Proteoglycan-based diversification of disease outcome in head and neck cancer patients identifies NG2/CSPG4 and syndecan-2 as unique relapse and overall survival predicting factors

Analyses of the relative mRNA expression levels of the eleven prevalent cell surface-associated PGs conducted on a total of 119 primary oral cavity HNSCC lesions revealed that 3 of the PGs, including NG2/CSPG4, GPC2 and GPC5, were de novo expressed in neoplastic cells, i.e. were not detectable in the healthy control tissues, but were detectable in cancer cells. These were transcribed in 71% (NG2/CSPG4), 36% (GPC2) and 72% (GPC5) of the lesions, respectively. The remaining 8 PGs, for which transcripts were expressed at a frequency of 84% (GPC1), 86% (GPC3), 88% (GPC4), 70% (GPC6), 100% (SDC1), 94% (SDC2), 93% (SDC3) and 95% (SDC4) of the tumour cases, respectively, were found to be differently modulated. Thus, SDC2, SDC3 and SDC4 were up-regulated in 79-84% of the patients, whereas GPC4 was enhanced in 11% and GPC3 in 57% of the specimens. However, GPCs were more frequently down-regulated (GPC3, 22%; GPC4, 21%; GPC1, 24%; and GPC6 30%) than SDCs (SDC1, 8%; SDC2, 8%; SDC3, 11%; and SDC4, 7%; Figure 1a; Additional file 4: Table S2).

We next compared the PG expression patterns exhibited by discrete groups of patients differentiated by tumour staging, i.e. T1-T2/N- versus T1-T2/N+ and T3-T4/N- versus T3-T4/N+. GPC2 was expressed in a mere 19% of the T3-T4/N- classified lesions, whereas it was a two-fold more frequently transcribed in patients belonging to the other three classes (T1-T2/N-, 36%, T1-T2/N+, 40% and T3-T4/N+, 42%). GPC5 and NG2/CSPG4 were detectable in samples of a large proportion of patients, ranging from 62% in T3-T4/N- to 80% in T1-T2/N-, but their relative expression levels did not discriminate between the above patient subsets (Figure 1b). GPC1 was similarly differently expressed in the distinct groups of patients with a 1.5-fold higher frequency in the T3-T4/N- patients compared to the other patient subsets. GPC4 was down-regulated in about 27% of the T1-T2/N- and T3-T4/N+ subgroups, and <13% in the T1-T2/N+ and T3-T4/N- patient subgroups (Figure 1b). Syndecans were generally up-regulated in most of the lesions (Figure 1b), with SDC4 showing enhanced expression in 100% of T1-T2/N+ patients.

Intralesional distribution of PGs was further examined in oral cavity HNSCC lesions and control healthy tissue using empirically validated, pre-selected antibodies against each of the PGs (Figure 2; Figure 3; Additional file 5: Figure S2). The percentage of cases in which GPCs could be disclosed on the epithelial neoplastic cells varied from 18% (30 out of 163 cases) for GPC3 to 72% for GPC1 (118 out of 163 cases; Table 2). Relative frequency of expression was in the order: GPC1 > GPC4 (41%; 67 out of 163 cases) > GPC6 (37%; 61 out of 163 cases) > GPC3. GPCs were often detected within the cytoplasm as well as on the cell membrane, consistent with their thoroughly described internalization and recycling patterns. The hybrid cell membrane/cytoplasmic distribution of these PGs was characteristically observed for GPC1, GPC3 and GPC4, with GPC1 being most strongly associated with these two cellular compartments in keratinizing neoplastic cells (Additional file 5: Figure S2). GPC3 was entirely absent in healthy tissue, while GPC4 showed a widespread distribution both on normal epithelial cells and in the intralesional stromal compartment (Additional file 5: Figure S2). GPC6 appeared to be preferentially retained within intracellular vesicles (Figure 2), as deduced by the appearance of GPC6-positive granules throughout the cytoplasm of neoplastic cells. This seemed rather specific for tumour cells since it was not observed in healthy epithelial cells (Additional file 5: Figure S2). GPC1 and GPC6 were rarely seen in the intralesional stroma of oral cavity HNSCC lesions (13 out of 163 and 16 out of 163 of the cases, respectively), whereas GPC3 was consistently absent from this compartment and GPC4 showed a somewhat more frequent expression in stromal cells (19%; 31 out of 163 of the cases; Table 2; Figure 2).

HNSCC lesions showed variable expression of SDCs with a relative frequency of positive cases decreasing in the order: SDC1 > SDC3 > SDC4 > SDC2 (Table 2). In fact, a total of 149 cases out of 163 lesions that were evaluated for the in situ expression of the SDC1/CD138 protein had epithelial neoplastic cells presenting the PG on the cell surface, or in intracellular locations (Figure 3). In contrast, a mere 57 (35%), 32 (19.6%), and 19 (12%) out of 163 examined lesions had epithelial neoplastic cells staining positively for respectively SCD2, SDC3 and SDC4 (Table 2). The relative number of cancer cells that expressed these PGs in each lesion markedly differed and a similar divergence was seen in terms of subcellular localization of the molecules. Thus, neoplastic cells with plasma membrane-associated SDC1 were mainly keratinizing cells located at the center of the neoplastic nests (Figure 3). Fibroblasts of the tumour stroma that surrounded the neoplastic nests were positive for SDC1 in 34 out of 163 (21%) of the tumours. In 14 out of 163 of the lesions (8.6%), SDC2 was immunolocalized within the cytoplasm of neoplastic cells, while it appeared widespread in the stromal cells of the majority of the lesions (73.6%, 120 out of 163 of the cases; Figure 3) and was particularly enriched in lesions containing desmoplastic stroma. Intriguing was the fact that in 100% of the lesions, SDC2 could be observed in the wall of both normal and intra-lesional blood vessels, suggesting that it was associated with both endothelial and neovascular pericytes (Figure 3). In contrast to SDC1 and SDC2, both SDC3 and SDC4 were undetectable in the healthy epithelium, or the tumour stroma, but could be immunolocalized both subcellularly and on defined portions of the cell surface of neoplastic cells, with a particular concentration in focal plaque-like structures (Figure 3; Additional file 5: Figure S2). Finally, the diversity of SDCs expressions in oral cavity HNSCC lesions was even more remarkable when considering the relative distribution of these PGs in the stromal compartment. In this case, the frequency of occurrence of the PGs was largely reversed with respect to that seen in the cancer cells and decreased in the order: SDC2 > SDC1 > SDC3 = SDC4 (Table 2). Deviating from the pronounced intracellular distribution of SDC3 and SDC4 was that of NG2/CSPG4 which showed an exclusive cell membrane localization in all samples in which the PG could be disclosed (63%). NG2/CSPG4 was rarely detected in the stromal compartment (5 out of 163; Table 2; Figure 2), where, if occurring, was concentrated on the membrane of basal cells (Additional file 5: Figure S2).

All demographic and clinical-pathological traits of the patients were initially compared by univariate analysis of the cumulative PG expression data, except for patient categories comprised of less than 13 patients (independently defined as a cut-off level of “statistical” exclusion). In these correlation analyses we considered five primary disease outcomes, including loco-regional tumour recurrence, lymphnodal metastasis, distant metastases, disease-related death and a situation in which at least one of the former disease outcomes was manifested (Table 3).

In order to test whether there is a relationship between PGs transcript and protein expressions and clinicopathological parameters, a Chi-Square test was applied and just three such correlation resulted statistically significant: N classification and SDC2 stromal positivity, p = 0.002; alcohol consumption and SDC1 mRNA up-regulation, p = 0.021, and presence of precancerous lesion and SDC1 mRNA up-regulation, p = 0.016.

Although radiation therapy and excessive alcohol consumption independently correlated with one or more of the above clinical outcomes (Table 3), these parameters were not considered in the multivariate logistic regression analyses because of being potentially confounding indicators. The first because almost all patients presenting lymphnodal infiltrations had been routinely subjected to radiation therapy, the second because, despite of its well-recognized importance as a risk factor in HNSCC, the admission of this habit was measured by a self-provided questionnaire and no details were available on the accuracy of the information provided by the patients. A further consideration is that self-reported excessive alcohol intake is often denied, causing underestimation of the cohort of patient that may fall under this “risk category”.

Advanced T classification (p = 0.007), T3-T4 grouping (p = 0.001; Figure 4), positive NG2/CSPG4 transcript expression (p = 0.029; Figure 4), or GPC1 positivity in stromal cells (p = 0.007) were all conditions strongly associated with a high loco-regional tumour relapse rate (Table 3). Stromal GPC6 expression could, however, not be included as a parameter in the multivariate logistic regression model due to the low number of cases contained within this category and the borderline statistical significance in univariate analyses (p = 0.058). Application of the Cox proportional hazard model revealed that T3-T4 classification of the tumour (HR, 6.36, p = 0.001) and de novo expression of NG2/CSPG4 mRNA (HR, 6.76, p = 0.017) were independent, robust prognostic factors for local tumour recurrence (Table 4; Figure 5). If combining T-grouping and mRNA expression of NG2/CSPG4, the probability to develop a secondary loco-regional lesion was further increased (p < 0.001; Table 3; Figure 4).

Significant correlations with post-surgical lymphnodal metastases were further disclosed between gender (p = 0.043), the presence of precancerous lesions (p = 0.003; Figure 4), the occurrence of lymphnodal infiltration at time of surgery (p = 0.001), stromal expression of SDC2 (p = 0.015; Figure 4) and altered GPC6 protein expression in tumour cells (p = 0.01; Table 3). As predictable, “sentinel lymphnodes” at diagnosis significantly influenced the later appearance of more prominent lymphnodal lesions, but this factor was not considered in the multivariate logistic regression model because of its unuttered prognostic implication. Multivariate analysis corroborated that the occurrence of precancerous lesions (HR, 3.773, p = 0.005), and more incisively the presence of SDC2 in the stromal compartment (HR, 7.652, p = 0.007), but not GPC6 expression or gender of the patient, were independent prognostic markers for post-surgery secondary infiltrations of lymphnodes (Table 4; Figure 5). If we then accounted for both a history of precancerous lesions and SDC2 stromal expression, the probability to develop post-surgical lymphnodal infiltration was significantly increased (p = 0.001; Table 3; Figure 4).

Contrary to the above associations, univariate logistic analyses revealed that the N classification , at time of surgical removal of the primary tumour mass, and up-regulated transcription of SDC1 (Figure 4) were significantly associated with the formation of distant metastases ( p = 0.016 for both correlations; Table 3). The frequency of GPC6 expression in neoplastic cells did not satisfy the limit of significance (p = 0.052), but could be a potentially interesting indicator to take into account in future investigations on larger cohorts of patients. Conversely, multivariate analyses reinforced the impact of N classification (HR, 4.38, p = 0.012) and down-regulated or unaltered SDC1 expression (HR, 0.232, p = 0.013) as independent factors predicting the formation of distant metastases, albeit with opposite trends, (Table 4; Figure 5). When we next considered the combination of the unfavourable conditions represented by lymphnodal infiltration and up-regulation of SDC1 transcription, we unfolded a significantly increased probability to develop distant metastases (p = 0.004; Table 3; Figure 4). Noteworthy was also the fact that 91% of patients with up-regulated SDC1 transcription that developed distant metastases within the follow-up period invariably succumbed the disease.

Advanced T classification (p = 0.038), T3-T4 grouping (p = 0.043), positive N classification (p < 0.001; Figure 4), up-regulation of SDC1 transcription (p = 0.036) and stromal expression of SDC2 or GPC1 (p = 0.001 and p = 0.012 respectively; Table 3, Figure 4) were also found to be strongly associated with disease-related death. Cervical lymphnodal involvement (HR, 2.971, p = 0.005) and, more markedly, synthesis of SDC2 in the stromal cells (HR, 8.671, p = 0.003), established two independent predictors of survival (Table 4; Figure 5). The combination of these two conditions further decreased the survival probability of the patients (p < 0.001, Table 3; Figure 4).

In the evaluation of situations in which patients presented at any of the mentioned clinical outcomes, we similarly found a tight correlation between advanced T classification (p = 0.018), T3-T4 grouping (p = 0.005), positive N classification (p < 0.001; Figure 4), up-regulation of SDC1 mRNA (p = 0.013; Figure 4) and SDC2 stromal reactivity (p = 0.002). Finally, in multivariate analyses, N involvement (HR, 3.203, p < 0.001) and down-regulation or unaltered SDC1 expression (HR, 0.429, p = 0.012), but not T classification or SDC2 detection in stromal cells, were independent factors with opposite trends for the prediction of poor prognosis (Table 4; Figure 5). Even in this case the combination of positive N classification and up-regulation of SDC1 mRNA expression significantly increased the probability of the patients to incur into a dismal disease course (p < 0.001;Table 3; Figure 4).

We finally evaluated the disease course in patients scoring positively for the 3 dismal prognostic indicators, i.e. de novo expression of NG2/CSPG4, stromal abundance of SDC2 and up-regulation of SDC1 mRNA, which, in an independent manner, associated with one or more of the adverse clinical outcomes. This conditions was found in 36 of the 173 patients (21%) and within this patient subgroup 17% and 25%, respectively, developed loco-regional secondary lesions or distant metastases. Lymphnodal metastasis was observed in 11% of the patients, whereas 33% of them succumbed to the disease. Survival analyses revealed a strong association between the PG pattern analyzed and the presence of distant metastases (p < 0.002), disease-related deaths (p = 0.004) and a cumulative bad prognosis (p < 0.004). The Cox regression model ascertained that patients not expressing simultaneously the three bad independent disease course markers resulted to have a best prognosis in terms of putative development of distant metastases (HR, 0.146, p = 0.002), for survival (HR, 0.272, p = 0.004) and for incurring into any of the unfavourable clinical events under consideration (HR, 0.363, p = 0.003) (Table 4; Figure 5).