Gene Expression Classifier Research Articles

Gene expression signature for predicting homologous recombination deficiency in triple-negative breast cancer

Triple-negative breast cancers (TNBCs) are a subset of breast cancers that have remained difficult to treat. A proportion of TNBCs arising in non-carriers of BRCA pathogenic variants have genomic features that are similar to BRCA carriers and may also benefit from PARP inhibitor treatment. Using genomic data from 129 TNBC samples from the Malaysian Breast Cancer (MyBrCa) cohort, we developed a gene expression-based machine learning classifier for homologous recombination deficiency (HRD) in TNBCs. The classifier identified samples with HRD mutational signature at an AUROC of 0.93 in MyBrCa validation datasets and 0.84 in TCGA TNBCs. Additionally, the classifier strongly segregated HRD-associated genomic features in TNBCs from TCGA, METABRIC, and ICGC. Thus, our gene expression classifier may identify triple-negative breast cancer patients with homologous recombination deficiency, suggesting an alternative method to identify individuals who may benefit from treatment with PARP inhibitors or platinum chemotherapy.

A single-sample gene-expression classifier for the intrinsic consensus molecular subtypes in colorectal cancer.

e15587 Background: An analysis of colorectal cancer epithelial cells showed two intrinsic subtypes called iCMS2 and iCMS3, in addition to the bulk consensus subtypes (CMS1-4). The intrinsic subtypes are prognostically important. We present a robust method for calculating the iCMS subtypes in individual samples, as opposed to batches, applicable in a clinical setting. Methods: We selected the subset of iCMS genes (N = 201) with the strongest epithelial expression in colorectal cancer, aiming to reduce the signal from other cell types. We then calculated gene expression centroids by taking the per-gene mean expression of 8 random samples from three non-batch corrected datasets (1779 samples total) that were previously used to derive the iCMS subtypes. Thirty-two centroids were calculated per dataset and per iCMS, for a total of 192 centroids. The single sample classifier (SSC) calculates the most likely iCMS class based on the class of the nearest centroid by non-parametric correlation distance. A confident call is made when the correlation is at least 0.1 and distance from the nearest centroid of the opposite iCMS class is greater than 0.05. Results: The SSC achieved 98.3% agreement with the reference calls in training data. The SSC was applied to unseen data for validation, by classifying non-batch corrected samples one at a time. The same data were clustered with the reference published iCMS public dataset and the reference iCMS was estimated at the batch level by proximity to reference samples in the hierarchical tree structure. Concordance between reference and SSC calls was 91.4% in non-batch corrected data and only modestly improved to 93.9% when batch correction was applied, showing the robustness of this approach. Concordance remained acceptable even when non-confident SSC calls were included (86.9%). The proportion of calls that were judged confident with the SSC was 77.9%, slightly lower than the proportion of confident reference calls (87.6%). In addition, the SSC was applied to data from the VELOUR trial, for which reference iCMS calls were also available. The SSC iCMS was prognostic (HR 0.601 for OS in iCMS2 vs iCMS3, P < 0.00001) and performed at least as well as the reference iCMS (HR 0.669, P = 0.0004). Conclusions: The SSC reproduces the iCMS classification and appears robust to batch effect. The SSC calls are prognostic in clinical trial data. The SSC R package is available for download (https://github.com/CRCrepository/iCMS.SSC). [Table: see text]

Diagnostic accuracy of Afirma gene expression classifier, Afirma gene sequencing classifier, ThyroSeq v2 and ThyroSeq v3 for indeterminate (Bethesda III and IV) thyroid nodules: a meta-analysis.

The management of thyroid nodules with indeterminate cytology (ITN) is still a challenge. To evaluate the performance of commercial molecular tests for ITN, we performed this comprehensive meta-analysis. We performed an electronic search using PubMed/Medline, Embase, and the Cochrane Library. Studies assessing the diagnostic accuracy of Afirma gene expression classifier (GEC), Afirma gene sequencing classifier (GSC), ThyroSeq v2 (TSv2), or ThyroSeq v3 (TSv3) in patients with ITN (only Bethesda category III or IV) were selected; Statistical analyses were performed by using Stata. Seventy-one samples (GEC, n = 38; GSC, n = 16; TSv2, n = 9; TSv3, n = 8) in 53 studies, involving 6490 fine needle aspirations (FNAs) with ITN cytology with molecular diagnostics (GEC, GSC, TSv2, or TSv3), were included in the study. The meta-analysis showed the following pooled estimates: sensitivity 0.95 (95% CI: 0.94-0.97), specificity 0.35 (0.28-0.43), positive likelihood ratio (LR+) 1.5 (1.3-1.6), and negative likelihood ratio (LR-) 0.13 (0.09-0.19), with the best performance for TSv3 (area under the ROC curve 0.95 (0.93-0.96), followed by TSv2 (0.90 (0.87-0.92)), GSC (0.86 (0.82-0.88)), and GEC (0.82 (0.78-0.85)); the best rule-out property was observed for GSC (LR-, 0.07 (0.02-0.19)), followed by TSv3 (0.11 (0.05-0.24)) and GEC (0.16 (0.10-0.28), and the best rule-in was observed for TSv2 (LR+, 2,9 (1.4-4.6)), followed by GSC (1.9 (1.6-2.4)). A meta-regression analysis revealed that study design, Bethesda category, and type of molecular test were independent factors. We showed that in patients with ITN, TSv3 has the best molecular diagnostic performance, followed by TSv2, GSC, and GEC. As regards rule-out malignancy, GSC, and rule-in, TSV2 is superior to other tests.

Expert Perspectives on Controversies in Castration-Sensitive Prostate Cancer Management: Narrative Review and Report of the First US Prostate Cancer Conference Part 1.

Castration-sensitive prostate cancer (CSPC) is a complex and heterogeneous condition encompassing a range of clinical presentations. As new approaches have expanded management options, clinicians are left with myriad questions and controversies regarding the optimal individualized management of CSPC. The US Prostate Cancer Conference (USPCC) multidisciplinary panel was assembled to address the challenges of prostate cancer management. The first annual USPCC meeting included experts in urology, medical oncology, radiation oncology, and nuclear medicine. USPCC co-chairs and session moderators identified key areas of controversy and uncertainty in prostate cancer management and organized the sessions with multidisciplinary presentations and discussion. Throughout the meeting, experts responded to questions prepared by chairs and moderators to identify areas of agreement and controversy. The USPCC panel discussion and question responses for CSPC-related topics are presented. Key advances in CSPC management endorsed by USPCC experts included the development and clinical utilization of gene expression classifiers and artificial intelligence (AI) models for risk stratification and treatment selection in specific patient populations, the use of advanced imaging modalities in patients with clinically localized unfavorable intermediate or high-risk disease and those with biochemical recurrence, recommendations of doublet or triplet therapy for metastatic CSPC (mCSPC), and consideration of prostate and/or metastasis-directed radiation therapy in select patients with mCSPC. CSPC is a diverse disease with many therapeutic options and the potential for adverse outcomes associated with either undertreatment or overtreatment. Future studies are needed to validate and clinically integrate novel technologies, including genomics, AI, and advanced imaging, to optimize outcomes among patients with CSPC.

Host-response transcriptional biomarkers accurately discriminate bacterial and viral infections of global relevance

Diagnostic limitations challenge management of clinically indistinguishable acute infectious illness globally. Gene expression classification models show great promise distinguishing causes of fever. We generated transcriptional data for a 294-participant (USA, Sri Lanka) discovery cohort with adjudicated viral or bacterial infections of diverse etiology or non-infectious disease mimics. We then derived and cross-validated gene expression classifiers including: 1) a single model to distinguish bacterial vs. viral (Global Fever-Bacterial/Viral [GF-B/V]) and 2) a two-model system to discriminate bacterial and viral in the context of noninfection (Global Fever-Bacterial/Viral/Non-infectious [GF-B/V/N]). We then translated to a multiplex RT-PCR assay and independent validation involved 101 participants (USA, Sri Lanka, Australia, Cambodia, Tanzania). The GF-B/V model discriminated bacterial from viral infection in the discovery cohort an area under the receiver operator curve (AUROC) of 0.93. Validation in an independent cohort demonstrated the GF-B/V model had an AUROC of 0.84 (95% CI 0.76–0.90) with overall accuracy of 81.6% (95% CI 72.7–88.5). Performance did not vary with age, demographics, or site. Host transcriptional response diagnostics distinguish bacterial and viral illness across global sites with diverse endemic pathogens.

The Prediction Analysis of Microarray 50 (PAM50) Gene Expression Classifier Utilized in Indeterminate-Risk Breast Cancer Patients in Hungary: A Consecutive 5-Year Experience.

Breast cancer has been categorized into molecular subtypes using immunohistochemical staining (IHC) and fluorescence in situ hybridization (FISH) since the early 2000s. However, recent research suggests that gene expression testing, specifically Prosigna® Prediction Analysis of Microarray 50 (PAM50), provides more accurate classification methods. In this retrospective study, we compared the results of IHC/FISH and PAM50 testing. We also examined the impact of various PAM50 parameters on overall survival (OS) and progression-free survival (PFS). We analyzed 42 unilateral breast cancer samples, with 18 classified as luminal A, 10 as luminal B, 8 as Human epidermal growth factor receptor 2 (HER2)-positive, and 6 as basal-like using PAM50. Interestingly, 17 out of the 42 samples (40.47%) showed discordant results between histopathological assessment and the PAM50 classifier. While routine IHC/FISH resulted in classification differences for a quarter to a third of samples within each subtype, all basal-like tumors were misclassified. Hormone receptor-positive tumors (hazard rate: 8.7803; p = 0.0085) and patients who had higher 10-year recurrence risk scores (hazard rate: 1.0539; p = 0.0201) had shorter OS and PFS. Our study supports the existing understanding of molecular subtypes in breast cancer and emphasizes the overlap between clinical characteristics and molecular subtyping. These findings underscore the value of gene expression profiling, such as PAM50, in improving treatment decisions for breast cancer patients.

Aberrant activation of five embryonic stem cell-specific genes robustly predicts a high risk of relapse in breast cancers

BackgroundIn breast cancer, as in all cancers, genetic and epigenetic deregulations can result in out-of-context expressions of a set of normally silent tissue-specific genes. The activation of some of these genes in various cancers empowers tumours cells with new properties and drives enhanced proliferation and metastatic activity, leading to a poor survival prognosis.ResultsIn this work, we undertook an unprecedented systematic and unbiased analysis of out-of-context activations of a specific set of tissue-specific genes from testis, placenta and embryonic stem cells, not expressed in normal breast tissue as a source of novel prognostic biomarkers. To this end, we combined a strict machine learning framework of transcriptomic data analysis, and successfully created a new robust tool, validated in several independent datasets, which is able to identify patients with a high risk of relapse. This unbiased approach allowed us to identify a panel of five biomarkers, DNMT3B, EXO1, MCM10, CENPF and CENPE, that are robustly and significantly associated with disease-free survival prognosis in breast cancer. Based on these findings, we created a new Gene Expression Classifier (GEC) that stratifies patients. Additionally, thanks to the identified GEC, we were able to paint the specific molecular portraits of the particularly aggressive tumours, which show characteristics of male germ cells, with a particular metabolic gene signature, associated with an enrichment in pro-metastatic and pro-proliferation gene expression.ConclusionsThe GEC classifier is able to reliably identify patients with a high risk of relapse at early stages of the disease. We especially recommend to use the GEC tool for patients with the luminal-A molecular subtype of breast cancer, generally considered of a favourable disease-free survival prognosis, to detect the fraction of patients undergoing a high risk of relapse.

Construction and validation of a gene expression classifier to predict immunotherapy response in primary triple-negative breast cancer

BackgroundImmune checkpoint inhibitors (ICI) improve clinical outcomes in triple-negative breast cancer (TNBC) patients. However, a subset of patients does not respond to treatment. Biomarkers that show ICI predictive potential in other solid tumors, such as levels of PD-L1 and the tumor mutational burden, among others, show a modest predictive performance in patients with TNBC.MethodsWe built machine learning models based on pre-ICI treatment gene expression profiles to construct gene expression classifiers to identify primary TNBC ICI-responder patients. This study involved 188 ICI-naïve and 721 specimens treated with ICI plus chemotherapy, including TNBC tumors, HR+/HER2− breast tumors, and other solid non-breast tumors.ResultsThe 37-gene TNBC ICI predictive (TNBC-ICI) classifier performs well in predicting pathological complete response (pCR) to ICI plus chemotherapy on an independent TNBC validation cohort (AUC = 0.86). The TNBC-ICI classifier shows better performance than other molecular signatures, including PD-1 (PDCD1) and PD-L1 (CD274) gene expression (AUC = 0.67). Integrating TNBC-ICI with molecular signatures does not improve the efficiency of the classifier (AUC = 0.75). TNBC-ICI displays a modest accuracy in predicting ICI response in two different cohorts of patients with HR + /HER2- breast cancer (AUC = 0.72 to pembrolizumab and AUC = 0.75 to durvalumab). Evaluation of six cohorts of patients with non-breast solid tumors treated with ICI plus chemotherapy shows overall poor performance (median AUC = 0.67).ConclusionTNBC-ICI predicts pCR to ICI plus chemotherapy in patients with primary TNBC. The study provides a guide to implementing the TNBC-ICI classifier in clinical studies. Further validations will consolidate a novel predictive panel to improve the treatment decision-making for patients with TNBC.

Chronic Lymphocytic Thyroiditis May Limit the Utility of Molecular Testing in AUS/FLUS Thyroid Nodules

IntroductionChronic lymphocytic thyroiditis (CLT) may increase the likelihood of atypia of undetermined significance/follicular lesion of undetermined significance (AUS/FLUS) results in thyroid nodules by fine needle aspiration (FNA). Gene expression classifier (GEC) and Thyroid Sequencing (ThyroSeq) may better stratify rate of malignancy (ROM) of AUS/FLUS thyroid nodules. This study compares the utility of molecular tests in determining malignancy in surgical patients with concomitant AUS/FLUS thyroid nodules and CLT. MethodsA retrospective review of 1648 patients with index thyroid nodules who underwent FNA and thyroidectomy at a single institution was performed. Patients with concomitant AUS/FLUS thyroid nodules and CLT were subdivided into three diagnostic groups: FNA only, FNA with GEC, and FNA with ThyroSeq. Patients with AUS/FLUS thyroid nodules without CLT were subdivided into similar groups. Final histopathology of the cohorts was further stratified into benignity and malignancy and analyzed using Chi-squared statistics. ResultsOf 463 study patients, 86 had concomitant AUS/FLUS thyroid nodules and CLT with a 52% ROM, and the difference of ROM among FNA only (48%), suspicious GEC (50%), or positive ThyroSeq (69%) was not significant. In 377 patients with AUS/FLUS thyroid nodules without CL, ROM was 59%. ROM among these patients was significantly higher when molecular testing was used (FNA only 51%, suspicious GEC 65%, and positive ThyroSeq 68%; P < 0.05). ConclusionsMolecular tests may have limited value in predicting malignancy in surgical patients with concomitant AUS/FLUS thyroid nodules and CLT.

Genomic Classifier Performance in Intermediate-Risk Prostate Cancer: Results From NRG Oncology/RTOG 0126 Randomized Phase 3 Trial

Intermediate-risk prostate cancer is a heterogeneous disease state with diverse treatment options. The 22-gene Decipher genomic classifier (GC) retrospectively has shown to improve risk stratification in these patients. We assessed the performance of the GC in men with intermediate-risk disease enrolled in NRG Oncology/RTOG 01-26 with updated follow-up. After National Cancer Institute approval, biopsy slides were collected from NRG Oncology/RTOG 01-26, a randomized phase 3 trial of men with intermediate-risk prostate cancer randomized to 70.2 Gy versus 79.2 Gy of radiation therapy without androgen deprivation therapy. RNA was extracted from the highest-grade tumor foci to generate the locked 22-gene GC model. The primary endpoint for this ancillary project was disease progression (composite of biochemical failure, local failure, distant metastasis, prostate cancer-specific mortality, and use of salvage therapy). Individual endpoints were also assessed. Fine-Gray or cause-specific Cox multivariable models were constructed adjusting for randomization arm and trial stratification factors. Two-hundred fifteen patient samples passed quality control for analysis. The median follow-up was 12.8 years (range, 2.4-17.7). On multivariable analysis, the 22-gene GC (per 0.1 unit) was independently prognostic for disease progression (subdistribution hazard ratio [sHR], 1.12; 95% confidence interval [CI], 1.00-1.26; P=.04), biochemical failure (sHR, 1.22; 95% CI, 1.10-1.37; P < .001), distant metastasis (sHR, 1.28; 95% CI, 1.06-1.55; P=.01), and prostate cancer-specific mortality (sHR, 1.45; 95% CI, 1.20-1.76; P < .001). Ten-year distant metastasis in GC low-risk patients was 4% compared with 16% for GC high-risk patients. In patients with lower GC scores, the 10-year difference in metastasis-free survival rate between arms was -7%, compared with 21% for higher GC patients (P-interaction=.04). This study represents the first validation of a biopsy-based gene expression classifier, assessing both its prognostic and predictive value, using data from a randomized phase 3 trial of intermediate-risk prostate cancer. Decipher improves risk stratification and can aid in treatment decision-making in men with intermediate-risk disease.

Integrated host/microbe metagenomics enables accurate lower respiratory tract infection diagnosis in critically ill children.

BACKGROUNDLower respiratory tract infection (LRTI) is a leading cause of death in children worldwide. LRTI diagnosis is challenging because noninfectious respiratory illnesses appear clinically similar and because existing microbiologic tests are often falsely negative or detect incidentally carried microbes, resulting in antimicrobial overuse and adverse outcomes. Lower airway metagenomics has the potential to detect host and microbial signatures of LRTI. Whether it can be applied at scale and in a pediatric population to enable improved diagnosis and treatment remains unclear.METHODSWe used tracheal aspirate RNA-Seq to profile host gene expression and respiratory microbiota in 261 children with acute respiratory failure. We developed a gene expression classifier for LRTI by training on patients with an established diagnosis of LRTI (n = 117) or of noninfectious respiratory failure (n = 50). We then developed a classifier that integrates the host LRTI probability, abundance of respiratory viruses, and dominance in the lung microbiome of bacteria/fungi considered pathogenic by a rules-based algorithm.RESULTSThe host classifier achieved a median AUC of 0.967 by cross-validation, driven by activation markers of T cells, alveolar macrophages, and the interferon response. The integrated classifier achieved a median AUC of 0.986 and increased the confidence of patient classifications. When applied to patients with an uncertain diagnosis (n = 94), the integrated classifier indicated LRTI in 52% of cases and nominated likely causal pathogens in 98% of those.CONCLUSIONLower airway metagenomics enables accurate LRTI diagnosis and pathogen identification in a heterogeneous cohort of critically ill children through integration of host, pathogen, and microbiome features.FUNDINGSupport for this study was provided by the Eunice Kennedy Shriver National Institute of Child Health and Human Development and the National Heart, Lung, and Blood Institute (UG1HD083171, 1R01HL124103, UG1HD049983, UG01HD049934, UG1HD083170, UG1HD050096, UG1HD63108, UG1HD083116, UG1HD083166, UG1HD049981, K23HL138461, and 5R01HL155418) as well as by the Chan Zuckerberg Biohub.

Geographic variation of decipher genomic classifier in patients with early prostate cancer.

376 Background: External factors such as environmental or lifestyle-related exposure may influence genomic composition of cancer, and therefore its progression and outcome. Thus geographic data is of importance. Decipher is a gene expression classifier used in risk stratification of early non metastatic prostate cancer. Currently, there are no data reporting geographic variation of the test. We aimed to analyze Decipher genomic score results among Israeli patients, in comparison to American patients. Methods: The Decioher database (Decipher Biosciences Inc) contains data from patients who have given consent for broad data-sharing. We identified and analyzed Decipher test results ordered among Israeli patients with localized prostate cancer. Matched analysis to the US population was done. Results: 144 patients (median age 69.5, range 64-74) were included in the present analysis. The clinical NCCN risk grouping was low in 42.4% (n=61), favorable intermediate in 38.9% (n=56), unfavorable intermediate in 17.4% (n=25), and high risk in 1.4% (n=2). Physicians primarily obtained Decipher for low-intermediate NCCN risk. Median Decipher score was 0.57 (Q1 0.385, Q3 0.823). Decipher risk grouping was low in 38.1% (n=61), intermediate in 16.9% (n=27), and high in 45% (n=72). Median Decipher score increased with Grade Group. 46% of NCCN low risk patients had a higher risk as per Decipher. 82% of NCCN favorable intermediate and unfavorable intermediate risk were reclassified to lower or higher risk. 9% of NCCN high risk patients had a lower risk as per Decipher. The present study cohort had a higher proportion of higher Decipher scores than observed in similar grade US population (n=43065) in the Decipher database (Median 0.43, Q1 0.27, Q3 0.66; 52%/18%/30% with low/intermediate/high risk). Conclusions: The preliminary results of the present study suggest a higher proportion of high risk Decipher among Israeli patients with early prostate cancer. These results should be interpreted cautiously. The study and further data collection are ongoing.

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

Abstract Background: PDAC is a highly morbid disease with no validated biomarkers for first-line (1L) treatment selection. The PurIST single-sample molecular subtyping gene signature, initially described by Rashid et al. (Clin. Cancer Res., 2020), classifies PDAC tumors as basal or classical. Prior work showed that these subtypes are associated with prognosis and that basal subtype patients have significantly lower objective response rate to FOLFIRINOX (FFX) compared to classical subtype patients. This suggests potential implications for treatment with FFX versus gemcitabine and nab-paclitaxel (GnP). Here, we retrospectively demonstrate the clinical validity of PurIST implemented as a laboratory developed test (LDT) on the Tempus Labs sequencing platform using a real-world dataset of advanced PDAC patients. Methods: De-identified PDAC patients were selected from the Tempus Oncology Data Ecosystem under an IRB-approved protocol according to the following inclusion criteria: no systemic treatment, surgically unresectable/metastatic, available RNA-sequencing data from primary or metastatic tissue, and FFX or GnP as 1L systemic therapy with available outcomes. Sequencing was performed by the CAP/CLIA validated Tempus xT assay. The PurIST model was applied to normalized RNA-sequence abundance files to compute a basal or classical subtype label. Predefined statistical analysis parameters included 12-month survival rate and median overall survival (OS) in FFX treated patients. OS was compared using Kaplan-Meier estimates, hazard ratios, and log-rank statistical tests. Results: 258 PDAC patients (64.9 +/- 9.9 yrs, 42% female), were identified for analysis with median OS of 11.7 months (95% CI: 10.6-13.7) and 12-month censorship rate of 12.8%. 151 patients were treated with FFX and 107 with GnP. 42 patients (28%) receiving FFX and 31 patients (29%) receiving GnP were classified as basal. Among FFX treated patients, median OS was 14.4 months (95% CI: 12.5-16.9) in classical patients vs. 9.4 months (95% CI: 8.3-14.5) in basal patients (HR = 1.72, 95% CI = 1.2-2.6, p=0.006). The 12-month survival rate was significantly lower in basal patients receiving FFX vs. classical patients (33.3% vs. 59.4%, p=0.011). In basal patients, no difference in OS was observed between FFX and GnP groups (p=0.6). Classical patients receiving FFX had a 14.4 months median OS vs. 10.8 months for patients receiving GnP (p=0.046). Conclusions: In this real-world cohort, we validate the association between PurIST subtypes and PDAC patient survival when administered FFX or GnP. Among FFX-treated patients, classical patients had significantly better outcomes compared to basal patients. Moreover, classical patients appeared to have improved outcomes with FFX vs. GnP. These findings represent underlying biological PDAC differences and demonstrate the clinical validity of PurIST as a prognostic marker in PDAC patients when performed as an LDT on the Tempus xT platform. Ongoing evaluation of PurIST performance will be continuously monitored through the Tempus Oncology Data ecosystem. Citation Format: Stephane Wenric, James M. Davison, Yun E. Wang, Gregory M. Mayhew, Kirk Beebe, Hyunseok P. Kang, Michael V. Milburn, Vincent Chung, Tanios Bekaii-Saab, Charles M. Perou. Purity Independent Subtyping of Tumor (PurIST): Real-world data validation of a pancreatic ductal adenocarcinoma (PDAC) gene expression classifier and its prognostic implications [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer; 2022 Sep 13-16; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2022;82(22 Suppl):Abstract nr A002.

A Transcriptional Classifier Identifies Pediatric T-Cell Acute Lymphoblastic Leukemias at High Risk for End of Induction Minimal Residual Disease

Upfront risk stratification with incorporation of risk-adapted therapy continues to be a pressing need for pediatric T-ALL. However, robust molecular or cytogenetic biomarkers are currently lacking. Instead, initial response to therapy, measured as end of induction minimal residual disease (MRD), is the only prognostic factor used in most cooperative groups. This inability to stratify patients at diagnosis limits the early introduction of rationally-designed therapies that may better achieve long-term remissions. To address this, we asked whether gene expression profiling performed at diagnosis may identify patients at high risk for future MRD+, enabling earlier therapeutic intervention. To test this, we used a training dataset of RNA-seq data from diagnostic samples from patients enrolled on the Children's Oncology Group (COG) AALL0434 trial. Using leave-one-out cross-validation, we defined a risk score (RS) for MRD+ based on the expression of 119 genes. This RS indicates the probability of being MRD+ for a given gene expression pattern. Gene ontology analysis of the genes in our model demonstrated enrichment for genes involved in signal transduction, lymphocyte differentiation and activation, and cell death. Using this model, we found that early T-cell precursor (ETP) samples (n=19), of which 79% were MRD+, had uniformly high RS, with an average of 78 (SD=17). In contrast, non-ETP samples (n=146), of which 26% were MRD+, had an average RS of 26 (SD=21), with MRD+ patients having a significantly higher RS relative to MRD- patients (p<0.0001). When we analyzed the expression pattern of the 119 genes, we found that ETP samples had a relatively uniform pattern, consistent with their high rate of MRD+. Non-ETP samples with RS<50 had a pattern that was inverted relative to the ETP samples, while those with RS>50 closely mirrored the ETP samples. Using a RS cutoff of 50 to identify patients at high risk for MRD+ gives a 97% specificity in this training cohort. Given the similar gene expression pattern between ETPs and RS>50 non-ETPs, we hypothesized that this RS classification may relate to differences in developmental stage. We found that ETP and RS>50 non-ETP samples were significantly enriched for the LMO2/LYL1, HOXA, and TLX3 subgroups of T-ALL, with the subgroup distribution in the RS>50 non-ETP samples differing significantly from the RS<50 non-ETP samples (p=0.001). Furthermore, we found that the RS>50 non-ETP samples, relative to the RS<50 samples, were enriched for lacking a TCRb rearrangement (p<0.0001), indicative of an earlier developmental stage. To determine the generalizability of our gene expression classifier, we applied our fixed model to an independent validation cohort of 826 non-ETP samples from the same COG AALL0434 trial. Here, we again found that RS>50 non-ETP samples (n=130) were significantly enriched for MRD+ relative to RS<50 samples (n=696; p<0.0001). Next, we applied our classifier model to a pilot cohort of samples from patients enrolled on COG AALL1231, which used dexamethasone rather than prednisone as the induction steroid. In this group of 31 diagnostic non-ETP samples, RS>50 samples were similarly enriched for MRD+ (p=0.05), suggesting that this classifier may retain its prognostic value across multiple induction steroids. Finally, to facilitate clinical implementation of this prognostic tool, we transitioned to the NanoString nCounter platform, a more rapid and cost-effective means of performing targeted gene expression analysis. Using 364 samples from COG AALL0434 for which we had corresponding RNA-seq data, we assessed the performance of our custom NanoString assay and found a strong correlation between the RS calculated using each of the two platforms (concordance correlation coefficient = 0.98), suggesting that the prognostic value of the classifier is retained on the NanoString assay. In conclusion, while ETP samples are commonly MRD+ and are readily identifiable by their unique cell surface immunophenotype, non-ETP samples lack a surface immunophenotype that correlates with MRD+. Here, we present a targeted gene expression classifier, validated on a clinically-tractable platform, that identifies at diagnosis a subset of non-ETP T-ALLs with a gene expression pattern that resembles that of ETP T-ALLs and that are at high risk for future MRD+.

A Multi-Cohort Gene Expression Classifier (ALLCatchR) Identifies B-Precursor ALL Subtypes and Their Developmental Trajectories across Age Groups

The molecular landscape of B cell precursor acute lymphoblastic leukemia (BCP-ALL) describes >20 disease subtypes defined by initiating genomic alterations and corresponding gene expression signatures. Some subtypes represent phenocopies with shared gene expression signatures but distinct or absent genomic drivers. The majority of BCP-ALL subtypes have been acknowledged as diagnostic entities by the currently revised classifications of hematologic malignancies (WHO-HAEM5 / International Consensus Classification). Thus, unified diagnostic approaches are needed to systematically identify subtypes in individual BCP-ALL cases in clinical routine. Gene expression signatures represent the shared functional equivalent of heterogeneous genomic driver alterations. Therefore, we established a gene expression-based classifier to identify BCP-ALL subtypes and related clinical and biological phenotypes. We analyzed a total of n=3477 BCP-ALL gene expression profiles obtained from transcriptome sequencing of adult (n=1253) and pediatric patients (n=2224) including publicly available (n=2074; Gu Z et al. Nat. Genet. 2019; Schmidt B. et al. Blood Adv. 2022) and own data sets (n=1403) from adult and pediatric European study groups (GMALL, AIEOP-BFM, CELL). Raw read counts from 6 cohorts representing heterogeneous library preps (i.e., stranded/unstranded; mRNA/total RNA), sequencing platforms/depths, and gene count quantifications were included. All samples had been previously allocated to molecular BCP-ALL subtypes based on integration of independent genomic profiling, gene fusion calling and gene expression. For classification, data with defined subgroup allocation were randomly split into training (n=1831; 70%) and testing sets (n=785; 30%) and two hold-out studies (n=305). To define gene expression signatures of 21 BCP-ALL subtypes, we used training set samples and extracted specific gene sets employing different feature selection algorithms as well as ranking of genes in individual subtypes. Based on these definitions we established a BCP-ALL subtype classifier (ALLCatchR), which integrates machine learning models (support vector machines) and nearest-neighbors associations of single sample gene set enrichment analyses (Fig. A). ALLCatchR performed molecular subtype allocation for the testing set at high sensitivity (0.934±0.165) and specificity (0.997±0.005) on average across subtypes. A similar performance was achieved in the two hold-out cohorts despite completely independent data structures which have not been shown to the classifier (sensitivity: 0.946±0.097 / specificity: 0.996±0.007). Overall accuracies of 0.952 and 0.92 for testing and hold-out-data sets indicate that ALLCatchR reliably identifies molecular BCP-ALL subtypes in mixed adult / pediatric cohorts based on gene expression profiles alone (Fig. A, below). Among n=44/1090 (4%) samples of the combined hold-out/testing set with ambiguous subtype allocation (deemed 'unclassified'), ALLCatchR provides candidate subtype allocations (sensitivity: 0.671±0.393 / specificity: 0.978±0.036) for confirmation based on genomic driver alterations. In addition, we modified our classifier to predict baseline variables such as patient's sex (accuracy: 0.991), immunophenotype (accuracy: 0.935) or blast count (R=0.633). To define the corresponding B cell differentiation stage of origin, we compared BCP-ALL subtypes to differentiation stage-specific gene sets extracted from 7 FACS-sorted lymphopoietic subsets from healthy bone marrow donors (n=4). We observed defined BCP-ALL subtype-specific enrichment patterns which can be grouped along 4 major differentiation stages (Pro-B to Pre-B-II-Large, Fig. 1B) with a high degree of similarity in independent adult and pediatric cohorts. Thus, B cell developmental trajectories underlying BCP-ALL subtypes are dissected at high resolution by gene expression profiling and differentiation stages may favor the selection of specific leukemogenic drivers. ALLCatchR is a gene expression classifier for 21 established BCP-ALL molecular subtypes, which performs at independently validated cohorts with an accuracy of >90% and allows imputation of clinical baseline variables as well as underlying B cell developmental trajectories. It provides a reliable basis for systematic diagnostic approaches in pediatric and adult BCP-ALL patients. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

A predictive ensemble classifier for the gene expression diagnosis of ASD at ages 1 to 4 years

Autism Spectrum Disorder (ASD) diagnosis remains behavior-based and the median age of diagnosis is ~52 months, nearly 5 years after its first-trimester origin. Accurate and clinically-translatable early-age diagnostics do not exist due to ASD genetic and clinical heterogeneity. Here we collected clinical, diagnostic, and leukocyte RNA data from 240 ASD and typically developing (TD) toddlers (175 toddlers for training and 65 for test). To identify gene expression ASD diagnostic classifiers, we developed 42,840 models composed of 3570 gene expression feature selection sets and 12 classification methods. We found that 742 models had AUC-ROC ≥ 0.8 on both Training and Test sets. Weighted Bayesian model averaging of these 742 models yielded an ensemble classifier model with accurate performance in Training and Test gene expression datasets with ASD diagnostic classification AUC-ROC scores of 85–89% and AUC-PR scores of 84–92%. ASD toddlers with ensemble scores above and below the overall ASD ensemble mean of 0.723 (on a scale of 0 to 1) had similar diagnostic and psychometric scores, but those below this ASD ensemble mean had more prenatal risk events than TD toddlers. Ensemble model feature genes were involved in cell cycle, inflammation/immune response, transcriptional gene regulation, cytokine response, and PI3K-AKT, RAS and Wnt signaling pathways. We additionally collected targeted DNA sequencing smMIPs data on a subset of ASD risk genes from 217 of the 240 ASD and TD toddlers. This DNA sequencing found about the same percentage of SFARI Level 1 and 2 ASD risk gene mutations in TD (12 of 105) as in ASD (13 of 112) toddlers, and classification based only on the presence of mutation in these risk genes performed at a chance level of 49%. By contrast, the leukocyte ensemble gene expression classifier correctly diagnostically classified 88% of TD and ASD toddlers with ASD risk gene mutations. Our ensemble ASD gene expression classifier is diagnostically predictive and replicable across different toddler ages, races, and ethnicities; out-performs a risk gene mutation classifier; and has potential for clinical translation.

Development of a Clinically Applicable NanoString-Based Gene Expression Classifier for Muscle-Invasive Bladder Cancer Molecular Stratification

Simple SummaryMolecular subtyping of muscle-invasive bladder cancer (MIBC) via gene expression can improve therapeutic decision-making and disease prognosis. However, the currently used molecular classification tools are based on complex transcriptomic profiling methodology that hinders timely translation to clinical practice. In this study, we evaluated the NanoString nCounter platform and conventional GATA3-CK5/6 immunohistochemistry for the molecular classification of MIBC in primary care settings. The methodologies were highly concordant and allowed us to explore differences in clinicopathologic parameters and prognosis between intrinsic MIBC molecular subtypes in a cohort of 138 MIBCs.Transcriptional profiling of muscle-invasive bladder cancer (MIBC) using RNA sequencing (RNA-seq) technology has demonstrated the existence of intrinsic basal and luminal molecular subtypes that vary in their prognosis and response to therapy. However, routine use of RNA-seq in a clinical setting is restricted by cost and technical difficulties. Herein, we provide a single-sample NanoString-based seven-gene (KRT5, KRT6C, SERPINB13, UPK1A, UPK2, UPK3A and KRT20) MIBC molecular classifier that assigns a luminal and basal molecular subtype. The classifier was developed in a series of 138 chemotherapy naïve MIBCs split into training (70%) and testing (30%) datasets. Further, we validated the previously published CK5/6 and GATA3 immunohistochemical classifier which showed high concordance of 96.9% with the NanoString-based gene expression classifier. Immunohistochemistry-based molecular subtypes significantly correlated with recurrence-free survival (RFS) and disease-specific survival (DSS) in univariable (p = 0.006 and p = 0.011, respectively) and multivariate cox regression analysis for DSS (p = 0.032). Used sequentially, the immunohistochemical- and NanoString-based classifiers provide faster turnaround time, lower cost per sample and simpler data analysis for ease of clinical implementation in routine diagnostics.

Assessing Bias and Limitations of Clinical Validation Studies of Molecular Diagnostic Tests for Indeterminate Thyroid Nodules: Systematic Review and Meta-Analysis.

Background: Molecular tests for thyroid nodules with indeterminate fine needle aspiration results are increasingly used in clinical practice; however, true diagnostic summaries of these tests are unknown. A systematic review and meta-analysis were completed to (1) evaluate the accuracy of commercially available molecular tests for malignancy in indeterminate thyroid nodules and (2) quantify biases and limitations in studies that validate those tests. Summary: PubMed, EMBASE, and Web of Science were systematically searched through July 2021. English language articles that reported original clinical validation attempts of molecular tests for indeterminate thyroid nodules were included if they reported counts of true-negative, true-positive, false-negative, and false-positive results. We performed screening and full-text review, followed by assessment of eight common biases and limitations, extraction of diagnostic and histopathological information, and meta-analysis of clinical validity using a bivariate linear mixed-effects model. Forty-nine studies were included. Meta-analysis of Afirma Gene expression classifiers (GEC; n = 38 studies) revealed a sensitivity of 0.92 (confidence interval: 0.90-0.94), specificity of 0.26 (0.20-0.32), negative likelihood ratio (LR-) of 0.32 (0.23-0.44), positive LR+ of 1.24 (1.15-1.35), and area under the curve (AUC) of 0.83 (0.74-0.89). Afirma Genomic Sequencing Classifier (GSC; n = 10) had a sensitivity of 0.94 (0.89-0.96), specificity of 0.38 (0.27-0.50), LR- of 0.18 (0.10-0.30), LR+ of 1.52 (1.28-1.87), and AUC of 0.91 (0.62-0.92). ThyroSeq v1 and v2 (n = 10) had a sensitivity of 0.86 (0.82-0.90), specificity of 0.74 (0.59-0.85), LR- of 0.19 (0.13-0.26), LR+ of 3.52 (2.08-5.92), and AUC of 0.86 (0.81-0.90). ThyroSeq v3 (n = 6) had a sensitivity of 0.92 (0.86-0.95), specificity of 0.41 (0.18-0.69), LR- of 0.24 (0.09-0.62), LR+ of 1.67 (1.09-2.98), and AUC of 0.90 (0.63-0.92). Fourteen percent of studies conducted a blinded histopathologic review of excised thyroid nodules, and 8% made the decision to go to surgery blind to molecular test results. Conclusions: Meta-analyses reveal a high diagnostic accuracy of molecular tests for thyroid nodule assessment of malignancy risk; however, these studies are subject to several limitations. Limitations and their potential clinical impacts must be addressed and, when feasible, adjusted for using valid statistical methodologies.

Molecular Cytogenetic Profiling Reveals Similarities and Differences Between Localized Nodal and Systemic Follicular Lymphomas.

Recently, we have developed novel highly promising gene expression (GE) classifiers discriminating localized nodal (LFL) from systemic follicular lymphoma (SFL) with prognostic impact. However, few data are available in LFL especially concerning hotspot genetic alterations that are associated with the pathogenesis and prognosis of SFL. A total of 144 LFL and 527 SFL, enrolled in prospective clinical trials of the German Low Grade Lymphoma Study Group, were analyzed by fluorescence in situ hybridization to detect deletions in chromosomes 1p, 6q, and 17p as well as BCL2 translocations to determine their impact on clinical outcome of LFL patients. The frequency of chromosomal deletions in 1p and 17p was comparable between LFL and SFL, while 6q deletions and BCL2 translocations more frequently occurred in SFL. A higher proportion of 1p deletions was seen in BCL2-translocation–positive LFL, compared with BCL2-translocation–negative LFL. Deletions in chromosomes 1p, 6q, and 17p predicted clinical outcome of patients with SFL in the entire cohort, while only deletions in chromosome 1p retained its negative prognostic impact in R-CHOP–treated SFL. In contrast, no deletions in one of the investigated genetic loci predicted clinical outcome in LFL. Likewise, the presence or absence of BCL2 translocations had no prognostic impact in LFL. Despite representing a genetic portfolio closely resembling SFL, LFL showed some differences in deletion frequencies. BCL2 translocation and 6q deletion frequency differs between LFL and SFL and might contribute to distinct genetic profiles in LFL and SFL.

A diagnostic classifier for gene expression-based identification of early Lyme disease

BackgroundLyme disease is a tick-borne illness that causes an estimated 476,000 infections annually in the United States. New diagnostic tests are urgently needed, as existing antibody-based assays lack sufficient sensitivity and specificity.MethodsHere we perform transcriptome profiling by RNA sequencing (RNA-Seq), targeted RNA-Seq, and/or machine learning-based classification of 263 peripheral blood mononuclear cell samples from 218 subjects, including 94 early Lyme disease patients, 48 uninfected control subjects, and 57 patients with other infections (influenza, bacteremia, or tuberculosis). Differentially expressed genes among the 25,278 in the reference database are selected based on ≥1.5-fold change, ≤0.05 p value, and ≤0.001 false-discovery rate cutoffs. After gene selection using a k-nearest neighbor algorithm, the comparative performance of ten different classifier models is evaluated using machine learning.ResultsWe identify a 31-gene Lyme disease classifier (LDC) panel that can discriminate between early Lyme patients and controls, with 23 genes (74.2%) that have previously been described in association with clinical investigations of Lyme disease patients or in vitro cell culture and rodent studies of Borrelia burgdorferi infection. Evaluation of the LDC using an independent test set of samples from 63 subjects yields an overall sensitivity of 90.0%, specificity of 100%, and accuracy of 95.2%. The LDC test is positive in 85.7% of seronegative patients and found to persist for ≥3 weeks in 9 of 12 (75%) patients.ConclusionsThese results highlight the potential clinical utility of a gene expression classifier for diagnosis of early Lyme disease, including in patients negative by conventional serologic testing.