In silico strategies to improve insight in breast cancer
Bense, Rico
DOI:
10.33612/diss.101935267
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from
it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
2019
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Bense, R. (2019). In silico strategies to improve insight in breast cancer. Rijksuniversiteit Groningen.
https://doi.org/10.33612/diss.101935267
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the
author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the
number of authors shown on this cover page is limited to 10 maximum.
Rico D. Bense
1Christos Sotiriou
2Martine J. Piccart-Gebhart
2John B. A. G. Haanen
3Marcel A. T. M. van Vugt
1Elisabeth G. E. de Vries
1Carolien P. Schröder
1*
Rudolf S. N. Fehrmann
1*
1Department of Medical Oncology, University Medical Center Groningen,
University of Groningen, Groningen, the Netherlands
2
Department of Medical Oncology and Breast Cancer Translational Research Laboratory,
Institut Jules Bordet, Université Libre de Bruxelles, Brussels, Belgium
3
Division of Immunology, Netherlands Cancer Institute, Amsterdam, the Netherlands
*These authors contributed equally to this work
Relevance of tumor-infiltrating
immune cell composition and functionality
for disease outcome in breast cancer
ABSTRACT
Background: Not all breast cancer patients benefit from neoadjuvant or adjuvant therapy,
resulting in considerable undertreatment or overtreatment. New insights into the
role of tumor-infiltrating immune cells suggest that their composition, as well as their
functionality, might serve as a biomarker to enable optimal patient selection for current
systemic therapies and upcoming treatment options such as immunotherapy.
Methods: We performed several complementary unbiased in silico analyses on gene
expression profiles of 7,270 unrelated tumor samples of nonmetastatic breast cancer
patients with known clinical follow-up. CIBERSORT was used to estimate the fraction of
22 immune cell types to study their relations with pathological complete response (pCR),
disease-free survival (DFS), and overall survival (OS). In addition, we used four previously
reported immune gene signatures and a CD8+ T-cell exhaustion signature to assess their
relationships with breast cancer outcome. Multivariable binary logistic regression and
multivariable Cox regression were used to assess the association of immune cell–type
fractions and immune signatures with pCR and DFS/OS, respectively.
Results: Increased fraction of regulatory T-cells in human epidermal growth factor
receptor 2 (HER2)–positive tumors was associated with a lower pCR rate (odds ratio [OR]
= 0.15, 95% confidence interval [CI] = 0.03 to 0.69), as well as shorter DFS (hazard ratio
[HR] = 3.13, 95% CI = 1.23 to 7.98) and OS (HR = 7.69, 95% CI = 3.43 to 17.23). A higher
fraction of M0 macrophages in estrogen receptor (ER)–positive tumors was associated
with worse DFS (HR = 1.66, 95% CI = 1.18 to 2.33) and, in ER-positive/HER2-negative
tumors, with worse OS (HR = 1.71, 95% CI = 1.12 to 2.61). Increased fractions of γδ
T-cells in all breast cancer patients related to a higher pCR rate (OR = 1.55, 95% CI =
1.01 to 2.38), prolonged DFS (HR = 0.68, 95% CI = 0.48 to 0.98), and, in HER2-positive
tumors, with prolonged OS (HR = 0.27, 95% CI = 0.10 to 0.73). A higher fraction of
activated mast cells was associated with worse DFS (HR = 5.85, 95% CI = 2.20 to 15.54)
and OS (HR = 5.33, 95% CI = 2.04 to 13.91) in HER2-positive tumors. The composition
of relevant immune cell types frequently differed per breast cancer subtype. Furthermore,
a high CD8+ T-cell exhaustion signature score was associated with shortened DFS in
patients with ER-positive tumors regardless of HER2 status (HR = 1.80, 95% CI = 1.07
to 3.04).
Conclusions: The main hypothesis generated in our unbiased in silico approach is that a
multitude of immune cells are related to treatment response and outcome in breast cancer.
3
Breast cancer outcome has clearly improved in recent decades. Advances in neoadjuvant
and adjuvant treatment have contributed in large part to this progress. However, not
all patients benefit from standard treatment regimens,
1,2resulting in undertreatment or
overtreatment in many women. Predicting treatment response is particularly challenging
for upcoming treatment options such as immunotherapy,
3,4especially in view of the
potentially severe side effects of immunotherapeutic drugs. Consequently, optimal patient
selection for systemic therapy is crucial.
Breast cancer has long been thought of as a nonimmunogenic malignancy, but a
growing body of evidence suggests that this might not always be the case. The most widely
studied immune cells in this context are tumor-infiltrating lymphocytes (TILs). Presence
of TILs has been shown to be potentially predictive and prognostic in specific breast
cancer subtypes. Specifically in patients with human epidermal growth factor receptor 2
(HER2)–positive and triple-negative breast cancer (TNBC), large adjuvant studies have
shown that higher levels of TILs in primary biopsies are associated with improved overall
survival (OS) and fewer recurrences, regardless of therapy.
5–7In patients with TNBC
and HER2-positive tumors, increased levels of TILs are also associated with a higher
pathological complete response (pCR) rate following neoadjuvant therapy.
8–10Moreover,
patients with HER2-positive breast cancer and higher levels of TILs benefit more from
adjuvant trastuzumab treatment.
6Besides lymphocytes, tumors commonly contain tumor-associated macrophages
(TAMs). In breast cancer patients, these TAMs have been associated with a shorter
disease-free survival (DFS) and OS.
11–13TILs and TAMs are thus potential biomarkers.
In addition, several broader immune gene signatures have been developed and related to
breast cancer outcome.
14–17However, the number of TILs does not always predict response to treatment,
indicating that additional factors play a role. One possibility is that the functionality of
various tumor-infiltrating immune cells should also be taken into account. For example,
a CD8+ T-cell exhaustion signature, developed in purified circulating CD8+ T-cells,
has recently been related to favorable prognosis of patients with autoimmune and
inflammatory disease.
18It is still unknown whether CD8+ T-cell exhaustion might also be
relevant in tumors as a possible explanation for tumor immune evasion.
These new insights into the role of tumor-infiltrating immune cells suggest that their
composition as well as their functionality might be relevant for breast cancer management.
In the present study, we therefore performed several complementary unbiased in silico
analyses in an extensive data set comprising gene expression profiles of 7,270 unrelated
tumor samples of nonmetastatic breast cancer patients with known clinical follow-up
and 172 normal breast samples from women without breast disease. In this
hypothesis-generating study, we used CIBERSORT
19to estimate the fractions of 22 immune cell types,
which enabled us to study their independent associations with pCR, DFS, and OS in breast
cancer in general and its subtypes in a large number of patients. In addition, we assessed
the relationships with breast cancer outcome of four previously identified immune gene
signatures
14–17and a CD8+ T cell exhaustion signature.
18METHODS
Detailed methods information is provided in the Supplementary Methods.
Data acquisition
Publicly available raw microarray expression data from newly diagnosed primary tumors
of nonmetastasized breast cancer patients (prior to any treatment) and normal breast
tissue were collected from the Gene Expression Omnibus (GEO), as well as relevant
clinicopathological data and information on treatment regimen, pCR, and survival,
whenever available.
20Analysis was confined to samples hybridized to the HG-U133A
(GEO accession number GPL96) or Affymetrix HG-U133 Plus 2.0 (GEO accession
number GPL570) platforms. Preprocessing and aggregation of raw data was performed
according to the robust multi-array average algorithm. Quality control of the resulting
expression data was executed as previously described.
21–23Clinicopathological data collection
Information was collected on age, tumor histotype, grade, tumor size, TNM stage, lymph
node involvement, ER, progesterone receptor and HER2 status, treatment regimen, pCR,
DFS, and OS. Data on ER, progesterone receptor status, and HER2 status was collected
and scored according to immunohistochemistry staining guidelines of the American
Society of Clinical Oncology and College of American Pathologists.
24,25Whenever
immunohistochemistry data for receptor status were not reported, we determined
receptor status by means of inference (see details in the Supplementary Methods). For the
treatment regimen, we labeled all samples with missing information about treatment as a
separate category (“unknown”). DFS was defined as the interval between date of diagnosis
until date of development of distant metastasis. OS was defined as the interval between
date of diagnosis until date of death from any cause. The number of samples we used to
assess the independent predictive and prognostic value of immune cell–type fractions,
immune signatures, and CD8+ T-cell exhaustion signatures in breast cancer in general
and in subtypes are provided in Supplementary Tables 1-3.
3
Breast cancer subtypes
We performed analyses in several breast cancer subtypes based on receptor status and
in the intrinsic molecular subtypes as defined by Sorlie et al., Parker et al., and Hu et
al.
26–28In addition, Lehmann et al. described seven TNBC subgroups that were identified
by means of cluster analysis of gene expression profiles: basal-like 1, basal-like 2, unstable,
immunomodulatory, mesenchymal, mesenchymal stem-like, and luminal androgen
receptor.
29We applied the Lehmann classification to the collected TNBC tumors in order
to compare estimated immune cell–type fractions within TNBC subgroups.
Estimated immune cell type fractions
CIBERSORT is a method for characterizing cell composition of complex tissues from
their gene expression profiles that has been shown to have strong agreement with ground
truth assessments in bulk tumors.
19,30We used the leukocyte gene signature matrix,
termed LM22, which contains 547 genes that distinguish 22 human hematopoietic cell
phenotypes, including seven T-cell types, naive and memory B cells, plasma cells, natural
killer (NK) cells, and myeloid subsets. We used CIBERSORT in combination with the
LM22 signature matrix to estimate the fractions of 22 immune cell types in our collected
breast cancer and normal breast samples. For each sample, the sum of all estimate immune
cell–type fractions equals 1.
Immune gene signatures
We investigated the relationships between immune cell–type fractions and four published
immune signatures. Desmedt et al. identified an immune response gene signature
associated with prognosis in HER2-positive and ER-negative/HER2-negative breast
cancer subtypes.
14Teschendorff et al. determined that downregulation of a seven-gene
immune signature was related to a higher risk of distant metastases in patients with
ER-negative breast cancer.
15Perez et al. identified a set of immune function genes that
may provide a means of predicting benefit from adjuvant trastuzumab treatment.
16Gu-Trantien et al. defined an eight-gene CD4+ follicular helper T-cell signature (Tfh
signature) that predicted pathological tumor response following neoadjuvant therapy or
survival.
17To compute the immune signature scores—often derived from gene signatures
developed on other microarray platforms—for various data sets (distinct patient cohorts
and laboratories), we used the weighted average method previously described.
31We only
evaluated tumors that were hybridized to the Affymetrix HGU133 Plus 2 platform. This
ensured that we could use almost all genes that were part of individual immune signatures
to calculate the scores.
Statistical analysis
Distributions of the estimated immune cell–type fraction in normal breast tissue samples
and breast cancer samples were compared by Mann-Whitney U test. All areas under the
curves (AUCs) were rescaled within a range from -0.5 to 0.5. A negative AUC represented
a relatively lower fraction of immune cell type in breast cancer compared with normal
breast tissue, whereas a positive AUC represented a relatively higher fraction of an
immune cell type in breast cancer.
The predictive value of estimated immune cell–type fractions in the neoadjuvant
setting was assessed by multivariable binary logistic regression using pCR as outcome
variable and age, T-stage (because of a low number of reported tumor size), grade, lymph
node involvement, ER status, HER2 status, and treatment regimen as covariates. The
prognostic value of estimated immune cell–type fractions in neoadjuvant and adjuvant
settings was assessed by multivariable Cox regression analysis with time to distant
metastasis and time to death as outcome variables and age, tumor size, grade, lymph
node involvement, ER status, HER2 status, and treatment regimen as covariates. We used
the listwise deletion method for handling of missing data. With this method, an entire
sample is excluded from analysis if any single value is missing for the variables used in
the multivariable Cox regression and multivariable binary logistic regression. Analyses
were performed within a multivariable permutation testing framework for controlling
the proportion of false discovery.
32For each breast cancer subset analysis, we used the
multivariable permutation testing framework with 100 permutations and a false discovery
rate (FDR) of 25%. A FDR of 25% indicates that the result is likely to be valid three out of
four times. All results were considered statistically significant when P values were less than
0.05. All statistical tests were two-sided.
RESULTS
Data set containing 7,270 breast cancer samples and 172 normal breast tissue
samples
A summary of available baseline patient and primary tumor characteristics is presented in
Table 1. We also assembled a reference group of 172 normal breast tissue samples obtained
during reduction mammoplasty. Samples are classified according to their inferred ER and
HER2 status, intrinsic molecular subtype,
26–28or TNBC subgroup as defined by Lehmann
et al. (Fig. 1).
29Composition of tumor-infiltrating immune cells
Figure 2 shows the immune cell composition in normal breast tissue versus breast cancer
tissue (subtypes). Detailed results are provided in Supplementary Tables 4–20. Compared
3
Figure 1. Overview of breast cancer subtypes based on inferred receptor status, intrinsic molecular subtype, and triple-negative breast cancer subgroup classification. TNBC subgroups are classified as defined by Lehmann et al.29 The ER–positive (n = 4,906) and HER2–positive (n = 1,580) subtypes contain double cases, being the ER-positive/HER2-positive tumors (n = 812). ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; TNBC, triple-negative breast cancer.
Table 1. Baseline patient and primary tumor characteristics
Variable No. of samples % Valid %
Age at diagnosis, y ≤50 >50 Missing 1,854 2,408 3,008 25.5 33.1 41.4 43.5 56.5 Tumor grade 1 2 3 Missing 406 1,260 1,370 4,234 5.6 17.3 18.8 58.2 13.4 41.5 45.1 T-stage T0 T1 T2 T3 T4 Missing 8 445 1,466 467 306 4,578 0.1 6.1 20.2 6.4 4.2 63.0 0.3 16.5 54.2 17.3 11.7 Lymph node involvement
True False Missing 2,134 4,715 2,555 29.4 35.5 35.1 45.3 54.7 Stage I II III IV Missing 193 1,038 537 35 5,467 2.7 14.3 7.4 0.5 75.2 10.7 57.6 29.8 1.9 ER status Positive Negative Missing 1,294 476 5,500 17.8 6.5 75.7 73.1 26.9 HER2 status Positive Negative Missing 388 448 6,434 5.3 6.2 88.5 46.4 53.6 ER, estrogen receptor; HER2, human epidermal growth factor receptor 2.
with normal breast tissue, breast cancer tissue generally contained a higher fraction for
macrophages M0 (AUC = 0.34) and M1 (AUC = 0.22), T-cells follicular helper (AUC =
0.21), and regulatory T-cells (AUC = 0.28), whereas the plasma cell fraction was lower
(AUC = -0.25) (Fig. 2, left box). This pattern was similar for receptor-based breast cancer
subtypes compared with normal breast tissue. Within the intrinsic molecular subtypes,
especially Her2 and the basal subtype showed an increased fraction of macrophages M1
(AUC = 0.26 and AUC = 0.24, respectively). A relatively lower plasma cell fraction (AUC =
-0.11) was seen in the Her2 subtype compared with the other intrinsic molecular subtypes.
Within the Lehmann TNBC subgroups, the γδ T-cell fraction was higher (AUC = 0.11)
in the immunomodulatory subgroup compared with normal breast tissue (and relative to
the other TNBC subgroups), whereas it was lower in the mesenchymal subgroup (AUC
= -0.17). The CD8+ T-cell fraction was highest in the immunomodulatory (AUC = 0.17)
and luminal androgen receptor (AUC = 0.16) subgroups.
Immune cell–type fractions as independent predictive or prognostic factors
Figure 3 shows the statistical significance of all immune cell–type fractions as independent
predictive or prognostic factors for breast cancer subtypes. In the bubble heat map, a blue
bubble indicates that a higher fraction is associated with lower pCR rate, shorter DFS,
or shorter OS; a yellow bubble indicates that a higher fraction is associated with higher
pCR rate, prolonged DFS, or prolonged OS. The size of a bubble indicates the statistical
significance level. Detailed results are provided in Supplementary Tables 21–41. For
regulatory T-cells, in the HER2-positive subtype, a higher fraction was associated with
a lower pCR rate (odds ratio [OR] = 0.15, 95% confidence interval [CI] = 0.03 to 0.69),
worse DFS (hazard ratio [HR] = 3.13, 95% CI = 1.23 to 7.98), and worse OS (HR = 7.69,
95% CI = 3.43 to 17.23). A higher fraction of γδ T-cells was associated with a higher pCR
rate (OR = 1.55, 95% CI = 1.01 to 2.38) and prolonged DFS (HR = 0.68, 95% CI = 0.48 to
0.98) independent of receptor status, and OS in the HER2-positive subtype (HR = 0.27,
95% CI = 0.10 to 0.73). For macrophages M1, a higher fraction was associated with a
higher pCR rate (particularly in ER-positive disease; OR = 3.65, 95% CI = 1.51 to 8.82),
as well as prolonged DFS (irrespective of subtype; HR = 0.53, 95% CI = 0.35 to 0.80).
In the HER2-positive/ER-positive subtype, a higher macrophage M1 fraction was most
prominently associated with improved OS (HR = 0.22, 95% CI = 0.05 to 0.93). However,
the opposite association was observed for a higher macrophage M0 fraction, particularly
in ER-positive disease (irrespective of HER2 status) with DFS (HR = 1.66, 95% CI = 1.18
to 2.33), and for ER-positive/HER2-negative tumors with OS (HR = 1.71, 95% CI = 1.12 to
2.61). A higher activated mast cell fraction was associated with worse DFS and OS, most
clearly in HER2-positive disease (HR = 5.85, 95% CI = 2.20 to 15.54, and HR = 5.33, 95%
3
B cells naiv e B cells memo ry Plasma cell s T cells CD 8 T cells CD4 naiv e T cells CD4 memo ry restin g ry ac tivate d T cells fo llicular helper T cells regulato ry (T regs )T cells gamma delt
a NK cells restin g NK cells ac tivate d Mono cy te s Macr ophages M0 Macr ophages M1 Macr ophages M2 Dendritic cells restin g
Dendritic cells activate
d
Mast cells restin
g
Mast cells activate
d Eosinophil s Neutrophil s Breast cancer ER-neg/HER2-neg ER-neg/HER2-pos ER-pos/HER2-neg Normal-like Luminal A Luminal B Her2 Basal Basal-like 1 Basal-like 2 Unstable Mesenchymal stem-like Mesenchymal Immuno-modulatory Luminal androgen receptor
Receptor status based subgr
oups
In
trinsic molecular subtypes
Lehmann TNBC subgroups ER-pos/HER2-pos Figur e 2. Dis tri bu tio n o f imm une ce ll–ty pe f rac tio ns in b re as t c ance r s ub ty pes co m pa re d w ith he al th y b re as t tis su e. F rac tio ns o f e ac h imm un e ce ll t yp e w er e co m pa re d b y m ea ns o f t w o-side d M ann-W hi tn ey U t es t f or b re as t c an cer s ub typ es b as ed o n inf er re d r ecep to r s ta tu s, in tr in sic m ole cu la r s ub typ es, a nd t rip le-n ega tiv e b re as t c an cer s ubg ro ups a s defin ed b y L ehm ann et a l. 23 A gre en tri an gl e in dic at es a hig her imm un e ce ll–t yp e f rac tio n in b re as t c an cer a s co m pa re d w ith n or m al b re as t t iss ue . A pur pl e tri an gl e in dic at es a lo w er f rac tio n in b re as t c an cer a s co m pa re d w ith n or m al br ea st t iss ue . Th e size o f t he t ria ng le r ep res en ts t he a re a un der t he c ur ve o f t he eff ec t size o f t he s hift in t he di str ib ut io n o f imm un e ce ll–t yp e f rac tio ns. ER , es trog en r ecep to r; HER2, h um an ep ider m al gr owt h fac to r r ecep to r 2; TNB C, t rip le-n ega tiv e b re as t c an cer .
B cells naiv e B cells memo ry Plasma cell s T cells CD 8 T cells CD4 naiv e T cells CD4 memo ry restin g T cells CD4 memo ry ac tivated T cells fo llicular helper T cells regulato ry
T cells gamma delt
a NK cells restin g NK cells ac tivated Mono cy te s Macr ophages M0 Macr ophages M1 Macr ophages M2 Dendritic cells restin g
Dendritic cells activated
Mast cells
restin
g
Mast cells activated
Eosinophil
s
Neutrophil
s
Breast cancer (all) ER-pos HER2-pos ER-pos/HER2-pos ER-pos/HER2-neg ER-neg/HER2-pos ER-neg/HER2-neg Disease -free sur vival Pa
thological complete response
B cells naiv e B cells memo ry Plasma cell s T cells CD 8 T cells CD4 naiv e T cells CD4 memo ry restin g ry ac tivate d T cells fo llicular helper T cells regulato ry
T cells gamma delt
a NK cells restin g NK cells ac tivate d Mono cy te s Macr ophages M0 Macr ophages M1 Macr ophages M2 Dendritic cells restin g
Dendritic cells activate
d
Mast cells r
estin
g
Mast cells activate
d
Eosinophils
Neutrophil
s
Breast cancer (all) ER-pos HER2-pos ER-pos/HER2-pos ER-pos/HER2-neg ER-neg/HER2-pos ER-neg/HER2-neg
AB
B cells naiv e B cells memo ry Plasma cell s T cells CD 8 T cells CD4 naiv e T cells CD4 memo ry restin g T cells CD4 memo ry ac tivate d T cells fo llicular helper T cells regulato ryT cells gamma delt
a NK cells restin g NK cells ac tivate d Mono cy te s Macr ophages M0 Macr ophages M1 Macr ophages M2 Dendritic cells restin g
Dendritic cells activate
d
Mast
cells restin
g
Mast cells activate
d
Eosinophils
Neutrophil
s
Breast cancer (all) ER-pos HER2-pos ER-pos/HER2-pos ER-pos/HER2-neg ER-neg/HER2-pos ER-neg/HER2-neg Ov erall sur vival
C
Figur e 3. B ub bl e he at ma p f or the p re di ctiv e a nd p rog nos tic va lu es o f imm une ce ll–ty pe f rac tio ns in b re as t c ance r s ub ty pes. A ss oci at io ns b et w een f rac tio ns a nd ( A ) pCR , ( B) D FS, a nd ( C) OS w er e an al yze d. A bl ue b ub bl e in dic at es t ha t a hig her f rac tio n i s a ss oci at ed w ith lo w er pCR ra te , s ho rt er D FS, o r s ho rt er OS; a ye llow bu bb le in dic at es t ha t a hig her f rac tio n i s a ss oci at ed w ith hig her pCR ra te , p ro lo ng ed D FS, o r p ro lo ng ed OS. Th e siz e o f the bu bb le in dic at es t he s ta tis tic al sig nific an ce le ve l. Th e p re dic tiv e va lue o f imm un e ce ll–t yp e f rac tio ns in t he n eo ad ju va nt s et tin g wa s a ss es se d b y m ul tiva ria ble b in ar y log ist ic r eg res sio n u sin g pCR a s o ut co m e va ria ble a nd a ge , T -s ta ge , g rade , l ym ph n ode in vo lv em en t, ER s ta tu s, HER2 s ta tu s, a nd t re at m en t r eg im en a s co va ria tes. Th e p rog nos tic va lue of imm un e ce ll–t yp e frac tio ns in th e ne oad ju va nt an d ad ju va nt set tin gs wa s a ss es se d by m ul tiva ria ble C ox reg res sio n an al ysi s, w ith tim e to di sta nt m et as ta sis a nd tim e to de at h as ou tco m e va ria bles an d ag e, tum or size , g rade , l ym ph no de in vo lv em en t, ER sta tu s, HER2 sta tu s, an d tre at m en t r eg im en as co va ria tes. D FS. di se as e-f re e s ur vi va l; ER , es trog en re cep to r; HER2, hum an ep ider m al gr owt h fac to r r ecep to r 2; NK, n at ura l k iller ; OS, o vera ll s ur vi va l; pCR , p at ho log ic al co m plet e r es po ns e; TNB C, t rip le-n ega tiv e b re as t c an cer .3
Pathological complete response Disease-free survival
A
B
CD8-exhaustion Perez Tfh Teschendorff DesmedtBreast cancer (all) ER-pos HER2-pos ER-pos/HER2-po
s
ER-pos/HER2-neg ER-neg/ HER2-pos ER-neg/HER2-neg
CD8-exhaustion Perez Tfh Teschendorff Desmedt
Breast cancer (all
)
ER-po
s
HER2-po
s
ER-pos/ HER2-pos ER-pos/HER2-neg ER-neg/ HER2-po
s ER-neg/HER2-neg Overall survival
C
CD8-exhaustion Perez Tfh Teschendorff DesmedtBreast cancer (all) ER-po
s
HER2-po
s
ER-pos/ HER2-pos ER-pos/HER2-neg ER-neg/ HER2-po
s
ER-neg/HER2-neg
Figure 4. Bubble heat map for the predictive and prognostic values of immune gene signatures in breast cancer subtypes. Associations between fractions and (A) pCR, (B) DFS, and (C) OS were analyzed. Signatures identified by Desmedt et al.,14 Teschendorff et al.,15 Perez et al.,16 Gu-Trantien et al. (Tfh signature),17 and a CD8+ T-cell exhaustion signature18 were investigated. A blue bubble indicates that a higher fraction is associated with lower pCR rate, shorter DFS, or shorter OS; a yellow bubble indicates that a higher fraction is associated with higher pCR rate, prolonged DFS, or prolonged OS. The size of the bubble indicates the statistical significance level. DFS, disease-free survival; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; NK, natural killer; OS, overall survival, pCR, pathological complete response; TNBC, triple-negative breast cancer.
CI = 2.04 to 13.91, respectively). Also in HER2-positive disease, a higher activated NK cell
fraction was associated with prolonged DFS (HR = 0.39, 95% CI = 0.16 to 0.97), whereas a
higher resting NK cell fraction indicated the opposite (HR = 3.73, 95% CI = 1.30 to 10.68).
In addition, in the TNBC subtype, a higher fraction of resting NK cells was also associated
with worse DFS (HR = 18.91, 95% CI = 3.05 to 117.14) and OS (HR = 19.65, 95% CI = 1.66
to 232.56). For plasma cells, a higher fraction was associated with improved DFS (HR =
0.59, 95% CI = 0.40 to 0.88) regardless of receptor status.
Immune signatures as independent predictive or prognostic factors
Figure 4 shows the statistical significance of immune signatures as independent predictive
or prognostic factors. Detailed results are provided in Supplementary Tables 42–65. A
higher Tfh signature score was more statistically significantly associated in breast cancer
(irrespective of receptor status) with a higher pCR rate (OR = 1.68, 95% CI = 1.05 to 2.71),
prolonged DFS (HR = 0.42, 95% CI = 0.29 to 0.61), and prolonged OS (HR = 0.49, 95%
CI = 0.33 to 0.73) in comparison with the other three signatures. This applies to almost
all subtypes based on receptor status. A high CD8+ T-cell exhaustion signature score was
associated with shorter DFS in patients with ER-positive disease regardless of HER2 status
(HR = 1.80, 95% CI = 1.07 to 3.04).
DISCUSSION
We investigated the independent predictive and prognostic value of several in silico
immune phenotypes in a large set of breast cancer patients. In our analyses, we included
the clinicopathological parameters that are currently used in the clinical decision-making
for neoadjuvant and adjuvant treatment. This provided insight into multiple immune
parameters and their potential relevance for breast cancer management. This is of particular
interest in light of the current clinical developments of immune-modulating therapies.
Previously, it was thought that breast cancer was not an immunogenic cancer type, in
contrast to melanoma or renal cell cancer. However, our unbiased approach suggests the
hypothesis that the immune system is indeed involved in breast cancer. More specifically,
our data indicate that specific immune cells, depending on breast cancer subtypes, are
associated with highly relevant measures such as treatment response and survival.
First, we observed differences in subtypes with regard to immune cell fractions
associated with response to neoadjuvant chemotherapy and survival. An estimated high
regulatory T-cell fraction was associated with a lower pCR rate, as well as shortened DFS
and OS, particularly in patients with HER2- positive breast cancers, irrespective of ER
status. Previous studies have reported conflicting results regarding the prognostic value of
regulatory T-cell infiltration for OS and DFS in breast cancer patients. These studies were
either smaller, with 93 to 237 patients, or took a lower number of covariates into account
in their analyses.
33–37These associations of high regulatory T-cell fraction with worse
disease outcome parameters are of interest in the light of possible intervention strategies.
For instance, the anti-CTLA-4 antibody ipilimumab has been shown to downregulate
regulatory T-cell tumor infiltration in both melanoma and early-stage breast cancer.
38,39A higher estimated γδ T-cell fraction was associated with a higher pCR rate, especially
in patients with ER-positive breast cancer, irrespective of HER2 status. In addition, in
patients with HER2-positive/ER-negative tumors, a high γδ T-cell fraction was associated
with a prolonged DFS and OS. This is in line with recent findings from Gentles et al.,
30who
reported that γδ T-cells are the most statistically significant favorable prognostic immune
cell population for 39 malignancies, including breast cancer. However, in that study no
analysis of breast cancer subtypes was conducted, and fewer covariates were included to
assess the independent prognostic value.
A high estimated M1 macrophage fraction was associated with a higher pCR rate in
patients with ER-positive breast cancer (irrespective of HER2 status) and prolonged OS
particularly in patients with ER-positive disease. This supports the current hypothesis that
3
these macrophages are tumoricidal and therefore beneficial for prognosis.
40TAMs were
previously associated with shorter survival in breast cancer patients,
11–13which has been
attributed to their polarization towards the M2 subtype.
41In our analysis, however, we did
not find an association between M2 macrophage fraction and response to neoadjuvant
therapy, DFS, or OS. In contrast to M1 macrophages, a higher estimated fraction of M0
macrophages was associated with poor DFS, as well as shortened OS in patients with
ER-positive breast cancer. These macrophages are formed from monocytes when entering
the tissue and are not yet polarized toward either the M1 or M2 macrophage subtypes.
The hypothesis that M0 macrophage fractions seem relevant in both OS and DFS
underlines its possible impact on intrinsic ER-positive breast cancer biology and deserves
further attention in future studies. These apparently varying associations of macrophage
subpopulations with disease outcome parameters is of great interest, particularly in light
of the development of interventions affecting monocytes and macrophages.
42In patients with TNBC, we observed that a higher fraction of activated mast cells
was associated with a higher pCR rate. This is in accordance with several studies in breast
cancer that have linked mast cells to a good prognosis.
43–46However, in the present study,
an increased fraction of activated mast cells was also associated with poor DFS and OS in
patients with HER2-positive breast cancer. Indeed, mast cells are hypothesized to possess
both antitumoral and protumoral properties,
47which might vary according to breast
cancer subtype.
In patients with TNBC or HER2-positive breast cancer, we found that a higher fraction of
resting NK cells was associated with worse DFS and OS. Interestingly, NK cells have the
capacity to inhibit cytotoxic T-cell responses in mice and humans.
48The association with
worse DFS is in line with the lower pCR rate we observed for a higher fraction of NK cells
(resting and activated) for patients with breast cancer in general. The role of NK cells in
the clinical outcome in TNBC may provide for a future therapeutic target in TNBC.
With regard to functionality of immune cells in breast cancer, our data suggest that
a high score of the McKinney signature for CD8+ T-cell exhaustion
18is associated with
poor DFS in patients with ER-positive breast cancer. The relevance of T-cell exhaustion
in breast cancer, particularly in light of its apparent subtype relatedness, has hardly been
considered in previous studies. In chronic viral infection, CD8+ T-cell exhaustion has
recently been related to poor outcome,
49indicating its relation to immune system evasion.
In addition, Poschke et al. reported signs of exhaustion, such as loss of CD28, on
tumor-associated as compared with blood-derived CD8+ T-cells in early-stage breast cancer.
50Together with our results, these data suggest the hypothesis that CD8+ T-cell exhaustion
is also related to tumor immune evasion in breast cancer.
into which immune cell–type fractions and signatures could be of interest as independent
predictive or prognostic factors, we wanted to keep the power to detect potentially relevant
signals as high as possible (i.e., lower type II error). Therefore, we chose not to pursue a
split-sample approach with a discovery and validation cohort, which would decrease the
type I error (i.e., false-positive findings). We think that any future use of immune cell–type
fractions and signature as independent predictive and prognostic factors in breast cancer
management warrants additional validation in well-designed studies controlling the type
I error.
The main hypothesis generated in our unbiased in silico approach is that a multitude
of immune cells are related to treatment response and outcome in breast cancer. Varying
immune cell fractions seem to be important in particular breast cancer subtypes,
indicating the complexity of immune system involvement in breast cancer. The results of
our study also justify an unbiased approach for gaining insight into this system. The recent
study by Nanda et al. has provided initial indications that immunotherapy can be effective
for treating breast cancer.
51Even in ER-positive breast cancer, which was previously
considered a particularly nonimmunogenic disease, preliminary data have shown clinical
efficacy of immunotherapy.
52However, as in TNBC, this was the case only in a subset
of patients. Insight into how to select the best treatment for the right patient is urgently
needed. The present study may provide a further step in that direction.
Funding
This research was supported by Dutch Cancer Society grant RUG 2010-4739 to C. P.
Schröder and NWO-VENI grant (916-16025), the Bas Mulder award of Alpe d’HuZes/
Dutch Cancer Society (RUG 2013-5960), and a Mandema Stipendium to R.S.N. Fehrmann.
Notes
The study funders had no role in the design of the study; the collection, analysis, or
interpretation of the data; the writing of the manuscript; or the decision to submit the
manuscript for publication.
We thank T.N. Schumacher for the critical discussion of the results presented in this
manuscript.
CPS and RSNF were responsible for the conception and design of this study.
RDB and RSNF collected and assembled data. All authors contributed to data analysis
and interpretation, the writing of this manuscript, and the final decision to submit the
manuscript.
3
References
1. Harris LN, You F, Schnitt SJ, et al. Predictors of resistance to preoperative trastuzumab and vinorelbine for HER2-positive
early breast cancer. Clin Cancer Res. 2007;13:1198–1207.
2. Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). Relevance of breast cancer hormone receptors and other
factors to the efficacy of adjuvant tamoxifen: patient-level meta-analysis of randomised trials. Lancet. 2011; 378:771–784.
3. Emens L, Braiteh F, Cassier P, et al. Inhibition of PD-L1 by MPDL3280A leads to clinical activity in patients with metastatic
triple-negative breast cancer. San Antonio Breast Cancer Symposium. 2014;abstr PD1-6.
4. Nanda R, Chow L, Dees E, et al. A phase Ib study of pembrolizumab (MK-3475) in patients with advanced triple-negative
breast cancer. San Antonio Breast Cancer Symposium. 2014;abstr S1-09.
5. Loi S, Sirtaine N, Piette F, et al. Prognostic and predictive value of tumor-infiltrating lymphocytes in a phase III randomized
adjuvant breast cancer trial in node-positive breast cancer comparing the addition of docetaxel to doxorubicin with doxorubicin-based chemotherapy: BIG 02-98. J Clin Oncol. 2013;31:860–867.
6. Loi S, Michiels S, Salgado R, et al. Tumor infiltrating lymphocytes are prognostic in triple negative breast cancer and
predictive for trastuzumab benefit in early breast cancer: results from the FinHER trial. Ann Oncol. 2014;25:1544–1550.
7. Salgado R, Denkert C, Campbell C, et al. Tumor-infiltrating lymphocytes and associations with pathological complete
response and event-free survival in HER2-positive early-stage breast cancer treated with lapatinib and trastuzumab. JAMA Oncol. 2015;1:448–454.
8. 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.
9. Ono M, Tsuda H, Shimizu C, et al. Tumor-infiltrating lymphocytes are correlated with response to neoadjuvant
chemotherapy in triple-negative breast cancer. Breast Cancer Res Treat. 2012;132:793–805.
10. 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 2-positive and triple-negative primary breast cancers. J Clin Oncol. 2015;33:983–991.
11. Leek RD, Lewis CE, Whitehouse R, Greenall M, Clarke J, Harris AL. Association of macrophage infiltration with angiogenesis and prognosis in invasive breast cancer. Cancer Res. 1996;56:4625–4629.
12. Campbell MJ, Tonlaar NY, Garwood ER, et al. Proliferating macrophages associated with high grade, hormone receptor negative breast cancer and poor clinical outcome. Breast Cancer Res Treat. 2011;128:703–711.
13. Zhang Y, Cheng S, Zhang M, et al. High-infiltration of tumor-associated macrophages predicts unfavorable clinical outcome for node-negative breast cancer. PLoS One. 2013;8:e76147.
14. Desmedt C, Haibe-Kains B, Wirapati P, et al. Biological processes associated with breast cancer clinical outcome depend on the molecular subtypes. Clin Cancer Res. 2008;14:5158-5165.
15. Teschendorff AE, Miremadi A, Pinder SE, Ellis IO, Caldas C. An immune response gene expression module identifies a good prognosis subtype in estrogen receptor negative breast cancer. Genome Biol. 2007;8:R157.
16. Perez EA, Thompson EA, Ballman KV, et al. Genomic analysis reveals that immune function genes are strongly linked to clinical outcome in the North Central Cancer Treatment Group N9831 Adjuvant Trastuzumab Trial. J Clin Oncol. 2015;33:701–708.
17. Gu-Trantien C, Loi S, Garaud S, et al. CD4+ follicular helper T-cell infiltration predicts breast cancer survival. J Clin Invest. 2013;123:2873-2892.
18. McKinney EF, Lyons PA, Carr EJ, et al. A CD8+ T-cell transcription signature predicts prognosis in autoimmune disease. Nat Med. 2010;16:586–591.
19. Newman AM, Liu CL, Green MR, et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods. 2015;12:453–457.
20. Barrett T, Wilhite SE, Ledoux P, et al. NCBI GEO:archive for functional genomics data sets - update. Nucleic Acids Res. 2013;41:991–995.
21. Crijns APG, Fehrmann RSN, De Jong S, et al. Survival-related profile, pathways, and transcription factors in ovarian cancer. PLoS Med. 2009;6:e1000024.
22. Heijink DM, Fehrmann RSN, de Vries EGE, et al. A bioinformatical and functional approach to identify novel strategies for chemoprevention of colorectal cancer. Oncogene. 2011;30:2026–2036.
23. Fehrmann RSN, Karjalainen JM, Krajewska M, et al. Gene expression analysis identifies global gene dosage sensitivity in cancer. Nat Genet. 2015;47:115–125.
guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer. J Clin Oncol. 2010;28:2784–2795.
25. Wolff AC, Hammond MEH, Hicks DG, et al. Recommendations for human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists clinical practice guideline update. J Clin Oncol. 2013;31:3997–4013.
26. Sorlie T, Tibshirani R, Parker J, et al. Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc Natl Acad Sci U S A. 2003;100:8418–8423.
27. Parker JS, Mullins M, Cheung MCU, et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol. 2009;27:1160–1167.
28. Hu Z, Fan C, Oh DS, et al. The molecular portraits of breast tumors are conserved across microarray platforms. BMC Genomics. 2006;7:96.
29. Lehmann BD, Bauer JA, Chen X, et al. Identification of human triple-negative breast cancer subtypes and preclinical models for selection of targeted therapies. J Clin Invest. 2011;121:2750–2767.
30. Gentles AJ, Newman AM, Liu CL, et al. The prognostic landscape of genes and infiltrating immune cells across human cancers. Nat Med. 2015;21:938-945.
31. 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.
32. Korn EL, Troendle JF, McShane LM, Simon R. Controlling the number of false discoveries: application to high-dimensional genomic data. J Stat Plan Inference. 2014;124:379–398.
33. Bates GJ, Fox SB, Han C, et al. Quantification of regulatory T-cells enables the identification of high-risk breast cancer patients and those at risk of late relapse. J Clin Oncol. 2006;24:5373–5380.
34. Liu F, Lang R, Zhao J, et al. CD8+ cytotoxic T-cell and FOXP3+ regulatory T-cell infiltration in relation to breast cancer survival and molecular subtypes. Breast Cancer Res Treat. 2011;130:645–655.
35. Liu S, Foulkes WD, Leung S, et al. Prognostic significance of FOXP3. Tumor-infiltrating lymphocytes in breast cancer depends on estrogen receptor and human epidermal growth factor receptor-2 expression status and concurrent cytotoxic T-cell infiltration. Breast Cancer Res. 2014;16:432.
36. Merlo A, Casalini P, Carcangiu ML, et al. FOXP3 expression and overall survival in breast cancer. J Clin Oncol. 2009;27:1746–1752.
37. Aruga T, Suzuki E, Saji S, et al. A low number of tumor-infiltrating FOXP3-positive cells during primary systemic chemotherapy correlates with favourable anti-tumor response in patients with breast cancer. Oncol Rep. 2011;25:223–230. 38. Romano E, Kusio-Kobialka M, Foukas PG, et al. Ipilimumab-dependent cell-mediated cytotoxicity of regulatory T-cells ex
vivo by nonclassical monocytes in melanoma patients. Proc Natl Acad Sci U S A. 2015;112:6140–6145.
39. Diab A, McArthur H, Solomon S, Sacchin I, Comstock C, Maybody M. A pilot study of preoperative (Pre-op), single-dose ipilimumab (Ipi) and/or cryoablation (Cryo) in women (pts) with early-stage/resectable breast cancer. J Clin Oncol. 2014;32;5s(suppl; abstr 1098).
40. Mantovani A, Sozzani S, Locati M, Allavena P, Sica A. Macrophage polarization: tumor-associated macrophages as a paradigm for polarized M2 mononuclear phagocytes. Trends Immunol. 2002;23:549–555.
41. Mantovani A, Sica A. Macrophages, innate immunity and cancer: balance, tolerance, and diversity. Curr Opin Immunol. 2010;22:231–237.
42. Nywening TM, Wang-Gillam A, Sanford DE, et al. Targeting tumour-associated macrophages with CCR2 inhibition in combination with FOLFIRINOX in patients with borderline resectable and locally advanced pancreatic cancer: a single-centre, open-label, dose-finding, non-randomised, phase 1b trial. Lancet Oncol. 2016;17:651–662.
43. Aaltomaa S, Lipponen P, Palpinaho S, Kosma V. Mast cells in breast cancer. Anticancer Res. 1993;13:785–788. 44. Dabiri S, Huntsman D, Makretsov N, et al. The presence of stromal mast cells identifies a subset of invasive breast cancers
with a favorable prognosis. Mod Pathol. 2004;17:690–695.
45. Rajput A, Turbin D, Cheang M, et al. Stromal mast cells in invasive breast cancer are a marker of favourable prognosis: a study of 4,444 cases. Breast Cancer Res Treat. 2008;107:249–257.
46. Rovere F Della, Granata A, Familiari D, D’Arrigo G, Mondello B, Basile G. Mast cells in invasive ductal breast cancer: different behavior in high and minimum hormone-receptive cancers. Anticancer Res. 2007;27:2465–2471.
47. Khazaie K, Blatner NR, Khan MW, et al. The significant role of mast cells in cancer. Cancer Metastasis Rev. 2011;30:45–60. 48. Crome S, Lang P, Lang K, Ohashi P. Natural killer cells regulate diverse T-cell responses. Trends Immunol. 2013;34:342–
349.
3
autoimmunity and infection. Nature. 2015;523:612–616.50. Poschke I, De Boniface J, Mao Y, Kiessling R. Tumor-induced changes in the phenotype of blood-derived and tumor-associated T-cells of early stage breast cancer patients. Int J Cancer. 2012;131:1611–1620.
51. Nanda R, Chow LQM, Dees EC, et al. Pembrolizumab in patients with advanced triple-negative breast cancer: phase Ib KEYNOTE-012 study. J Clin Oncol. 2016;34:2460-2467.
52. Rugo H, Delord J-P, Im S-A, et al. Preliminary efficacy and safety of pembrolizumab (MK-3475) in patients with PD-L1–positive, estrogen receptor-positive (ER+)/HER2-negative advanced breast cancer enrolled in KEYNOTE-028. San Antonio Breast Cancer Symposium. 2015;abstr S5-S07.
SUPPLEMENTARY METHODS
Data acquisition
Publicly available raw microarray expression data from patient-derived nonmetastasized
breast cancer samples (prior to any treatment) and normal breast tissue collected from
GEO, as well as relevant clinicopathological data and information on treatment regimen,
survival and response to neoadjuvant therapy whenever available.
1We confined analyses
to samples hybridized to the HG-U133A (GEO accession number GPL96) or Affymetrix
HG-U133 Plus 2.0 (GEO accession number GPL570) platforms. For these two platforms we
downloaded the Simple Omnibus Format in Text (SOFT) file. This file is publicly available
per platform and provides per sample information as provided by the experimenter who
uploaded the data to GEO (downloaded March 2014).
Search strategy
SOFT files were parsed to extract sample descriptions. To identify breast cancer relevant
samples we used a two-step search strategy: automated filtering on keywords followed by
manual curation. We automatically filtered potential relevant samples from the two SOFT
files by retaining samples that contained at least one of the following keywords
(case-insensitive) in any description field in the SOFT format: ‘breast’, ‘TNBC’, ‘HER2’, ‘estrogen’,
‘progesterone’, ‘HER-2’, ‘mamma’, ‘ERBB2’, ‘HER2/neu’. These keywords were aimed at
sensitivity, not specificity, and manual curation was necessary to assess the relevance of
samples that were retained after filtering on keywords. During this manual curation step,
samples were retained if they represented normal breast tissue or breast cancer samples
obtained from patients. Samples of patients treated before biopsy were removed. Cell line
samples were deemed irrelevant and excluded from further analysis.
Sample processing
Non-corrupted raw data CEL files were downloaded from GEO for the selected samples.
To identify samples that were uploaded to GEO multiple times we generated a MD5 hash
for each CEL file. A MD5 hash acts like a unique fingerprint for each file: duplicate CEL
files have an identical MD5 hash. After removal of duplicate CEL files, pre-processing and
aggregation of CEL files was performed with Affymetrix Power Tools version 1.15.2, using
apt-probe set-summarize and applying the robust multi-array average RMA algorithm,
using the Affymetrix GeneChip Array CDF layout files (downloaded March 2014). We
used the latest probe annotation file provided by Affymetrix (version 34). Principal
component analysis quality control of the resulting expression data was performed as
previously described.
2–4Clinicopathological data collection
Initially all data were processed as provided on GEO by the researchers. Subsequently,
we searched the articles corresponding to the included samples and their supplementary
files for additional clinicopathological data that we could not find in GEO. We collected
information concerning age, tumor histotype, grade, TNM–stage, lymph node
involvement, ER, progesterone receptor (PR) and human HER2 status, treatment regimen,
pCR, DFS and OS.
For age we created an extra variable by classifying samples into two groups using an
age cut-off of 50 years. Tumor grade was collected according to the Nottingham system.
5TNM staging was collected according to the guideline of the American Joint Committee
on Cancer.
6Tumor samples were classified into the following histotype categories:
“invasive ductal carcinoma”, “invasive lobular carcinoma”, “mixed carcinoma”, and “other”.
Data on hormone (ER, PR) and HER2 status were collected, including the cut-off values
used to define a positive or negative receptor status. Cut-off values for receptor statuses
were not uniformly defined. For example, some studies used an immunohistochemistry
(IHC) cut-off value of 10% for ER status, while others used 1%. For clarity, we based
our cut off-values on the most recent guidelines of the American Society of Clinical
Oncology.
7,8In accordance with this guideline, for ER and PR status, we used an IHC
cut-off value of 1%. Samples were considered HER2-positive when they reached an IHC
score of 3+. Samples with a HER2 IHC score of 2+ with a HER2/CEP17 ratio ≥2.0 were
also considered positive. A HER2 IHC score of +1 or 0 was considered negative. If we
could not redefine the receptor status according to the ASCO guidelines (e.g. in case of a
negative status when a cut-off of 10% on IHC was used by the corresponding study) we
filled in a blank. Information on molecular subtyping using gene expression profiles were
collected for three methods developed by Sorlie et al., Parker et al., and Hu et al.
9–11For
the treatment regimen we labeled all samples with missing information for treatment as a
separate category (“unknown”). Reported response to neoadjuvant therapy of samples was
processed as either pCR or residual disease. DFS was defined as the interval between date
of diagnosis until date of distant metastasis development. OS was defined as the interval
between date of diagnosis until date of death from any cause.
3
Information of duplicate samples was checked for concordance. First, in duplicate
samples in which different methods of measurement were used, we retained information
from the method that is most commonly used in oncological practice (e.g. a grade
according to Nottingham was preferred to Black’s modified nuclear grade). Next, when
information of one duplicate was described in greater detail compared to the other,
information of the first was retained. When the situations described above did not apply
to discordant sample information, we retained the information on the duplicate that was
uploaded first on GEO.
After correcting discordant information, we used the information of duplicates
to supplement missing data of the paired samples. After this, duplicate samples were
considered redundant and therefore removed.
Receptor status inference
For receptor status inference we used both regular mRNA expression and functional
genomic mRNA (FGM) expression. FGM profiling is a novel method that corrects
gene expression data (i.e. mRNA expression data) for major, non-genetic factors (e.g.
physiological, metabolic, cell-type-specific and experimental factors). For a detailed
description of FGM profiling, we refer to Fehrmann et al.
4For all probes we explored the
empirical expression distributions (regular mRNA expression and FGM expression) for
ER, PR and HER2 status of negative and positive breast cancer samples (receptor status
defined according to guidelines). Both ER and PR positive and negative status were best
discriminated by the regular mRNA expression levels of the Affymetrix probe 205225_at.
HER2 positive and negative status was best discriminated by the FGM expression level of
Affymetrix probe 216836_s_at. We used these probes to infer receptor status of samples
that were missing ER, PR or HER2 status according to guidelines. Thresholds were defined
by selecting the mRNA or FGM expression value that resulted in the optimal balance in
sensitivity between the reported guideline negative and positive samples. Samples with
an expression value below the defined threshold were assigned a negative receptor status.
Likewise, when samples had an expression value above the threshold, they were considered
ER, PR or HER2 positive. If the inferred receptor status was discordant with an available
guideline receptor status, the given receptor status overruled the inferred receptor status.
Breast cancer subtypes
We investigated estimated immune cell type fractions in breast cancer subtypes based on
hormone receptor and HER2 status as well as intrinsic molecular subtypes as defined by
Sorlie et al., Parker et al. and Hu et al.
9–11In addition, Lehmann et al. described 7 TNBC
basal-like 1, basal-like 2, unstable, immunomodulatory, mesenchymal, mesenchymal
stem-like, and luminal androgen receptor.
12This cluster analysis was repeated for our
collected breast cancer samples to compare estimated immune cell type fractions within
TNBC subgroups.
Immune cell type fraction estimation
The recently introduced CIBERSORT method is a tool for characterizing cell composition
of complex tissues from their gene expression profiles (https://cibersort.stanford.edu/).
13We used the leukocyte gene signature matrix, called LM22, which contains 547 genes that
distinguish 22 human hematopoietic cell phenotypes, including seven T-cell types, naive
and memory B cells, plasma cells, natural killer cells and myeloid subsets. CIBERSORT
in combination with the LM22 signature matrix was used to estimate the fraction for 22
immune cell types in our collected breast cancer and normal breast tissue samples. For
each sample, the sum of all estimated immune cell type fractions equals 1.
Immune signatures
We investigated the relationship between immune cell type fractions and four published
immune signatures.
14–17Spearman rank correlation coefficients were calculated between
immune cell type fractions and immune signature scores. To compute the immune
signature scores— often derived from gene signatures developed on other microarray
platforms—for different datasets (distinct patient cohorts and laboratories), we used the
weighted average method as previously described.
18We only used tumor samples that
were hybridized to the Affymetrix HG-U133 Plus 2 platform. This ensured that we could
use almost all genes that were part of individual immune signatures to calculate the scores.
Statistical analysis
Distributions of the estimated immune cell type fractions in breast cancer samples and
normal breast tissue were compared by Mann-Whitney U test. All AUCs were rescaled
within a range from -0.5 to 0.5. A negative AUC represented a relatively lower fraction of
immune cell type in breast cancer compared to normal breast tissue, whereas a positive
AUC represented a relatively higher fraction of an immune cell type in breast cancer.
The predictive value of immune cell type fractions in the neoadjuvant setting was
assessed by multivariate binary logistic regression using pCR as outcome variable and age,
T-stage (due to a low number of reported tumor size), grade, lymph node involvement,
ER status, HER2 status and treatment regimen as covariates. The prognostic value of
estimated immune cell type fractions in the neoadjuvant and adjuvant setting was assessed
by multivariate Cox regression analysis with time to distant metastasis and time to death
3
as outcome variables, and age, tumor size, grade, lymph node involvement, ER status,
HER2 status and treatment regimen as covariates. All results were considered significant
when P <0.05. We applied the listwise deletion method for handling missing data. With
this method, an entire sample is excluded from analysis if any single value is missing
for the variables used in the multivariate cox regression and multivariate binary logistic
regression. Analyses were performed within a multivariate permutation testing framework
for controlling the proportion of false discovery.
19For each breast cancer subset analysis,
we applied the multivariate permutation testing framework with 100 permutations and a
FDR of 25%. A FDR of 25% indicates that the result is likely to be valid 3 out of 4 times.
References
1. Barrett T, Wilhite SE, Ledoux P, et al. NCBI GEO: archive for functional genomics data sets - update. Nucleic Acids Res.
2013;41:991-995.
2. Crijns APG, Fehrmann RSN, de Jong S, et al. Survival-related profile, pathways, and transcription factors in ovarian
cancer. PLoS Med. 2009;6:e1000024.
3. Heijink DM, Fehrmann RSN, de Vries EGE, et al. A bioinformatical and functional approach to identify novel strategies
for chemoprevention of colorectal cancer. Oncogene. 2011;30:2026-2036.
4. Fehrmann RSN, Karjalainen JM, Krajewska M, et al. Gene expression analysis identifies global gene dosage sensitivity in
cancer. Nat Genet. 2015;47:115-125.
5. Elston CW, Ellis IO. Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer:
experience from a large study with long-term follow up. Histopathology. 1991;19:403-410.
6. Edge SB, Byrd DR, Compton CC, Fritz AG, Greene EL, Trotti A. AJCC Cancer Staging Manual. 7th ed. New York, NY:
Springer-Verlag; 2010.
7. Hammond MEH, Hayes DF, Dowsett M, et al. American Society of Clinical Oncology/College of American Pathologists
guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer. J Clin Oncol. 2010;28:2784-2795.
8. Wolff AC, Hammond MEH, Hicks DG, et al. Recommendations for human epidermal growth factor receptor 2 testing in
breast cancer: American Society of Clinical Oncology/College of American Pathologists clinical practice guideline update. J Clin Oncol. 2013;31:3997-4013.
9. Sorlie T, Tibshirani R, Parker J, et al. Repeated observation of breast tumor subtypes in independent gene expression data
sets. Proc Natl Acad Sci U S A. 2003;100:8418-8423.
10. Parker JS, Mullins M, Cheung MCU, et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol. 2009;27:1160-1167.
11. Hu Z, Fan C, Oh DS, et al. The molecular portraits of breast tumors are conserved across microarray platforms. BMC Genomics. 2006;7:96.
12. Lehmann BD, Bauer JA, Chen X, et al. Identification of human triple-negative breast cancer subtypes and preclinical models for selection of targeted therapies. J Clin Invest. 2011;121:2750-2767.
13. Newman AM, Liu CL, Green MR, et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods. 2015;12:453-457.
14. Desmedt C, Haibe-Kains B, Wirapati P, et al. Biological processes associated with breast cancer clinical outcome depend on the molecular subtypes. Clin Cancer Res. 2008;14:5158-5165.
15. Teschendorff AE, Miremadi A, Pinder SE, Ellis IO, Caldas C. An immune response gene expression module identifies a good prognosis subtype in estrogen receptor negative breast cancer. Genome Biol. 2007;8:R157.
16. Perez EA, Thompson EA, Ballman KV, et al. Genomic analysis reveals that immune function genes are strongly linked to clinical outcome in the North Central Cancer Treatment Group N9831 Adjuvant Trastuzumab Trial. J Clin Oncol. 2015;33:701-708.
17. Gu-Trantien C, Loi S, Garaud S, et al. CD4+ follicular helper T cell infiltration predicts breast cancer survival. J Clin Invest. 2013;123:2873-2892.
18. 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.
19. Korn EL, Troendle JF, McShane LM, Simon R. Controlling the number of false discoveries: application to high-dimensional genomic data. J Stat Plan Inference. 2014;124:379-398.
SUPPLEMENTARY DATA
Supplementary Table 1. Cases included in multivariate analysis for response to neoadjuvant therapy
Breast cancer subtype Total (n) pCR (n) Residual disease (n) All ER+ HER2+ ER+/HER2-ER+/HER2+ ER-/HER2+ ER-/HER2-611 373 103 306 67 36 202 132 47 22 39 8 14 71 479 326 81 267 59 22 131
The predictive value of immune cell type fractions in the neoadjuvant setting was assessed by multivariate binary logistic regression using pCR as outcome variable and age, T-stage (due to a low number of reported tumor size), grade, lymph node involvement, ER status, HER2 status and treatment regimen as covariates. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; pCR, pathological complete response.
Supplementary Table 2. Cases included in multivariate analysis for disease-free survival analysis
Breast cancer subtype Total (n) Event (n) Censored (n) All ER+ HER2+ ER+/HER2-ER+/HER2+ ER-/HER2+ ER-/HER2-846 657 147 576 78 69 119 147 115 31 97 18 13 19 699 542 116 479 60 56 100
The prognostic value of estimated immune cell type fractions in the neoadjuvant and adjuvant setting was assessed by multivariate Cox regression analysis with time to distant metastasis and time to death as outcome variables and age, tumor size, grade, lymph node involvement, ER status, HER2 status and treatment regimen as covariates. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2.
Supplementary Table 3. Cases included in multivariate analysis for overall survival
Breast cancer subtype Total (n) Event (n) Censored (n) All ER+ HER2+ ER+/HER2-ER+/HER2+ ER-/HER2+ ER-/HER2-632 472 119 416 55 64 95 153 107 39 87 20 19 27 479 365 80 329 35 45 68
The prognostic value of estimated immune cell type fractions in the neoadjuvant and adjuvant setting was assessed by multivariate Cox regression analysis with time to death and time to distant metastasis as outcome variables and age, tumor size, grade, lymph node involvement, ER status, HER2 status and treatment regimen as covariates. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2.
