The prognostic value of stroma in pancreatic cancer in patients receiving adjuvant therapy

Abstract

Background

Pancreatic ductal adenocarcinoma (PDA) is comprised of a prominent desmoplastic stromal compartment and only 10–40% of the tumour consists of PDA cells. However, how stromal components should be assessed and how the characteristics of the stromal compartment determine clinical outcomes in PDA patients remain unknown.

Method

A cohort of 66 consecutive patients who underwent pancreaticoduodenectomy and were primarily followed at Johns Hopkins Hospital between 1998 and 2004, and treated with adjuvant therapy, were included in a retrospective analysis. Resected PDA blocks with good tissue preservation were available for all patients. A new, computer-aided, quantitative method was developed to assess the density and activity of stroma in PDAs and the associations of these characteristics with clinical outcomes.

Results

High stromal density in resected PDA was found to be significantly associated with longer disease-free [adjusted hazard ratio (aHR) 0.39; P = 0.001] and overall (aHR 0.44; P = 0.004) survival after adjusting for the use of pancreatic cancer vaccine therapy, as well as gender and resection margin positivity. Stromal activity, representing activated pancreatic stellate cells in PDAs, was not significantly associated with the prognosis of resected PDAs.

Conclusions

These results illustrate the complexity of the role of stroma in PDAs. Further exploration of the prognostic ability of the characteristics of stroma is warranted.

Introduction

Pancreatic ductal adenocarcinoma (PDA) is currently the 10th leading cause of cancer in the USA. However, its associated mortality is disproportionally high as PDA is the country's fourth leading cause of cancer-related death.¹ The overall poor prognosis has been attributed to both late detection, typically at a relatively advanced stage, and the lack of effective treatments.²

Pancreatic ductal adenocarcinoma cells constitute only 10–40% of PDA tumour volume and are surrounded by a dense, collagenous, hypovascular, desmoplastic stromal compartment. This stroma is composed of extracellular matrix proteins, fibroblasts, stellate cells and inflammatory cells. Pancreatic stellate cells (PSCs) in their activated form produce extracellular matrix components, lose their vitamin A storage capacity, and have enhanced α-smooth muscle actin (α-SMA) expression.^3–5 In turn, PSCs have been shown to promote tumorigenesis and chemoresistance in vitro.^3–5 Emerging evidence suggests that the depletion of tumour-associated stromal fibroblasts by disrupting Hedgehog signalling, the ablation of hyaluronic acid from the extracellular matrix by pegylated recombinant hyaluronidase, and the depletion of stromal fibroblasts expressing fibroblast activation protein-α (FAP-α) in mouse models of PDA result in improved delivery and greater antitumour activity of chemotherapy agents or immune-mediated hypoxic necrosis of the tumour. These results indicate that the stromal compartment may play a determinant role in the prognosis of PDA.^6–8

The present paper reports the development of a new, computer-aided, quantitative method to assess the density and activity of stroma in PDAs and their association with clinical outcomes. High stromal density, but not SMA expression, represented by stromal activity, in resected PDA was found to be significantly associated with longer disease-free survival (DFS) and overall survival (OS).

Materials and methods

Patients and specimens

This retrospective study refers to data for consecutive patients submitted to pancreaticoduodenectomy between 1998 and 2004 at Johns Hopkins Hospital (JHH) and subsequently treated with adjuvant chemoradiation therapy. Only patients primarily followed at JHH were included. The study inclusion criteria also required the existence of archived paraffin-embedded tissue blocks with good tissue preservation. A total of 66 patients were eligible for analysis. Among them, 38 patients participated in an adjuvant pancreatic cancer vaccine study and received an allogeneic granulocyte-macrophage colony-stimulating factor (GM-CSF)-secreting whole-cell pancreatic cancer vaccine (GVAX) in addition to standard chemotherapy and radiation.⁹ None of the 66 patients were treated with neoadjuvant therapy.

Immunohistochemistry

In this study, a novel method for the quantification of tumour stroma was employed, in which tumours were co-stained for epithelial tumour cells and stromal collagen and then quantified using computer-based image analysis. Sections of formalin-fixed, paraffin-embedded tissue of 3 μm in thickness were stained for pankeratin [anti-pankeratin (AE1/AE3/PCK26), 1:100; Ventana Medical Systems, Inc., Tucson, AZ, USA] using a Dako autostainer (Dako, Agilent Technologies, Santa Clara, CA, USA) with a high pH retrieval and 3,3′-diaminobenzidine detection, and using a modified protocol in which counterstaining with haematoxylin was not performed prior to fixing in Bouin's 2000 overnight at room temperature and subsequent counterstaining with aniline blue. Staining for α-SMA on consecutive sections of the same blocks was performed at the JHH Immunopathology Laboratory using a Ventana autostainer with mouse anti-SMA immunoglobulin G (IgG) antibody (clone: 1A4; catalogue no. 760–2833; Ventana Medical Systems, Inc.) according to the manufacturer's protocols, with 3,3′-diaminobenzidine detection and haematoxylin counterstaining.

Image analysis

All slides were scanned and images analysed using ImageScope Viewer (Aperio Technologies, Inc., Vista, CA, USA). The area of tumour to be analysed was circled by a blinded pathologist (RAA) on haematoxylin and eosin-stained slides in such a way as to include the largest area possible of continuous neoplastic tissue and to exclude, as far as possible, normal pancreatic, intestinal and muscle tissue. The positive pixel count algorithm was used to count positive (brown) and negative (blue) pixels in the specified area. The stromal density, defined by the ratio of the area of stroma to that of the total tumour mass, was calculated as the total number of negative (blue, staining for collagen) pixels divided by the total number of pixels [blue (staining for collagen) plus brown (staining for pankeratin)] in the selected tumoral area in slides stained for pankeratin and collagen. Similarly, slides stained for SMA were scanned and positive (brown, staining for smooth-muscle actin) and negative (blue, staining for nuclei) pixels in the predefined tumoral area were counted. As most stromal cells express SMA at a certain level, whereas PSC is characterized by a high level of SMA expression, stromal activity was defined as the percentage of stromal cells demonstrating the strong expression of SMA among all cells expressing SMA and scored as the total number of strongly positive pixels divided by the total number of positive pixels in the selected area. The default threshold of 100 was used to select strongly positive pixels as per the software user's guide. The expression of SMA in a randomly selected area of smooth muscle was used to normalize the variation in staining between slides from different batches of staining.

Statistical analysis

Comparisons of the clinical and pathological characteristics of the two cohorts were made using Fisher's exact test and the Wilcoxon rank-sum test. Disease-free survival was defined as the time from surgery to the first evidence of local or metastatic recurrent disease or death from any cause. Overall survival was defined as the time from surgery to death from any cause. Individuals were censored for DFS or OS at the date of the last known scan or contact, respectively, if no events were observed. Kaplan–Meier estimates of the survival function were used to graphically compare the time-to-event outcomes based upon treatment, as well as stromal characteristics. Comparisons between groups were made with log-rank tests and Cox proportional hazards regression models. All analyses were performed using R: A Language and Environment for Statistical Computing (http://www.R-project.org; R Foundation for Statistical Computing, Vienna, Austria).

Results

A total of 66 patients submitted to pancreaticoduodenecotmy followed by adjuvant chemoradiation, who were primarily followed at JHH, and for whom well-preserved tissue blocks were available, were identified and included in the retrospective analysis (Table 1 and Table S1, online). Pankeratin staining in combination with collagen staining was performed on these tissue specimens and used to quantify stromal density (Fig. 1a, b). Median OS was 22.2 months [95% confidence interval (CI) 17.7–26.3] in the entire cohort, 25.1 months (95% CI 21.7–41.4) in individuals with high stromal density (≥0.8), and 14.4 months (95% CI 11.1–24.2) in individuals with low stromal density (<0.8) (Fig. 2a). In unadjusted analyses, high stromal density was associated with significantly longer OS than low stromal density [unadjusted hazard ratio (uHR) 0.43, 95% CI 0.24–0.74; P < 0.001]. Median DFS was 14.8 months (95% CI 10.5–18.8) in the entire cohort, 18.7 months (95% CI 14.7–30.8) in individuals with high stromal density, and 9.6 months (95% CI 7.3–17.0) in those with low stromal density (Fig. 2b). Individuals with high stromal density remained disease-free for significantly longer than those with low-density stroma (uHR 0.44, 95% CI 0.25–0.77; P = 0.003). Gender, tumour volume, tumour grade and type of adjuvant therapy (chemoradiation alone or with GVAX) were not significantly associated with OS. The association between positive resection margin status and shorter OS was of borderline significance (uHR 1.60; P = 0.08), as anticipated (Table S1). These results suggest that high stromal density appears to be a good prognostic factor for PDAs following pancreaticoduodenectomy and adjuvant therapy.

Table 1.

Clinical and pathological characteristics of patients

Characteristic	Overall (n = 66)	Unvaccinated patients (n = 28)	Vaccinated patients (n = 38)	P-value
Age at surgery, years, median (range)	63 (43–83)	63 (44–80)	62 (43–83)	0.690
Female, n (%)	28 (42%)	11 (39%)	17 (45%)	0.800
Node-positive, n (%)	62 (94%)	27 (96%)	35 (92%)	0.630
Margin-positive, n (%)	31 (47%)	17 (61%)	14 (37%)	0.081
Tumour characteristics
Size, cm, median (range)	3.0 (1.0–6.0)	3.0 (1.0–5.5)	3.0 (1.3–6.0)	0.220
≥3 cm, n (%)	38 (58%)	18 (64%)	20 (53%)	0.450
Tumour grade 3–4, n (%)	24 (36%)	11 (39%)	13 (34%)	0.800
Stroma characteristics
Stromal density, median (range)	0.83 (0.46–0.98)	0.83 (0.46–0.92)	0.83 (0.54–0.98)	0.440
Stromal activity, median (range)	1.51 (0.16–9.72)	1.50 (0.21–9.72)	1.51 (0.16–5.04)	0.200

Figure 1.

Immunohistochemistry staining and quantification of stromal density and activity in pancreatic ductal adenocarcinoma (PDA). (a) Representative immunohistochemistry staining of pankeratin (brown signals) and collagens (blue signals). (b) Representative quantification of stromal density in the intratumoral area (enclosed). (c) Representative immunohistochemistry staining of smooth muscle actin (SMA) (brown signals). (d) Representative quantification of stromal activity in the intratumoral area (enclosed). (e) Representative PDA with high stromal activity/strong SMA expression. (f) Representative PDA with low stromal activity/weak SMA expression

Figure 2.

Kaplan–Meier estimates of (a) overall and (b) disease-free survival in individuals with high stromal density (≥0.8, light blue) and low stromal density (<0.8, dark blue) over the first 60 months of follow-up. Numbers of individuals at risk in each stromal density category are shown below the graphs

As preclinical studies have suggested that stroma promotes PDA growth and metastasis, it is not anticipated that high stromal density would be associated with a better prognosis. Thirty-eight (58%) of the 66 patients were treated with GVAX, which was developed through a clinical trial in the adjuvant setting, in addition to standard chemoradiation (Table S1). There was no significant difference between the vaccinated patient group and the unvaccinated patient group in terms of age, gender, node positivity, tumour grade or the proportion of patients with a tumour size of ≥3 cm. This raised the question of whether it was possible that the inclusion of these 38 vaccinated patients may have changed the prognostic value of stromal density. It should be noted that the effects of gender on OS and DFS differed between the vaccinated patient group and the group of patients treated with adjuvant chemoradiation only (test of interaction: P = 0.008 and P < 0.001, respectively). In the vaccinated group, there was no significant difference in OS and DFS between men and women; however, among individuals receiving adjuvant chemoradiation therapy only, men had significantly longer OS and DFS than women (Tables 2 and 3). Therefore, the analysis was adjusted for gender, margin status and the receipt of vaccine treatment to explore the potential influence of the vaccine. Nonetheless, the association of high stromal density with longer OS remained significant after simultaneously adjusting for the receipt of vaccine treatment as well as for gender and margin status. The effect of stromal density was virtually unchanged (aHR 0.44, 95% CI 0.25–0.77; P = 0.004) (Table 2), whereas the risk increased slightly for margin-positive individuals and became statistically significant (aHR 1.97, 95% CI 1.08–3.58; P = 0.027). Similarly to observations for OS, the association of high stromal density with longer DFS persisted (aHR 0.39, 95% CI 0.21–0.70; P = 0.001) after adjusting for the use of vaccine therapy, as well as gender and margin positivity.

Table 2.

Adjusted analysis of risk factors associated with overall survival (OS)

	At risk, n	Deaths, n	Median OS (95% CIa)	HR (95% CI)	P-value
Unvaccinated patientsb					0.006
Female	11	10	14.7 (8.4–25.1)	1.00
Male	17	13	21.7 (12.1–83.9)	0.25 (0.09–0.67)
Vaccinated patientsb					0.500
Female	17	13	25.0 (13.8–41.4)	1.00
Male	21	20	23.6 (17.8–31.2)	1.28 (0.62–2.64)
Margin status					0.027
Negative	35	28	26.3 (16.1–40.0)	1.00
Positive	31	28	18.1 (14.0–25.0)	1.97 (1.08–3.58)
Stromal density					0.004
<0.8	26	24	14.4 (10.6–23.7)	1.00
≥0.8	40	32	25.1 (20.9–34.3)	0.44 (0.25–0.77)

^aConfidence intervals for median OS are computed based upon the log hazard.

^bTest of interaction between gender and treatment was statistically significant (P = 0.008).

95% CI, 95% confidence interval; HR, hazard ratio.

Table 3.

Adjusted analysis of risk factors associated with disease-free survival (DFS)

	At risk, n	Eventsa	Median DFS (95% CIb)	HR (95% CI)	P-value
Unvaccinated patientsc					<0.001
Female	11	11	8.7 (4.8–16.1)	1.00
Male	17	14	14.8 (9.2–39.1)	0.20 (0.08–0.50)
Vaccinated patients					0.130
Female	17	11	18.0 (9.8–52.0)	1.00
Male	21	20	18.3 (10.1–20.7)	1.83 (0.84–3.97)
Margin status					0.150
Negative	35	28	18.3 (9.9–25.8)	1.00
Positive	31	28	11.5 (9.2–16.1)	1.54 (0.85–2.78)
Stromal density					0.001
<0.8	26	23	9.6 (6.4–12.7)	1.00
≥0.8	40	33	18.7 (14.7–25.8)	0.39 (0.21–0.70)

^aAn event is defined as the evidence of disease or death.

^bConfidence intervals for the median DFS are computed based upon the log hazard.

^cTest of interaction between gender and treatment was statistically significant (P < 0.001).

95% CI, 95% confidence interval; HR, hazard ratio.

Next, SMA expression, designated stromal activity, which is a surrogate marker of activated PSCs, was examined (Fig. 1c–f). Stromal activity was not significantly associated with OS (HR 0.85, 95% CI 0.54–1.35; P = 0.49 comparing stromal activity of ≥1.46 versus <1.46) or DFS (HR 0.79, 95% CI 0.49–1.27; P = 0.33). Stromal activity was similar across the vaccinated patient group and the unvaccinated patient group (1.51 versus 1.50; P = 0.20). Therefore, immunohistochemical assessment of SMA expression may not have a prognostic value in resected PDA.

Discussion

The present report has described a new method of quantifying stromal density and stromal activity in resected PDAs. Using this method, the current study has demonstrated the prognostic value of stromal density in patients submitted to the surgical resection of PDAs followed by adjuvant chemoradiation with or without the GM-CSF-secreting pancreatic cancer vaccine treatment.

Although the desmoplastic component of pancreatic adenocarcinoma has long been appreciated, until recently its role in tumorigenesis was poorly understood. There is now increasing evidence that the tumour stroma is a dynamic compartment that may promote tumorigenesis.^3–5 However, this has been difficult to validate clinically. This study employed a novel method of quantifying stroma in clinical samples, in which the amounts of both collagen and tumour were directly measured. Using this measure, high stromal density was significantly associated with improved survival. By contrast, stromal activity does not appear to have prognostic value. Although these results are unexpected, they illustrate the complexity of the role of the stroma in PDAs. The quantity of stroma or the extent of desmoplasia itself is not necessarily associated with the aggressiveness of PDA. On the contrary, it is strongly associated with a better prognosis. In addition, SMA expression is also not prognostic. One possible explanation is that PDA growth is proinflammatory and thus slow-growing PDAs accumulate more fibrotic stroma as a result of the longer duration of inflammatory response. Another possibility is that PDAs with favourable outcomes are associated with an immunoactive tumour microenvironment that results in a high stromal density, whereas those with unfavourable outcomes are associated with an immunosuppressive tumour microenvironment that results in a low stromal density. A third possibility is that high stromal density has selected more preferentially the infiltration of antitumour effector immune cells or less preferentially the infiltration of pro-cancerous immunosuppressive cells. However, no association between stromal density or activity and CD8+ or Foxp3+ cells was observed, which suggests that the mechanisms, if they exist, are more complex than a single marker will demonstrate. This hypothesis is consistent with recently published studies showing that the depletion of stroma in the mouse model of PDA facilitates tumour growth and metastasis.^10,11 One of these published studies suggests that stroma-rich PDAs are associated with differentiated histology, whereas stroma-poor PDAs are associated with less differentiated histology,¹¹ and provides a potential mechanistic explanation for the present results. Thus, further exploration of the mechanisms underlying the prognostic value of the stroma is warranted. The present study and these recently published studies have suggested that an effective antitumour treatment should target a specific stromal component rather than targeting the stroma in general.

Several ongoing clinical trials are investigating the targeting of the stroma as a new therapeutic strategy in patients with pancreatic cancer and have shown conflicting results, some of which have been promising (Phase I hyaluronidase study), although others have been disappointing (Hedgehog inhibitor IPI926 Phase II study).^12,13 The contradictions among the findings of different preclinical studies and between the findings of preclinical studies and the outcomes of clinical trials highlight the need to better elucidate the mechanistic roles of stromal components in tumorigenesis, antitumour immune response, and drug resistance.^14,15 The present method of quantifying the stromal compartment may have utility in ongoing trials that have collected pre- and post-treatment human specimens to further investigate the effects of these therapies.

The present study is limited by its small sample size and retrospective nature. An unexpected result of this study refers to the significantly different influences of gender on clinical outcomes in those treated with and without vaccine; however, the sample size was quite small and the results require to be validated in a future study. The aetiology of this gender effect is unknown, although previous studies have hinted at the possibility of such an effect.¹⁶

After adjusting for potential confounders including gender and treatment, stromal density remained significantly associated with clinical outcome. Given the simplicity of the method for measuring stromal density presented in this study, stromal density can be further explored for its prognostic value in PDAs and in other solid malignancies, as well as for its predictive value in selecting patients who may respond to novel stroma-targeting therapies.

Conflicts of interest

Under a licensing agreement between Aduro BioTech, Inc. (Berkeley, CA, USA) and the Johns Hopkins University and Dr Elizabeth Jaffee, the university is entitled to milestone payments and royalties on sales of the vaccine product described in this manuscript. The authors have no other relevant conflicts of interest.

Supporting Information

Table S1 Assessment of risk factors for overall survival and disease-free survival based upon unadjusted models.