BRCA1 secondary splice-site mutations drive exon-skipping and PARP inhibitor resistance

BRCA1 splice isoforms Δ11 and Δ11q can contribute to PARP inhibitor (PARPi) resistance by splicing-out the mutation-containing exon, producing truncated, partially-functional proteins. However, the clinical impact and underlying drivers of BRCA1 exon skipping remain undetermined. We analyzed nine ovarian and breast cancer patient derived xenografts (PDX) with BRCA1 exon 11 frameshift mutations for exon skipping and therapy response, including a matched PDX pair derived from a patient pre- and post-chemotherapy/PARPi. BRCA1 exon 11 skipping was elevated in PARPi resistant PDX tumors. Two independent PDX models acquired secondary BRCA1 splice site mutations (SSMs), predicted in silico to drive exon skipping. Predictions were confirmed using qRT-PCR, RNA sequencing, western blots and BRCA1 minigene modelling. SSMs were also enriched in post-PARPi ovarian cancer patient cohorts from the ARIEL2 and ARIEL4 clinical trials. We demonstrate that SSMs drive BRCA1 exon 11 skipping and PARPi resistance, and should be clinically monitored, along with frame-restoring secondary mutations.

CIS controls the functional polarization of GM-CSF-derived macrophages

The cytokine granulocyte-macrophage-colony stimulating factor (GM-CSF) possesses the capacity to differentiate monocytes into macrophages (MØs) with opposing functions, namely, proinflammatory M1-like MØs and immunosuppressive M2-like MØs. Despite the importance of these opposing biological outcomes, the intrinsic mechanism that regulates the functional polarization of MØs under GM-CSF signaling remains elusive. Here, we showed that GM-CSF-induced MØ polarization resulted in the expression of cytokine-inducible SH2-containing protein (CIS) and that CIS deficiency skewed the differentiation of monocytes toward immunosuppressive M2-like MØs. CIS deficiency resulted in hyperactivation of the JAK-STAT5 signaling pathway, consequently promoting downregulation of the transcription factor Interferon Regulatory Factor 8 (IRF8). Loss- and gain-of-function approaches highlighted IRF8 as a critical regulator of the M1-like polarization program. In vivo, CIS deficiency induced the differentiation of M2-like macrophages, which promoted strong Th2 immune responses characterized by the development of severe experimental asthma. Collectively, our results reveal a CIS-modulated mechanism that clarifies the opposing actions of GM-CSF in MØ differentiation and uncovers the role of GM-CSF in controlling allergic inflammation.

Epithelial-to-Mesenchymal Transition Supports Ovarian Carcinosarcoma Tumorigenesis and Confers Sensitivity to Microtubule Targeting with Eribulin

Ovarian carcinosarcoma (OCS) is an aggressive and rare tumor type with limited treatment options. OCS is hypothesized to develop via the combination theory, with a single progenitor resulting in carcinomatous and sarcomatous components, or alternatively via the conversion theory, with the sarcomatous component developing from the carcinomatous component through epithelial-to-mesenchymal transition (EMT). In this study, we analyzed DNA variants from isolated carcinoma and sarcoma components to show that OCS from 18 women is monoclonal. RNA sequencing indicated that the carcinoma components were more mesenchymal when compared with pure epithelial ovarian carcinomas, supporting the conversion theory and suggesting that EMT is important in the formation of these tumors. Preclinical OCS models were used to test the efficacy of microtubule-targeting drugs, including eribulin, which has previously been shown to reverse EMT characteristics in breast cancers and induce differentiation in sarcomas. Vinorelbine and eribulin more effectively inhibited OCS growth than standard-of-care platinum-based chemotherapy, and treatment with eribulin reduced mesenchymal characteristics and N-MYC expression in OCS patient-derived xenografts. Eribulin treatment resulted in an accumulation of intracellular cholesterol in OCS cells, which triggered a downregulation of the mevalonate pathway and prevented further cholesterol biosynthesis. Finally, eribulin increased expression of genes related to immune activation and increased the intratumoral accumulation of CD8+ T cells, supporting exploration of immunotherapy combinations in the clinic. Together, these data indicate that EMT plays a key role in OCS tumorigenesis and support the conversion theory for OCS histogenesis. Targeting EMT using eribulin could help improve OCS patient outcomes.

SIGNIFICANCE

Genomic analyses and preclinical models of ovarian carcinosarcoma support the conversion theory for disease development and indicate that microtubule inhibitors could be used to suppress EMT and stimulate antitumor immunity.

svaRetro and svaNUMT: modular packages for annotating retrotransposed transcripts and nuclear integration of mitochondrial DNA in genome sequencing data

Nuclear integration of mitochondrial genomes and retrocopied transcript insertion are biologically important but often-overlooked aspects of structural variant (SV) annotation. While tools for their detection exist, these typically rely on reanalysis of primary data using specialised detectors rather than leveraging calls from general purpose structural variant callers. Such reanalysis potentially leads to additional computational expense and does not take advantage of advances in general purpose structural variant calling. Here, we present svaRetro and svaNUMT; R packages that provide functions for annotating novel genomic events, such as nonreference retrocopied transcripts and nuclear integration of mitochondrial DNA. The packages were developed to work within the Bioconductor framework. We evaluate the performance of these packages to detect events using simulations and public benchmarking datasets, and annotate processed transcripts in a public structural variant database. svaRetro and svaNUMT provide modular, SV-caller agnostic tools for downstream annotation of structural variant calls.

From a large-scale genomic analysis of insertion sequences to insights into their regulatory roles in prokaryotes

Background

Insertion sequences (ISs) are mobile repeat sequences and most of them can copy themselves to new host genome locations, leading to genome plasticity and gene regulation in prokaryotes. In this study, we present functional and evolutionary relationships between IS and neighboring genes in a large-scale comparative genomic analysis.

Results

IS families were located in all prokaryotic phyla, with preferential occurrence of IS3, IS4, IS481, and IS5 families in Alpha-, Beta-, and Gammaproteobacteria, Actinobacteria and Firmicutes as well as in eukaryote host-associated organisms and autotrophic opportunistic pathogens. We defined the concept of the IS-Gene couple (IG), which allowed to highlight the functional and regulatory impacts of an IS on the closest gene. Genes involved in transcriptional regulation and transport activities were found overrepresented in IG. In particular, major facilitator superfamily (MFS) transporters, ATP-binding proteins and transposases raised as favorite neighboring gene functions of IS hotspots. Then, evolutionary conserved IS-Gene sets across taxonomic lineages enabled the classification of IS-gene couples into phylum, class-to-genus, and species syntenic IS-Gene couples. The IS5, IS21, IS4, IS607, IS91, ISL3 and IS200 families displayed two to four times more ISs in the phylum and/or class-to-genus syntenic IGs compared to other IS families. This indicates that those families were probably inserted earlier than others and then subjected to horizontal transfer, transposition and deletion events over time. In phylum syntenic IG category, Betaproteobacteria, Crenarchaeota, Calditrichae, Planctomycetes, Acidithiobacillia and Cyanobacteria phyla act as IS reservoirs for other phyla, and neighboring gene functions are mostly related to transcriptional regulators. Comparison of IS occurrences with predicted regulatory motifs led to ~ 26.5% of motif-containing ISs with 2 motifs per IS in average. These results, concomitantly with short IS-Gene distances, suggest that those ISs would interfere with the expression of neighboring genes and thus form strong candidates for an adaptive pairing.

Conclusions

All together, our large-scale study provide new insights into the IS genetic context and strongly suggest their regulatory roles.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-022-08678-3.

Lung Imaging Reveals Stroke-Induced Impairment in Pulmonary Intravascular Neutrophil Function, a Response Exacerbated with Aging

In stroke patients, infection is a significant contributor to morbidity and mortality. Moreover, older stroke patients show an increased risk of developing stroke-associated infection, although the mechanisms underlying this increased susceptibility to infection are unknown. In this study, using an experimental mouse model of ischemic stroke, we showed that older (12-15 mo of age) mice had elevated lung bacterial infection and inflammatory damage after stroke when compared with young (8-10 wk of age) counterparts, despite undergoing the same degree of brain injury. Intravital microscopy of the lung microvasculature revealed that in younger mice, stroke promoted neutrophil arrest in pulmonary microvessels, but this response was not seen in older poststroke mice. In addition, bacterial phagocytosis by neutrophils in the lung microvasculature was reduced by both aging and stroke, such that neutrophils in aged poststroke mice showed the greatest impairment in this function. Analysis of neutrophil migration in vitro and in the cremaster muscle demonstrated that stroke alone did not negatively impact neutrophil migration, but that the combination of increased age and stroke led to reduced effectiveness of neutrophil chemotaxis. Transcriptomic analysis of pulmonary neutrophils using RNA sequencing identified 79 genes that were selectively altered in the context of combined aging and stroke, and they were associated with pathways that control neutrophil chemotaxis. Taken together, the findings of this study show that stroke in older animals results in worsening of neutrophil antibacterial responses and changes in neutrophil gene expression that have the potential to underpin elevated risk of stroke-associated infection in the context of increased age.

Blood-Derived Extracellular Vesicle-Associated miR-3182 Detects Non-Small Cell Lung Cancer Patients

Simple Summary

Lung cancer is the leading cause of cancer-related death worldwide as patients are burdened with incredibly poor prognosis. Low survival rates are primarily attributed to lack of early detection and, therefore, timely therapeutic interventions. Late diagnosis is essentially caused by absent and non-specific symptoms, and compounded by inadequate diagnostic tools. We show here that a lung cancer biomarker, based on a simple blood test, might provide promising advantages for diagnostic assessment. Small extracellular vesicles (sEVs) are miniscule messengers that carry cancer biomarkers and are easily detected in the blood. We identify that the abundance of a specific micro-RNA, miR-3182, in these sEVs can be detected in the blood of lung cancer patients but not in controls with benign lung conditions. This demonstrates the potential use of miR-3182 as a biomarker for lung cancer diagnosis.

Abstract

With five-year survival rates as low as 3%, lung cancer is the most common cause of cancer-related mortality worldwide. The severity of the disease at presentation is accredited to the lack of early detection capacities, resulting in the reliance on low-throughput diagnostic measures, such as tissue biopsy and imaging. Interest in the development and use of liquid biopsies has risen, due to non-invasive sample collection, and the depth of information it can provide on a disease. Small extracellular vesicles (sEVs) as viable liquid biopsies are of particular interest due to their potential as cancer biomarkers. To validate the use of sEVs as cancer biomarkers, we characterised cancer sEVs using miRNA sequencing analysis. We found that miRNA-3182 was highly enriched in sEVs derived from the blood of patients with invasive breast carcinoma and NSCLC. The enrichment of sEV miR-3182 was confirmed in oncogenic, transformed lung cells in comparison to isogenic, untransformed lung cells. Most importantly, miR-3182 can successfully distinguish early-stage NSCLC patients from those with benign lung conditions. Therefore, miR-3182 provides potential to be used for the detection of NSCLC in blood samples, which could result in earlier therapy and thus improved outcomes and survival for patients.

Abstract PO026: Building the infrastructure required to research novel therapies for 35% of women with endometrial cancer who have rare subtypes

Whilst the majority of Endometrial Cancers (ECs) are common Type 1 endometrioid cancers, accounting for 75-80% of cases, with a good prognosis, Type 2 ECs include high-grade, clinically aggressive histologies, with poor response rates to hormonal therapies. Serous endometrial carcinoma is the second most common type, accounting for ~10 percent of cases, most with a p53 abnormality and a lesser proportion with HER2 overexpression/amplification/mutation. Clear cell endometrial carcinoma accounts for <5 % of EC, with the most aggressive cases also having p53 mutations. Mixed histologies and undifferentiated ECs are also aggressive. Carcinosarcoma (ECS) is a rare, aggressive, biphasic carcinoma that accounts for <5 percent of ECs, 90% with p53 abnormalities. Endometrial stromal sarcomas (ESS) can be low-grade or high-grade, or undifferentiated or resemble ovarian sex cord tumors. Uterine leiomyosarcoma (uLMS) are epithelioid or myxoid in type. Adenosarcoma involves a benign epithelial component mixed with a malignant stromal element. Whilst there have been considerable improvements in death rates for all cancers combined over the last 20 years, these improvements have not been seen for most rare cancers (RC). In order to ensure that those diagnosed with RC have access to research and novel therapies, we designed the Walter and Eliza Hall Institute of Medical Research (WEHI) Stafford Fox Rare Cancer Program (SFRCP). The WEHI-SFRCP has streamlined ethics, governance, consenting processes, including remote consent at home anywhere in Australia, and data collection protocols to allow the analysis of data and tissue of any type of RC. We have interconnected clinical and laboratory RC Databases within BioGrid Australia using the online REDCap platform. We have extensive laboratory processes in place for processing of tumour and blood samples, including the generation of PBMCs, DNA, RNA, to generate NGS including WGS; patient-derived xenografts (PDX), organoids, cell lines and other derivatives. We have high grade serous endometrial cancer (33 cases, including 4 PDX with 3 pending); uLMS (32 cases, including 2 PDX with 3 pending); endometrial carcinosarcoma (13 cases, including 3 PDX, 1 pending); rare (other, adenosarcoma, STUMP). We perform NGS testing on cases, depending on tumour purity, including Whole Genome Sequencing from fresh tumour samples; characterize PDX according to current chemotherapy and relevant novel therapeutics; study rare endometrial subtypes in specific projects and also provide information back to patients to guide therapy in the clinic. By integrating data sets with endometrial projects of similar depth we will drive forward the study of rare endometrial cancer subtypes.

Citation Format: Clare L. Scott, Ratana Lim, Amandine Carmagnac, Cassandra Vandenberg, Gayanie Ratnayake, Genevieve Dall, Joshua Tram, Justin Bedo, Jocelyn Penington, Joep Vissers, Sean Grimmond, Matthew Wakefield, Anthony Papenfuss, Holly Barker. Building the infrastructure required to research novel therapies for 35% of women with endometrial cancer who have rare subtypes [abstract]. In: Proceedings of the AACR Virtual Special Conference: Endometrial Cancer: New Biology Driving Research and Treatment; 2020 Nov 9-10. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(3_Suppl):Abstract nr PO026.

Abstract PO037: Identifying effective combinations of targeted therapies, using novel pre-clinical models, to improve treatment options for high-grade serous endometrial cancer

High-grade serous endometrial carcinoma (HGSEC) accounts for just 10% of endometrial cancer (EC) cases but is responsible for at least 40% of all EC-related deaths. It typically arises in post-menopausal women, with 70% of patients presenting with stage III or IV disease, does not respond to hormone therapy unlike the less aggressive forms of EC, and has a lower overall survival rate of just 18-27%, which has not improved over the past two decades. The primary treatment for HGSEC is surgery, followed by a combination of standard chemotherapies (platinum and taxane) with or without localised radiotherapy. However, recurrent HGSEC is less responsive to chemotherapy than are other subtypes of EC and even initial responses to chemotherapy are poor. Therefore, there is a great unmet clinical need to find better treatment options for women with this aggressive cancer. Apart from TP53 (mutated in up to 90% of cases), the other most frequently mutated genes in HGSEC are PPP2R1A (31%), PIK3CA (22%), FBXW7 (28%), CHD4 (17%) and BRCA2 (12%). Focal amplifications of the genes MYC, ERBB2, CCNE1, FGFR3 and SOX17 are also common. The presence of ERBB2 amplification and/or HER2 over-expression in around 30% of HGSEC suggests these patients may respond to HER2-targeting drugs, such as trastuzumab. However, only modest benefit has so far been seen for single-agent HER2-targeted therapies (ie trastuzumab or lapatinib) against HGSEC, suggesting resistance mechanisms are present. Another feature of HGSEC that could be exploited therapeutically is homologous recombination deficiency (HRD), which may be targeted with PARP inhibitors (PARPi). It is not clear what proportion of HGSEC are HRD and neither HER2-targeting drugs or PARPi have been approved for the treatment of HGSEC. Due to its rarity and a lack of pre-clinical models, HGSEC has so far been understudied, resulting in a lack of effective treatment options. We currently have 33 HGSEC patients consented to the WEHI-Stafford Fox Rare Cancer Program and have developed pre-clinical models from fresh patient tumour samples received (4 patient-derived xenograft (PDX) models validated, with 3 pending). Preliminary molecular analysis of whole-genome sequencing (5 samples, one of which gave rise to a PDX model), whole-exome sequencing (4 samples), and cancer panel sequencing (3 samples, 2 of which gave rise to PDX models; one harbouring ERBB2 amplification and one harbouring an AKT mutation) data from our HGSEC cohort has been performed. This has identified potential treatment targets, including ERBB2 amplifications and mutations in HR genes. I am using the PDX models for initial in vivo therapeutic characterization studies and to develop organoid models for use in high-throughput drug assays in vitro. This will guide subsequent novel drug combination testing in our PDX models. By combining specific targeted drugs I hope to overcome de novo resistance mechanisms and prevent acquired resistance. Results from this study will guide future decisions about therapeutic strategies to improve survival of women with HGSEC.

Citation Format: Holly E. Barker, Ratana Lim, Amandine Carmagnac, Cassandra Vandenberg, Gayanie Ratnayake, Genevieve Dall, Briony Milesi, Angela Komiti, Emily O'Grady, Joshua Tram, Kym Pham Stewart, Justin Bedo, Jocelyn Penington, Joep Vissers, Sean Grimmond, Matthew Wakefield, Tony Papenfuss, Clare Scott. Identifying effective combinations of targeted therapies, using novel pre-clinical models, to improve treatment options for high-grade serous endometrial cancer [abstract]. In: Proceedings of the AACR Virtual Special Conference: Endometrial Cancer: New Biology Driving Research and Treatment; 2020 Nov 9-10. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(3_Suppl):Abstract nr PO037.

Abstract PO021: Developing pre-clinical models of uterine leiomyosarcoma

Uterine sarcomas make up 1-4% of uterine malignancies. Of these 60% are classified as leiomyosarcoma (uLMS). The 5-year survival rate of uLMS is 35-65.2% for tumours that have not spread beyond the uterus. However, women are often diagnosed at a late stage due to a lack of screening options by which time the tumour has often spread to adjacent and distant tissues. Current standard of care for uLMS patients is surgical de-bulking followed by adjuvant chemotherapy, but significant improvement in progression free survival and overall survival is not consistently observed. The lack of advances for the treatment and screening of uLMS is due in part to the scarcity of appropriate research resources for this rare disease. Genetic analyses have been performed but on relatively small samples and there are just 14 reported patient-derived xenograft (PDX) models of uLMS in the literature to date. Through the WEHI Stafford Fox Rare Cancer Program, as well as collaborations (facilitated by ANZGOG) throughout the country and internationally, we have access to a large biobank of uLMS tissue. We have received 8 fresh uLMS samples in the laboratory, 2 of which have established PDX lines that were validated as uLMS by our anatomical pathologist. All fresh samples received into the laboratory are snap frozen for whole-genome sequencing as well as viably frozen to enable regeneration of the tissue for future applications. One application is organoid culturing, which allows the tissue to retain its 3D growth properties and is significantly cheaper than growing tissue as a PDX. Organoid culturing also allows for higher throughput of samples in drug screening assays, enabling us to fast-track the selection of drugs for validation in our PDX models. In addition to these fresh samples we also have 23 archival uLMS samples (formalin fixed, paraffin embedded) that can be used for lower coverage genetic analysis, and protein expression by immunohistochemistry. This unique biobank of uLMS tissue is the first of its kind in Australia and with it we will endeavour to gain a comprehensive understanding of this disease. Through our PDX modelling we also have the opportunity to predict resistance to therapy and test emerging therapies in a clinically relevant context. We believe this biobank will provide a critical resource which will ultimately lead to better outcomes for uLMS patients.

Citation Format: Genevieve Dall, Cassandra Vandenberg, Amandine Carmagnac, Ratana Lim, Briony Milesi, Angela Komiti, Emily O'Grady, Joshua Tram, Gayanie Ratnayake, Kym Pham Stewart, Justin Bedo, Jocelyn Penington, Joep Vissers, Inger Olesen, Sean Grimmond, Holly Barker, Tony Papenfuss, Clare Scott. Developing pre-clinical models of uterine leiomyosarcoma [abstract]. In: Proceedings of the AACR Virtual Special Conference: Endometrial Cancer: New Biology Driving Research and Treatment; 2020 Nov 9-10. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(3_Suppl):Abstract nr PO021.

Improved HLA-based prediction of coeliac disease identifies two novel genetic interactions

Human Leucocyte Antigen (HLA) testing is useful in the clinical work-up of coeliac disease (CD) with high negative but low positive predictive value. We construct a genomic risk score (GRS) using HLA risk genotypes to improve CD prediction and guide exclusion criteria. Imputed HLA genotypes for five European CD case-control GWAS (n > 15,000) were used to construct and validate an interpretable HLA-based risk model (HDQ₁₅), which shows statistically significant improvements in predictive performance upon all previous HLA-based risk models. Conditioning on this model, we find two novel associations, HLA-DQ6.2 and HLA-DQ7.3, that interact significantly with HLA-DQ2.5 (p = 2.51 × 10^-9, 1.99 × 10^-7, respectively). Integrating these novel alleles into a new risk model (HDQ₁₇) leads to predictive performance equivalent or better than the strongest reported GRS (GRS₂₂₈) using 228 single nucleotide polymorphisms (SNPs). We also demonstrate that our proposed HLA-based models can be implemented using only six HLA tagging SNPs with statistically equivalent predictive performance. Using insights from our model to guide exclusionary criteria, we find the positive predictive value of CD testing in high-risk populations can be increased by 55%, from 17.5 to 27.1%, while maintaining a negative predictive value above 99%. Our results suggest that HLA typing is currently undervalued in CD assessment.

MTOR signaling orchestrates stress-induced mutagenesis, facilitating adaptive evolution in cancer

In microorganisms, evolutionarily conserved mechanisms facilitate adaptation to harsh conditions through stress-induced mutagenesis (SIM). Analogous processes may underpin progression and therapeutic failure in human cancer. We describe SIM in multiple in vitro and in vivo models of human cancers under nongenotoxic drug selection, paradoxically enhancing adaptation at a competing intrinsic fitness cost. A genome-wide approach identified the mechanistic target of rapamycin (MTOR) as a stress-sensing rheostat mediating SIM across multiple cancer types and conditions. These observations are consistent with a two-phase model for drug resistance, in which an initially rapid expansion of genetic diversity is counterbalanced by an intrinsic fitness penalty, subsequently normalizing to complete adaptation under the new conditions. This model suggests synthetic lethal strategies to minimize resistance to anticancer therapy.

RAR/RXR binding dynamics distinguish pluripotency from differentiation associated cis-regulatory elements

In mouse embryonic cells, ligand-activated retinoic acid receptors (RARs) play a key role in inhibiting pluripotency-maintaining genes and activating some major actors of cell differentiation. To investigate the mechanism underlying this dual regulation, we performed joint RAR/RXR ChIP-seq and mRNA-seq time series during the first 48 h of the RA-induced Primitive Endoderm (PrE) differentiation process in F9 embryonal carcinoma (EC) cells. We show here that this dual regulation is associated with RAR/RXR genomic redistribution during the differentiation process. In-depth analysis of RAR/RXR binding sites occupancy dynamics and composition show that in undifferentiated cells, RAR/RXR interact with genomic regions characterized by binding of pluripotency-associated factors and high prevalence of the non-canonical DR0-containing RA response element. By contrast, in differentiated cells, RAR/RXR bound regions are enriched in functional Sox17 binding sites and are characterized with a higher frequency of the canonical DR5 motif. Our data offer an unprecedentedly detailed view on the action of RA in triggering pluripotent cell differentiation and demonstrate that RAR/RXR action is mediated via two different sets of regulatory regions tightly associated with cell differentiation status.

Blood-based protein biomarkers for diagnosis of Alzheimer disease

OBJECTIVE

To identify plasma biomarkers for the diagnosis of Alzheimer disease (AD).

DESIGN

Baseline plasma screening of 151 multiplexed analytes combined with targeted biomarker and clinical pathology data.

SETTING

General community-based, prospective, longitudinal study of aging.

PARTICIPANTS

A total of 754 healthy individuals serving as controls and 207 participants with AD from the Australian Imaging Biomarker and Lifestyle study (AIBL) cohort with identified biomarkers that were validated in 58 healthy controls and 112 individuals with AD from the Alzheimer Disease Neuroimaging Initiative (ADNI) cohort.

RESULTS

A biomarker panel was identified that included markers significantly increased (cortisol, pancreatic polypeptide, insulinlike growth factor binding protein 2, β(2) microglobulin, vascular cell adhesion molecule 1, carcinoembryonic antigen, matrix metalloprotein 2, CD40, macrophage inflammatory protein 1α, superoxide dismutase, and homocysteine) and decreased (apolipoprotein E, epidermal growth factor receptor, hemoglobin, calcium, zinc, interleukin 17, and albumin) in AD. Cross-validated accuracy measures from the AIBL cohort reached a mean (SD) of 85% (3.0%) for sensitivity and specificity and 93% (3.0) for the area under the receiver operating characteristic curve. A second validation using the ADNI cohort attained accuracy measures of 80% (3.0%) for sensitivity and specificity and 85% (3.0) for area under the receiver operating characteristic curve.

CONCLUSIONS

This study identified a panel of plasma biomarkers that distinguish individuals with AD from cognitively healthy control subjects with high sensitivity and specificity. Cross-validation within the AIBL cohort and further validation within the ADNI cohort provides strong evidence that the identified biomarkers are important for AD diagnosis.

The poisson margin test for normalization-free significance analysis of NGS data

The current methods for the determination of the statistical significance of peaks and regions in next generation sequencing (NGS) data require an explicit normalization step to compensate for (global or local) imbalances in the sizes of sequenced and mapped libraries. There are no canonical methods for performing such compensations; hence, a number of different procedures serving this goal in different ways can be found in the literature. Unfortunately, the normalization has a significant impact on the final results. Different methods yield very different numbers of detected "significant peaks" even in the simplest scenario of ChIP-Seq experiments that compare the enrichment in a single sample relative to a matching control. This becomes an even more acute issue in the more general case of the comparison of multiple samples, where a number of arbitrary design choices will be required in the data analysis stage, each option resulting in possibly (significantly) different outcomes. In this article, we investigate a principled statistical procedure that eliminates the need for a normalization step. We outline its basic properties, in particular the scaling upon depth of sequencing. For the sake of illustration and comparison, we report the results of re-analyzing a ChIP-Seq experiment for transcription factor binding site detection. In order to quantify the differences between outcomes, we use a novel method based on the accuracy of in silico prediction by support vector machine (SVM) models trained on part of the genome and tested on the remainder. See Kowalczyk et al. ( 2009 ) for supplementary material.

Precision-mapping and statistical validation of quantitative trait loci by machine learning

Background

We introduce a QTL-mapping algorithm based on Statistical Machine Learning (SML) that is conceptually quite different to existing methods as there is a strong focus on generalisation ability. Our approach combines ridge regression, recursive feature elimination, and estimation of generalisation performance and marker effects using bootstrap resampling. Model performance and marker effects are determined using independent testing samples (individuals), thus providing better estimates. We compare the performance of SML against Composite Interval Mapping (CIM), Bayesian Interval Mapping (BIM) and single Marker Regression (MR) on synthetic datasets and a multi-trait and multi-environment dataset of the progeny for a cross between two barley cultivars.

Results

In an analysis of the synthetic datasets, SML accurately predicted the number of QTL underlying a trait while BIM tended to underestimate the number of QTL. The QTL identified by SML for the barley dataset broadly coincided with known QTL locations. SML reported approximately half of the QTL reported by either CIM or MR, not unexpected given that neither CIM nor MR incorporates independent testing. The latter makes these two methods susceptible to producing overly optimistic estimates of QTL effects, as we demonstrate for MR. The QTL resolution (peak definition) afforded by SML was consistently superior to MR, CIM and BIM, with QTL detection power similar to BIM. The precision of SML was underscored by repeatedly identifying, at ≤ 1-cM precision, three QTL for four partially related traits (heading date, plant height, lodging and yield). The set of QTL obtained using a 'raw' and a 'curated' version of the same genotypic dataset were more similar to each other for SML than for CIM or MR.

Conclusion

The SML algorithm produces better estimates of QTL effects because it eliminates the optimistic bias in the predictive performance of other QTL methods. It produces narrower peaks than other methods (except BIM) and hence identifies QTL with greater precision. It is more robust to genotyping and linkage mapping errors, and identifies markers linked to QTL in the absence of a genetic map.

Stratification bias in low signal microarray studies

BACKGROUND

When analysing microarray and other small sample size biological datasets, care is needed to avoid various biases. We analyse a form of bias, stratification bias, that can substantially affect analyses using sample-reuse validation techniques and lead to inaccurate results. This bias is due to imperfect stratification of samples in the training and test sets and the dependency between these stratification errors, i.e. the variations in class proportions in the training and test sets are negatively correlated.

RESULTS

We show that when estimating the performance of classifiers on low signal datasets (i.e. those which are difficult to classify), which are typical of many prognostic microarray studies, commonly used performance measures can suffer from a substantial negative bias. For error rate this bias is only severe in quite restricted situations, but can be much larger and more frequent when using ranking measures such as the receiver operating characteristic (ROC) curve and area under the ROC (AUC). Substantial biases are shown in simulations and on the van 't Veer breast cancer dataset. The classification error rate can have large negative biases for balanced datasets, whereas the AUC shows substantial pessimistic biases even for imbalanced datasets. In simulation studies using 10-fold cross-validation, AUC values of less than 0.3 can be observed on random datasets rather than the expected 0.5. Further experiments on the van 't Veer breast cancer dataset show these biases exist in practice.

CONCLUSION

Stratification bias can substantially affect several performance measures. In computing the AUC, the strategy of pooling the test samples from the various folds of cross-validation can lead to large biases; computing it as the average of per-fold estimates avoids this bias and is thus the recommended approach. As a more general solution applicable to other performance measures, we show that stratified repeated holdout and a modified version of k-fold cross-validation, balanced, stratified cross-validation and balanced leave-one-out cross-validation, avoids the bias. Therefore for model selection and evaluation of microarray and other small biological datasets, these methods should be used and unstratified versions avoided. In particular, the commonly used (unbalanced) leave-one-out cross-validation should not be used to estimate AUC for small datasets.

Gene selection via the BAHSIC family of algorithms

MOTIVATION

Identifying significant genes among thousands of sequences on a microarray is a central challenge for cancer research in bioinformatics. The ultimate goal is to detect the genes that are involved in disease outbreak and progression. A multitude of methods have been proposed for this task of feature selection, yet the selected gene lists differ greatly between different methods. To accomplish biologically meaningful gene selection from microarray data, we have to understand the theoretical connections and the differences between these methods. In this article, we define a kernel-based framework for feature selection based on the Hilbert-Schmidt independence criterion and backward elimination, called BAHSIC. We show that several well-known feature selectors are instances of BAHSIC, thereby clarifying their relationship. Furthermore, by choosing a different kernel, BAHSIC allows us to easily define novel feature selection algorithms. As a further advantage, feature selection via BAHSIC works directly on multiclass problems.

RESULTS

In a broad experimental evaluation, the members of the BAHSIC family reach high levels of accuracy and robustness when compared to other feature selection techniques. Experiments show that features selected with a linear kernel provide the best classification performance in general, but if strong non-linearities are present in the data then non-linear kernels can be more suitable.

AVAILABILITY

Accompanying homepage is http://www.dbs.ifi.lmu.de/~borgward/BAHSIC.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.