Virtual Library
Start Your Search
B. Rikke
Author of
-
+
P1.06 - Poster Session/ Screening and Early Detection (ID 218)
- Event: WCLC 2015
- Type: Poster
- Track: Screening and Early Detection
- Presentations: 1
- Moderators:
- Coordinates: 9/07/2015, 09:30 - 17:00, Exhibit Hall (Hall B+C)
-
+
P1.06-009 - Volatolomic Signatures to Assess Sensitivity to FGFR Tyrosine Kinase Inhibitors (ID 1711)
09:30 - 09:30 | Author(s): B. Rikke
- Abstract
Background:
Targeted therapy is transforming the treatment of lung cancer. Such therapies are critically dependent on companion diagnostics that can predict the response to therapy. An ideal test is one that is quick, inexpensive, and non-invasive. In this regard, artificial intelligence nanosensor-based devices that profile volatolomic signatures (through volatile organic compounds (VOCs) analysis) have shown exciting potential. Numerous studies have shown cancer cells produce characteristic patterns of VOCs as a byproduct of their metabolism. These patterns can be used to diagnose patients with cancer using exhaled-breath samples. Here we asked whether the VOC patterns emanating from cancer cells could also be used to guide targeted therapy. In particular, we investigated whether lung cancer cell lines known to be sensitive to FGFR tyrosine kinase inhibitors (TKIs) can be distinguished from cell lines known to be resistant using an array of cross reactive, highly sensitive chemiresistors composed of gold nanoparticles (GNP) and carbon nanotubes (CNTs) coated with various recognition layers previously shown to be highly effective at profiling VOCs.
Methods:
Fourteen sensitive cell lines having an IC~50~ ≤ 50 nM for Ponatinib and AZD4547 (nonspecific and specific FGFR TKIs, respectively) and 21 resistant cell lines representing small cell and non-small cell lung cancers were cultured in complete media (RPMI 1640, 10% fetal bovine serum, and penicillin/streptomycin) under standard conditions to 50% to 75% confluency. SKC Tenax® TA Adsorbent resin was used to collect the VOCs from the head space of each cell line over a period of 60 to 72 hours. Triplicate measures were collected on each sample along with biological replicates. VOCs were also collected at the same time from control plates containing media only. After thermal desorption, the VOC pattern of each sample was characterized using a chemiresistor array of 36 sensors and 4 features per sensor. A statistical pattern recognition analysis was then conducted using a discriminant function analysis (DFA) algorithm to identify the most informative sensors and features.
Results:
We found that sensitive cell lines could be distinguished from resistant cell lines using only 4 sensors and one feature from each (GNP+dodecanethiol, CNT+PAH, GNP+thiol and CNT+β dextrin). Leave-one-out cross validation indicated a sensitivity of 88% for the FGFR TKI-sensitive cell lines with 100% specificity and 92% accuracy. The area under the receiver-operating characteristic curve was 70% and Wilcoxon p-value of 0.06.
Conclusion:
Profiling the VOCs emanating from lung cancer cells shows excellent diagnostic potential as a means of gauging initial sensitivity to FGFR1 TKIs. Consequently, this study suggests that the electronic nose devices currently being developed to profile exhaled breath for cancer detection could also play an important role in predicting responses to targeted therapies. Although cell lines are useful for identifying the VOC pattern that predicts the cancer cell response to therapy, they do not necessarily reflect the complexity that occurs in vivo due to interactions with the microenvironment. Therefore, future studies are needed to confirm if these results can be extended to project efficacy in patients assigned to FGFR TKI therapy.
-
+
P2.04 - Poster Session/ Biology, Pathology, and Molecular Testing (ID 234)
- Event: WCLC 2015
- Type: Poster
- Track: Biology, Pathology, and Molecular Testing
- Presentations: 1
- Moderators:
- Coordinates: 9/08/2015, 09:30 - 17:00, Exhibit Hall (Hall B+C)
-
+
P2.04-006 - MiRNA Signature to Assess Sensitivity to FGFR Tyrosine Kinase Inhibitors (ID 1717)
09:30 - 09:30 | Author(s): B. Rikke
- Abstract
Background:
Increased signaling through the FGF/FGFR signaling pathway has been implicated as a driver in a number of different malignancies including lymphomas, prostate cancer, breast cancer, and lung cancer. This pathway also appears to play a role in conferring de novo and acquired resistance to cancers driven by EGFR mutations. Consequently, drugs that inhibit FGFRs are being investigated as potential therapeutics for cancer. Here we screened a large panel of miRNAs as potential predictors of sensitivity to FGFR tyrosine kinase inhibitors (TKIs).
Methods:
A panel of 377 miRNAs (Megaplex Card A, Life Technologies) was screened for expression level differences between four lung cancer cell lines that are sensitive (IC~50~< 50 nM) and four lines that are resistant (IC~50~ > 100 nM) to ponatinib (non-specific FGFR TKI) and AZD4547 (FGFR-specific TKI). Expression levels were assayed by RT-qPCR and analyzed using the Statistical Analysis of Microarrays (SAM) method. Thirty-nine miRNAs having an estimated false discover rate (FDR) of zero and large median fold differences (> 8) between the sensitive and resistant lines were selected for signature development. RT-qPCR assays were incorporated into a custom microfluidics card (Life Technologies), which was used to profile the original 8 cell lines and 10 additional sensitive lines and 16 additional resistant lines (34 lines total). Logistic regression was then used to identify the best signature panel for distinguishing sensitive cell lines from resistant.
Results:
Univariate analysis indicated three miRNAs (let-7c, miR-338, and miR-218) that differed between the sensitive and resistant lines at p < .05. The best signature panel consisted of let-7c, miR-200a and miR-200b, which gave an area under the receiver operator characteristic (AUROC) curve of 0.90 (95% CI = 0.8 to 1). This performance was nearly as good as using FGFR1 mRNA alone (AUROC = 0.94). The predominant miRNA in our 3-miRNA signature was let-7c, which also exhibited a suggestive additive effect to using FGFR1 as a biomarker (p = 0.09). We also tested whether cell lines with high sensitivity to ponatinib can be made resistant by reducing the high level of let-7c in these lines. We have found that transient transfection of let-7c silencing RNA (Life Technologies) produces a decrease in FGFR1 mRNA levels for some cell lines but not others.
Conclusion:
It appears possible to predict sensitivity to an FGFR1 inhibitor using miRNA expression signatures. More studies, however, are needed to confirm the 3-marker signature developed in this study. Modulating let-7c, the predominant predictor within the signature, appears to modulate FGFR1 levels in a manner consistent with altering ponatinib sensitivity. This effect is most likely indirect as the mRNA of FGFR1 does not contain predicted binding sites for let-7c.
-
+
P2.06 - Poster Session/ Screening and Early Detection (ID 219)
- Event: WCLC 2015
- Type: Poster
- Track: Screening and Early Detection
- Presentations: 1
- Moderators:
- Coordinates: 9/08/2015, 09:30 - 17:00, Exhibit Hall (Hall B+C)
-
+
P2.06-007 - A miRNA Signature Derived From Independently Replicated Biomarkers of Non-Small Cell Lung Cancer (ID 1728)
09:30 - 09:30 | Author(s): B. Rikke
- Abstract
Background:
miRNAs have shown exceptional promise as biomarkers of lung cancer; however, no miRNA signatures have yet reached the clinic. Towards developing a signature with a high likelihood of being validated externally for clinical use, we screened a panel of 50 miRNAs shown to be effective biomarkers in at least two previous studies for distinguishing human lung cancer samples from non-cancer samples.
Methods:
Sixty tumor-normal pairs (33 adenocarcinoma, 27 squamous cell carcinoma) were used to identify the best-performing combination of 4 miRNAs for distinguishing tumor samples from normal. The miRNA levels were measured by RT-qPCR using Taqman custom-made microfluidics cards and primer pools purchased from Life Technologies. All possible combinations of 4 miRNAs were tested, and best performance was defined as the highest median area-under the receiver operating curve (AUC) obtained from 1000 bootstrap replicates. A second, independent set of 68 tumor-normal samples (half adenocarcinoma, half squamous) was used as a test set, and bootstrapping was used to determine the 95% confidence interval for the AUC.
Results:
The median AUC for the top-performing panel of 4 miRNAs in our training set was 0.96. Several other miRNA combinations exhibited AUCs > 0.95 as well. In our test set, the top-performing panel (and only panel tested) exhibited an AUC of 0.97 (0.93, 0.99). This panel consisted of miRs 26a, 145, 183 and 486. miRs 145 and183 have previously been shown, when used individually, to be significant lung tumor biomarkers in at least 4 previous studies; miR-486 has been replicated 8 times.Figure 1
Conclusion:
Consistent with previous studies, we’ve identified a panel of 4 miRNAs that shows excellent potential for diagnosing lung tumors. Each of these miRNAs has been replicated as a biomarker of lung cancer in at least two previous studies, suggesting a high likelihood of achieving clinical validation. Several previous studies have also shown that these four miRNAs are potentially useful as biomarkers for diagnosing lung cancer using blood samples, and we are currently pursuing such validation studies.
-
+
P3.06 - Poster Session/ Screening and Early Detection (ID 220)
- Event: WCLC 2015
- Type: Poster
- Track: Screening and Early Detection
- Presentations: 1
- Moderators:
- Coordinates: 9/09/2015, 09:30 - 17:00, Exhibit Hall (Hall B+C)
-
+
P3.06-008 - Meta-Analysis Criteria Used to Rank Biomarkers for Validation Testing: What Works? (ID 81)
09:30 - 09:30 | Author(s): B. Rikke
- Abstract
Background:
Hundreds of biomarkers are being developed for the screening and early detection of lung cancer. The vast majority, however, even after extensive internal validation, will likely fail during external validation. For biomarkers to reach the clinic, therefore, it’s imperative that external validation studies focus on the most promising candidates. Towards this end, various strategies have been proposed to rank order and prioritize biomarker candidates. These strategies range from simple, highly intuitive ideas to highly sophisticated statistical analyses. To our knowledge, however, none of these strategies has itself been validated externally, which is an important consideration given that each strategy involves making subjective decisions. Here we conducted an independent validation test to assess the performance of the “vote-counting strategy”, a straightforward, commonly used strategy that ranks biomarkers on the basis of three highly intuitive criteria: the number of supporting studies in the literature, the combined sample size in the supporting studies, and the average fold change difference associated with the biomarker.
Methods:
We obtained vote-counting biomarker rankings from two recent meta-analyses that together surveyed over 180 miRNAs reported to distinguish lung tumor tissue from normal. We compared the rankings of 50 top candidates and 22 unranked miRNAs to our RT-qPCR results obtained from 45 tumor-normal pairs. We tested for a statistically significant Pearson correlation (r) between biomarker performance and the rankings according to each of the three ranking criteria.
Results:
We found that the number of supporting studies in the literature was indeed a statistically significant predictor of biomarker performance (r = 0.44, n = 50, p = .0006). Our results also suggested that markers supported by two studies in the literature had approximately a 50% chance of being confirmed, markers supported by 3 studies about a 67% chance, and markers supported by 6 studies about a 90% chance. Our unranked markers showed only a 5% chance of being confirmed. At the same time, we found that the combined sample size in the supporting studies was not a predictor of biomarker performance (r = 0.11, n = 50, p = 0.29). We also found that the mean fold change associated with each biomarker was not a predictor (r = 0.12, n = 47, p = 0.22) because large fold-change differences were also associated with large amounts of variability between studies.
Conclusion:
Considering that vote counting has obvious limitations (such as selection bias, not counting negative votes, and the variation in how different studies define significance) counting the number of supporting studies in the literature appears to work remarkably well for ranking biomarker candidates. On the other hand, using total sample size or mean fold change in the supporting studies to rank biomarker candidates appears to provide little, if any, added value. Our results also indicate a need for external validation testing of the current strategies being used to rank biomarkers across studies.
