Human papillomavirus (HPV) drives high-grade intraepithelial neoplasia and cancer; for unknown reasons, this occurs most often in the cervical transformation zone. Either mutation or HPV E6–driven inhibition of Notch1 can drive neoplastic development in stratified squamous epithelia. However, the contribution of Notch1 and its Delta-like ligands (DLL) to site susceptibility remains poorly understood. Here, we map DLL1/DLL4 expression in cell populations present in normal cervical biopsies by immunofluorescence. In vitro keratinocyte 2D monolayer models, growth assays, and organotypic raft cultures were used to assess the functional role of DLL–Notch signaling in uninfected cells and its modulation by HPV16 in neoplasia. An RNA sequencing–based gene signature was used to suggest the cell of origin of 279 HPV-positive cervical carcinomas from The Cancer Genome Atlas and to relate this to disease prognosis. Finally, the prognostic impact of DLL4 expression was investigated in three independent cervical cancer patient cohorts. Three molecular cervical carcinoma subtypes were identified, with reserve cell tumors the most common and linked to relatively good prognosis. Reserve cells were characterized as DLL1/DLL4+, a proliferative phenotype that is temporarily observed during squamous metaplasia and wound healing but appears to be sustained by HPV16 E6 in raft models of low-grade and, more prominently, high-grade neoplasia. High expression of DLL4 was associated with an increased likelihood of cervical cancer–associated death and recurrence. Taken together, DLL4–Notch1 signaling reflects a proliferative cellular state transiently present during physiologic processes but inherent to cervical reserve cells, making them strongly resemble neoplastic tissue even before HPV infection has occurred.

Significance:

This study investigates cervical cancer cell-of-origin populations and describes a DLL–Notch1 phenotype that is associated with disease prognosis and that might help identify cells that are susceptible to HPV-induced carcinogenesis.

High-risk human papillomaviruses (hrHPV) are detected in precancerous lesions, for example, low- and high-grade squamous intraepithelial lesions (LSIL and HSIL, respectively), and cancers at several epithelial sites including the cervix, anus, oropharynx, penis, and vulva. Cervical cancer is the most prevalent HPV-associated cancer. The general notion is that for persistent lesions to arise, HPV must either target a cell with stem cell(–like) properties or confer these properties on cells through deregulation of viral oncogenes (1–4). Keratin (KRT)17+/TP63+ reserve cells in the endocervix and transformation zone (TZ), KRT7+ cuboidal cells at the squamocolumnar junction (SCJ) and KRT5+/TP63+ basal cells in the ectocervix have previously been proposed as cancer cell of origin (5–7). Differences in somatic mutation profiles and HPV type distribution between the two main histologic subtypes of cervical cancer, adenocarcinoma, and squamous cell carcinoma (SCC), suggest that multiple distinct cell populations exist in the cervix, with equally distinct molecular regulation, from which tumors can develop through clonal expansion, albeit at different frequencies (8–10). It remains unclear why the cervix is such a susceptible site (1), or why abortive infection, as seen in HSIL and cancer, occurs either within the endocervical columnar epithelium or the squamous metaplasia inherent to the TZ, but rarely in the ectocervix (11, 12).

When viral deregulation does occur in stratified squamous epithelia, the increased HPV E6 expression has been shown to trigger a decrease in Notch1 receptor activation in basal cells (4, 13). Inactivating NOTCH1 mutations are common in HPV-negative, non-TZ–associated squamous neoplasms (14, 15). Contrastingly, in some HPV-positive SCCs and cervical SCC-derived cell lines, (cleaved) NOTCH1 can be detected and has been linked to cell growth inhibition as well as proliferation, increased cell viability, and poor prognosis, in a dose-dependent manner (16–19). The Notch signaling pathway governs squamous cell fate through transcription of target genes, including HES and HEY. Although the proteolytic mechanism by which all ligands, i.e., Jagged (JAG)1–2, Delta-like ligand (DLL)1 and DLL4, activate the Notch1 receptor is identical, ligand-specific functional effects have been reported. In myocytes, pulsatile cleaved NOTCH1 (N1ICD) through DLL1 binding promotes differentiation, while sustained N1ICD through DLL4 drives proliferation (20). Studies point to a similar dual role for DLL–Notch1 signaling in epithelial cell fate regulation: in the human epidermis, commitment to differentiation is triggered in basal keratinocytes through binding of NOTCH1 to DLL1 on neighboring cells (21, 22). Contrastingly, during reepithelialization, leader cell identity is regulated by DLL4–Notch1 signaling (23), and these cells are targeted by HPV at sites of (micro)trauma to establish initial infection (24–26). Although the microenvironment within the cervix is known to dynamically alter tissue phenotype and to play a key role in the quiescence or activation of cervical (stem) cell lineages (27), both the expression of DLL1 and DLL4 at this site and their contribution to HPV-associated (pre)cancer biology remains to be delineated.

To investigate DLL expression in cell populations across the cervix and evaluate how these Notch pathway–associated properties relate to those present in HPV-induced neoplasia, we analyzed uninfected, normal cervical biopsies, (HPV16) keratinocyte 2D monolayers, and organotypic raft cultures. We used an RNA sequencing (RNA-seq)–based gene signature to identify the potential cell of origin of 279 HPV-positive cervical carcinomas from The Cancer Genome Atlas (TCGA; ref. 8). We additionally assessed the link between DLL4–Notch1 signaling and disease outcome in this dataset and validated our prognostic findings in two independent cohorts of 300 (GEO-GSE44001; ref. 28) and 77 carcinoma patients (institutional cohort).

Cell IHC

For immunofluorescence analysis, NIKS were seeded on 8-well chamber slides at six densities to reflect conditions at subconfluence (1 × 104, 2 × 104, 4 × 104, 8 × 104), confluence (1.6 × 105), and postconfluence (3.2 × 105) in a 2D monolayer. Cells were washed once in ice-cold PBS for 5 minutes, incubated in PBS-EDTA for 10 minutes to remove feeder cells, and washed again twice in ice-cold PBS. Subsequently, cells were fixed in 4% paraformaldehyde for 10 minutes at room temperature. After washing, cells were permeabilized for 30 minutes with 0.1% Triton in case of nuclear epitope detection (N1ICD), followed by blocking in either normal horse serum (for N1ICD staining, 35 minutes) or normal goat serum (for all other antibodies, 1 hour) in a humidified chamber at room temperature. Cells were thereafter incubated with the appropriate antibody in a humidified chamber at 4°C overnight, followed by a secondary antibody at room temperature for 1 hour. For N1ICD, slides were then incubated with ImmPRESS anti-rabbit HRP (Vector Laboratories; MP-7401) at room temperature for 1 hour, followed by tetramethylrhodamine tyramide reagent (1:50; PerkinElmer; #FP1014) for 6 to 8 minutes. Chromogenic detection of DLL4 was performed following the abovementioned protocol with minor modifications: after incubation with the primary antibody, ImmPRESS was added for 1 hour at room temperature and staining was developed chromogenically with 3,3′-Diaminobenzidine (DAB) for 10 minutes. Cells were briefly counterstained with hematoxylin. A Zeiss Axiovert A1 microscope and Zeiss LSM 700 confocal microscope (Carl Zeiss AG) were used for image acquisition. For quantification of the IHC data, 10 random fields were acquired for each sample and the percentage of positive (stained) cells was calculated using ImageJ (Fiji 2) software.

Growth assays

LXSN (control) and LXSN 16E6 NIKS were seeded in 6-well plates at a high density (6 × 105 cells/well), on top of 1 × 105 feeders in F-medium without EGF (FI-medium). At day 1 postseeding (subconfluence), cells were given fresh FC-medium supplemented with either DAPT (D5942, Sigma-Aldrich, a γ-secretase inhibitor) dissolved in ethanol at a final concentration of 10 μmol/L, or ethanol (control). NIKS are normally given fresh medium every other day: to enable the addition of DAPT/ethanol every 24 hours without disturbing the growth conditions, additional NIKS on top of feeder cells were plated, and conditioned medium was used to feed the cells of the growth assay every other day, alternating with fresh FC-medium. To harvest, cells were washed in PBS and feeder cells were removed by a 2-minute incubation with trypsin-versene at 37°C, 5% CO2. Trypsin was aspirated and 1 mL of fresh trypsin was added for 15 minutes to dislodge cells. Four milliliters of fresh FI-medium was used to deactivate the trypsin and cells were pipetted up and down to get a single-cell suspension. Cells were counted in duplicate on day 1 (seeding efficiency), day 2, day 4, and day 6 postseeding with a Z1 Coulter Counter (Beckman).

Cervical tissue and tissue microarray selection

The study was conducted in accordance with ethical guidelines as outlined by the Declaration of Helsinki. Hysterectomy material from patients undergoing hysterectomy for reasons other than cervical abnormality were obtained from the Cambridge University Hospitals Human Research Tissue Bank (HRTB NRES 11/OEE/0011) and the Biology of the Human Uterus Tissue Bank (NRES 17/EE/0151). Each patient provided written informed consent for use of tissue for medical research. Specifically, three formalin-fixed paraffin-embedded (FFPE) normal cervical tissue blocks were selected based on the presence of adjacent endocervical, TZ, and ectocervical epithelium. Tissue was confirmed to be HPV negative. In addition, this study included an institutional cohort of patients with a primary cervical carcinoma, who had undergone curative treatment between 1991 and 2012 at the Antoni van Leeuwenhoek Hospital (NKI-AVL, Amsterdam, the Netherlands). Ethical approval was obtained from the NKI-AVL Institutional Review Board (IRBd19076) and written informed consent was not mandatory as the study was within the scope of non-WMO research. Only patients with HPV-positive tumors of evident histology, who had achieved complete remission after treatment, were considered. A total of 77 tumor samples were included based on these criteria, consisting of 17 adenocarcinomas and 60 SCCs. Patient and tumor characteristics were assembled from medical records and all patient identities were pseudonymized. For each patient, FFPE tissue blocks were selected to create tissue microarrays (TMA).

Public datasets

For this study, two publicly accessible patient cohorts were evaluated. The TCGA CESC cohort consisted of 279 patients with HPV-positive cervical SCC and adenocarcinoma (endocervical, mucinous, and endometrioid), of which, RNA-seq data were gathered and preprocessed (8). Log-transformed RSEM gene expression values of genes characterizing potential cervical cell-of-origin populations were downloaded directly from the UCSC Xena Browser in February 2019. Values were scaled by row and Z scores were used to create the heatmap. We used KRT8 and KRT18 as markers for columnar cells and KRT5, TP63, HES2 (a differentiation-associated downstream Notch1 target; ref. 29), KRT1 and KRT13 for squamous cells. We additionally included hypothesized cancer cell-of-origin markers KRT7 (SCJ) and KRT17 (reserve cell). Tumor samples were subjectively assigned a position within the heatmap based on mutual similarities in gene expression profile. Clinical data were retrieved from firebrowse.org and the TCGA portal. The second cohort (GEO-GSE44001; ref. 28) had previously been clinically characterized and consisted of 300 patients with early-stage cervical SCC, adenocarcinoma, or adenosquamous carcinomas with unknown HPV status. Log-transformed, quantile normalized expression data were obtained.

Statistical analysis

Differences in gene expression between tumors of different molecular origin were tested using the Mann–Whitney U test. The χ2 or Fisher exact test was used to test associations between subgroups and patient and/or tumor characteristics. Correlations between genes were assessed with Pearson correlation. Two approaches were used to investigate the association between potential prognostic factors and patient survival. The log-rank test was employed to evaluate the significance of differences between groups in Kaplan–Meier survival plots. To explore whether clinicopathologic variables, the molecular subgroups, and DLL4 gene expression could explain outcome, we included the variables in univariate and multivariable Cox proportional hazards models. Statistical analyses were carried out in IBM SPSS 22 and R 3.5.0.; GraphPad Prism 8.0.2 (GraphPad Software) and Microsoft Excel 2016 were used to generate survival curves and graphs. A P value lower than 0.05 was considered statistically significant. In case of multiple testing, a Bonferroni correction was applied.

The Supplementary Materials and Methods provide more detail on cell culture; antibodies; scratch assay; immunofluorescence and IHC; TMA analysis; Western blots; and statistical analysis.

Cervical cell populations show differential expression of DLL1 and DLL4

To map initiators of Notch1 signaling in the cervix that may help characterize the cellular state at the time of virus entry, we analyzed DLL1 and DLL4 expression in uninfected, normal human cervical tissue (Fig. 1A). KRT17 was used to identify reserve cells (5, 30). In the endocervix, columnar cells showed low levels of DLL1, DLL4, and KRT17 (Fig. 1B). DLL1/DLL4+/KRT17+ reserve cells were present under columnar cells in areas of the endocervix and around the SCJ (Supplementary Fig. S1A). In the TZ, a focus of reserve cell hyperplasia appeared to undermine columnar cells and was, on the basis of histology, identified as immature squamous metaplasia. In contrast, suprabasal cells of the mature squamous epithelium in the TZ showed high DLL1 and DLL4 and weak KRT17 staining. In the ectocervix, DLL4 expression was more prominent in parabasal layers than in the basal layer, declining upon stratification and differentiation (Fig. 1C, bottom right). DLL1 expression was similarly highest in parabasal layers and absent in terminally differentiated cells. Expression within the basal layer was not uniform, with DLL1-overexpressing cell clusters (Fig. 1C, top right: arrowhead). KRT17 was not expressed in the ectocervical epithelium. Together, these observations indicate that Notch1 ligands are differentially expressed within the cervix and that the reserve cell population, the infection of which is presumed to facilitate deregulated viral gene expression and neoplasia formation (1, 2), is characterized by high DLL4 and low DLL1 expression.

Figure 1.

Differential expression of Notch1 ligands DLL1 and DLL4 and reserve-cell marker KRT17 in the human cervix. A, Overview of hematoxylin and eosin–stained uninfected, normal cervical tissue (low magnification). B and C, Representative images showing the localization and levels of DLL1, DLL4, and KRT17 in endocervical columnar epithelium, reserve cell hyperplasia, immature squamous metaplasia, mature squamous epithelium (B) and ectocervical squamous epithelium (C). Hematoxylin and eosin–stained tissue is included for evaluation of morphology. Top, red box, arrowhead, DLL1+ group of cells within the basal layer; bottom, differential expression of DLL4 in the basal layer (low) and parabasal layers (high). DAPI was used for nuclear counterstaining. Because of inevitable tissue damage during immunostaining, tissue folds are visible in some areas. BL, basal layer; Co, columnar epithelium; Imm metapl, immature (squamous) metaplasia; Mature sq, mature squamous epithelium; PBL, parabasal layers; Res, reserve cells. Scale bar, 50 μm unless otherwise indicated.

Figure 1.

Differential expression of Notch1 ligands DLL1 and DLL4 and reserve-cell marker KRT17 in the human cervix. A, Overview of hematoxylin and eosin–stained uninfected, normal cervical tissue (low magnification). B and C, Representative images showing the localization and levels of DLL1, DLL4, and KRT17 in endocervical columnar epithelium, reserve cell hyperplasia, immature squamous metaplasia, mature squamous epithelium (B) and ectocervical squamous epithelium (C). Hematoxylin and eosin–stained tissue is included for evaluation of morphology. Top, red box, arrowhead, DLL1+ group of cells within the basal layer; bottom, differential expression of DLL4 in the basal layer (low) and parabasal layers (high). DAPI was used for nuclear counterstaining. Because of inevitable tissue damage during immunostaining, tissue folds are visible in some areas. BL, basal layer; Co, columnar epithelium; Imm metapl, immature (squamous) metaplasia; Mature sq, mature squamous epithelium; PBL, parabasal layers; Res, reserve cells. Scale bar, 50 μm unless otherwise indicated.

Close modal

HPV16 E6 selectively inhibits Notch1 activation to prevent differentiation

The deregulated expression of HPV E6 and E7 expression in HSIL compared with LSIL leads to reduced responsiveness of basal cells to contact inhibition and increased cell proliferation, effects that are largely attributable to E6 (4, 31). To assess more precisely the effect of high E6 expression on Notch-driven modulation of cell fate, we used a normal human keratinocyte cell line (NIKS) capable of recapitulating the process of epithelial differentiation and proliferation in vitro (32) and compared previously generated NIKS expressing HPV16 E6 (NIKS 16E6) or an empty vector (NIKS LXSN; ref. 4). Immunostaining of the full-length Notch1 receptor (NOTCH1), the cleaved intracellular domain (N1ICD) and KRT10, an early differentiation marker, was carried out in a 2D monolayer model. Commitment to terminal differentiation is triggered by culture confluence: a setting wherein keratinocytes have formed a cohesive monolayer (33), mimicking the epithelial basal layer. Nuclear N1ICD was occasionally detected in confluent control cells (NIKS LXSN and NIKS; Fig. 2A and D), reminiscent of the expression pattern typical of the ectocervical basal layer. KRT10 induction occurred in postconfluent control cells that had migrated upwards to form a second cell layer. KRT10 and N1ICD/NOTCH1 did not appear to be coexpressed in cells, supporting a transient and rapid induction of N1ICD in the context of differentiation (Fig. 2B and D; ref. 4). HPV16 E6 abolished N1ICD expression almost entirely from confluence onwards (Fig. 2A and C; P = 0.009 and P = 0.013 for LXSN vs. LXSN 16E6 at confluence and postconfluence, respectively, Student t test). A similar reduction in NOTCH1 was observed, suggesting that abrogation of Notch1 activation by HPV16 E6 can at least partially be attributed to a reduction in the number of available Notch1 receptors from the point of confluence (4). In subconfluent cells, HPV16 E6 had no significant effect on N1ICD/NOTCH1 abundance (P = 0.111, Student t test; Fig. 2C). In line with inhibition of Notch1 controlled differentiation, the expression of HPV16 E6 led to a strong downregulation of KRT10 in postconfluent cells (P = 0.007, Student t test; Fig. 2B; Supplementary Fig. S1B).

Figure 2.

HPV16 E6–driven modulation of Notch1-regulated keratinocyte fate. A and B, N1ICD (A) and NOTCH1 and KRT10 (B) levels in cells expressing HPV16 E6 (NIKS 16E6) and control cells with an empty vector (NIKS LXSN) at subconfluent, confluent, and postconfluent conditions in a 2D monolayer. Arrowheads, mitotic figures. C, Quantification of N1ICD IHC data as the percentage of N1ICD-positive LXSN and LXSN 16E6 NIKS at various densities. Cells in 10 random fields were counted for each sample. Bars, mean ± SD (n = 3). *, P < 0.05; **, P < 0.01; n.s., nonsignificant; Student t test. D, N1ICD and KRT10 expression in control NIKS (no vector) at different monolayer densities. The bottom right-hand panel is a representative image of N1ICD expression in normal ectocervical epithelium (DAB staining). E, Western blot analysis of N1ICD, TP53 (as a surrogate marker for E6), and GAPDH levels in LXSN and LXSN 16E6 NIKS, which were plated and treated with either DAPT, dissolved in ethanol (+), or ethanol (−), following a growth assay format. Cells were harvested in duplicate at postconfluence. F, Effect of DAPT on the growth of LXSN 16E6 and LXSN NIKS. DAPT or ethanol was added from day 1 postseeding onwards and replaced every 24 hours. Growth was monitored for 6 days, and cells were counted in duplicate on days 1, 2, 4, and 6 (data depicted as mean ± SD). Red dashed line represents the cell number at which the monolayer reaches confluence. Scale bar, 50 μm.

Figure 2.

HPV16 E6–driven modulation of Notch1-regulated keratinocyte fate. A and B, N1ICD (A) and NOTCH1 and KRT10 (B) levels in cells expressing HPV16 E6 (NIKS 16E6) and control cells with an empty vector (NIKS LXSN) at subconfluent, confluent, and postconfluent conditions in a 2D monolayer. Arrowheads, mitotic figures. C, Quantification of N1ICD IHC data as the percentage of N1ICD-positive LXSN and LXSN 16E6 NIKS at various densities. Cells in 10 random fields were counted for each sample. Bars, mean ± SD (n = 3). *, P < 0.05; **, P < 0.01; n.s., nonsignificant; Student t test. D, N1ICD and KRT10 expression in control NIKS (no vector) at different monolayer densities. The bottom right-hand panel is a representative image of N1ICD expression in normal ectocervical epithelium (DAB staining). E, Western blot analysis of N1ICD, TP53 (as a surrogate marker for E6), and GAPDH levels in LXSN and LXSN 16E6 NIKS, which were plated and treated with either DAPT, dissolved in ethanol (+), or ethanol (−), following a growth assay format. Cells were harvested in duplicate at postconfluence. F, Effect of DAPT on the growth of LXSN 16E6 and LXSN NIKS. DAPT or ethanol was added from day 1 postseeding onwards and replaced every 24 hours. Growth was monitored for 6 days, and cells were counted in duplicate on days 1, 2, 4, and 6 (data depicted as mean ± SD). Red dashed line represents the cell number at which the monolayer reaches confluence. Scale bar, 50 μm.

Close modal

A higher mitotic index was observed in NIKS 16E6 compared with control cells (P = 0.011, Student t test; Supplementary Fig. S1C). To examine the consequences of Notch signaling inhibition on keratinocyte growth, we monitored the growth of NIKS LXSN and NIKS16E6 in monolayer culture, with and without the addition of DAPT. DAPT inhibits γ-secretase, a membrane-associated protease complex responsible for Notch1 receptor cleavage, and thereby indirectly inhibits Notch1 signaling, as confirmed in NIKS through Western blot analysis (Fig. 2E; Supplementary Fig. S1D). DAPT did not influence TP53 levels (a surrogate marker for E6). In the presence of HPV16 E6, the rate of cell growth was higher than that of NIKS LXSN and growth continued after reaching confluence (∼1.5 × 106 cells; Fig. 2F). Treatment of NIKS 16E6 with DAPT from day 1 postseeding resulted in a decrease in growth compared with untreated E6 cells, although growth still exceeded that of NIKS LXSN. DAPT did not influence the growth rate of NIKS LXSN. Collectively, these results suggest that Notch1 signaling might contribute both to keratinocyte differentiation and proliferation and that increased E6 expression selectively inhibits Notch1-associated differentiation.

HPV16 E6 limits keratinocyte differentiation through downregulation of DLL1

To further evaluate how HPV might restrict cell differentiation during neoplastic development and whether DLL1 might play a role in this process, immunostaining of NIKS-derived organotypic raft cultures was carried out (Fig. 3A). Parental NIKS were used as well as six NIKS HPV16 clonal cell lines, which produce rafts possessing either an LSIL-like (1L, 2L, and 3L) or HSIL-like phenotype (4H, 5H, and 6H) and are comparable with the lesions observed upon in vivo infection (4, 31). In parental rafts, DLL1 expression resembled that seen in the ectocervix (Fig. 1C), with heterogeneous expression in the basal layer, including a subset of cells showing high expression, and expression extending into parabasal layers and declining during terminal differentiation. In HPV16 LSIL-like rafts, DLL1 expression was downregulated most prominently in the lower layers while maintaining suprabasal expression. In contrast, DLL1 was almost undetectable in the entire epithelium of HSIL-like rafts, in line with the deregulated viral oncogene expression and the dramatic reduction in both N1ICD and KRT10 in these rafts (4).

Figure 3.

Inhibition of DLL1 expression by HPV16 E6 during neoplastic development. A, Representative images of the expression pattern and level of DLL1 in parental NIKS, LSIL-like (1L, 2L, and 3L), and HSIL-like organotypic rafts (4H, 5H, and 6H) created from NIKS episomal HPV16 cell lines. The NIKS parental raft had folded, explaining the second area of epithelium seen in the lower part of the image. White dotted line, “basal lamina.” B, Expression of DLL1 in subconfluent, confluent, and postconfluent NIKS. C, Expression of DLL1 in confluent LXSN NIKS and LXSN 16E6 NIKS. The right-hand graph is a quantification of the DLL1 IHC data, depicted as number of DLL1+ cells compared with the total number of cells. Ten random fields were counted for each sample. Data are presented as mean ± SD [n = 3 for (pre)confluent-, n = 2 for postconfluent samples]. **, P < 0.01; ***, P < 0.001; n.s., nonsignificant; Student t test. Scale bar, 50 μm.

Figure 3.

Inhibition of DLL1 expression by HPV16 E6 during neoplastic development. A, Representative images of the expression pattern and level of DLL1 in parental NIKS, LSIL-like (1L, 2L, and 3L), and HSIL-like organotypic rafts (4H, 5H, and 6H) created from NIKS episomal HPV16 cell lines. The NIKS parental raft had folded, explaining the second area of epithelium seen in the lower part of the image. White dotted line, “basal lamina.” B, Expression of DLL1 in subconfluent, confluent, and postconfluent NIKS. C, Expression of DLL1 in confluent LXSN NIKS and LXSN 16E6 NIKS. The right-hand graph is a quantification of the DLL1 IHC data, depicted as number of DLL1+ cells compared with the total number of cells. Ten random fields were counted for each sample. Data are presented as mean ± SD [n = 3 for (pre)confluent-, n = 2 for postconfluent samples]. **, P < 0.01; ***, P < 0.001; n.s., nonsignificant; Student t test. Scale bar, 50 μm.

Close modal

We tried to further define the effect of HPV16 E6 on DLL1-mediated induction of differentiation in our 2D monolayer model (Fig. 3B and C). In subconfluent growth conditions, DLL1 expression occurred in a subset of NIKS cells. At confluence, when scattered N1ICD starts to appear (Fig. 2A and D), DLL1 expression was comparably heterogeneous with clusters of cells showing high expression (Fig. 3B and C), consistent with the results seen in the basal layer of parental rafts and the ectocervix. In NIKS that had committed to differentiation at postconfluence, DLL1 was strongly induced (Fig. 3B; Supplementary Fig. S1E). In line with the observations in HSIL-like rafts, expression of HPV16 E6 almost completely abolished the induction of DLL1 in NIKS from confluence onwards (Fig. 3C; Student t test; P = 0.087, P = 0.0001, P = 0.0031 for the comparison of preconfluent, confluent and postconfluent LXSN and LXSN 16E6, respectively). Together, this suggests that the downregulation of DLL1 might be an additional mechanism through which HPV modulates basal cell fate and restricts the differentiation of infected cells.

HPV16 E6 induces sustainable expression of DLL4 in keratinocytes

During (micro)wound healing HPV encounters (transiently)dividing cells, which is a critical condition for initial infection (3, 34). As DLL4 is induced in both HPV-targeted leader cells and reserve cells, we examined how HPV might modulate the expression of DLL4 in LSIL and HSIL, through immunostaining of raft cultures. In parental rafts, expression of DLL4 was low in the basal layer and induced in parabasal layers, with marked membranous expression, in accordance with the levels in the ectocervix (Fig. 4A). LSIL-like rafts showed an increase in DLL4 expression in basal keratinocytes, in line with the restricted E6 activity and HPV-mediated cell proliferation in these lower epithelial layers (Supplementary Fig. S2A; refs. 4, 11). Notably, in 1L rafts, DLL4 upregulation in the basal layer was less prominent than in 2L and 3L rafts, which is consistent with the observation that 1L cells express E6 at the lowest level and that 1L raft cultures only contain a small number of MCM7-positive cells in the basal layer (4, 31). In HSIL-like rafts, as E6 expression becomes deregulated and cell proliferation extends into the middle layers (4, 31), cytoplasmic expression of DLL4 had similarly extended into suprabasal layers. ΔNp63 is downmodulated by DLL1–Notch1 activation and in turn maintains keratinocyte proliferative potential through induction of HEY expression, a downstream DLL4 target (20, 35). The increase in DLL4 was most prominent in 5H rafts, which uniquely express (suprabasal) ΔNp63 and which show the most considerable disruption of N1ICD, TP53, and KRT10 expression compared with the other rafts (4).

Figure 4.

Induction of DLL4 in leading-edge cells and by HPV16 E6 during neoplastic development. A, Representative images of the expression pattern and levels of DLL4 in rafts from the parental NIKS cell line and LSIL (2L)- and HSIL-like rafts (4H) created from NIKS episomal HPV16 cell lines. B, Representative images of DLL4 immunostaining in LXSN and 16E6 NIKS at different densities. At postconfluence, some (differentiating) cells had been excluded from the monolayer. DAB-stained cells were briefly counterstained with hematoxylin. C, Confocal 2D and 3D images of postconfluent control NIKS. DLL4 expression is absent or confined to the intracellular compartment in cells of the bottom layer (BL), with an upregulation and translocation of DLL4 to the membrane in cells of the upper layer (UL). Scale bar, 25 μm. D, Cells at the leading edge (LE) of large cell-free gaps in the NIKS monolayer induce DLL4 (box *). Cells spanning the gap, with an elongated, polarized phenotype, are DLL4+/KRT17+ (bottom, inset, higher magnification). When gaps are smaller and the edge is less defined (box **), DLL4+/KRT17+ cells are absent and DLL4 is endocytosed. IM, inner, confluent monolayer. E, DLL4 expression in keratinocytes of the leading edge (dotted line) and of the confluent monolayer across a 22-hour scratch assay. At 22 hours postscratching, wound fronts had met. Arrow, direction of cell migration. Scale bar, 50 μm unless otherwise indicated.

Figure 4.

Induction of DLL4 in leading-edge cells and by HPV16 E6 during neoplastic development. A, Representative images of the expression pattern and levels of DLL4 in rafts from the parental NIKS cell line and LSIL (2L)- and HSIL-like rafts (4H) created from NIKS episomal HPV16 cell lines. B, Representative images of DLL4 immunostaining in LXSN and 16E6 NIKS at different densities. At postconfluence, some (differentiating) cells had been excluded from the monolayer. DAB-stained cells were briefly counterstained with hematoxylin. C, Confocal 2D and 3D images of postconfluent control NIKS. DLL4 expression is absent or confined to the intracellular compartment in cells of the bottom layer (BL), with an upregulation and translocation of DLL4 to the membrane in cells of the upper layer (UL). Scale bar, 25 μm. D, Cells at the leading edge (LE) of large cell-free gaps in the NIKS monolayer induce DLL4 (box *). Cells spanning the gap, with an elongated, polarized phenotype, are DLL4+/KRT17+ (bottom, inset, higher magnification). When gaps are smaller and the edge is less defined (box **), DLL4+/KRT17+ cells are absent and DLL4 is endocytosed. IM, inner, confluent monolayer. E, DLL4 expression in keratinocytes of the leading edge (dotted line) and of the confluent monolayer across a 22-hour scratch assay. At 22 hours postscratching, wound fronts had met. Arrow, direction of cell migration. Scale bar, 50 μm unless otherwise indicated.

Close modal

Next, we considered how HPV16 E6 might affect DLL4 expression in various growth conditions. In the 2D monolayer, DLL4 was similarly induced in subconfluent control and 16E6-expressing cells (Fig. 4B). In the presence of HPV16 E6, increased cytoplasmic expression of DLL4 was observed at confluence, alongside the punctate, intracellular DLL4 staining seen in control cells. More evidently, postconfluent 16E6 cells displayed high levels of DLL4 with a predominantly cytoplasmic and perinuclear localization, as opposed to the clusters of membrane-associated DLL4 seen in top-layer control cells (Fig. 4C; Supplementary Fig. S2B). Notably, as cells approached confluence, the leading-edge cells of monolayer gaps showed increased DLL4 as well as sporadic nuclear N1ICD (Fig. 4D; Supplementary Fig. S2C). During wound healing, keratinocytes induce, among other keratins, KRT17 expression, presumably facilitating the cell migration and proliferation needed during the early phases of reepithelialization (36). Here, KRT17 coincided with DLL4 in some leading-edge cells, including morphologically migratory cells, protruding into the gap (Fig. 4D, inset). When cells were less neatly aligned and gaps were smaller, induction of DLL4 was rare and DLL4+/KRT17+ cells were absent (Fig. 4D, box* and **). To further corroborate the involvement of DLL4 in cell migration and proliferation, we performed a scratch assay in a confluent NIKS monolayer and assessed the expression of DLL4 (Fig. 4E). At the time of scratching, DLL4 expression was restricted to the intracellular compartment. Thirteen hours postscratching, DLL4 was significantly induced in cells at the leading edge when compared with the inner monolayer (Supplementary Fig. S2D). When the scratch had closed and all cell–cell contact had been reestablished, DLL4 expression returned to the basal level and pattern seen at confluence. Taken together, these results indicate that proliferation and migration of keratinocytes correlates with an increase in DLL4 expression and Notch1 receptor activation. Our data suggest that deregulated viral gene expression, which underlies high-grade neoplasia, leads to increased and persistent levels of DLL4.

Molecular analysis identifies three cervical cancer cell-of-origin populations

To get more insight into the cells from which tumors arise, we assessed mRNA expression levels of a gene signature in 279 HPV-positive cervical carcinomas from TCGA (Fig. 5A). Tumors could broadly be classified into two groups: a columnar-like (CL: KRT8+/KRT18+/TP63/KRT5/HES2) and a, more heterogeneous, reserve cell/squamous metaplasia–like group (RSL: KRT8/KRT18/TP63+/KRT5+/HES2+). We do not rule out the possibility of a third, rare subgroup (16/279, 6%), consisting of KRT8/KRT18/KRT5+/TP63+/HES2+/KRT17/KRT7 (ectocervical) basal cell–derived tumors. For further analyses, we chose to include this group in the RSL group. Remarkably, 19 of 241 (8%) tumors histologically defined as SCC had expressional profiles highly similar to CL tumors (Supplementary Fig. S3A), with one adenocarcinoma resembling RSL tumors (1/38; 3%).

Figure 5.

mRNA-based gene signature in cervical carcinomas (TCGA) and the association of cancer subtypes with patient survival. A, Gene expression profiling of HPV-positive cervical carcinomas from the TCGA (n = 279): top, heatmap clustered manually, based on a predefined gene signature of columnar and squamous cell markers (mRNA expression: row scaled, Z scores). Additional features included are histologic type (based on pathology review), DLL4 (<3rd quartile vs. >3rd quartile), KRT17 (<1st quartile, 1st to 3rd quartile, and >3rd quartile), KRT7, KRT1, and KRT13 expression. Bottom, model of the range of epithelial sites within the cervix, from which tumors in the heatmap are presumed to have originated and differential regulation of Notch signaling therein. B and C, (Five-year) DFS of patients with cervical carcinoma (n = 187), classified based on histology (B) or molecular origin (C). SCC-diagnosed tumors of the columnar subtype were compared with SCC of the reserve/squamous subtype (D) and adenocarcinoma of the columnar subtype (E). AC, adenocarcinoma.

Figure 5.

mRNA-based gene signature in cervical carcinomas (TCGA) and the association of cancer subtypes with patient survival. A, Gene expression profiling of HPV-positive cervical carcinomas from the TCGA (n = 279): top, heatmap clustered manually, based on a predefined gene signature of columnar and squamous cell markers (mRNA expression: row scaled, Z scores). Additional features included are histologic type (based on pathology review), DLL4 (<3rd quartile vs. >3rd quartile), KRT17 (<1st quartile, 1st to 3rd quartile, and >3rd quartile), KRT7, KRT1, and KRT13 expression. Bottom, model of the range of epithelial sites within the cervix, from which tumors in the heatmap are presumed to have originated and differential regulation of Notch signaling therein. B and C, (Five-year) DFS of patients with cervical carcinoma (n = 187), classified based on histology (B) or molecular origin (C). SCC-diagnosed tumors of the columnar subtype were compared with SCC of the reserve/squamous subtype (D) and adenocarcinoma of the columnar subtype (E). AC, adenocarcinoma.

Close modal

Columnar tumors are associated with a higher recurrence rate

To assess whether the cell of origin is associated with cancer prognosis, we compared the disease-free survival (DFS) and disease-specific survival (DSS) of cervical carcinomas stratified by histologic or molecular subtype. Histologic type did not influence DFS within 5 years after diagnosis (n = 187, P = 0.278, log-rank test; Fig. 5B). Contrastingly, molecular CL tumors correlated with a significantly shorter DFS compared to RSL tumors (P < 0.001, log-rank test; Fig. 5C). Multivariate Cox regression analysis confirmed that patients with CL tumors had a higher recurrence rate (HR, 3.00; confidence interval (CI), 1.08–8.35; P = 0.035), independent of lymphovascular space invasion (Table 1). SCC-diagnosed CL-tumor patients were more likely to experience disease recurrence than both SCC-RSL (P < 0.0001, log-rank test; Fig. 5D) and adenocarcinoma-CL patients (P = 0.054, log-rank test; Fig. 5E). Notably, eight (42%) of the SCC-CL cases showed no lymph node (LN) involvement. No significant difference in clinicopathologic characteristics was observed between the molecular subtypes (Supplementary Table S1). In CL tumors, HPV16 (50%), HPV18 (39%), and HPV45 (11%) were detected. In RSL tumors, HPV16 predominated (62%), followed by HPV45 (8%), HPV18 (7%), and other hrHPVs (23% combined). HPV18 was significantly enriched in CL tumors compared with RSL tumors (P < 0.00001, χ2 test).

Table 1.

Cox proportional hazards analysis (TCGA cohort).

Univariate
5-year DSS (n = 254)a5-year DFS (n = 187)a
Clinical variable HR (95% CI) P valueb HR (95% CI) P valueb 
Lymphovascular space invasion 7.68 (1.76–33.61) 0.007 5.47 (1.56–19.20) 0.008 
Lymph node involvement 3.39 (1.38–8.29) 0.008 1.26 (0.48–3.31) 0.644 
Histologic type  0.870  0.283 
 SCC Reference  Reference  
 Adenocarcinoma 0.93 (0.36–2.35)  1.71 (0.64–4.53)  
FIGO stage  0.015  0.244 
 Early (I–IIA) Reference  Reference  
 Late (IIA1–IVB) 2.11 (1.16–3.86)  1.72 (0.69–4.29)  
Grade  0.339  0.050 
 I–II (well/moderate) Reference  Reference  
 III (poor) 1.74 (0.56–5.39)  2.21 (1.00–4.88)  
Gene variable     
Molecular subtype  0.805  0.001 
 Reserve/squamous Reference  Reference  
 Columnar 1.10 (0.53–2.29)  3.54 (1.62–7.71)  
DLL4c 1.57 (1.12–2.20) 0.008 1.74 (1.11–2.73) 0.017 
Multivariable 
Lymph node involvement 4.78 (1.79–12.74) 0.002 — — 
 Present vs. absent     
FIGO stage 0.76 (0.26–2.26) 0.625 — — 
 Early (I–IIA) vs. late (IIA1–IVB)     
DLL4c 2.49 (1.32–4.72) 0.005 — — 
Lymphovascular space invasion — — 6.14 (1.74–21.72) 0.005 
 Present vs. absent     
Molecular subtype — — 3.00 (1.08–8.35) 0.035 
 Columnar vs. reserve/squamous     
Univariate
5-year DSS (n = 254)a5-year DFS (n = 187)a
Clinical variable HR (95% CI) P valueb HR (95% CI) P valueb 
Lymphovascular space invasion 7.68 (1.76–33.61) 0.007 5.47 (1.56–19.20) 0.008 
Lymph node involvement 3.39 (1.38–8.29) 0.008 1.26 (0.48–3.31) 0.644 
Histologic type  0.870  0.283 
 SCC Reference  Reference  
 Adenocarcinoma 0.93 (0.36–2.35)  1.71 (0.64–4.53)  
FIGO stage  0.015  0.244 
 Early (I–IIA) Reference  Reference  
 Late (IIA1–IVB) 2.11 (1.16–3.86)  1.72 (0.69–4.29)  
Grade  0.339  0.050 
 I–II (well/moderate) Reference  Reference  
 III (poor) 1.74 (0.56–5.39)  2.21 (1.00–4.88)  
Gene variable     
Molecular subtype  0.805  0.001 
 Reserve/squamous Reference  Reference  
 Columnar 1.10 (0.53–2.29)  3.54 (1.62–7.71)  
DLL4c 1.57 (1.12–2.20) 0.008 1.74 (1.11–2.73) 0.017 
Multivariable 
Lymph node involvement 4.78 (1.79–12.74) 0.002 — — 
 Present vs. absent     
FIGO stage 0.76 (0.26–2.26) 0.625 — — 
 Early (I–IIA) vs. late (IIA1–IVB)     
DLL4c 2.49 (1.32–4.72) 0.005 — — 
Lymphovascular space invasion — — 6.14 (1.74–21.72) 0.005 
 Present vs. absent     
Molecular subtype — — 3.00 (1.08–8.35) 0.035 
 Columnar vs. reserve/squamous     

Abbreviations: CI, confidence interval; FIGO, Fédération Internationale de Gynécologie et d'Obstétrique.

aFive-year DSS: 43 events; 5-year DFS: 26 events.

bWald test. Statistically significant values are bold.

cDLL4 mRNA expression was used as a continuous variable in the model.

Increased DLL4 expression correlates with poor cervical cancer prognosis

Although various Notch pathway genes (JAG1, JAG2, NOTCH1–3, HES2, HES5, and HEY2) were more highly expressed in RSL tumors, NOTCH4 was more abundant in CL tumors (Supplementary Fig. S3B). The expression of DLL1 and DLL4 was comparable between molecular subtypes (P = 0.863 and P = 0.435, respectively, Student t test). DLL4–Notch4 signaling is vital for vascular development and is commonly detected in tumor vessels, implying a potential tumor-origin–independent role of DLL4 in angiogenesis (37). In CL tumors, DLL4 and vascular markers CD31 (PECAM1) and VEGFR2 (KDR) expression were significantly correlated (r = 0.659 and r = 0.615, respectively, Pearson correlation, P < 0.0001), with similar correlations in RSL tumors (DLL4–PECAM1: r = 0.535, DLL4–KDR: r = 0.691, P < 0.0001). Interestingly, a relative upregulation of other pathways involved in epithelial homeostasis, like Hedgehog and Wnt, was seen in CL tumors.

Next, we explored the prognostic impact of DLL4 in the TCGA dataset through Cox's survival analyses (Table 1). Independent of LN involvement (HR, 4.78; CI, 1.79–12.74; P = 0.002) and Fédération Internationale de Gynécologie et d'Obstétrique (FIGO) stage (HR, 0.76; CI, 0.26–2.26; P = 0.625), high DLL4 expression was linked to poor 5-year DSS (HR, 2.49; CI, 1.32–4.72; P = 0.005). DLL4 was also significantly associated with a higher likelihood of recurrence (HR, 1.74; CI, 1.11–2.73; P = 0.017). The low number of events in the TCGA cohort restricted our multivariate DFS-analysis to two variables; however, we were able to confirm the association of DLL4 with 5-year DFS in an independent cohort of 300 patients with cervical carcinoma (GSE44001; 36 events; HR, 1.69; CI, 1.09–2.62; P = 0.019; ref. 28). To validate the link between DLL4–Notch1 signaling and prognosis, we assessed immunostaining of N1ICD and DLL4 in a third, independent cohort (Fig. 6A and B; Supplementary Table S2). N1ICD and/or DLL4 (tumoral) abundance was low in 51 patients (66%; low-risk), whereas in 26 patients, high abundance of both N1ICD and DLL4 in SCCs or only DLL4 in adenocarcinomas was observed (34%; high-risk). N1ICD was undetectable in all adenocarcinomas. High-risk patients showed a significantly reduced 5-year DFS and DSS compared with low-risk patients (P = 0.023 and P = 0.035, respectively, log-rank test; Fig. 6C and D).

Figure 6.

DLL4 and N1ICD protein heterogeneity and its clinical significance in cervical carcinomas (institutional cohort). A, Heterogeneity in tumoral protein abundance of N1ICD in cervical SCCs. Vascular endothelium was used as a positive internal control (black arrows). DAB was used for immunostaining. B, Heterogeneity in cytoplasmic protein abundance of DLL4 in cervical SCCs. Scoring consisted of “positive,” “weak,” or “negative” based on DLL4 protein levels in tumor cells. Representative areas were captured. Sq, squamous tumor. Scale bar, 50 μm. C and D, DFS (C) and DSS (D) according to DLL4/N1ICD abundance in cervical tumor cells (n = 77). High-risk patients, high DLL4 and high N1ICD in SCC or high DLL4 in adenocarcinoma, red line; low-risk patients, low DLL4 and/or low N1ICD in SCC or adenocarcinoma, blue line.

Figure 6.

DLL4 and N1ICD protein heterogeneity and its clinical significance in cervical carcinomas (institutional cohort). A, Heterogeneity in tumoral protein abundance of N1ICD in cervical SCCs. Vascular endothelium was used as a positive internal control (black arrows). DAB was used for immunostaining. B, Heterogeneity in cytoplasmic protein abundance of DLL4 in cervical SCCs. Scoring consisted of “positive,” “weak,” or “negative” based on DLL4 protein levels in tumor cells. Representative areas were captured. Sq, squamous tumor. Scale bar, 50 μm. C and D, DFS (C) and DSS (D) according to DLL4/N1ICD abundance in cervical tumor cells (n = 77). High-risk patients, high DLL4 and high N1ICD in SCC or high DLL4 in adenocarcinoma, red line; low-risk patients, low DLL4 and/or low N1ICD in SCC or adenocarcinoma, blue line.

Close modal

In this study, we have examined DLL–Notch1 signaling in normal squamous cells and tissue and the modulation of DLL–Notch1–driven cell fate regulation by E6 during infection. We present data showing that low and high-grade neoplasia are, through deregulation of E6, associated with an expansion of a cell population with a DLL1/DLL4+ phenotype, reflecting a proliferation-skewed state, the expression of which is tightly restricted during physiologic processes that require temporarily induced proliferation, and that reserve cells are the only cervical population inherently characterized by these traits, likely rendering them more susceptible to HPV infection and/or induced neoplastic transformation.

The RNA-seq–based gene signature analysis of the TCGA cohort has revealed that cervical cancer is a heterogeneous disease, which appears to arise from at least three distinct cell-of-origin populations (1, 2), consisting of reserve cells, endocervical columnar cells, and ectocervical basal cells. We were unable to identify a specific KRT7+ SCJ-derived subset of tumors as described by Herfs and colleagues (6). The molecular subtypes differed in their recurrence rate and the prevalence of HPV types; columnar-derived tumors were associated with a significantly shorter DFS than reserve-cell tumors and a higher frequency of HPV18, in accordance with previous studies (10) and possibly reflecting differences in tissue tropism. Our survival analysis additionally revealed an important group of histologically misdiagnosed CL-tumor patients with a shorter DFS, offering a possible explanation for why some studies (38) report a difference in prognosis between histologic types, while others (39) and our study have failed to detect it, and highlighting a patient group that might benefit from additional treatment.

Reserve cell(–like) tumors were common, suggesting that these cells are particularly susceptible to HPV infection and/or (pre)malignant transformation (5, 30, 40). Of all cervical cell populations, reserve cells were uniquely characterized by high DLL4 and low to absent DLL1 expression. This phenotype is similar to that observed in 16E6 NIKS and LSIL/HSIL-like organotypic rafts, suggesting that the baseline state of reserve cells closely resembles that of neoplastic epithelial cells. Elaborating on the risk of transformation, the development of both LSIL- and HSIL-like organotypic rafts from HPV16 isogenic NIKS clonal cell lines (31) suggests that the state of the target cell at and after the time of HPV introduction may be an important determinant of the final phenotype. The presence of a proliferative/migratory cell state, as reflected by DLL4–Notch1 signaling in (basal) leader cells during (micro)wound healing, and particularly a durable one, seems to ease carcinogenesis. Illustratively, although cancers arising from mature squamous epithelia are rare, hyperproliferative or chronically injured tissue shows a predisposition to SCC (41). Earlier studies (42, 43) detect other wound healing–associated proteins in various TZs and interestingly, their expression in cancer correlates with invasive growth (44) and poor patient survival (30). Here, we show that high DLL4 expression in cervical tumors is associated with a significantly shorter DFS and DSS, consistent with previous studies describing the contribution of DLL4–Notch1 signaling to tumor cell survival, invasion (45), and poor prognosis (46). In both columnar and squamous-derived tumors, DLL4 might function as an initiator of endothelial Notch4 signaling, potentially promoting tumor angiogenesis (47). In SCCs, endothelial-expressed DLL4 might induce Notch1 signaling in adjacent tumor cells still further (48). Hence, DLL4 inhibitors, which are currently tested in phase I trials, seem more broadly applicable and efficacious (49) than Notch1 inhibitors in (HPV-associated) epithelial cancers (50).

These observations raise questions about the added value of E6-driven DLL–Notch regulation to transformation risk in, for example, infected reserve cells, as these already inherently possess a DLL1/DLL4+ phenotype. In the uninfected cervix, the proliferative capacity of (reserve) cells appears to be under the strict control of the stromal environment (Supplementary Fig. S3C; refs. 27, 51, 52), as demonstrated by the rapid disappearance of the squamous metaplasia–associated DLL phenotype once the epithelium matures. Further research is needed to determine the contribution of other pathways, like Wnt and Hedgehog, to columnar (53, 54) and reserve cell (de)regulation. Similarly, we, and others (23, 55), show that DLL4–Notch1 signaling is temporary and strictly regulated in squamous epithelial cells during physiologic processes. E6 might render the infected cell resistant to environmental signals that would normally alter its DLL1/DLL4+ phenotype and cause the cell to become quiescent again. Additionally, wound reepithelization, which encompasses proliferation and migration, is accelerated by HPV16 E6/E7 expression in transgenic mice (56), suggesting that increased oncogene expression may reinforce preexistent proliferative cellular capacities.

Although attempting to not oversimplify HPV functions during squamous neoplastic development, we believe that our findings highlight important Notch-associated mechanisms that HPV influences, ultimately skewing cell fate. In uninfected ectocervical epithelium and rafts, DLL1-enriched cell clusters were seen in the basal layer; this pattern is common for squamous epithelia, wherein stem(-like) cells overexpress DLL1, rendering them unresponsive to differentiation cues, but capable of driving Notch1-induced differentiation in neighboring cells (21, 22). A downregulation and marked internalization of DLL4 was observed in this setting, which fits well with the need for dynamic modulation of competitive ligands like DLL1 and DLL4 during Notch1-induced cell fate regulation (57–59). Differences in ligand–receptor affinity might contribute to the opposing, dose-dependent effects of Notch1 in cervical cancer cell lines (18, 58). Implicit in this model is a role for Notch1 during squamous cell proliferation and migration (60). This is supported by studies that show that Notch1 is upregulated at the wound edge and that depletion of Notch1 leads to delayed wound closure in scratch-wounded keratinocytes (23, 55). These findings fit well with the detection of NOTCH1 cleavage, coinciding with the appearance of DLL4 in subconfluent cells and cells adjacent to large gaps, where migration is lamellipodium-mediated (61). This hypothesis is further supported by the reduction in cell growth seen after DAPT-induced abrogation of N1ICD in (subconfluent) HPV16 E6 cells, suggesting that the postconfluent growth advantage of E6 cells might in part be dependent on NOTCH1 activation. In addition, it depicts the increase of E6 from confluence onwards in HSIL-like cells, but not in LSIL-like cells (31), as a way to selectively inhibit Notch1 signaling to specifically prevent differentiation (4).

Taken together, our study highlights that reserve cells are, owing to their proliferative nature, unlikely to follow the traditional model of neoplasia, where infection of a squamous cell initially leads to productive infection, which eventually progresses to abortive infection. DLL4–Notch1 signaling may offer a promising way to identify (i) susceptible epithelial cell populations in the human body and (ii) cervical cancer patients with poor prognosis, that might benefit from additional treatment.

H. Griffin reports grants from MRC outside the submitted work. R.D.M. Steenbergen reports other funding from Self-screen BV, a university spin-off company, outside the submitted work, as well as a patent for biomarkers for cervical cancer pending and issued. J. Doorbar reports grants from Medical Research Council during the conduct of the study. No disclosures were reported by the other authors.

M. Khelil: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. H. Griffin: Supervision, investigation, visualization, methodology. M.C.G. Bleeker: Writing–review and editing, pathological assessment. R.D.M. Steenbergen: Resources, methodology, writing–review and editing. K. Zheng: Validation, investigation, writing–review and editing. T. Saunders-Wood: Investigation, visualization, methodology, writing–review and editing. S. Samuels: Data curation, writing–review and editing. J. Rotman: Data curation, writing–review and editing. W. Vos: Investigation. B.E. van den Akker: Investigation, methodology. R.X. de Menezes: Formal analysis, writing–review and editing. G.G. Kenter: Resources, supervision, writing–review and editing. J. Doorbar: Conceptualization, resources, supervision, visualization, methodology, writing–review and editing. E.S. Jordanova: Conceptualization, resources, supervision, validation, investigation, visualization, methodology, project administration, writing–review and editing.

The authors thank Drs. Erin Isaacson Wechsler for organotypic raft cultures, Christina Holleywood for Western blot analysis and growth assays, and Nagayasu Egawa for helpful discussions and support. This study was supported by the UK Medical Research Council through grant MR/S024409/1 and the Louise Vehmeijer Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Doorbar
J
,
Griffin
H
. 
Refining our understanding of cervical neoplasia and its cellular origins
.
Papillomavirus Res
2019
;
7
:
176
9
.
2.
Egawa
N
,
Egawa
K
,
Griffin
H
,
Doorbar
J
. 
Human papillomaviruses; epithelial tropisms, and the development of neoplasia
.
Viruses
2015
;
7
:
3863
90
.
3.
Doorbar
J
,
Egawa
N
,
Griffin
H
,
Kranjec
C
,
Murakami
I
. 
Human papillomavirus molecular biology and disease association
.
Rev Med Virol
2015
;
25
:
2
23
.
4.
Kranjec
C
,
Holleywood
C
,
Libert
D
,
Griffin
H
,
Mahmood
R
,
Isaacson
E
, et al
Modulation of basal cell fate during productive and transforming HPV-16 infection is mediated by progressive E6-driven depletion of notch
.
J Pathol
2017
;
242
:
448
62
.
5.
Martens
JE
,
Arends
J
,
van der Linden
PJ
,
de Boer
BA
,
Helmerhorst
TJ
. 
Cytokeratin 17 and p63 are markers of the HPV target cell, the cervical stem cell
.
Anticancer Res
2004
;
24
:
771
5
.
6.
Herfs
M
,
Yamamoto
Y
,
Laury
A
,
Wang
X
,
Nucci
MR
,
McLaughlin-Drubin
ME
, et al
A discrete population of squamocolumnar junction cells implicated in the pathogenesis of cervical cancer
.
Proc Natl Acad Sci U S A
2012
;
109
:
10516
21
.
7.
Jiang
M
,
Li
H
,
Zhang
Y
,
Yang
Y
,
Lu
R
,
Liu
K
, et al
Transitional basal cells at the squamous-columnar junction generate Barrett's oesophagus
.
Nature
2017
;
550
:
529
33
.
8.
Cancer Genome Atlas Research Network
. 
Integrated genomic and molecular characterization of cervical cancer
.
Nature
2017
;
543
:
378
84
.
9.
Spaans
VM
,
Trietsch
MD
,
Peters
AAW
,
Osse
M
,
ter Haar
N
,
Fleuren
GJ
, et al
Precise classification of cervical carcinomas combined with somatic mutation profiling contributes to predicting disease outcome
.
PLoS One
2015
;
10
:
e0133670
.
10.
Bulk
S
,
Berkhof
J
,
Bulkmans
NWJ
,
Zielinski
GD
,
Rozendaal
L
,
van Kemenade
FJ
, et al
Preferential risk of HPV16 for squamous cell carcinoma and of HPV18 for adenocarcinoma of the cervix compared to women with normal cytology in The Netherlands
.
Br J Cancer
2006
;
94
:
171
5
.
11.
Schiffman
M
,
Doorbar
J
,
Wentzensen
N
,
de Sanjosé
S
,
Fakhry
C
,
Monk
BJ
, et al
Carcinogenic human papillomavirus infection
.
Nat Rev Dis Primers
2016
;
2
:
16086
.
12.
Reich
O
,
Regauer
S
. 
Thin HSIL of the cervix: detecting a variant of high-grade squamous intraepithelial lesions with a p16INK4a antibody
.
Int J Gynecol Pathol
2017
;
36
:
71
5
.
13.
Nyman
PE
,
Buehler
D
,
Lambert
PF
. 
Loss of function of canonical notch signaling drives head and neck carcinogenesis
.
Clin Cancer Res
2018
;
24
:
6308
18
.
14.
Agrawal
N
,
Frederick
MJ
,
Pickering
CR
,
Bettegowda
C
,
Chang
K
,
Li
RJ
, et al
Exome sequencing of head and neck squamous cell carcinoma reveals inactivating mutations in NOTCH1
.
Science
2011
;
333
:
1154
7
.
15.
Wang
NJ
,
Sanborn
Z
,
Arnett
KL
,
Bayston
LJ
,
Liao
W
,
Proby
CM
, et al
Loss-of-function mutations in Notch receptors in cutaneous and lung squamous cell carcinoma
.
Proc Natl Acad Sci U S A
2011
;
108
:
17761
6
.
16.
Rettig
EM
,
Chung
CH
,
Bishop
JA
,
Howard
JD
,
Sharma
R
,
Li
RJ
, et al
Cleaved NOTCH1 expression pattern in head and neck squamous cell carcinoma is associated with NOTCH1 mutation, HPV status, and high-risk features
.
Cancer Prev Res
2015
;
8
:
287
95
.
17.
Zagouras
P
,
Stifani
S
,
Blaumueller
CM
,
Carcangiu
ML
. 
Artavanis-Tsakonas S. Alterations in Notch signaling in neoplastic lesions of the human cervix
.
Proc Natl Acad Sci U S A
1995
;
92
:
6414
8
.
18.
Henken
FE
,
De-Castro Arce
J
,
Rösl
F
,
Bosch
L
,
Meijer
CJ
,
Snijders
PJ
, et al
The functional role of Notch signaling in HPV-mediated transformation is dose-dependent and linked to AP-1 alterations
.
Cell Oncol
2012
;
35
:
77
84
.
19.
Yousif
NG
,
Sadiq
AM
,
Yousif
MG
,
Al-Mudhafar
RH
,
Al-Baghdadi
JJ
,
Hadi
N
. 
Notch1 ligand signaling pathway activated in cervical cancer: poor prognosis with high-level JAG1/Notch1
.
Arch Gynecol Obstet
2015
;
292
:
899
904
.
20.
Nandagopal
N
,
Santat
LA
,
LeBon
L
,
Sprinzak
D
,
Bronner
ME
,
Elowitz
MB
. 
Dynamic ligand discrimination in the notch signaling pathway
.
Cell
2018
;
172
:
869
80
.
21.
Lowell
S
,
Jones
P
,
Le Roux
I
,
Dunne
J
,
Watt
FM
. 
Stimulation of human epidermal differentiation by delta-notch signalling at the boundaries of stem-cell clusters
.
Curr Biol
2000
;
10
:
491
500
.
22.
Williams
SE
,
Beronja
S
,
Pasolli
HA
,
Fuchs
E
. 
Asymmetric cell divisions promote notch-dependent epidermal differentiation
.
Nature
2011
;
470
:
353
8
.
23.
Riahi
R
,
Sun
J
,
Wang
S
,
Long
M
,
Zhang
DD
,
Wong
PK
. 
Notch1-Dll4 signalling and mechanical force regulate leader cell formation during collective cell migration
.
Nat Commun
2015
;
6
:
6556
.
24.
Doorbar
J
,
Quint
W
,
Banks
L
,
Bravo
IG
,
Stoler
M
,
Broker
T
, et al
The biology and life-cycle of human papillomaviruses
.
Vaccine
2012
;
30
:
F55
70
.
25.
Flores
ER
,
Allen-Hoffmann
BL
,
Lee
D
,
Lambert
PF
. 
The human papillomavirus type 16 e7 oncogene is required for the productive stage of the viral life cycle
.
J Virol
2000
;
74
:
6622
31
.
26.
Kines
RC
,
Thompson
CD
,
Lowy
DR
,
Schiller
JT
,
Day
PM
. 
The initial steps leading to papillomavirus infection occur on the basement membrane prior to cell surface binding
.
Proc Natl Acad Sci U S A
2009
;
106
:
20458
63
.
27.
Chumduri
C
,
Kumar Gurumurthy
R
,
Berger
H
,
Koster
S
,
Brinkmann
V
,
Klemm
U
, et al
Transition of Wnt signaling microenvironment delineates the squamo-columnar junction and emergence of squamous metaplasia of the cervix
.
BioRxiv
443770
[Preprint]. 2018. Available from
: https://www.biorxiv.org/content/10.1101/443770v1.
28.
Lee
YY
,
Kim
TJ
,
Kim
JY
,
Choi
CH
,
Do
IG
,
Song
SY
, et al
Genetic profiling to predict recurrence of early cervical cancer
.
Gynecol Oncol
2013
;
131
:
650
4
.
29.
Oittinen
M
,
Popp
A
,
Kurppa
K
,
Lindfors
K
,
Mäki
M
,
Kaikkonen
MU
, et al
Polycomb repressive complex 2 enacts wnt signaling in intestinal homeostasis and contributes to the instigation of stemness in diseases entailing epithelial hyperplasia or neoplasia
.
Stem Cells
2017
;
35
:
445
57
.
30.
Escobar-Hoyos
LF
,
Yang
J
,
Zhu
J
,
Cavallo
JA
,
Zhai
H
,
Burke
S
, et al
Keratin 17 in premalignant and malignant squamous lesions of the cervix: proteomic discovery and immunohistochemical validation as a diagnostic and prognostic biomarker
.
Mod Pathol
2014
;
27
:
621
30
.
31.
Isaacson Wechsler
E
,
Wang
Q
,
Roberts
I
,
Pagliarulo
E
,
Jackson
D
,
Untersperger
C
, et al
Reconstruction of human papillomavirus type 16-mediated early-stage neoplasia implicates E6/E7 deregulation and the loss of contact inhibition in neoplastic progression
.
J Virol
2012
;
86
:
6358
64
.
32.
Allen-Hoffmann
BL
,
Schlosser
SJ
,
Ivarie
CAR
,
Meisner
LF
,
O'Connor
SL
,
Sattler
CA
. 
Normal growth and differentiation in a spontaneously immortalized near-diploid human keratinocyte cell line
,
NIKS. J Invest Dermatol
2000
;
114
:
444
55
.
33.
Poumay
Y
,
Pittelkow
MR
. 
Cell density and culture factors regulate keratinocyte commitment to differentiation and expression of suprabasal K1/K10 keratins
.
J Invest Dermatol
1995
;
104
:
271
6
.
34.
Pyeon
D
,
Pearce
SM
,
Lank
SM
,
Ahlquist
P
,
Lambert
PF
. 
Establishment of human papillomavirus infection requires cell cycle progression
.
PLoS Pathog
2009
;
5
:
e1000318
.
35.
Nguyen
BC
,
Lefort
K
,
Mandinova
A
,
Antonini
D
,
Devgan
V
,
Della Gatta
G
, et al
Cross-regulation between Notch and p63 in keratinocyte commitment to differentiation
.
Genes Dev
2006
;
20
:
1028
42
.
36.
Patel
GK
,
Wilson
CH
,
Harding
KG
,
Finlay
AY
,
Bowden
PE
. 
Numerous keratinocyte subtypes involved in wound re-epithelialization
.
J Invest Dermatol
2006
;
126
:
497
502
.
37.
Gale
NW
,
Dominguez
MG
,
Noguera
I
,
Pan
L
,
Hughes
V
,
Valenzuela
DM
, et al
Haploinsufficiency of delta-like 4 ligand results in embryonic lethality due to major defects in arterial and vascular development
.
Proc Natl Acad Sci U S A
2004
;
101
:
15949
54
.
38.
Galic
V
,
Herzog
TJ
,
Lewin
SN
,
Neugut
AI
,
Burke
WM
,
Lu
YS
, et al
Prognostic significance of adenocarcinoma histology in women with cervical cancer
.
Gynecol Oncol
2012
;
125
:
287
91
.
39.
Teke
F
,
Yoney
A
,
Teke
M
,
Inal
A
,
Urakci
Z
,
Eren
B
, et al
Lack of any impact of histopathology type on prognosis in patients with early-stage adenocarcinoma and squamous cell carcinoma of the uterine cervix
.
Asian Pac J Cancer Prev
2014
;
15
:
2815
9
.
40.
Reich
O
,
Pickel
H
,
Regauer
S
. 
Why do human papillomavirus infections induce sharply demarcated lesions of the cervix?
J Low Genit Tract Dis
2008
;
12
:
8
10
.
41.
Arwert
EN
,
Hoste
E
,
Watt
FM
. 
Epithelial stem cells, wound healing and cancer
.
Nat Rev Cancer
2012
;
12
:
170
80
.
42.
McNairn
AJ
,
Guasch
G
. 
Epithelial transition zones: merging microenvironments, niches, and cellular transformation
.
Eur J Dermatol
2011
;
21
:
21
8
.
43.
Pearton
DJ
,
Yang
Y
,
Dhouailly
D
. 
Transdifferentiation of corneal epithelium into epidermis occurs by means of a multistep process triggered by dermal developmental signals
.
Proc Natl Acad Sci U S A
2005
;
102
:
3714
19
.
44.
Bates
RC
,
Bellovin
DI
,
Brown
C
,
Maynard
E
,
Wu
B
,
Kawakatsu
H
, et al
Transcriptional activation of integrin beta6 during the epithelial-mesenchymal transition defines a novel prognostic indicator of aggressive colon carcinoma
.
J Clin Invest
2005
;
115
:
339
47
.
45.
Guo
X
,
Duan
Y
,
Ye
X
,
Hu
L
,
Xu
T
,
Tong
L
, et al
Stable silencing of dll4 gene suppresses the growth and metastasis of esophagus cancer cells by attenuating Akt phosphorylation
.
Oncol Rep
2018
;
40
:
495
503
.
46.
Yang
S
,
Liu
Y
,
Xia
B
,
Deng
J
,
Liu
T
,
Li
Q
, et al
DLL4 as a predictor of pelvic lymph node metastasis and a novel prognostic biomarker in patients with early-stage cervical cancer
.
Tumour Biol
2016
;
37
:
5063
74
.
47.
Fish
JE
,
Cantu Gutierrez
M
,
Dang
LT
,
Khyzha
N
,
Chen
Z
,
Veitch
S
, et al
Dynamic regulation of VEGF-inducible genes by an ERK/ERG/p300 transcriptional network
.
Development
2017
;
144
:
2428
44
.
48.
Bernier-Latmani
J
,
Cisarovsky
C
,
Demir
CS
,
Bruand
M
,
Jaquet
M
,
Davanture
S
, et al
DLL4 promotes continuous adult intestinal lacteal regeneration and dietary fat transport
.
J Clin Invest
2015
;
125
:
4572
86
.
49.
Gurney
A
,
Hoey
T
. 
Anti-DLL4, a cancer therapeutic with multiple mechanisms of action
.
Vasc Cell
2011
;
3
:
18
.
50.
Piha-Paul
SA
,
Munster
PN
,
Hollebecque
A
,
Argilés
G
,
Dajani
O
,
Cheng
JD
, et al
Results of a phase 1 trial combining ridaforolimus and MK-0752 in patients with advanced solid tumours
.
Eur J Cancer
2015
;
51
:
1865
73
.
51.
Vega
ME
,
Giroux
V
,
Natsuizaka
M
,
Liu
M
,
Klein-Szanto
AJ
,
Stairs
DB
, et al
Inhibition of Notch signaling enhances transdifferentiation of the esophageal squamous epithelium towards a Barrett's-like metaplasia via KLF4
.
Cell Cycle
2014
;
13
:
3857
66
.
52.
van der Marel
J
,
van Baars
R
,
Alonso
I
,
del Pino
M
,
van de Sandt
M
,
Lindeman
J
, et al
Oncogenic human papillomavirus-infected immature metaplastic cells and cervical neoplasia
.
Am J Surg Pathol
2014
;
38
:
470
9
.
53.
Gray
GE
,
Mann
RS
,
Mitsiadis
E
,
Henrique
D
,
Carcangiu
ML
,
Banks
A
, et al
Human ligands of the notch receptor
.
Am J Pathol
1999
;
154
:
785
94
.
54.
Wang
DH
,
Tiwari
A
,
Kim
ME
,
Clemons
NJ
,
Regmi
NL
,
Hodges
WA
, et al
Hedgehog signaling regulates FOXA2 in esophageal embryogenesis and Barrett's metaplasia
.
J Clin Invest
2014
;
124
:
3767
80
.
55.
Na
J
,
Shin
JY
,
Jeong
H
,
Lee
JY
,
Kim
BJ
,
Kim
WS
, et al
JMJD3 and NF-κB-dependent activation of Notch1 gene is required for keratinocyte migration during skin wound healing
.
Sci Rep
2017
;
7
:
6494
.
56.
Valencia
C
,
Bonilla-Delgado
J
,
Oktaba
K
,
Ocádiz-Delgado
R
,
Gariglio
P
,
Covarrubias
L
. 
Human papillomavirus E6/E7 oncogenes promote mouse ear regeneration by increasing the rate of wound re-epithelization and epidermal growth
.
J Invest Dermatol
2008
;
128
:
2894
903
.
57.
Andrawes
MB
,
Xu
X
,
Liu
H
,
Ficarro
SB
,
Marto
JA
,
Aster
JC
, et al
Intrinsic selectivity of Notch 1 for delta-like 4 over Delta-like 1
.
J Biol Chem
2013
;
288
:
25477
89
.
58.
Luca
VC
,
Jude
KM
,
Pierce
NW
,
Nachury
MV
,
Fischer
S
,
Garcia
KC
. 
Structural basis for Notch1 engagement of Delta-like 4
.
Science
2015
;
347
:
847
53
.
59.
Lowell
S
,
Watt
FM
. 
Delta regulates keratinocyte spreading and motility independently of differentiation
.
Mech Dev
2001
;
107
:
133
40
.
60.
Palazzo
E
,
Morandi
P
,
Lotti
R
,
Saltari
A
,
Truzzi
F
,
Schnebert
S
, et al
Notch cooperates with survivin to maintain stemness and to stimulate proliferation in human keratinocytes during
ageing. Int J Mol Sci
2015
;
16
:
26291
302
.
61.
Anon
E
,
Serra-Picamal
X
,
Hersen
P
,
Gauthier
NC
,
Sheetz
MP
,
Trepat
X
, et al
Cell crawling mediates collective cell migration to close undamaged epithelial gaps
.
Proc Natl Acad Sci U S A
2012
;
109
:
10891
6
.