Formalin-fixed, paraffin-embedded (FFPE) tumour samples were collected as part of the PeriOperative Endocrine Therapy-Individualising Care (POETIC) trial (CRUK/07/15) from postmenopausal women with primary ER + breast cancer. For eligibility to the trial, ER status was determined locally. Full details of the eligibility criteria, the conduct of the trial and its primary outcome data are published elsewhere [11]. In brief, the trial randomised 4,480 postmenopausal women recruited between 2008 and 2014 to receive aromatase inhibitor (AI) 2 weeks before and 2 weeks after surgery or no perisurgical treatment. Patients were recruited irrespective of HER2 status, which was also established locally by immunohistochemistry and fluorescent in situ hybridisation on the surgery sample. Tumours needed to be palpable or at least 1.5 cm by ultrasound.

In this exploratory substudy, only baseline samples with Ki67 ≥ 10% (to maximise the precision of estimates of AI-induced proportional changes) were selected from the AI-treatment group. These were categorised by (i) HER2 status and then (ii) by degree of Ki67 response as a percentage of baseline after 2 weeks of AI. For HER2 − cases, all samples from the 15% worst responders (PR), 30% of the 35% of patients with intermediate response (IR) and 30% of the 50% best responders (GR) were analysed. Samples for all 3 groups in this HER2 − subset were matched by baseline Ki67 within groups 10–20%, 20–30%, 30–40% or 40 + %. All HER2 + samples were analysed, and the same Ki67 cut-offs were applied to define PR, IR and GRs.

Tissue for IHC and mRNA were taken as sequential sections from the same block.

Staining was undertaken at the Ralph Lauren Centre for Breast Cancer Research. ER and PgR staining were analysed in the baseline sample, whilst Ki67 staining was evaluated in the baseline and 2-week sample. ER staining was done using the ER antibody (NCL-L-ER6F11, Novocastra, Leica) at 1:400 dilution (8.5 µg/ml) and PgR antibody (L-PgR-312 clone 16, Novocastra, Leica) at 1:300 dilution (11.6 µg/ml). Antigen retrieval for both was done in Envision FLEX Target Retrieval Solution low PH (DM829, Dako Agilent) in a Dako PT Link Tissue Processor at 97 °C for 20 min. The Envision FLEX detection system on the Autostainer from DAKO was used. Ki67 staining was done with the Anti-MIB1 clone (M7240, Agilent DAKO UK).

The ER scoring method was adapted from a global scoring protocol that has been developed by the IKBCWG for Ki67 scoring [12, 13]. The whole section was examined in order to estimate the percentages of the invasive tumour component exhibiting relatively high, medium, low or negligible ER staining frequencies. Based on these estimates, the assessor decided on which fields to score for each ER staining distribution in the whole tumour. In total, four representative high-power fields (HPF) of invasive breast cancer were selected and in each HPF, 100 invasive tumour cells were scored. The number of ER-positive nuclei was counted irrespective of the staining intensity. ER positivity was calculated as the percentage of the total number of ER-positive invasive tumour cells in all assessed fields relative to the total number of invasive tumour cells. Where there were no positive cells or the score was < 1%, the presence or absence of normal/benign ducts was noted and whether they contained ER-positive nuclei; if tissue was available the staining/scoring of cases scored as < 1% was redone to confirm the negative result and to avoid the possibility that this was a false-negative/low result in the central laboratory.

The percentage of PgR positivity was assessed by visual estimation using a light microscope. The whole section was examined at low, medium and high power in order to estimate manually by eye the % PgR-positive nuclei present in the invasive breast cancer. PgR was deemed positive when ≥ 1% of cells were positive.

RNA was coextracted with DNA from three 10 µm FFPE sections from the baseline block of patients using the ROCHE High Pure miRNA isolation kit for RNA (Roche, Basel, Switzerland) and the Allprep FFPE kit for DNA (Qiagen) following SOP M027 from The Cancer Genome Atlas (TCGA) Program developed by the Biospecimen Core Resource (BCR) at Nationwide Children’s Hospital in Columbus, Ohio. Quantitation was done using the high-sensitivity RNA Qubit assay (Thermo Fisher Scientific, Carlsbad, CA).

Two hundred nanograms of RNA was reverse-transcribed using SuperScript IV VILO Master mix (Thermo Fisher Scientific, Carlsbad, CA) following the manufacturer’s instructions, with thermocycling conditions: 25 °C for 10 min, 50 °C for 10 min and 85 °C for 5 min.

ESR1, ACTB and TFRC levels were measured by RT-qPCR based on TaqMan commercial probes in a QuantStudio 6 FLEX PCR Detection System (Applied Biosystems, Foster City, CA). Hs00951083_m1 (TFRC), Hs01060665_g1 (ACTB) and Hs01046816_m1 (ESR1) were the probes used. Data analysis was performed with QuanStudio FLEX software V.1.7.1.

The RT-qPCR mix was prepared according to TaqMan Fast Universal PCR Master Mix instructions (Applied Biosystems, Foster City, CA). The PCR mix contained 2.5 µl Universal Master Mix, 0.25 µl of the Assay of interest, 0.25 µl of Water and 10 ng of cDNA. The reactions were performed in triplicate, and any of these with standard deviation > 0.3 was excluded. The PCR cycler conditions were the same for all 3 assays: 50 °C 2 min, 95 °C for 5 min and 40 cycles of 95 °C for 2 s and 60 °C for 25 s.

The relative mRNA level of ESR1 was determined as 2^−Delta Ct (2^-(Ct ESR1 in test – Ct average of ACTB/TFRC in test)). The primary result of a real-time PCR is a ct value, and by calculating the delta-ct value we obtain the relative mRNA level measured as units of relative expression.

Ki67 residual percentage is calculated by dividing Ki67 expression levels in the surgery sample by Ki67 expression levels in the baseline sample. Spearman rank correlation was used to analyse the correlations between ER, PgR, Ki67 residual percentage and ESR1 mRNA levels. This analysis results in a rho value, which measures the strength of the association between the two variables analysed. For IHC, comparisons were made between ER categories of < 1%, ≥ 1 < 10 and ≥ 10%. Among the cases ≥ 10%, further categories were considered as follows: ≥ 10 < 20, ≥ 20 < 40, ≥ 40 < 60, ≥ 60 < 80 and ≥ 80.

As described above, ALL 15% poor responder patients from the HER2 − POETIC patient group were selected; however, only 30% of the 35% intermediate and 50% good responders were selected due to the time/cost constraints of analysing over 2000 patients if we had analysed the whole HER2 − POETIC population. Thus, when making comparisons between the prevalence of PR vs GR and/or IR, the number of cases that were GR or IR was multiplied by 3.3 to account for our assessing only 30% of that category. This results in some of the estimated total cases being non-integer. Cut-points would be identified according to the proportion of PRs, with at least 90% being considered desirable.