A total of 164 PC tumour patients operated on between 1990 and 2020 at Helsinki University Hospital, Finland, were included in this retrospective study. Most of these tumours were TCs (n = 123), and the rest were ACs (n = 41). In all, 30 of the tumours had metastasized, and of these, 20 showed lymph node involvement at the time of diagnosis. Formalin-fixed and paraffin-embedded (FFPE) primary tumour samples were retrieved through the Helsinki Biobank. Associated clinical information was collected from patient records (Table 1).

Each tumour was re-evaluated from diagnostic whole slides by an experienced pulmonary pathologist following the 2021 classification of pulmonary neuroendocrine tumours by the WHO [2]. Neuroendocrine differentiation and epithelial origin were confirmed by routine immunohistochemical labelling for chromogranin A, synaptophysin, and pan-cytokeratin.

This study was approved by the HUS Regional Committee on Medical Research Ethics (HUS/1258/2020). As the Finnish Biobank Act provides a lawful basis for research use of patient samples and clinical data, a project-specific consent was not obtained from the patients who had given a biobank consent (n = 146). Eighteen patients had not given a biobank consent, but their samples and data were used based on the statements of the Finnish Medicines Agency Fimea (FIMEA/2020/002937) and the Finnish Social and Health Data Permit Authority Findata (THL/4427/14.02.00/2020).

Fresh slides were sectioned from the original representative tissue blocks, stained with haematoxylin and eosin (H&E), and digitized with a Pannoramic II (samples 1990–2015) or Pannoramic 250 Flash III (samples 2016–2020) (3DHISTECH, Budapest, Hungary) microscope slide scanner using a 20 × objective. Annotations for the TMAs were marked on the digitized slides with the CaseViewer 2.4 software (3DHISTECH) in accordance with the following principles: two cores from the middle of the tumour, two cores from the tumour border, two cores from the non-tumour area, and one core from the bronchus, if applicable. The TMAs were constructed with an automated TMA Grand Master tissue microarrayer (3DHISTECH) using 1 mm punches.

OTP immunohistochemical stainings were performed with a semiautomated AutoStainer instrument (Lab Vision Corp., Fremont, CA, USA). After deparaffinization, a heat-induced antigen retrieval at pH 6 was used before incubating the TMA sections with three different primary OTP antibodies: polyclonal OTP from Sigma-Aldrich (HPA059342, dilution 1:200) and two monoclonal OTP antibodies from Atlas Antibodies (CL11222 [AMAb91695] and CL11225 [AMAb91696], dilutions 1:200). EnVision FLEX + Mouse (LINKER) (Dako, Agilent Pathology Solutions, Santa Clara, CA, USA) was used for signal amplification for both monoclonal antibodies. Antibody binding was visualized with EnVision FLEX kit (Dako). All sections were counterstained with haematoxylin.

Ki-67 immunohistochemical staining was performed in the HUS Diagnostic Centre, Department of Pathology (Helsinki, Finland) with an automated BenchMark ULTRA (Ventana Medical Systems, Inc., Tucson, AZ, USA) instrument. Primary antibody clone MIB-1 (Dako) was used with CC1 standard pretreatment.

Immunohistochemically stained TMA slides were imaged with a Pannoramic 250 Flash III (3DHISTECH) digital slide scanner. A 20 × objective was used for brightfield imaging. Digitized slides were scored manually by two independent raters (J.N. and S.M.) blinded to patients’ clinical data, using a SlideViewer 2.6 (3DHISTECH) software and Aiforia® platform (Aiforia Technologies, Helsinki, Finland).

Nuclear expression of the OTP protein was evaluated by staining intensity on a scale from 0 to 3 (0 = negative, 1 = weak, 2 = moderate, and 3 = strong) (Fig. 1) and by the proportion of positive tumour cells (0%, 25%, 50%, 75%, and 100%) in each tumour core [20]. In the evaluation of heterogeneously stained tumour cores, a threshold of 40% was used to interpret the intensity. The total H-score was calculated from 2–4 different tumour cores for each sample by multiplying the staining intensity and the proportion of positive tumour cells in each core and averaging them. The staining result was classified as positive with H-score of ≥ 50, following the scoring criteria presented by Moonen et al. [20]. The results of the two raters were averaged to obtain consensus H-scores.

Scoring intensities for immunohistochemical OTP stainings: a OTP pAb, b CL11222 mAb, and c CL11225 mAb. Images were obtained with SlideViewer 2.6 (3DHISTECH, Budapest, Hungary) software with 40 × magnification (scale bar 50 µm). 0 = negative, 1 = weak, 2 = moderate, 3 = strong. OTP, orthopaedia homeobox protein; pAb, polyclonal antibody; mAb, monoclonal antibody

Ki-67 PI was analysed with deep-learning-based Aiforia software (Aiforia Technologies), as described earlier [21]. The highest Ki-67 PI of four parallel TMA spots per tumour was used in further statistical analysis.

The association between OTP status and dichotomous clinicopathological variables was calculated using Fisher's exact test, whereas the association between continuous variables was analysed with the Mann–Whitney U test. The level of concordance between the H-scores obtained with three different OTP antibody clones was tested with intraclass correlation coefficient (ICC) two-way mixed model using the definition of absolute agreement. Agreement between raters was assessed accordingly. Cohen's kappa was calculated for dichotomous data derived from consensus H-scores.

The association of OTP expression with disease-specific survival (DSS) was evaluated using Kaplan‒Meier survival analysis with a log-rank test. DSS was determined as the time elapsed between the primary tumour surgery and death caused by PC disease. Survival probabilities were also assessed using time to progression (TTP). TTP was defined as the time elapsed from the day of the primary tumour surgery to the time of disease progression (metastasis or tumour recurrence) or death from PC tumour.

Univariable Cox survival regression analysis was applied to obtain hazard ratios (HRs) and 95% confidence intervals (CIs) for OTP status and other factors contributing to survival prognosis. Based on the univariable analysis, a multivariable Cox regression model was calculated for the most significant risk factors to evaluate the effect of other variables on the risk caused by OTP status in the patient's outcome. Testing of the Cox model assumption of a constant HR over time involved plotting the Schoenfeld residuals over time and testing for a correlation, with no relevant non-proportionality of HRs identified. The possibility of interaction terms was explored; no interactions were identified.

Two-tailed tests were used and p values less than 0.05 were considered statistically significant. Analyses were performed with IBM SPSS Statistics for Windows, version 29.0 (IBM Corp., Armonk, NY, USA).