Li et al.

Molecular Medicine (2022) 28:43

https://doi.org/10.1186/s10020-022-00467-8
Molecular Medicine

RESEARCH ARTICLE Open Access

Identification and prognostic analysis

of biomarkers to predict the progression
of pancreatic cancer patients
Wei Li1,5* , Tiandong Li6, Chenguang Sun2, Yimeng Du5, Linna Chen5, Chunyan Du3*, Jianxiang Shi4* and
Weijie Wang2*

Abstract
Background: Pancreatic cancer (PC) is a malignancy with a poor prognosis and high mortality. Surgical resection is
the only “curative” treatment. However, only a minority of patients with PC can obtain surgery. Improving the overall
survival (OS) rate of patients with PC is still a major challenge. Molecular biomarkers are a significant approach for
diagnostic and predictive use in PCs. Several prediction models have been developed for patients newly diagnosed
with PC that is operable or patients with advanced and metastatic PC; however, these models require further valida-
tion. Therefore, precise biomarkers are urgently required to increase the efficiency of predicting a disease-free survival
(DFS), OS, and sensitivity to immunotherapy in PC patients and to improve the prognosis of PC.
Methods: In the present study, we first evaluated the highly and selectively expressed targets in PC, using the
GeoMxTM Digital Spatial Profiler (DSP) and then, we analyzed the roles of these targets in PCs using TCGA database.
Results: LAMB3, FN1, KRT17, KRT19, and ANXA1 were defined as the top five upregulated targets in PC compared
with paracancer. The TCGA database results confirmed the expression pattern of LAMB3, FN1, KRT17, KRT19, and
ANXA1 in PCs. Significantly, LAMB3, FN1, KRT19, and ANXA1 but not KRT17 can be considered as biomarkers for
survival analysis, univariate and multivariate Cox proportional hazards model, and risk model analysis. Furthermore, in
combination, LAMB3, FN1, KRT19, and ANXA1 predict the DFS and, in combination, LAMB3, KRT19, and ANXA1 predict
the OS. Immunotherapy is significant for PCs that are inoperable. The immune checkpoint blockade (ICB) analysis
indicated that higher expressions of FN1 or ANXA1 are correlated with lower ICB response. In contrast, there are no
significant differences in the ICB response between high and low expression of LAMB3 and KRT19.
Conclusions: In conclusion, LAMB3, FN1, KRT19, and ANXA1 are good predictors of PC prognosis. Furthermore, FN1
and ANXA1 can be predictors of immunotherapy in PCs.

*Correspondence: [email protected]; [email protected];

[email protected]; [email protected]
1
Department of Hematology, The First Affiliated Hospital of Zhengzhou
University, Zhengzhou 450052, Henan, China
2
Department of Hepatobiliary Surgery, The First Affiliated Hospital
of Zhengzhou University, Zhengzhou 450052, Henan, China
3
Laboratory Animal Center, School of Medical Sciences, Zhengzhou
University, Zhengzhou 450052, Henan, China
4
BGI College and Henan Institute of Medical and Pharmaceutical
Sciences in Academy of Medical Science, Zhengzhou University,
Zhengzhou 450052, Henan, China
Full list of author information is available at the end of the article

© The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.
Li et al. Molecular Medicine (2022) 28:43 Page 2 of 16

Keywords: Pancreatic cancer, Biomarkers, LAMB3, FN1, KRT19, ANXA1

Introduction features (Sorg et al. 2020). As such, direct interrogation

Pancreatic cancer (PC) is a highly malignant cancer of tools are required to enable characterization of local-
the gastrointestinal tract that is characterized by late ized transcriptomic changes in discrete tissue while
diagnosis, limited treatment success and a poor prog- preserving additional tissue for testing. The GeoMxTM
nosis (Chen et al. 2021). The morbidity and mortality of Digital Spatial Profiling (DSP) platform robustly quan-
PC continues to rise and constitutes a major challenge tifies high-plex RNA expression data from single, 5 μm
for basic and clinical oncologists (Chen et al. 2021). To sections, capturing genome-wide expression patterns in
date, surgical excision remains the only method with spatially resolved locations throughout the tissue, and
curative potential (Strobel et al. 2019). However, at the it has already been used to study many types of cancers
time of diagnosis, approximately 80–85% of patients (Van and Blank 2019; Chandramohan et al. 2021; Gong
have already advanced to either an unresectable or et al. 2020; Kalita-de Croft et al. 2021). In this article,
a metastatic state, which accounts for the 5-year sur- we present the DSP method to study PC. The GeoMx
vival rate of less than 10% of patients (Mizrahi et al. Cancer Transcriptome Atlas (CTA) targets are designed
2020). Even for the minority of patients who have the for comprehensive profiling of tumors, tumor microen-
opportunity to undergo surgery, the prognosis remains vironment, and tumor immune status with 1833 RNA
poor, with only 20% surviving for 5 years (Mizrahi et al. targets, including negative controls. We conducted
2020). Therefore, there is an urgent need to explore 1,833 RNA targets on both cancer and paracancer tis-
new biomarkers for a clinically meaningful impact on sues, and then we identified differentially expressed
the screening of patients with high-risk PC. genes between cancer and paracancer tissue samples.
Several studies have demonstrated that the carbohy- We then confirmed the expression pattern of the iden-
drate antigen 19-9 (CA 19-9) and the carcinoembryonic tified differentially expressed markers with the Cancer
antigen (CEA) can be used to predict the outcomes in Genome Atlas (TCGA) data set. Survival analysis, the
a variety of cancers, including PC (Stojkovic Lalose- univariate and multivariate Cox proportional hazards
vic et al. 2017; Zhou et al. 2017; Distler et al. 2013; Liu model, risk model analysis and ICB responses analy-
et al. 2018a). However, these biomarkers are not suf- sis were used to confirm the predictive roles of these
ficiently specific or sensitive for use in PC (Zhu et al. markers in PCs. Our approach for combining these
2017; Swords et al. 2016). In addition, high-throughput two powerful technologies has the potential to provide
sequencing has identified a large number and different meaningful biological insight in PCs.
varieties of biomarkers; however, few hold future prom-
ise as a preferred marker for PC (Ballehaninna and Materials and methods
Chamberlain 2013). This is because most of these dis- Fresh frozen section preparation and hematoxylin
covered biomarkers have significant limitations, such and eosin (H&E) staining
as a lack of methodological standardization and quality Cancer and paracancer tissue samples were obtained
control; limited or no correlation with tumor stage and from the same patient during surgery and immediately
tumor invasives; and minimal utility to assess progno- embedded into an optimal cutting temperature (OCT)
sis, predict tumor recurrence, or evaluate tumor immu- compound. Thereafter, H&E staining was performed, as
notherapy response (Ballehaninna and Chamberlain previously described (Cao et al. 2022). Briefly, the fro-
2013; Nixon et al. 2013; Sahni et al. 2020; Pereira et al. zen sections are allowed to warm at room temperature
2020). for 5–10 min, and then the rewarmed slices are soaked
PC is a special kind of tumor; the number of tumors in water for approximately 30–60 s. The tissues are
that are operable are small, and the paraneoplastic immersed in distilled water for 1–2 min, while ensur-
and cancerous tissues are not easily obtained or dis- ing that the distilled water covers the tissues entirely and
tinguished from cancerous tissue (Nixon et al. 2013). that the water is evenly distributed. The tissues are then
Pathology testing often requires the use of as few or stained with hematoxylin staining solution for 5 min and
limited tissue samples as possible, while testing for mul- washed with water for 3–5 s. This is followed by stain-
tiple biomarkers. High-plex profiling of gene expression ing with eosin for 2 min, washing with distilled water for
under bulk or single-cell analysis provides rich contex- 1–2 min, and blotting with filter paper or natural dry-
tual information but consumes large fractions of tis- ing. Finally, CaseViewer software was used to analyze the
sue and lacks spatial context of the key morphological results of the cancer and paracancer tissue samples.
Li et al. Molecular Medicine (2022) 28:43 Page 3 of 16

Immunofluorescence (IF) study. (1) Technical signal QC is a sequencing quality

The frozen sections were removed and allowed to dry at evaluation for each AOI/ROI with three metrics: raw
room temperature for 15 min. They were then soaked in reads, aligned reads percentage, and sequencing satura-
PBS solution for 10 min to remove the OCT compound. tion. Raw reads include all the reads sequences of AOI/
The tissue to be stained was circled with a histochemi- ROI at the time of sequencing. Aligned reads percentage
cal grease pencil to allow effective incubation of the is the percentage of reads sequences in the AOI/ROI that
antibodies. The sections were permeabilized with 0.5% are compared to the template sequences. Sequencing sat-
TritonX-100 at room temperature for 20 min and closed uration are the sequencing reads of an AOI/ROI that can
with PBS solution containing 10% normal goat serum for be sequenced once or more times; sequencing saturation
1 h at room temperature to improve the accuracy of the is the percentage of reads that are detected at least twice
target protein and to reduce the background. The anti- in raw reads, and it is recommended that it should exceed
bodies CD45-Alexa 594 and PanCK-488 were added, and 50%.
the sections were incubated at 4 °C overnight. Then, PBS (2) Technical background QC is a running compari-
washed three times for 10 min each. Syto 13 was incu- son of GeoMx DSP settings, divided into two metrics, no
bated for in the dark for 5 min to stain the nuclei and PBS template control count (NTC count) and negative probe
washed three times for 10 min each. Finally, the fluores- count. NTC count is a negative control experiment set
cence microscope was used to photograph the results. up during each CTA experiment, and it does not contain
a template for detecting template contamination during
GeoMx DSP for CTA profiling library construction. The NTC value for this CTA experi-
GeoMx Cancer Transcriptome Atlas Panel was used ment is 3000. Negative probe count is the number of
in this study and fresh surgically excised cancerous and negative probes in each AOI/ROI, which is used to meas-
paracancerous tissue samples were immediately embed- ure the technical noise level of the CTA experiment. Low
ded in OCT compound to prepare frozen sections. technical noise requires consideration of whether the
Then, the fresh-frozen tissues were processed, follow- AOI/ROI area is too small, whether the library build is
ing the GeoMx DSP slide prep user manual as previ- successful, and whether the sequencing depth is too low.
ously described (Desai et al. 2020). Briefly, (1) RNA target Based on the empirical value of the negative probe meas-
exposure: proteinase K was added prior to the incuba- urement, a technical noise of five or more is required for
tion with RNA probe mix (CTA panel). (2) Staining: each CTA experiment. GeoMx DSP will limit the nuclei
after overnight incubation, the slides were washed with counts and surface area for each AOI/ROI, and too low
buffer and stained with CD45-Alexa 594, PanCK-488, nuclei counts or surface area is not recommended.
and SYTO 13 for 1 h and loaded into the GeoMx DSP (3) Probe QC includes two aspects, namely, probe out-
machine to scan the fluorescent images. (3) Regions lier detection and target LOQ detection. Target is the
of interest (ROIs) selection: ROIs were placed by align- gene to be detected, and GeoMx DSP will design multiple
ing to those ROIs identified during protein profiling. (4) probes to detect each target. LOQ is the limit of quan-
Sequencing library construction: oligos were cleaved titation, and, if a target’s count value is below LOQ, it
and collected into 96-well plates. Oligos from each area does not mean it is not expressed, while, if it is greater
of interest (AOI) were uniquely indexed using Illumina’s than LOQ, its expression can be confirmed. LOQ repre-
i5 × i7 dual-indexing system. Four μL of a GeoMx DSP sents the limit value of a target confirmation expression,
sample was used in the PCR reaction. The PCR reac- which is calculated as follows: LOQ = GeoMean (Neg-
tions were purified with two rounds of AMPure XP Probe) × GeoSD (NegProbe)threshold. In CTA, a threshold
beads (Beckman Coulter) at a 1:2 bead-to-sample ratio. value of 2.5 represents a more stringent standard, while
(5) Sequencing: Libraries were paired-end sequenced a threshold value of 2.0 represents a slightly more leni-
(2 × 75) on a NextSeq550 up to 400 M total aligned reads. ent standard. In this CTA experiment, the LOQ value
(6) Analytical process: Fastq files were processed by the was 102 and the threshold value 2.0. GeoMx DSP was
NanoString DND pipeline to generate count files for each designed with multiple probes for target genes, and, if a
target probe and saved as DCC files. probe performed too abnormally, it was rejected. There
are two rejection indicators, low outlier detection and
Grubb’s outlier test. Low outlier detection is defined as
GeoMx RNA data quality control (QC) low outlier if the count value of a probe is less than 10%
GeoMx DSP data analysis software was used for primary of the average count value of other probes. Grubb’s out-
analysis and then referring to annotation information lier test was performed for all probes in each target. If a
included in Additional file 3: Table S1 for data interpre- probe was marked as an outlier in more than 20% of the
tation. Three methods of QC were used in the present AOI/ROI, this probe was removed.
Li et al. Molecular Medicine (2022) 28:43 Page 4 of 16

Differential expression analysis of GeoMx RNA data the median value of LAMB3, FN1, KRT19, or ANXA1. A
For each tissue and each gene, differential expression univariate Cox analysis was performed to identify poten-
versus cancer presence/absence was evaluated with tial prognostic factors, and a multivariate Cox analysis
an unpaired, heteroscedastic t-test of the gene’s log2- was used to determine risk score as an independent risk
transformed normalized data. For analysis of expression factor for OS in PC. A ROC curve was generated to vali-
within the tissues, genes were defined as consistently date the accuracy of the risk model in predicting patients’
upregulated if they had (1) a log2 fold change > 0.2 and a DFS and OS via the survival ROC R package.
Benjamini–Hochberg FDR < 0.1 in at least 2 ROIs, and (2)
never had a log2 fold change < 0 and a Benjamini–Hoch- Immune checkpoint blockade (ICB) response analysis
berg FDR < 0.1 in any ROI. Genes were defined as consist- Raw counts of RNA-sequencing data (level 3) and corre-
ently downregulated by an equivalent rule. For analysis sponding clinical information from 178 PC patients were
across patient samples, expression was modeled using obtained from The TCGA dataset (https://​portal.​gdc.​
linear mixed effect models that allow for random slope cancer.​gov/), in which the methods of acquisition and
and intercept terms per patient sample. P-values were application complied with the guidelines and policies.
estimated with Satterthwaite’s method for approximation Potential ICB response was predicted with a TIDE algo-
and adjusted with the Benjamini–Hochberg FDR. Only rithm, as previously described (Jiang et al. 2018). Briefly,
genes that rose twofold above-background in at least one TIDE uses a set of gene expression markers to assess 2
ROI were considered. different mechanisms of tumor immune escape, includ-
ing dysfunction of tumor-infiltrating cytotoxic T lympho-
Differentially expressed genes analysis and KEGG cytes (CTL) and exclusion of CTL by immunosuppressive
pathways analysis of PCs from the TCGA database factors. High TIDE scores are associated with poor effi-
Tumoral RNA-seq data for the 178 PC patients on the cacy of immune checkpoint blockade therapy (ICB) and
TCGA database were downloaded from the Genomic short survival after receiving ICB therapy.
Data Commons (GDC) data portal (TCGA), and four
of the tumors also had RNA-seq data of paired normal Statistics analysis
tissue samples. All data of normal tissue samples were R version 4.0.3. software was used for all statistical analy-
obtained from 328 pancreases in GTEx V8 release ver- ses. The statistical details of all experiments are reported
sion (https://​gtexp​ortal.​org/​home/​datas​ets). The limma in the materials and methods, including statistical analy-
package of R software was used to analyze the differ- sis performed and statistical significance.
entially expressed genes as our previously described
(Cao et al. 2022; Li et al. 2019; Wang et al. 2021). Briefly, Results
“Adjusted P < 0.05 and Log2 (FC) > 1 or Log2 (FC) < − 1” Selection ROIs between pancreatic cancer and paracancer
were defined as the thresholds for the screening of dif- To generate unbiased transcriptomic maps of the PC tis-
ferentially expressed genes. To better understand the car- sue sections, we mounted cryosections of freshly frozen
cinogenesis of mRNA, the clusterProfiler package in R cancer tissue and paracancer tissue that originated from
software was employed to analyze the enrich the Kyoto the same patient onto the spatially barcoded microar-
Encyclopedia of Genes and Genomes (KEGG) pathway. ray slides. The age of this patient is 61-years old with
ECOG score of 0. The patient was received surgery at
Kaplan–Meier curves, univariate and multivariate Cox March, 22, 2021. And the patient’s pathological type is
proportional hazards model and risk model analysis a medium- to low-differentiated adenocarcinoma of the
RNA-sequencing expression profiles and correspond- tail of the pancreatic body. The TNM stage of this patient
ing clinical information for PC patients were down- is stage II. Then, we processed the freshly frozen cancer
loaded from the TCGA dataset (https://​portal.​gdc.​com). tissue and paracancer tissue by H&E staining (Fig. 1A).
The Kaplan–Meier curves, univariate and multivariate Although the two tissue sections are from cancer tissue
Cox proportional hazards model, and risk model analy- and paracancer tissue, the paracancer tissue circled in
sis (Lasso cox regression) based on the expression of red resembles cancerous tissue morphologically, indicat-
LAMB3, FN1, KRT19, and ANXA1 were constructed, as ing that paracancer tissues contain normal pancreatic
previously described (Lin et al. 2020). Disease-free sur- tissue and cancer tissue. To obtain ROIs from the two
vival time (DFS) and Overall-survival time (OS) were tissue samples, we stained them with CK, CD45, and
compared between the high and low LAMB3, FN1, Syto 13 (Fig. 1B). In the paracancer tissue, we selected 11
KRT19, and ANXA1 risk groups via a Kaplan–Meier ROIs (Fig. 1C), which included five cancer tissues (001,
analysis, using the survival and survminer packages in R. 002, 003, 004, and 011), and six normal tissues (005, 006,
The cutoff value for the high and low risk was defined as 007, 008, 009, and 010). In the cancer tissue sample, we
Li et al. Molecular Medicine (2022) 28:43 Page 5 of 16

Fig. 1 A Paracancer and cancer samples from a PC patient as confirmed by H&E staining. B Paracancer and cancer samples from a PC as assessed
by immunofluorescence (IF) staining to select appropriate ROIs. Green for CK staining, red for CD45 staining, and blue for Syto 13 staining. C The
selected ROIs shown in the paracancer sample. D The selected ROIs shown in the cancer sample

selected 12 ROIs (Fig. 1D, 001–012). The ROIs were GeoMx DSP parameters. Additional file 1: Fig. S1A–C
then processed for CTA analysis, including complemen- shows raw reads, aligned reads percentage, sequenc-
tary DNA synthesis, amplification by in vitro transcrip- ing saturation, negative probe count, nuclei counts, and
tion, library construction, and sequencing, as previously surface area. We next performed the probe QC, which
described (Chandramohan et al. 2021). The GeoMx CTA includes two aspects, probe outlier detection and the
Panel was used to analyze the expression pattern of the target LOQ detection. Additional file 1: Fig. S1D and E
two sections. demonstrate the low outlier detection, Grubbs outlier
test, and the LOQ. All the metrics demonstrate satisfac-
Comparison of the differentially expressed genes tory sequencing quality. Thereafter, we used an unbiased
between cancer and paracancer tissues by DSP analysis tSNE clustering of all ROIs and identified three distinct
We then analyzed the data with the GeoMx Digital Spa- clusters (Fig. 2A). The cluster circled in red contains the
tial Profiler (DSP). To begin, we performed quality control ROIs 001, 002, 003, 004, and 011 from the paracancer tis-
(QC) on the data. Segment QC was performed to quality sue, which resembles cancerous tissue morphologically.
control the area of interest (AOI)/ROI, which includes The other ROIs from the cancer tissue, 001–007, are also
technical signal QC, technical background QC, and contained in this cluster. This suggests that these ROIs
Li et al. Molecular Medicine (2022) 28:43 Page 6 of 16

Fig. 2 A Principal component analysis (PCA) of DSP data demonstrate paracancer ROIs 001–011 and cancer ROIs 001–012. B Principal component
analysis (PCA) of DSP data demonstrates paracancer ROIs 005–010 and cancer ROIs 008–012. C Principal component analysis (PCA) of DSP data
demonstrates paracancer ROIs 001–004 and 011 and cancer ROIs 001–007. D Volcano map display of the differentially expressed genes between
paracancer ROIs 001–011 and cancer ROIs 001–012. E Volcano map display of the differentially expressed genes between paracancer ROIs 005–010
and cancer ROIs 008–012. F Volcano map display of the differentially expressed genes between paracancer ROIs 001–004 and 011 and cancer ROIs
001–07. G The mean expression level of upregulated target of LAMB3 in paracancer ROIs 005–010 and cancer ROIs 008–012. H The mean expression
level of upregulated target of LAMB3, FN1, KRT17, and KRT19 in paracancer ROIs 005–010 and cancer ROIs 008–012. I The pathway network in
cancer ROIs 001–012. J The pathway network in cancer ROIs 008–012
Li et al. Molecular Medicine (2022) 28:43 Page 7 of 16

from the two tissue samples are similar, which is consist- more, all of which were highly associated with the upreg-
ent with H&E staining. We then performed PCA cluster- ulated genes (Fig. 2I and J). Taken together and using the
ing separately for those with and without differences in GeoMx CTA Panel, we identified five highly and selec-
ROIs again to compare the differentially expressed genes tively expressed genes in the PC tissues.
(Fig. 2B and C). In the present study, we selected genes
with Log2 FC > 1 as significantly differentially expressed. Survival analysis demonstrated distinct prognostic effects
Figure 2D demonstrates the significantly upregulated of LAMB3, FN1, KRT17, KRT19, and ANXA1 in pancreatic
and downregulated genes between all the ROIs from the cancer
paracancer tissue and the cancer tissue. Figure 2E illus- The expression pattern of LAMB3, FN1, KRT17,
trate the significantly upregulated and downregulated KRT19, and ANXA1 was then confirmed in 178 PC
genes between ROIs with differences. However, there patients (Stage I: 24 patients, Stage II: 146, Stage III:
are no significant differentially expressed genes between 3, Stage IV: 5) from the TCGA database. Compared
ROIs without differences (Fig. 2F). Consistent with PCA with the adjacent tissues (N = 4) and normal tis-
analysis, similar ROIs showed no differentially expressed sues (N = 328), LAMB3, FN1, KAR17, KAR19, and
genes, which further confirms that ROIs 001,002,003,004, ANXA1 were significantly upregulated (Fig. 3A). The
and 011 from the paracancer tissue are cancer tissue. Kaplan Maier survival analysis with log-rank tests
Then, we analyzed the upregulated top five differentially was used to compare the difference in DFS and OS
expressed genes between the paracancer tissue and the between high and low expression of LAMB3, FN1,
cancer tissue, which included LAMB3, FN1, KRT17, KAR17, KAR19 and ANXA1. Sixty-nine of the 178
KRT19, and ANXA1 (Fig. 2G and H). All the expression patients had undergone surgery, and we analyzed the
of the 1833 targets from the paracancer tissue and the DFS based on those 69 patients. The analysis of OS
cancer tissue are shown in Additional file 4: Table S2. The was based on all 178 patients. It is noteworthy that the
KEGG pathway enrichment analysis of the cancer tissue ­ AMB3hi, ­FN1hi, ­K RT19hi,
analysis indicated that the L
hi
included laminin interactions, degradation of the extra- and ­ANXA1 groups demonstrated shorter DFS
cellular matrix, extracellular matrix organization, and (undefined VS. 1.962 years, P = 0.0057; undefined VS.

Fig. 3 A The mRNA expression levels of LAMB3, FN1, KRT17, KRT19, and ANXA1 as assessed by RNA-seq between PCs and controls (including
paracancer or normal pancreas). B The Kaplan–Meier DFS curves for PC patients assigned to high and low expression groups of LAMB3, FN1, KRT17,
KRT19, and ANXA1 based on the expression level, respectively. The median was selected as the cutoff value for high or low groups. The blue lines
are the low expression groups and the red lines are the high expression groups. C The Kaplan–Meier OS curves for PC patients assigned to high and
low expression groups of LAMB3, FN1, KRT17, KRT19, and ANXA1 based on the expression level, respectively. The median was selected as the cutoff
value for high or low groups. The blue lines are the low expression groups and the red lines are the high expression groups
Li et al. Molecular Medicine (2022) 28:43 Page 8 of 16

1.485 years, P = 0.0107; undefined VS. 2.277 years, Univariate analysis and multivariate Cox proportional
P = 0.0392; undefined VS. 1.962 years, P = 0.0101) and hazards model
OS (1.923 years vs 1.364 years, P = 0.0084; 1.874 years Subsequently, we conducted univariate and multivari-
VS 1.652 years, P = 0.0368; 1.811 years vs 1.364 years, ate Cox analyses to evaluate the independent prognos-
P = 0.0162; 1.904 years VS 1.493 years, P = 0.0054). In tic value of LAMB3, FN1, KRT19, and ANXA1 in terms
contrast, there were no significant differences for both of both DFS and OS of patients with PC. The univari-
DFS and OS based on the high and low expression of ate analysis indicated that the group with high LAMB3,
KRT17 (Fig. 3B and C). FN1, KRT19, and ANXA1 were significantly correlated
with shorter DFS (Fig. 4A) and OS (Fig. 5A). Multivariate
analyses revealed that the group with high LAMB3, FN1,

Fig. 4 A The univariate analysis revealing the hazard ratio, P‐values, and some parameters of LAMB3, FN1, KRT19, and ANXA1 in terms of DFS in PC
patients. B The multivariate Cox proportional hazards model revealing the hazard ratio, P‐values, and some parameters of LAMB3, FN1, KRT19, and
ANXA1 in terms of DFS in PC patients
Li et al. Molecular Medicine (2022) 28:43 Page 9 of 16

Fig. 5 A The univariate analysis evaluating the hazard ratio, P‐values, and some parameters of LAMB3, FN1, KRT19, and ANXA1 in terms of OS in PC
patients. B The multivariate Cox proportional hazards model evaluating the hazard ratio, P‐values, and some parameters of LAMB3, FN1, KRT19, and
ANXA1 in terms of OS in PC patients

KRT19, and ANXA1 were still independently associated FN1, KRT19, and ANXA1 in PC patients, which can
with significantly poorer DFS (Fig. 4B) and OS (Fig. 5B) effectively discern most of the available forecast mark-
of patients with PC, which could serve as independent ers and produce prognostic indicators for predicting
prognostic factors for PC. clinical results. We first analyzed the prognostic effects
of LAMB3, FN1, KRT19, and ANXA1 for predicting the
Construction and validation of a risk model according DFS of PC patients. The coefficients of the selected fea-
to LAMB3, FN1, KRT19, and ANXA1 in PC patients tures of LAMB3, FN1, KRT19, and ANXA1 are shown
Next, we constructed and validated a risk model (LASSO through the lambda parameter (Fig. 6A). Partial likeli-
Cox regression), according to the expression of LAMB3, hood deviance versus log (λ) was drawn using the LASSO
Li et al. Molecular Medicine (2022) 28:43 Page 10 of 16

Fig. 6 A LASSO coefficients profiles of LAMB3, FN1, KRT19, and ANXA1 are shown by lambda parameter in relation to DFS. B LASSO Cox regression
with ten-fold cross-validation obtained using minimum lambda value related to DFS. C Prognostic analysis of LAMB3, FN1, KRT19, and ANXA1
signatures in the TCGA set related to DFS. Top: The dotted line represents the median risk score and divides the patients into low-risk and high-risk
groups. Middle: Survival status of the patients are shown in red (alive) and blue (dead) dots. Down: Heatmap of the expression profiles of the four
prognostic genes (LAMB3, FN1, KRT19, and ANXA1) in the low- and high-risk groups relating to DFS. D The Kaplan–Meier DFS curves for PC patients
assigned to low-risk and high-risk groups. The blue line represents the low-risk group and the red line the high-risk group. E ROC curves showing
the predictive efficiency of the LAMB3, FN1, KRT19, and ANXA1 signatures relating to DFS on the 1-year, 3-year, and 4-year survival rate. (F) LASSO
coefficients profiles of LAMB3, FN1, KRT19, and ANXA1 are shown by lambda parameter relating to OS. G LASSO Cox regression with ten-fold
cross-validation obtained using minimum lambda value related to OS. H Prognostic analysis of LAMB3, FN1, KRT19, and ANXA1 signatures in the
TCGA set relating to OS. The dotted line represents the median risk score and divides the patients into a low-risk and a high-risk group. Survival
status of the patients are shown in red (alive) and blue (dead) dots. Heatmap of the expression profiles of the prognostic genes (LAMB3, FN1, KRT19,
and ANXA1) showing in the low- and high-risk groups related to OS. (I) The Kaplan–Meier OS curves for PC patients assigned to the low-risk and
high-risk groups. The blue line represents the low-risk group and the red line the high-risk group. J ROC curves showing the predictive efficiency of
the LAMB3, FN1, KRT19, and ANXA1 signatures relating to OS on the 1-year, 3-year, and 4-year survival rate

Cox regression model (Fig. 6B). Hence, LAMB3, FN1, was calculated as previously described (Lin et al. 2020).
KRT19, and ANXA1 were selected for the subsequent PC patients were categorized into low-risk group (n = 89)
multivariate analysis. And the risk score of every patient and high-risk group (n = 89), based on the median
Li et al. Molecular Medicine (2022) 28:43 Page 11 of 16

cut-off point obtained by “survminer” R package analy- The differentially expressed genes and KEGG pathways
sis (Fig. 6C top). The survival status and survival time analysis based on the high and low expression of LAMB3,
of patients in the two different risk groups are shown in FN1, KRT19, or ANXA1
the Fig. 6C middle, and the relative expression standards Given that LAMB3, FN1, KRT19, and ANXA1 can
of LAMB3, FN1, KRT19, and ANXA1 are presented in be considered as prognostic biomarkers for PCs, we
Fig. 6C down. The survival analysis demonstrated that finally analyzed the transcriptional expression differ-
the DFS of the high-risk group was overall shorter than ences between high and low expression of LAMB3, FN1,
that of the low-risk group (P = 0.0004; Fig. 6D). The KRT19, or ANXA1 in patients with PC. Additional file 2:
AUC was 0.803 at 1 year, 0.744 at 3 years, and 0.849 at Fig. S2A demonstrates the 433 upregulated genes and 89
5 years, indicating the high predictive value of combined downregulated genes in the L ­ AMB3hi group compared
low
LAMB3, FN1, KRT19, and ANXA1 (Fig. 6E). Using the with the L­ AMB3 group in volcano plots (Additional
same analysis method, we next examined the prognostic file 3: Table S1). The enriched KEGG signaling pathways
effects of LAMB3, FN1, KRT19, and ANXA1 for pre- were selected to demonstrate the primary biological
dicting the OS of PC patients, and only LAMB3, KRT19, actions of the major potential mRNA. Additional file 2:
and ANXA1 were selected for the multivariate analysis Fig. S2B shows the enrichment of KEGG pathways in the
for the OS (Fig. 6F and G). Figure 6H shows the two risk ­L AMB3hi group, including tight junction, PI3K-Akt sign-
groups, based on the expression of combined LAMB3, aling pathway, focal adhesion, ECM-receptor interaction,
KRT19, and ANXA1, and Fig. 6I shows that the low- and cell adhesion molecules. Additional file 2: Fig. S2C
risk group had a longer OS compared with the high-risk shows the enrichment of KEGG pathways in L ­ AMB3low
group. The AUC was 0.694 at 1 year, 0.694 at 3 years, and group, including pancreatic secretion, protein digestion
0.673 at 5 years, indicating the predictive value of com- and absorption, fat digestion and absorption, and drug
bined LAMB3, KRT19, and ANXA1 for the OS of PC metabolism-cytochrome P450. Using the same analysis
patients (Fig. 6J). method, we found that there are 387 upregulated genes
and seven downregulated genes in the ­FN1hi group com-
Immune checkpoint blockade (ICB) response analysis of PC pared with the ­FN1low group (Additional file 2: Fig. S2D,
patients, based on the high and low expression of LAMB3, Additional file 3: Table S3). The enrichment of KEGG
FN1, KRT19, and ANXA1 pathways in the F ­ N1hi group included the PI3K-Akt sign-
Cancer treatment with ICB can provide long-lasting clin- aling pathway, focal adhesion, and ECM-receptor inter-
ical benefits, but only a fraction of patients responds to action (Additional file 2: Fig. S2E). We then analyzed
this treatment. Tumor immune dysfunction and exclu- the differentially expressed genes and pathways between
sion (TIDE) were used to predict the potential ICB the ­KRT19hi group and ­KRT19low group. Four hundred
response in previous studies (Jiang et al. 2018; Wang and forty-three upregulated genes and 87 downregu-
et al. 2020). We evaluated the TIDE scores, based on the lated genes were identified in the K ­ RT19hi group com-
low
high and low expression of LAMB3, FN1, KRT19, and pared with K ­ RT19 group (Additional file 2: Fig. S2F,
ANXA1, in 89 PC patients undergoing ICB treatment. Additional file 6: Table S4). In addition, the enrichment
Remarkably, high expressions of FN1 or ANXA1 aligned of KEGG pathways in the ­KRT19hi group and ­KRT19low
with higher TIDE scores, compared with low expression group are similar to the L ­ AMB3hi group and ­L AMB3low
of FN1 (Fig. 7B) or ANXA1 (Fig. 7D). Significantly, higher groups (Additional file 2: Fig. S2G and H). Finally, there
TIDE scores were correlated to lower ICB response. In were 223 upregulated genes and 45 downregulated genes
contrast, there were no significant differences in TIDE in the ANXA1 hi group compared with the ­ANXA1low
scores based on the expression of LAMB3 (Fig. 7A) or group (Additional file 2: Fig. S2I, Additional file 7:
KRT19 (Fig. 7C). Next, we compared the expression of Table S5). The enrichment of KEGG pathways in the
immune checkpoints based on the expression of FN1 or ­ANXA1hi and A ­ NXA1low groups is presented in Addi-
ANXA1. Figure 7E indicated that FN1hi PC patients have tional file 2: Fig. S2J and K.
higher PDCD1LG1, CTLA4, TIM3, LAG3, PDCD1LG2,
and TIGIT, compared with low expression of FN1. Fig- Discussion
ure 7F demonstrated similar expression pattern of PC is a deadly disease with a 5-year survival rate of an
immune checkpoints for ANXA1hi PC patients. Taken approximately 10% in the USA, and it is becoming an
together, FN1 or ANXA1 expression can predict the increasingly common cause of cancer deaths (Miz-
response of PC patients to ICB. rahi et al. 2020). Surgical resection represents the only
option for a possible cure, and advancements in adjuvant
chemotherapy have improved long-term outcomes in
these patients (Mizrahi et al. 2020). In addition, targeted
Li et al. Molecular Medicine (2022) 28:43 Page 12 of 16

Fig. 7 A The TIDE score for PC patients showing in the ­LAMB3hi and ­LAMB3low groups. B The TIDE score for PC patients showing in the ­FN1hi and
­FN1low groups. C The TIDE score for PC patients showing in the ­KRT19hi and ­KRT19low groups. D The TIDE score for PC patients showing in the
­ANXA1hi and ­ANXA1low groups. E The mRNA expression of immune checkpoints for PC patients showing in the F­ N1hi and ­FN1low groups. F The
mRNA expression of immune checkpoints for PC patients showing in the A ­ NXA1hi and ­ANXA1low groups

therapy and immunotherapy have increased the OS of personalized treatment measures will be a more rational
advanced and metastatic patients (Strobel et al. 2019). treatment approach.
However, PC is still an uncurable malignant disease. Previous studies have reported several predictive mark-
Many strategies aimed at finding novel targeted therapy ers for risk estimation, such as S100P, ERO1LB, SULF1,
and immunotherapy, such as deconstructing the sur- ITGA2, GPRC5A, ACTN4, LMO2, p16INK4a, CLPS,
rounding desmoplastic stroma and targeting the immu- COL11A1, GJB2, CTRL, CEL, CPA1, POSTN, PM20D1,
nosuppressive pathways have largely failed (Ho et al. and MARCO (Lin et al. 2020; Jiang et al. 2018; Wang
2020). Therefore, considering the functional complex- et al. 2020; Ho et al. 2020; Watanabe et al. 2015; Nakata
ity of PC, finding novel biomarkers to predict the prog- et al. 2009; Shi et al. 2021; Gerdes et al. 2002; Lyu et al.
nosis of PC patients at different stages and to provide 2018; Zhang et al. 2013; Ji et al. 2014). Furthermore, many
miRNAs also demonstrate prognostic roles in PCs (Khan
Li et al. Molecular Medicine (2022) 28:43 Page 13 of 16

et al. 2015; Eid et al. 2021). It is noteworthy that most confirmed that ICB treatment response can be predicted
existing prognostic models for PC involve only one gene with TIDE scores (Jiang et al. 2018; Wang et al. 2020),
or are only related to short-term clinical response (DFS) but, with this method, we found that there was no sig-
or long-term clinical effects (OS). Significantly, not all of nificant difference in TIDE scores of samples with high
the biomarkers have been used in the clinic or evaluated and low expression of LAMB3. In conclusion, our results
for their roles in predictions for immunotherapy. In the show that LAMB3 can be a predictor indicator for the
present study, we first presented a comprehensive profil- DFS and OS of PC patients.
ing of the tumor, tumor microenvironment, and tumor The adhesive extracellular matrix protein, fibronectin
immune status with 1,833 RNA targets of PC tissue, with its integrin receptors play important roles at sev-
using the GeoMx CTA Panel. We found five significantly eral stages of multiple tumors development (Erdogan
upregulated markers in the PC tissue, including LAMB3, et al. 2017; Glasner et al. 2018; Liang et al. 2020; Liu
FN1, KRT17, KRT19, and ANXA1. The KEGG pathway et al. 2020). In a pivotal study, Cristina P.R. Xavier et al.
enrichment analysis of the cancer tissue included laminin reported that FN1 is the most abundant cargo protein
interactions, degradation of the extracellular matrix, released by human macrophage extracellular vesicles
extracellular matrix organization, and more, all of which (EVs), and it is correlated with lower OS of PC patients
were highly associated with the upregulated genes. The (Xavier et al. 2021). Other studies have also demon-
expression pattern of LAMB3, FN1, KRT17, KRT19, strated that FN1 has an unfavorable prognostic impact
and ANXA1 was confirmed by the TCGA dataset. In the for PC patients (Akiyama et al. 1995; Munasinghe et al.
178 PC patients, the LAMB3, FN1, KRT17, KRT19, and 2020; Hiroshima et al. 2020). In addition, Hu et al. after
ANXA1 were also upregulated in cancer tissue, com- performing a immunohistochemical analysis on 138 PC
pared with adjacent tissues and normal tissues. Signifi- patients, reported that stromal FN1 expression was not
cantly, survival analysis demonstrated high expression of associated with long-term survival (Hu et al. 2019). In
LAMB3, FN1, KRT19, and ANXA1, but KRT17 was asso- our study, we used the GeoMx CTA Panel to perform
ciated with a shorter DFS and OS. The DFS benefits in a 1833 targets analysis that included the tumor, tumor
the low expression of LAMB3, FN1, KRT19, and ANXA1 microenvironment, and tumor immune status of PC
group indicate that these markers can be used to pre- patients. We identified FN1 precisely as the differentially
dict the treatment efficacy of surgery, and the treatment expressed targets in PCs. The increased expression of
approaches of patients with high expression of LAMB3, FN1 in PCs was confirmed by the TCGA dataset, which
FN1, KRT19, and ANXA1 may need to be improved. Fur- showed the same expression pattern. Furthermore, high
ther studies are necessary to examine this issue. expression of FN1 is highly associated with shorter DFS,
Previous studies have demonstrated that LAMB3 could OS, and resistance to immunotherapy compared with
mediate cell cycle arrest and apoptosis in PC cells and low expression of FN1 in PC patients. Co-expression
alter the proliferative, invasive, and metastatic behav- networks are significant for the development of can-
iors of PC by regulating the PI3K/Akt signaling path- cer (Zhou et al. 2019). Significant transcriptional differ-
way (Zhang et al. 2019; Huang et al. 2020). Inhibition ences were found in PC patients with high expression of
of LAMB3 abrogated the tumorigenic effects of PI3K/ FN1. The top upregulated genes included COL11A1, (Jia
Akt signaling pathway activation. Our study found that et al. 2016) MMP11 (Zhang et al. 2020), MARCO (Shi
­L AMB3hi PCs upregulated the PI3K/Akt signaling path- et al. 2021), GJB2 (Zhou et al. 2019), CD163 (Shi et al.
way, which is consistent with the findings of the previous 2021), and CCN4 (Banerjee et al. 2016) (Additional file 3:
study. However, the prognostic role of LAMB3 in dis- Table S3) in the high expression of FN1 group, which
tinct stages and treatment groups of PC patients is still were highly correlated with a poor prognosis of patients
unclear. In this study, we found that LAMB3 is signifi- with PC. The enriched KEGG pathway included ECM-
cantly upregulated and selectively expressed in PC tissue. receptor interaction and focal adhesion, which was corre-
High expression of LAMB3 was associated with shorter lated with the regulation of FN1, and FN1 increased cell
DFS and OS, suggesting its use as a predicting indicator. proliferation and enhanced chemoresistance in PC cells
Significantly, the univariate analysis and multivariate Cox (Miyamoto et al. 2004). Collectively, our results indicate
analysis revealed that LAMB3 can be a negative prog- that FN1 can be a biomarker for predicting short-term
nostic indicator for both DFS and OS. This data indicates clinical response, long-term clinical effects, and clinical
that LAMB3 can predict the treatment efficacy of surgery response to immunotherapy.
and the treatment efficacy of first-line therapy. Immuno- Keratins (KRTs) are intermediate filament proteins,
therapy has been become the main treatment option for responsible for the structural integrity of epithelial cells
PC; however, it is still unknown whether LAMB3 expres- and markers of epithelial tissue, activating signaling net-
sion can predict ICB response. Previous studies have works that regulate cell migration, invasion, metastasis,
Li et al. Molecular Medicine (2022) 28:43 Page 14 of 16

cell cycle, and apoptosis (Coulombe and Wong 2004; PC. In addition, the combination of LAMB3, KRT19, and
Hendrix et al. 1996; Saha et al. 2017). KRT19 is a known ANXA1 can be considered as a biomarker to predict OS
prognostic biomarker for multiple cancer types, includ- in PC.
ing PCs, and increased KRT19 was closely correlated
with a poor prognosis of patients with PC (Saha et al. Conclusions
2017; Tang et al. 2014; Yao et al. 2016). Experiments with However, this study has limitations. We only evaluated
mice have revealed that K ­ rt19+/Lgr5−cells are radiore- 1833 targets in one PC patient. More studies should be
sistant cancer-initiating stem cells in the colon and intes- conducted to confirm our results and early or late stage
tine (Asfaha et al. 2015), which can partially explain the pancreatic cancer patients’ samples should also be used
poorer prognosis of tumors with higher KRT19 expres- to evaluate the differences. In addition, the way that
sion. Consistent with this finding, our survival analysis, LAMB3, FN1, KRT19, and ANXA1 regulate the prog-
univariate analysis and multivariate Cox proportional nosis of PCs should be studied further. Nevertheless, the
hazards model showed that the K ­ RT19hi group of PC expression of LAMB3, FN1, KRT19, and ANXA1 are still
patients had a shorter DFS and OS. In contrast, when we effective predictors of PC prognosis. Furthermore, FN1
examined the sensitivity of PC patients with high and low and ANXA1 can be predictors of the efficacy of immuno-
expression of KRT19 to immunotherapy, we did not find therapy in PC. Last, independent studies are required to
significant differences. confirm the findings of this study.
ANXA1 is a calcium-dependent phospholipid-binding
protein considered to play an important role in tumo-
rigenesis in multiple types of cancer, including breast Abbreviations
AOI: Area of interest; CTA​: Cancer transcriptome atlas; DFS: Disease-free sur-
cancer, colorectal cancer, and more (Zhang et al. 2010; vival time; DSP: Digital spatial profiler; EVs: Extracellular vesicles; ICB: Immune
Graauw et al. 2010; Liang and Li 2021; Wang et al. 2010). checkpoint blockade; KEGG: Kyoto encyclopedia of genes and genomes; OS:
In addition, ANXA1 has been shown to have immu- Overall survival time; PC: Pancreatic cancer; QC: Quality control; ROIs: Regions
of interest; TIDE: Tumor immune dysfunction and exclusion.
nomodulatory effects on T-cells, macrophages, and den-
dritic cells (Dempsey et al. 2021). Significantly, targeting
ANXA1 with humanized antibodies inhibits tumorigenic Supplementary Information
The online version contains supplementary material available at https://​doi.​
processes and induces an immune response in tumors org/​10.​1186/​s10020-​022-​00467-8.
with an overexpression of ANXA1 (Dempsey et al. 2021).
However, there are no conclusions on the prognostic role
Additional file 1: Figure S1. (A) Technical signal QC reveals raw reads,
of ANXA1 in PC, especially in the prediction of sensi- aligned reads percentage, and sequencing saturation. (B) Technical
tivity to immunotherapy. In the present study, we used background QC indicates no template control count (NTC count) and
multiple analytic approaches, including survival analysis, negative probe count. (C) DSP parameters demonstrate the nuclei counts
and surface area. (D) Target QC was measured to determine the LOQ.
univariate analysis, multivariate Cox proportional haz- LOQ = GeoMean (NegProbe) × GeoSD (NegProbe)threshold. (E) Probe
ards model, and analysis of immune response, all of which outlier QC demonstrates the low outlier detection and Grubbs outlier test.
indicated the predictive roles of ANXA1 in PC patients. Additional file 2: Figure S2. (A) Volcano map indicating the differentially
Furthermore, compared with the ­ ANXA1low group of expressed genes in LAMB3hi and LAMB3low groups of PC patients. (B) The
hi upregulated KEGG pathways in the LAMB3hi group of PC patients. (C) The
PC patients, the ­ANXA1 group exhibited significantly downregulated KEGG pathways in the LAMB3hi group of PC patients. (D)
increased upregulated genes. In addition, the enrichment Volcano map indicating the differentially expressed genes in FN1hi and
of the KEGG pathway in ­ANXA1hi PC patients included FN1low groups of PC patients. (E) The upregulated KEGG pathways in the
FN1hi group of PC patients. (F) Volcano map indicating the differentially
the PI3K–Akt signaling pathway (Zhang et al. 2019), focal expressed genes in KRT19hi and KRT19low groups of PC patients. (G) The
adhesion (Sawai et al. 2005), and ECM—receptor inter- upregulated KEGG pathways in the KRT19hi group of PC patients. (H) The
action (Hosein et al. 2020), all of which are correlated downregulated KEGG pathways in the KRT19hi group of PC patients. (I)
Volcano map indicating the differentially expressed genes in ANXA1hi and
with the progression of cancers. Collectively, we identi- ANXA1low groups of PC patients. (J) The upregulated KEGG pathways in
fied a biomarker that predicts DFS, OS, and sensitivity to the ANXA1hi group of PC patients. (K) The downregulated KEGG pathways
immunotherapy in PC. In the future, continued screening in the ANXA1hi group of PC patients.
of ANXA1 in the PC immune microenvironment could Additional file 3. The upregulated and downregulated genes in the
­LAMB3hi group compared with the L­ AMB3low group.
help us to develop novel targets more accurately and pro-
vide a theoretical basis for the pathological mechanism of Additional file 4. The expression of the 1833 targets from the paracancer
tissue and the cancer tissue by DSP analysis.
PCs.
Additional file 5. The upregulated and downregulated genes in the F­ N1hi
Notably, the predictive roles of combined LAMB3, group compared with the ­FN1low group.
FN1, KRT19, and ANXA1 have never been reported. We Additional file 6. The upregulated and downregulated genes in the
found that the combination of LAMB3, FN1, KRT19, and ­KRT19hi group compared with the K­ RT19low group.
ANXA1 can be used as a biomarker to predict DFS in
Li et al. Molecular Medicine (2022) 28:43 Page 15 of 16

Cao W, Fan W, Wang F, et al. GM-CSF impairs erythropoiesis by disrupt-

Additional file 7. The upregulated and downregulated genes in the ing erythroblastic island formation via macrophages. J Transl Med.
­ANXA1hi group compared with the A ­ NXA1low group. 2022;20(1):11.
Chandramohan L, Au Q, Nagy M, Parnell E, Peters T. Abstract 2771: comprehen-
sive analysis of immuno oncology markers in the tumor microenviron-
Acknowledgements
ment of solid tumor samples using GeoMxTM digital spatial profiler (DSP)
We thank all our authors listed in this manuscript. We thank “Home for
and MultiOmyxTM hyperplexed immunofluorescence (IF). Cancer Res.
Researchers” for the analysis of TCGA data.
2021;81(13 Supplement):2771.
Chen X, Zeh HJ, Kang R, Kroemer G, Tang D. Cell death in pancreatic
Author contributions
cancer: from pathogenesis to therapy. Nat Rev Gastroenterol Hepatol.
WL designed experiments, performed experiments, analyzed the data and
2021;18(11):804–23.
drafted manuscript. CD, JS and WW designed and supervised the study and
Coulombe PA, Wong P. Cytoplasmic intermediate filaments revealed as
edited the manuscript. TL, CS, YD, and LC collected tissues and analyzed the
dynamic and multipurpose scaffolds. Nat Cell Biol. 2004;6(8):699–706.
data. All authors approved the final manuscript. All authors read and approved
de Graauw M, van Miltenburg MH, Schmidt MK, et al. Annexin A1 regulates
the final manuscript.
TGF-beta signaling and promotes metastasis formation of basal-like
breast cancer cells. Proc Natl Acad Sci U S A. 2010;107(14):6340–5.
Funding
Dempsey FC, Al-Ali H, Crichton SJ, Pepper C, Parris CN. Abstract 1874: MDX-
This work was supported by Natural Science Foundation of China (WW,
124, a novel annexin-A1 antibody, induces an anti-tumor immune
81900558), Provincial Ministry of Henan Province Youth Project (WW,
response and wide-ranging anti-cancer activity in multiple preclinical
SBGJ202003035), Postdoctoral Research Start-up Funding of The First Affili-
models. Cancer Res. 2021;81(13 Supplement):1874.
ated Hospital of Zhengzhou University (WL), Young Postdoctoral Innovators
Desai N, Neyaz A, Szabolcs A, et al. Temporal and spatial heterogeneity
in Henan Province (WL), and China Postdoctoral Science Foundation (WL,
of host response to SARS-CoV-2 pulmonary infection. Nat Commun.
2021M692930).
2020;11(1):6319.
Distler M, Pilarsky E, Kersting S, Grützmann R. Preoperative CEA and CA 19–9
Availability of data and materials
are prognostic markers for survival after curative resection for ductal
All data generated and materials in the study are included in the present
adenocarcinoma of the pancreas—a retrospective tumor marker
article and supplementary data.
prognostic study. Int J Surg. 2013;11(10):1067–72.
Eid M, Karousi P, Kunovský L, et al. The role of circulating MicroRNAs in
Declarations patients with early-stage pancreatic adenocarcinoma. Biomedicines.
2021;9(10):1468.
Ethics approval and consent to participate Erdogan B, Ao M, White LM, et al. Cancer-associated fibroblasts promote
Informed consent was given in accordance with the Declaration of Helsinki directional cancer cell migration by aligning fibronectin. J Cell Biol.
and under approval from The First Affiliated Hospital of Zhengzhou University. 2017;216(11):3799–816.
Gerdes B, Ramaswamy A, Ziegler A, et al. p16INK4a is a prognostic marker
Consent for publication in resected ductal pancreatic cancer: an analysis of p16INK4a, p53,
Informed consent has been obtained from the patient. MDM2, an Rb. Ann Surg. 2002;235(1):51–9.
Glasner A, Levi A, Enk J, et al. NKp46 receptor-mediated interferon-γ
Competing interests production by natural killer cells increases fibronectin 1 to alter tumor
The authors declare that they have no conflict of interest relating to the publi- architecture and control metastasis. Immunity. 2018;48(1):107-119.e4.
cation of this manuscript. Gong J, Farren MR, Shaib W, et al. Abstract 6499: small slide, big world: fast

Nanostring nCounter®GeoMxTM Digital Spatial Profiler. Cancer Res.

detection of altered proteins and RNA levels in clinical samples using
Author details
1
Department of Hematology, The First Affiliated Hospital of Zhengzhou 2020;80(16 Supplement):6499.
University, Zhengzhou 450052, Henan, China. 2 Department of Hepatobil- Hendrix MJ, Seftor EA, Chu YW, Trevor KT, Seftor RE. Role of intermediate
iary Surgery, The First Affiliated Hospital of Zhengzhou University, Zheng- filaments in migration, invasion and metastasis. Cancer Metastasis Rev.
zhou 450052, Henan, China. 3 Laboratory Animal Center, School of Medical 1996;15(4):507–25.
Sciences, Zhengzhou University, Zhengzhou 450052, Henan, China. 4 BGI Col- Hiroshima Y, Kasajima R, Kimura Y, et al. Novel targets identified by inte-
lege and Henan Institute of Medical and Pharmaceutical Sciences in Academy grated cancer-stromal interactome analysis of pancreatic adenocarci-
of Medical Science, Zhengzhou University, Zhengzhou 450052, Henan, China. noma. Cancer Lett. 2020;469:217–27.
5
The Academy of Medical Science, College of Medical, Zhengzhou University, Ho WJ, Jaffee EM, Zheng L. The tumour microenvironment in pancreatic
Zhengzhou 450052, Henan, China. 6 College of Public Health, Zhengzhou cancer—clinical challenges and opportunities. Nat Rev Clin Oncol.
University, Zhengzhou 450000, Henan, China. 2020;17(9):527–40.
Hosein AN, Brekken RA, Maitra A. Pancreatic cancer stroma: an update
Received: 13 January 2022 Accepted: 4 April 2022 on therapeutic targeting strategies. Nat Rev Gastroenterol Hepatol.
2020;17(8):487–505.
Hu D, Ansari D, Zhou Q, et al. Stromal fibronectin expression in patients
with resected pancreatic ductal adenocarcinoma. World J Surg Oncol.
2019;17(1):29.
References Huang W, Gu J, Tao T, et al. MiR-24-3p inhibits the progression of pancreatic
Akiyama SK, Olden K, Yamada KM. Fibronectin and integrins in invasion and ductal adenocarcinoma through LAMB3 downregulation. Front Oncol.
metastasis. Cancer Metastasis Rev. 1995;14(3):173–89. 2020;9:1499.
Asfaha S, Hayakawa Y, Muley A, et al. Krt19+/Lgr5− cells are radioresistant Ji Y-F, Huang H, Jiang F, Ni R-Z, Xiao M-B. S100 family signaling network
cancer-initiating stem cells in the colon and intestine. Cell Stem Cell. and related proteins in pancreatic cancer (Review). Int J Mol Med.
2015;16(6):627–38. 2014;33(4):769–76.
Ballehaninna UK, Chamberlain RS. Biomarkers for pancreatic cancer: Jia D, Liu Z, Deng N, et al. A COL11A1-correlated pan-cancer gene signature
promising new markers and options beyond CA 19–9. Tumor Biol. of activated fibroblasts for the prioritization of therapeutic targets.
2013;34(6):3279–92. Cancer Lett. 2016;382(2):203–14.
Banerjee SK, Maity G, Haque I, et al. Human pancreatic cancer progression: an Jiang P, Gu S, Pan D, et al. Signatures of T cell dysfunction and exclusion pre-
anarchy among CCN-siblings. J Cell Commun Signal. 2016;10(3):207–16. dict cancer immunotherapy response. Nat Med. 2018;24(10):1550–8.
Li et al. Molecular Medicine (2022) 28:43 Page 16 of 16

Kalita-de Croft P, Sadeghi Rad H, Gasper H, et al. Spatial profiling tech- Sun D, Jin H, Zhang J, Tan X. Integrated whole genome microarray analysis
nologies and applications for brain cancers. Expert Rev Mol Diagn. and immunohistochemical assay identifies COL11A1, GJB2 and CTRL as
2021;21(3):323–32. predictive biomarkers for pancreatic cancer. Cancer Cell Int. 2018;18:174.
Khan MA, Zubair H, Srivastava SK, Singh S, Singh AP. Insights into the role of Swords DS, Firpo MA, Scaife CL, Mulvihill SJ. Biomarkers in pancreatic adeno-
microRNAs in pancreatic cancer pathogenesis: potential for diagnosis, carcinoma: current perspectives. Onco Targets Ther. 2016;9:7459–67.
prognosis, and therapy. Adv Exp Med Biol. 2015;889:71–87. Tang J, Zhuo H, Zhang X, et al. A novel biomarker Linc00974 interacting with
Li C, Zeng X, Yu H, Gu Y, Zhang W. Identification of hub genes with diag- KRT19 promotes proliferation and metastasis in hepatocellular carci-
nostic values in pancreatic cancer by bioinformatics analyses and noma. Cell Death Dis. 2014;5(12):e1549.
supervised learning methods. World J Surg Oncol. 2018;16(1):223. Van TM, Blank CU. A user’s perspective on GeoMxTM digital spatial profiling.
Li W, Wang Y, Zhao H, et al. Identification and transcriptome analysis of Immuno-Oncology Technol. 2019;1:11–8.
erythroblastic island macrophages. Blood. 2019;134(5):480–91. Wang LP, Bi J, Yao C, et al. Annexin A1 expression and its prognostic signifi-
Liang Z, Li X. Identification of ANXA1 as a potential prognostic biomarker cance in human breast cancer. Neoplasma. 2010;57(3):253–9.
and correlating with immune infiltrates in colorectal cancer. Autoim- Wang Q, Li M, Yang M, et al. Analysis of immune-related signatures of
munity. 2021;54(2):76–87. lung adenocarcinoma identified two distinct subtypes: implica-
Liang H, Yu M, Yang R, et al. A PTAL-miR-101-FN1 axis promotes EMT and tions for immune checkpoint blockade therapy. Aging (albany NY).
invasion-metastasis in serous ovarian cancer. Mol Ther Oncolytics. 2020;12(4):3312–39.
2020;16:53–62. Wang Y, Li W, Schulz VP, et al. Impairment of human terminal eryth-
Lin W, Wu S, Chen X, et al. Characterization of hypoxia signature to evaluate roid differentiation by histone deacetylase 5 deficiency. Blood.
the tumor immune microenvironment and predict prognosis in glioma 2021;138(17):1615–27.
groups. Front Oncol. 2020;10:796. Watanabe T, Ueno H, Watabe Y, et al. ACTN4 copy number increase as a predic-
Liu B, Yang H, Taher L, et al. Identification of prognostic biomarkers by com- tive biomarker for chemoradiotherapy of locally advanced pancreatic
bined mRNA and miRNA expression microarray analysis in pancreatic cancer. Br J Cancer. 2015;112(4):704–13.
cancer. Transl Oncol. 2018a;11(3):700–14. Xavier CPR, Castro I, Caires HR, et al. Chitinase 3-like-1 and fibronectin in the
Liu B, Yang H, Pilarsky C, Weber GF. The effect of GPRC5a on the proliferation, cargo of extracellular vesicles shed by human macrophages influ-
migration ability, chemotherapy resistance, and phosphorylation of ence pancreatic cancer cellular response to gemcitabine. Cancer Lett.
GSK-3β in pancreatic cancer. Int J Mol Sci. 2018b;19(7):1870. 2021;501:210–23.
Liu X, Meng L, Li X, et al. Regulation of FN1 degradation by the p62/SQSTM1- Yao H, Yang Z, Liu Z, et al. Glypican-3 and KRT19 are markers associating with
dependent autophagy-lysosome pathway in HNSCC. Int J Oral Sci. metastasis and poor prognosis of pancreatic ductal adenocarcinoma.
2020;12(1):34. Cancer Biomark. 2016;17(4):397–404.
Lyu Y, Cheng Y, Wang B, Chen L, Zhao S. Sulfatase 1 expression in pancreatic Zhang Z, Huang L, Zhao W, Rigas B. Annexin 1 induced by anti-inflammatory
cancer and its correlation with clinicopathological features and postop- drugs binds to NF-kappaB and inhibits its activation: anticancer effects
erative prognosis. Cancer Biomark. 2018;22(4):701–7. in vitro and in vivo. Cancer Res. 2010;70(6):2379–88.
Miyamoto H, Murakami T, Tsuchida K, et al. Tumor-stroma interaction of Zhang G, He P, Tan H, et al. Integration of metabolomics and transcriptomics
human pancreatic cancer: acquired resistance to anticancer drugs and revealed a fatty acid network exerting growth inhibitory effects in human
proliferation regulation is dependent on extracellular matrix proteins. pancreatic cancer. Clin Cancer Res. 2013;19(18):4983–93.
Pancreas. 2004;28(1):38–44. Zhang H, Pan Y, Cheung M, et al. LAMB3 mediates apoptotic, proliferative,
Mizrahi JD, Surana R, Valle JW, Shroff RT. Pancreatic cancer. Lancet. invasive, and metastatic behaviors in pancreatic cancer by regulating the
2020;395(10242):2008–20. PI3K/Akt signaling pathway. Cell Death Dis. 2019;10(3):230.
Munasinghe A, Malik K, Mohamedi F, et al. Fibronectin acts as a molecular Zhang X, Lu J, Zhou L, et al. Matrix metalloproteinase 11 as a novel tumor
switch to determine SPARC function in pancreatic cancer. Cancer Lett. promoter and diagnostic and prognostic biomarker for pancreatic ductal
2020;477:88–96. adenocarcinoma. Pancreas. 2020;49(6):812–21.
Nakata K, Ohuchida K, Nagai E, et al. LMO2 is a novel predictive marker for a Zhou G, Liu X, Wang X, et al. Combination of preoperative CEA and CA19-9
better prognosis in pancreatic cancer. Neoplasia. 2009;11(7):712–9. improves prediction outcomes in patients with resectable pancreatic
Nixon AB, Pang H, Starr MD, et al. Prognostic and predictive blood-based adenocarcinoma: results from a large follow-up cohort. Onco Targets
biomarkers in patients with advanced pancreatic cancer: results from Ther. 2017;10:1199–206.
CALGB80303 (Alliance). Clin Cancer Res. 2013;19(24):6957–66. Zhou Y-Y, Chen L-P, Zhang Y, et al. Integrated transcriptomic analysis reveals
Pereira SP, Oldfield L, Ney A, et al. Early detection of pancreatic cancer. Lancet hub genes involved in diagnosis and prognosis of pancreatic cancer. Mol
Gastroenterol Hepatol. 2020;5(7):698–710. Med. 2019;25(1):47.
Saha SK, Choi HY, Kim BW, et al. KRT19 directly interacts with β-catenin/RAC1 Zhu L, Xue H-D, Liu W, et al. Enhancing pancreatic mass with normal serum
complex to regulate NUMB-dependent NOTCH signaling pathway and CA19-9: key MDCT features to characterize pancreatic neuroendocrine
breast cancer properties. Oncogene. 2017;36(3):332–49. tumours from its mimics. Radiol Med. 2017;122(5):337–44.
Sahni S, Nahm C, Krisp C, et al. Identification of novel biomarkers in pancreatic
tumor tissue to predict response to neoadjuvant chemotherapy. Front
Oncol. 2020;10:237. Publisher’s Note
Sawai H, Okada Y, Funahashi H, et al. Activation of focal adhesion kinase Springer Nature remains neutral with regard to jurisdictional claims in pub-
enhances the adhesion and invasion of pancreatic cancer cells via extra- lished maps and institutional affiliations.
cellular signal-regulated kinase-1/2 signaling pathway activation. Mol
Cancer. 2005;4(1):37.
Shi B, Chu J, Huang T, et al. The scavenger receptor MARCO expressed by
tumor-associated macrophages are highly associated with poor pancre-
atic cancer prognosis. Front Oncol. 2021;11:771488.
Sorg K, Gong J, Zimmerman S, et al. Performance of GeoMx CTA and
WTA, high-plex, spatial gene expression profiling tools. J Mol Diagn.
2020;22(11):S95–S95.
Stojkovic Lalosevic M, Stankovic S, Stojkovic M, et al. Can preoperative CEA and
CA19-9 serum concentrations suggest metastatic disease in colorectal
cancer patients? Hell J Nucl Med. 2017;20(1):41–5.
Strobel O, Neoptolemos J, Jäger D, Büchler MW. Optimizing the outcomes of
pancreatic cancer surgery. Nat Rev Clin Oncol. 2019;16(1):11–26.