Tissue biopsies from 3,114 patients were retrospectively collected, and representative tumor specimens were selected and sectioned into 0.6 mm cores for incorporation into tissue microarrays (TMA) using standard procedures [14]. All tissue samples were collected with approval from the Yale Human Investigation Committee protocol #9505008219. Written informed consent or a waiver of consent was obtained from all patients with the approval of the Yale Human Investigation Committee. The following TMA cohorts were studied- (1) one multi-tumor array, YTMA (Yale TMA)-395 (containing 235 cores of 13 different tumors retrieved from year 1997–2015), (2) two NSCLC (non-small cell lung cancer) arrays, YTMA-423 and YTMA-553 (containing 287 and 296 cases retrieved from year 2011–2016 and 2017–2020, respectively), (3) two ovarian carcinoma arrays, YTMA-69 and YTMA-264 (containing 358 and 335 cases retrieved from year 1996–2003 and 1963–2001, respectively), (4) two CRC (colorectal carcinoma) arrays, YTMA-221 and YTMA-410 (containing 254 and 151 cases retrieved from year 2000–2005 and 2009–2017, respectively), (5) two breast carcinoma arrays, YTMA-489 and YTMA-499 (containing 263 and 183 cases retrieved from year 2011–2012 and 2013–2014, respectively), (6) two bladder carcinoma arrays, YTMA-361 and 497 (containing 230 and 42 cases retrieved from year 2005–2014 and 2015–2019, respectively), (7) two head and neck squamous cell carcinoma (HNSCC) arrays YTMA-465 and 579 (containing 204 and 247 cases retrieved from year 1998–2018, and 2019–2021, respectively), (8) one pancreatic carcinoma array, YTMA-454 (containing 238 cases retrieved from year 2010–2017) (Supplementary Table 1). The clinicopathological characteristics of each cohort are presented in supplementary Tables 2–7. A minimum of three slides of the above-mentioned tumor cohorts were stained using a QIF protocol (two serial tissue sections from the same block of TMA and a separate section from a different block to make it two-fold redundant) to assess the reproducibility and BCAM expression heterogeneity within the tumor. An index TMA (YTMA-558) with BCAM-positive and-negative cell line controls and a few tumor cores, including ovarian cancer, lung cancer (NSCLC), and CRC cases, was constructed for antibody validation and optimization purposes.

Five commercially available BCAM antibodies and five biotinylated antibodies provided by NextCure Inc. (10 antibodies in total) were used. The signal-to-noise ratio (SNR) was used as a measure of the QIF signal for each tested antibody. The antibody with the highest SNR was selected for further evaluation of BCAM protein expression in different tumor populations. Monoclonal anti-human BCAM antibodies included AB111181 (Abcam Inc., Cambridge, Massachusetts, USA), MCA 1982 (Bio-Rad Laboratories Inc., Hercules, California, USA), MM0107 (Novus Biologicals, LLC, Colorado, USA), and FQS5276 (Creative Diagnostics, Shirley, New York, USA). The polyclonal antibody HPA005654 (Sigma Aldrich, Inc., St. Louis, Missouri, USA) along with biotinylated antibodies NP095, NP638, NP639, NP640, and NP641 (Next Cure Inc., Maryland, USA) were also tested. The details of these antibodies are provided in the Supplementary Information. The optimal working concentration of each antibody was determined by titration using a log2 range serial dilution for each antibody. The antibody concentration with the highest SNR in the index array YTMA-558 was considered optimal. Antibody validation for protein expression was performed according to the previously formulated standards [15].

The protocol for the QIF assay, including sample preparation, blocking, primary antibody incubation, secondary antibody incubation, and imaging, has been previously documented [15]. Briefly, TMA slides were deparaffinized in xylene and rehydrated in graded ethanol baths. Heat-induced antigen retrieval was performed using a citrate buffer (pH 6.0) at 97 ℃ for 20 min. Permeabilization and blocking of endogenous peroxidases were performed by incubating the slides for 30 min in 0.3% hydrogen peroxide in methanol and then in a blocking solution of 0.3% bovine serum albumin and 0.05% tween-20 for 30 min [15]. Tissue slides were then incubated at room temperature for 1 h with a primary antibody mixture consisting of anti-BCAM (AB111181), which was the target antibody, and anti-cytokeratin (clone AE1/AE3, Dako), an epithelial marker that was used to mark the tumor area later for analysis, followed by a secondary antibody mixture consisting of anti-rabbit Envision (Dako) and anti-mouse Alexa Fluor 546 (Invitrogen). Subsequently, tyramide cyanine 5 (PerkinElmer) was used for signal amplification, followed by nuclear staining with DAPI (4′,6-diamidino-2-phenylindole). The slides were rinsed in TBST and TBS buffer before and after incubation. Prolonged gold antifade reagent (Invitrogen) was used for slide mounting. A more detailed description of the staining protocol is provided in Supplementary Table 8.

TMA slides from five tumor types, including ovarian, lung (NSCLC), breast, bladder, and HNSCC, were stained for PD-L1 using E1L3N clone, a rabbit monoclonal antibody (Cell Signaling Technology, CST, Massachusetts, USA), following the above-mentioned QIF staining protocol. Antigen retrieval was performed at pH 8.0. using ethylenediaminetetraacetic acid (EDTA) buffer. The optimal concentration of the antibody used was 1.1 µg/ml as previously established in a standardized staining protocol [16]. Details of the staining protocol are provided in the (supplementary information Table 9).

AQUA (Automated Quantitative Analysis) version 3.0, a software tool for digital pathology, was used to measure BCAM protein expression on fluorescence images acquired using a PM-2000 microscope system (Navigate BioPharma) according to standard operating procedure [17]. The total compartment of all cells generated by thresholding the DAPI signal. Target antibody-BCAM was measured within the tumor mask/compartment defined by masking cytokeratin generated in Alexa 546 channel. Stromal compartment was created by excluding the tumor compartment from the total. Expression of the protein in tumor tissues was measured based on AQUA scores generated for each protein by dividing the sum of target pixel staining intensities by the area of the designated compartment. The final value or measurement of signal/staining intensity of the antibody in tumor compartment was calculated by averaging the AQUA scores generated for each tumor spot in the three slides (two serial sections cut from same block of the tumor and a separate section/slide cut from another block of the TMA). Exclusion criteria of tissues from analysis were spots with inadequate tumor tissue, missing spots/exhausted tissue, or artifacts such as tissue folding and air bubbles. Cases with no clinical information were excluded from survival outcome analysis as well.

Statistical analyses were performed using GraphPad Prism 9.2.0 (GraphPad Software Inc., CA, USA). For all tumor types and cohorts used in this study, we estimated the prognostic value of BCAM using the Kaplan–Meier product-limit method and compared it using the log-rank test in GraphPad Prism. We used the median as the cut-off to divide the BCAM-expressing tumors into two groups (Low BCAM group ≤ median, High BCAM group > median). We also established a visual cut-off point for positive BCAM-stained cases by inspecting all cases and selecting the one where minimal membranous staining could be recognized for each cohort. This allowed for the characterization of the proportion of positive BCAM cases and was used as an alternative cutoff point. Comparisons were performed using the log-rank test. All statistical tests were two-sided, with a level of significance < 0.05. A linear regression model was used to compare the correlation coefficient (R2 value) between the expression of PD-L1 and BCAM, as well as to plot the reproducibility between the serial sections and intratumoral heterogeneity.