The Contrasting Role of p16Ink4A Patterns of Expression in Neuroendocrine and Non-Neuroendocrine Lung Tumors: A Comprehensive Analysis with Clinicopathologic and Molecular Correlations

Lung cancer encompasses a constellation of malignancies with no validated prognostic markers. p16Ink4A expression has been reported in different subtypes of lung cancers; however, its prognostic value is controversial. Here, we sought to investigate the clinical significance of p16Ink4A immunoexpression according to specific staining patterns and its operational implications. A total of 502 tumors, including 277 adenocarcinomas, 84 squamous cell carcinomas, 22 large cell carcinomas, 47 typical carcinoids, 12 atypical carcinoids, 28 large cell neuroendocrine carcinomas, and 32 small cell carcinomas were reviewed and subjected to immunohistochemical analysis for p16Ink4A and Ki67. The spectrum of p16Ink4A expression was annotated for each case as negative, sporadic, focal, or diffuse. Expression at immunohistochemical level showed intra-tumor homogeneity, regardless tumor histotype. Enrichments in cells expressing p16Ink4A were observed from lower- to higher-grade neuroendocrine malignancies, whereas a decrease was seen in poorly and undifferentiated non-neuroendocrine carcinomas. Tumor proliferation indices were higher in neuroendocrine tumors expressing p16Ink4A while non-neuroendocrine malignancies immunoreactive for p16Ink4A showed a decrease in Ki67-positive cells. Quantitative statistical analyses including each histotype and the p16Ink4A status confirmed the independent prognostic role of p16Ink4A expression, being a high-risk indicator in neuroendocrine tumors and a marker of good prognosis in non-neuroendocrine lung malignancies. In this study, we provide circumstantial evidence to suggest that the routinary assessment of p16Ink4A expression using a three-tiered scoring algorithm, even in a small biopsy, may constitute a reliable, reproducible, and cost-effective substrate for a more accurate risk stratification of each individual patient.


Introduction
Lung cancer represents a heterogeneous group of tumors, embracing distinct entities with different morphologies, molecular features, clinical outcomes, and responses to therapy [1]. The prognosis of these patients is generally poor, with a 5-year overall survival rate of approximately 17% only slightly improved in recent decades [2]. In this era of precision medicine, the search for reliable prognostic markers is of remarkable clinical importance; however, no protein marker has been sufficiently validated for clinical use in lung malignancies, since the results from more than five hundreds indexed studies are inconsistent and/or contradictory [3][4][5]. To date, tumor histotype and disease stage remain the most important prognostic factors, that drive clinical management and treatment decision [1].
Since p16 Ink4A first characterization in the early nineties as a cyclin-dependent kinase inhibitor [6] it has continued to gain widespread importance in a plethora of malignancies, including lung cancer [3,[7][8][9][10][11]. This tumor suppressor protein is encoded by the cyclin-dependent kinase inhibitor 2A (CDKN2A, 9p21) and regulates gene expression at different levels by modifying the functional equilibrium of transcription factors, microRNAs, post-transcriptional regulators and ultimately inhibiting transition of the cell cycle from G1 to S phase [12]. To date, the role of CDKN2A and its transcript is of burgeoning interest in target therapies. For example, oropharyngeal squamous cell carcinomas overexpressing p16 Ink4A respond more favorably to intensity-modulated radiotherapy treatment compared to p16 Ink4A -negative controls [13]. Furthermore, delivery of the whole p15/p16/p14ARF locus in bacterial artificial chromosomes and induction of p16 Ink4A using DNA methyltransferase inhibitors combined with histone deacetylation inhibitors result in the suppression of cell development in human glioma and lung epithelial tumor cells lines, respectively [14,15]. Overall, ectopic induction of p16 Ink4A in cancer cells inhibits cell growth and induces apoptosis and senescence, whereas CDKN2A silencing reduces the p53-mediated response to chemotherapeutic agents [16].
The association between p16 Ink4A overexpression and patients' survival in lung cancer has been widely investigated, and there are several lines of evidence to suggest that p16 Ink4A could play a part in lung cancer tumorigenesis [3,[17][18][19]. However, only a handful of previously published studies showed results potentially translatable into clinical practice, highlighting the complexity and the ambiguity underpinning the role of p16 Ink4A and p16 Ink4A -related pathways in lung tumors. At present, the diagnostic, prognostic and predictive values of this essential cell-cycle regulator remain controversial in the lung [20][21][22][23]. This could be related to the small sample size of previously published studies focusing only on single major histotypes of lung cancer, often individually, and the lack of correlation with clinical and molecular data. Furthermore, the lack of standardization and reproducibility of previously proposed p16 Ink4A scoring systems in lung malignancies, often restricted to a dichotomous record of positive versus negative cases might have limited the significance of the previously reported results [17].
In this scenario, the purpose of our study was to investigate the clinical implication of specific patterns of p16 Ink4A expression in a large series of lung malignancies using a three-tiered scoring algorithm. Given the histologic and molecular intra-and inter-tumor heterogeneity of lung tumors, and the biologic diversity of each histologic entity, we first sought to define p16 Ink4A -expression heterogeneity, and subsequently to characterize the prognostic significance of p16 Ink4A testing in lung cancers using a semi-quantitative score. Moreover, as a hypothesis-generating aim, we explored the correlation between p16 Ink4A status and Ki67 as well as the most common molecular aberrations in well-known targetable cancer genes, such as epidermal growth factor receptor (EGFR) and anaplastic lymphoma kinase (ALK).

Patients and tissue specimens
The study was conducted in accordance to local ethical guidelines. After obtaining institutional review board (IRB) approval (San Paolo IRB #10664/13; Fondazione IRCCS Ca' Granda IRB #179/13), the routinely prepared formalin-fixed paraffin-embedded (FFPE) tissue blocks and the archived slides of consecutive primary lung tumors were retrieved from the archives of the Department of Pathology, A.O. San Paolo Hospital, Milan, Italy and integrated with selected cases retrieved from the archives of the Department of Pathology, Fondazione IRCCS Ca' Granda-Ospedale Maggiore Policlinico, including rarer histotypes. All patients have provided signed consent forms. Taken together, 502 primary lung tumors, either surgical specimens (n = 383) or core biopsies (n = 119), diagnosed between 2000 and 2015, were included. The study group was composed of 277 adenocarcinomas (ADC), 84 squamous cell carcinomas (SCC), 22 large cell carcinomas (LCC), 47 typical carcinoids (TC), 12 atypical carcinoids (AC), 28 large cell neuroendocrine carcinomas (LCNEC), and 32 small cell carcinomas (SCLC). All cases were centrally reviewed by four pathologists (NF, ADG, EGR, and SF) that performed ex novo the histologic classification and pathologic staging, according to the latest editions of the WHO classification of tumors of the lung and of the AJCC staging system, respectively [1,24]. Specifically, the ADC group included glandular or solid tumors immunoreactive for thyroid transcription factor (TTF1) and/or napsin A; SCCs encompassed keratinizing and non-keratinizing p40-positive, TTF1-negative malignancies; undifferentiated non-small cell malignancies with "null" phenotype (i.e. lacking TTF1, napsin A, p40, synaptophysin, chromogranin, and CD56 expression) were recorded as LCC. Furthermore, the diagnostic criteria employed for subtyping tumors with neuroendocrine morphology included the number of mitoses per 2 mm 2 , the presence of necrosis, and Ki67 proliferation index [1]. The medical records were subsequently reviewed to obtain patients' data, including age at diagnosis, gender, smoking history, and survival data. Follow-up data were available for 444 patients until 2015 with a mean follow-up time of 31 months (range 1-144). Clinicopathologic features of the cases included in the study are summarized in Table 1 and detailed in S1 Table. Among 502 cases included in the study, 476 primary lung tumors (either surgical resections or core biopsies) were amenable for multiple sampling and therefore used to construct 15 FFPE tissue microarrays (TMA); the remaining cases were studied on conventional full-face sections. All samples were anonymized prior to processing.

Tissue microarrays construction
Using a semi-automatic arrayer (Alphelys Minicore2, Plaisir, France), 1-mm and 3-mm cores were randomly generated based on the amount of tumor tissue available as previously described [25]. Each TMA block contained up to 180 tumor cores with a total number of 2190 spots of neoplastic tissue. For each case, a mean of 4.7 tumor tissue cores (range 2 to 5 cores), including the tumor invasive front, was sampled and one spot of matched non-neoplastic lung tissue for each case was incorporated in the TMAs when available.
Immunostaining for p16 Ink4A was evaluated independently by two pathologists (NF and SF), and nuclear as well as cytoplasmic staining was considered a positive reaction, as described by Klaes et al. [10]. The pattern of p16 Ink4A positivity was therefore scored on a semi-quantitative scale for each single tissue core, as follows: negative (less than 1% of positive neoplastic cells), sporadic (isolated positive cells accounting for less than 5% of all neoplastic cells), focal (small clusters of positive cells accounting for less than 25% of all neoplastic cells) and diffuse (more than 25% of positive neoplastic cells) [10]. This system has shown higher levels of reproducibility and accurateness for p16 Ink4A scoring in gynecologic malignancies, compared to the semi-quantitative German score [27]. Ki67 was scored independently by three pathologists (NF, ADG and SF) as the percentage of cells with nuclear staining in at least 1000 neoplastic cells randomly selected over 10 high-power (magnification, 40X) fields [28]. The expression of biomarkers for EGFR mutations was evaluated as reported elsewhere [29], following the  [30]. For each immunohistochemical analysis, discordant results were resolved on a multi-headed microscope.

DNA extraction, PCR amplification, and Sanger sequencing
The presence of somatic alterations in EGFR exons 19 and 21 was validated by Sanger sequencing analysis in all cases displaying equivocal immunohistochemical stains, as recommended [29,30]. For this analysis, genomic DNA of each tumor and matched normal tissue was collected from the corresponding FFPE blocks and extracted using the DNeasy 1 Blood & Tissue Kit (Qiagen, Valencia, CA), according to the manufacturer's instructions. Primers set to amplify all coding regions of exon 19 (chromosomal position: chr.7: 55,174,722-55,174,820) and exon 21 (chromosomal position: chr.7: 55,191,719-55,191,874) of the EGFR gene were employed as described [31]. PCR amplification of 60 ng of genomic DNA was performed using the AmpliTaq 360 master mix (Life Technologies, Grand Island, NY). PCR fragments were purified and sequenced on an ABI3730 capillary sequencer as described [32]. Sequences of the forward and reverse strands were analyzed using SeqScape (version 2.5, Applied Biosystems, Carlsbad, CA).

Fluorescence in situ hybridization (FISH)
The presence of the most common inversion event occurring on the short arm of chromosome 2, resulting in the fusion of ALK with the echinoderm microtubule-associated protein-like 4 (EML4) gene loci, was investigated in all non-small cell lung carcinomas included in the study by means of fluorescence in situ hybridization (FISH) [1,30]. This analysis was performed with a two-color probe mix consisting of BACs for 5 0 ALK and 3 0 EML4 using validated protocols established at Istituto Nazionale Tumori, Naples, Italy [33]. At least 60 non-overlapping, interphase nuclei of morphologically unequivocal neoplastic cells were analyzed.

Statistical analysis
Differences in p16 Ink4A expression across tumor types were investigated using the Χ 2 test (MedCalc Software, Acacialaan, Ostend, Belgium). In order to assess the correlation between histotype and p16 Ink4A overexpression, a Cox regression analysis was performed assessing first the p16 Ink4A status of each case in a dual fashion (negative/positive). A second proportionalhazards regression analysis taking into account the various spectra of p16 Ink4A expression was subsequently performed. For the purpose of this work, carcinoids, either typical or atypical, LCNEC, and SCLC were identified as neuroendocrine tumors, whereas ADC, SCC, and LCC were defined non-neuroendocrine carcinomas; survival analysis was performed separately for these two broad groups. Correlation of the protein's patterns of expression to patients' overall and disease-free survival was assessed using the Cox proportional-hazards regression model; survival curves were built according to the Kaplan-Meier method (MedCalc software). Patients who died from causes other than lung cancer were excluded for survival analysis. Quantitative analyses were performed using multiple the Cox proportional hazards regression in order to assess the independence of p16 Ink4A as a prognostic factor [34]. Probability values (p) less than 0.05 were considered statistically significant. Furthermore, to assess the correlation between proliferative indices and p16 Ink4A overexpression, a differential Ki67 index (ΔKi67) between p16 Ink4A -negative and p16 Ink4A -positive cases was calculated for each histologic subtype, using a combinatorial analysis scheme [35]. This absolute value allows the detection of the variation in the average proliferation indices of each histotype according to p16 Ink4A status.
Primary lung tumors exhibit intra-tumor homogeneous p16 Ink4A expression Intra-tumor heterogeneity analysis of p16 Ink4A overexpression was performed for the subgroup of neoplasms incorporated in the TMAs that displayed any pattern of positivity (n = 177). Among them, 171 (96.6%) tumors showed immunoreactive neoplastic cells in each tumor core, whereas heterogeneous p16 Ink4A expression across tumor cores was restricted to 6 (3.4%) cases, namely 5 ADCs and 1 SCC, as detailed in Table 3. Of note, the p16 Ink4A expression pattern was homogeneous across all cores of the 66 tumors displaying diffuse positivity, suggesting p16 Ink4A expression in neuroendocrine lung tumors The frequency of neuroendocrine carcinomas expressing p16 Ink4A ranged from 14.9% to 100% ( Table 2, Fig 2). In particular, only a subset of 7 (14.9%) TCs showed sporadic staining pattern, as exemplified in Fig 1, (Fig 1), whereas 10 (35.7%) LNECs and 2 (6.3%) SCLCs, displayed sporadic or focal staining patterns. Overall, all SCLCs included in this study displayed immunohistochemical reactivity for p16 Ink4A , whereas in a subset of LCNECs (n = 4, 14.3%) no expression of p16 Ink4A was detected. Therefore, we observed progressive higher p16 Ink4A -positive tumor cells populations in higher-grade neuroendocrine malignancies (p <0.05, Anova) (Fig 2).
In agreement with the results reported above, the correlation between p16 Ink4A status and Ki67 index revealed a significantly higher tumor proliferation index in those cases expressing p16 Ink4A in at least a subset of neoplastic cells, with a progressive lower ΔKi67 values for TCs, ACs, and LNECs, respectively (Fig 3). A similar association was found for the two SCLCs with sporadic p16 Ink4A staining pattern compared to the SCLCs with strong and diffuse positivity (92.5% versus 94.4%). No significant correlation was identified between p16 Ink4A status and other well-known clinicopathologic parameters such as age, smoke status, stage, and lymph node status (p >0.05). p16 Ink4A overexpression was significantly associated with a shorter survival time in lung neuroendocrine tumors (HR = 0.08, 95% CI, 0.03 to 0.2, p <0.0001) (Fig 4A). Taking into account the various patterns of p16 Ink4A expression, a shorter time to recurrence (HR = 0.0826, 95% CI, 0.03 to 0.2, p <0.0001) and a decreased overall survival time (HR = 2.48, 95% CI, 1.8 to 3.4, p <0.0001) were observed in neuroendocrine malignancies overexpressing p16 Ink4A , irrespective of staining patterns (S1 Fig). These results were expected, since progressive higher p16 Ink4A expression was observed in the aggressive subtypes of neuroendocrine malignancies. However, univariate analyses including each histotype and the p16 Ink4A status confirmed that p16 Ink4A expression is an independent indicator of poor prognosis in neuroendocrine lung tumors (p ranging 0.001 to 0.01). Overall, the non-neuroendocrine lung carcinomas with p16 Ink4A immunoreactivity ranged from 13.6% to 40.1% (Table 2). Among them, 166 (59.9%) ADCs were p16 Ink4A -negative and 31 (11.2%) ADCs expressed p16 Ink4A in less than 5% of the neoplastic cells, whereas 47 (17.0%) tumors showed focal immunoreactivity, as exemplified in Fig 1; for this group, the diffuse pattern of p16 Ink4A staining was observed in 33 (11.9%) cases. Nine (10.7%) SCCs displayed sporadic or focal p16 Ink4A expression and the same number (10.7%) were diffusely immunoreactive, whereas most of the SCCs analyzed (78.6%) were completely negative (Fig 1). Among LCCs, 1 (4.6%) case displayed a focal staining pattern, 2 (10.0%) cases were diffusely immunoreactive, whereas 19 (86.4%) cases did not express p16 Ink4A . In contrast to neuroendocrine neoplasms, a significant decrease in the overall p16 Ink4A expression was observed for the In addition, decreased Ki67 indices were observed in p16 Ink4A -positive non-neuroendocrine carcinomas, as confirmed by the presence of inverted ΔKi67 values compared to neuroendocrine tumors (Fig 3). Correlation of p16 Ink4A status with other clinicopathologic parameters such as age, smoke status, stage, and lymph node status failed to reveal any statistical significance.
Interestingly, in patients with non-neuroendocrine lung malignancies, diffuse or focal p16 Ink4A -expression was a favorable prognostic factor (HR = 0.08, 95% CI, 0.03 to 0.2, p = 0.0123 by Cox proportional-hazards regression analysis). This association was also evident taking into account the tumor pathologic stage as a covariate in the Cox proportional-hazards regression analysis (p <0.05), where the survival analysis showed a triple-fashion plot (Fig 4B). Specifically, patients with non-neuroendocrine tumors displaying strong and diffuse p16 Ink4A overexpression had a considerable better outcome compared to p16 Ink4A -negative non-neuroendocrine carcinomas, while patients with tumors showing sporadic and focal staining patterns clustered together in an intermediate-risk group. These data were also confirmed separately for either SCCs (HR = 0.079, 95% CI, 0.03 to 0.211, p <0.001) or ADCs, showing decreased overall survival times in patients with tumors displaying patterns of p16 Ink4A expression ranging from negative to sporadic (p = 0.0123) (Fig 4C). At the molecular level, all lung carcinomas showing diffuse overexpression of p16 Ink4A and harboring genetic alterations involving EGFR and ALK had an excellent outcome, as represented in S2 Table.

Discussion
In this study, we have characterized the specific spectra of p16 Ink4A expression in a large series of neuroendocrine and non-neuroendocrine lung tumors using a three-tiered semi-quantitative score. Our findings provide strong circumstantial evidences to suggest that, despite the well-known histologic and molecular intra-tumor heterogeneity of lung malignancies, the expression of p16 Ink4A is homogeneous across each individual neoplastic entity, allowing the analysis of a small area of the tumor, such as a core biopsy, as representative of p16 Ink4A status of the entire lesion. Moreover, our observations corroborate the notion that the tumor suppressor protein p16 Ink4A plays a dichotomous role in the lung, being a high-risk feature in neuroendocrine tumors and a marker of good prognosis in non-neuroendocrine carcinomas. Finally, we have described a subset of lung carcinomas harboring EGFR and ALK mutations and overexpressing p16 Ink4A that shows better clinical outcome compared to p16 Ink4A -wild-type tumors.
Spatial heterogeneity analysis of p16 Ink4A expression in neuroendocrine and non-neuroendocrine tumors, revealed that p16 Ink4A immunoreactivity is homogeneous across different areas of lung tumors, with the same recurring pattern of expression. Since its discovery, the subcellular localization of p16 Ink4A has been a matter of great controversy [36]. Hence, although it is intuitive that p16 Ink4A acts as a cell cycle inhibitor in the nucleus, there are functional evidences to suggest that the cytoplasmic immunohistochemical expression represent the phenotypic evidence of protein inactivation similar to that observed in other tumor suppressor genes [36,37]. However, the operational implications of the assessment of defined p16 Ink4A expression patterns have not been taken into account in previous studies on lung cancer [4,20,22]. In particular, we reasoned that a dualistic immunohistochemical scoring system could be not representative of the actual p16 Ink4A status of each individual case, nor sufficiently reproducible. Indeed, it is probable that a considerable proportion of tumors (i.e. p16 Ink4A -sporadic and p16 Ink4A -focal lung neoplasms) has been clustered differently in previous works, determining a substantial heterogeneity in the reported results and therefore leading to a nihilistic and skeptical vision of p16 Ink4A as a trustable biomarker in the lung. For this reason, we decided to score the p16 Ink4A immunohistochemical nuclear and/or cytoplasmic expression using a semi-quantitative scale, as previously reported for gynecologic malignancies [10]. Hence, we observed that 17.5% of the tumors were strongly and diffusely immunoreactive for p16 Ink4A , whereas a submodal positivity was restricted to 22.3% tumors. It should be noted that p16 Ink4A staining pattern does not necessarily mirror the underlying status of CDKN2A; hence, negative immunostaining for p16 Ink4A can be seen in some tumors that lack any detectable copy number anomaly in chromosome 9, as assessed by comparative genomic hybridization [38]. Our observations have potential clinical implications, as the assessment of a specific pattern of p16 Ink4A expression in a small area of the tumor could provide information on the biological aggressiveness of each individual lung neoplasm.
The assessment of p16 Ink4A immunoreactivity in the different histotypes of lung tumor using a semi-quantitative scoring system showed a dichotomous scenario involving neuroendocrine and non-neuroendocrine entities. Our data provide further credence to the notion that these two broad and heterogeneous group of tumors are indeed genetically, and clinically, distinct diseases harboring distinct of mechanisms of p16 Ink4A alterations (i.e. methylation, hemi/ homozygous deletion, somatic mutations), as previously reported [18,19,39]. In particular, the number of cases overexpressing p16 Ink4A even in a submodal proportion of tumor cells, progressively increased from lower-to higher-grade neuroendocrine lung tumors, with percentages varying from 14.9% in TC to 100% in SCLC. Furthermore, the specific fraction of cancer cells immunoreactive for p16 Ink4A increased among higher grade neuroendocrine tumor. Our findings confirm that p16 Ink4A /retinoblastoma (RB) pathway is consistently compromised in LCNEC and SCLC, and less frequently compromised in AC [19]. However, in contrast with previous observations, we observed that in a subset of TC, p16 Ink4A display an aberrant pattern in a proportion of tumor cells (i.e. sporadic staining pattern in TCs and sporadic to focal staining pattern in ACs), suggesting that the immunohistochemical assessment of this tumor suppressor protein should be performed and interpreted carefully in the lung, recording every single positive cell. Intriguingly, p16 Ink4A seems to play an opposite role in lung non-neuroendocrine carcinomas, with the percentage of cases displaying the entire spectrum of p16 Ink4A positivity significantly higher in well/moderately differentiated tumors (41.3%) compared to poorly differentiated and undifferentiated malignancies (26.1%).These results confirm that CDKN2A deregulation through distinct mechanisms, including rare point mutations, promotor methylation and frequent homozygous deletions, is a relatively frequent event occurring in non-small cell lung cancer [18]. A mechanicistic explanation of the clear-cut dichotomous role of p16 Ink4A that we observed in lung neoplasms could be provided by recent molecular studies [40]. In particular, our data corroborate previous molecular and phylogenetic findings that could not identify a clonal relationship between neuroendocrine and non-neuroendocrine lung tumors [39,41]. Hence, the presence of completely different repertoires of somatic genetic alterations between neuroendocrine and non-neuroendocrine tumors of the lung suggests that these two group of neoplasms are, actually, distinct diseases [42,43]. Furthermore, specific genetic and epigenetic mechanisms are involved in CDKN2A instability, according to malignancies originating from distinct anatomical sites [15]. Previous studies demonstrated that promoter silencing of p16 Ink4A through methylation could lead to loss of control of the restriction point in the G1 phase of the cell cycle and favor cellular transformation in the lung [19,40]. Moreover, it has been speculated that the p16 Ink4A alteration in lung cancer is a rather early event that may accelerate tumor aggressiveness through inactivation of both the RB and p53 pathways and may link to a poor disease outcome [16,18,19,44]. To this end, large multicentric prospective and functional studies focusing on CDKN2A alterations in lung malignancies, including a representative number of rare entities, are warranted. Our observations are consistent with the notion that, while in neuroendocrine tumors p16 Ink4A status mirrors the biologic features of higher-grade neoplasms, the expression of this protein in lung non-neuroendocrine malignancies is related to earlier step of tumorigenesis. Based on our results, the assessment of p16 Ink4A status, even in a biopsy sample, could improve the characterization of lung tumors.
The correlation analysis between p16 Ink4A expression and the proliferative index confirmed the dualistic direction of p16 Ink4A immunoreactivity in lung neoplasms. In fact, each neuroendocrine histotype displaying p16 Ink4A overexpression had higher tumor proliferative indices, while p16 Ink4A -positive non-neuroendocrine malignancies showed lower Ki67 rates. These data suggest that the routinary assessment of both p16 Ink4A and Ki67 could provide significant information on the specific behavior of each individual histologic subtype of lung malignancies, allowing a fruitful patient's management.
The contrasting direction of p16 Ink4A effect in lung tumors is particularly noticeable in its prognostic value. Despite the great heterogeneity in the previously reported results on the prognostic significance of p16 Ink4A in lung malignancies, our analyses demonstrate that p16 Ink4A status, if assessed using an accurate semi-quantitative system at the immunohistochemical level, is an independent prognostic factor in lung tumors, being related to poorer prognosis in neuroendocrine entities, and to a more favorable outcome in non-neuroendocrine malignancies. As a matter of circumstance, tumors showing sporadic and focal staining patterns clustered together in an intermediate-risk group in both families of lung tumors. This observation is not trivial, given the evidence that the presence in a small biopsy of a submodal neoplastic population harboring p16 Ink4A immunohistochemical overexpression could indicate a worse outcome.
To our knowledge, this is the first study aiming to explore the correlation between p16 Ink4A protein expression and mutations in potentially targetable genes. In the small set of cases harboring molecular aberrations in EGFR and ALK included in this study, the null or heterogeneous p16 Ink4A immunohistochemical expression conferred a poor prognosis, suggesting that the assessment of p16 Ink4A protein status might help to tailor the prognostic evaluation of EGFR or ALK aberrant lung cancers. This study has several limitations. First, given the rarity of ACs, LNECs, and LCCs, we could analyze only a relatively small number of cases. It should be noted, however, that to the best of our knowledge this study represents the first comprehensive analysis of lung neoplasms that provides preliminary data on the prognostic significance of defined patterns of p16 Ink4A expression in ACs, LNECs, and LCCs. Second, additional markers mirroring the CDKN2A molecular pathways, including cyclin D1 and RB proteins, were not investigated due to the study design focused on the exploration of p16 Ink4A routinary testing using a specific scoring system. Our work, however, should be considered as hypothesis generating and validations in wider independent cohorts, with additional and comprehensive molecular data, are warranted.
Despite these limitations, our results suggest that each individual neoplasm of the lung exhibit intra-tumor homogeneous specific patterns of p16 Ink4A expression. This tumor suppressor protein plays a contrasting role, being an independent poor prognostic factor in neuroendocrine lung tumors and an independent good prognostic factor in non-neuroendocrine malignancies. In conclusion, p16 Ink4A semi-quantitative immunohistochemical analysis is a reliable, reproducible, and cost-effective testing that can potentially be employed to improve the characterization of lung neoplasms, allowing a more accurate patient's management. Large clinical prospective studies coupled with immunohistochemical and molecular analyses are warranted to delineate the ramifications of our findings for both pathologist and oncologists.

Author Contributions
Conceived and designed the experiments: NF SB SF. Performed the experiments: NF VV GE RF FZM. Analyzed the data: NF EGR ADG. Contributed reagents/materials/analysis tools: GB MN AP. Wrote the paper: NF EGR SB SF. Supervised the work: SB SF.