close

Вход

Забыли?

вход по аккаунту

?

annonc%2Fmdx691

код для вставкиСкачать
Title page
A gene signature to predict high tumour-infiltrating lymphocytes after neoadjuvant
chemotherapy and outcome in patients with triple negative breast cancer
C. Criscitiello1‡, M.A. Bayar2,3‡, G. Curigliano1, F.W. Symmans4, C. Desmedt5, H.
Bonnefoi6, B. Sinn4, G. Pruneri7, C. Vicier8,9, J.Y. Pierga10, C. Denkert11, S. Loibl11, C.
Sotiriou5, S. Michiels2,3*, F. André8,9*
1
Division of Experimental Therapeutics/European Institute of Oncology, Milano, Italy
2
Biostatistics and Epidemiology Department/Gustave Roussy, Villejuif, France
3
CESP, Faculty of Medicine, University of Paris Sud, Faculty of Medicine, INSERM,
University Paris Saclay, Villejuif, France
4
Department of Pathology, University of Texas MD Anderson Cancer Center, Houston, TX
5
Translational Breast Cancer Laboratory, Free University of Brussels, Institute Jules
Bordet, Brussels, Belgium
6
Departement of Medical Oncology, Institute Bergonié, Bordeaux, France
7
Biobank for Translational Medicine (B4MED) European Institute of Oncology, University
of Milan, Italy
8
Department of Medical Oncology, Gustave Roussy, Villejuif, France
9
INSERM U981, University Paris Sud, France
10
Institut Curie, Department of Medical Oncology, Paris Cedex 05, France
11
GBG German Breast Group, Neu Isenburg, Germany
‡These authors contributed equally to this work as first authors
*These authors contributed equally to this work as senior authors.
© The Author 2017. Published by Oxford University Press on behalf of the European Society for Medical
Oncology. All rights reserved. For permissions, please email: journals.permissions@oup.com.
Corresponding author:
Prof. Fabrice André
Department of Medical Oncology, Gustave Roussy, Villejuif, France
INSERM U981, Université Paris Sud, Villejuif, France
Email: Fabrice.ANDRE@gustaveroussy.fr
Phone: +33 1 42 11 42 93
2
Abstract
Introduction: In patients with triple-negative breast cancer (TNBC), the extent of tumorinfiltrating lymphocytes (TILs) in the residual disease (RD) after neoadjuvant
chemotherapy (NACT) is associated with better prognosis. Our objective was to develop a
gene signature from pre-treatment samples to predict the extent of TILs after NACT, and
then to test its prognostic value on survival.
Patients and Methods: Using 99 pre-treatment samples, we generated a four-gene
signature associated with high post-NACT TILs. Prognostic value of the signature on
distant relapse-free survival (DRFS) was first assessed on the training set (n=99) and then
on an independent validation set (n=115).
Results: A four-gene signature combining the expression levels of HLF, CXCL13,
SULT1E1, and GBP1 was developed in baseline samples to predict the extent of
lymphocytic infiltration after NACT. In a multivariate analysis performed on the training set,
this signature was associated with DRFS (HR: 0.28 for a one-unit increase in the value of
the four-gene signature, 95%CI: 0.13-0.63). In a multivariate analysis performed on an
independent validation set, the four-gene signature was significantly associated with DRFS
(HR: 0.17, 95%CI: 0.06-0.43). The four-gene signature added significant prognostic
information when compared to the clinicopathologic pre-treatment model (likelihood ratio
test in the training set p=0.004 and in the validation set p=0.002).
Conclusion: A four-gene signature predicts high levels of TILs after anthracyclinecontaining NACT and outcome in patients with TNBC and adds prognostic information to a
clinicopathological model at diagnosis.
Key words: breast cancer, gene signature, neoadjuvant therapy, tumor-infiltrating
lymphocytes, residual disease.
Key message: Molecular tests predicting outcome in patients with TNBC who receive
NACT are needed. We developed a 4-gene signature to predict at diagnosis which
cancers will be infiltrated by lymphocytes after chemotherapy. In multivariate analysis, this
3
signature was associated with outcome in training and validation sets, thus identifying
patients with poor outcome who require new drugs.
Introduction
Triple-negative breast cancer (TNBC) is characterized by the absence of expression of the
estrogen receptor (ER), progesterone receptor (PgR) and human epidermal growth factor
receptor 2 (HER2). While the TNBC subtype is traditionally perceived as an aggressive
disease, recent prospective trials suggest that not all patients immediately relapse after
(neo)adjuvant chemotherapy. As illustration, in the BEATRICE trial, the three-year invasive
disease free interval (IDFS) was 83% (95% CI 80.5-85.0) [1]. Hence, one of the major
stakes in TNBC will be the identification of those patients who have excellent survival
outcome when treated with conventional therapies as opposed to those who require new
drugs. Several efforts are ongoing to identify prognostic parameters of survival in TNBC
patients. Pathological complete response (pCR) after neoadjuvant chemotherapy (NACT)
is strongly correlated with more favorable outcome [2, 3]. Absence of pCR has been
recently used as an inclusion criterion to enrich randomized trials with patients at high risk
of relapse, therefore requiring new therapies. Tumor-infiltrating lymphocytes (TILs) are
mononuclear immune cells that infiltrate tumor tissue and likely reflect an individual
immunological response. In early breast cancer (BC), the presence of TILs and high
immune infiltration defined by gene expression related signatures are associated with
more favorable outcome [4, 5]. We reported that the presence of >60% TILs in residual
disease (RD) after NACT is associated with better prognosis in patients with TNBC [6]. In
this study, 10% of patients had TILs >60% in their RD after NACT, and the 5-year OS rate
was 91% in the high-TILs versus 55% in the low-TILs groups, respectively. This
observation was recently confirmed in an independent study [7]. Therefore, the extent of
post-NACT TILs is also associated with good outcome. Based on these data, we
4
hypothesized that developing, from pre-NACT tumor samples, a molecular test able to
predict the quantity of TILs within residual TNBC after NACT would be of prognostic value.
The present study aimed to develop a genomic predictor of chemotherapy-mediated
increase in TILs in patients with TNBC. As at baseline it is not known which patients will
achieve a pCR or not, we also tested its prognostic value in an unselected population
including both patients with pCR and RD. Such a tool could help identifying - at diagnosis patients at high risk of recurrence, who might require novel immuno-stimulatory treatments
within clinical trials.
Patients and Methods
Patients
In the training set, patients were selected if they had an ER-/HER2- BC, treated with NACT
(anthracycline-containing), RD after treatment, a gene expression array performed on pretreatment samples and hematoxylin and eosin (H&E)-stained tumor sections available to
assess post-treatment TILs (See supplementary Figure A1). Two patient series were
included in the training set: 30 patients from Trial of Principle (TOP) study and 69 patients
from MDACC neoadjuvant series [8, 9] (See Supplementary Table A1 and A5). In the
validation set, patients were selected if they had an ER-/HER2- BC treated with NACT
(anthracycline-containing), RD after treatment and a gene expression array performed on
pre-treament samples. Five cohorts of patients were included in the validation set: 24
patients from I-SPY-1, 16 patients from LBJ/INEN/GEICAM, 20 patients from
MAQCII/MDACC (those who were not included in the training set), 41 patients from TOP
trial (those who were not included in the training set) and 14 patients from USO-02103 [912] (See Supplementary Tables A2 and A5 and Figure A2). The TOP trial and the MDACC
series were split between the training and validation set based on the availability of H&E5
slides. A larger validation set from the same five cohorts was considered for exploratory
analyses and included patients with either pCR or RD. Detailed methods of TILs
assessment and gene expression computation are provided in supplementary material
(See Supplementary Figure A4 and Tables A3-A4).
Tumors were identified as ER-/HER2- based on ER assessment by IHC and
HER2 assessment by IHC and fluorescent in situ hybridization, as originally reported [812]. When unavailable, ER and HER2 status were assigned according to ESR1 and
ERBB2 gene expression [5]. In the validation set 11 tumours were PgR+ by IHC. These
tumours were included in our analysis, because considered triple-negative by gene
expression in the original manuscripts [8-12].
TILs were quantified on RD after NACT in H&E slides from surgical samples from MDACC
neoadjuvant series and TOP trial (training set) [8, 9], closely adhering to the criteria
proposed by the TILs WG [13], although they were developed for scoring pre-treatment
samples. Briefly, all mononuclear cells (i.e., lymphocytes and plasma cells) in the stromal
compartment within the borders of the invasive tumor were evaluated and reported as a
percentage (TILs score). TILs outside of the tumor border, around DCIS and normal breast
tissue, as well as in areas of necrosis, if any, were not included in the scoring. TILs have
been evaluated for the stromal compartment only (=% stromal TILs), as per the TILs WG
recommendations [13]. TILs were assessed as a continuous measure (score). For each
surgical specimen, all the slides containing invasive RD have been evaluated. The
reproducibility of this method has been described [5]. H&E slides from TOP samples have
been sent to IEO, where they have been independently read for TIL-infiltration by two
investigators (CC and GP). MDACC H&E slides have been read on-site by two
investigators (CC and BS). In case of disagreement between investigators, the score was
assigned by consensus.
6
Methods
Spearman correlation was used to quantify the correlation between pre- and posttreatment TILs and a Wilcoxon signed-rank test was performed to assess whether there
was an increase in the average post-treatment TILs as compared to baseline samples. To
assess the prognostic value of post-NACT TILs on survival, we used a Cox model
adjusted for age (continuous), cT (T0-1-2 vs. T3-4), cN (N0 vs. N+) and grade (1-2 vs. 3),
and stratified by patient series (TOP vs. MDACC).
The first step in constructing the genomic signature was gene selection. As the most
appropriate cutoff is controversial, we fitted a general linear model of the continuous level
of TILs after Box-Cox transformation as a function of gene expression, while controlling for
the effect of potential confounders, i.e. series (TOP vs. MDACC), age (continuous), cT
(T0-1-2 vs. T3-4), cN (N0 vs. N+) and grade (1-2 vs. 3). To identify a parsimonious set of
genes, we used the lasso penalization for variable selection with the optimal value of the
tuning parameter obtained from a ten-fold cross validation process [14, 15] (See
Supplementary Figure A5). The clinicopathologic covariates included in the model were
not penalized.
We estimated pairwise correlation values between the genes included in the 4-gene
signature using the Spearman correlation; 95% confidence intervals (CI) were obtained
using 1 000 bootstrap repetitions.
The four-gene signature was defined as the linear combination of the gene expressions
weighted by the regression coefficients in the general linear model; to facilitate
interpretation of the values of the four-gene signature thus obtained, the signature was
scaled within the training set, so that the 2.5% and 97.5% quantiles equaled respectively 0
and +1. To assess the accuracy of the model to predict the observed value of TILs, we
7
computed the root mean squared prediction error (RMSE) using 1000 repetitions of a tenfold crossvalidation.
We computed the four-gene signature on the validation set using the same probes and the
same coefficients as for the training set. We scaled the four-gene signature in the
validation set using the same values as used in the training set. Distant relapse-free
survival (DRFS) was defined as time from initial diagnosis to distant relapse or death.
Patients who did not experience the event were censored at the date of last follow-up.
The median follow-up was computed using inverse Kaplan-Meier method applied on
DRFS [16]. To assess the prognostic value of the signature on DRFS in the training set
and then in validation set, we computed hazard ratios (HRs) and 95% CIs using a Cox
model stratified on series and adjusted on clinicopathologic variables. As defining if a
patient achieved pCR or not requires a six-month observation period, for patients in the
validation set with RD, landmarking was used with a landmark time at six months [17]. This
technique was not used for the entire validation set. We built two risk groups (four-gene
signature low vs. four-gene signature high) using the median value of the four-gene
signature in the training set as a cutoff. The cutoff was then frozen for the entire study and
used to build risk groups in the validation set.
We used Uno’s concordance indice (C-index) to quantify the capacity of the prediction
models to discriminate subjects with different event times [18]. We considered two
truncation times: a horizon at 3 years and 5 years. The resulting Cs tell how well the given
prediction models work in predicting events that occur in the time window from zero to
three years and zero to five years, respectively. 95% CIs were obtained using 1000
bootstrap repetitions. We used the likelihood ratio test to assess the added value of the 4four-gene signature to the clinical model including the clinicopathologic covariates
determined at diagnosis in the training set and then in the validation set.
8
We explored the association between the four-gene signature and the probability to
achieve pCR in the entire validation set, we computed odds ratios (ORs) and 95% CI using
a conditional logistic model stratified on series and adjusted on clinicopathologic variables.
For all statistical models, we used restricted cubic splines with two degrees of freedom to
investigate non-linear associations.
Secondary analysis consists in modeling TILs from the training set using univariate
selection (one gene at a time) adjusted on clinicopathological variables with correction for
multiple comparisons. For detailed statistical methods see supplementary Table A19 [19].
Results
Study population
Overall, 99 patients with ER-/HER2- BC were selected to generate the signature. Although
5 of these patients did not have survival data, they were included because H&E sections to
determine post-treatment TILs and baseline gene expression data were available. Hence,
they could be entered in our primary model to determine a gene expression score
associated with post-treatment TILs. 115 patients with ER-/HER2- BC and RD were
selected to validate the prognostic value of the signature on DRFS. The entire validation
set included 185 patients whether they achieved pCR or not. Patients’ characteristics in
the training set and validation set are given in Table 1. Flowcharts for the training set and
validation set are described in supplementary material (see supplementary Figures A1A2).
Tumor-infiltrating lymphocytes (TILs)
The median value of stromal-TILs after NACT was 15% (range 0-95%) in the training set.
Detailed summary statistics of TILs after NACT and description of the Box cox
9
transformation are provided in supplementary material (See Supplementary Table A4).
Pre- and post-NACT TILs were both available for 29 patients from the training set. There
was no statistically significant correlation between pre- and post-treatment TILs
(Spearman correlation coefficient=0.17, p=0.384). There was a significant absolut increase
in post-treatment TILs as compared to baseline samples (at baseline, mean (SD) = 14.14
(20.765), post-treatment, mean (SD) = 29.56 (26.894), absolut increase 18.28, [95%CI
6.21 - 30.34], Wilcoxon signed-rank test p- value = 0.002) (See supplementary Figure A3).
Post-chemotherapy TILs were associated with improved DRFS in a multivariate analysis
including age, cT, cN and grade, and stratified by series (hazard ratio, HR for a 10%
increase in TILs: 0.83, 95% CI: 0.68-0.99, p=0.043). A similar trend was observed for OS
(HR for a 10% increase in TILs: 0.83, 95% CI: 0.69-1.01, p=0.063).
Development of a four-gene signature on pretreatment samples to predict postNACT TILs
The LASSO penalised regression model of the level of TILs ater NACT led to a
parsimonious set of 4 probes of 4 different genes (HLF, CXCL13, SULTE1, GBP1) that are
provided with their corresponding coefficients in Table 2. Detailed information about model
selection and pairwise correlation values among the 4 probes are provided in
supplementary material (See Supplementary Tables A6-A7-A16-A17-A20-A21-A22-A23
and Figure A6-A8). The four-gene signature was strongly associated with TILs; the
corrected RMSE using cross-validation was 2.21 (95% CI; 2.15 – 2.28; for more details
see Supplementary Table A9-A10 and Figure A7).
Prognostic value of the four-gene signature in the training set
The prognostic value of the four-gene signature was assessed in 94 patients from the
training set, for whom survival data were available. All patients had RD after NACT.
Median (Q1 – Q3) follow-up was 7.6 years (3.7 – 8.8). In a multivariate analysis (Table 3),
10
the four-gene signature was significantly associated with better DRFS (HR for a one-unit
increase in the value of the 4-gene signature: 0.28, 95%CI: 0.13-0.63, p=0.002). KaplanMeier DRFS curves of the risk groups (low four-gene signature vs. high four-gene
signature) constructed using the median value of the four-gene signature (median=0.51)
are shown in Figure 1a. There was no evidence of a non-linear association between the
four-gene signature and DRFS. The four-gene signature added significant prognostic
information to the clinicopathological characteristics at diagnosis, as shown by the
likelihood ratio test (p=0.004). The discrimination was also improved; at five years, the Cindex increased from 0.617 to 0.673 (Table 4). Similar results were obtained for OS (HR
for a one-unit increase in the value of the four-gene signature: 0.35, 95%CI: 0.16-0.75,
p=0.007; likelihood ratio test, p=0.012; the C-index increased from 0.631 to 0.668).
Prognostic value of the four-gene signature in the validation set
In the validation set, 68 (37%) patients achieved pCR and 115 (63%) had RD (2 missing
information). We first evaluated the prognostic value of the four-gene signature in patients
with RD (n=115), as in the training set (See Supplementary Tables A7-A8-A9). In a
multivariate analysis (Table 3), the four-gene signature was significantly associated with
DRFS (HR for a one-unit increase: 0.17, 95%CI: 0.06-0.43, p=<0.001). Kaplan-Meier
DRFS curves of the risk groups constructed using the same cutoff (0.51) as in the training
set are shown in Figure 1b. No significant non-linear association between four-gene
signature and DRFS was brought out. When adding the four-gene signature to the
clinicopathologic characteristics, there was an added prognostic value, as shown by the
likelihood ratio test (p=0.002). Discrimination ability was also improved, at five years, the
C-index increased from 0.712 to 0.749 in the validation set for patients with RD.
As part of the exploratory analyses, we then assessed the prognostic value in the overall
population irrespective of the pathological response to NACT (pCR vs. RD). In a
11
multivariate analysis (n=185), the four-gene signature was significantly associated with
improved DRFS (HR: 0.29, 95%CI: 0.13-0.67, p=0.004, Table 3), after controlling for
clinicopathologic covariates and response to NACT. Kaplan-Meier DRFS curves of the risk
groups constructed using the same cutoff (0.51) as in the training set are shown in Figure
1c. No significant non-linear association between the four-gene signature and DRFS was
brought out. When adding the four-gene signature to the clinicopathologic characteristics,
there was an added prognostic value, as shown by the likelihood ratio test (p=0.008).
Discrimination ability was also improved, at five years, the C-index increased from 0.686 to
0.700 in the validation set for all patients (pCR and RD).
Results of the conditional logistic model showed no statistically significant association
between the four-gene signature and the probability to achieve pCR in the validation set
(OR for a one-unit increase in the four-gene signature: 0.96, 95% CI: 0.30-3.08, p=0.947).
No significant non-linear association between the four-gene signature and the probability
to achieve pCR was brought out (See Supplementary Table A18).
Discussion
There is a large body of evidence supporting the immunogenic role of chemotherapy and
the association between immune system activation and good outcome [20]. However, the
molecular predictors of this effect are still to be found. Preliminary data have suggested
that autophagic markers and genetic polymorphisms on Toll-like receptor 4 (TLR4), P2X
purinoceptor 7 (P2RX7), fusion regulatory protein-1 (FRP1) may define a subset in which
chemotherapy could generate an adaptive immune response [21]. In this study, we report
the development of a four-gene signature that predicts high levels of TILs after
anthracycline-containing chemotherapy. Guanylate-binding protein 1 (GBP1) and
chemokine (C-X-C motif) ligand 13 (CXCL13) are two proteins involved in anti-tumor
immune response. GBP1 is an interferon-regulated protein involved in cancer cell
12
apoptosis following inflammatory response. Interferon-response has previously been
proven to be a major component of chemotherapy-induced adaptive immune response.
CXCL13 is a chemokine involved in B Cell attraction. CXCL13 expression is regulated
through Th17 response, a mechanism of immunogenic cell death [22]. The link between
immune system and Sulfotransferase Family 1E Member 1 (SULT1E1)/hepatic leukemia
factor (HLF) is less obvious, and will deserve further preclinical explorations. SULT1E1 is a
gene involved in the transformation from E1 to E2, and its suppression could create a
more immunogenic microenvironment [23]. HLF is a transcription factor that could be
involved in treatment induced immunogenic cell death [24].
Pre-treatment TILs have been associated with very good outcome in patients with TNBC.
The four-gene signature developed in the training set may predict the extent of posttreatment TILs and prognosis. In the validation set, only the prognostic value of the fourgene signature was confirmed. The predictive value of the signature on post-treatment
TILs has not been independently validated yet.
To what extent the four-gene signature adds information to TILs is a matter of discussion.
Unfortunately, we could not assess pre-treatment TILs in the vast majority of patients
included in the present study. Nevertheless, there are three arguments suggesting that the
four-gene signature could be complementary and/or add information to pre-treatment TILs.
First, the integrated prognostic model of clinicopathological factors plus the four-gene
signature will allow to detect more patients with low risk of relapse as compared to the use
of pre-treatment TILs and its conventional cutoff (>50% TILs), which only identifies 10% of
TNBC as TILs-positive [6]. Second, the vast majority of patients with high TILs after NACT
do not present TILs in baseline samples. As illustration, in the study reported by Dieci et al,
13 out of 18 patients with post-NACT TILs had <60% TILs in baseline samples [6]. In the
present study, there was no strong correlation between pre- and post-treatment TILs
(p=0.384) in the limited set of samples for which both TILs values were obtaind, and post13
treatment TILs were increased as compared to baseline samples (Wilcoxon signed-rank
test p- value = 0.002). Third, we believe that pre-treatment TILs and the four-gene
signature reflect two separate mechanisms of antitumor activity. Pre-treatment TILs have
been associated with pCR in several independent studies, while the four-gene signature
has been developed specifically in patients with RD [25, 26]. Altogether, these
considerations suggest that pre-treatment TILs and our four-gene signature will be
complementary in assessing outcome in patients with TNBC.
Furthermore, we compared the added value of the four-gene signature to 2 immunerelated signatures (immune 1 and immune 2, one at a time) in both training and validation
sets [5] (See Supplementary Tables A9-A10-A11-A12). The four-gene signature was not
strongly correlated to any of these two signatures in any dataset. Our signature adds
statistically significant independent prognostic information to a model including
clinicopathological variables and either of these two immune-signatures already known to
predict outcome in TNBC [5] (See Supplementary Tables A13-A14-A15). Hence, this
significantly improves the strength of our finding, partially circumventing the lack of
information about pre-treatment TILs (see Appendix).
There are several ways to further position this signature in daily practice. First, the fourgene signature could be used at diagnosis in an intergrated clinicopathological model to
predict patients at high risk of relapse when treated with neoadjuvant anthracyclinecontaining chemotherapy, who may therefore require additional therapies. In the validation
set, the signature was associated with an increased likelihood of metastatic relapse
(HR=1.94, 95%CI: 1.05-3.60). Second, since the signature is not associated with pCR, it
lends itself naturally to a model including a molecular predictor of pCR. Increased pCR
rate has not been shown to correlate with prolonged event free survival. Our findings
contribute to explain this lack of correlation. Indeed, we have shown that - even in patients
with RD - a treatment induced TILs upregulation may correlate with clinical benefit that
14
cannot be captured by the simple pCR rate. This may be another way to look at treatment
efficacy, even in the absence of significant change in pCR rate. As mentioned earlier on, a
score combining pre-treatment TILs and the four-gene signature could be a good
candidate to optimally predict relapse by combining a predictor of pCR and a predictor of
high TILs following NACT.
If we can define a subset of patients who will not show chemo-immunisation, then the
major issue is how to improve their outcome. One possible strategy could consist in
targeting the inhibitors of chemo-immunisation, like MEK [7]. HLF and SULT1E1, two
genes included in our signature, are targetable and further preclinical studies will
determine whether their modulation could enhance the chemotherapy induced immune
response .
In conclusion, the four-gene signature identifies, based on baseline samples, those
patients who will present high TILs after NACT and is associated with good outcome.
Further steps will consist of integrating the signature with clinical parameters and pretreatment TILs, as well as the evaluation of targetable proteins (HLF/SULT1E1) in
preclinical studies.
Funding: The study was funded by Transcan-2011, Operation Parrain Chercheurs,
Odyssea, and Fondation Dassault.
Authors’ disclosure of potential conflicts of interests
CC, MAB, GC, SM, and FA are inventors of a patent describing the prognostic value of the
four-gene signature.
15
References
1.
Cameron D, Brown J, Dent R et al. Adjuvant bevacizumab-containing therapy in triple-negative
breast cancer (BEATRICE): primary results of a randomised, phase 3 trial. Lancet Oncol 2013; 14: 933-942.
2.
Liedtke C, Mazouni C, Hess KR et al. Response to neoadjuvant therapy and long-term survival in
patients with triple-negative breast cancer. J Clin Oncol 2008; 26: 1275-1281.
3.
Cortazar P, Zhang L, Untch M et al. Pathological complete response and long-term clinical benefit in
breast cancer: the CTNeoBC pooled analysis. Lancet 2014; 384: 164-172.
4.
Savas P, Salgado R, Denkert C et al. Clinical relevance of host immunity in breast cancer: from TILs
to the clinic. Nat Rev Clin Oncol 2016; 13: 228-241.
5.
Ignatiadis M, Singhal SK, Desmedt C et al. Gene modules and response to neoadjuvant
chemotherapy in breast cancer subtypes: a pooled analysis. J Clin Oncol 2012; 30: 1996-2004.
6.
Dieci MV, Criscitiello C, Goubar A et al. Prognostic value of tumor-infiltrating lymphocytes on
residual disease after primary chemotherapy for triple-negative breast cancer: a retrospective multicenter
study. Ann Oncol 2014; 25: 611-618.
7.
Loi S, Dushyanthen S, Beavis PA et al. RAS/MAPK Activation Is Associated with Reduced TumorInfiltrating Lymphocytes in Triple-Negative Breast Cancer: Therapeutic Cooperation Between MEK and PD1/PD-L1 Immune Checkpoint Inhibitors. Clin Cancer Res 2016; 22: 1499-1509.
8.
Hatzis C, Pusztai L, Valero V et al. A genomic predictor of response and survival following taxaneanthracycline chemotherapy for invasive breast cancer. JAMA 2011; 305: 1873-1881.
9.
Desmedt C, Di Leo A, de Azambuja E et al. Multifactorial approach to predicting resistance to
anthracyclines. J Clin Oncol 2011; 29: 1578-1586.
10.
Esserman LJ, Berry DA, DeMichele A et al. Pathologic complete response predicts recurrence-free
survival more effectively by cancer subset: results from the I-SPY 1 TRIAL--CALGB 150007/150012, ACRIN
6657. J Clin Oncol 2012; 30: 3242-3249.
11.
Shi L, Campbell G, Jones WD et al. The MicroArray Quality Control (MAQC)-II study of common
practices for the development and validation of microarray-based predictive models. Nat Biotechnol 2010;
28: 827-838.
12.
Iwamoto T, Bianchini G, Booser D et al. Gene pathways associated with prognosis and
chemotherapy sensitivity in molecular subtypes of breast cancer. J Natl Cancer Inst 2011; 103: 264-272.
13.
Salgado R, Denkert C, Demaria S et al. The evaluation of tumor-infiltrating lymphocytes (TILs) in
breast cancer: recommendations by an International TILs Working Group 2014. Ann Oncol 2015; 26: 259271.
14.
Tibshirani R. Regression shrinkage and selection via the Lasso. Journal of the Royal Statistical
Society Series B-Methodological 1996; 58: 267-288.
15.
Tibshirani R. The lasso method for variable selection in the Cox model. Stat Med 1997; 16: 385-395.
16.
Schemper M, Smith TL. A note on quantifying follow-up in studies of failure time. Control Clin Trials
1996; 17: 343-346.
17.
Anderson JR, Cain KC, Gelber RD. Analysis of survival by tumor response. J Clin Oncol 1983; 1: 710719.
18.
Uno H, Cai T, Pencina MJ et al. On the C-statistics for evaluating overall adequacy of risk prediction
procedures with censored survival data. Stat Med 2011; 30: 1105-1117.
19.
Benjamini Y, Hochberg Y. Controlling the False Discovery Rate - a Practical and Powerful Approach
to Multiple Testing. Journal of the Royal Statistical Society Series B-Methodological 1995; 57: 289-300.
20.
Kroemer G, Senovilla L, Galluzzi L et al. Natural and therapy-induced immunosurveillance in breast
cancer. Nat Med 2015; 21: 1128-1138.
16
21.
Galluzzi L, Buque A, Kepp O et al. Immunogenic cell death in cancer and infectious disease. Nat Rev
Immunol 2016.
22.
Andre F, Dieci MV, Dubsky P et al. Molecular pathways: involvement of immune pathways in the
therapeutic response and outcome in breast cancer. Clin Cancer Res 2013; 19: 28-33.
23.
Falany CN, Krasnykh V, Falany JL. Bacterial expression and characterization of a cDNA for human
liver estrogen sulfotransferase. J Steroid Biochem Mol Biol 1995; 52: 529-539.
24.
Waters KM, Sontag RL, Weber TJ. Hepatic leukemia factor promotes resistance to cell death:
implications for therapeutics and chronotherapy. Toxicol Appl Pharmacol 2013; 268: 141-148.
25.
Denkert C, Loibl S, Noske A et al. Tumor-associated lymphocytes as an independent predictor of
response to neoadjuvant chemotherapy in breast cancer. J Clin Oncol 2010; 28: 105-113.
26.
Denkert C, von Minckwitz G, Brase JC et al. Tumor-infiltrating lymphocytes and response to
neoadjuvant chemotherapy with or without carboplatin in human epidermal growth factor receptor 2positive and triple-negative primary breast cancers. J Clin Oncol 2015; 33: 983-991.
17
Figures
A
B
C
D
Figure 1: Kaplan-Meier survival curves for (A) distant relapse-free survival in the training set, (B)distant relapse-free survival in the patients from the validation
set who did not achieve pCR (primary analysis) (C) distant relapse-free survival in the entire validation (exploratory analysis) set for patients who achieved pCR or
not, and (D) overall survival in the training set
1
Risk groups for the training set and the validation set as well are constructed using the training set median value of the four-gene signature (median=0·51). HR, Hazard ratio.
CI, confidence interval. p, p-value.
2
Tables
Training set
Age, years
Mean (SD)
Median (Q1 – Q3)
Min – Max
cT
T1
T2
T3
T4
cN
N0
N+
ER status
Negative
Positive
PR status
Negative
Positive
Missing
Histologic grade
1-2
3
Unknown
Missing
Post-chemo Stromal TILs
Mean (SD)
Median (Q1 – Q3)
Min – Max
Response
pCR
RD
Missing
No. of distant relapses
included)
(deaths
n = 99
Validation set
Residual disease
n = 115
Validation set
Entire set
n = 185
49 (11.3)
47 (40 – 59)
27 – 75
48 (10.5)
48 (40 – 57)
26 – 72
48 (10.1)
48 (40 – 57)
26 – 75
5 (5%)
60 (61%)
19 (19%)
15 (15%)
13 (11%)
50 (43%)
30 (26%)
22 (19%)
18 (10%)
90 (49%)
46 (25%)
31 (17%)
35 (35%)
64 (65%)
32 (28%)
83 (72%)
51 (28%)
134 (72%)
99 (100%)
0 (0%)
115 (100%)
0 (0%)
185 (100%)
0 (0%)
87 (100%)
0 (0%)
12
61 (85%)
11 (15%)
43
117 (87%)
17 (13%)
51
17 (17%)
81 (83%)
0 (0%)
1
20 (19%)
78 (73%)
9 (8%)
8
24 (14%)
138 (80%)
10 (6%)
13
21 (21.4)
10 (5 – 30)
0 – 90
N/A
N/A
N/A
N/A
N/A
N/A
0 (0%)
99 (100%)
0
45 (46%)
0 (0%)
115 (100%)
0
49 (43%)
68 (37%)
115 (63%)
2
57 (31%)
1
No. of deaths
Median follow-up in years (Q1 – Q3)
GEO
References
43 (44%)
7.59 (3.74 – 8.82)
GSE16446
GSE25066
GSE20271
7
Hatzis et al
8
Desmedt et al
N/A
3.30 (2.26 – 4.16)
GSE25066
GSE20194
GSE16446
GSE23988
7
Hatzis et al
8
Desmedt et al
10
Shi et al
Iwamoto et al11
N/A
3.24 (2.26 – 4.46)
GSE25066
GSE20194
GSE16446
GSE23988
7
Hatzis et al
8
Desmedt et al
10
Shi et al
Iwamoto et al11
Data are mean (SD), median (Q1 – Q3), min – max, or n (%). Patients of the training set were from
7, 8
MDACC neoadjuvant series and TOP study. SD, standard deviation; Q1, 25th percentile; Q3, 75th
percentile; Min, Minimum; Max, Maximum; cT, clinical tumor size; cN, clinical nodal status; ER, estrogen
receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2; TILs, tumorinfiltrating lymphocytes; GEO, gene expression omnibus; N/A, not available.
Table 1: Patients’ characteristics in the training and validation set
2
PROBEID
202269_x_at
204753_s_at
205242_at
219934_s_at
Gene
GBP1
HLF
CXCL13
SULT1E1
Description
guanylate binding protein 1, interferon-inducible
hepatic leukemia factor
chemokine (C-X-C motif) ligand 13
sulfotransferase family 1E, estrogen-preferring, member 1
Coefficient
0.288
-1.027
0.392
-1.726
The four-gene signature is the linear combination of the gene expressions weighted by the regression
coefficients; to facilitate the interpretation of the values of the four-gene signature thus obtained, the
signature was scaled within the training set, so that the 2.5% and 97.5% quantiles equaled 0 and +1. A
positive coefficient indicates that an increasing gene expression is associated with an increased quantity
of TILs. A negative coefficient indicates that an increasing gene expression is associated with a
decreased quantity of TILs.
Table 2: A four-gene signature to predict post-chemotherapy tumor-infiltrating lymphocytes
3
Age
cT
T0-1-2
T3-4
Training set
Validation set - Residual disease
Validation set - Entire set
(n=94)
(n=90)
(n=160)
HR
95% CI
p†
HR
95% CI
p†
1·01
0·98 – 1·03
0·695
0·310
1.00
0.97 – 1.05
0.767
0.002
1
1·39
0·74 – 2·62
1
4.67
1.78 – 12.25
cN
0·559
N0
N+
1
1·23
1-2
3
1-unit increase in the
four-gene signature
0·48 – 2·10
0·28
0·13 – 0·63
0·002
1.00
0.97 – 1.03
1
2.96
1.54 – 6.67
0.59 – 4.04
0.17
0.06 – 0.43
< 0·001
0.880
0.001
1.30 – 7.83
0.370
1
1.55
p†
0.011
1
3.19
0.85 – 6.76
0·100
1
1·00
95% CI
0.098
1
2.40
0·61 – 2·47
Grade
HR
0.981
1
1.01
0.43 – 2.37
0.29
0.13 – 0.67
0.004
†Wald test p-values to assess the prognostic value of each variable within the multivariate analysis.
cT, clinical tumor size; cN, clinical nodal status; HR, Hazard ratio; CI, confidence interval
Table 3: Prognostic value of the four-gene signature on distant relapse-free survival
4
CM
Training
DRFS (n = 94)
Validation set - Residual disease
DRFS (n = 90)
Validation set - Entire set
DRFS (n = 160)
Training OS (n = 94)
CM + 4-gene signature
3 –year C-index
5 –year C-index
3 –year C-index
5 –year C-index
[95%CI]
[95%CI]
[95%CI]
[95%CI]
0.657
0.617
0.681
[0.507 – 0.807]
[0.488 – 0.745]
0.694
0.712
[0.557 – 0.832]
[0.574 – 0.850]
Difference
3-year C-index
increase [95%CI]
5-year C-index
increase [95%CI]
χ² increase
0.673
0.024
0.056
8.23
0.004
[0.558 – 0.804]
[0.566 – 0.779]
[-0.082 – 0.130]
[-0.051 – 0.163]
0.734
0.749
0.040
0.037
9.66
0.002
[0.617 – 0.851]
[0.631 – 0.867]
[-0.037 – 0.116]
[-0.032 – 0.105]
7.10
0.008
6.25
0.012
0.681
0.686
0.693
0.700
0.012
0.014
[0.584 – 0.779]
[0.592 – 0.780]
[0.598 – 0.788]
[0.606 – 0.795]
[-0.033 – 0.058]
[-0.028 – 0.056]
0.643
0.631
0.663
0.668
0.020
0.036
[0.504 – 0.783]
[0.449 – 0.764]
[0.544 – 0.782]
[0.554 – 0.781]
[-0.069 – 0.108]
[-0.051 – 0.123]
p
Uno’s concordance indices were computed to quantify the capacity of the prediction models in discriminating among subjects with different event times. Two
truncation times were considered: 3 years and 5 years. The concordane indices indicate how well the given prediction models work in predicting events that occur
in the time range from 0 to 3 years and 0 to 5 years, respectively. The likelihood ratio statistics was used in Cox regression models stratified on center to estimate
the added value of the 4-gene signature to the clinical models. 95% confidence intervals were obtained using 1000 bootstrap repetitions. CM, clinical model; Cindex, Concordance index; p, p-value; DRFS, distant relapse-free survival; OS, overall survival.
Table 4: Assessing the prognostic value of the four-gene signature to a clinical model
5
Документ
Категория
Без категории
Просмотров
0
Размер файла
818 Кб
Теги
2fmdx691, annonc
1/--страниц
Пожаловаться на содержимое документа