A practical distribution pattern of α-SMA-positive carcinoma associated fibroblasts indicates poor prognosis of patients with pancreatic ductal adenocarcinoma

Highlights

• •Distinct α-SMA intensity highlights CAFs' spatial patterns in PDAC tissues.
• •Type Ⅰ CAFs are associated with advanced N and TNM stages, indicating worse prognosis.
• •CAF spatial distribution independently predicts patient outcomes in PDAC.
• •Transcriptomic data reveal unique immune and oncogenic pathways in type Ⅰ CAFs and corresponding tumors.
• •CAF spatial distribution may guide stroma-targeting therapies in PDAC.

Abstract

Purpose The present study aimed to clarify the distribution pattern of carcinoma associated fibroblasts (CAFs) across pancreatic ductal adenocarcinoma (PDAC) and its prognostic prediction value.

Methods Data of two cohorts were retrospectively collected from consecutive patients who underwent primary pancreatic resection from January 2015 to December 2017. We used tumor specimens to screen out the most suitable markers for the spatial distribution analysis for CAFs subpopulations. We utilized a tissue microarray to assess the spatial intensity of α-SMA expression within the tumor microenvironment. Specifically, we classified CAFs into two types based on their α-SMA spatial expression. Type II CAFs were designated as those located in the juxtatumoural stroma with α-SMA expression that was moderate or higher, and those in the peripheral stroma with α-SMA expression that was less than moderate. All other cases, where the α-SMA expression did not meet these criteria, were categorized as Type I CAFs. Multivariable Cox proportional hazards regression was used to assess risk factors associated with patient outcomes. RNA sequencing data were obtained from bulk tumor samples and isolated CAFs from patients to reveal the distinct pattern and elucidated their fundamental characteristics.

Results The α-SMA spatial intensity was the most suitable variable for representative of CAFs spatial characteristics. Patients with Type Ⅰ CAFs were more likely to be allocated into N1 or N2 of the N stage and Ⅱ and Ⅲ of the TNM stage. The spatial distribution pattern of CAFs (Type Ⅰ v.s. Type Ⅱ: HR, 1.568; 95 % CI, 1.053–2.334; P = 0.027) was an independent prognostic factor in the discovery cohort, so as in the validation (Type Ⅰ vs. Type Ⅱ: HR, 2.197; 95 % CI, 1.410–3.422; P = 0.001). RNA sequencing analysis revealed that the differentially expressed genes (DEGs) in Type I CAFs are closely associated with those in corresponding tumor tissues, highlighting the enhanced biological significance of immune-related and oncogenic invasive pathways.

Conclusions Our findings that two types of α-SMA-positive CAFs with different spatial patterns present heterogeneously across tissues of PDACs and correlated with patients’ outcomes. The spatial location of CAFs may facilitate patients’ selection in precision medicine of PDACs.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is the fourth leading cause of cancer-related death worldwide, with a dismay survival and 5-year survival rate of <12 % [1]. Over the past several decades, compared with other solid malignancies, improvements in patient outcomes of PDAC have been modest, reflecting both the lack of effective tools for early detection and a paucity of treatment options [2]. Besides, the specific characteristics should not be ignored, such as the rich stroma of tumor microenvironment (TME) [[3], [4], [5]]. The three fundamental components of PDAC stroma are carcinoma-associated fibroblasts (CAFs), extracellular matrix (ECM), and vasculature. The contribution of CAFs to the biology of PDAC has generally been held to be tumor-suppression or tumor-promoting [6], which derives from myofibroblastic CAFs (myo-CAFs) or inflammatory CAFs (i-CAFs) based on the identification of subpopulations in both mouse and human PDAC [7,8]. And the myo-CAFs with elevated expression of α-smooth muscle actin (α-SMA) locate immediately adjacent to neoplastic cells, whereas the i-CAFs which lacked elevated α-SMA expression and instead secreted interleukin (IL) −6 and additional inflammatory mediators, locate more distantly from tumor cells [7]. Based on an increasing number of researches [[9], [10], [11], [12], [13]] for the complex function of fibroblasts, the CAFs features of PDAC have been clarified as spatial distribution, plastic and dynamic transform, and sequential and hierarchical activation between the two subpopulations of fibroblasts [6]. Therefore, the anti-fibroblast therapies, those targeting very specific subpopulation of pro-tumorigenic fibroblasts, may be the promising silver bullet against PDAC.

Additionally, the preclinical study have demonstrated the specific protein markers to identify these subpopulations by single-cell transcriptome sequencing, and the myo-CAFs markers consist of α-SMA, Periostin (POSTN), metalloproteinase (MMP) −11, et al., so as for i-CAFs are IL-6, stromal cell-derived factor (SDF) −1, platelet-derived growth factor receptor (PDGFR) -α, et al. [14]. In further, there were several studies which investigated the significance of the markers of CAFs for predicting prognosis, such as α-SMA [15] and POSTN [16], the markers of myo-CAFs. Unexpectedly, the high expression of both α-SMA [15] and POSTN [16] indicated worse prognosis, which was in contrary to the tumor-suppressive function of myo-CAFs. The reasons for the conflict between basic and clinical research may contribute to the bias related with the small number cohort of the clinical studies and the defective methods for detecting markers. With a consideration for the complexity of CAFs, the advancement and improvement of evaluation methods for CAFs may facilitate the patient's selection for precision management. Hence, referring to the research methods used for immune cells in TME [17], we also take more attention on the spatial distribution to optimize the evaluation methods of CAFs. To our knowledge, the relationship between spatial distribution of CAFs and the prognosis of PDAC has not been extensively investigated.

Based on the previous definition of juxtatumoural stroma and peripheral stroma [18,19], we improved spatial pattern analysis methods allow for better characterization of CAFs distribution in the TME and its potential clinical implications. Furthermore, we analyzed the biological function of CAFs derived from the two subgroups of patients classified by the present method, in order to verify the reliability of the classification. In this study, we attempted to explore the prognostic values of distribution patterns of CAFs in patients with PDACs.

Materials and methods

Study population and data collection

A total of 529 consecutive patients with a final histopathological diagnosis of PDAC who underwent primary pancreatic resection at the Department of Hepatobiliary Pancreatic Surgery in Changhai Hospital (Shanghai, China) during January 2015 to December 2017 were enrolled in this study. For all patients, the following demographic and clinicopathological variables were recorded in the database: sex, age, preoperative carbohydrate antigen 19–9 (CA19–9), tumor location (head/neck/uncinate, body/tail), perineural invasion (PNI), lymphovascular invasion (LVI), R status (R1 or R0), tumor grade (G1/2 or G3/4), operation type (pancreaticoduodenectomy, PD / distal pancreatectomy, DP / total pancreatectomy, TP) and information on neoadjuvant therapy and postoperative adjuvant therapy. The staging was performed following the WHO recommendations (AJCC, 8th edition) [20]. Clinical and follow-up data were obtained from a prospective digital database. The inclusion criteria were patients who underwent surgery with curative intent and formalin-fixed paraffin-embedded (FFPE) tumor specimens available to obtain from the pathological archives room. The exclusion criteria for this study were as follows: (1) patients with intraoperative metastasis (excluded lymph node metastases) or macroscopic evidence of margin involvement (R2); (2) patients who received neoadjuvant therapy; (3) patients with other malignancies in the past; (4) patients who died within 90 days after surgery; (5) patients who were lost to follow up; and (6)patients cannot retrieve suitable FFPE for tissue microarrays (TMA). Subsequently, 390 patients were included; of these patients, 190 who underwent primary pancreatic resection from January 2015 to June 2016 composed the discovery cohort, and 200 who underwent primary pancreatic resection from July 2016 to December 2017 composed the validation cohort. This study was approved by the Institutional Review Board of Changhai Hospital (approval no. CHEC2018–112), and no additional informed consent was required to review the patients’ medical records.

Tissue microarray (TMA) and immunohistochemistry

The included 390 FFPE tumor specimens were undertook for TMA construction. To reduce the effects of intratumoral heterogeneity, two representative 1.5 mm tissue cores were selected for the construction of TMA using a manual tissue microarray (Beecher Instruments). Six markers were selected for immunohistochemistry (IHC) staining, including cytokeratin (CK) −19 (antibody, Cat# GB12198, Servicebio, RRID:AB_3,665,700) as marker for tumor cell, fibroblast activation protein (FAP) -α (antibody, Cat# ab53066, Abcam, RRID:AB_880,077) as a pan-marker for CAFs, α-SMA (antibody, Cat# ab7817, Abcam, RRID:AB_262,054) and POSTN (antibody, Cat# ab219056, Abcam, RRID:AB_2,920,576) as markers for myo-CAFs, and IL-6 (antibody, Cat# 66,146–1-Ig, Proteintech, RRID:AB_2,881,543) and SDF-1 (antibody, Cat# ab9797, Abcam, RRID:AB_296,627) as markers for i-CAFs (Supplementary Fig. 1). Three markers were selected for IHC staining of immune cells, including CD3+ (antibody, Cat# ab135372, Abcam, RRID:AB_2,884,903), CD4+ (antibody, Cat# ab133616, Abcam, RRID: AB_2,750,883), and CD8+ (antibody, Cat# ab217344, Abcam, RRID:AB_2,890,649). The α-SMA was also stained by fluorescent-IHC antibody.

The stained slides were scanned using a Hamamatsu S60 whole slide scanner (Hamamatsu Photonics, Hamamatsu City, Japan) to obtain digitalized images, which could also be observed using NanoZoomer Digital Pathology view 2 software version 2.7.25. Expression of the markers was classified independently by two senior pathologists (Dr. Yang Wang and Dr. Hui Jiang), and any disagreement between the two pathologists was resolved by recording the mean value. [21].

Evaluation of tumor cells

The overall spatial intensity (negative, weak, and strong staining) of CK-19 in tumor cells and the percentage of stained tumor cells in the three groups (0 %, <90 %, ≥90 %) were semiquantitative defined for each tumor. >90 % of tumor cells with strong staining was regarded as high expression. Tumor cells with negative staining was regarded as negative expression. And others were regarded as low expression. CK-19 was a highly sensitive marker for tumor cells of PDAC. Therefore, the PDAC could be diagnosed by hematoxylin-eosin (H&E) and CK-19 staining definitely.

Evaluation of CAFs spatial distribution

Stromal cells were firstly identified by the CAFs markers relied on the slides derived from FFPE tissue sample with both the tumor and the normal pancreatic tissue, so that the sensitivity and specificity of CAFs markers for detecting CAFs in the tumor could be evaluated comprehensively. The spatial intensity of staining for CAFs markers were classified as negative, weak, moderate, and strong, recorded as 0, 1, 2 and 3, respectively (Supplementary Fig. 2A-D). In addition, the strong spatial intensity was regarded as the spatial intensity of α-SMA staining in vessels (Supplementary Fig. 2E). The frequency of staining for CAFs markers were classified as < 5 %, 5–24 %, 25–49 % and ≥ 50 %, recorded as 0, 1, 2 and 3, respectively. The evaluation score equals to multiply spatial intensity by frequency, so the minimum is zero and maximum is nine. And three compartments of stroma were defined as the previous paper [18]. The juxtatumoural stroma is defined as the stroma at a distance of ≤ 100 μm away from the tumor cells (tumor-c). The peripheral stroma is located > 100 μm away from the cancer cells. The septal stromal is defined as the stroma in the perilobular areas, surrounding the pancreatic lobuli. To investigate spatial distribution characteristics of CAFs in the tumors, we conducted spatial analyses using TMA based on the CAFs distribution findings in FFPE tissue derived slides. In this part of the research, only juxtatumoural stromal and peripheral stromal were evaluated by staining of H&E, CK-19, FAP-α, α-SMA, POSTN, and SDF-1.

Based on the α-SMA staining patterns, we classified CAFs into two distinct types. Type II CAFs were defined by a specific α-SMA staining pattern where the spatial intensity in the juxtatumoural stroma was ≥ 2 (moderate to strong), while the spatial intensity in the peripheral stroma was < 2 (negative to weak). This indicates a higher concentration of activated CAFs in close proximity to tumor cells. Conversely, Type I CAFs encompassed all other α-SMA staining distributions that did not meet the Type II criteria, representing various patterns of α-SMA spatial intensity across both juxtatumoural stromal and peripheral stromal regions (Supplementary Fig. 3). This classification system allows for a more nuanced understanding of CAF distribution and activation in the tumor microenvironment, potentially providing insights into tumor-stroma interactions and their impact on cancer progression.

Evaluation of immune cells

The positive staining frequency of immune cell in tumor samples was evaluated using cutoffs of 5 % and 25 %.

CAFs isolation and culture

The written informed consents were obtained before operation. In brief of isolation of pancreatic CAFs, surgically resected pancreatic cancer tissues were obtained from 6 patients with pancreatic ductal adenocarcinoma (PDAC) and 3 patients with pancreas benign tumor from July 1 to September 30, 2020. The fresh pancreatic tumor tissues were divided into two parts. One part was treated with FFPE procedure, in order to evaluate the CAFs distribution. Another part was minced into 1–3 mm³ fragments and digested with 1 mg/ml collagenase I (Cat#C0130, Sigma), hyaluronidase (Cat#37,326–33–3, Sigma) and Dnase (Cat#11,284,932,001, Roche) at 37 °C for 1 h. The solution was centrifuged at 1000 rpm for 5 min, washed with phosphate-buffered saline twice and filtered with a 100-μm filter. Subsequently, the isolated cells were seeded in six-well culture plates. IHC staining and RNA sequencing was performed following subculture and passage less than three times of these cells.

Transcriptome library preparation

Tumor samples from 24 pancreatic ductal adenocarcinoma (PDAC) patients were collected (Type Ⅰ, n = 13; Type Ⅱ, n = 11). Additionally, previously cultured CAFs (Type Ⅰ, n = 3; Type Ⅱ, n = 3; normal, n = 3) were included. These samples were used to construct a transcriptome sequencing library. Total RNA was extracted from cultured cells using TRIzol reagent (Cat#15,596–026, Invitrogen) according to the manufacturer's instructions. RNA was extracted using an AllPrep DNA/RNA/miRNA Universal Kit (QIAGEN Cat# 80,224) following the manufacturer's protocol, and the sample quality was confirmed (RNA integrity number, 7.3 ± 0.7) using an Agilent 2100 Bioanalyser (Agilent). Index-coded samples were processed on a cBot Cluster Generation System using the Illumina PE Cluster Kit (Illumina, San Diego, California, USA) to generate clusters, and the cDNA libraries were sequenced on an Illumina HiSeq X TEN platform to produce 150 bp paired-end reads (Illumina, San Diego, California, USA).

RNA-seq analysis

Raw fastq data underwent quality control using fastp (v0.23.2, RRID:SCR_016962) [22], eliminating adapters and low-quality reads. Clean reads were subsequently aligned to GRCh38 using HISAT2 (v2.2.1, RRID:SCR_015530) [23]. Read quantification was performed using featureCounts (v2.0.6, RRID:SCR_012919) [24] based on GENCODE v.42. Differential gene expression analysis was conducted using DESeq2 (v1.40.2, RRID:SCR_015687) [25], with differentially expressed genes (DEGs) defined by |log2(fold change)| ≥ 1 and adjusted p-value < 0.05. Gene Ontology (GO) and Gene Set Enrichment Analysis (GSEA) were performing using the R package clusterProfiler (v4.8.3, RRID:SCR-016884) [26], applying thresholds of q-value < 0.05 for GO and |NES| > 1 with FDR < 0.25 for GSEA. Cancer-related functional gene signatures (Fges) were calculated using single-sample GSEA (ssGSEA) based on previously published data [27]. Immune scores were calculated using the ESTIMATE (v1.0.13, RRID:SCR_026090) [28].

Follow-up protocol

The institutional follow-up was jointly completed by department follow-up specialists, and third-party (LinkDoc Technology Co. Ltd. Beijing, China). The frequency of follow-ups was determined once per 2-months during the first half-year after the operation, followed by once per half-year until December 31, 2022, the cutoff date of follow-ups in this study. The follow-up methods included outpatient visits, contact by phone, mail, chatting software, or address. The follow-up endpoint OS was defined as the time from operation to death. Patients who were still alive at the cutoff date of follow-ups were censored when they were last confirmed to be alive. We defined loss to follow-up as a no-show at the clinical follow-ups or the inability to contact patients or their family members by phone, mail, or address.

Statistical analysis

Categorical data are presented as percentages. Distributional differences in baseline variables between the two cohorts and the association of CAFs distribution patterns with clinicopathological features were examined using the chi-squared test or Wilcoxon rank-sum test. Univariate and multivariate Cox regression analyses were performed to identify independent prognostic factors, and hazard ratios (HRs) were calculated. Variables with a P value < 0.1 in univariate analyses were included in multivariate analyses using a forward selection algorithm. The Kaplan–Meier method and log-rank test were used to analyze time to endpoints. Analyses were performed using SPSS version 25.0 (IBM Corp., Armonk, NY, USA, RRID:SCR_002865). R (version 4.3.1; https://cran.r-project.org/, RRID:SCR_001905) was used for analyzing data of RNA sequence. For all analyses, a two-tailed P value < 0.05 was considered statistically significant.

Results

Study population

Of the 529 consecutive patients in our study, 139 were excluded because they were not suitable for TMA construction (n = 40), had intraoperative metastasis or R2 (n = 18), had other malignancies in the past (n = 5), died within 90 days after surgery (n = 22), were lost to follow-up (n = 13) or received neoadjuvant therapy (n = 41). All patients enrolled were of Asian descent. The discovery cohort comprised 190 patients, whereas the validation cohort consisted of 200 patients. In the discovery cohort, 63 and 127 patients were deemed as Type Ⅱ and Type Ⅰ, respectively, corresponding to those 67 and 133 patients in the validation cohort (Supplementary Fig. 4). Relevant baseline variables such as age, sex, tumor location, operation type, preoperative CA19–9, T stage, N stage, TNM stage, tumor grade, α-SMA spatial intensity, PNI, LVI, R status, and postoperative adjuvant therapy were similarly distributed in the discovery and validation cohorts (Supplementary Table 1).

CAFs distribution characteristics in resected pancreatic cancer

We firstly screened out the most suitable markers for the spatial analysis for CAFs based on the criteria including 1) significant different expression between juxtatumoural stromal and peripheral stromal, and 2) with or without expression in tumor cells (tumor-c) and peripheral stromal. Of the 13 selected cases (Supplementary Table 2), we found that pan-marker FAP-α expression (Supplementary Fig. 1A) was significantly higher in peripheral stromal than juxtatumoural stromal in spatial intensity (P < 0.01, Fig. 1A), frequency (P < 0.001, Fig. 1B) and evaluation score (P < 0.001, Fig. 1C). In addition, tumor cells also had a high expression. In further analysis of myo-CAFs markers, α-SMA expression (Supplementary Fig. 1B) and POSTN expression (Supplementary Fig. 1C) were significantly higher in juxtatumoural stromal than peripheral stromal in spatial intensity (P < 0.001, Fig. 1D; P < 0.01, Fig. 1G) and evaluation score (P < 0.001, Fig. 1F; P < 0.01, Fig. 1I), respectively. However, for i-CAFs markers IL-6 (Supplementary Fig. 1D) and SDF-1 (Supplementary Fig. 1E), only SDF-1 expression was significantly higher in juxtatumoural stromal than peripheral stromal in spatial intensity (P < 0.01, Fig. 1M), frequency (P < 0.001, Fig. 1N) and evaluation score (P < 0.001, Fig. 1O). Therefore, based on the criteria for spatial analysis for CAFs, we chose markers FAP-α, α-SMA, POSTN and SDF-1 for further research in TMA. Of 190 patients in discovery cohort, the expression of FAP-α, α-SMA, POSTN and SDF-1 in spatial intensity, frequency and evaluation score were significantly different between juxtatumoural stromal and peripheral stromal (P < 0.001, Supplementary Table 3). And except for FAP-α, the expression of other 3 markers were higher in juxtatumoural stromal compared with that in peripheral stromal, which agreed with that showed in the 13 cases analyzed with slides from FFPE tissue (Supplementary Table 3). To identify the most suitable markers for characterizing the spatial distribution of CAFs, we conducted correlation coefficient analyses between the juxtatumoural stromal and peripheral stromal regions. Our analysis revealed that α-SMA spatial intensity exhibited the lowest correlation coefficient (0.65) between these regions (Fig. 2A), followed by POSTN spatial intensity with a coefficient of 0.71 (Fig. 2G). This finding suggests that α-SMA spatial intensity demonstrates the highest degree of spatial variability, making it the most suitable candidate for representing CAF spatial characteristics. Additionally, α-SMA spatial intensity can be easily and reproducibly assessed in routine pathological evaluations. Based on these results, we established a classification system using α-SMA spatial intensity to define two distinct CAF subpopulations: Type I and Type II - based on the spatial intensity and spatial distribution of α-SMA staining (Fig. 3). This classification was then applied to categorize CAF distribution patterns in both our developmental and validation cohorts.

Fig. 1.

Scatter plots showing pairwise comparisons of CAFs markers (FAP-α, α-SMA, POSTN, IL-6 and SDF-1) expression among four compartments of each tumor tissue. Each marker was evaluated by staining spatial intensity (A, D, G, J, M), frequency (B, E, H, K, N) and score (C, F, I, L, O). The P value was calculated using the chi-squared test. Abbreviations: CAFs, Carcinoma Associated Fibroblasts; FAP-α, fibroblast activation protein-α; α-SMA, α-smooth muscle actin; POSTN, Periostin; IL-6, interleukin-6; SDF-1, stromal cell-derived factor-1; tumor-c, tumor cells.

Fig. 2.

Correlations between juxtatumoural stromal and peripheral stromal of CAFs markers (α-SMA, FAP-α, POSTN and SDF-1) expression in TMA of pancreatic cancer. Each marker was evaluated by staining intentity (A, D, G, J), frequency (B, E, H, K) and score (C, F, I, L). Correlation coefficients (r) were calculated using the Spearman correlation test. Abbreviations: CAFs, Carcinoma Associated Fibroblasts; FAP-α, fibroblast activation protein-α; α-SMA, α-smooth muscle actin; POSTN, Periostin; SDF-1, stromal cell-derived factor-1; TMA, tissue microarrays.

Fig. 3.

The CAFs distribution types identifies by α-SMA spatial intensity in TMA of pancreatic cancer. (A) Examples of CAFs type Ⅱ featured as moderate in juxtatumoural stromal and weak in peripheral stromal for α-SMA spatial intensity by IHC staining. (B) Examples of CAFs typeⅠ featured as weak in juxtatumoural stromal and negative in peripheral stromal for α-SMA spatial intensity by IHC staining. H&E and CK19 were both stained for tumor evaluating. The scale bar is 500 μm for magnification of 40 times (x40), 250 μm for magnification of 80 times (x80), and 100 μm for magnification of 200 times (x200). Abbreviations: CAFs, Carcinoma Associated Fibroblasts; α-SMA, α-smooth muscle actin; TMA, tissue microarrays; IHC, immunohistochemistry; H&E, hematoxylin-eosin; CK19, cytokeratin-19.

CAFs distribution patterns and clinicopathological variables

The CAFs patterns were significantly associated with the N stage, and TNM stage in the discovery and validation cohorts (P < 0.05). Patients with Type Ⅱ CAFs were more likely to be allocated into N1 and N2 stages, Ⅱ and Ⅲ of the TNM stages (Table 1).

Table 1.

Association between clinicopathological features and α-SMA spatial intensity.

	Discovery cohort (%)		P	Validation cohort (%)		P
	Type Ⅰ	Type Ⅱ		Type Ⅰ	Type Ⅱ
Total	127 (66.8)	63(33.2)		133(66.5)	67 (33.5)
Sex			0.953			0.245
Male	72 (56.7)	36 (57.1)		76 (57.1)	44 (65.7)
Female	55 (43.3)	27 (42.9)		57 (42.9)	23 (34.3)
Age (years)			0.876			0.453
≤65	65 (51.2)	33 (52.4)		62 (46.6)	35 (52.2)
>65	62 (48.8)	30 (47.6)		71 (53.4)	32 (47.8)
Tumor location			0.228			0.092
Head/neck/uncinate	106 (83.5)	48 (76.2)		114 (85.7)	51 (76.1)
Body/tail	21 (16.5)	15 (23.8)		19 (14.3)	16 (23.9)
Operation type			0.417			0.565
PD	72 (56.7)	37 (58.7)		77 (57.9)	43 (64.2)
DP	54 (42.5)	24 (38.1)		55 (41.4)	24 (35.8)
TP	1 (0.8)	2 (3.2)		1 (0.8)	0 (0)
Grade			0.07			0.597
1/2	87 (68.5)	51 (81.0)		98 (73.7)	47 (70.1)
3/4	40 (31.5)	12 (19.0)		35 (26.3)	20 (29.9)
LVI			0.983			0.585
Without	109(85.8)	54 (85.7)		115 (86.5)	56 (83.6)
With	18 (14.2)	9 (14.3)		18 (13.5)	11 (16.4)
PNI			0.558			0.727
Without	9 (7.1)	6 (9.5)		14 (10.5)	6 (9.0)
With	118 (92.9)	57 (90.5)		119 (89.5)	61 (91.0)
T stage			0.733			0.096
1	21 (16.5)	9 (14.3)		19 (14.3)	18 (26.9)
2	71 (55.9)	39(61.9)		71 (53.4)	30 (44.8)
3/4	35 (27.6)	15 (23.8)		43 (32.3)	19 (28.4)
N stage			0.002			<0.001
0	57 (44.9)	36 (57.1)		47 (35.3)	41 (61.2)
1	40 (31.5)	25 (39.7)		59 (44.4)	11 (16.4)
2	30 (23.6)	2 (3.2)		27 (20.3)	15 (22.4)
TNM stage			0.002			0.001
Ⅰ	46 (36.2)	26 (41.3)		36 (27.1)	33 (49.3)
Ⅱ	48 (37.8)	34 (54.0)		70 (52.6)	18 (26.9)
Ⅲ	33 (26.0)	3 (4.8)		27 (20.3)	16 (23.9)
R status			0.878			0.238
R0	102(80.3)	50 (79.4)		107 (80.5)	49 (73.1)
R1	25 (19.7)	13 (20.6)		26 (19.5)	18 (26.9)
CD3+ frequency			0.950
< 5%	14 (11.0)	6 (9.5)
5-24%	52 (40.9)	26(41.3)
≥25%	61 (48.0)	31(49.2)
CD4+ frequency			0.001
< 5%	16 (12.6)	3 (4.8)
5-24%	62 (48.8)	48(76.2)
≥25%	49 (38.6)	12(19.0)
CD8+ frequency			0.123
< 5%	16 (12.6)	3 (4.8)
5-24%	72 (56.7)	44(69.8)
≥25%	39 (30.7)	16(25.4)

Abbreviation: TNM, tumor–node–metastasis; α-smooth muscle actin; LVI, lymphovascular invasion; PNI, perineural invasion; PD, pancreaticoduodenectomy; DP, distal pancreatectomy; TP, total pancreatectomy.

CAFs distribution types and patient outcomes

We performed a Cox regression analysis to examine the effect of postoperative clinicopathological parameters on prognosis. Univariate analyses revealed that TNM stage (II vs. I: HR, 1.827; 95 % CI, 1.257 - 2.656; P = 0.002; III vs. I: HR, 3.359; 95 % CI, 2.136 - 5.283; P < 0.001), spatial intensity of α-SMA (Type Ⅰ vs. Ⅱ: HR, 1.934; 95 % CI, 1.322 - 2.828; P = 0.001; Fig. 4A) and tumor grade (G3/4 vs. G1/2: HR, 1.794; 95 % CI, 1.253 - 2.567; P = 0.001) were significantly associated with OS in the discovery cohort (Supplementary Table 4). The Univariate analysis results of the validation cohort were parallel to those of the discovery cohort (Supplementary Table 4). Furthermore, the multivariable analysis confirmed that TNM stage (II vs. I: HR, 1.620; 95 % CI, 1.093 - 2.402; P = 0.016; III vs. I: HR, 2.144; 95 % CI, 1.307 - 3.519; P = 0.003), spatial intensity of α-SMA (Type Ⅰ vs. Type Ⅱ: HR, 1.568; 95 % CI, 1.053 - 2.334; P = 0.027), and tumor grade (G3/4 vs. G1/2: HR, 1.752; 95 % CI, 1.197 - 2.566; P = 0.004) were independent prognostic factors in the discovery cohort (Fig. 4C, Supplementary Table 4). The above-mentioned independent prognostic factors were also proved in the validation cohort (Fig. 4B, D, Supplementary Table 4).

Fig. 4.

Two CAFs types differentially associated with patient survival of PDAC. Kaplan–Meier curves for overall survival according to CAFs type in discovery cohort (A) and validation cohort (B). The P value was calculated using the log-rank test. The table shows the number of patients who remained alive and at risk of death at each time point. Forest plot showing multivariable hazard ratios in the Cox regression model for overall survival in discovery cohort (C) and validation cohort (D). Detailed results of the multiplex Cox regression analysis are shown in Supplementary Table 3. Abbreviations: CAFs, Carcinoma Associated Fibroblasts; PDAC, pancreatic ductal adenocarcinoma.

CAFs spatial patterns revealed tumor heterogeneity

To investigate the heterogeneity of tumor tissues associated with different CAF spatial patterns, we conducted transcriptomic analysis on samples from 24 PDAC patients (Type Ⅰ, n = 13; Type Ⅱ, n = 11). Our results demonstrated significant transcriptomic differences between tumor tissues associated with Type I and Type II subtypes (Fig. 5A), with more pronounced inter-subtype distinctions despite some observed intra-subtype heterogeneity. Subsequent differential expression analysis was performed on tumor tissues associated with Type I and Type II subtypes. This analysis identified 1038 significantly differentially expressed genes (DEGs) with a false discovery rate (FDR) < 0.05 (Supplementary Table 5). Notably, the majority of these differentially expressed genes were found to be present in Type I tumors (Fig. 5B). To elucidate the functional implications of the DEGs, we performed GO enrichment analysis and ssGSEA analysis. Our analysis demonstrated that Type I tumors, compared to Type II, exhibited significantly enhanced activation of immune-related biological pathways. Furthermore, Type I tumors showed enrichment across nearly all immune and stromal-related functional gene signatures (Fges) (Fig. 5C, D) [27]. Moreover, we utilized the ESTIMATE algorithm to evaluate the immune status of the tumors. Additionally, IHC analysis of immune cells revealed a significantly higher CD4+ cell content in Type I tumors compared to Type II (Table 1 and Supplementary Fig. 5). This analysis demonstrated that Type I tumors had significantly higher immune scores compared to Type II tumors, corroborating the enhanced immune-related pathway activation observed in our initial analyses (Fig. 5E). Further GSEA analysis demonstrated upregulation of inflammatory and oncogenic pathways in Type I tumors, including IL-6/JAK/STAT3 signaling and p53-mediated processes (Fig. 5F, G and Supplementary Table 6). Moreover, analysis of the overall Ki-67 positivity in tumor samples revealed significantly higher Ki-67 expression in Type I tumors compared to Type II (Supplementary Table 7). These findings demonstrate that Type I tumors exhibit increased immune cell infiltration and enhanced proliferative potential. These molecular characteristics correlate with our clinical observations, where patients with Type I present with more advanced TNM stages and poorer prognosis. This alignment between molecular profiles and clinical outcomes underscores the potential significance of CAF subtypes in tumor progression and patient survival.

Fig. 5.

CAFs spatial patterns correlate with distinct transcriptomic characteristics (A) The PCA plot visualizes the overall transcriptomic landscape based on CAFs patterns. (B) Heatmap plot showing the DEGs in Type I tumors and Type II tumors. Only genes with adjusted p-values < 0.05 and absolute log2 fold change of 1 or above will be considered significantly differentially expressed. (C) GO enrichment analysis of biological processes associated with DEGs. (D) Heatmap presents a comprehensive view of 29 tumor-related Fges across 24 PDAC samples (Type I, n = 13; Type II, n = 11). (E) Comparison of Immune Scores between different CAFs patterns. (F) and (G) GSEA analysis plots for the IL-6/JAK/STAT3 signaling and p53-mediated processes. (H) Comparison of Type I and Type II CAFs. (I) GO enrichment analysis of biological processes associated with DEGs unique to Type I CAFs. (J) Heatmap present the expression patterns of top 300 unique DEGs to Type I CAFs expressed in tumors. Abbreviations: PCA, principal components analysis; CAFs, Carcinoma Associated Fibroblasts; DEGs, differently expressed genes; GO, Gene Ontology; Fges, Functional gene signatures.

To further elucidate the distinctions between CAF patterns, we conducted transcriptome analysis on isolated CAFs from six pancreatic ductal adenocarcinoma (PDAC) patients. The study included three patients for each CAF pattern (Type I and Type II) and a control group of normal CAFs (n = 3). Our analysis revealed a core set of 1499 differentially expressed genes (DEGs) shared between the two CAF patterns, along with pattern-specific DEG signatures. Notably, we identified 131 genes that overlapped with the DEGs observed in tumor tissues, suggesting a potential functional link between CAF patterns and tumor characteristics. (Fig. 5H). Gene Ontology analysis highlighted enrichment of antigen presentation-related pathways among unique DEGs in Type Ⅰ CAFs (Fig. 5I). Notably, these findings parallel the GO results observed in the associated tumor tissues, suggesting a functional alignment between Type I CAFs and their corresponding tumors in terms of antigen presentation processes. Given the observed similarities between the DEGs and GO enrichment results in CAFs and tumors, we sought to determine whether the DEGs identified in CAFs could effectively distinguish between different tumor subtypes. Our analysis revealed that these CAF-derived DEGs indeed exhibited significant differential expression patterns in their corresponding tumor samples. (Fig. 5J). This finding not only confirms a strong molecular correspondence between CAF subtypes and their associated tumor microenvironments but also suggests that CAF-derived gene signatures may serve as potential biomarkers for tumor subtype classification. This alignment between CAF and tumor molecular profiles underscores the intricate interplay within the tumor microenvironment and highlights the potential of CAF-based analyses in enhancing our understanding of tumor heterogeneity and progression.

Discussion

The PDAC microenvironment is characterized by a dense and fibrotic stroma made up of CAFs, which are thought to contribute to tumor cell proliferation, invasion, immunosuppression and drug resistance in vitro and in vivo models of the pancreatic cancer [6,[29], [30], [31], [32]]. Whereas the evidently different outcomes of targeted depletion of α-SMA positive fibroblasts and FAP-α positive fibroblasts in PDAC indicated that the distinctions of CAFs subpopulations are vital [6]. In further, the intratumoral heterogeneity of CAFs function in PDAC were extensively investigated [10,33], and it suggested that pro-tumorigenic subpopulations will be crucial to delete in stroma-targeted therapies but not these tumor suppressive subpopulations. However, due to the intertumoral heterogeneity[10], the tumors with more pro-tumorigenic CAFs may be the best candidates for stroma-targeted therapies. Therefore, a robust and standardized method for detecting and identifying CAFs subtypes in tissue is urgently needed to be explored [34,35]. Considering the convenience of clinical studies and trials for patients’ selection, detecting markers of CAFs subtypes by IHC may uncover features of specific CAFs subpopulations within the appropriate tissue context [34,36]. The improved manner for identifying CAFs, together with enhanced understanding of CAF complexity and function, have the potential to foster the development of effective therapies for this highly malignant cancer. In the present study, we analyzed the spatial characteristics and prognostic significance of CAFs distribution patterns in two cohorts of 390 PDAC specimens. We found that α-SMA spatial intensity was the most reliable candidate for distinguishing tumors with different spatial characteristics of CAFs across the patients, and we firstly defined CAFs feature as Type Ⅰ and Type Ⅱ according to spatial characteristics by staining spatial intensity of α-SMA. In further, the function of patterns of CAFs were investigated by isolated CAFs from patients. Notably, spatial characteristics of CAFs were a strong and independent predictor of patient outcomes and Type Ⅰ CAFs were significantly associated with shorter survival time. Many of the previous studies focused on CAFs in PDAC have investigated several markers by IHC, including α-SMA [15], MMP-11 [37],POSTN [16], FAP-α [38] and PDGFR-β [39]. However, these studies have been limited by small sample size or shortcomings of detection methods for CAFs markers. Moreover, the spatial distribution feature of CAFs were thoroughly explored in both murine and human PDAC [5,7,40,41], and the function of CAFs subpopulations were also extensively investigated[14] by single-cell analysis. Meanwhile, previous research reported that the intertumoral heterogeneity of CAFs subpopulations were revealed in PDAC, potentially reflecting a intimate relationship of tumor invasiveness and CAFs [42]. Consequently, the function combined with spatial feature of CAFs may suggest a new insight of CAFs complexity, so as the immune cells of TME [43,44].

Previous studies with human tissue cohorts have indicated the spatial expression profiles of several CAFs markers [18,45], but there is still no research focused on associations between spatial distribution characteristics of CAFs and worse prognosis in patients with PDAC. In the current study, based on routine pathological examination system of IHC staining, we firstly explored the most appropriate candidate marker for CAFs to demonstrate the spatial difference of CAFs distribution patterns. Moreover, α-SMA was firstly recommended for recording of CAFs in clinical studies and trials [34]. Meanwhile, α-SMA was also recommended as a biomarker related to the epithelial-to-mesenchymal transition by the multiplexed tissue imaging mass cytometer in the previous research [46]. Therefore, we intended to explore the importance of spatial distribution of α-SMA positive CAFs on prognosis prediction. In further, we utilized TMA by IHC staining to characterize the α-SMA expression in two cohorts of PDAC tissue. We found that PDAC with Type Ⅱ CAFs, α-SMA spatial intensity in juxtatumoural stromal beyond than or equal to moderate and in peripheral stromal less than moderate, had a good prognosis. Interestingly, the good outcome may contribute to the Type Ⅱ CAFs, which characterized by much more myo-CAFs surrounding the malignant tumor cells, and the myo-CAFs’ biological function was antitumor activity as previous researches [[47], [48], [49], [50]]. In addition, we also found that Type Ⅰ CAFs was closely associated with late N and TNM stages, which could explain the worse prognosis of PDAC with Type Ⅰ CAFs. Meanwhile, we speculated that tumor-restraining property [51] of Type Ⅰ CAFs was weakened compared with that of Type Ⅱ CAFs, and late N stage may be caused by the Type Ⅰ CAFs. As a latest study further clarified that collagen I produced mainly by myo-CAFs can negatively regulate the immune-suppressive environment in PDAC [52,53]. Moreover, we investigated the function of Type Ⅰ CAFs by RNA sequence analysis and indicated that the Type Ⅰ CAFs were correlated with upregulation of inflammatory and oncogenic pathways, as previous reports [54,55]. Moreover, we found the intricate interplay of CAFs and tumor cells within the tumor microenvironment and highlights the potential of CAF-based analyses in enhancing our understanding of tumor heterogeneity and progression.

Our study has important strengths. Firstly, reviewing the previous research for CAFs markers comprehensively and arduously, we finally selected out a reliable and easy-detecting marker by a qualified IHC method and a reasonable flow path, enabling high confidence for exhibition of CAFs’ spatial distribution patterns, which could not be identified using dissociative techniques such as flow cytometry or single cell RNA sequencing. Secondly, the two patterns of CAFs based on classification of spatial distribution feature had a good performance for outcome prediction in our two cohorts of resected, previously untreated tumors from a high-volume center. Finally, the two patterns of CAFs in the present research could be identified conveniently by the pathologists in the routine pathological examination. Hence, the classification method is more likely to be used in other center for clinical practice and clinical trials.

The present study also has several limitations that require consideration. Firstly, our study has the intrinsic shortcomings of any retrospective study. Secondly, the approaches of IHC provide only semi-quantitative measurements and have limitations such as false negatives and false positives, although both negative control and positive control for α-SMA are setted in the present study. Additionally, we only included a limited number of markers in our analysis. To address this, we plan to use single-cell sequencing and other methods to identify and analyze additional CAF-related markers. In the future studies, the digital image analysis could be used for the precise quantification of intensities, but which may compromise the convenience of the clinical routine practice. Finally, as we were unable to obtain an external validation cohort based on public databases, external and functional validation are currently ongoing.

Conclusion

The current study shows that two patterns of CAFs with different spatial distribution for α-SMA spatial intensity are present heterogeneously across patients with PDAC. The spatial characteristics of CAFs subpopulations is associated with patient outcomes, revealing complexity of the CAFs’ function, and which may be important to facilitate precision medicine for patients with PDAC.

CRediT authorship contribution statement

Bo Li: Data curation. Meilong Shi: Data curation. Yang Wang: Data curation. Penghao Li: Data curation. Xiaoyi Yin: Data curation. Guoxiao Zhang: Data curation. Xiaochao Kang: Data curation. Huan Wang: Data curation. Suizhi Gao: Data curation. Kailian Zheng: Data curation. Xiaohan Shi: Data curation. Xiongfei Xu: Data curation. Yukun Zhou: Data curation. Hui Jiang: Data curation. Wei Jing: Data curation. Shiwei Guo: Data curation. Gang Jin: Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.