mTOR pathway activation assesed by immunohistochemistry in cervical biopsies of HPV-associated endocervical adenocarcinomas (HPVA): Correlation with Silva invasion patterns

Abstract

The Silva pattern of invasion, recently introduced to stratify patients at risk for lymph node metastases (LNM) in HPVAs, can only be assessed in cone and LEEP excisions with negative margins or in a hysterectomy specimen. Previous studies found associations between destructive stromal invasion patterns (Silva patterns B and C) and mutations in genes involved in the MEK/PI3K pathways that activate the mTOR pathway. The primary aim of this study was to use cervical biopsies to determine whether markers of mTOR pathway activation associate with aggressive invasion patterns in matched excision specimens. The status of the markers in small biopsy specimens should allow us to predict the final and biologically relevant pattern of invasion in a resection specimen. Being able to predict the final pattern of invasion is important, since prediction as Silva A, for example, might encourage conservative clinical management. If the pattern in the resection specimen is B with LVI or C, further surgery can be performed 34 HPVA biopsies were evaluated for expression of pS6, pERK and HIF1α. Immunohistochemical stains were scored semiquantitatively, ranging from 0-4+ with scores 2-4+ considered positive, and Silva pattern was determined in follow-up excisional specimens.

Silva patterns recognized in excisional specimens were distributed as follows: pattern A (n=8), pattern B (n=4) and pattern C (n=22). Statistically significant associations were found comparing pS6 and pERK immunohistochemistry with Silva pattern (p=0.034 and 0.05, respectively). Of the three markers tested, pERK was the most powerful for distinguishing between pattern A and patterns B and C (p=0.026 OR=6.75 CI95%[1.111-41.001]). Although the negative predictive values were disappointing, the positive predictive values were encouraging: 90% for pERK, 88% for pS6 and 100% for HIF1α. mTOR pathway activation assesed by immunohistochemistry in cervical biopsies of HPV-associated endocervical adenocarcinomas (HPVA) correlate with Silva invasion patterns.

INTRODUCTION

Endocervical adenocarcinoma (ECA) accounts for 15% to 20% of cervical carcinomas, and often occurs in young women^1-3. The management of ECA depends on the International Federation of Gynecology and Obstetrics (FIGO) staging system, based primarily on findings at physical examination, tumor size, disease involvement of other sites, extent/depth of invasion for clinically occult lesions and more recently on imaging and pathologic findings^1-7.

The Silva pattern–based classification system was developed to stratify ECAs into categories that correlate with risk of lymph node metastasis (LNM) and recurrence^8,9. This system is based on the pattern of stromal invasion and presence of lymphovascular invasion (LVI). The Silva pattern can only be assessed in cervical cone biopsy, LEEP excisions with negative margins and hysterectomy specimens. Three patterns were proposed: (A) nondestructive stromal invasion without LVI; (B) focal destructive stromal invasion with or without LVI; (C) diffuse destructive stromal invasion with or without LVI^8,9. Independent of tumor size or depth of invasion, pattern A tumors lack LNM, disease recurrence or disease-related death, and pattern B tumors are only rarely associated with LNM. However, pattern C tumors have been associated with higher FIGO stage, presence of LNM, disease recurrence and disease-associated death^1,2,9-11.

Predicting the Silva pattern from a cervical biopsy containing adenocarcinoma, the aim of this study, could theoretically lead to conservative clinical management for a patient with a pattern A cervical adenocarcinoma and more aggressive management for a pattern C tumor. We note previously reported associations between Silva pattern and gene mutations that activate the mTOR (mammalian target of rapamycin) pathway ^11-13. The PI3K/AKT/mTOR pathway is an intracellular signaling pathway important in regulating the cell cycle. We evaluated whether the immunohistochemical expression of selected markers of mTOR pathway activation in small cervical biopsies can be used to predict the Silva pattern of invasion in resection specimens.

MATERIAL AND METHODS

Appropriate institutional review board approval was obtained by each of the participating institutions.

Case Selection

All Human Papillomavirus (HPV)-associated invasive endocervical adenocarcinomas in the International Endocervical Adenocarcinoma Criteria and Classification (IECC) database, diagnosed in a hysterectomy or a cone biopsy with negative margins, were selected from 2 centers (USA: Memorial Sloan Kettering Cancer Center, New York, NY, and Romania: University of Medicine, Pharmacy, Sciences and Technology of Targu Mures). Cases were considered eligible if a previous biopsy was available for review, diagnosed as HPVA, and sufficient tissue for immunohistochemical testing was present^14,15. The Silva pattern was originally assigned by consensus review of excisional specimens using the Silva criteria^8,9,14. Two pathologists (SS and SES) subsequently reviewed all available slides from the excisional specimens (n=34), independently of each other, without knowledge of the tumor’s appearance in the preceding biopsy.

The following demographic and clinicopathologic data, obtained from the electronic patient record and pathology reports, were recorded: type of gynecological excisional procedure (cone biopsy, trachelectomy, hysterectomy, or exenteration) FIGO stage and tumor size, using measurements reported in the excisional specimen. All cases were restaged using FIGO 2018 staging system⁷. We defined “large” tumors as greater than 20 mm, which was informed by the median tumor size, 21.5 mm.

Immunohistochemical staining

The markers studied have direct or indirect interactions with mTOR pathway activation, as will be discussed subsequently. The immunohistochemical panel chosen was informed in part by availability of reagents and in part by institutional experience with the chosen markers. Immunohistochemical staining of tissue of whole slides from biopsies was used to evaluate the expression of pERK, pS6 and HIF1α.

A multi tissue block with ten normal tissues (spleen, tonsil, lung, kidney, testis, liver, skin, placenta, lung, pancreas) was used as positive control for pERK (clone D13.14.4E (#4370; Cell Signaling Technologies/CST, Danvers, MA), pS6 (clone 91B2 (#4857; CST) and HIF1α (HIF1α, clone H1alpha67 (NB100-123; Novus, Centennial, CO). Primary incubation time was 30 minutes. Antigen retirval was perfromed at 99°C for 30 minutes. All staining was performed on a Leica-Bond-3 (Leica, Buffalo Grove, IL) automated stainer platform. For the detection of the primary, a polymer based secondary kit (Refine, Leica) was employed.

The intensity of cytoplasmic expression (pERK and pS6) or nuclear staining (HI1α) was recorded using a four-value intensity score 0 (negative), 1 (weak), 2 (moderate) and 3 (strong) and the percentage of expression also using a four-value score 0 (0%), 1 (1-25%), 2 (26-50%), 3 (51-75%) and 4 (76-100%), were evaluated separately. This represents a slightly modified version of one reported previously¹⁶. For statistical analysis, we considered positive cases with a percentage score of 3-4 and a moderate to strong intensity.

Statistical Analysis

We studied associations between the immumnohistochemical results in biopsies, the Silva pattern assessed in matched excisional specimens and correlation with size and FIGO stage (2018). Data were analyzed using Chi-squared test. Cross tab test was used to determine sensibility, specificity and predictive value. P<0.05 was considered statistically significant.

RESULTS

We identified a total of 34 eligible cases with available tissue for analysis. Biopsy was followed by either cone biopsy (7 cases) or hysterectomy (27 cases). Tumor sizes ranged from 1-60 mm (median 21.5 mm). Out of 34 cases, 2 cases (5.88%) were FIGO stage IA and 28 cases (82.35%) were stage IB. Of these, 11 cases (39.28%) were IB1, 10 cases (35.71%) were IB2 and 7 cases (25%) were FIGO IB3. Size and FIGO stage were not available (NA) in 2 cases of Silva pattern A, 1 cases of Silva pattern B and 1 cases of Silva pattern C.

Eight (24%), four (12%) and twenty-two (64%) cases were classified as Silva pattern A (n=8), B (n=4) and C (n=22), respectively in excisional specimen. When Silva pattern of invasion was compared on surgical specimen (cone or hysterectomy) versus small biopsies, we found a similar Silva pattern in most cases (89% of cases) while in 11% of cases (4 cases of Silva B pattern) the Silva pattern on surgical specimen was upgraded from Silva A (diagnosed on biopsy) to Silva B.

pERK

The percentages of positive cases for Silva pattern A, B and C were 25%, 50% and 73%, respectively (Table 1). pERK expression was associated with destructive invasion patterns B/C, when compared with non-destructive pattern A (p=0.026 OR=6.75 CI95% [1.111-41.001]) (Tables 1 and 2). The sensitivity of pERK expression was 69% and specificity was 75%, positive predictive value was 90% and negative predictive value was 43%.

Table 1.

Analysis of pS6, pERK and HIF1 α expression stratified by Silva invasion pattern

	pS6		pERK		HIF1 α
	Negative (0-1-2)	Positive (3-4)	Negative (0-1-2)	Positive (3-4)	Negative (0-1-2)	Positive (3-4)	NA
SILVA A (8 cases)	7 (87.5%)	1 (12.5%)	6 (75%)	2 (25%)	7 (100%)	0	1
SILVA B (4 cases)	1 (25%)	3 (75%)	2 (50%)	2 (50%)	3 (100%)	0	1
SILVA C (22 cases)	18 (81%)	4 (19%)	6 (27%)	16 (73%)	15 (68%)	7 (32%)	0
	p=0.034		p=0.05		p=0.131

(NA: not available)

Table 2:

Analysis of pS6, pERK and HIF1a expression stratified by dichotomized Silva invasion pattern (A versus B/C)

	pS6		pERK		HIF1 α
	Negative (0-1-2)	Positive (3-4)	Negative (0-1-2)	Positive (3-4)	Negative (0-1-2)	Positive (3-4)	NA
SILVA A	7 (87.5%)	1 (12,5%)	6 (75%)	2 (25%)	7 (100%)	0	1
SILVA B/C	19 (73.1%)	7 (26.9%)	8 (30.8%)	18 (69.2%)	18 (72%)	7(28%)	1
	p=0.400		p=0.026 OR=6.75 CI95% (1.111-41.001)		p=0.113

(NA: not available)

pS6

The percentages of positive cases for Silva pattern A, B and C were 12.5%, 75% and 19%, respectively (Table 1). Although pS6 expression was present in pattern B, when stratified by non-destructive and destructive patterns (A vs B/C), it was not statistically significant (p=0.400) (Tables 1 and 2). The sensitivity of pS6 expression was 27% and specificity was 88%, positive predictive value was 88% and negative predictive value was 27%.

HIF1α

The percentages of positive cases for Silva pattern A, B and C were 0%, 0% and 32%, respectively (Table 1). For Silva pattern A versus B/C cases, this marker was not statistically significant (p=0.113) (Tables 1 and 2) (Figure 1, 2). The sensitivity of HIF1α expression was 28% and specificity was 100%, positive predictive value was 100% and negative predictive value was 28%.

Figure 1:

Case 1. Fragments of HPV-associated endocervical adenocarcinomas of usual type with papillary architecture in a cervical biopsy (A); pS6 score 1+ cytoplasmic staining in the biopsy (B); ERK score 3+ cytoplasmic staining in the biopsy (C); same case, invasive HPV-associated endocervical adenocarcinomas of usual type in the hysterectomy specimen, Silva pattern C (D) (100x).

Figure 2:

Case 2. Fragments of HPV-associated endocervical adenocarcinomas of usual type with papillary architecture in a cervical biopsy (A); pS6 score 3+ cytoplasmic staining in the biopsy (B); ERK score 4+ cytoplasmic staining in the biopsy (C); HIF1α score 3+ nuclear staining in the biopsy (D) (100x).

Tumor Size

Relationships between tumor size and Silva pattern in excisional specimens were also studied using a size cutoff separating “small” and “large” (20 mm) carcinomas. Of the 8 pattern A cases, 5 (83.34%) were <20 mm and 1 (16.66%) ≥20 mm; all 3 (100%) pattern B cases were < 20 mm, and for pattern C, 5 (23.8%) cases were <20 mm and 16 (76.2%) ≥20 mm (1 case:NA) (Supplemental Table 1).

Large tumor size (≥20 mm) correlated with Silva pattern (16.66% in non-destructive pattern A versus 66.7% in destructive pattern B/C, (p=0.027)(OR=10, 95%CI=0.994-100.612) (Supplemental Table 2).

FIGO stage

Relationships between tumor stage and Silva pattern were also studied. Most Silva A cases were FIGO stage IB1 (4 cases, 66.66%), followed by 1 (16.66%) stage IA case and 1 stage IB2 (16.66%) case. Silva pattern B was identified in 2 IB1cases, one 1 stage IA adenocarcinoma. All cases with Silva pattern C were stage IB, of which 5 cases (23.8%) were IB1, 9 cases (42.85%) IB2 and 7 cases (33.35%) IB3 (Supplemental Table 1). Chi-squared test demonstrates a significant statistical correlation betweeen FIGO stage and Silva pattern (p=0.033). However no statistical correlation was found between FIGO stage and dichotomized non-destructive Silva pattern A versus destructive Silva pattern B/C (p=0.148) (Supplemental Table 2). The analysis of pS6, pERK and HIF1α expression stratified by FIGO stage is illustrated in Supplemental Table 3. pERK was the only marker analysed that was significantly correlated with FIGO stage (p=0.027).

DISCUSSION

Dysregulation of the PI3K/AKT/mTOR signaling pathway has been linked to the development of various solid malignancies, including cervical carcinomas¹⁷. Recent work by Hodgson et al. and Spaans et al., described a high frequency of genetic abnormalities in the PI3K/AKT pathway in ECAs with Silva patterns B and C (destructive invasion) when compared to pattern A (nondestructive) tumors^12,13. To the best of our knowledge, this is the first study to explore the use immunohistochemistry to indirectly assess activation of the PI3K/AKT/mTOR pathways in endocervical adenocarcinomas. Strong correlations between markers of mTOR pathway regulation and final Silva status are reported here.

The Cancer Genome Atlas (TCGA) evaluation of cervical cancer concluded that over 70% of cervical cancers exhibited genomic alterations in either one or both the PI3K–MAPK and TGFβ pathways¹⁷. Other studies have reported mutations in the PIK3CA, KRAS and PTEN genes in endocervical adenocarcinomas, all members of the PI3K/AKT/mTOR signaling cascade^18-21. However, the ECAs included in these studies were not classified by the IECC or the Silva pattern of invasion.

The PI3K/AKT/mTOR signaling pathway is one of the most frequently altered pathways that is involved in cell proliferation, differentiation, development, survival and progression of several solid tumors. Multiple upstream and downstream components such as Epidermal Growth Factor Receptor (EGFR), phosphatidylinositol 3-kinase (PI3K), protein kinase B (AKT/PKB), phosphatase and tensin homolog (PTEN) and mammalian target of rapamycin (mTOR) have been found to be dysregulated in multiple cancer types^22-24. This pathway is activated by growth factors and other cellular stimuli to regulate protein synthesis through two downstream messengers: ribosomal protein S6 kinase beta-1 (S6K1) and eukaryotic translation initiation factor 4E binding protein 1 (4EBP1)^23,25. Overexpression of pS6 has been correlated with poor prognosis in sinonasal carcinomas, nasopharyngeal tumors, laryngeal carcinomas, salivary gland carcinomas and esophageal tumors^25-28.

The endoplasmic reticulum (ER) regulates calcium homeostasis, protein folding, and transportation in eukaryotic cells. In response to intrinsic or extrinsic stress inducers, several responses may occur, including ER-associated degradation (ERAD), the unfolded protein response (UPR), and apoptosis. The UPR pathway transiently inhibits protein synthesis via the induction of protein kinase RNA-like endoplasmic reticulum kinase (pERK), which phosphorylates eukaryotic translation initiation factor 2α (eIF2α), in order to restore the homeostasis of the ER and to promote cell survival. The activation of the UPR is thought to contribute to tumor development^29-32. pERK, is described to regulate the PI3K/AKT/mTORC1 pathway by activating AKT, increasing AMPK activity or inactivating TSC2³³.

Hypoxia in solid tumors has been associated with tumor progression, metastasis and treatment resistance^34-36. Activation of the hypoxia-inducible factor 1 (HIF-1) regulates the expression of numerous genes that promote adaptive response to hypoxia, including angiogenesis, anaerobic glycolysis and erythropoiesis. HIF-1 is composed of two subunits: HIF-1α (oxygen labile) and HIF-1β (constitutively expressed in cells). In response to hypoxic conditions HIF-1α becomes stable and accumulates in the nucleus where it forms a heterodimer with the HIF-1β subunit. The activated HIF-1 complex controls the expression of various hypoxia-regulated genes, such as Vascular Endothelial Growth Factor (VEGF) and Erythropoietin (EPO), which regulate angiogenesis and erythropoiesis, respectively. Overexpression of HIF-1 has been reported in many types of cancers as one of the cellular responses to hypoxic stress that results in adaptation of the cancer cells to oxygen deprivation^31,32. Hypoxia inducible factor (HIF) is generally thought to inhibit mTOR by regulation of the DNA damage response 1 (REDD1) and by repression of the DEP domain-containing mTOR-interacting protein^37,38 .

In addition to studies of mTOR pathway regulation, investigators have evaluated more easily determined parameters, such as size and growth patterns, to assess whether a biopsy might be predictive of final Silva pattern. Djordjevic et al. studied associations between Silva pattern in biopsy and final Silva pattern and also between tumor size and final Silva pattern¹¹. They reported that, in contrast with pattern A cases, pattern B and C cases more frequently presented at higher stages and that pattern Cs were, on average, the largest (on average 24 mm). The presence of glandular complexity and confluence in a biopsy was also predictive of the final Silva pattern. Despite these encouraging results, the authors cautioned that prediction of the final Silva pattern using biopsy material alone was suboptimal. These data are similar to ours since in the present study, in 4 cases (11%) , the Silva pattern on surgical specimen was upgraded from Silva A (diagnosed on biopsy) to Silva B.

Rivera-Colon, et al. reported that the presence of grade 3 nuclei and/or intraluminal necrosis in a biopsy was strongly associated with Silva pattern C on excision³⁹. Our data regarding size is concordant with the data reported by Djordjevic¹¹ (see supplemental tables). Attempts to predict the final Silva pattern at the time of biopsy could require knowledge of the carcinoma’s size before excisional therapy is planned, but studies performed to date have not incorporated tumor measurements derived from imaging studies, such as MRI. Such studies should be performed. Interestingly, our results also demonstrate significant statistical correlation betweeen FIGO stage and Silva pattern (p=0.033) with pERK being the only marker analysed that was significantly correlated with FIGO stage (p=0.027) although no statistical correlation was found between FIGO stage and dichotomized non-destructive Silva pattern A versus destructive Silva pattern B/C (p=0.148) (see supplemental tables).

One of the potential weaknesses of this study was the distribution and number of cases assigned to the 3 Silva categories. In the seminal publication describing the Silva methodology, 21% of ECAs were pattern A, 26% were pattern B and 54% were pattern C⁸. A smaller study by Hodgson reported a similar distribution of cases¹³. The frequency of pattern A ECAs studied here is similar to that reported in the two papers, but our numbers were small. In contrast, the frequency of pattern B ECAs was lower in our material and the frequency of pattern C was higher; this reflects the relatively low frequency of pattern B tumors in the study set from which the current cases were drawn¹⁴. In fact, the range of frequency of pattern B across comparable studies was 9-37%⁴⁰. The reasons underlying these apparent differences are uncertain. Criteria used for patient selection for hysterectomy might have differed from other studies and across participating centers, leading to inclusion of a higher number of large, Silva pattern C tumors. For example, tumors from one center were of more advanced stage and of larger size that tumors from the other center (15 out of 22 cases of Silva C pattern were larger than 20mm in one center). Next, many pattern A and pattern B tumors are not preceded by an incisional biopsy because these do not generally form grossly visible lesions; this observation may also account for the limited number of biopsies from patterns A and B ECAs. The distinction between patterns B and C ECAs can also be subject to interobserver variation, which we attempted to mitigate here by having a group of pathologists review all slides from excisional specimens, with full consensus agreeement. A study of the diagnostic reproducibility of the various Silva patterns indicated that reproducibility for pattern B (kappa=0.32) was lower compared to pattern A and C⁴¹. The lack of a precise cut-off that distinguished between focal destructive invasion (pattern B) and diffuse invasion (pattern C) was noted in this early paper as one of several problems with interobserver agreement, but recent guidelines have introduced a size cut-off, distinguishing between patterns B and C^{7, 42}.

In this hypothesis generating study, we show that markers of mTOR pathway activation assessed in cervical biopsies correlate with final Silva pattern on excision, with high positive predictive values for all markers assessed. Data from other groups suggest that a tumor’s mutational profile, size and biopsy morphology are also associated with final Silva pattern, although none is perfect. In conclusion, these data lead to a hypothesis that further study of markers of mTOR pathway activation, alone or in combination, might provide more precise predictive information about the Silva pattern of invasion, particularly for use in clinical settings.