal therapy, with patient 011 missing just the final week (week 36) of AI treatment. Patient 002 completed C1 eribulin/AI followed by C2 eribulin only (no further AI). Patient 008 completed C1 eribulin/AI followed by C2 eribulin plus 1 additional week (week 28) of AI treatment (Table 2). AI drug compliance data (the proportion of days with dose completion) were available for 5 patients whereby the median daily compliance was 100% (IQR: 95–100%) (based only on days with available treatment records) (Supp Table 2).

The median duration of treatment completion (from date of entry to either study completion or withdrawal from the study) for the 8 enrolled patients was 202 days (Range: 28–293 days). For the 6 patients with at least one tumour assessment, the median duration of treatment completion was 261 days (Range: 78–293 days) whereby the median tumour size changes were 27%, 15% and 2% decrease at months 3, 6 and 9, respectively (Table 3). Looking first at the sum diameters for all target lesions (Supp Table 1) by 3 months (n = 6), 3/6 patients had stable disease (SD) and 3/6 had a partial response (PR). By month 6 (n = 5), 3/5 had SD and 2/5 had progressive disease (PD). By month 9 (n = 4), 3/4 had SD and 1/4 PD. Focussing on the 3 patients who completed two alternating cycles of eribulin/AI, patients 006 and 011 remained stable over the 9-month duration and patient 004 remained stable for 6 months with PD by month 9. The remaining 2 patients (002 and 008) continued the treatment regime, albeit with reduced AI duration (total of 9 and 10 weeks of AI, respectively); these both showed a PR by month 3 with PD by month 6 (C1 eribulin/AI + C2 eribulin).

The median PFS at month 3 cannot be calculated because no patient experienced disease progression at this follow-up time. At month 6, the observed median PFS was 202 days (95%CI: 135-undefined), while at month 9, the median PFS was 235 days (95%CI: 235-undefined) (Table 4). The 95% CI cannot be accurately computed due to the small sample size and limited number of events. Notably, the 3 patients reporting the longest PFS (004, 006 and 011, Table 4) were the 3 patients who received 2 alternating cycles of eribulin/AI treatments, followed by patients 008 and 002 who continued the treatment regime with reduced AI duration (Table 2). Overall, CB was reported in 4 out of 8 (50%) patients (004, 006, 008 and 011), these being 4 of the 5 per-protocol analysis set (Table 4).

Safety data were reported among all the patients (n = 8) who received at least one dose of study treatment (Table 5). In total, there were 121 adverse events (AEs) recorded from all patients, of which 10% were Grade 3 AEs and no Grade 4 and 5 AEs. Almost all (94%) of observed AEs were not serious and 86% are expected events. In terms of relationship with the study drug, 15% of the AEs are classified as ‘definitely related’, 13% are ‘probably related’ and 25% are ‘possibly related’. There were 8 serious adverse events (SAEs) observed from 5 patients during the study period (Supp Table 3). Patient 007 experienced neutropenic sepsis, classified as ‘definitely related’ to the study medication and also mucositis, deemed probably related. Patient 009 experienced dizziness and confusion, probably related to the study. Patient 010 experienced haematemesis, possibly related to the study. The 4 remaining SAEs, including hypercalcemia (patient 002), dyspnoea (patient 008), respiratory distress syndrome, confusion and multi-foci acute ischemia (patient 009) were classified as unlikely or not related to the study (Supp Table 3).

The concentration of total plasma cell-free DNA (cfDNA) at screening and 9-week intervals thereafter (n = 24) was monitored (Supp Table 4). At the point of screening, the median cfDNA concentration of the 8 patients was 1.1 ng/ul (range 352 pg/ul to 13.8 ng/ul). For patients who received 2 alternating cycles of eribulin/AI treatment (004, 006 and 011), cfDNA concentrations showed little fluctuation from screening level for patient 004 (consistent with CB, Table 4), a fivefold increase at end of treatment (EoT) above screening level for patient 006 (consistent with overall CB) (Table 4) but inconsistent with the SD indicated at 9 months (Supp Table 1) and finally little fluctuation from screening in the week 9 and 18 samples for patient 11, consistent with the SD indicated at month 3 (Supp Table 1). Of the remaining patients, longitudinal cfDNA samples were collected for patients 002 and 008. Patient 002 cfDNA levels decreased post C1 eribulin, increased post C1 AI therapy and decreased again upon C2 eribulin, consistent with no overall CB (Table 4). Patient 008 presented with a high level of cfDNA at screening, a significant decrease (89% that of screening) by week 9 (post C1 Eribulin) with further decrease (91% that of screening) by week 18 (post AI) and again by week 36 (post C2 eribulin plus week 28 AI only) before increasing at EoT, consistent with the CB reported in Table 4.

Oncomine™ Breast cfDNA assay analysis was carried out in all 24 cfDNA samples for detection of ctDNA and compared with matched genomic DNA samples and FFPE tumour DNA (available for 6 patients). Analysis detected SNVs in breast cancer driver genes in all 6 FFPE tumour DNA samples; 5 patients had a PIK3CA driver mutation and concurrent ESR1 mutation(s) and all 6 had multiple low-level polyclonal variants in TP53 (Supp Table 5). Furthermore, SNVs were detected in the cfDNA of 4 of the 8 patients, 002, 006, 008 and 011 (Fig. 2A–D, respectively, and Supp Table 5). A single germline TP53 mutation (p.H214Y) was detected (15% VAF) for patient 009 (Supp Table 6).

Oncomine™ cfDNA Breast Assay analysis of longitudinally collected cfDNA samples from patients (A) 002, (B) 006, (C) 008 and (D) 011. Treatment received, including the weeks of AI treatment is show for each patient. Time points at which cfDNA samples were not collected are indicated in grey font on the x-axis

Concerning plasma cfDNA, patients 006 and 011 (Fig. 2B, D, respectively, and Supp Table 6) each completed 2 cycles of alternating eribulin/AI treatment. For patient 006, 3 mutations were detected in plasma cfDNA at screening (ESR1 p.Y537S 0.18%, PIK3CA p.E542K at 0.12% and ERBB2 p.V104M 0.11%). ESR1 p.Y537S persisted but reduced to 0.08% on switch to AI and the other 2 mutations resolved. This pattern remained stable following 9 weeks of AI, consistent with CB (Table 4). However, after switching back to eribulin, p.Y537S and p.E542K significantly increased to 4% and 4.9% VAF, respectively, in the EoT samples (Supp Table 6) indicative of progression.

For patient 011, TP53 p.K132E was detected at low levels (< 0.1%) at screening, 9 weeks (post C1 eribulin) and 18 weeks (post C1 AI) with no significant change in VAF. Other variants (KRAS p.G12V, PIK3CA p.E542K and TP53 p.E258K) were detected at low levels, but each in just a single sample. ESR1 (p.V392I, 0.23% VAF) and ERBB2 (p.V104M, 0.59% VAF) mutations detected in the FFPE tumour were undetected in plasma suggesting that these were subclonal mutations not detectable in ctDNA.

Two other patients (002 and 008) had ctDNA detected (Fig. 2A, C, respectively, and Supp Table 5), each completed one full cycle of eribulin/AI (Table 2), with patient 008 receiving additional AI treatment for one week only (week 28). Patient 002 had 6 SNVs detected in ctDNA at screening (ESR1 p.Y537N 9.4%, ESR1 p.Y357S 2.5% and ESR1 p.D538G 0.5%, PIK3CA p.N345K 18%, KRAS p.G12D 0.11% and TP53 p.V272L 0.17%). By week 9 (post C1 eribulin), 3 variants had resolved and 3 were still detected but at a lower VAF (PIK3CA p.N345K at 2.4%, ESR1 p.Y357N at 1%, and ESR1 p.Y357S at 0.2%) suggesting response to treatment. By week 18 (post 9 weeks AI), the VAF had increased significantly to 30.8% for both PIK3CA p.N345K and ESR1 p.Y537N; ESR1p.Y357S also increased to 1.2% and KRAS p.G12D reappeared. By the EoT sample (post C2 eribulin but no further AI), detectable ctDNA had reduced significantly to 4.1% (PIK3CA p.N345K), 2.4% (ESR1 p.Y537N) and 0.1% (ESR1 p.Y537S) (Supp Table 5).

Patient 008 had high ctDNA levels detected at screening (PIK3CA p.H1047R, 60.7% VAF) with a low frequency putative subclonal T53 p.R213Q variant at 0.08% VAF. The PIK3CA p.H1047R VAF reduced to 2.2% by week 9 (post C1 eribulin); T53 p.R213Q resolved, but a second putative subclonal PIK3CA p.E524K mutation was detected at 0.05% VAF, which was then undetected at week 18. However, PIK3CA p.H1047R VAF increased to 4.4% by week 18 (post C1 AI), 13.6% by week 36 (post C2 eribulin plus week 28 AI) and stabilised at 13.6% by EoT (Supp Table 5).