Immunohistochemical and molecular imaging biomarker signature for the prediction of failure site after chemoradiation for head and neck squamous cell carcinoma

Abstract

Objective: To identify a failure site-specific prognostic model by combining immunohistochemistry (IHC) and molecular imaging information to predict long-term failure type in squamous cell carcinoma of the head and neck.

Patient and methods: Tissue microarray blocks of 196 head and neck squamous cell carcinoma cases were stained for a panel of biomarkers using IHC. Gross tumor volume (GTV) from the PET/CT radiation treatment planning CT scan, maximal Standard Uptake Value (SUVmax) of fludeoxyglucose (FDG) and clinical information were included in the model building using Cox proportional hazards models, stratified for p16 status in oropharyngeal carcinomas. Separate models were built for time to locoregional failure and time to distant metastasis.

Results: Higher than median p53 expression on IHC tended toward a risk factor for locoregional failure but was protective for distant metastasis, χ² for difference p = .003. The final model for locoregional failure included p53 (HR: 1.9; p: .055), concomitant cisplatin (HR: 0.41; p: .008), β-tubulin-1 (HR: 1.8; p: .08), β-tubulin-2 (HR: 0.49; p: .057) and SUVmax (HR: 2.1; p: .046). The final model for distant metastasis included p53 (HR: 0.23; p: .025), Bcl-2 (HR: 2.6; p: .08), SUVmax (HR: 3.5; p: .095) and GTV (HR: 1.7; p: .063).

Conclusions: The models successfully distinguished between risk of locoregional failure and risk of distant metastasis, which is important information for clinical decision-making. High p53 expression has opposite prognostic effects for the two endpoints; increasing risk of locoregional failure, but decreasing the risk of metastatic failure, but external validation of this finding is needed.

Introduction

The emergence of human papilloma virus (HPV) related oropharyngeal carcinomas as a new, prognostically favorable [1–4] entity among head and neck squamous cell carcinoma (HNSCC) has led to a flurry of treatment de-escalation trials of HPV + patients [5]. As a result, p16 as a surrogate for HPV infection may become the first example of a predictive biomarker guiding management decisions in HNSCC. It has also been shown that genetic and immunohistochemical (IHC) biomarker profiles differ according to HPV status [6,7]. However, despite intensive research, there is a paucity of useful biomarkers in the non-HPV-related cases, etiologically linked with a history of heavy tobacco and alcohol use [8].

Prognostic models in the literature often focus on composite endpoints, such as disease-free survival or time to any failure [9–12]. There is a twofold downside of using such composite endpoints; first, combining distant and locoregional failure (LRF) in one single analysis may result in the loss of information about the biology of the disease. Second, it is of substantial interest to know if the patient's risk is dominated by distant metastasis (DM) or LRF in the clinical management of HNSCC and the pooling of endpoints in composite analysis prohibits the extraction of such information. We have previously hypothesized [13] that some biomarkers may be associated with a specific type of failure in HNSCC. This is potentially of interest because it could lead to individualization of treatment based on each patient’s failure risk.

In the current analysis, we focus on combining IHC assays and molecular imaging assays to build separate prognostications for DM and LRF. As a preplanned substudy, we intended to validate a previously published prognostic model on survival [14] on which the selection of IHC biomarkers were based rather than on a biologic rationale.

Material and methods

This is a retrospective cohort study (n = 287) of consecutive Danish adults diagnosed and receiving (chemo-) radiotherapy for previously untreated, locally advanced HNSCC in a single center according to national guidelines made by the Danish Head and Neck Group (http://www.dahanca.dk). The population has previously been described in detail [15–17].

The inclusion criteria were as follows: (1) diagnosis and intensity-modulated radiotherapy treatment for HNSCC between January 2005 and October 2009; (2) sufficient amount of diagnostic biopsy material to include in a tissue microarray (TMA); (3) a pretreatment 18F-FDG PET/CT scan performed as part of radiotherapy planning.

The study was approved by both the Capital Regional Committee on Health Research Ethics (ID: H-2-2013-122) and the Data Protection Authority (Journal no. 30-1129). The Danish registry for ‘Tissue Usage’ had no records of any of the included patients opposing to their tissue being used for research purposes.

Tissue microarray and immunohistochemistry

Pre-treatment diagnostic hematoxylin–eosin (H&E) stained sections were reevaluated for relevant and representative tumor tissue by one of the authors (GBR) and validated by an experienced pathologist with head and neck cancer sub specialization (MHT). TMA blocks were constructed from 2 mm core punch outs from each patient’s pretreatment, formalin-fixed paraffin-embedded (FFPE) diagnostic tumor biopsy blocks, stored since diagnosis at room temperature in the archive of the Department of Pathology. Due to sparseness of diagnostic tissue in many of the diagnostic biopsies, multiple coring was not possible. The part of the cohort (n = 98) collected after a previously published pilot series (n = 102) [15] was stained at the Department of Pathology, Rigshospitalet, Copenhagen, Denmark, except for glutathione-S-transferase-π (GST-π) and β-tubulin-1 and β-tubulin-2 which were stained manually at the Greenebaum Comprehensive Cancer Center, Baltimore, MD.

Sections of 4–5 µm were cut from the TMA blocks and mounted on positively charged glass slides, air-dried overnight and incubated at 60 °C for 60 min (Thermo Scientific section dryer, Waltham, MA, USA). Fully automated IHC stains were performed according to standard guidelines of the pathology departments and according to the protocols of the manufacturers of the primary antibodies including staining of relevant control tissues (tonsil, colon, pancreas and liver, lung) in the same batch. For the 102 patients described in the pilot series [15], fully automated IHC stains were performed on ‘Leica Bond III’ instruments staining for Bcl-2 (clone bcl-2/100/D5, Leica Biosystems, Newcastle, UK), p53 (clone DO-7, Leica Biosystems), and p16 (clone E6H4, Ventana Medical Systems, Inc. Tucson, AZ). Staining for EGFR, (ready to use Clone 5B7, Ventana Medical Systems) and Ki-67 (clone MIB1, DAKO, Glostrup, Denmark) were performed on the fully automated BenchMark ULTRA IHC/ISH Staining Module (Ventana Medical Systems).

The remaining 98 patients were stained using BenchMark ULTRA IHC/ISH Staining Module, using sections of the TMAs cut at four micron thickness, mounted on Dako FLEX IHC microscope slides. Staining was performed using the following clones: Bcl-2 (clone 124, DAKO), p53 (clone DO-7, DAKO), p16 (clone E6H4, Ventana Medical Systems), EGFR (ready-to-use Clone 5B7, Ventana Medical Systems), Ki-67 (clone MIB1, DAKO). Relevant control tissues, tonsil, colon, pancreas, liver and lung for the staining of p53, Ki-67, Bcl-2, p16, were costained in each batch. Both the GST-π and the β-tubulin stains had internal controls in the HNSCC tissue. Negative controls were performed by omitting the primary antibody.

Manual IHC was performed for the GST-π (clone LW29, Leica Biosystems, Newcastle, UK) and β-tubulin-1 and β-tubulin-2 (clone JDR.3B8, Santa Cruz Biotechnology, Inc., Dallas, TX) stains as previously described [15].

The first of the consecutive TMA sections from each TMA block was stained with H&E to ensure the presence of relevant and sufficient amount of tumor tissue in each core.

The quantification of the immunostained TMA sections were performed following D.C. Allred [18]. The Allred scoring consists of a proportion score (0–5; 0: 0% positive tumor cells; 1: up to 1% positive tumor cells; 2: up to 10% positive tumor cells; 3: up to 33% positive tumor cells, 4: up to 66% positive tumor cells; 5: up to 100% positive tumor cells) which is added to an intensity score (0–3; 0: no staining of tumor cells; 1: weak; 2: moderate; 3: intense) for the total score of (0; 2–8). IHC stains were scored by a single observer (GBR), supervised by another author (MHT). Difficult cases were discussed and consensus was reached using a double-headed microscope. The clinical patient data and the treatment outcome were blinded to GBR and MHT until the IHC scoring had been completed.

For statistical analysis, IHC variables were dichotomized according to the population median of the total Allred score (except for p16, which was scored according to EORTC/DAHANCA guidelines [19], regardless of tumor location). In short, to be classified as p16 positive the staining must be strong and uniform in more than 70% of carcinoma cells in both nucleus and cytoplasm [19].

During the IHC-staining process, some patients’ TMA cores were lost, fragmented, used up cutting through the block or otherwise unable to being scored. If more than two of a patient’s IHC stains were non-evaluable, the patient was excluded from the current analysis.

FDG PET/CT imaging and delineation

As part of the radiotherapy planning, integrated 18F-FDG PET/CT scans were performed on fasting patients as described recently [17]. Immobilized patients in individually molded masks were scanned one hour after the injection of the 18F-FDG tracer in accordance with the department’s guidelines. Intravenous contrast was given unless contraindicated. The standard uptake values (SUV) were calculated to correct for bodyweight and injected activity. PET/CT scans were used for delineation. A multibed PET scan was performed after the CT, covering the area from vertex to mid-thigh.

Delineation of the 18F-FDG PET positive area was performed by a nuclear medicine physician; the gross tumor volume (GTV) from the CT scan was delineated by a radiologist in collaboration with an oncologist.

Statistics

An attempt to perform an external validation of a previously published IHC-based model [14] was made. The methods used in the original publication were applied to the present dataset. The endpoint in this analysis was overall survival. Quantification of IHC was performed as stated in the original paper.

Multivariable failure-type specific Cox models were built using following candidate predictors and coding: Concomitant cisplatin prescribed (no = 0; yes = 1), tobacco use (never smoker: 1; former smoker: 2; current smoker: 3), the natural logarithm of the GTV measured continuously in cubic centimeters (GTV (ln)) and the maximal SUV (SUVmax (ln)) (continuous, natural logarithm of SUVmax)). IHC candidate predictors were dichotomized stains of Bcl-2, p53, GST-π; cytoplasmic β-tubulin-1; nuclear β-tubulin-2, EGFr, Ki-67 as described above. Patients were stratified between p16-positive oropharyngeal squamous cell carcinoma (p16 + OPSCC) versus others. The ‘other’ stratum includes the patients with p16-negative OPSCC and all patients with SCC in subsites other than oropharynx. p16 status was used as stratification leading to p16 status-specific baseline hazards, but no HR for p16 positivity versus p16 negativity (proportionality not assumed).

Multivariable model reduction was performed by backwards elimination using the Wald score statistics, retaining parameters with a significance level of p < .15. Two-year patient-specific Kaplan–Meier (KM) estimates of time to locoregional failure (TLRF)/time to distant metastasis (TDM) were calculated based on the reduced multivariate models. A predefined analysis plan described codings, backwards elimination procedure and stratification.

LRF was defined as failure in either the original tumor site or in the regional lymph nodes with censoring for DM or death. LRF or death was treated as censorings in analysis of TDM. Patients with synchronous LRF and DM failure were included as events in both models in accordance with a predefined analysis plan. Two-year KM estimates for both TLRF and TDM of the reduced multivariate models were calculated.

SPSS version 22 (IBM Armonk, NY, USA) was used for statistical analysis, supplemented with R version 3.3.0 and R-studio version 0.99.896 (Boston, MA, USA).

Results

The characteristics of the included patients are shown in Table 1. A total of 287 patients were eligible for the collection of diagnostic biopsies. Eighty-four patients were excluded due to scarcity of tumor material in the primary diagnostic biopsies. A total of 203 patients' samples were included in the TMA analysis. Another seven patients (4 from the pilot series, 3 from the present series) were excluded from analyses after IHC staining because of missing stains due to scarcity of actual tumor tissue in the TMA sections, or due to cores falling off the slide during the staining process. A total of 196 patients were included in the final model building analyses (Figure 1). Intensity of some of the included IHC-stains (p53 and Bcl-2) was missing in one patient, and this patient was excluded from the validation attempt and the p53-based patient characteristics. Median follow-up in censored patients was 32 months.

Figure 1. Patient flowchart: IMRT: intensity-modulated radiotherapy; FFPE: formalin-fixed paraffin-embedded; TMA: tissue microarray; IHC: immunohistochemistry.

Table 1. Patient characteristics based on p53 expression.

		High p53 expression		Low p53 expression
Covariates		No.	%	No.	%
No. of patients		91	47	104	53
Primary localization
	Nasopharynx	11	50	11	50
	Oral cavity	11	61	7	39
	Oropharynx	30	34	57	66
	Hypopharynx	22	52	20	48
	Larynx	17	65	9	35
Histological subtype
	Well-differentiated SCC	5	56	4	44
	Moderately differentiated SCC	39	49	41	51
	Poorly differentiated SCC	20	67	10	33
	Nonkeratinizing OPSCC	18	33	36	67
	Nasopharyngeal carcinoma	6	46	7	54
	Other and missing classification	3	33	6	67
p16-positive OPSCC
	p16-positive OPSCC	19	33	39	67
	Other	72	53	65	47
Gender
	Female	17	39	27	61
	Male	74	49	77	51
Mean age at RT-start	(Range)	60.1	35–83	58.4	26–83
Tobacco
	Never smoker	9	45	11	55
	Former smoker	27	42	37	58
	Current smoker	55	50	56	50
Mean SUVmax (ln)	(Range)	2.55	1.19–3.53	2.52	0.83–3.21
Mean GTV (ln)	(Range)	3.72	0.94-7.05	3.94	0.93-5.98
Concomitant cisplatin
	Cisplatin	47	44	61	56
	No cisplatin	44	51	43	49
Immunohistochemistry
	β-tubulin-1 high expression	50	49	52	51
	β-tubulin-1 low expression	41	44	52	56
	β-tubulin-2 high expression	42	51	41	49
	β-tubulin-2 low expression	49	44	63	56
	GST-π high expression	34	43	46	58
	GST-π low expression	57	50	58	50
	Ki-67 high expression	49	45	60	55
	Ki-67 low expression	39	48	43	52
	Bcl-2 high expression	39	41	57	59
	Bcl-2 low expression	52	53	47	47
	EGFr high expression	50	50	50	50
	EGFr low expression	39	42	53	58
N-stage
	N0	8	42	11	58
	N1-N2b	31	44	40	56
	N2c-N3 (Incl. N2 nasopharynx)	52	50	53	50
T-stage
	T-stage 1	13	52	12	48
	T-stage 2	28	44	36	56
	T-stage 3	32	49	34	52
	T stage 4	18	45	22	55
Failure site
	No failure	65	47	74	53
	Locoregional failure	23	59	16	51
	Distant metastatic failure	3	18	14	82

The ?Other and missing classification? histology subgroup included basaloid SCC (n = 1); transitional cell carcinoma (n = 1) hybrid type SCC (n = 2) all with low p53 expression. Five patients had missing classification (n = 3 with high p53 expression; n = 2 with low p53 expression).

The attempted validation of the IHC-based model performed by Cullen et al [14] is shown in Figure 2. The prognostication from the Cullen model could not be reproduced.

Figure 2. Kaplan–Meier curves of overall survival by number of positive IHC biomarkers. Sum of positive IHC-stains as performed in the reference [14].

Failure site-specific cox models

The failure site-specific Cox models are shown in Table 2. The models for the two failure types are different, suggesting a possibility to distinguish between locoregional and distant failure. Most importantly, high expression of p53 is found of borderline significance for increased risk of LRF (HR = 1.9, 95% CI [0.99; 3.77]) but significant for decreased risk of DM (HR = 0.23, 95% CI [0.065; 0.83]) with a χ² p value of p = .003 for differential effect, cf. Figure 3. This p value remains significant after Bonferroni correction for the 11 comparisons. Cumulative incidence curves (1-KM) for high/low p53 expression in the two strata are shown in Figure 4.

Figure 3. Forest plot of hazard ratios for locoregional failure (closed circles, full line, top) versus distant metastatic failure (open circles, broken line, bottom) of each of the investigated covariables in the current study. Data are from the reduced models presented in Table 2 when included, or from the full model if not included in the reduced model (marked with dagger and gray color on the forest plot). χ² p values for differential effect on the two endpoints are given, the p value of p .003 for p53 is significant after correcting for multiple comparisons using the Bonferroni method.

Figure 4. Cumulative incidence curves (1-KM) for locoregional failure and distant metastasis in the two strata (p16-positive oropharyngeal cancers and ‘others’). The ‘other’ stratum includes the patients with p16-negative OPSCC and all patients with SCC in subsites other than oropharynx.

Table 2. Failure site-specific Cox models.

	Full multivariate model				Reduced multivariate model
Covariates	HR	p value	95% CI		HR	p value	95% CI
Time to locoregional failure
Tobacco	1.45	.23	0.79	2.66	–	–	–	–
Concomitant cisplatin yes versus no	0.39	.009	0.19	0.79	0.41	.008	0.21	0.79
β-Tubulin-1 (cytoplasmic)	1.92	.08	0.92	4.00	1.84	.084	0.92	3.67
β-Tubulin-2 (nuclear)	0.54	.11	0.26	1.15	0.49	.057	0.24	1.02
GST-π	1.08	.83	0.53	2.19	–	–	–	–
Ki-67	0.79	.53	0.38	1.66	–	–	–	–
Bcl-2	1.30	.50	0.61	2.80	–	–	–	–
EGFr	1.04	.92	0.50	2.17	–	–	–	–
p53	1.73	.13	0.85	3.53	1.93	.055	0.99	3.77
GTV (ln)	1.00	.98	0.70	1.45	–	–	–	–
SUVmax (ln)	2.65	.036	1.06	6.60	2.14	.046	1.01	4.53
Time to distant metastatic failure
Tobacco	1.70	.24	0.71	4.10	–	–	–	–
Concomitant cisplatin yes vs. no	0.61	.34	0.22	1.68	–	–	–	–
β-Tubulin-1 (cytoplasmic)	0.90	.85	0.31	2.61	–	–	–	–
β-Tubulin-2 (nuclear)	0.64	.45	0.20	2.02	–	–	–	–
GST-π	2.17	.15	0.75	6.31	–	–	–	–
Ki-67	1.54	.46	0.49	4.81	–	–	–	–
Bcl-2	2.67	.11	0.81	8.80	2.61	.083	0.88	7.70
EGFr	0.48	.18	0.16	1.42	–	–	–	–
p53	0.21	.029	0.05	0.85	0.23	.025	0.07	0.83
GTV (ln)	1.43	.23	0.80	2.56	1.71	.063	0.97	3.01
SUVmax (ln)	5.32	.048	1.02	27.83	3.47	.095	0.81	14.94

Full and reduced multivariate models of time to locoregional failure and time to distant metastasis, stratified for patient status of p16 positive OPSCC. Tobacco was included as a linear variable: 1: never smoker, 2: former smoker, 3: current smoker. IHC stains included as dichotomized variables: high and low expression, split on the median of each stain. Model building was performed as backwards elimination a.m. Wald, including factors at p < .15. Significant values ( p < .05) are shown in bold.

A visual presentation of individual risk estimates based on the two-year KM estimates of the site-specific reduced multivariate models is shown in Figure 5. The suggested possibility to distinguish LRF and DM is reflected in the dispersion in distribution. Figure 5(B,C) show predicted versus observed cumulative incidence of the two outcomes.

Figure 5. (A) Cumulative incidence (1-KM) plot of estimates for 2-year risk of DM or LRF from multivariate reduced models. Each patient had risk estimates of the two failure types plotted against each other. The gray lines represent the medians of the respective 1-KM estimates. (B) and (C) calibration plots of observed versus predicted cumulative incidence (1 minus two-year KM estimate) of risk of the two failure types. The patients were grouped in quintiles based on their predicted two-year estimates from the reduced multivariate models.

Discussion

The main purpose of this study was to combine imaging and IHC biomarkers in an attempt to distinguish between risk of LRF and risk of DM. A failure-specific model would be more clinically useful than prognostication in itself as it would allow selection of patients for the most promising experimental treatment strategies – that is, intensified local therapy for patients with high risk of LRF and intensified systemic therapy for patients with a high risk of DM. Also, we hypothesized that failure-type specific models may add biological insights compared to composite endpoints.

Our main worry at the onset of the project was the risk of not being able to distinguish between the two types of failures, if patients with a high risk of LRF would generally also be at high risk of DM. Figure 5 shows that our model does indeed identify patients at a high risk of LRF and low risk of DM and vice versa as a result of the selective risk profiling of p53 and other markers, cf. Figures 3 and 4. This is indeed promising for future research moving away from pure prognostic models and into models capable of distinguishing between the risk of different failure types for individual patients, thus allowing a tailored package of locoregional and systemic therapy in the individual patient. However, the limited sample size – in particular the low number of distant failure events – and the lack of external validation should be acknowledged as limitations.

The main driver of the failure site prediction was p53 status, indicating a higher risk of LRF and lower risk of DM in the current dataset. While it should be acknowledged that the dataset is limited and external validation would be required to draw firm conclusions, the result is intriguing. When comparing with the literature, studies distinguishing between LRF and DM are rare [13] and none of the previously investigated biomarkers had an as distinct different prognostic value on locoregional versus distant failures as p53 in this study, albeit direct comparison is difficult as model building, staining procedures and quantifications of IHC stains are different. Quite a few studies have focused on p53 and with some discrepancy in conclusions; immunohistochemical expression of p53 has previously been shown to predict locoregional failure in the head and neck area [20,21]. Koch et al. [20] found an increased risk of locoregional failure, but no effect of TP53 mutation on the overall survival. Ma et al. [21] found that mutated p53 gene, but not overexpression of the protein, could be a useful prediction tool for local tumor recurrences. In other words, the literature suggests that p53 overexpression may be correlated to higher risk of LRF, but to our knowledge, this study is the first to associate high expression of p53 with reduced risk of DM, as found in the present material.

Traditionally, p53 expression has been associated with SCC typically located in the larynx or hypopharynx and characterized by heavy use of tobacco and alcohol [22,23] and with no HPV infection. This is in accordance with the demographics of Table 1 where it is seen that the p53 high expression and p53 low expression patient groups differ, most importantly with fewer patients with high p53 expression in p16 + OPSCC-group. Note, however, that the analyses in the present study are stratified by p16 status and so the effect of p16 + in OPSCC is already accounted for in the analyses.

In the analyses, tobacco use was found to increase the risk of both outcomes, but not significantly so. Other studies have found smoking status significantly prognostic [1]. Most likely, the lack of significant association in the study is due to the limited power – it should be noted that the point estimate of the full model in Table 2 is in good agreement with the literature data.

The data also allowed for a preplanned external validation of a previously published prognostic model [14]. Such external validations are much needed and called for by several important papers [24,25]. Unfortunately, the previously published model [14] did not provide good prognostication of overall survival in our series. The differences in treatment between the studies should definitely be acknowledged; most importantly the patients in the present study did not receive induction chemotherapy. This could be one explanation of the discrepancy. Another difference that might explain some of the discrepancy is the use of a different IHC clone used to stain for GST-π. In the present analysis, clone LW29 (Leica Biosystems) was used, whereas in the original model clone, 353-10 was used (Dako, Carpinetria, CA, USA). Due to the original manufacturers cessation of production of the original clone, this could not be obtained. Also, in our cohort, 22 nasopharyngeal patients were included – a site not included in the Cullen paper. Nevertheless, the result is disappointing, as a strong prognostic model should be robust across different patient series in order to be clinically useful. The Cullen prognostic model has not been externally validated. Well described challenges with this kind of modeling are overfitting of included variables.

The limitations of the present study should be recognized. The sample size is limited and in particular the number of metastatic failures is modest. Nevertheless the present study demonstrates a method to distinguish between competing types of failure of utmost clinical relevance and the data suggest that such separation between risks is indeed possible with commonly available IHC techniques. Furthermore, the separate analysis of risk of locoregional failure and distant metastasis revealed intriguing biological information and warrants a deeper analysis, using for example DNA sequencing to elucidate the underlying reasons for the different recurrence pattern according to p53 status.

Acknowledgments

The authors group are grateful to histotechnician Pernille Frederiksen, Rigshospitalet, histotechnicians accredited by the American Society of Clinical Pathology (ASCP) Kimberly Tuttle and Loretta Kendall, UMSOM, for helping with the IHC staining. The departments of pathology in the Capital Region of Denmark (Rigshospitalet, Herlev, Hvidovre), and in Region Zealand (Roskilde, Slagelse, Naestved) are acknowledged for helping with retrieval of the diagnostic FFPE material.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

Arvid Nilsson Foundation

Freddy Johansen’s Foundation

Aase and Ejnar Danielsen’s Foundation

Danish Cancer Research Foundation10.13039/501100005940

National Institutes of Health10.13039/100000002P30 CA 134274-04Region Hovedstaden, Global Excelence AwardAase og Ejnar Danielsens Fond10.13039/501100003035Dansk Kræftforsknings Fond10.13039/501100005940Kirsten og Freddy Johansens Fond10.13039/501100007505

This work was supported by grants from the Arvid Nilsson Foundation, the ‘Global Excellence’ award from the Capital Region of Denmark, Kirsten and Freddy Johansen’s Foundation, Aase and Ejnar Danielsen’s Foundation, Danish Cancer Research Foundation (Dansk Kraeftforskningsfond), as well as the National Institutes of Health, NIH-grant-[P30 CA 134274-04].