Characterization of undifferentiated carcinomas of the pancreas with and without osteoclast-like giant cells

Abstract

Background

Undifferentiated carcinoma (UC) is a rare subtype of pancreatic cancer distinguished from UC with osteoclast-like giant cells (UC-OGC) in 2019, affecting interpretation of literature that does not distinguish these subtypes. We sought to identify translationally relevant differences between these 2 variants and compared with pancreatic ductal adenocarcinoma.

Methods

We characterized clinical and multiomic differences between UC (n = 32) and UC-OGC (n = 15) using DNA sequencing, RNA sequencing, and multiplex immunofluorescence and compared these findings with pancreatic ductal adenocarcinoma.

Results

Characteristics at diagnosis were similar between UC and UC-OGC, though the latter was more resectable (P = .009). Across all stages, median overall survival was shorter for UC than for UC-OGC (0.4 years vs 10.8 years, respectively; P = .003). This shorter survival was retained after stratification by resection, albeit without statistical significance (1.8 years vs 11.9 years, respectively; P = .08). In a subset of patients with available tissue, the genomic landscape was similar among UC (n = 9), UC-OGC (n = 5), and pancreatic ductal adenocarcinoma (n = 159). Bulk RNA sequencing was deconvoluted and, along with multiplex immunofluorescence in UC (n = 13), UC-OGC (n = 5), and pancreatic ductal adenocarcinoma (n = 16), demonstrated statistically significantly increased antigen-presenting cells, including M2 macrophages and natural killer cells, and decreased cytotoxic and regulatory T cells in UC and UC-OGC vs pancreatic ductal adenocarcinoma. Findings were similar between UC and UC-OGC , except for decreased regulatory T cells in UC-OGC (P = .04).

Conclusions

In this series, UC was more aggressive than UC-OGC, with these variants having more antigen-presenting cells and fewer regulatory T cells than pancreatic ductal adenocarcinoma, suggesting potential for immune-modulating therapies in the treatment of these pancreatic cancer subtypes.

Pancreatic cancer is among the most lethal malignancies, with a 5-year overall survival rate of 12%¹. Despite an increase in novel treatment options in oncology, including targeted and immune therapies, pancreatic cancer remains largely refractory to these interventions². The most common subtype is pancreatic ductal adenocarcinoma, which accounts for approximately 85% of all pancreatic cancers³. As a result, most clinical trials focus on treatment of pancreatic ductal adenocarcinoma and contribute to the void in understanding the biology, treatment response, and prognosis of less common variants⁴.

Initially, the World Health Organization designated 1 variant group of undifferentiated pancreatic carcinomas⁵. In 2010 and 2019, revisions to the World Health Organization classification system defined 8 distinct subtypes: colloid (mucinous noncystic), medullary, adenosquamous, rhabdoid, undifferentiated (anaplastic) carcinoma (UC), undifferentiated carcinomas with osteoclast-like giant cells (UC-OGC), hepatoid, and signet ring/poorly cohesive^4,6. Discord remains in the field as to whether some of these histologic variations should be considered distinct entities or if they represent nested subsets of UCs with differing biologic behavior⁴. For example, undifferentiated sarcomatoid carcinomas have been studied as a specific subtype of UC, with similar genetic background but different expression profiles that suggest the possibility of distinct responses to immune-mediated therapeutic interventions such as anti–programmed cell death 1 or ligand 1 inhibitors^7,8. Conversely, undifferentiated rhabdoid carcinomas have distinct patterns of genomic alterations compared with other UCs, with unknown implications for therapeutic response⁹. Efforts are ongoing to elucidate the molecular details of each specific histologic subtype, although information remains limited due to their rarity.

The definition of UC has evolved to include the absence of osteoclast-like giant cells, thus establishing at least 2 distinct histologic subtypes that were previously described together⁴. Prior retrospective studies have reported variable prognostic data, possibly because of the combination of both subtypes and limited cohort size^3,10. Retrospective studies have also attempted to distinguish the clinical features of UC, with and without UC-OGC, from pancreatic ductal adenocarcinoma and reported the variants to be more prevalent in younger patients and female patients (2:1 ratio) and to present with advanced disease and harbor a worse prognosis^11–13. With recent advancements in molecular analytics, it is important not only to further discriminate the clinical features of these distinct entities but also to further evaluate the differences in the tumor immune microenvironment between UC and UC-OGC and compared with pancreatic ductal adenocarcinoma, which may have implications for tumor biology and treatment response.

Methods

Study design and cohort identification

This single-center, retrospective cohort study was conducted with a waiver of consent under 2 institutional review board–approved applications. Patients aged 18 years or older with pancreatic carcinoma diagnosed between January 2005 and May 2021 and with records at the University of Michigan were identified by International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) codes 157 and C25, respectively. As there are no ICD codes specific to UC of the pancreas, patients’ structured and unstructured health records were then searched using the Electronic Medical Record Search Engine¹⁴ for terms indicating UC. Following identification, pathology slides were reviewed by a board-certified gastrointestinal pathologist (J.S.) for confirmation of the diagnosis and subtype according to the 2019 World Health Organization guidelines¹⁵ (Figure 1). Tissues from archived formalin-fixed, paraffin-embedded blocks underwent assessment for tissue integrity before DNA and RNA extraction as well as sectioning for multiplex immunofluorescence.

Figure 1.

CONSORT diagram showing patient selection and study overview. ^aOf the samples that underwent bulk RNA sequencing, 2 from each tumor type did not pass quality control analysis and were excluded from the final analysis. TCGA = The Cancer Genome Atlas.

Data collection

Data were collected on demographics, pathology, disease stage, comorbidities, treatment, and survival by manual abstraction of the electronic health records. Disease stage or resectability status was established by review of imaging reports in the electronic health record using the National Comprehensive Cancer Network, version 2.2023, guidelines¹⁶ as well as tumor board and surgical evaluation notes. Overall survival was defined as the time from pathological diagnosis to date of death from any cause or censored at last available date of contact. Comorbidities were based on diagnoses at the time of presentation. Heavy alcohol use was defined as patient-reported consumption of more than 7 drinks per week for women or more than 14 drinks per week for men according to the Centers for Disease Control and Prevention guidelines¹⁷.

DNA and RNA extraction

Available hematoxylin-eosin slides from all cases of UC, UC-OGC, and a matched set of pancreatic ductal adenocarcinoma cases were reviewed by an experienced gastrointestinal pathologist (J.S.) to identify corresponding formalin-fixed, paraffin-embedded blocks containing sufficient tumor tissue for next-generation sequencing. For each case, representative flash frozen and paraffin-embedded tissue was manually scraped from glass slides under a dissecting microscope, and DNA and RNA were extracted using the AllPrep DNA/RNA FFPE Kit (Qiagen).

Targeted next-generation sequencing and analysis

Next-generation sequencing libraries were constructed from up to 20 ng of extracted formalin-fixed, paraffin-embedded DNA using the Oncomine Comprehensive Assay Plus (Thermo Fisher Scientific), a commercially available AmpliSeq DNA panel targeting more than 500 genes commonly altered in human cancers or up to 20 ng of extracted formalin-fixed, paraffin-embedded RNA using the Ion AmpliSeq Transcriptome Human Gene Expression Kit (Thermo Fisher Scientific), a commercially available AmpliSeq RNA panel targeting more than 20 000 human RefSeq genes, templated using an Ion Chef System (Thermo Fisher Scientific) and sequenced using an Ion GeneStudio S5 Prime System (Thermo Fisher Scientific). All DNA libraries were sequenced to at least 500× mean depth with more than 80% uniformity, and all RNA libraries had at least 6 million mapped reads. Next-generation sequencing reads were processed, aligned, and analyzed using Ion Torrent Suite software (Thermo Fisher Scientific), cloud-based standard Ion Reporter workflows (Thermo Fisher Scientific), and/or custom in-house bioinformatic pipelines^18,19. Annotated variants were filtered using standard criteria and manually curated by an experienced molecular pathologist (A.M.U.). Prioritized variations were visualized using the OncoPrinter tool (cBioPortal)^20,21. Raw RNA sequencing read counts were obtained from the coverageAnalysis plug-in for the Ion Torrent Suite and for all pancreatic ductal adenocarcinoma cases in The Cancer Genome Atlas (TCGA) pancreatic cancer cohort (ie, pancreatic adenocarcinoma) from the University of California, Santa Cruz Xena Cancer Browser website (xena.ucsc.edu)²² and normalized using the Trimmed Mean of the M-values approach. Differential expression analysis was performed using the edgeR package in R (R Foundation for Statistical Computing)²³. Two samples from each of the UC and UC-OGC groups did not pass the accepted quality control measures and were excluded from final analysis.

Digital cytometry

Digital cytometry was performed as described previously using raw RNA sequencing data and online web-based CIBERSORTx tools (cibersortx.stanford.edu) using the standard LM22 signature matrix file^24-26. Cell type enumeration was performed using the Impute Cell Fractions module with batch correction = enabled; batch correction mode = B-mode (LM22 matrix); disable quantile normalization = true; run mode = absolute; and permutations = 100. Samples with deconvolution P greater than .05 were excluded from subsequent analyses.

Multiplex immunofluorescence

Sections from archived tissue blocks identified by J.S. were processed as previously described²⁷. Briefly, blocks were cut into 5-µm slices and placed onto charged slides, baked in a hybridization chamber for 1 hour at 60 °C, subjected to deparaffinization and rehydration, then fixed with formalin. Following an established protocol²⁸, multiplex staining was conducted on the slides through 6 successive rounds of staining. Slides were prepared for staining using either an antigen retrieval buffer with pH 9 or pH 6 (AR9 and AR6, Akoya Biosciences) preceded by a primary antibody (CD3, CD8, CD163, pancytokeratin, and FoxP3), followed by secondary antibody application (Opal polymer, Akoya Biosciences) and fluorescent tyramide signal amplification (Akoya Biosciences). Slides were counterstained with Spectral 40 and DAPI and imaged with the Mantra Quantitative Pathology WorkStation (Akoya Biosciences) at 20× magnification in all channels: DAPI, FITC, CY3, CY5, CY7, Texas Red, and Qdot, with an exposure of 250 milliseconds.

Statistics

All statistics were performed using packages in R, version 4.3.1. Continuous and categorical variables were compared with the t test and χ² test, respectively. Cox proportional hazards regression was implemented to estimate hazard ratios and 95% CIs describing risk of death. Models were unadjusted because of limited data. Kaplan-Meier curves visually represent unadjusted, median survival differences. P values from bulk RNA sequencing differential expression analyses were adjusted for multiple hypothesis testing using the Benjamini-Hochberg method, and only genes with an adjusted P < .05 were considered statistically significant. Absolute cell type enumeration estimates were compared among samples using a standard nonparametric approach (ie, Kruskal-Wallis), with manual Bonferroni adjustment for multiple hypothesis testing.

Results

Demographics and presenting symptoms were similar in patients with UC and UC-OGC

Following selection of the patients as described (Figure 1), the data were sorted by pathology type: UC (n = 32) and UC-OGC (n = 15). None of the specimens contained sarcomatoid features, rhabdoid features, or features of other histological variants. The characteristics of patients in the 2 groups are presented in Table 1 and were similar. Patients were not more likely to be female in either group. The known risk factors for pancreatic ductal adenocarcinoma and patient comorbidities were distributed similarly between patients with UC and UC-OGC, although UC trended toward a higher percentage of individuals who smoke (P = .29). Similarly, presenting symptoms (jaundice, abdominal pain), select laboratory measures, and initial cancer-related mortality of UC and UC-OGC did not demonstrate statistically significant differences (Supplementary Table 1, available online).

Table 1.

Patient characteristics at diagnosis

Variable	Undifferentiated carcinoma (n = 32)	Undifferentiated carcinoma with osteoclast-like giant cells (n = 15)	P
Age, median (IQR), y	65 (57-73)	70 (65-75)	.12
Female sex, No. (%)	17 (53.1)	9 (60.0)	.89
Race,a No. (%)			.33
African American	2 (6.2)	1 (6.7)
Asian	0 (0)	1 (6.7)
White	30 (93.8)	13 (86.7)
Tobacco smoking status, No. (%)			.19
Never	14 (43.8)	10 (66.7)
Current	7 (21.9)	0 (0)
Former	10 (31.2)	5 (33.3)
Tobacco smoking pack-years, median (IQR)	8 (0-30)	0 (0-3)	.22
Alcohol consumption, No. (%)			.27
Never	12 (37.5)	5 (33.3)
Current	14 (43.8)	10 (66.7)
Former	5 (15.6)	0 (0)
Alcohol consumed per week, median (IQR), oz	0.0 (0.0-0.8)	0.6 (0-0.6)	.19
Ever diagnosed with, No. (%)
Diabetes	9 (28.1)	6 (40.0)	.63
Pancreatitis	8 (25.0)	3 (20.0)	.99

^aSelf-reported.

Patients with UC-OGC had earlier-stage disease at diagnosis than patients with UC

Characteristics of the primary tumors in the UC and UC-OGC cohorts are shown in Table 2. Tumor size was similar, although notably, there were outliers, with a 17 cm tumor as well as 2 other tumors larger than 10 cm in the UC-OGC group. The median tumor size in the cohorts presented here was 4.0 cm (range = 1.7-12.3 cm) for UC and 3.5 cm (range = 1.3-17.0 cm) for UC-OGC. Distribution of primary tumor location within the pancreas was similar between the tumor subtypes.

Table 2.

Tumor characteristics at diagnosis^a

Variable	Undifferentiated carcinoma (n = 32)	Undifferentiated carcinoma with osteoclast-like giant cells (n = 15)	P
Radiologic tumor size, median (IQR), cm	4.0 (3.0-7.4)	3.5 (2.6-9.5)	.53
Cancer antigen 19-9, median (IQR), U/mL	49.0 (18.0-149.7)	24.0 (13.2-51.8)	.58
Tumor site, No, (%)			.33
Ampulla	1 (3.1)	1 (6.7)
Body	9 (28.1)	5 (33.3)
Head or neck	16 (50.0)	9 (60.0)
Tail	6 (18.8)	0 (0)
Disease stage, No. (%)			.009
Resectable	13 (40.6)	12 (80.0)
Borderline resectable	2 (6.2)	2 (13.3)
Unresectable	17 (53.1)	1 (6.7)
Distant metastasis location, No. (%)			.39
Abdominal cavity	1 (3.1)	0 (0)
Liver	6 (18.8)	1 (6.7)
Lungs	3 (9.4)	1 (6.7)
Multiple	6 (18.8)	1 (6.7)
None	16 (50.0)	12 (80.0)
Cause of death, No. (%)			.006
Progression of malignancy	25 (78.1)	5 (33.3)
N/A, alive at analysis	6 (18.8)	6 (40.0)
Other	1 (3.1)	4 (26.7)
Resected, No. (%)	13 (40.6)	13 (86.7)	.008
Pathologic tumor size, median (IQR), cm	4.5 (2.7-5.3)	4.2 (3.1-11.3)	.10
Operation, No. (%)			.69
Pancreaticoduodenectomy	7 (53.8)	9 (69.2)
Pancreatectomy	6 (46.2)	4 (30.8)
Outcome, No. (%)			.18
R0	10 (76.9)	13 (100.0)
R1	2 (15.4)	0 (0)
R2	1 (7.7)	0 (0)

^aBold values denote statistical significance. N/A, not applicable; R0, resection margin negative; R1, resection margin microscopically positive; R2, resection margin macroscopically positive.

SI conversion factor: To convert cancer antigen 19-9 to kU/L, mulitply by 1.

A majority of patients (87%) with UC-OGC presented with localized, resectable disease, whereas less than half of the patients (40.6%) with UC had resectable disease, in large part because 50% of the patients with UC presented with distant metastasis compared with 13% of patients with UC-OGC. Sites of metastases in UC included lungs, liver, distant lymph nodes, abdominal cavity/peritoneum, and bone; 5 of 16 patients had multiple sites of metastasis at presentation. The 2 patients with UC-OGC who had distant disease had liver (n = 1) and lung (n = 1) metastases.

Patients with UC-OGC were more likely to undergo resection and receive systemic therapy

Surgical interventions in patients with UC and patients with UC-OGC are summarized in Table 2. Sixteen patients with UC (12 resectable, 1 borderline, 3 unresectable) underwent a surgical intervention. Two patients underwent operations for bowel perforations; 1 patient underwent celiac artery aneurysm repair; and the remaining 13 patients underwent tumor resections, either pancreaticoduodenectomy (n = 7) or distal pancreatectomy (n = 6). Two additional patients with localized disease (1 resectable, 1 borderline) were judged to have inoperable disease for medical reasons. Within the UC-OGC group, 13 patients (12 with resectable disease and 1 with borderline disease) underwent resection with pancreaticoduodenectomy (n = 9) or distal pancreatectomy (n = 4). The difference in the proportion of patients who underwent resection was statistically significant between the 2 groups (50% for UC and 87% for UC-OGC; P = .008). Outcomes with surgical intervention were favorable in UC-OGC, with all 13 patients achieving R0 resection (Table 2). In UC, however, only 10 of the 13 patients achieved R0 resection; 2 had R1 resection, and 1 had R2 resection.

Because little histology-specific data exist for management of UC and UC-OGC and because rates of recurrence are high in pancreatic ductal adenocarcinoma, most patients in our cohorts underwent adjuvant therapy (61.5% and 77% for resected UC and UC-OGC, respectively). Following surgery, 5 patients with UC (38.5%) and 3 patients with UC-OGC (23%) underwent active surveillance rather than adjuvant therapy. None of the patients in either cohort received neoadjuvant therapy or therapy targeted toward cytotoxic T-lymphocyte–associated antigen 4, programmed cell death 1 protein, or programmed cell death 1 ligand 1. Patients with unresected disease received palliative systemic therapy in both groups (22% and 13% for UC and UC-OGC, respectively), and 12 (37.5%) patients with UC notably had best supportive care as the initial intervention, whereas all patients with UC-OGC had at least 1 cancer-directed therapy (Figure 2, A). Individual therapy regimens were highly variable (Supplementary Table 2, available online).

Figure 2.

Overall survival in patients with UC-OGC compared with UC. A) Pie chart depicting the percentage of patients with UC (left) vs UC-OGC (right) receiving each initial intervention. B) Kaplan-Meier survival curve for all patients with UC (n = 32, median overall survival = 0.40 years [4.8 months]) and UC-OGC (n = 15, median overall survival = 10.82 years [131.6 months]). C) Kaplan-Meier survival curve for patients with resected UC (n = 13, median overall survival = 1.84 years [22.4 months]) and UC-OGC (n = 13, median overall survival = 11.92 years [145.0 months]). D) Kaplan-Meier survival curve for patients with unresected UC (n = 19, median overall survival = 0.17 years [2.1 months]) and UC-OGC (n = 2, median overall survival = 0.74 years [9.0 months]). HR = hazard ratio; UC = undifferentiated carcinoma; UC-OGC = undifferentiated carcinoma with osteoclast-like giant cells.

Undifferentiated carcinoma with osteoclast-like giant cells correlates with longer survival than  UC

Patients with UC and UC-OGC had dramatically different outcomes. At the time of data collection, 27 of the 32 (84.4%) patients with UC died from their malignancy compared with only 5 patients (33.3%) with UC-OGC (Table 2). Median overall survival for the entire UC cohort was 0.4 years (range = 0.02-9.37) and 10.8 years (range = 0.13-17.0) in the UC-OGC cohort (P = .003) (Figure 2, B). Four of the 9 deceased patients with UC-OGC died from a cause unrelated to pancreatic cancer or its treatment.

To evaluate the contribution of surgery to survival outcomes, median overall survival was determined for patients who underwent resection compared with those patients with advanced disease at presentation (Figure 2, C and D, respectively). For resected UC (n = 13), median overall survival was 1.8 years (range = 0.24-9.37) vs 11.9 years (range = 0.13-17.0) for resected UC-OGC (n = 13; P = .08). Median overall survival for advanced UC (n = 19) was 2.1 months (range = 0.26-22.21) vs 9.0 months (range = 3.28-14.45) for advanced UC-OGC (n = 2), although this difference was not statistically significant, likely because of the small number of patients with advanced UC-OGC.

Undifferentiated pancreatic carcinomas have similar genomic alterations and transcriptomes

To investigate the factors contributing to the difference in survival between UC and UC-OGC, we assessed the histological, genomic, transcriptomic, and composition of the tumor immune milieu. Representative images of pancreatic ductal adenocarcinoma, UC, and UC-OGC histological sections are shown in Figure 3, A. Genomic analysis of tumors demonstrated KRAS, TP53, and copy number alterations across the genome, with overall genomic diversity similar to that seen in pancreatic ductal adenocarcinoma (Figure 3, B).

Figure 3.

Histology and genomic characteristics by tumor type. A) Representative images of the histology of (a) pancreatic ductal adenocarcinoma, (b) UC, and (c) UC-OGC (arrowhead = tumor cell, arrow = osteoclast-like giant cell). B) Top genomic alterations, by targeted whole-exome sequencing in UC (n = 9; left) and UC-OGC (n = 5; right) tumors. Percentages represent the frequency of the alteration in the total (N = 14) tumor set. UC = undifferentiated carcinoma; UC-OGC = undifferentiated carcinoma with osteoclast-like giant cells.

Transcriptomic analysis demonstrated few differentially expressed genes between UC-OGC (n = 3) and UC (n = 7), and these genes did not correlate with clinically significant biological processes in pancreatic cancer (Supplementary Figure 1, A, available online). To characterize the different cellular components of UC and UC-OGC, bulk RNA sequencing data from these samples and TCGA datasets were deconvoluted using the extensively validated LM22 matrix²⁵ (Supplementary Figure 1, B, available online). Compared with pancreatic ductal adenocarcinoma, UC and UC-OGC tumors together had differential enrichment of various immune cell types (P < .05) (Figure 4). Enriched cell types in the undifferentiated variants included activated natural killer cells, activated mast cells, plasma cells, follicular helper T cells, γ-δ T cells, and M2 macrophages (Figure 4, A-F). Enriched cell types in pancreatic ductal adenocarcinoma were regulatory T cells (Tregs) and resting states of natural killer cells and naive B cells (Figure 4, G-I).

Figure 4.

Undifferentiated carcinoma and UC-OGC tumors show differential enrichment of immune cells in the tumor microenvironment compared with pancreatic ductal adenocarcinoma. Relative prevalence of immune cell subtypes between pancreatic ductal adenocarcinoma (n = 159), UC (n = 7), and UC-OGC (n = 3) tumors after deconvolution using the LM-22 dataset (P < .05 for panels A-I using combined UC/UC-OGC [n = 10] compared with pancreatic ductal adenocarcinoma). UC = undifferentiated carcinoma; UC-OGC = undifferentiated carcinoma with osteoclast-like giant cells.

Undifferentiated carcinoma with osteoclast-like giant cells tumors are deficient in Tregs but enriched in antigen-presenting cells

To validate the immune cell composition predicted by our deconvolution analysis, immune cell markers were assessed by multiplex immunofluorescence (Figure 5). Representative images of pancreatic ductal adenocarcinoma, UC, and UC-OGC fluorescent labeling are presented in Figure 5, A. Pancytokeratin, a marker of epithelial cells, was more abundant in pancreatic ductal adenocarcinoma than in UC and UC-OGC, which do not vary significantly from one another (Figure 5, Ba). This finding validated the histological assessment by the gastrointestinal pathologist.

Figure 5.

Multiplex immunofluorescence in UC, UC-OGC, and pancreatic ductal adenocarcinoma. A) Representative multiplex immunofluorescence images of (a) pancreatic ductal adenocarcinoma (matched controls, n = 16), (b) UC (n = 13), and (c) UC-OGC (n = 5) tumors. B) Quantification of multiplex immunofluorescence stains for (a) tumor cells (PanCK+), (b) cytotoxic T lymphocytes (CD3+CD8+FoxP3‒), (c) Tregs (CD3+CD8‒FoxP3+), and (d) antigen-presenting cells (CD163+). C) Quantification of the M2 (CD163+arginase 1+) macrophages in UC vs UC-OGC. NS = nonsignificant; Treg = regulatory T cell; UC = undifferentiated carcinoma; UC-OGC = undifferentiated carcinoma with osteoclast-like giant cells.

Cytotoxic T lymphocytes, positive for CD8 and CD3, were statistically significantly more prevalent in pancreatic ductal adenocarcinoma than in UC or UC-OGC (Figure 5, Bb). Tregs (CD3+FoxP3+) were statistically significantly decreased in UC-OGC compared with UC and pancreatic ductal adenocarcinoma, and trended toward decreased expression in UC vs pancreatic ductal adenocarcinoma (Figure 5, Bc). This finding corroborates the RNA sequencing deconvolution demonstrating decreased Tregs in the UCs compared with pancreatic ductal adenocarcinoma (Figure 4, I).

The defining feature of UC-OGC is the presence of osteoclast-like giant cells, which have been reported to act as anti-tumorigenic macrophages^4,29,30. UC-OGC had significantly more antigen-presenting cells than did pancreatic ductal adenocarcinoma (P = .039), identified by expression of CD163 (Figure 5, Bd). Parsing of the CD163-expressing population by expression of arginase 1, a marker historically associated with immunosuppressive tumor-associated macrophages^31-33, demonstrated a trend toward decreased tumor-associated macrophages in UC-OGC compared with UC (P = .12) (Figure 5, C). No specific patterns were identified in the expression markers on the individual osteoclast-like giant cells with the probes used in this study.

Discussion

Deeper characterization of UCs of the pancreas has been limited by their rare occurrence, the paucity of available tissue, and evolving definitions of rare variants. Many prior reports have combined UC and UC-OGC, but we demonstrate here that the behavior of these cancer subtypes differs substantially, and therefore separation of these 2 histological variants is essential to understanding the complex biology underlying these tumors. This study is among the largest retrospective cohort analysis comparing UC, UC-OGC, and pancreatic ductal adenocarcinoma. Patients with UC and UC-OGC had similar demographics, clinical characteristics, and genomic alterations, but patients with UC-OGC were more frequently diagnosed at a resectable stage than UC, which may have resulted in a statistically and clinically significantly higher rate of resectability and survival. Patients with UC-OGC who underwent resection, however, also lived longer than patients with UC, suggesting a difference in tumor biology.

The immune profiles of UC and UC-OGC have been assessed previously by using immunohistochemistry, primarily with a goal to assess programmed cell death 1 protein/programmed cell death 1 ligand 1 expression^3,31,34, but global investigatory techniques exploring the composition of the tumor immune microenvironment as a major factor contributing to the striking difference in survival between these variants was yet to be performed. The increased antigen-presenting cells, decreased Tregs, and increased cytotoxic T cells in UC-OGC compared with pancreatic ductal adenocarcinoma are novel findings that provide a foundation for further study of the tumor immune microenvironment in subtypes of pancreatic cancer that may affect overall survival and have implications for treatment decisions.

Differences in the behavior of immune cells, including the osteoclast-like giant cells themselves, have long been hypothesized to contribute to the increased survival of patients with UC-OGC^31,35-37. We identified enrichment of antigen-presenting cells in UC-OGC, which may be indicative of increased immune awareness in these tumors. We also identified statistically significantly decreased Tregs in the composition of UC-OGC tumors compared with UC and pancreatic ductal adenocarcinoma, suggesting that the tumor immune microenvironment in UC-OGC may be primed to respond to immune-modulating agents. These novel findings support further investigation of the possible differences in the tumor immune microenvironment of UC and UC-OGC tumors to determine whether alternative treatment strategies, such as immune checkpoint inhibitors and macrophage inhibitors, which alter recruitment of macrophage subtypes and increase recruitment of CD4-positive T cells that are not Tregs toward an antitumor phenotype³³ may be considered to improve survival in UC.

Although this study expands the knowledge of the tumor immune microenvironment in these 2 variants of pancreatic cancer, there are several limitations. The major limitation was tissue availability and adequacy for the analyses (Figure 1). The CIBERSORTx sequencing analysis is a computational model of transcriptomic analysis with inherent limitations, and further validation is critical to confirm the differences in the tumor immune microenvironment of UC, UC-OGC, and pancreatic ductal adenocarcinoma. Even if additional data continue to support exploration of novel therapeutic targets in UC, such as checkpoint and macrophage inhibitors, prospective trials will be challenging to perform given the rarity of this disease subtype as well as intertumoral and intratumoral heterogeneity. Retrospective analyses such as this one do have limitations, but they continue to be the most viable means by which rare histological variants of pancreatic cancer can be studied.

Precision medicine requires depth of understanding of rare and common subtypes of disease alike. Despite the challenges associated with studying rare variants, ongoing exploration of the underlying biology is critical to the incorporation of precision medicine for the treatment of pancreatic carcinomas of all types. Undifferentiated carcinoma with osteoclast-like giant cells provides a unique opportunity to study such biologic variation, similar in genomic alterations to pancreatic ductal adenocarcinoma but with distinct composition of the tumor immune milieu. Continued effort to understand the underlying mechanisms by which the tumor immune microenvironment alters patient outcomes can inform future studies focused on incorporation of immune-modulating therapies for rare pancreatic cancer subtypes.

Data availability

The sequencing data analyzed in this study are publicly available from the TCGA database: https://www.cancer.gov/tcga. The sequencing data generated in this study are available upon request from the corresponding author. The clinical data analyzed in this study are not publicly available due to patient privacy requirements but are available upon reasonable request from the corresponding author.

Author contributions

Jamie N. Mills, MD, PhD (Conceptualization; Data curation; Formal analysis; Investigation; Project administration; Visualization; Writing—original draft; Writing—review & editing), Valerie Gunchick, MS (Data curation; Formal analysis; Project administration; Visualization; Writing—review & editing), Jake McGue, BS (Formal analysis; Visualization), Zhaoping Qin, BS (Formal analysis; Investigation), Chandan Kumar-Sinha, PhD (Data curation; Resources; Visualization; Writing—review & editing), Filip Bednar, MD, PhD (Resources; Writing—review & editing), Noah Brown, MD (Conceptualization; Data curation; Writing—review & editing), Jiaqi Shi, MD, PhD (Conceptualization; Data curation; Formal analysis; Investigation; Resources; Validation; Visualization; Writing—review & editing), Aaron Udager, MD, PhD (Formal analysis; Investigation; Resources; Software; Visualization; Writing—review & editing), Timothy Frankel, MD (Funding acquisition; Investigation; Resources; Supervision; Writing—review & editing), Mark M. Zalupski, MD (Conceptualization; Supervision; Writing—review & editing), Vaibhav Sahai, MBBS (Conceptualization; Funding acquisition; Project administration; Supervision; Writing—review & editing).

Funding

This work was supported by the Rogel Scholar Award (V.S.), the University of Michigan Oncology Research Training Grant (T32CA009357 to J.N.M.), and the National Cancer institute of the National Institutes of Health award (R37CA262209 to J.S.).

Conflicts of interest

V.S. received institutional grant funding from Agios, Aravive, Boehringer-Ingram, Bristol-Myers Squibb, Celgene, Clovis, Exelixis, Fibrogen, Incyte, Ipsen, Medimmune, Merck, Syneos, and Cornerstone as well as consultant fees from AstraZeneca, Autem, Delcath Systems, GlaxoSmithKline, Histosonics, Incyte, Ipsen, QED, and Cornerstone. V.G. has received consultant fees from Cornerstone.

Supplementary material

Supplementary material is available at JNCI Cancer Spectrum online.