Tumor-Infiltrating Lymphocytes in Patients Receiving Trastuzumab/Pertuzumab-Based Chemotherapy

A TRYPHAENA Substudy

Michail Ignatiadis; Gert Van den Eynden; Salgado Roberto; Marco Fornili; Yacine Bareche; Christine Desmedt; Françoise Rothé; Marion Maetens; David Venet; Esther Holgado; Virginia McNally; Astrid Kiermaier; Heidi M. Savage; Timothy R. Wilson; Javier Cortes; Andreas Schneeweiss; Karen Willard-Gallo; Elia Biganzoli; Christos Sotiriou

Disclosures

J Natl Cancer Inst. 2019;111(1):69-77. 

In This Article

Results

Patient Characteristics

The patient characteristics for the entire TRYPHAENA cohort, the subgroups of patients with evaluable TILs at baseline and at the point of surgery, and available gene expression data at baseline are summarized in Table 1. TIL status was successfully evaluated in 213 (94.7%) of 225 patients at baseline and in 64 (73.6%) of 87 patients with invasive residual disease at the point of surgery. The reasons behind unsuccessful evaluation are presented in the CONSORT diagram for TILs (Supplementary Figure 1, available online). The main reason for missing results at the point of surgery was either because patients had pCR or because the pathologists were not able to identify the tumor bed in available H&E slides (n = 140 patients) (Supplementary Figure 1, available online). Agreement between the two pathologists was good (Supplementary Figure 2, available online). For Log TILs at baseline, the concordance correlation coefficient was 0.62 (95% CI = 0.55 to 0.68), while at surgery it was 0.66 (95% CI = 0.59 to 0.73).

Supplementary Figure 1.

CONSORT diagram for TILs.
5FEC: 5-Fluorouracil, Epirubicin, Cyclophosphamide, H: Trastuzumab, P:Pertuzumab, T: Docetaxel, C: Carboplatin, N: Numberof patients, D Samples: Diagnosis Samples, S Samples: Surgery Samples, TILs: Tumor Infiltrating Lymphocytes, pCR pathological complete response, *Surgery Samples were not evaluable because no tumor bed could be identified or patients had pCRD

Supplementary Figure 2.

Agreement between the two pathologists for logTILs at baseline and at surgery.

The expression of immune-related single genes and gene signatures at baseline in tumor samples was successfully evaluated using NanoString technology in 173 of 225 (76.9%) patients. A lack of sample was the main reason for missing data (n = 41), with additional reasons described in the CONSORT diagram for NanoString gene expression data (Supplementary Figure 3, available online).

Supplementary Figure 3.

CONSORT diagram for NanoString gene expression data.
8FEC: 5-Fluorouracil, Epirubicin, Cyclophosphamide, H: Trastuzumab, P:Pertuzumab, T: Docetaxel, C: Carboplatin, N: Number of patients, D Samples: Diagnosis Samples, pCR pathological complete response

TIL Percentage at Baseline and Baseline Clinicopathological Characteristics

We first tested whether there was any association between TIL percentage and clinicopathological characteristics at baseline. No evidence of association was found between TIL percentage and age, ER status, or type of neoadjuvant chemotherapy (anthracycline-based vs other) (Supplementary Table 1, available online). However, there was an association between an increased TIL percentage at baseline and histological grade III tumors (P = .03).

TIL Percentage at Baseline and pCR

No evidence for a nonlinear relationship between pCR probability on the logit scale and TIL percentage at baseline was found. Figure 1 and Supplementary Figure 4 (available online) depict pCR rates according to TIL percentage at baseline assessed as a categorical variable using quartiles and a scatterplot of pCR rates across several TIL percentage cutoffs, respectively. In a univariate analysis, TIL percentage at baseline was not significantly associated with pCR (odds ratio [OR] = 1.15, 95% CI = 0.99 to 1.32, P = .06) (Table 2). In this study, among other baseline clinicopathological characteristics, only ER status was significantly associated with pCR, with ER-positive tumors having a lower probability for pCR (OR = 0.32, 95% CI = 0.19 to 0.56, P < .001) (Table 2). In a multivariable model including baseline clinicopathological characteristics and percentage TILs, TILs were not significantly associated with pCR (OR for every 10-percentage unit increase in TILs = 1.12, 95% CI = 0.95 to 1.31, P = .17) (Supplementary Table 2, available online). Only ER-positive and higher–clinical stage tumors were independently associated with a lower probability of pCR.

Figure 1.

Pathological complete response rates by tumor-infiltrating lymphocyte (TIL) quartiles at baseline, assessed as a categorical variable. The error bars represent 95% confidence intervals, computed according to the Wilson method. pCR = pathological complete response; TIL = tumor-infiltrating lymphocyte.

Supplementary Figure 4.

Distribution of pCR rate according to several baseline TILs cutoff (10% increment).

TIL Percentage at Baseline and EFS

Next, we investigated whether there was association between TILs at baseline and EFS. After a median follow-up of 4.7 years, for every 10% increase in TILs at baseline, EFS increased, but this change was not statistically significant, with a hazard ratio (HR) of 0.80 (95% CI = 0.63 to 1.02, P = .08) (Table 2). When we examined associations between baseline clinicopathological characteristics and EFS, a higher clinical stage was associated with a shorter EFS (HR = 2.45, 95% CI = 1.11 to 5.42, P = .03), whereas the presence of pCR was associated with longer EFS (HR = 0.35, 95% CI = 0.16 to 0.73, P = .005) (Table 2). In a multivariable model including baseline clinicopathological characteristics, baseline TIL percentage, and pCR, only pCR and baseline TIL percentage were independently associated with EFS (Table 3). In this model, for every 10-percentage unit increase in baseline TILs, there was a 25% reduction in the hazard for the event (adjusted HR = 0.75, 95% CI = 0.56 to 1.00, P = .05). Interestingly, when we examined the EFS of patients according to the presence of pCR (yes vs no) and the level of percentage TIL (high vs low based on the median TIL value of 14.1% [interquartile range = 7.1%–32.4%] in this series), we found that patients with high baseline percentage TIL and pCR had the best prognosis compared with the other groups (Figure 2). No evidence of interaction between pCR and baseline TIL percentage, dichotomized with cutoff of 14%, was observed (P = .10).

Figure 2.

Kaplan-Meier event-free survival curves of patients according to pathological complete response (pCR) status (yes vs no) and tumor-infiltrating lymphocytes (TILs) at baseline (>14% vs ≤14%). Patients were divided into four groups: TILs >14% and pCR, TILs >14% and non-pCR, TILs ≤14% and pCR, TILs ≤14% and non-pCR. pCR = pathological complete response; TIL = tumor-infiltrating lymphocyte.

TIL Percentage at the Point of Surgery

TIL percentage at the point of surgery was evaluated in 64 of 87 patients with invasive residual disease (Supplementary Figure 1, available online). TILs at surgery were not significantly associated with baseline clinicopathological characteristics (Supplementary Table 3, available online), or with EFS (HR = 1.05, 95% CI = 0.73 to 1.50, P = .80) (Table 2), although the small number of patients and events does not allow definitive conclusions to be drawn.

Association Between TIL Percentage at Baseline and Gene Expression (NanoString) in Baseline Tumor Samples

In the 173 patients for whom baseline tumor gene expression data were available, we found that TIL percentage at baseline strongly correlated with most immune genes and gene signatures, but not with ESR1 or ERBB2 (Figure 3). No association was found between TIL percentage and the PAM50 intrinsic subtype (Kruskal-Wallis P = .48) within HER2 breast cancer patients (Supplementary Figure 5, available online). To derive further insights into the tumor biology associated with an immune response as measured by TIL percentage, we evaluated associations between TIL percentage at baseline and more than 800 cancer-related genes (Figure 4A). Interestingly, gene ontology analysis showed that TIL percentage was statistically significantly inversely correlated with genes associated with epithelial mesenchymal transition (EMT), angiogenesis, and T-cell inhibition (Figure 4B), such as SNAIL1, ZEB1, NOTCH3, and B7-H3 (Figure 4C).

Figure 3.

Heat map reflecting the hierarchical clustering of pairwise correlation between baseline tumor-infiltrating lymphocytes and baseline immune-related single genes/gene signatures in 173 patients. Cells are colored according to Spearman's correlation coefficient values, with blue indicating a positive correlation and red indicating a negative correlation.

Figure 4.

A) Heat map reflecting the hierarchical clustering of the statistically significantly correlated genes with baseline tumor-infiltrating lymphocytes (TILs). Two-sided Spearman correlation was used in the analysis, and correction for multitesting (False Discovery Rate [FDR]) was performed. Only genes with |rho| ≥.30 and FDR <0.05 were considered statistically significant. B) Gene ontology analysis of the statistically significantly inversely correlated and positively correlated genes with baseline TILs. C)Correlation plot for the epithelial-to-mesenchymal transition (EMT) markers with baseline TIL levels. Two-sided Spearman correlation was used in the analysis. Only genes with P <.05 were considered statistically significant. pCR = pathological complete response; TIL = tumor-infiltrating lymphocyte.

Supplementary Figure 5.

Distribution of baseline TILs according to PAM50 subtypes within HER2-positive breast cancer. Association between TILs and PAM50 subtype was assessed using the two-sided Kruskal-Wallis test.

Gene Expression (NanoString) in Baseline Tumor Samples and pCR

Next, we investigated the association between pCR and the baseline expression of the following: single genes (ESR1, ERBB2), the PAM50 subtype (HER2-enriched subtype vs others), three immune-related gene signatures, and single immune-related genes (GZMA and GZMB, PRF1, FOXP3, CTLA4, PDL1, PD1, IL4, IFNG). After correction for multiple testing, increased pCR rates were observed in patients with a HER2-enriched subtype, higher expression of ERBB2, the immune_TLS and immune_trastuzumab gene signatures, and GZMA and GZMB, PRF1, PD1 and IFNG in a univariate analysis (Supplementary Figure 6, available online). Following multivariable analysis (after adjustment for baseline clinicopathological characteristics and baseline TILs) and correction for multiple testing, increased pCR rates were observed in patients with a higher expression of ERBB2, immune_TLS and immune_trastuzumab gene signatures, and GZMA and GZMB, PRF1 and IFNG(Figure 5). When we examined the distribution of pCR rates across PAM50 subtypes (Supplementary Figure 7, available online), we observed the highest response (57 out of 82 patients or 69.5%) for the HER2-enriched subtype (P = .02).

Figure 5.

Adjusted odds ratios (ORs) for pathological complete response for a unit increase in baseline single gene/gene signature score using logistic regression (adjusted for baseline clinicopathological characteristics including age, histology grade, estrogen receptor, clinical stage, chemotherapy, and baseline tumor-infiltrating lymphocytes). Horizontal bars represent the 95% confidence intervals of the ORs. The Wald test was used to obtain P values and correction for multiple testing using the Benjamini-Hochberg procedure. Characteristics with statistically significant (FDR < 0.05) and non–statistically significant effects are shown in black and gray, respectively. CI = confidence interval; FDR = false discovery rate; OR = odds ratio.

Supplementary Figure 6.

Unadjusted odds ratios (ORs) for pCR for a unit increase in baseline single genes/gene signatures score and for change in baseline PAM50 subtype (HER2-enriched vs other) using logistic regression. Horizontal bars represent the 95% confidence intervals of ORs. Characteristics with significant effect (p<.05) after correction for multiple testing using the Benjamini-Hochberg procedure are shown in blue, n=number of patients.

Supplementary Figure 7.

A. Distribution of the baseline PAM50 intrinsic subtypes across the 173 patients. B. PCR rate distribution across the PAM50 intrinsic subtype. Differences between Her2-enriched subtype and the remaining intrinsic subtypes taken together was assessed using a chi-square test (p<.05).

Gene Expression (NanoString) in Baseline Tumor Samples and EFS

There was no association between EFS and the expression of any of the immune genes/gene signatures tested (Supplementary Figure 8, available online). In a multivariable analysis including baseline clinicopathological characteristics, PAM50 subtype, pCR, and baseline TIL percentage, pCR remained statistically significantly associated with EFS (P = .02) but not with TIL percentage (P = .06) (Supplementary Table 4, available online).

Supplementary Figure 8.

Unadjusted hazard ratios (HRs) for event-free survival (EFS) for a unit increase in baseline single genes/gene signatures score and for change in baseline PAM50 subtype (HER2-enriched vs other) using Cox regression. Horizontal bars represent the 95% confidence intervals of HRs. Correction for multiple testing was performed using the Benjamini-Hochberg procedure.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....