Skip to main content
NCBI home page
The Journal of Molecular Diagnostics : JMD logoLink to The Journal of Molecular Diagnostics : JMD
. 2024 Nov;26(11):962–970. doi: 10.1016/j.jmoldx.2024.07.002

Purity Independent Subtyping of Tumors (PurIST) Pancreatic Cancer Classifier

Analytic Validation of a 16-RNA Expression Signature Distinguishing Basal and Classical Subtypes

Yan Li ∗,†, Jason D Merker ∗,†, Rachana Kshatriya ∗, Dimitri G Trembath ∗, Ashley B Morrison †, Peyton C Kuhlers †,‡, Naim U Rashid †,§, Jen Jen Yeh †,‡, Margaret L Gulley ∗,†,∗
PMCID: PMC11524322  PMID: 39181325

Abstract

The two major molecular subtypes of pancreatic adenocarcinoma reportedly have differential response to FOLFIRINOX-based therapy. To promote rapid assignment of basal versus classical subtypes, an array-based single-sample classifier assay was developed and applied to 74 formalin-fixed, paraffin-embedded biopsy or resection specimens of known subtype based on transcriptomics. The Purity Independent Subtyping of Tumors (PurIST) algorithm assigns subtype based on relative expression of 16 RNAs counted by RNA sequencing (RNAseq) versus more practical array-based NanoString nCounter Elements XT technology. Subtype calls were largely concordant between RNAseq and array methods (72/74, 97% agreement). Compared with the lengthy RNAseq protocol, the array-based assay takes just 3 working days to analyze, permitting rapid reporting of tumor subtype. In conclusion, the PurIST pancreatic cancer classifier has robust performance to classify pancreatic adenocarcinoma into basal versus classical subtypes. Clinical validation studies are underway to evaluate outcome in patients whose standard-of-care chemotherapy regimen is selected on the basis of rapid subtype assignment (NCT04683315).

Pancreatic ductal adenocarcinoma (PDAC) is an aggressive cancer that is the third most common cause of cancer death in the United States. Molecular profiling studies suggest that PDAC has two major molecular subtypes, basal and classical.1, 2, 3 The basal subtype is so named because the expression profile is similar to basal epithelial cells and to basal subtypes of breast and bladder adenocarcinomas. Data from clinical trials suggest that basal tumors, compared with classical tumors, have inferior survival and are resistant to standard first-line FOLFIRINOX-based treatment and possibly are more responsive to gemcitabine-based therapy.4, 5, 6, 7 This implies that improved outcome might be achieved if a practical laboratory assay was available to assist in selecting first-line chemotherapy as informed by knowledge of classical versus basal subtype.

In prior work, our group used full-transcriptome sequencing [RNA sequencing (RNAseq)] of malignant and stroma fractions to identify transcripts characteristic of carcinoma or that distinguish classical and basal subtypes, and then machine learning strategies were used to select eight pairs of mRNAs for which relative expression levels are predictive of subtype.8,9 As previously described, expression levels of the 16 RNAs are input to an algorithm that weights each pair's contribution to generate a subtype call.8 The 16 informative RNAs were selected from a larger candidate list of tumor-specific RNAs.1 This specificity led to a hypothesis, supported by empirical data,8 that the algorithm is informative of subtype even when malignant cell percentage is low. For this reason, this assay to predict classical versus basal subtype was named the Purity Independent Subtyping of Tumors (PurIST) pancreatic cancer classifier.

RNAseq is a powerful technology for characterizing gene expression profiles and tumor subtypes; however, it is cumbersome to use in clinical settings because it is labor intensive, with a long turnaround-time, whereas prompt treatment decisions are needed in patients with pancreatic cancer. In prior preclinical work, specimens from 51 patients had paired RNAseq and array-based expression profiling (NanoString, Seattle, WA), and only a single specimen demonstrated discordance between pancreatic cancer subtypes.8 In the current study performed in a clinical laboratory, a rapid array-based expression assay was devised to quickly assign classical or basal subtype to pancreatic cancer tissues. Sixteen informative RNAs were quantified by each of two methods, gold standard RNAseq and a faster NanoString array method, and then a PurIST algorithm was used to compare subtype assignments in 74 pancreatic cancer tissues. The findings define a precision medicine–based approach to tumor subtyping that is practical for informing selection of therapy in clinical trials.

Materials and Methods

Tissue Selection and RNA Preparation

After institutional review board approval of this study, formalin-fixed, paraffin-embedded (FFPE) tumor specimens (n = 74) were obtained from pathology archives at University of North Carolina Health. A pathologist reviewed the diagnosis of pancreatic ductal adenocarcinoma and estimated the percentage of malignant cells by microscopy of a hematoxylin and eosin–stained section. Eight serial adjacent unstained sections, 10 μm thick on coated slides, were input to RNA extraction using the FFPET RNA Isolation Kit (NanoString) on deparaffinized sections. For 19 of 74 tumors, a second sampling of the same tumor block was required to obtain sufficient RNA for both RNAseq and array protocols, whereas the remaining 55 tumors had the same RNA extract used for both platforms. Ultraviolet light absorbance at OD 260 and 280 was quantified by spectrophotometry (Nanodrop, Thermo Fisher, Waltham, MA), and the degree of nucleic acid fragmentation was assessed by a Tape Station 4200 using the manufacturer's protocol to calculate the fraction of RNA that was >200 nucleotides in length (DV200; Agilent, Santa Clara, CA).

Full-Transcriptome RNAseq

Total RNA was sequenced on a NextSeq500 instrument (Illumina, San Diego, CA), and reads were aligned to the human reference genome (Genome Reference Consortium Human Build 38). Libraries prepared using either the TruSeq RNA Exome Kit (Illumina) or KAPA RNA HyperPrep Kit with RiboErase, with 17 cycles for the final PCR step (Roche, Indianapolis, IN), were pooled and diluted to 1.65 pmol/L and were sequenced using the NextSeq 500/550 High Output Kit version 2.5 (150 cycles; Illumina) to generate approximately 60 million reads per sample. Base call files were converted to FASTQ files using bcl2fastq2 Conversion software 2.20.0 (Illumina), and data were collapsed into one lane for alignment and to quantify transcripts. Read counts were input to the PurlST algorithm that generates a subtype call and calculates a PurIST score based on relative abundance of each of eight pairs of RNAs. Prior work established a threshold PurIST score of >0.5 to call basal subtype and ≤0.5 to call classical subtype.8 Figure 1 displays laboratory workflow and lists the 16 RNAs.

Figure 1.

Figure 1

Open in a new tab

Assay workflow to assign basal versus classical subtype. RNA from each tumor was tested in parallel by NanoString array and gold standard RNA-sequencing (RNAseq) protocols. Numeric counts for each of 16 RNAs were input to a Purity Independent Subtyping of Tumors (PurIST) algorithm to calculate a PurIST score varying from 0 (least basal) to 1 (strongly basal) based on relative expression of eight RNA pairs. The software issues an objective subtype call, with a score >0.5 defining basal subtype and ≤0.5 defining classical subtype. Note that the terms PurIST, PurIST algorithm, PurIST score, basal, and classical are as defined in this testing laboratory, and these terms may differ in other settings. FFPE, formalin fixed, paraffin embedded. RefSeq, Reference Sequence.

NanoString Array

Total RNA input varied from 100 to 300 ng, calculated to adjust for RNA quality using the formula input = 100 ng/DV200. NanoString nCounter Elements XT technology was applied using a custom probe set to capture the same 16 target RNAs plus 15 additional housekeeper RNAs and spiked system controls using an nCounter Dx Analysis System in FLEX mode, according to manufacturer instructions (NanoString). Raw counts from nAnalyzer software 5s (NanoString) were input to a PurlST algorithm to calculate a PurIST score. A score of >0.5 was assigned as basal subtype and ≤0.5 was assigned as classical subtype.8

The NanoString system includes a series of internal controls to check analytic performance in each run of up to 12 specimens, including negative controls to assess background counts for six probes that should have no targets, and positive controls for sensitivity and linearity of the test system based on probes targeting RNAs that are spiked into each RNA specimen at known amount. In this study, internal positive and negative control RNAs and system controls were within limits recommended by the manufacturer (NanoString). These system controls test sensitivity, linearity, and evenness of signal across the solid surface of each array, and serve as quality indicators to ensure that hybridization, capture, and counting operate as expected.

Three external quality controls confirmed each run performed as expected for basal versus classical subtype call as assigned by gold standard RNAseq of aliquots stored at –80°C for one-time use in each run. Commercial cell line control is Universal Human Reference RNA (Invitrogen, Thermo Fisher Scientific) from 10 human cell lines, with an expected PurIST score near to the cutoff of 0.5 and a classical subtype call. Classical subtype control and Basal subtype control are each patient-derived xenografts fixed and embedded in paraffin, with an expected subtype call of classical or basal, respectively.

Data Analysis and Statistical Analysis

Agreement in subtype call (a discrete variable) was examined between the NanoString versus gold standard RNAseq. PurIST scores (numeric values) of <0.1 or >0.9 were so far from the 0.5 cutoff that they were considered strong evidence of classical or basal subtype, respectively.

Results

The 74 pancreatic ductal carcinoma tumor specimens included 42 FFPE large biopsy or resection tissues and 32 FFPE core biopsies collected by an ultrasound-guided fine needle. Extracted RNA from these 74 specimens and from three exogenous controls was tested in parallel by the NanoString array and by the gold standard comparator, which was full-transcriptome RNAseq. For each platform, the PurIST algorithm was applied to raw counts of eight pairs of RNA to yield a PurIST score and a call of basal versus classical subtype.

NanoString array was performed in duplicate for 26 of the 74 tumor RNAs. Of these 26 duplicate analyses performed on separate days, 18 were large biopsy or resection specimens (Table 1), and 7 were core biopsies (Table 2). Results replicated well, as evidenced by the same subtype call in all cases and consistent PurIST scores (R2 = 0.92).

Table 1.

Molecular Subtype of 42 Large Pancreatic Cancer Tissues Analyzed by NanoString Array and by Gold Standard RNAseq, 18 of Which Were Tested in Duplicate

Case no. Malignant cell % NanoString subtype RNAseq subtype NanoString PurIST score RNAseq PurIST score NanoString RNA input, ng DV200 RNA quality metric
1 40 Classical Classical 0.001 0.001 159 63
1 Repeat Classical Classical 0.001
2 30 Classical Classical 0.001 0.001 181 55
2 Repeat Classical Classical 0.001
3 20 Classical Classical 0.137 0.033 212 47
3 Repeat Classical Classical 0.137
4 20 Classical Classical 0.003 0.001 169 59
4 Repeat Classical Classical 0.003
5 10 Classical Classical 0.032 0.011 169 59
5 Repeat Classical Classical 0.032
6 20 Classical Classical 0.003 0.003 184 54
6 Repeat Classical Classical 0.003
7 30 Classical Classical 0.001 0.001 172 58
7 Repeat Classical Classical 0.001
8 10 Classical Classical 0.001 0.003 167 60
8 Repeat Classical Classical 0.001
9 5 Classical Classical 0.005 0.003 181 55
9 Repeat Classical Classical 0.005
10 15 Classical Classical 0.003 0.001 161 62
10 Repeat Classical Classical 0.003
11 40 Classical Classical 0.137 0.004 164 61
11 Repeat Classical Classical 0.137
12 40 Classical Classical 0.055 0.049 153 66
12 Repeat Classical Classical 0.055
13 20 Classical Classical 0.001 0.001 133 75
13 Repeat Classical Classical 0.001
14 2.5 Classical Classical 0.200 0.080 137 73
14 Repeat Classical Classical 0.200
15 5 Classical Classical 0.002 0.001 176 57
15 Repeat Classical Classical 0.001
16 10 Classical Classical 0.003 0.003 300 16
16 Repeat Classical Classical 0.003
17 2.5 Classical Classical 0.003 0.003 140 71
17 Repeat Classical Classical 0.003
18 5 Classical Classical 0.008 0.003 237 42
18 Repeat Classical Classical 0.008
19 50 Classical Classical 0.005 0.002 238 42
20 2.5 Classical Classical 0.087 0.420 300 27
21 20 Classical Classical 0.001 0.001 167 60
22 30 Classical Classical 0.001 0.001 137 73
23 25 Classical Classical 0.001 0.008 142 70
24 25 Classical Classical 0.001 0.001 158 63
25 40 Classical Classical 0.001 0.003 139 72
26 50 Classical Classical 0.008 0.003 139 72
27 30 Classical Classical 0.001 0.001 156 64
28 30 Classical Classical 0.023 0.036 128 78
29 20 Classical Classical 0.005 0.004 170 59
30 20 Classical Classical 0.198 0.020 158 63
31 30 Classical Basal 0.194 0.902 183 55
31 40 Classical Classical 0.008 0.039 300 27
33 20 Classical Classical 0.003 0.003 235 43
34 30 Classical Classical 0.001 0.001 239 42
35 20 Classical Classical 0.001 0.003 231 43
36 30 Classical Classical 0.007 0.281 244 41
37 30 Classical Classical 0.008 0.003 259 39
38 40 Classical Classical 0.001 0.001 249 40
39 20 Classical Classical 0.020 0.051 228 44
40 40 Basal Basal 0.548 0.548 137 73
41 20 Classical Classical 0.001 0.001 136 74
42 20 Classical Classical 0.001 0.001 144 69
Open in a new tab

Discrepant subtype call or weak PurIST score is shown in bold.

DV200, fraction of RNA that was >200 nucleotides in length; PurIST, Purity Independent Subtyping of Tumors; RNAseq, RNA sequencing.

Table 2.

Molecular Subtype of 32 Core Biopsy Pancreatic Cancer Tissues Analyzed by NanoString Array and by Gold Standard RNAseq, Eight of Which Were Tested in Duplicate

Case no. NanoString subtype RNAseq subtype NanoString PurIST score RNAseq PurIST score NanoString RNA input, ng DV200 RNA quality metric
43 Classical Classical 0.003 0.013 181 55
44 Basal Basal 0.986 0.986 136 74
45 Classical Classical 0.003 0.001 142 70
46 Classical Classical 0.013 0.038 287 35
47 Classical Classical 0.003 0.002 144 69
48 Classical Classical 0.032 0.002 119 84
49 Classical Classical 0.008 0.001 293 34
50 Classical Classical 0.001 0.001 162 62
51 Classical Classical 0.013 0.013 115 87
52 Classical Classical 0.003 0.002 146 68
53 Classical Classical 0.003 0.001 143 70
54 Classical Classical 0.129 0.034 138 72
55 Classical Classical 0.003 0.001 133 75
56 Classical Classical 0.003 0.033 119 84
57 Basal Basal 0.965 0.692 146 68
58 Classical Classical 0.056 0.003 136 74
59 Classical Classical 0.002 0.001 151 66
60 Classical Classical 0.003 0.021 136 74
61 Classical Classical 0.013 0.004 262 38
62 Classical Classical 0.007 0.003 164 61
63 Classical Classical 0.001 0.005 138 73
64 Classical Classical 0.087 0.087 151 66
65 Classical Classical 0.008 0.092 137 73
66 Classical Classical 0.020 0.008 160 62
67 Basal Classical 0.557 0.093 163 61
67 Repeat Basal Classical 0.784
68 Basal Basal 0.957 0.902 147 68
68 Repeat Basal Basal 0.647
69 Basal Basal 0.937 0.904 148 67
69 Repeat Basal Basal 0.937
70 Classical Classical 0.003 0.002 126 79
70 Repeat Classical Classical 0.003
71 Classical Classical 0.008 0.039 119 84
71 Repeat Classical Classical 0.008
72 Classical Classical 0.003 0.011 146 69
72 Repeat Classical Classical 0.003
73 Classical Classical 0.008 0.001 117 86
73 Repeat Classical Classical 0.003
74 Classical Classical 0.003 0.004 117 86
74 Repeat Classical Classical 0.004
Open in a new tab

Discrepant subtype call or weak PurIST score is shown in bold.

DV200, fraction of RNA that was >200 nucleotides in length; PurIST, Purity Independent Subtyping of Tumors; RNAseq, RNA sequencing.

When subtype call was compared between the NanoString array and gold standard RNAseq, subtype call was correct by array technology in 72 of 74 tumors (97% agreement) (Figure 2).

Figure 2.

Figure 2

Open in a new tab

Cross-platform comparison of Purity Independent Subtyping of Tumors (PurIST) scores. When 74 tumors were tested in parallel by NanoString array and gold standard RNA-sequencing (RNAseq) protocols, two tumors had discordant subtype calls, whereas 72 tumors (97%) had the same subtype call. Red lines mark the 0.5 cutoffs for distinguishing basal from classical subtypes. Blue dots represent 42 large tumor specimens, whereas orange dots represent 32 core biopsies. Case 31 was classical by NanoString array and basal by RNAseq. Case 67 was basal by both replicates of NanoString array and classical by RNAseq.

One outlier (Case 31) had a score of 0.194 with a classical call by NanoString array versus a score of 0.902 with a basal call by RNAseq; there was insufficient specimen for repeat testing. The other discrepant subtype call was in Case 67, a small biopsy that had a PurIST score of 0.557, a value which is just above the threshold of 0.5, leading to basal subtype assignment by array. In contrast, the PurIST score by RNAseq was below the threshold at 0.093 with a classical subtype call. Repeat analysis of this specimen by NanoString array again generated a basal call, with a PurIST score of 0.784. In this specimen, raw data analysis revealed that one of the eight RNA pairs (S100A2/SLC40A1) had read counts that were equivalent, suggesting that relative measurements of these two transcripts could randomly favor basal versus classical, potentially accounting for discrepant scores. Additional quality metrics were scrutinized in both tumors having discrepant subtype calls, and metrics were not substantially different from the 72 concordant tumors with respect to percentage of malignant cells, input RNA amount, or DV200.

Among the 72 subtype concordant cases, cross-platform PurIST scores were either strong basal or strong classical in 62 cases, whereas the other 10 tumors had at least one weaker score closer to the cutoff (plotted in Figure 2 and boldfaced in Tables 1 and 2), of which three were less basal by NanoString array, one was equally basal by both platforms, and six were more basal by NanoString array. These findings do not support a systematic scoring bias between the two platforms.

In summary, a discordant subtype call was made in just 2 of 74 tumors (3%), suggesting that NanoString and RNAseq technologies are generally concordant when distinguishing the two major subtypes of pancreatic cancer.

Quality Indicators for the NanoString Array

PurIST scores and subtype calls for three exogenous controls are shown in Table 3. These control RNAs were derived from FFPE xenograft tissues of a basal cancer versus a classical cancer propagated in mice, or a commercial cell line mixture. Subtype call reproducibility was 100%. Note that the commercial cell line control had a PurIST score near the threshold of 0.5, yet PurIST scores did not vary in replicate assays. Later, this commercial cell line control exhibited score deviations that might be explained by equivalent counts in two of the eight RNA pairs (C16ORF74/DDC and ANXA10/KRT6A), suggesting that relative levels of these transcripts could randomly favor basal versus classical and account for variable PurIST scores observed during production mode.

Table 3.

Replicate Test Results on Three Controls by NanoString Array

Basal subtype control
 Classical subtype control
 Commercial RNA control
Run no. Subtype call PurIST score Run no. Subtype call PurIST score Run no. Subtype call PurIST score
1 Basal 0.991 1 Classical 0.001 1 Classical 0.427
2 Basal 0.991 2 Classical 0.001 1 Classical 0.427
3 Basal 0.991 3 Classical 0.001 7 Classical 0.427
4 Basal 0.991 4 Classical 0.001 7 Classical 0.427
5 Basal 0.991 5 Classical 0.001 8 Classical 0.427
6 Basal 0.991 6 Classical 0.001 9 Classical 0.427
7 Basal 0.937 7 Classical 0.001 10 Classical 0.427
8 Basal 0.991 8 Classical 0.001 11 Classical 0.427
9 Basal 0.991 9 Classical 0.001 12 Classical 0.427
10 Basal 0.991 10 Classical 0.001 13 Classical 0.427
11 Basal 0.991 11 Classical 0.001 14 Classical 0.427
12 Basal 0.991 12 Classical 0.001 14 Classical 0.427
13 Basal 0.991 13 Classical 0.001 14 Classical 0.427
14 Basal 0.991 14 Classical 0.001
15 Basal 0.991 15 Classical 0.001
Open in a new tab

One outlier value is shown in bold.

PurIST, Purity Independent Subtyping of Tumors.

The findings informed an evidence-based standard operating procedure incorporating multiple strategies for quality assurance of the PurIST array-based classifier, including the following: i) malignancy confirmed in FFPE tissue selected for testing, ii) 100 to 300 ng of RNA input, and iii) minimum DV200 of 30 consistent with intact RNA having minimal degradation. In addition, documented review of NanoString system flags and correct results for exogenous run controls were required for each batch of assays.

Turnaround Time

In production mode, the NanoString array test system yields a subtype call in 3 working days from the time that tissue sections are received in the testing laboratory after tumor histopathology is vetted by a pathologist. Deparaffinization is done on day 1, followed by RNA extraction, hybridization, and capture on day 2, and counting and algorithm calculations with subtype call and reporting on day 3. Although the bench work for RNAseq was not systematically evaluated in this study, it is clearly more labor intensive and has more specimen transfers, each of which introduce risk of specimen switches. Experience suggests that NanoString array is a more rapid and practical strategy for clinical tumor subtype assignment.

Discussion

This study confirmed that a full-transcriptome gene expression test system could be simplified to devise a rapid and informative laboratory assay to classify cancer tissue into major molecular subtypes. There was good concordance (97%) between the rapid NanoString array and the lengthier and labor-intensive RNAseq-based subtype calls.

The PurIST algorithm is weighted so that the most informative markers of classical versus basal subtype proportionally impact the PurIST score.8 Importantly, the algorithm does not require data normalization to assess which transcript in each of eight informative pairs of RNAs is more highly expressed, and this feature should facilitate use of the algorithm across various platforms capable of quantifying target analytes.

Among the 74 tumors evaluated, two had discrepant subtype calls across the two platforms. One of these discrepancies (Case 31) might be a technical error given that scores were dramatically different across the two methods. The other discrepant case had a score near to the threshold of 0.5, and it was subject to random error related to equivalent counts in one of the eight RNA pairs, which could explain, at least in part, the observed variance.

A range of biologic or technical issues might result in discrepant calls. Biologically, intertumoral and intratumoral heterogeneity may impact PurIST score.10,11 Prior work12 and recent single-cell sequencing11 showed that different cells within the same tumor mass can express basal or classical genes. Unusual types of cancer can have outlier gene expression that might reflect alternate mechanisms of carcinogenesis.10,13 On the technical side, the two protocols had different capture probes that could target different exons or splice variants. Assay design can help overcome this concern by evaluating multiple segments of the same coding gene or evaluating many more informative target RNAs. Measurement seems more accurate when natural RNA is probed directly without the need for amplification, theoretically favoring NanoString over RNAseq quantification but practically yielding good cross-platform correlation.14 Assay design that accommodates poor-quality nucleic acid is favored when nucleic acid preservation is difficult to control, as is typical of archival FFPE tissue having variable degrees of cold ischemia, fixation type and duration, oxidation, RNA adducts and cross-linking, contaminants, or fragmentation. RNA stability concerns are overcome only partly by designing a rank order algorithm to help control for RNA quantity and quality. Sampling error can result in different proportions of cell types as sections are cut deeper into a given tissue block. Complex multistep assays, like RNAseq, seem more prone to human error than simpler, automated, or user-friendly assays requiring less expertise. Pancreatic cancer is notoriously rich in densely desmoplastic stroma, and it is feasible that some target analytes derive less from malignant cells and more from stromal elements. In short, there are many factors impacting analytic results, and definitive explanations were not feasible for the few discrepancies noted in this study.

Importantly, this study showed that small core needle biopsy specimens were as robust as larger biopsy and resection specimens for completing RNA-based subtype assignment of FFPE cancer tissue. This is clinically relevant because core biopsies are commonly used for pancreatic cancer diagnosis.

Multiple prior published studies have applied a PurIST strategy to classify pancreatic cancers using RNAseq. An advantage of RNAseq over NanoString array technology is the ability to evaluate many more analytes and biochemical pathways that could reveal underlying pathogenesis and refine tumor classification. Emerging data support validity of a PurIST-based single-sample classifier for prognosis, at least upfront at diagnosis of PDAC. Using RNAseq, Zhou et al15 revealed differential prognosis of classical versus basal subtypes before neoadjuvant treatment, but not in residual tumor remaining after treatment. Tempus laboratory offers a clinical-grade PurIST assay as an optional component of whole-transcriptome sequencing of pancreatic cancer tissue. Clinical validation data showed basal subtype was associated with poor response to FOLFIRINOX as well as worse survival in the cohort of patients with PDAC receiving FOLFIRINOX. Multivariate analysis of clinicopathologic prognostic features revealed PurIST subtype was independently associated with survival.16 Classical subtype was marginally associated with longer survival among the cohort of FOLFIRINOX-treated patients than among a gemcitabine + nanoparticle albumin-bound paclitaxel–treated cohort.17 The clinical-grade PurIST pancreatic cancer classifier by NanoString array technology has a shorter turnaround time and might be preferable to RNAseq, particularly if classical versus basal molecular subtype is the only genomic feature needed for clinical decision-making.

In conclusion, an analytically valid, rapid, robust assay was developed to classify pancreatic adenocarcinoma tissue into basal versus classical subtypes. This array-based single-sample classifier generated subtype calls that were 97% concordant with RNAseq, yet the array-based method has a less complex bench protocol with a 3-day turnaround time.

Ongoing work aims to gather evidence of clinical validity. The PurIST pancreatic cancer classifier assay serves as an integral biomarker in a prospective clinical trial of subtype-driven chemotherapy selection (https://www.clinicaltrials.gov/study/NCT04683315, last accessed July 26, 2024). The rationale for this and for several other ongoing clinical trials is that PDAC molecular characteristics are associated with differential response to therapy.4,6,13,15, 16, 17, 18 If shown to improve outcome in patients with pancreatic cancer, the PurIST pancreatic cancer classifier assay is a practical solution for predicting chemotherapy response in routine laboratory medicine practice.

Disclosure Statement

N.U.R. and J.J.Y. are patent holders on Purity Independent Subtyping of Tumors (PurIST) technology for subtyping pancreatic ductal adenocarcinoma (US17/601,002). GeneCentric Therapeutics, Inc., had no involvement in the current study but acquired an exclusive license from the University of North Carolina at Chapel Hill related to PurIST technology.

Footnotes

Supported by the NIH National Cancer InstituteR01 CA199064; the University of North Carolina Department of Pathology and Laboratory Medicine; and the Lineberger Comprehensive Cancer Center.

References

  • 1.Moffitt R.A., Marayati R., Flate E.L., Volmar K.E., Loeza S.G., Hoadley K.A., Rashid N.U., Williams L.A., Eaton S.C., Chung A.H., Smyla J.K., Anderson J.M., Kim H.J., Bentrem D.J., Talamonti M.S., Iacobuzio-Donahue C.A., Hollingsworth M.A., Yeh J.J. Virtual microdissection identifies distinct tumor- and stroma-specific subtypes of pancreatic ductal adenocarcinoma. Nat Genet. 2015;47:1168–1178. doi: 10.1038/ng.3398. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Cancer Genome Atlas Research Network Integrated genomic characterization of pancreatic ductal adenocarcinoma. Cancer Cell. 2017;32:185–203.e13. doi: 10.1016/j.ccell.2017.07.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Aguirre A.J., Nowak J.A., Camarda N.D., Moffitt R.A., Ghazani A.A., Hazar-Rethinam M., et al. Real-time genomic characterization of advanced pancreatic cancer to enable precision medicine. Cancer Discov. 2018;8:1096–1111. doi: 10.1158/2159-8290.CD-18-0275. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Xu Z., Hu K., Bailey P., Springfeld C., Roth S., Kurilov R., Brors B., Gress T., Buchholz M., An J., Wei K., Peccerella T., Büchler M.W., Hackert T., Neoptolemos J.P. Clinical impact of molecular subtyping of pancreatic cancer. Front Cell Dev Biol. 2021;9 doi: 10.3389/fcell.2021.743908. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Aung K.L., Fischer S.E., Denroche R.E., Jang G.H., Dodd A., Creighton S., Southwood B., Liang S.B., Chadwick D., Zhang A., O'Kane G.M., Albaba H., Moura S., Grant R.C., Miller J.K., Mbabaali F., Pasternack D., Lungu I.M., Bartlett J.M.S., Ghai S., Lemire M., Holter S., Connor A.A., Moffitt R.A., Yeh J.J., Timms L., Krzyzanowski P.M., Dhani N., Hedley D., Notta F., Wilson J.M., Moore M.J., Gallinger S., Knox J.J. Genomics-driven precision medicine for advanced pancreatic cancer: early results from the COMPASS trial. Clin Cancer Res. 2018;24:1344–1354. doi: 10.1158/1078-0432.CCR-17-2994. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.O'Kane G.M., Grünwald B.T., Jang G.H., Masoomian M., Picardo S., Grant R.C., Denroche R.E., Zhang A., Wang Y., Lam B., Krzyzanowski P.M., Lungu I.M., Bartlett J.M.S., Peralta M., Vyas F., Khokha R., Biagi J., Chadwick D., Ramotar S., Hutchinson S., Dodd A., Wilson J.M., Notta F., Zogopoulos G., Gallinger S., Knox J.J., Fischer S.E. GATA6 expression distinguishes classical and basal-like subtypes in advanced pancreatic cancer. Clin Cancer Res. 2020;26:4901–4910. doi: 10.1158/1078-0432.CCR-19-3724. [DOI] [PubMed] [Google Scholar]
  • 7.Seppälä T.T., Zimmerman J.W., Sereni E., Plenker D., Suri R., Rozich N., Blair A., Thomas D.L., 2nd, Teinor J., Javed A., Patel H., Cameron J.L., Burns W.R., He J., Tuveson D.A., Jaffee E.M., Eshleman J., Szabolcs A., Ryan D.P., Ting D.T., Wolfgang C.L., Burkhart R.A. Patient-derived organoid pharmacotyping is a clinically tractable strategy for precision medicine in pancreatic cancer. Ann Surg. 2020;272:427–435. doi: 10.1097/SLA.0000000000004200. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Rashid N.U., Peng X.L., Jin C., Moffitt R.A., Volmar K.E., Belt B.A., Panni R.Z., Nywening T.M., Herrera S.G., Moore K.J., Hennessey S.G., Morrison A.B., Kawalerski R., Nayyar A., Chang A.E., Schmidt B., Kim H.J., Linehan D.C., Yeh J.J. Purity independent subtyping of tumors (PurIST), a clinically robust, single-sample classifier for tumor subtyping in pancreatic cancer. Clin Cancer Res. 2020;26:82–92. doi: 10.1158/1078-0432.CCR-19-1467. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Rashid N.U., Li Q., Yeh J.J., Ibrahim J.G. Modeling between-study heterogeneity for improved replicability in gene signature selection and clinical prediction. J Am Stat Assoc. 2020;115:1125–1138. doi: 10.1080/01621459.2019.1671197. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Liu X., Wang W., Liu X., Zhang Z., Yu L., Li R., Guo D., Cai W., Quan X., Wu H., Dai M., Liang Z. Multi-omics analysis of intra-tumoural and inter-tumoural heterogeneity in pancreatic ductal adenocarcinoma. Clin Transl Med. 2022;12 doi: 10.1002/ctm2.670. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Chan-Seng-Yue M., Kim J.C., Wilson G.W., Ng K., Figueroa E.F., O'Kane G.M., et al. Transcription phenotypes of pancreatic cancer are driven by genomic events during tumor evolution. Nat Genet. 2020;52:231–240. doi: 10.1038/s41588-019-0566-9. [DOI] [PubMed] [Google Scholar]
  • 12.Peng X.L., Moffitt R.A., Torphy R.J., Volmar K.E., Yeh J.J. De novo compartment deconvolution and weight estimation of tumor samples using DECODER. Nat Commun. 2019;10:4729. doi: 10.1038/s41467-019-12517-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Schreyer D., Neoptolemos J.P., Barry S.T., Bailey P. Deconstructing pancreatic cancer using next generation-omic technologies-from discovery to knowledge-guided platforms for better patient management. Front Cell Dev Biol. 2021;9 doi: 10.3389/fcell.2021.795735. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Zheng C.M., Piao X.M., Byun Y.J., Song S.J., Kim S.K., Moon S.K., Choi Y.H., Kang H.W., Kim W.T., Kim Y.J., Lee S.C., Kim W.J., Yun S.J. Study on the use of Nanostring nCounter to analyze RNA extracted from formalin-fixed-paraffin-embedded and fresh frozen bladder cancer tissues. Cancer Genet. 2022;268-269:137–143. doi: 10.1016/j.cancergen.2022.10.143. [DOI] [PubMed] [Google Scholar]
  • 15.Zhou X., An J., Kurilov R., Brors B., Hu K., Peccerella T., Roessler S., Pfütze K., Schulz A., Wolf S., Hohmann N., Theile D., Sauter M., Burhenne J., Ei S., Heger U., Strobel O., Barry S.T., Springfeld C., Tjaden C., Bergmann F., Büchler M., Hackert T., Fortunato F., Neoptolemos J.P., Bailey P. Persister cell phenotypes contribute to poor patient outcomes after neoadjuvant chemotherapy in PDAC. Nat Cancer. 2023;4:1362–1381. doi: 10.1038/s43018-023-00628-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Wenric S., Davison J.M., Guittar J., Mayhew G.M., Beebe K.D., Wang Y.E., Iyer A.A., Hyunseok H.P., Milburn M.V., Chung V., Bekali-Saab T., Perou C.M. Real-world data validation of the purist pancreatic ductal adenocarcinoma gene expression classifier and its prognostic implications. medRxiv. 2023 doi: 10.1101/2023.02.23.23286356. [Preprint] [DOI] [Google Scholar]
  • 17.Wenric S., Davison J.M., Wang Y.E., Mayhew G.M., Beebe K., Kang H.P., Milburn M.V., Chung V., Bekaii-Saab T., Perou C.M. Abstract A002: purity independent subtyping of tumor (PurIST): real-world data validation of a pancreatic ductal adenocarcinoma (PDAC) gene expression classifier and its prognostic implications. Cancer Res. 2022;82(22_Supplement) [Google Scholar]
  • 18.Torre-Healy L.A., Kawalerski R.R., Oh K., Chrastecka L., Peng X.L., Aguirre A.J., Rashid N.U., Yeh J.J., Moffitt R.A. Open-source curation of a pancreatic ductal adenocarcinoma gene expression analysis platform (pdacR) supports a two-subtype model. Commun Biol. 2023;6:163. doi: 10.1038/s42003-023-04461-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from The Journal of Molecular Diagnostics : JMD are provided here courtesy of American Society for Investigative Pathology

RESOURCES