Tumour-infiltrating lymphocytes as a prognostic and tamoxifen predictive marker in premenopausal breast cancer: data from a randomised trial with long-term follow-up

The patients in this study participated in the SBII:2pre study and details of the study have been previously presented [18, 21, 22]. Briefly, during 1984–1991, 564 premenopausal women with stage II invasive BC (UICC TNM, third edition (1982)) were randomised between 2 years of adjuvant TAM or no systemic treatment. Two coordinating centres including 20 hospitals participated in the study: the South Eastern (Oncological Centre Lund) and Southern (Oncological Centre Linköping) Health Care Regions. Four patients were excluded in the latest update of the study due to protocol violations found by scrutiny of the patient records [18]. Among the included 560 patients, 284 were randomised to the control arm and 276 to the TAM treatment arm (Fig. 1). In the present study, the prognostic value of TILs was evaluated in patients allocated to no systemic therapy with tumours successfully scored for TILs and available IHC/in situ hybridisation (ISH) data for defining BC subtypes. All patients with ER+ tumours and successfully annotated TILs were included in the assessment of the prediction of TAM efficacy (Fig. 1).

Data on invasive distant, regional and local recurrence, contralateral BC, BC-related death and death due to other causes were determined based on a thorough review of all medical records and the Swedish Cause of Death Register as previously described [18].

Tissue microarrays were used for assessment of ER, PR, Ki67 and HER2. ER and PR were assessed by IHC and ER/PR-positivity was defined as tumours with > 10% stained nuclei according to Swedish Guidelines [23]. Data on both IHC and the cytosol-based method were available; for tumours with missing IHC data, the results from the original cytosol-based methods were used (ER: n = 32; PR: n = 46). Ki67 was assessed as a categorical variable (≤ 10%, 11–25%, ≥ 26%) [24]. Tumours were classified as HER2+ either by HER2 amplification by fluorescent ISH (n = 54) or by HER2 3+ as assessed by IHC in cases in which ISH data were missing (n = 12). Histological grade was evaluated as NHG according to Elston et al. [25]. The tumours were stratified into three subtypes based on IHC and ISH markers: ER+/HER2−, HER2+ (irrespective of ER status) and TNBC (ER−/PR−/HER2−).

Archival formalin-fixed paraffin-embedded tissues from breast tumours in the SBII:2pre trial were collected from seven regional biobanks and stained by haematoxylin-eosin (n = 520). Of these, 488 were available for TIL scoring and 486 for assessment of LVI. The scoring of TILs was performed according to the definition by the Immuno-Oncology International TILs Working Group, in which stromal TILs (referred to as TILs in the current study) are defined as the proportion of the stromal area containing infiltration of lymphocytes with no direct contact with invasive tumour cells [5]. Microscopic assessment was performed by a board-certified breast pathologist (Ute Krüger) blinded to the patient characteristics and outcomes. TILs were assessed under a light microscope (BX63F, Olympus, Japan) with a magnification of 40 and 100 (if necessary, 200). Tumours were categorised into the following groups based on TIL infiltration: < 10%, 10–49%, 50–74% and ≥ 75%. In the prognostic analyses, the two latter groups were merged into one category (high), and the three groups were then denoted as low, intermediate and high. Tumours with TILs ≥ 50% were also defined as LPBC and tumours with low/intermediate TILs (< 50%) were defined as non-LPBC in the predictive analyses. Photomicrographs of the different TIL categories are shown in Fig. 2.

According to the Swedish pathological guidelines, LVI was defined as present when tumour cells in cavities lined with endothelium (not by IHC endothelial markers) were verified in the peritumour area [23]. Patient and tumour characteristics have been reported previously [18, 21, 22] and are listed in Additional file 1, stratified by tumours with and without scored TILs.

Differences in distribution between clinico-pathological variables and TILs were analysed by chi2 test and chi2 test for trend. The primary endpoint was BC-free interval (BCFi), defined as the first event of local, regional or distant recurrence, contralateral BC (invasive or ductal cancer in situ (DCIS)) or BC-related death. The association with overall survival (OS) was also explored. The data cut-off date for events was November 30, 2016.

The cumulative incidence of BC-related events as a function of follow-up time was estimated and compared among patient subgroups to handle the problem with competing risks. To facilitate the comparison of results for the two endpoints, BCFi and OS, cumulative incidence, which is the same as one minus the Kaplan–Meier estimate, was also used for the endpoint death from all causes. The log-rank test was used to evaluate the evidence for difference between cumulative incidence curves and the trend version of the test was used for comparison with more than two ordered groups. Hazard ratios (HR) were calculated by Cox regression analyses, stratified by region.

The follow-up was censored in the analysis of BCFi if a patient died from a cause that was not BC-related without a preceding BC event included in the definition of an event for BCFi. Hence, the estimated HRs for BCFi in this cause-specific Cox regression analysis should be interpreted in an imaginary world where all other causes of death have been eliminated. In the predictive multivariable analyses, PR status was omitted due to collinearity with ER status. A Cox model with a term for interaction between TIL subgroup (LPBC vs. non-LPBC) and TAM was fitted to evaluate the evidence for differential effect of TAM in the two TIL subgroups. Exclusion of the seven cases with mostly DCIS in addition to microinvasion did not change the results, and these were therefore included in the analyses.

HRs are presented with 95% confidence intervals (CIs). All statistical tests are two-sided. The statistical calculations were performed with IBM SPSS Statistics, Version 25.0 (IBM Corp., Armonk, NY, USA), and the cumulative incidence curves were drawn using STATA, Version 16.1 (StataCorp LLC, College Station, TX, USA).

The distribution of TILs in relation to patient and tumour characteristics in the study cohort (n = 477) is presented in Table 1. A high proportion of immune cell infiltration was associated with younger age, high histological grade, ER−, PR− and HER2+ status, high Ki67 and medullary histological type. None of the lobular tumours had high infiltration of TILs. The frequency of high TILs in ER+/HER2−, HER2+ and TNBC subgroups was 6% (n = 16/257), 24% (n = 16/66) and 35% (n = 33/95), respectively. The frequency of low TILs was 69% (n = 176/257), 33% (n = 22/66) and 21% (n = 20/95), respectively.

The prognoses in terms of BCFi and OS stratified by the three TIL categories (low, intermediate, high) are displayed in Figs. 3a–d and 4a–d, respectively. The prognostic value of TILs was evaluated for all patients included in the control arm (n = 221) stratified by subtypes (ER+/HER2−, n = 136; HER2+, n = 38; and TNBC, n = 47) (Fig. 1). All patients with high TILs, irrespective of BC subtype, had improved prognosis compared with patients with low TILs (HR_BCFi 0.40; 95% CI 0.22–0.71; P = 0.002, and HR_OS 0.52; 95% CI 0.29–0.95; P = 0.03) (Table 2). This was true also in multivariable analysis adjusting for age, nodal status, tumour size, histological grade, ER, PR, HER2 and LVI (HR_BCFi 0.22; 95% CI 0.11–0.43; P < 0.001 and HR_OS 0.23; 95% CI 0.11–0.48; P < 0.001). The univariable prognostic effect of high vs. low TILs was essentially the same in patients with ER+/HER2− tumours (HR_BCFi 0.40; 95% CI 0.14–1.09; P = 0.07) as well as in HER2+ (HR_BCFi 0.28; 95% CI 0.06–0.97; P = 0.05) and TNBC tumours (HR_BCFi 0.27; 95% CI 0.08–0.88; P = 0.03). The prognostic effect of high TILs was also observed in the multivariable analysis, except for patients with TNBC (Table 2). Presence of LVI was associated with a worse prognosis in patients in the control arm in the univariable analysis (HR_BCFi 1.49; 95% CI 1.08–2.05; P = 0.02), and the results were essentially the same in multivariable analysis (HR_BCFi 1.39; 95% CI 0.99–1.95; P = 0.06) (Table 2).

The predictive value of TILs for TAM treatment was evaluated in the ER+ cohort (n = 321). In our previous follow-up study, TAM prolonged BCFi in the study population with ER+ tumours (n = 362) (HR_BCFi 0.62; 95% CI 0.47–0.82; P = 0.001) [18], and this was also true in the present study cohort (n = 321) (HR_BCFi 0.65; 95% CI 0.49–0.86; P = 0.002). The proportions of tumours with low, intermediate and high TILs were 63% (n = 107/171), 29% (n = 50/171) and 8% (n = 14/171) in the control group and 64% (n = 96/150), 29% (n = 44/150) and 7% (n = 10/150) in the TAM group, respectively. Figure 5a–c illustrates the outcome (BCFi) stratified by TIL categories and treatment allocation. In the univariable analysis, TAM improved the outcome for patients with low (HR_BCFi 0.66; 95% CI 0.46–0.93; P = 0.02) and intermediate TILs (HR_BCFi 0.59; 95% CI 0.35–1.00, P = 0.05). In contrast, the outcome of patients with high TILs was not affected by adjuvant TAM (HR_BCFi 0.89; 95% CI 0.26–3.07; P = 0.86). There was no clear association between OS and TIL categories as illustrated in Fig. 6a–c.

Furthermore, the differential effect of TAM treatment on BCFi in TIL subgroups was analysed in a Cox model including TAM treatment and TILs and an interaction term. The TIL variable was divided into two categories at an exploratory 50% cut-off (non-LPBC as < 50% vs. LPBC as ≥ 50%), based on the above predictive results. Among the ER+ samples, 93% (n = 297/321) were categorised as non-LPBC and 7% (n = 24/321) as LPBC. The effect of TAM was stronger in patients with non-LPBC (HR_BCFi 0.63; 95% CI 0.47–0.84; P = 0.002) than in patients with LPBC (HR_BCFi 0.84; 95% CI 0.24–2.86; P = 0.77), but the evidence for an interaction between TAM treatment and level of TIL infiltration on BCFi was weak (P_interaction = 0.65) (Table 3).