Low Transforming Growth Factor-β Pathway Activity in Cervical Adenocarcinomas

Cervical cancer is the fourth most common cancer in women worldwide. Squamous cell carcinoma (SCC) and adenocarcinoma (AC) are the most common histological types, with AC patients having worse prognosis. Over the last two decades, incidence rates of AC have increased, highlighting the importance of further understanding AC tumorigenesis, and the need to investigate new treatment options. The cytokine TGF-β functions as a tumour suppressor in healthy tissue. However, in tumour cells this suppressive function can be overcome. Therefore there is an increasing interest in using TGF-β inhibitors in the treatment of cancer. Here, we hypothesize that TGF-β plays a different role in SCC and AC. Analysis of RNA-seq data from the TCGA, using a TGF-β response signature, resulted in separate clustering of the two subtypes. We further investigated the expression of TGF-β-signalling related proteins (TβR1/2, SMAD4, pSMAD2, PAI-1, αvβ6 and MMP2/9) in a cohort of 62 AC patients. Low TβR2 and SMAD4 expression was associated with worse survival in AC patients and interestingly, high PAI-1 and αvβ6 expression was also correlated with worse survival. Similar correlations of TβR2, PAI-1 and αvβ6 with clinical parameters were found in previously reported SCC analyses. However, when comparing expression levels between SCC and AC patient samples, pSMAD2, SMAD4, PAI-1 and αvβ6 showed lower expression in AC compared to SCC. Because of the low expression of core TβR1/2, (p-)SMAD2 and SMAD4 proteins and the correlation with worse prognosis, TGF-β pathway most likely leads to tumour inhibitory effects in AC and therefore the use of TGF-β inhibitors would not be recommended. However, given the correlation of PAI-1 and αvβ6 with poor prognosis, the use of TGF- β inhibitors might be of interest in SCC and in the subsets of AC patients with high expression of these TGF-β associated proteins.


INTRODUCTION
Cervical cancer is the fourth most common cancer in women worldwide (1). It is caused by a persistent infection with high risk human papillomavirus (HPV) (2). Tumour progression is facilitated by the failing control of the host's immune response, combined with several immune escape mechanisms of the virally infected cells (3).
Transforming growth factor-b (TGF-b) is a multifunctional cytokine that functions as a tumour suppressor in healthy cells, by being a potent inducer of growth arrest and apoptosis, maintaining tissue (microenvironment) homeostasis (22). In malignancy, tumour cells can overcome the cytostatic effects of TGF-b signalling, mostly by acquiring mutations in the canonical signalling components (eliminating the tumour suppressor function) or by adaptation of TGF-b signalling through non-canonical signalling (leading to tumour promoting effects) (22)(23)(24). Accordingly, interest in TGF-b counteracting immunotherapies is gaining momentum (25), (26).
In normal tissue homeostasis, TGF-b is secreted into the extracellular matrix as part of a latent complex. TGF-b activators, such as integrins and proteases, release TGF-b from its latent state. Once activated, TGF-b initiates signalling by binding TGF-b receptor 2 (TbR2), which then forms a heteromeric complex with TGFb receptor 1 (TbR1). TbR2 phosphorylates TbR1, which subsequently phosphorylates Sma and Mad related protein 2 (SMAD2) or SMAD3. These SMADs form heteromeric complexes with common mediator SMAD4, allowing for translocation into the nucleus. There, they interact with various transcription factors to regulate transcription of genes such as plasminogen activator inhibitor 1 (PAI-1), SMAD7, matrix metalloproteinases (MMP's), collagen type I, and fibronectin (27,28).
The TGF-b signalling pathway has been investigated predominantly in cervical cancer cell lines and in SCC histology (29)(30)(31)(32)(33)(34)(35). Previous studies of our group found that TGF-b1 expression in SCC patients was strongly associated with tumour infiltration within surrounding stroma and thus poor outcome, and TGF-b1 expression positively correlated with PAI-1 expression (31,33). PAI-1 expression levels correlated with poor overall and disease free survival, while high SMAD4 expression was related to low infiltration depth (33,34). A potential difference in TGF-b signalling between SCC and AC subtypes was suggested in these studies. However, too few AC patients were included to allow for a meaningful comparison.
Research concerning TGF-b signalling in cervical AC is scarce. With the rising incidence rates of cervical AC, it is important to further distinguish the biological and immunological processes that drive the tumorigenesis of this histological subtype, as this knowledge will contribute to the development of more precise, tumour-specific treatment approaches. Here, we hypothesised that the TGF-b pathway plays a different role in cervical AC compared to SCC (19,33).
The aim of this study was to investigate the TGF-b pathway in cervical AC. We first analysed RNA-seq data from the TCGA, using a TGF-b response signature. Subsequently, we systematically investigated the protein expression of core canonical pathway members (SMAD4, pSMAD2, TbR1 and TbR2) and TGF-b pathway regulators and transcriptional targets (PAI-1, alpha-v beta-6 integrin (avb6 integrin), MMP2 and MMP9), and determined associations between protein expression and clinicopathological parameters in a welldefined, consecutive cohort of cervical AC patients.

Patients and Ethical Statement
This is a retrospective, single-centre-cohort study. All human tissues were used according to the Code of Conduct for responsible use of human tissues in the context of health research 2011 (https://www.bbmri.nl/sites/bbmri/files/styles/ Federa_code_of_conduct_english.pdf). (All women included, were diagnosed with cervical AC, International Federation of Gynaecology and Obstetrics (FIGO) stage 1b-2a, and underwent a radical hysterectomy with lymphadenectomy as primary treatment at the Leiden University Medical Center (LUMC) between January 1990 and December 2005. Patients of whom sufficient representative tumour material, preserved in formalinfixed, paraffin-embedded tissue (FFPE) blocks, was available in the archives of the department of Pathology of the LUMC were included. Clinical files were reviewed to collect data including age, FIGO stage, tumour diameter, invasion depth, lymphvascular space infiltration (LVSI), parametrial involvement, tumour positivity of the surgical resection margins, tumour positivity of the pelvic lymph nodes, adjuvant treatment, disease recurrence and survival.
Conventional histology sections were stained with haematoxylin and eosin, and reviewed by an experienced pathologist to select only the 'usual type' endocervical ACs, according to the World Health Organization (WHO) classification of tumours (4). All tumours were typed for HPV, previously (18).
Immunohistochemistry was performed and evaluated essentially as described before (29,31,32,36). See Supplementary Methods for further details. To compare AC and SCC, immunohistochemistry data from SCC performed in previous studies was used, in which immunohistochemistry and evaluation was essentially as described in this study (29,32,33,37). As a correction for comparing immunohistochemistry stainings, 3-5 SCC cases were taken along in the current AC immunohistochemistry staining.

Bioinformatical Analysis
RNA-seq and clinical data of AC and SCC were obtained from TCGA. The core set described in (19), was used for bioinformatical analysis. The most optimal number of groups within the TCGA data, was based on the silhouette value where a higher value indicates a better underlying number of groups. Overall similarities between AC and SCC, were visualized based on multidimensional scaling and the heatmap was based on scaled expression data and coloured by AC or SCC status.

Statistics
Statistical analysis was performed with IBM Statistical Package for the Social Sciences (SPSS) Statistics (version 23, IBM corp., Chicago, IL, USA). Data were processed using the Chi-square test or Fisher's exact test for categorical variables, Student's t-test for parametric continuous variables, or one-way analysis of variance for numerical data when comparing more than two groups. Correlation was tested by the Spearman rho correlation coefficient in non-parametric data and the Pearson correlation in cases of normality. Kaplan Meier survival curves and the Log Rank test were used to determine between-group differences in disease-specific survival (DSS) and recurrence-free survival (RFS). Multivariate analysis was performed using a Cox proportional hazard model with stepwise regression. For all reported tests, P values were two-sided and P values <0.05 were considered to indicate statistical significance.

TCGA Analyses
To determine if the TGF-b pathway plays a different role in AC versus SCC, expression data from TCGA on cervical cancer were analyzed (19). In this dataset, 144 squamous cell carcinomas and 31 adenocarcinomas were present. We determined the expression of a 153-gene TGF-b response signature (38) (Supplementary Table 2) in every tumor, and used these data to analyze the clustering of AC and SCC patients. The optimal amount of clusters in this data was two, based on the silhouette value ( Figure 1A). Multidimensional scaling ( Figure 1B) showed the clustering into two major subclasses; subclass A, presenting mostly AC samples (yellow), and subclass B, presenting mostly SCC patients (dark blue). Unsupervised hierarchical clustering of both the patient samples and signature genes, resulted in a heatmap confirming clustering of the SCC samples (dark gray) separately from the AC samples (light gray) ( Figure 1C). As AC and SCC clustered mostly separately based on the TGF-b response signature, this indeed confirmed our hypothesis that AC is different from SCC in their response to TGF-b. We investigated this further at the protein level.

Patients
In total 62 patients with usual type cervical adenocarcinoma were included in this study. Clinicopathological characteristics are summarized in Table 1. Protein expression scores are summarized in Table 2 and Figure 2, examples of stainings are shown in Figure 2. Univariate survival analysis is shown in Table 3 and illustrative Kaplan Meier Curves for RFS and DSS are shown in Figure 3 and 4. A comparison of protein expression scores between this cohort of 62 ACs and a previously analysed and reported cohort of SCCs (29,32,33,37) is summarized in Table 4 and shown in graphs in Figure 2.

pSMAD2 and SMAD4 Expression
Overall, pSMAD2 and nuclear SMAD4 staining was more weakly expressed in AC compared to SCC ( Table 4, Nuclear pSMAD2 staining was observed in almost all AC tumours (only one tumour was completely negative, Table 2 and Figures 2C, D). Nuclear SMAD4 staining was present in 68% of the tumours, and cytoplasmic SMAD4 staining in 87% of the tumours ( Table 2; Figures 2E, F). For SMAD4, four cases showed weak nuclear staining but absent cytoplasmic staining, whilst 13 cases showed absent nuclear staining but positive cytoplasmic staining. For SMAD4, as well as for pSMAD2, a typical staining pattern was observed with an intense positive nuclear staining on the tumour edges, and a weak/negative staining pattern towards the centre of the tumour fields.

Avb6 Expression
Overall, avb6 was weakly expressed in AC compared to SCC ( Table 4; Figures 2I, J). The characteristic avb6 staining of intense staining at peripheral tumour borders and weaker staining centrally within the tumour nest was less obvious in AC compared to SCC, and was only found in larger, undifferentiated AC samples (32).

MMP2 and MMP9 Expression
Moderate to strong MMP2 expression was shown in 76% of the cases ( Table 2; Figures 2K, L). The typical tumour border staining [as previously reported in SCC (37)] was less obvious in AC. Although a stronger staining pattern on the edges of the tumour outline was observed in 20 tumours (32%) versus homogenous staining in all other tumours.
Low MMP2 expression was associated with deeper tumour infiltration (mean infiltration depth 13 ± 8mm vs. 9 ± 4mm, p=0.025), but not with any of the other clinicopathological characteristics. MMP2 expression was not associated with RFS         Figure 3, 4). High MMP2 expression occurred more frequently in AC compared to SCC ( Table 4; Figure 2L). Weak MMP9 expression was observed in 40% of the tumours, whilst 60% stained negative ( Table 2; Figures 2M,  N). MMP9 expression was also seen in different amounts at the tumour stroma near the tumour border and scored as abundant (n=15, 24%), sporadic (n=34, 55%), or negative (n=13, 21%). Neither MMP9 expression in tumour cells, nor MMP9 stromal expression was associated with any of the clinicopathological characteristics, and there were no associations with survival ( Table 3; Figures 3 and 4). No difference in MMP9 expression intensity was found between AC and SCC (Table 4; Figure 2N).

TbR2 null and SMAD4 null Analysis
Previously, the TCGA research network identified a high number of aberrations in tumour suppressor genes related to the TGF-b pathway in both SCC and AC (19). In SCC, both SMAD4 and TGFBR2 were mutated in 4% and 8% of the cases, respectively, whereas in AC only SMAD4 and not TGFBR2 mutations were observed in 12% of cases. Mutations in these genes are expected to result in absence of or aberrant protein expression. Hence, we assessed the absence of TbR2 and SMAD4 expression in our AC dataset. Indeed, SMAD4 was completely absent (SMAD4 null , for both nuclear and cytoplasmic staining) in 4/52 AC cases (7.54%) (Figures 5A, B). Unexpectedly, also TbR2 expression was absent (TbR2 null ) in 5/58 (8.62%) AC cases ( Figure 5C). PAI-1 expression and MMP2 expression were both significantly reduced in TbR2 null patients ( Figure 5D). Kaplan Meier survival analysis indicated that the TbR2 null group and the TbR2 null or SMAD4 null group had a significantly worse RFS compared to the wild type group ( Figures 5C, D and E).

DISCUSSION
Research concerning TGF-b signalling in cervical AC is scarce. Farley et al., studied TGF-b and TbR protein expression in AC (N=7) and its precursor lesions, and suggested that the neoplastic transformation of the endocervix might be related to deregulated TGF-b, and therewith loss of cell cycle control (39). Fan et al., studied TGF-b protein expression in 66 AC cases (who all received chemo-and immunotherapy prior to surgery), and described a positive expression being an independent predictor for worse survival (40). Previous studies of our group concerning the TGF-b pathway in cervical cancer demonstrated differences in TGF-b pathway activity comparing AC to SCC, however, too few AC samples were included to lead to meaningful conclusions (33). By comparing the expression of a 153-gene TGF-b response signature in AC and SCC patient samples from the TCGA dataset (19), we identified that most AC and SCC cluster separately. This suggests that the TGF-b pathway might indeed play a different role in the tumorigenesis of these cervical carcinoma subtypes. To better characterize cervical AC, we performed a comprehensive IHC analysis of various proteins of the TGF-b canonical pathway and downstream TGF-b pathway targets in a well-defined, consecutive cohort of 62 cervical AC patients. The AC series of our study was carefully classified using an additional mucus staining (14).
First, we assessed the expression of the core TGF-b canonical pathway proteins: TGF-b receptors and SMAD proteins. These proteins, are crucial in the signalling cascade to transfer the extracellular signal from activated TGF-b into transcription of target genes, including PAI-1, avb6 integrin, MMP's and several cell cycle inhibitors ( Figure 6). We showed that low expression of the core TGF-b pathway proteins (TbR1, TbR2, pSMAD2 and SMAD4) was associated with a poor prognosis in AC. Interestingly, a similar correlation to disease progression was observed in SCC (29). Previously, our group reported that all cervical SCC tumours show nuclear pSMAD2 expression and cytoplasmic SMAD4 expression (29). In comparison with SCC, here, pSMAD2 and SMAD4 staining was found to be lower in AC (Table 4; Figures 2D, F). Overall, the association of low expression of pSMAD2 and SMAD4 with poor prognosis, suggests that inactivation of the TGF-b pathway in AC might result in hampering cytostatic effects of the pathway.
In a few samples, we observed a complete lack of staining for either TbR2 or SMAD4. These TbR2 null and SMAD4 null cases were identified in 9% and 8% of our cohort, respectively. In the TCGA data, approximately 12% of AC patients showed a mutation or DNA methylation of SMAD4 (19). This could fit with our findings; the slightly higher percentage might be due to the low sample size in the TCGA dataset (31 AC samples) and of the group deficient in TbR2. In contrast, the TCGA analysis revealed no mutations or DNA methylation for TGFBR2, whereas our results suggest that about 9% of the cases had a complete lack of expression. Again, the small AC sample size in the TCGA cohort, and the small sample size of patients deficient in TbR2, might be the cause. Alternatively, this might be the result of epigenetic or posttranslational silencing. Future research should address these questions. Next, we investigated TGF-b regulating and target proteins avb6 integrin and PAI-1. Avb6 is a transmembrane cell surface receptor that mediates cell adhesion. It binds and activates latent TGF-b (41). In turn, TGF-b upregulates the expression of avb6 integrin on human keratinocytes (42). The serine protease inhibitor PAI-1, regulates the cleavage of plasminogen into plasmin by inhibiting the urokinase-and tissue plasminogen activators (uPA and tPA), which is an important mechanism in the regulation, formation, and degradation of the extracellular matrix (ECM) (33,43). In normal cervical epithelium, PAI-1 and avb6 are not-or weakly expressed, while their expression in cervical intraepithelial neoplasia is strong (32). Interestingly, in AC we observed predominantly weak PAI-1 and avb6 staining. This was in contrast with SCC, where PAI-1 and avb6 were strongly expressed (Table 4; Figures 2H, J). Moreover, in AC the characteristic staining pattern of avb6 on the tumourstromal interface was less obvious compared to SCC (32). Despite the overall weak staining, high PAI-1 and avb6 expression were associated with a worse outcome. These findings were opposite to the expression of core TGF-b pathway members (TbR and SMAD proteins), suggesting these proteins might be induced by other signalling pathways associated with worse survival outcome (e.g. TNFa/NFkB) (44). Indeed, neither PAI-1 nor avb6 showed correlative (inverse) expression with one of the core TGF-b pathway proteins in AC. This was in contrast to SCC, where TGF-b1 expression inversely correlated with PAI-1 and avb6  (31,32). Thus, although in both AC and SCC high PAI-1 expression correlates with worse outcome, the expression pattern and staining intensity of PAI-1 and of avb6 was different between the two subtypes.
In AC, TGF-b expression was found to be higher compared to SCC (33,45). If TGF-b1 expression is high in AC, but the core pathway proteins and target/regulating proteins are lower in tumour cells, this suggests that TGF-b1 might be acting predominantly on the microenvironment in AC rather than on the tumour cells. TGF-b affects many cell types in the microenvironment, including immune cells, and it can promote tumour progression by evasion of the immune system (24). By activating pro-tumorigenic microenvironment cells like M2 macrophages or by inhibiting CD8 T cell activation, TGF-b1 might result in a tumour promoting effect. Future work should address these points.
Similar to the differences observed in protein expression scores for SCC and AC, our bioinformatical analysis showed that SCC and AC samples fell into separate clusters when looking at the TGF-b response signature. This suggests, that these subsets do show differences in TGF-b signalling. Part of these differences could be explained by the lower staining intensity of pSMAD2, SMAD4, PAI-1 and avb6 in SCC. The TGF-b pathway-related protein expression differences observed between AC and SCC are summarized in Figure 6. Interestingly, 6/31 of AC samples clustered in the predominant SCC cluster, indicating that these 6 AC patients might respond to TGF-b similarly to SCC patients, and might potentially have higher TGF-b related protein expression. Further investigation is necessary to identify the differences in TGF-b response between AC and SCC.
In conclusion, in AC, similar to SCC, PAI-1 and avb6 integrin are unfavourable prognostic factors although this may not be TGF-b pathway related -given the inverse correlation in protein expression levels. Moreover, we showed that in AC there is a low core TGF-b pathway protein expression compared to SCC, and low expression is associated with poor patient survival and worse prognosis. As such, the use of TGF-b inhibitors to treat AC patients with moderate to strong staining for core TGF-b pathway proteins, would not be recommended, as tumour cell proliferation might increase. Alternatively, for AC patients with high expression of TGF-b target proteins such as PAI-1 and avb6 integrin, as well as for SCC patients, TGF-b inhibitors could be of clinical interest. Few clinical trials using these inhibitors are including patients with cervical cancer, and targeting TGF-b remains challenging (46,47). The role of TGF-b on the immune system is gaining interest, and there is increasing evidence that TGF-b affects immunotherapy response in cervical cancer as well (26,(48)(49)(50)(51). Differences between AC and SCC, should be taken into account in these studies, as immune cell recruitment has been found to differ between the two subtypes (51). Future studies including validation in in vitro and in vivo models will be necessary to further elucidate the role of TGF-b in cervical AC and SCC. Our results could also be of relevance for other cancer types with this subclassification, such as lung cancer. Based on the here presented data, characterisation of TGF-b related protein expression in individual AC patients will be important upon patient inclusion in future clinical trials with TGF-b inhibitors, as monotherapy or in combination with other therapies.

DATA AVAILABILITY STATEMENT
Publicly available datasets were analyzed in this study. This data can be found here: TCGA database, discussed in DOI:10.1038/nature21386.

AUTHOR CONTRIBUTIONS
VMS carried out the experiments. RCS and MK performed out the bioinformatical analysis. EMS, ESJ, LR and LM analysed data. Study design was done by VMS, and EJ. VMS and CDdK were responsible for recruitment of patients and collecting samples. VMS and DLM wrote the draft manuscript. ESJ, LR, PtD and CDdK revised the manuscript. All authors had final approval of the submitted and published versions.

FUNDING
This study was supported by Cancer Genomics Netherlands (CGC.NL) to LR and PtD, and the Louise Vehmeijer stichting Amsterdam to ESJ.

ACKNOWLEDGMENTS
We acknowledge Enno Dreef and Natalja ter Haar for their assistance performing the immunological staining's. We kindly thank Cees Sier for providing the MMP9 antibody. We thank Arko Gorter for his structural comments on the design of this study.