reconstructing tumor history in breast cancer: signatures

ORIGINAL ARTICLE

Reconstructing tumor history in breast cancer: signatures of mutationalprocesses and response to neoadjuvant chemotherapy5

C. Denkert1,2*y, M. Untch3y, S. Benz4, A. Schneeweiss5, K. E. Weber6, S. Schmatloch7, C. Jackisch8, H. P. Sinn9,10,J. Golovato4, T. Karn11, F. Marmé12, T. Link13, J. Budczies2,9,10, V. Nekljudova6, W. D. Schmitt2, E. Stickeler14, V. Müller15,P. Jank1,2, R. Parulkar4, E. Heinmöller16, J. Z. Sanborn4, C. Schem17, B. V. Sinn2, P. Soon-Shiong4, M. van Mackelenbergh18,P. A. Fasching19, S. Rabizadeh4 & S. Loibl6,20

1Institute of Pathology, Philipps-University Marburg and University Hospital Marburg (UK-GM), Marburg; 2Charité e Universitätsmedizin Berlin, Institute of Pathology,Berlin; 3Helios Klinikum Berlin-Buch, Department of Obstetrics and Gynaecology, Berlin, Germany; 4NantOmics, LLC, Culver City, USA; 5Nationales Centrum fürTumorerkrankungen, Universitätsklinikum und Deutsches Krebsforschungszentrum Heidelberg, Heidelberg; 6German Breast Group (GBG), Neu-Isenburg;7Brustzentrum Kassel, Elisabeth Krankenhaus, Kassel; 8Department of Obstetrics and Gynecology and Breast Cancer Center, Sana Klinikum Offenbach, Offenbach;9Institute of Pathology, University Hospital Heidelberg, Heidelberg; 10German Cancer consortium (DKTK), Heidelberg; 11Klinik für Frauenheilkunde und Geburtshilfe,Goethe Universität, Frankfurt; 12Universitätsfrauenklinik Mannheim, Mannheim; 13Department of Gynecology and Obstetrics, Technische Universität Dresden,Dresden; 14Department of Gynecology, RWTH Aachen, Aachen; 15Department of Gynecology, Universitätsklinikum Hamburg-Eppendorf, Hamburg; 16PathologieNordhessen, Kassel; 17Mammazentrum Hamburg am Krankenhaus Jerusalem, Hamburg; 18Universitätsklinikum Schleswig-Holstein, Klinik für Gynäkologie undGeburtshilfe, Kiel; 19Department of Gynecology and Obstetrics, University Hospital Erlangen, Comprehensive Cancer Center Erlangen-EMN, Erlangen; 20University ofFrankfurt, Frankfurt am Main, Germany

*CorrespBaldingersGermany.E-mail: c

5Note: Ty Co-firs0923-75

500

Available online 6 January 2021

Background: Different endogenous and exogenous mutational processes act over the evolutionary history of amalignant tumor, driven by abnormal DNA editing, mutagens or age-related DNA alterations, among others, togenerate the specific mutational landscape of each individual tumor. The signatures of these mutational processescan be identified in large genomic datasets. We investigated the hypothesis that genomic patterns of mutationalsignatures are associated with the clinical behavior of breast cancer, in particular chemotherapy response andsurvival, with a particular focus on therapy-resistant disease.Patients and methods: Whole exome sequencing was carried out in 405 pretherapeutic samples from the prospectiveneoadjuvant multicenter GeparSepto study. We analyzed 11 mutational signatures including biological processes suchas APOBEC-mutagenesis, homologous recombination deficiency (HRD), mismatch repair deficiency and also age-relatedor tobacco-induced alterations.Results: Different subgroups of breast carcinomas were defined mainly by differences in HRD-related andAPOBEC-related mutational signatures and significant differences between hormone-receptor (HR)-negativeand HR-positive tumors as well as correlations with age, Ki-67 and immunological parameters wereobserved. We could identify mutational processes that were linked to increased pathological completeresponse rates to neoadjuvant chemotherapy with high significance. In univariate analyses for HR-positivetumors signatures, S3 (HRD, P < 0.001) and S13 (APOBEC, P ¼ 0.001) as well as exonic mutation rate(P ¼ 0.002) were significantly correlated with increased pathological complete response rates. The signaturesS3 (HRD, P ¼ 0.006) and S4 (tobacco, P ¼ 0.011) were prognostic for reduced disease-free survival ofpatients with chemotherapy-resistant tumors.

ondence to: Prof. Carsten Denkert, UKGM (University Hospital of Gießen and Marburg) - Universitätsklinikum Marburg, Philipps-Universität Marburg,traße 1, D-35043 Marburg,Tel: +49-06421-58-62270; Fax: [email protected] (C. Denkert).

his study was previously presented in part at the ASCO Meeting 2018, Chicago, IL, USA.t authors.34/© 2021 European Society for Medical Oncology. Published by Elsevier Ltd. All rights reserved.

https://doi.org/10.1016/j.annonc.2020.12.016 Volume 32 - Issue 4 - 2021

mailto:[email protected]

http://crossmark.crossref.org/dialog/?doi=10.1016/j.annonc.2020.12.016&domain=pdf

https://doi.org/10.1016/j.annonc.2020.12.016









C. Denkert et al. Annals of Oncology

Vo

Conclusion: The results of this investigation suggest that the clinical behavior of a tumor, in particular, response toneoadjuvant chemotherapy and disease-free survival of therapy-resistant tumors, could be predicted by thecomposition of mutational signatures as an indicator of the individual genomic history of a tumor. After additionalvalidations, mutational signatures might be used to identify tumors with an increased response rate to neoadjuvantchemotherapy and to define therapy-resistant subgroups for future therapeutic interventions.Key words: breast cancer, mutational signatures, neoadjuvant therapy, whole exome sequencing, response, prognosis

INTRODUCTION

Signatures of mutational processes have been described inseveral comprehensive analyses based on whole exome orwhole genome sequencing.1-3 These signatures are theconsequence of the activity of endogenous and exogenousmutational processes acting in combination over long pe-riods of time to generate the specific mutational landscapeof each individual tumor.4

The molecular basis for identification of these mutationalsignatures is the observation that different mutationalprocesses do not act completely randomly, but that eachmutational process generates a specific pattern that can beidentified by sequencing analysis. For example, specificpatterns of mutations caused by UV light5 (with strongprevalence in melanoma) or by tobacco carcinogens6 (withstrong prevalence in tumors of patients with a smokinghistory) have been identified.

The main challenge in identification of mutational sig-natures is that many different processes act over time togenerate an individual pattern in each tumor. The theoret-ical model and computational framework to separate thesedifferent mutational signatures has been established byAlexandrov et al.1,2 More than 30 different types of muta-tional signatures have been identified.7-9 Based on theseresults, it is now possible to identify the contribution ofdifferent mutational processes to the genomic landscapeand to reconstruct the history of each individual tumor.

For breast cancer, the predominant mutational processesinclude mutations induced by pathological activation ofAPOBEC enzymes,10 mutations caused by alterations ofBRCA-related11 or mismatch repair (MMR)-related12 DNArepair pathways and age-related mutational processes.13

The concept of mutational signatures provides a mech-anistic biological framework explaining the complex inter-action of different endogenous and exogenous carcinogenicpathways. The most important next step is the translation ofthis biological concept into clinical applications. Despite theprogress in identifying oncogenic pathways and individualtargetable mutations, the prediction of therapy response insolid tumors is still a major challenge. These tumors developover a very long time by a combination of parallel muta-tional processes; therefore, there is a very strong hypothesisthat the specific pattern of mutational signatures might alsobe relevant for the response of a tumor to a given therapy.Furthermore, these mutational processes do not act only onthe cells of the clinically relevant tumor; they act on all cellsin the patient’s body. The majority of neoplastic cells or

lume 32 - Issue 4 - 2021

premalignant lesions are eliminated by cellular controlprocesses or immunological processes, and progression intoa clinically relevant tumor is a comparably rare event. Weand others have provided evidence that immunologicalparameters are highly relevant for patient prognosis andsuccess of therapy in breast cancer and other types oftumors.14,15

This elimination of premalignant lesions could be seen asa continuous training situation for the immune system, andthis training may be different based on the differentmutagenic processes that act over the lifetime of eachhuman being. The mutational pattern observed in a clini-cally manifest tumor, therefore, may also allow us to lookback into the history of all other premalignant lesions thathave been successfully eliminated.

In this project, we have evaluated the hypothesis thatsignatures of mutational processes that have contributed tothe training of the immune system and other antineoplasticprocesses are relevant for response to neoadjuvantchemotherapy and prognosis in breast cancer. To evaluatethis hypothesis, we have carried out whole exomesequencing of a cohort of 405 pretherapeutic formalin-fixed, paraffin-embedded (FFPE) core biopsies from pa-tients included in the neoadjuvant GeparSepto trial.

We investigated the prevalence of the different muta-tional signatures in different breast cancer subgroups aswell as the role of the most predominant signatures forresponse to neoadjuvant chemotherapy.

PATIENTS AND METHODS

Patients and treatment

In the GeparSepto study (NCT01583426)16,17 women withpreviously untreated, primary invasive breast cancer wereincluded after written informed consent for study partici-pation and biomaterial collection. This translational projectwas restricted to patients with human epidermal growthfactor receptor 2 (HER2)-negative tumors (n ¼ 810). Detailson the clinical trial parameters are given in theSupplementary material, Methods section, available athttps://doi.org/10.1016/j.annonc.2020.12.016.

Objectives and endpoints of the analysis

The current analysis is an exploratory retrospective analysisof a cohort from a prospective clinical trial. The primaryobjective was to investigate mutational signatures andmutational load in the tumor to predict response and

https://doi.org/10.1016/j.annonc.2020.12.016 501




Annals of Oncology C. Denkert et al.

resistance to neoadjuvant therapy. The secondary objectiveswere to correlate mutational signatures with clinicopatho-logical parameters. Predefined endpoints were pathologicalcomplete response (pCR) defined as ypT0ypN0 and disease-free survival (DFS). Predefined covariables for multivariateregression models were age (continuous), tumor stage (T1-2versus T3-4), nodal stage (N0 versus Nþ), Ki-67 (contin-uous), hormone receptor (HR) status (negative versus pos-itive) and treatment (nab-P versus P).

DNA isolation, whole exome sequencing and statisticalanalysis

A total of 703 FFPE core biopsies were selected based onthe availability of sufficient tumor tissue after histopatho-logical quality control and paired blood samples to controlfor germline alterations and single-nucleotide poly-morphisms (SNPs). A total of 703 samples were shipped toNantOmics (Culver City, CA). Automated DNA extractionwas carried out using QIAsymphony DSP DNA Kit (Qiagen,Germantown, MD) on the QIAsymphony. Quantification wasdone using the Invitrogen� Quant-iT� dsDNA Assay Kit,Broad Range (Thermofisher, Waltham, MA) and the BioTekSynergy HTX Multi-Mode Reader (BioTek, Winooski, VT).Samples were sequenced on an Illumina HiSeq sequencer(Illumina, San Diego, CA) according to standard NantOmicsprotocols. Two hundred and eight samples were not suc-cessfully sequenced due to failures during DNA extraction(<50 ng of extracted DNA) or failures during library prep-aration for sequencing which were established by analysisof electropherograms and yield post-library preparation. Ofthe 495 samples that were sequenced, 15 had to beincluded due to provenance fail (mismatch) and 75 due tocontamination fail. Therefore, a total of 298 samples wereexcluded and successful whole exome sequencing wascarried out for 405 samples after filtering for contaminationand tumor/normal mismatches.

Mutational signatures were determined by the bio-informatical team at NantOmics LLC, using the methoddescribed by Alexandrov et al.,1 which was reimplementedin Python via NMF in scikit learn, using the package versionsscikit-learn ¼ 0.14.1, scipy ¼ 0.13.3 and numpy ¼ 1.9.2(https://scikit-learn.org/stable/). The matrix weights wereused from the mutational signature website18 using versionv2 of mutational signatures. In addition, the exonic muta-tion rate (EMR) per Mb was calculated. Using the publishedalgorithm for detection of mutational signatures, we iden-tified the absolute integer number of mutations assigned toeach signature in each tumor. In addition, we determinedthe presence of each mutational signature (with at least onemutation assigned to this signature) as a binary variable.

For a detailed evaluation, we selected only those 11mutational signatures that were relevant for this dataset.The strategy for selection of mutational signature and thedetails of the statistical analysis are described inSupplementary methods, available at https://doi.org/10.1016/j.annonc.2020.12.016.

502 https://doi.org/10.1016/j.annonc.2020.12.016

RESULTS

Baseline clinical data

The set of successfully sequenced patients (n ¼ 405;Supplementary Figure S1, available at https://doi.org/10.1016/j.annonc.2020.12.016, consort diagram) did notsignificantly differ from the HER2-negative GeparSepto pa-tients not sequenced with respect to tumor size, nodalstatus, grading, treatment arm and pCR rate(Supplementary Table S1, available at https://doi.org/10.1016/j.annonc.2020.12.016). In the sequenced cohort,there were fewer HR-negative tumors and fewer invasive-ductal tumors. The patients in the whole exomesequencing group were also slightly younger (mean age 49years versus 50 years in the non-analyzed cohort; P ¼0.027) and had slightly higher levels of stromal tumor-infiltrating lymphocytes (TILs) (mean 26% versus 23% innon-analyzed cohort; P ¼ 0.006).

Frequency of mutational signatures in HR-positive and-negative tumors

In the complete dataset of 405 tumors, the median numberof mutations was 153 [minimum: 26, maximum: 13 072;interquartile range (IQR) 159]. In the subset of 284 HR-positive/HER2-negative tumors, the median number ofmutations was 142 (minimum: 27, maximum: 13 072; IQR135). In the subset of 121 triple-negative tumors, the me-dian number of mutations was 220 (minimum: 26,maximum: 1059; IQR 168).

We evaluated the number of mutations of each signatureand the EMR in HR-positive and HR-negative tumors(Figure 1). Significant differences with higher numbers ofmutations in HR-negative tumors are observed for S3 (ho-mologous recombination deficiency, HRD; P < 0.001), S13(APOBEC; P < 0.001), S6 (MMR; P ¼ 0.015), S21 (MMR; P ¼0.015), S4 (tobacco; P < 0.001) and in total (EMR; P <0.001). Significantly higher mutation numbers in HR-positivetumors are observed for S16 (unknown process; P < 0.001)and S28 (unknown process; P ¼ 0.002). One single tumorhad a particularly high value for the MMR-related signa-tures S15, S21, S26 and a high EMR, which was validated bya reduced immunohistochemical expression of MSH2 andMSH6 indicating microsatellite-instability (MSI; Figure 1B).

Patterns of mutational signaturesdHRD and APOBEC

To give an overview on the patterns of mutational signa-tures, an unsupervised cluster analysis in HR-positive andHR-negative tumors was carried out. In HR-positive tumors,two main clusters and four smaller clusters were observed(Figure 2A and B). One main cluster (C1) was characterizedby the presence of signature S3 (HRD); the other maincluster (C4) showed an absence of S3 and a mixture of othersignatures with comparably high levels of signatures withunknown role (S16, S28). The cluster C2 contained mainlyAPOBEC-related signatures, while the cluster C3 consistedof age-related mutations (S1) with low mutation numbers inmost patients. A very small cluster (C5) included tumors

Volume 32 - Issue 4 - 2021

https://scikit-learn.org/stable/
















Sig

natu

re v

alue

0.3

1

3

10

30

100

300HR+ HR−

S3 S2 S13 S1 S6 S15 S21 S26 S4 S16 S28 EMR

HRD APOBEC Age MMR Tobacco Unknown

P <

0.0

01

P =

0.8

06

P <

0.0

01

P =

0.0

99

P =

0.0

15

P =

0.1

12

P =

0.0

15

P =

0.2

19

P <

0.0

01

P <

0.0

01

P =

0.0

02

P <

0.0

01

Exo

nic

mut

atio

n ra

te (

per

Mb)

013

1030

100300

10003000

10 000

A

BPMS2 MLH1 MSH2 MSH6

Figure 1. Comparison of mutational signatures in HR-positive and HR-negative breast carcinomas.(A) Absolute number of mutations of each signature for HR-negative and HR-positive tumors. In addition, EMR is shown (right part, separate y-axis). P values werecalculated using a two-sample Wilcoxon test. (B) Immunohistochemistry for microsatellite-instability (MSI) of one tumor with high EMR and high rates of MMR-relatedsignatures S15, S21, S26. This tumor is marked with gray boxes in A. A reduced nuclear expression of MSH2 and MSH6 is observed (inserts: positive controls).EMR, exonic mutation rate; HR, hormone receptor; HRD, homologous recombination deficiency; MMR, mismatch repair.


with high levels of the tobacco-related signature S4. Thecluster C6 consisted of a small group of tumors with ahigher contribution of other signatures. In HR-negative tu-mors (Figure 3A and B) the clusters C1-C5 were observed aswell, and the majority of tumors showed S3 (HRD)-relatedmutations (C1).

Mutational signatures, therapy response and prognosis

As expected in breast cancer, the overall pCR rate of pa-tients with HR-negative [triple-negative breast cancer(TNBC)] tumors was significantly higher than the pCR rate ofpatients with HR-positive tumors (38.8% versus 12.7%; P <0.001).

Differences in pCR rate for different mutational signa-tures were observed in HR-positive tumors. For HR-positivetumors, the two main mutational signatures S3 (HRD, P <0.001) and S13 (APOBEC, P ¼ 0.001), evaluated as contin-uous variables, were significantly correlated with anincreased pCR rate in univariable analysis (Figure 2C). Inmultivariable analysis, the correlation was highly significantfor S13 (APOBEC, P ¼ 0.008) and borderline significant forS3 (HRD, P ¼ 0.059, Figure 4A). Tumors with S3 levels abovethe median had a pCR rate of 18%, compared with 8% fortumors with S3 below the median (P ¼ 0.012, Figure 4B).Similarly, tumors with S13 above the median had a pCR rateof 22%, compared with 4% for tumors with low S13 levels(P < 0.001, Figure 4C).


For survival analysis, we focused on the clinically mostrelevant subgroup of therapy-resistant tumors notresponding to the neoadjuvant therapy (non-pCR). In HR-positive tumors without a pCR, the signatures S3 (HRD,P ¼ 0.006) and S4 (tobacco, P ¼ 0.011) were linked toreduced DFS (Figure 2D) in univariable analysis, but not inmultivariable analysis including the covariables age, cT, cN,Ki-67 and treatment (Supplementary Figure S2, available athttps://doi.org/10.1016/j.annonc.2020.12.016). SignatureS6 (MMR) was significant for reduced DFS only in multi-variate analysis (P ¼ 0.038). In KaplaneMeier analysis usingthe median as an exploratory cut point (Figure 2E and F), asignificant difference for S3 (HRD, P ¼ 0.034) and aborderline significant difference for S4 (tobacco, P ¼ 0.053)was observed.

In contrast, in HR-negative tumors there was no muta-tional signature that significantly predicted pCR (uni-variable: Figure 3C; multivariable: Supplementary Figure S3,available at https://doi.org/10.1016/j.annonc.2020.12.016).In therapy-resistant HR-negative tumors, S6 (MMR) had aborderline significance for reduced DFS (univariate: P ¼0.071, Figure 3D; multivariate: P ¼ 0.044, SupplementaryFigure S4, available at https://doi.org/10.1016/j.annonc.2020.12.016), which was borderline significant in KaplaneMeier analysis using the median as a cut point (P ¼0.083, Figure 3E). The univariable and multivariable analysisfor DFS in the complete cohort (including pCR- and non-pCR








DCSignature

S3 − HRDS2 − APOBECS13 − APOBECS1 − AgeS6 − MMRS15 − MMRS21 − MMRS26 − MMRS4 − TobaccoS16 − UnknownS28 − UnknownEMR

ORLow signature superior High signature superior

0.5 0.7 1 1.5 2

OR (95% CI)

1.460.951.550.851.181.091.051.211.000.900.982.01

(1.20-1.78)(0.72-1.25)(1.19-2.03)(0.62-1.16)(0.93-1.49)(0.87-1.37)(0.74-1.50)(0.90-1.63)(0.76-1.31)(0.75-1.08)(0.76-1.27)(1.28-3.14)

P value

<0.0010.7180.0010.3090.1760.4600.7650.1960.9970.2610.8910.002

Signature


0.5 0.7 1 1.5 2

Hazard ratio (95% CI)

0.941.001.020.930.901.081.081.261.020.931.20

1.22 (1.06-1.40)(0.77-1.15)(0.83-1.21)(0.79-1.32)(0.78-1.11)(0.74-1.09)(0.86-1.36)(0.85-1.37)(1.05-1.51)(0.90-1.17)(0.77-1.13)(0.89-1.62)

P value

0.0060.5650.9690.8630.4160.2680.4940.5450.0110.7350.4770.239

C1 – S3 (HRD)

pCR DFS - non-pCR patients

C2 C3 C5C6 C4 – unknown/mixed-S16, S28 (low S3)

APOBEC-related signatures Age-related signatures Tobacco-related signatures

A

B

HRD (S3) APOBEC (S2, S13) Age (S1) MMR (S6, S15, S21, S26)

Tobacco (S4) Unknown (S16) Unknown (S28) Other

HR-positive patients0%

20%

40%

60%

80%

100%HR-positive patients

0

500

1000

1500

2000

Por

tion

of s

igna

ture

(%

)S

igna

ture

val

ue (

cum

ul.)

F

E

0 12 24 36 48 60

0%

20%

40%

60%

80%

100%

P = 0.053

P = 0.034

Eve

nt-

free

su

rviv

al

S3 low S3 high

128120

121109

11295

10086

6552

43

0 12 24 36 48 60

0%

20%

40%

60%

80%

100%

Time (months)

Time (months)

Eve

nt-

free

su

rviv

al

S4 low S4 high

17078

16268

14958

13452

8829

61

Low signature inferior High signature inferiorHazard ratio

Figure 2. HR-positive tumors: patterns of mutational signatures and impact on therapy response as well as prognosis of therapy resistant tumors.(A) Overview on absolute number of mutations of each signature in each sample. (In A, one single HR-positive tumor with extremely high mutation numbers has beenremoved to achieve a comparable scale of the x-axis for comparison with the HR-negative tumors in Figure 3). (B) Overview on the proportion of each signature in eachsample. In A and B, mutational signatures are color-coded based on biological processes and patients are ordered by hierarchical clustering. (C) Mutational signaturesand pCR to neoadjuvant chemotherapy in HR-positive tumors. Univariate logistic regression (n ¼ 284). (D) Mutational signatures and DFS in therapy resistant HR-positive tumors (including only those patients with no pCR). Univariate (n ¼ 248) Cox-regression for DFS in no-pCR patients. (E, F) Univariate KaplaneMeier sur-vival analysis in no-pCR patients for S3 (HRD, E) and S4 (tobacco, E) using the median as a cut point (P: log rank test).APOBEC, apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like; CI, confidence interval; DFS, disease-free survival; HR, hormone receptor; HRD, homologousrecombination deficiency; MMR, mismatch repair; OR, odds ratio; pCR, pathological complete response.


tumors) is shown in Supplementary Figure S5, availableat https://doi.org/10.1016/j.annonc.2020.12.016 for HR-positive tumors and Supplementary Figure S6, availableat https://doi.org/10.1016/j.annonc.2020.12.016. for HR-negative tumors.

EMR, therapy response and prognosis

In addition to the individual signatures, we also investi-gated the EMR as a measurement of total mutationalburden. EMR correlated highly with increased pCR in HR-positive tumors (univariable P ¼ 0.002, Figure 2C; multi-variable P ¼ 0.005, Figure 4A). HR-positive tumors withEMR levels above the median had a pCR rate of 20%,compared with 6% for tumors with EMR below the median(P < 0.001, Figure 4D). EMR did not predict DFS in the HR-positive subcohort (Figure 2D; Supplementary Figures S2


and S5, available at https://doi.org/10.1016/j.annonc.2020.12.016).

In HR-negative tumors, EMR was not significantly asso-ciated with pCR (Figure 3C and Supplementary Figure S3,available at https://doi.org/10.1016/j.annonc.2020.12.016)or DFS (Figure 3D; Supplementary Figures S4 and S6,available at https://doi.org/10.1016/j.annonc.2020.12.016).

Correlation with biological parametersdpatient age,Ki-67, TILs

We evaluated the correlation between the number of mu-tations assigned to the different signatures and patient age,proliferation rate (Ki-67) and stromal TILs in HR-positive andHR-negative tumors (Figure 5). In HR-positive tumors,signature S1 had a highly significant correlation withincreased patient age (P < 0.001, Figure 5A), which

















A

E

DC

B

P = 0.083

pCR DFS - non-pCR patients

C1 – S3 (HRD)C4 mixed C2

APOBEC-related signatures

C3

Age-related

C5

Tobacco-related

HRD (S3) APOBEC (S2, S13) Age (S1) MMR (S6, S15, S21, S26)

Tobacco (S4) Unknown (S16) Unknown (S28) Other

HR-negative patients0

500

1000

1500

2000

Sig

natu

re v

alue

(cu

mul

.)

HR-negative patients 0%

20%

40%

60%

80%

100%

Por

tion

of s

igna

ture

(%

)

0 12 24 36 48 60

0%

20%

40%

60%

80%

100%

Time (months)

Eve

nt-

free

su

rviv

al

S6 low S6 high

4034

3628

3323

3019

2210

10

Signature


0.5 0.7 1 1.5 2

OR (95% CI)

0.990.990.991.050.971.011.000.751.041.111.071.09

(0.82-1.20)(0.71-1.37)(0.72-1.35)(0.75-1.46)(0.75-1.24)(0.78-1.30)(0.68-1.45)(0.45-1.24)(0.81-1.32)(0.92-1.35)(0.82-1.40)(0.60-2.00)

P value

0.9170.9320.9250.7830.7910.9540.9800.2580.7720.2810.6020.776

Signature


0.5 0.7 1 1.5 2

1.021.010.951.300.960.710.791.021.101.090.79

1.01 (0.83-1.22)(0.72-1.44)(0.73-1.38)(0.69-1.31)(0.98-1.73)(0.74-1.26)(0.44-1.13)(0.47-1.33)(0.81-1.30)(0.92-1.33)(0.84-1.41)(0.43-1.46)

P value

0.9580.9310.9710.7510.0710.7840.1460.3840.8480.3010.5380.452

ORLow signature superior High signature superior

Hazard ratio (95% CI)

Low signature inferior High signature inferiorHazard ratio

Figure 3. HR-negative ([TNBC) tumors: patterns of mutational signatures and impact on therapy response as well as prognosis of therapy resistant tumors.(A) Overview on absolute number of mutations of each signature in each sample. (B) Overview on the proportion of each signature in each sample. In A and B,mutational signatures are color-coded based on biological processes and patients are ordered by hierarchical clustering. (C) Mutational signatures and pCR to neo-adjuvant chemotherapy in HR-negative tumors. Univariate logistic regression (n ¼ 121). (D) Mutational signatures and DFS in therapy resistant HR-negative tumors(including only those patients with no pCR). Univariate (n ¼ 74) Cox-regression for DFS in no-pCR patients. (E) Univariate KaplaneMeier survival analysis in no-pCRpatients for S6 (MMR) using the median as a cut point (P: log rank test).APOBEC, apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like; CI, confidence interval; DFS, disease-free survival; HR, hormone receptor; HRD, homologousrecombination deficiency; MMR, mismatch repair; OR, odds ratio; pCR, pathological complete response.


validates S1 as an age-related signature, as described beforein other studies.13

In highly proliferating HR-positive tumors, significantlymore mutations were found for signatures S3 (HRD, P <0.001), S13 (APOBEC, P < 0.001) and S4 (tobacco, P <0.001; Figure 5B). In immunologically active HR-positivetumors with high numbers of TILs, more mutations werefound for S3 (HRD, P < 0.001) and S13 (APOBEC, P < 0.001;Figure 5C).

For HR-negative (TNBC) tumors (Figure 5A and B) therewere no significant correlations of the mutational signa-tures with patient age and proliferation rate; for TILs, therewere significant negative associations with S6 and S21 (bothMMR, P ¼ 0.003 and P ¼ 0.031; Figure 5C).

In Figure 5, we have also evaluated the EMR as aparameter of tumor mutational burden in correlation withage, Ki-67 and TILs. All three correlations were positive andsignificant for HR-positive tumors (Figure 5). In this subtype,significantly higher mutation rates were observed in tumors


with high TILs (P < 0.001), high proliferation (P < 0.001)and in tumors from older patients (P ¼ 0.046). In HR-negative tumors, in contrast, EMR is significantly nega-tively correlated with high TILs (P ¼ 0.005, Figure 5C). ForHR-negative tumors there was also a significant positivecorrelation of EMR with age (P ¼ 0.042).

In addition, we have also evaluated the correlation ofmutational signatures and the categorical clinicopatholog-ical parameters grade, tumor size, nodal status as well asthe therapy arm. The results are shown in SupplementaryTables S2 and S3, available at https://doi.org/10.1016/j.annonc.2020.12.016.

DISCUSSION

Signatures of mutational processes are indicators of multi-ple genetic events that act over years on all cells in thehuman body. Their biological relevance has been investi-gated in large institutional cohorts as well as tissue re-positories of large international consortia. As a next step for






DCB

0 10 20 30

pCR rate (%)

S3 highn = 140

S3 lown = 144 P = 0.012

0 10 20 30

pCR rate (%)

S13 highn = 137

S13 lown = 147 P < 0.001

0 10 20 30

pCR rate (%)

EMR highn = 142

EMR lown = 142 P < 0.001

SignatureA

S3 − HRD

S2 − APOBEC

S13 − APOBEC

S1 − age

S6 − MMR

S15 − MMR

S21 − MMR

S26 − MMR

S4 − tobacco

S16 − unknown

S28 − unknown

EMR

OR

Low signature superior High signature superior

0.5 0.7 1

(pCR)

1.5 2

OR (95% CI)

1.23

1.12

1.53

1.03

1.21

1.17

1.20

1.30

0.83

0.94

1.11

2.17

(0.99-1.52)

(0.81-1.53)

(1.12-2.10)

(0.71-1.50)

(0.91-1.59)

(0.90-1.52)

(0.79-1.82)

(0.91-1.85)

(0.61-1.12)

(0.77-1.16)

(0.83-1.48)

(1.26-3.72)

P value

0.059

0.492

0.008

0.882

0.189

0.237

0.386

0.148

0.216

0.573

0.490

0.005

Figure 4. Pathological complete response to neoadjuvant chemotherapy in HR-positive tumorsdmultivariable analysis and differences in pCR rate.(A) Multivariate logistic regression (n ¼ 274). The covariables in the multivariable models are age, cT, cN, Ki-67, and treatment. (B, C, D) pCR rate for the variables withsignificance in the univariate model using the median as cut point.APOBEC, apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like; CI, confidence interval; EMR, exonic mutation rate; HRD, homologous recombinationdeficiency; MMR, mismatch repair; OR, odds ratio; pCR, pathological complete response.


translation of these findings into the clinical setting, adetailed investigation of the clinical effects of differentmutational signatures on therapy response and survival isnecessary. In this study, we have used a large well-definedclinical trial cohort from a neoadjuvant clinical multicentertrial with known therapy response data and survival data,including HR-positive and HR-negative tumors.

The main result was that defined signatures could predictthe clinical behavior of HR-positive tumors, in particularresponse to neoadjuvant chemotherapy and DFS of non-pCR patients. We could identify mutational signatures S3(HRD) and S13 (APOBEC) that were linked to increased pCRrates to neoadjuvant chemotherapy with high significancein the HR-positive subcohort. The signatures S3 (HRD) andS4 (tobacco) were also associated with a reduced DFS intherapy-resistant (non-pCR) HR-positive tumors.

In addition, we investigated the association of S3 and S13with tumor proliferation and tumor immune infiltrate. Bothsignatures were positively correlated with increased Ki-67and also with increased levels of TILs, which might sug-gest involvement in the development of more aggressivesubtypes of luminal tumors which also have a higherimmunogenicity. These highly proliferating tumors respondbetter to chemotherapy, but may have a reduced prognosis


overall, which is in line with the results of our investigation.The correlation with Ki-67 might also partly explain the non-significance in multivariable analysis.

A multitude of translational investigations have describedTILs as one of the most relevant factors for chemotherapyresponse and prognosis. In our investigation, we haveobserved a possible explanation why some tumors have anaccumulation of TILs. These tumors have been induced bymutational processes involving HRD-like alterations and theaction of APOBEC signatures. Interestingly, there arecomprehensive data from preclinical investigations thatHRD and APOBEC might induce major immunological al-terations. A molecular link between inhibition of PARP inBRCA-mutated tumor and immune activation via STING-pathway activation, which is triggered by cytosolic DNA,has been described for ovarian cancer and TNBC.19,20

APOBEC enzymes are up-regulated by viral infections andphysiologically inhibit retrovirus and transposon replica-tion.21 They also play an important role in mutagenesis indifferent cancer types, including breast cancer.22

In a previous investigation, we have shown that TILsare predictive for increased neoadjuvant response in HR-positive and HR-negative breast cancer, while for survival,TILs are a positive survival factor in TNBC and a negative











−0.4

−0.2

0.0

0.2

0.4

A

B

C

Spe

arm

an c

orre

latio

n

HR+ HR−

S3 S2 S13 S1 S6 S15 S21 S26 S4 S16 S28 EMRHRD APOBEC Age MMR Tobacco Unknown

−0.4

−0.2

0.0

0.2

0.4

Spe

arm

an c

orre

latio

n

HR+ HR−


−0.4

−0.2

0.0

0.2

0.4

Spe

arm

an c

orre

latio

n

HR+ HR−


Figure 5. Correlation of absolute numbers of selected mutational signatures as well as EMR with the biological parameters.(A) Patient age, (B) proliferation rate (Ki-67) and (C) immune cell infiltrate (stromal TILs). The Spearman correlation coefficient is shown (error bars: 95% CI).APOBEC, apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like; EMR, exonic mutation rate; HR, hormone receptor; HRD, homologous recombinationdeficiency; MMR, mismatch repair; TILs, tumor-infiltrating lymphocytes.


survival factor in luminal/HER2-negative breast cancer.14

The differences observed in the present comparison ofmutational signatures between HR-positive and HR-negative tumors might at least partially explain thisdifferent prognostic effect of TILs. In luminal tumors,there is a high diversity of mutational processes and TILsare associated with those processes that are observed inmore aggressive tumors, in particular S3 (HRD). Therefore,TILs are also linked to worse prognosis.

In TNBC, the repertoire of mutational signatures is morehomogeneous, and there is no association with TILs or Ki-67. Therefore, in this subtype, tumors with different TILlevels are not distinguished by different mutational signa-tures. The higher homogeneity of the HR-negative tumorsmight be explained by the faster growth rate of thesetumors and the shorter history of development for eachindividual tumor. For a fast growing tumor, immune escape


might be more relevant compared with a slowly growingtumor, and only those fast growing tumor clones that sur-vive are able to accumulate non-immunogenic mutations.This could explain the observation that high EMR is linkedto reduced TILs in HR-negative (TNBC) tumors, which hasalso been observed in other studies.23

In our study, EMR and TILs, as an indicator of immuno-genicity of a tumor, are positively correlated in HR-positivetumors, but negatively correlated in HR-negative tumors,which is in line with previous studies of TCGA data.24

In HR-positive breast cancer, this correlation is driven bytwo signatures, S3 and S13, which are the predominantsignatures in HR-positive breast cancer with considerableintertumoral heterogeneity, resulting in a positive correla-tion of TILs and EMR and a highly significant association ofEMR and chemotherapy response. In contrast, in HR-negative tumors, EMR is not related to chemotherapy





response, and there is an inverse correlation of EMR andTILs. In contrast, in the GeparNuevo trial we observed anassociation between treatment response and TMB.25

For some of the signatures with a known biologicalfunction, we have been able to perform a validation in ourdataset. We have validated signature S1 by its correlationwith patient age and identified one tumor with MSI by thehigh EMR and the high levels of MMR-related signaturesS15, S21 and S26.

It should be noted that there are relevant limitations ofthis project. As recently discussed,26 the bioinformaticaldetection of mutational signatures is not always straight-forward, and results vary depending on the methods used.In our analysis, some unknown signatures have beendetected, particularly S16 and S28, which are not typical forbreast cancer. This limits the interpretation of this finding,and we cannot exclude that the unexpected high prevalenceof these unknown signatures might be related to artifactsincluded by the algorithms or the biomaterials used. Inparticular, we used small FFPE core biopsies for our study,which might limit the detection of some signatures andmight induce some changes due to paraffin artifacts. Wehave carried out and published a previous analysis withinour own laboratory indicating that limited bias is introducedfrom the FFPE process.27

In our study, we have decided to use a predefined stan-dard algorithm for detection of signatures, to test its rele-vance in clinical trial cohorts. A further optimization of thebioinformatical methods was not the aim of our study.Nevertheless, a further standardization of the detectionalgorithms for clinical applications, as well as validation inother cohorts, is essential for further translation into clinicalpractice.

We would also like to emphasize that mutational signa-tures are quasi-continuous variables and therefore will notresult in distinct, defined tumors types. From Figures 2 and3 it is evident that some patterns can be delineated byclustering algorithms, but these should not be interpretedas new subtypes of breast cancer. For most tumors, themutational signatures are heterogenous, suggesting thatmore than one mutational process is relevant for thedevelopment and progression of each individual tumor.

We would also like to emphasize that while mutationalsignatures in general provide a promising model for thedevelopment of malignant tumors, we cannot show cau-sality in our present investigation. This analysis is based on asample cohort from a clinical trial; therefore, it provides justcorrelative data of different markers, and no mechanisticand causal information. Many of the alterations are signif-icant only in univariate, but not in multivariate analysis,which might be partly explained by the correlation withknown biological parameters, in particular Ki-67. Therefore,these approaches will not replace classical diagnostic ap-proaches that can be conducted without advancedsequencing technology. However, for validation of themutational signatures, it is reassuring that they correlatewith central parameters of tumor biology. Of course, in thisstudy, we cannot draw conclusions of causality, but our


results provide a strong hypothesis for further validation ofgenomic alterations, for example of HRD-related alterations,which have a high prevalence.

In this project, we aimed to look back into the individualhistory of breast carcinomas to identify the processes thatlead to the accumulation of mutations in the developmentof the tumor. The setup of the GeparSepto cohort allowedus to study differences between fast growing triple-negativetumors and luminal/Her2-negative tumors. Our study un-derlines the fact that major differences exist between thesedifferent tumor types. In TNBC, the different mutationalsignatures are not associated with outcome parameterssuch as pCR and DFS, and most mutational signatures arecomparably high in this tumor type. This might be explainedby the fact that TNBC is a rapidly developing disease, andthe rapid onset of the disease leaves not much time inwhich differences between individual tumors coulddevelop. These tumors simply might not have a long historythat could lead to differences in composition of mutationalsignatures.

In contrast, luminal tumors typically develop over severalyears, and it was possible to identify differences betweenindividual tumors that suggest different roles for mutationalprocesses in the development of individual tumors. The twomain signatures involved are S3 (HRD) and S13 (APOBEC).The high number of HRD-associated mutations in subgroupsof luminal tumors, as well as its association with therapyresponse, could be a basis for additional validations, whichmight open new therapeutic options for the HR-deficientsubgroup of luminal tumors.

ACKNOWLEDGEMENTS

We thank all the patients who participated in the clinicalstudy and the translational research and all investigators,pathologists and study personnel at the sites. We would liketo thank Ines Koch, Britta Beyer, Peggy Wolkenstein, Bar-bara Meyer-Bartell and Silvia Handzik from Charité for theirexcellent technical assistance as well as Dr. Bärbel Felderand Stefanie Lettkemann from German Breast Group for thetranslational research organization.

FUNDING

This work was supported by the Translational OncologyProgramme of the German Cancer Aid (Deutsche Kreb-shilfe), project TransLuminal B [grant number 111636] andIntegrate -TN [grant number 70113450]. The sequencinganalysis was carried out and funded by NantOmics (no grantnumber).

DISCLOSURE

CD reports personal fees from Novartis, personal fees andtravel support from Roche, personal fees from MSDOncology, from Daiichi Sankyo, grants from Myriad Geneticsand GBG, other from Sividon Diagnostics/Myriad, outsidethe submitted work. In addition, CD has a patentEP18209672 pending, a patent EP20150702464 pendingand a patent Software (VMscope digital pathology) pending.












MU reports personal fees and non-financial support fromAbbvie, personal fees and non-financial support fromAmgen GmbH, personal fees and non-financial support fromAstraZeneca, personal fees from Bristol Myers Squibb(BMS), personal fees and non-financial support from Cel-gene GmbH, personal fees and non-financial support fromDaiji Sankyo, personal fees and non-financial support fromEisai GmbH, personal fees from Lilly Deutschland, personalfees and non-financial support from Lilly Int., personal feesand non-financial support from MSD Merck, personal feesand non-financial support from Mundipharma, personalfees and non-financial support from Myriad Genetics, per-sonal fees and non-financial support from Odonate, per-sonal fees and non-financial support from Pfizer GmbH,personal fees from PUMA Biotechnology, personal fees andnon-financial support from Roche Pharma AG, personal feesand non-financial support from Sanofi Aventis DeutschlandGmbH, personal fees and non-financial support from TEVAPharmaceuticals Ind Ltd, personal fees and non-financialsupport from Novartis, personal fees from Pierre Fabre,personal fees and non-financial support from ClovisOncology, outside the submitted work. SB reports otherfrom NantOmics, LLC, during the conduct of the study;other from NantOmics, LLC, outside the submitted work;and Employee of NantOmics, LLC with equity interest. ASreports grants from Celgene, grants from Roche, grantsfrom AbbVie, grants from Molecular Partner, personal feesfrom Roche, personal fees from AstraZeneca, personal feesfrom Celgene, personal fees from Roche, personal fees fromCelgene, personal fees from Pfizer, personal fees fromAstraZeneca, personal fees from Novartis, personal feesfrom MSD, personal fees from Tesaro, personal fees fromLilly, other from Roche, outside the submitted work. CJreports personal fees from Roche, personal fees from Cel-gene, personal fees from Amgen, during the conduct of thestudy. TL reports non-financial support from Pharma Mar,non-financial support from Daiichi Sankyo, personal feesand non-financial support from MSD, personal fees fromAmgen, personal fees and non-financial support from Pfizer,personal fees from Novartis, personal fees from Teva, per-sonal fees from Tesaro, personal fees and non-financialsupport from Roche, personal fees and non-financial sup-port from Clovis, non-financial support from Celgene,outside the submitted work.WDS reports grants fromGerman Breast Group, during the conduct of the study;personal fees from AstraZeneca, outside the submittedwork. VM reports grants and personal fees from Roche,during the conduct of the study; personal fees from Amgen,Astra Zeneca, Daiichi-Sankyo, Eisai, Pfizer, MSD, Novartis,Roche, Teva, Seattle Genetics and consultancy honorariafrom Genomic Health, Hexal, Roche, Pierre Fabre, Amgen,ClinSol, Novartis, MSD, Daiichi-Sankyo, Eisai, Lilly, Tesaro,Nektar, personal fees from Genomic Health, Hexal, Roche,Pierre Fabre, Amgen, ClinSol, Novartis, MSD, Daiichi-Sankyo,Eisai, Lilly, Tesaro and Nektar, other from I Novartis, Roche,Seattle Genetics, Genentech, outside the submitted work.PSS is the CEO of NantOmics and reports non-financialsupport from NantOmics, outside the submitted work.


MvM reports honoraria from Roche, Amgen, GenomicHealth, Astra Zeneca, as well as travel support from Lilly andNovartis. PAF reports grants from Novartis, grants fromBioNTech, personal fees from Novartis, personal fees fromRoche, personal fees from Pfizer, personal fees from Cel-gene, personal fees from Daiichi-Sankyo, personal fees fromAstraZeneca, personal fees from Merck Sharp & Dohme,personal fees from Eisai, personal fees from Puma, grantsfrom Cepheid, personal fees from Lilly, personal fees fromSeattle Genetics, during the conduct of the study. SR is theChief Scientific Officer for NantOmics and reports non-financial support from NantOmics, outside the submittedwork. SL reports grants and other from Celgene, grants andother from Roche, during the conduct of the study; grantsand other from Abbvie, grants and other from Amgen,grants and other from AstraZeneca, grants and other fromNovartis, grants and other from Pfizer, other from SeattleGenetics, other from PriME/Medscape, personal fees fromChugai, grants from Teva, grants from Vifor, grants andother from Daiichi-Sankyo, other from Lilly, other fromSamsung, other from Eirgenix, other from BMS, other fromPuma, other from MSD, grants from Immunomedics,outside the submitted work. In addition, SL has a patentEP14153692.0 pending. All other authors have declared noconflicts of interest.

CD reports personal fees from Novartis, personal fees andtravel support from Roche, personal fees from MSDOncology, from Daiichi Sankyo, grants from Myriad Geneticsand GBG, other from Sividon Diagnostics/Myriad, outsidethe submitted work. In addition, CD has a patentEP18209672 pending, a patent EP20150702464 pendingand a patent Software (VMscope digital pathology) pending.MU reports personal fees and non-financial support fromAbbvie, personal fees and non-financial support fromAmgen GmbH, personal fees and non-financial support fromAstraZeneca, personal fees from Bristol Myers Squibb(BMS), personal fees and non-financial support from Cel-gene GmbH, personal fees and non-financial support fromDaiji Sankyo, personal fees and non-financial support fromEisai GmbH, personal fees from Lilly Deutschland, personalfees and non-financial support from Lilly Int., personal feesand non-financial support from MSD Merck, personal feesand non-financial support from Mundipharma, personalfees and non-financial support from Myriad Genetics, per-sonal fees and non-financial support from Odonate, per-sonal fees and non-financial support from Pfizer GmbH,personal fees from PUMA Biotechnology, personal fees andnon-financial support from Roche Pharma AG, personal feesand non-financial support from Sanofi Aventis DeutschlandGmbH, personal fees and non-financial support from TEVAPharmaceuticals Ind Ltd, personal fees and non-financialsupport from Novartis, personal fees from Pierre Fabre,personal fees and non-financial support from ClovisOncology, outside the submitted work. SB reports otherfrom NantOmics, LLC, during the conduct of the study;other from NantOmics, LLC, outside the submitted work;and Employee of NantOmics, LLC with equity interest. ASreports grants from Celgene, grants from Roche, grants





from AbbVie, grants from Molecular Partner, personal feesfrom Roche, personal fees from AstraZeneca, personal feesfrom Celgene, personal fees from Roche, personal fees fromCelgene, personal fees from Pfizer, personal fees fromAstraZeneca, personal fees from Novartis, personal feesfrom MSD, personal fees from Tesaro, personal fees fromLilly, other from Roche, outside the submitted work. CJreports personal fees from Roche, personal fees from Cel-gene, personal fees from Amgen, during the conduct of thestudy. TL reports non-financial support from Pharma Mar,non-financial support from Daiichi Sankyo, personal feesand non-financial support from MSD, personal fees fromAmgen, personal fees and non-financial support from Pfizer,personal fees from Novartis, personal fees from Teva, per-sonal fees from Tesaro, personal fees and non-financialsupport from Roche, personal fees and non-financial sup-port from Clovis, non-financial support from Celgene,outside the submitted work.WDS reports grants fromGerman Breast Group, during the conduct of the study;personal fees from AstraZeneca, outside the submittedwork. VM reports grants and personal fees from Roche,during the conduct of the study; personal fees from Amgen,Astra Zeneca, Daiichi-Sankyo, Eisai, Pfizer, MSD, Novartis,Roche, Teva, Seattle Genetics and consultancy honorariafrom Genomic Health, Hexal, Roche, Pierre Fabre, Amgen,ClinSol, Novartis, MSD, Daiichi-Sankyo, Eisai, Lilly, Tesaro,Nektar, personal fees from Genomic Health, Hexal, Roche,Pierre Fabre, Amgen, ClinSol, Novartis, MSD, Daiichi-Sankyo,Eisai, Lilly, Tesaro and Nektar, other from I Novartis, Roche,Seattle Genetics, Genentech, outside the submitted work.PSS is the CEO of NantOmics and reports non-financialsupport from NantOmics, outside the submitted work.MvM reports honoraria from Roche, Amgen, GenomicHealth, Astra Zeneca, as well as travel support from Lilly andNovartis. PAF reports grants from Novartis, grants fromBioNTech, personal fees from Novartis, personal fees fromRoche, personal fees from Pfizer, personal fees from Cel-gene, personal fees from Daiichi-Sankyo, personal fees fromAstraZeneca, personal fees from Merck Sharp & Dohme,personal fees from Eisai, personal fees from Puma, grantsfrom Cepheid, personal fees from Lilly, personal fees fromSeattle Genetics, during the conduct of the study. SR is theChief Scientific Officer for NantOmics and reports non-financial support from NantOmics, outside the submittedwork. SL reports grants and other from Celgene, grants andother from Roche, during the conduct of the study; grantsand other from Abbvie, grants and other from Amgen,grants and other from AstraZeneca, grants and other fromNovartis, grants and other from Pfizer, other from SeattleGenetics, other from PriME/Medscape, personal fees fromChugai, grants from Teva, grants from Vifor, grants andother from Daiichi-Sankyo, other from Lilly, other fromSamsung, other from Eirgenix, other from BMS, other fromPuma, other from MSD, grants from Immunomedics,outside the submitted work. In addition, SL has a patentEP14153692.0 pending. All other authors have declared noconflicts of interest.


REFERENCES

1. Alexandrov LB, Nik-Zainal S, Wedge DC, et al. Signatures of mutationalprocesses in human cancer. Nature. 2013;500(7463):415-421.

2. Nik-Zainal S, Davies H, Staaf J, et al. Landscape of somatic mutations in560 breast cancer whole-genome sequences. Nature. 2016;534(7605):47-54.

3. Alexandrov LB, Kim J, Haradhvala NJ, et al. The repertoire of mutationalsignatures in human cancer. Nature. 2020;578(7793):94-101.

4. Harris RS. Cancer mutation signatures, DNA damage mechanisms, andpotential clinical implications. Genome Med. 2013;5(9):87.

5. Hayward NK, Wilmott JS, Waddell N, et al. Whole-genome land-scapes of major melanoma subtypes. Nature. 2017;545(7653):175-180.

6. Alexandrov LB, Ju YS, Haase K, et al. Mutational signatures associatedwith tobacco smoking in human cancer. Science. 2016;354(6312):618-622.

7. Alexandrov LB. Understanding the origins of human cancer. Science.2015;350(6265):1175.

8. Alexandrov LB, Nik-Zainal S, Wedge DC, Campbell PJ, Stratton MR.Deciphering signatures of mutational processes operative in humancancer. Cell Rep. 2013;3(1):246-259.

9. Alexandrov LB, Stratton MR. Mutational signatures: the patterns ofsomatic mutations hidden in cancer genomes. Curr Opin Genet Dev.2014;24:52-60.

10. Nik-Zainal S, Alexandrov LB, Wedge DC, et al. Mutational processesmolding the genomes of 21 breast cancers. Cell. 2012;149(5):979-993.

11. Davies H, Glodzik D, Morganella S, et al. HRDetect is a predictor ofBRCA1 and BRCA2 deficiency based on mutational signatures. NatMed. 2017;23(4):517-525.

12. Davies H, Morganella S, Purdie CA, et al. Whole-genome sequencingreveals breast cancers with mismatch repair deficiency. Cancer Res.2017;77(18):4755-4762.

13. Alexandrov LB, Jones PH, Wedge DC, et al. Clock-like mutational pro-cesses in human somatic cells. Nat Genet. 2015;47(12):1402-1407.

14. Denkert C, von Minckwitz G, Darb-Esfahani S, et al. Tumour-infiltratinglymphocytes and prognosis in different subtypes of breast cancer: apooled analysis of 3771 patients treated with neoadjuvant therapy.Lancet Oncol. 2018;19(1):40-50.

15. Savas P, Salgado R, Denkert C, et al. Clinical relevance of host immunityin breast cancer: from TILs to the clinic. Nat Rev Clin Oncol. 2016;13(4):228-241.

16. Untch M, Jackisch C, Schneeweiss A, et al. NAB-paclitaxel improvesdisease-free survival in early breast cancer: GBG 69-GeparSepto. J ClinOncol. 2019;37(25):2226-2234.

17. Untch M, Jackisch C, Schneeweiss A, et al. Nab-paclitaxel versussolvent-based paclitaxel in neoadjuvant chemotherapy for early breastcancer (GeparSepto-GBG 69): a randomised, phase 3 trial. LancetOncol. 2016;17(3):345-356.

18. Available at: https://cancer.sanger.ac.uk/cosmic/signatures_v2.Accessed June 12, 2020.

19. Pantelidou C, Sonzogni O, De Oliveria Taveira M, et al. PARP inhibitorefficacy depends on CD8(þ) T-cell recruitment via intratumoral STINGpathway activation in BRCA-deficient models of triple-negative breastcancer. Cancer Discov. 2019;9(6):722-737.

20. Ding L, Kim HJ, Wang Q, et al. PARP inhibition elicits STING-dependentantitumor immunity in Brca1-deficient ovarian cancer. Cell Rep.2018;25(11):2972-2980.e5.

21. Venkatesan S, Rosenthal R, Kanu N, et al. Perspective: APOBEC muta-genesis in drug resistance and immune escape in HIV and cancerevolution. Ann Oncol. 2018;29(3):563-572.

22. Roberts SA, Lawrence MS, Klimczak LJ, et al. An APOBEC cytidinedeaminase mutagenesis pattern is widespread in human cancers. NatGenet. 2013;45(9):970-976.

23. Karn T, Jiang T, Hatzis C, et al. Association between genomic metricsand immune infiltration in triple-negative breast cancer. JAMA Oncol.2017;3(12):1707-1711.


http://refhub.elsevier.com/S0923-7534(20)43224-7/sref1
















































https://cancer.sanger.ac.uk/cosmic/signatures_v2




























24. Safonov A, Jiang T, Bianchini G, et al. Immune gene expression isassociated with genomic aberrations in breast cancer. Cancer Res.2017;77(12):3317-3324.

25. Karn T, Denkert C, Weber KE, et al. Tumor mutational burden andimmune infiltration as independent predictors of response to neo-adjuvant immune checkpoint inhibition in early TNBC in GeparNuevo.Ann Oncol. 2020;31:1216-1222.


26. Nik-Zainal S, Morganella S. Mutational signatures in breast cancer:the problem at the DNA level. Clin Cancer Res. 2017;23(11):2617-2629.

27. Newton Y, Sedgewick AJ, Cisneros L, et al. Large scale, robust,and accurate whole transcriptome profiling from clinicalformalin-fixed paraffin-embedded samples. Sci Rep. 2020;10(1):17597.


















1

Appendix A

Supplementary material – methods section

Patients and Treatment

In the GeparSepto study (NCT01583426)1,2 women with previously untreated, primary invasive

breast cancer were included after written informed consent for study participation and

biomaterial collection. The relevant authorities and ethics committees approved the studies.

The clinical trial including the informed consent for biomarker analysis was approved by the

Ethics committee of the state of Berlin (dated 02.05.2012). The translational research was

approved by the Ethics committee of the Charité University Hospital Berlin (EA1/139/05;

Amendment 01.03.2018, Approval 29.03.2018). The REMARK criteria were followed.3 Patients

were randomized to either weekly nab-paclitaxel (nab-P) (Celgene, Germany) or solvent-

based paclitaxel (P) for 12 weeks followed by 4 cycles of conventionally-dosed

epirubicine/cyclophosphamide (EC). Patient inclusion criteria and treatment information have

been published.2 Pretherapeutic FFPE core biopsies were prospectively collected in the GBG

biobank. HER2 status (positive if IHC3+ or SISH ratio >2.0) and HR status (ER/PR positive if

³1% stained cells) were centrally assessed prior to randomization. To reduce the number of

different tumor subtypes, this translational project was restricted to patients with HER2-

negative tumors (n=810). Ki-67 as well as PMS2, MLH1, MSH2 and MSH6 were evaluated by

standard immunohistochemistry, stromal TILs were evaluated as previously described.4

DNA isolation and whole exome/whole genome sequencing

A total of 703 FFPE core biopsies were selected based on the availability of sufficient tumor

tissue after histopathological quality control and paired blood samples to control for germline-

alterations and SNPs. A total of 703 samples were shipped to Nantomics (Culver City, CA,

USA). Automated DNA extraction was performed using QIAsymphony DSP DNA Kit on the

QIAsymphony (Qiagen, Germantown, MD, USA). Quantification was done using the

Invitrogen™ Quant-iT™ dsDNA Assay Kit, Broad Range (Thermofisher, Waltham, MA, USA)

and the BioTek Synergy HTX Multi-Mode Reader (BioTek, Winooski, VT, USA). Samples were

sequenced on an Illumina HiSeq sequencer according to standard NantOmics protocols.

The following quality checks and quality criteria were used: After DNA isolation post extraction

yield had to exceed 50ng. Post-library preparation yields and electropherogram analysis were

performed to assure libraries were of sufficient quality. Sequencing flowcell yields, error rates,

alignment rates and overall Q30 were evaluated to assure sequencing was of sufficient quality.

Post-duplicate marking coverage of both tumor and normal were quantified, as well as the

coverage of 75%, 90% and 99% of known cancer-associated genes. Contamination detection

2

from external samples, a calculation of homozygosity of chromosome X, as well as a

provenance match of tumor and normal samples were performed to assure samples were

appropriately matched prior to downstream analysis.

208 samples were not successfully sequenced due to failures during DNA extraction (< 50ng

of extracted DNA) or failures during library preparation for sequencing which were established

by analysis of electropherogram and yield post-library preparation. Of the 495 samples that

were sequenced, 15 had to be included due to provenance fail (mismatch) and 75 due to

contamination fail. Therefore, a total of 298 samples were excluded and successful whole-

exome sequencing was performed for 405 samples after filtering for contamination and

tumor/normal mismatches.

From this clinical trial cohort, fresh frozen samples were unavailable for the patients in this

study for direct comparisons. However, we have performed and published previous analysis

within our own laboratory that indicate that limited bias is introduced from the FFPE process.5

Additionally, informatic techniques during the somatic calling exercise were performed to

identify potential errors that may be introduced into individual fragments and reduce false

positive variant calls.

Mutational signatures were determined by the bioinformatical team at NantOmics LLC, using

the method described by AlexandrovFehler! Textmarke nicht definiert., which was reimplemented in

Python via NMF in scikit learn6, using the package versions scikit-learn==0.14.1,

scipy==0.13.3 and numpy==1.9.2. The matrix weights were used from the mutational signature

website7 using version v2 of mutational signatures. In addition, the exonic mutation rate per

Mb was calculated (EMR). Using the published algorithm for detection of mutational

signatures, we identified the absolute integer number of mutations assigned to each signature

in each tumor. In addition, we determined the presence of each mutational signature (with at

least one mutation assigned to this signature) as a binary variable. For a detailed evaluation,

we selected only those mutational signatures that were relevant for this dataset.

Statistical analysis – predefined parameters and additional posthoc analyses

The project is based on a predefined statistical analysis plan. Predefined endpoints were

pathological complete response (pCR) defined as ypT0ypN0 and disease-free survival (DFS).

Predefined covariables for multivariate regression models were age (continuous), tumor stage

(T1-2 vs. T3-4), nodal stage (N0 vs. N+), Ki-67 (continuous), hormone receptor status

(negative vs. positive), and treatment (nab-P vs P). Mutational signatures were evaluated as

continuous variables and as binary variables. Associations between mutational signatures,

clinico-pathological characteristics and pathological complete response (pCR) rate were

investigated with Fisher´s exact tests (binary variables), χ2 tests (non-binary categorical

variables), Wilcoxon tests (binary vs continuous variable), and Spearman correlations

3

(continuous variables). The endpoint pCR was analyzed based on logistic regression models:

Odds ratios (ORs) with confidence intervals (CIs) and Wald p-values are presented. For the

endpoint DFS Cox regression models were constructed and hazard ratios (HRs) with CIs and

Wald p-values are reported. For regression models signature variables were transformed as

log(. +1) and variable EMR was transformed as log(. ); ORs and HRs refer to these

transformed variables. Interaction tests were reported as p-values from respective regression

models. All p-values are two-sided, with a p-value<5% considered to be statistically significant.

No correction for multiple testing was applied. All CIs correspond to 95%. Statistical analysis

was performed using R 3.3.2 (R Core Team, 2016). While the majority of endpoints and

clinicopathological covariables were predefined, it was necessary to adapt the analysis of the

individual mutational signatures to exclude signatures that were present only in small

subgroups or at very low levels (Supplemental table A.1). It was not possible to predefine these

analyses, since the distribution of the signatures was not known before the analysis.

Strategy for selection of mutational signatures for detailed analysis

The known causes of biological processes that generate the specific patterns of mutational

signatures were derived from the Sanger Center website,7 additional information on mutational

signatures relevant for breast cancer were derived from previous publications.Fehler! Textmarke nicht

definiert.,Fehler! Textmarke nicht definiert. During the course of the statistical analysis, an updated and

revised list of mutational signatures was published on the Sanger Center website; for this

project we used the previous signature list v2, which was predefined in our statistical analysis

plan.

Signatures for detailed analysis were selected based on prevalence, median and biological

function. An overview on the parameter used for selection of signatures is given in

Supplemental table A.1.

As main focus, we included signatures with known biological background in breast cancer:

age-related signatures (S1 and S5), APOBEC-related signatures (S2 and S13), a BRCA/HRD

related signature (S3), mismatch-repair related (MMR) signatures (S6, S15, S20, S21, and

S26), and a tobacco carcinogen-related signature (S4). From these signatures, we excluded

S5 (present in only 5.2% of tumors) and S20 (present in 15.1%, but very low absolute levels).

As a result of this selection process, we identified a set of 11 signatures for detailed evaluation

and correlation with the predefined clinical endpoints and clinicopathological parameters.

In this analysis of the prevalence of different signatures, it became evident that major

differences exist between HR-negative and HR-positive tumors which was not clear at the time

the statistical analysis plan was finalized. Therefore, for this manuscript, we made the post-

4

hoc decision to present the main statistical analyses including the correlation with clinical

factors as well as pCR and survival was performed separately for these two tumor subtypes.

Table A.1: Exploratory analysis and overview on distribution and prevalence of the

different mutational signatures in the complete cohort – selection for statistical analysis

Biological process

Mutational

Signature

Prevalence (%

of tumors)

Median absolute levels of tumors with at least one

mutation present* (interquartile range Q1-

Q3; maximum)

Include in statistical analysis

Justification

Defective HR S3 61.5 44 (15-102;406) include High levels APOBEC-rel. S2 44.4 6 (2-13;525) include High prevalence APOBEC-rel. S13 87.7 12 (5-26;392) include High prevalence

Clock-like S1 94.8 26 (15-41;258) include High prevalence Clock-like S5 5.2 8 (2-12;50) exclude Low prevalence MMR def. S6 69.9 16 (9-28;1299) include High prevalence MMR def. S15 45.7 10 (5-23;1263) include High prevalence MMR def. S20 15.1 5 (2-9;74) exclude low levels and prev. MMR def. S21 27.9 5 (3-9;7013) include High prevalence MMR def. S26 19.0 6 (4-11;2676) include High levels unknown S16 60.2 29 (14-64;1084) include High prevalence

Tobacco carc. S4 36.5 14 (5-26;120) include High prevalence UV-light S7 56.3 6 (3-11;143) exclude Low levels Aflatoxin S24 27.7 7 (3-12;66) exclude low levels

Tobacco chew. S29 14.3 4 (2-8;26) exclude low levels Polymerase n S9 6.4 4 (2-9;13) exclude low levels POLE polym. S10 26.2 3 (2-8;85) exclude low levels Alkyl. agents S11 22.5 4 (2-8;23) exclude low levels

Aristochol. acid S22 32.6 2 (1-5;16) exclude low levels unknown S8 1.5 4 (1-5;13) exclude Low prevalence unknown S12 14.8 6 (3-9;1187) exclude low levels and prev. unknown S14 2.5 3 (2-5;12) exclude low levels and prev. unknown S17 21.2 3 (2-6;1198) exclude low levels and prev. unknown S18 10.4 4 (2-7;20) exclude low levels unknown S19 14.1 6 (3-12;49) exclude low levels unknown S23 17.8 4 (2-6;17) exclude low levels unknown S25 2.7 6 (4-8;11) exclude low levels and prev. unknown S27 10.1 1 (1-3;27) exclude low levels unknown S28 60.7 8 (3-17;555) include High prev. / range unknown S30 8.4 9 (4-15;27) exclude low levels and prev.

* In this exploratory analysis, tumors without any mutation for a given signature were excluded from the calculation of median, interquartile range and maximum. These values are representing only those tumors in which the signature is present. As a result, the median values are higher than in figure 1, where all tumors are included.

5

References

1 Untch M, Jackisch C, Schneeweiss A, Schmatloch S, Aktas B, Denkert C, Schem C, Wiebringhaus H, Kümmel S, Warm M, Fasching PA, Just M, Hanusch C, Hackmann J, Blohmer JU, Rhiem K, Schmitt WD, Furlanetto J, Gerber B, Huober J, Nekljudova V, von Minckwitz G, Loibl S. NAB-Paclitaxel Improves Disease-Free Survival in Early Breast Cancer: GBG 69-GeparSepto. J Clin Oncol. 2019 Sep 1;37(25):2226-2234 2 Untch M, Jackisch C, Schneeweiss A, Conrad B, Aktas B, Denkert C, Eidtmann H, Wiebringhaus H, Kümmel S, Hilfrich J, Warm M, Paepke S, Just M, Hanusch C, Hackmann J, Blohmer JU, Clemens M, Darb-Esfahani S, Schmitt WD, Dan Costa S, Gerber B, Engels K, Nekljudova V, Loibl S, von Minckwitz G; German Breast Group (GBG); Arbeitsgemeinschaft Gynäkologische Onkologie—Breast (AGO-B) Investigators. Nab-paclitaxel versus solvent-based paclitaxel in neoadjuvant chemotherapy for early breast cancer (GeparSepto-GBG 69): a randomised, phase 3 trial. Lancet Oncol. 2016 Mar;17(3):345-356. doi: 10.1016/S1470-2045(15)00542-2. 3 McShane LM, Altman DG, Sauerbrei W, et al. Reporting recommendations for tumor marker prognostic studies (REMARK). J Natl Cancer Inst. 2005;97:1180-84. 4 Salgado R, Denkert C, Demaria S, Sirtaine N, Klauschen F, Pruneri G, Wienert S, Van den Eynden G, Baehner FL, Penault-Llorca F, Perez EA, Thompson EA, Symmans WF, Richardson AL, Brock J, Criscitiello C, Bailey H, Ignatiadis M, Floris G, Sparano J, Kos Z, Nielsen T, Rimm DL, Allison KH, Reis-Filho JS, Loibl S, Sotiriou C, Viale G, Badve S, Adams S, Willard-Gallo K, Loi S; International TILs Working Group 2014. The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Ann Oncol. 2015 Feb;26(2):259-71. doi: 10.1093/annonc/mdu450. 5 Newton Y, Sedgewick AJ, Cisneros L, Golovato J, Johnson M, Szeto CW, Rabizadeh S, Sanborn JZ, Benz SC, Vaske C. Large scale, robust, and accurate whole transcriptome profiling from clinical formalin-fixed paraffin-embedded samples. Sci Rep. 2020 Oct 19;10(1):17597. doi: 10.1038/s41598-020-74483-1. PMID: 33077815; PMCID: PMC7572424. 6 https://scikit-learn.org/stable/ accessed 1.12.2019 7 https://cancer.sanger.ac.uk/cosmic/signatures_v2, accessed at June, 12th, 2020

Appendix B -Supplementary tables

Supplementary table S1 – Clinicopathological characteristics of the cohort analyzed by WES – comparison with the complete cohort of patients with HER2-negative tumors treated

in the GeparSepto trial

Variable Category not analyzed analyzed Overall p

all 405 (100.0%) 405 (100.0%) 810 (100.0%)

T T1 132 (33.0%) 134 (33.3%) 266 (33.1%) 0.2958

T2 210 (52.5%) 217 (53.8%) 427 (53.2%)

T3 28 (7.0%) 34 (8.4%) 62 (7.7%)

T4 30 (7.5%) 18 (4.5%) 48 (6.0%)

missing 5 2 7

N N0 253 (63.4%) 261 (66.1%) 514 (64.7%) 0.4579

N+ 146 (36.6%) 134 (33.9%) 280 (35.3%)

missing 6 10 16

ER negative 175 (43.2%) 142 (35.1%) 317 (39.1%) 0.0212

positive 230 (56.8%) 263 (64.9%) 493 (60.9%)

PR negative 213 (52.6%) 181 (44.7%) 394 (48.6%) 0.0292

positive 192 (47.4%) 224 (55.3%) 416 (51.4%)

HR negative 155 (38.3%) 121 (29.9%) 276 (34.1%) 0.0144

positive 250 (61.7%) 284 (70.1%) 534 (65.9%)

HER2 negative 405 (100.0%) 405 (100.0%) 810 (100.0%)

positive 0 (0.0%) 0 (0.0%) 0 (0.0%)

subtype TNBC 155 (38.3%) 121 (29.9%) 276 (34.1%) 0.0144

luminal 250 (61.7%) 284 (70.1%) 534 (65.9%)

HistoType Ductal invasive (NST) 354 (87.4%) 324 (80.0%) 678 (83.7%) 0.0020

Lobular invasive 23 (5.7%) 22 (5.4%) 45 (5.6%)

other 28 (6.9%) 59 (14.6%) 87 (10.7%)

G G1-2 178 (44.0%) 174 (43.0%) 352 (43.5%) 0.8316

G3 227 (56.0%) 231 (57.0%) 458 (56.5%)

treatment pac 199 (49.1%) 204 (50.4%) 403 (49.8%) 0.7787

nab-pac 206 (50.9%) 201 (49.6%) 407 (50.2%)

ypT0_ypN0 no pCR 310 (76.5%) 322 (79.5%) 632 (78.0%) 0.3506

pCR 95 (23.5%) 83 (20.5%) 178 (22.0%)

Supplementary Table S2 – Luminal/Her2-negative tumors - Correlation of mutational signatures and clinicopathological parameters. Each clinicopathological variable defines two subgroups. The distribution of each (absolute) mutational signature and EMR is compared between

these subgroups: The median value for each subgroup and the p-value from a two-sample Wilcoxon test are shown. Row “n” contains the number of patients in each subgroup. Results are presented for HR-positive and HR-negative subtypes separately.

Grade Tumor stage Nodal status Histology Therapy

category G1-2 G3 cT1-2 cT3-4 cN0 cN+ Ductal

(NST) lobular/other paclitaxel

nab-

paclitaxel

n 144 140 238 44 170 106 228 56 138 146

median median p-value median median p-value median median p-value median median p-value median median p-value

S3(HRD) 0 20 <.0001 2 5 0.8951 2 4 0.8909 2 6 0.6029 6 1 0.3095

S2(APOBEC) 0 0 0.9309 0 2 0.1867 0 0 0.2539 0 2 0.0220 0 0 0.5551

S13(APOBEC) 5 11 0.0004 8 9 0.5200 7 9 0.4408 8 7 0.5106 9 7 0.1486

S1(age) 24 24 0.3210 24 25 0.9952 24 26 0.3239 24 27 0.2313 22 25 0.2985

S6(MMR) 6 12 0.0024 7 9 0.2898 10 3 0.0343 9 6 0.0757 6 10 0.1422

S15(MMR) 2 0 0.5345 0 3 0.2952 0 0 0.8833 1 0 0.1355 0 0 0.9430

S21(MMR) 0 0 0.6465 0 0 0.1854 0 0 0.1516 0 0 0.0723 0 0 0.2337

S26(MMR) 0 0 0.7551 0 0 0.3322 0 0 0.6695 0 0 0.5662 0 0 0.0841

S4(tobacco) 0 0 0.0595 0 0 0.0405 0 0 0.2952 0 0 0.7169 0 0 0.2124

S16(unknown) 16 14 0.6928 14 18 0.9664 16 12 0.4722 18 5 0.0378 14 16 0.6982

S28(unknown) 4 3 0.3365 3 6 0.1107 4 3 0.8872 4 1 0.0144 3 4 0.8108

EMR 1.3 2.2 <.0001 1.7 1.9 0.1950 1.7 1.6 0.9505 1.7 1.7 0.9580 1.8 1.7 0.4738

Supplementary table S3 – TNBC tumors - Correlation of mutational signatures and clinicopathological parameters Each clinicopathological variable defines two subgroups. The distribution of each (absolute) mutational signature and EMR is compared between

these subgroups: The median value for each subgroup and the p-value from a two-sample Wilcoxon test are shown. Row “n” contains the number of patients in each subgroup. Results are presented for HR-positive and HR-negative subtypes separately.

Grade Tumor stage Nodal status Histology Therapy

category G1-2 G3 cT1-2 cT3-4 cN0 cN+ Ductal

(NST) other* paclitaxel

nab-

paclitaxel

n 144 140 238 44 170 106 228 56 138 146

median median p-value median median p-value median median p-value median median p-value median median p-value

S3(HRD) 65 54 0.7588 56 103 0.7816 54 96 0.5654 54 69 0.6097 58 33 0.5156

S2(APOBEC) 0 0 0.6484 0 2 0.7626 0 0 0.6195 0 1 0.6442 0 0 0.5967

S13(APOBEC) 12 16 0.5683 14 32 0.0194 15 16 0.8484 16 14 0.8552 15 15 0.6674

S1(age) 25 30 0.8382 29 34 0.3585 29 28 0.9775 28 30 0.5345 28 31 0.9233

S6(MMR) 14 14 0.3793 14 12 0.9122 14 14 0.7814 14 10 0.6427 14 13 0.7216

S15(MMR) 0 0 0.6692 0 8 0.2396 0 0 0.1743 0 0 0.8240 0 0 0.3723

S21(MMR) 1 0 0.0326 0 0 0.1519 0 0 0.1450 0 0 0.9225 0 0 0.4690

S26(MMR) 0 0 0.7222 0 0 0.4838 0 0 0.7495 0 0 1.0000 0 0 0.4372

S4(tobacco) 0 1 0.7342 1 1 0.8238 0 0 0.9732 0 1 0.7455 0 1 0.9225

S16(unknown) 0 2 0.0173 0 6 0.5068 0 4 0.8326 0 7 0.2462 0 0 0.7298

S28(unknown) 0 1 0.1049 0 2 0.6360 0 0 0.9839 1 0 0.7351 1 0 0.8197

EMR 2.8 2.8 0.9808 2.7 4.1 0.0753 2.7 2.9 0.8657 2.7 3.2 0.4052 2.8 2.8 0.7528

*: There are no lobular HR-negative tumors in the analysis set.

GeparSepto breast cancer clinical trial cohort (n=810, HER2 neg. patients*)

Induction of mutations by different processes over months/years

HRD APOBEC aging MSI tobacco other

HER2neg tumors (ERpos or ERneg) (n=703) for NGS analysis

Successfull whole exome sequencing (n=405)HRneg (=TNBC): n=121; HRpos (=lum/HER2neg): n=284

30 mutational signatures11 signatures selected based on prevalence and level

Correlation with biological parameters (n=405)

Patient age

Proliferation (Ki-67)

Immune activ. (TILs)

Correlation with outcome (n=405)

Pathol. completeresponse

(pCR; ypT0ypN0)

Disease-freesurvival (DFS)

Hormone receptor status

Supplementary figure S.1: Analysis of mutational signatures in tumor samples from theGeparSepto multicenter trial - overview on clinical trial samples, consort statement and projectoutline (*the therapy of the HER2pos tumors (n=396) is not shown here, as these tumors werenot included in current sequencing study)

tumors without suitable tumor tissue orwithout paired blood sample

excluded (n=107)

WES sequencing not successful (total n=298)208: low DNA content

15: provenance fail75: contamination fail

Core

bio

psy

(bef

ore

stud

y en

try) Arm A

Arm B

R

Surgery

1:1 12 weeks 12 weeks

Paclitaxel80 mg/m2

weekly

Nab-paclitaxel 150 mg/m2

weekly; reduced to 125 mg/m2 after 464 pts

Epirubicin 90 mg/m2

Cyclophosphamide 600 mg/m2

Appendix C – supplementary figures

Denkert et al. Supplementary Figure S.1


Supplementary Figure S.2: Mutational signatures and disease-free survival (DFS) in therapyresistant HR-positive tumors (with no pCR). Multivariable analysis. Multivariate Cox-regressionfor DFS (n=239) using the covariables age, cT, cN, Ki67, and treatment in no pCR patients.

(DFS)


Supplementary Figure S.3: Mutational signatures and pathological complete response (pCR)to neoadjuvant chemotherapy in HR-negative tumors – multivariable analysis (n=119). Thecovariables in the multivariable models are age, cT, cN, Ki67, and treatment. The OR refers tothe transformed variables, see methods for details.

(pCR)

Denkert et al. Supplemental Figure S.4

Supplementary Figure S.4: Mutational signatures and disease-free survival (DFS) in therapyresistant triple-negative tumors (patients with no pCR) – multivariate analysis (n=73) Cox-regression. The covariables in the multivariable models are age, cT, cN, Ki67, and treatment.

(DFS)


Supplementary Figure S.5: Mutational signatures and disease-free survival (DFS) in all HR-positive tumors (n=284). a, Univariate (n=284) Cox-regression for DFS. b, Multivariable Cox-regression for DFS (n=274) using the covariables age, cT, cN, Ki67, and treatment. c, d, UnivariateKaplan-Meier survival analysis for S3 (HRD, c) and S4 (tobacco, d) using the median as a cutpoint(p: log rank test).

a

b

p=0.063 p=0.021

(DFS)

(DFS)

disease-free

survival

disease-free

survival

c d


Supplementary Figure S.6: Mutational signatures and disease-free survival (DFS) in all triple-negative tumors (n=121). a,b Univariate (a, n=121) and multivariable (b, n=119) Cox-regression.The covariables in the multivariable models are age, cT, cN, Ki67, and treatment. c, UnivariateKaplan-Meier survival analysis for S6 (MMR) using the median as a cutpoint (p: log rank test).

a

b

c

p=0.050

(DFS)

(DFS)

disease-free

survival

reconstructing tumor history in breast cancer: signatures

Documents