Improving accuracy of GPT-3/4 results on biomedical data using a retrieval-augmented language model.

Large language models (LLMs) have made a significant impact on the fields of general artificial intelligence. General purpose LLMs exhibit strong logic and reasoning skills and general world knowledge but can sometimes generate misleading results when prompted on specific subject areas. LLMs trained with domain-specific knowledge can reduce the generation of misleading information (i.e. hallucinations) and enhance the precision of LLMs in specialized contexts. Training new LLMs on specific corpora however can be resource intensive. Here we explored the use of a retrieval-augmented generation (RAG) model which we tested on literature specific to a biomedical research area. OpenAI's GPT-3.5, GPT-4, Microsoft's Prometheus, and a custom RAG model were used to answer 19 questions pertaining to diffuse large B-cell lymphoma (DLBCL) disease biology and treatment. Eight independent reviewers assessed LLM responses based on accuracy, relevance, and readability, rating responses on a 3-point scale for each category. These scores were then used to compare LLM performance. The performance of the LLMs varied across scoring categories. On accuracy and relevance, the RAG model outperformed other models with higher scores on average and the most top scores across questions. GPT-4 was more comparable to the RAG model on relevance versus accuracy. By the same measures, GPT-4 and GPT-3.5 had the highest scores for readability of answers when compared to the other LLMs. GPT-4 and 3.5 also had more answers with hallucinations than the other LLMs, due to non-existent references and inaccurate responses to clinical questions. Our findings suggest that an oncology research-focused RAG model may outperform general-purpose LLMs in accuracy and relevance when answering subject-related questions. This framework can be tailored to Q&A in other subject areas. Further research will help understand the impact of LLM architectures, RAG methodologies, and prompting techniques in answering questions across different subject areas.

HLA-DQA1*05 Not Associated With Ustekinumab Loss of Response and Antidrug Antibodies in Ulcerative Colitis and Crohn's Disease Patients.

It is unknown if inflammatory bowel disease patients carrying the HLA-DQA1*05 allele have a greater risk of immunogenicity to ustekinumab. We observed that ustekinumab-treated patients carrying the allele did not have increased immunogenicity or reduced efficacy/serum ustekinumab concentrations vs noncarriers.

Assessing reproducibility of inherited variants detected with short-read whole genome sequencing.

BACKGROUND: Reproducible detection of inherited variants with whole genome sequencing (WGS) is vital for the implementation of precision medicine and is a complicated process in which each step affects variant call quality. Systematically assessing reproducibility of inherited variants with WGS and impact of each step in the process is needed for understanding and improving quality of inherited variants from WGS. RESULTS: To dissect the impact of factors involved in detection of inherited variants with WGS, we sequence triplicates of eight DNA samples representing two populations on three short-read sequencing platforms using three library kits in six labs and call variants with 56 combinations of aligners and callers. We find that bioinformatics pipelines (callers and aligners) have a larger impact on variant reproducibility than WGS platform or library preparation. Single-nucleotide variants (SNVs), particularly outside difficult-to-map regions, are more reproducible than small insertions and deletions (indels), which are least reproducible when > 5 bp. Increasing sequencing coverage improves indel reproducibility but has limited impact on SNVs above 30×. CONCLUSIONS: Our findings highlight sources of variability in variant detection and the need for improvement of bioinformatics pipelines in the era of precision medicine with WGS.

Toward best practice in cancer mutation detection with whole-genome and whole-exome sequencing.

Clinical applications of precision oncology require accurate tests that can distinguish true cancer-specific mutations from errors introduced at each step of next-generation sequencing (NGS). To date, no bulk sequencing study has addressed the effects of cross-site reproducibility, nor the biological, technical and computational factors that influence variant identification. Here we report a systematic interrogation of somatic mutations in paired tumor-normal cell lines to identify factors affecting detection reproducibility and accuracy at six different centers. Using whole-genome sequencing (WGS) and whole-exome sequencing (WES), we evaluated the reproducibility of different sample types with varying input amount and tumor purity, and multiple library construction protocols, followed by processing with nine bioinformatics pipelines. We found that read coverage and callers affected both WGS and WES reproducibility, but WES performance was influenced by insert fragment size, genomic copy content and the global imbalance score (GIV; G > T/C > A). Finally, taking into account library preparation protocol, tumor content, read coverage and bioinformatics processes concomitantly, we recommend actionable practices to improve the reproducibility and accuracy of NGS experiments for cancer mutation detection.

Genome-wide association study of vitamin D concentrations and bone mineral density in the African American-Diabetes Heart Study.

Relative to European Americans, African Americans have lower 25-hydroxyvitamin D (25OHD) and vitamin D binding protein (VDBP) concentrations, higher 1,25-dihydroxyvitamin D (1,25(OH)2D3) concentrations and bone mineral density (BMD), and paradoxically reduced burdens of calcified atherosclerotic plaque (subclinical atherosclerosis). To identify genetic factors contributing to vitamin D and BMD measures, association analysis of >14M variants was conducted in a maximum of 697 African American-Diabetes Heart Study participants with type 2 diabetes (T2D). The most significant association signals were detected for VDBP on chromosome 4; variants rs7041 (ß = 0.44, SE = 0.019, P = 9.4x10-86) and rs4588 (ß = 0.17, SE = 0.021, P = 3.5x10-08) in the group-specific component (vitamin D binding protein) gene (GC). These variants were found to be independently associated. In addition, rs7041 was also associated with bioavailable vitamin D (BAVD; ß = 0.16, SE = 0.02, P = 3.3x10-19). Six rare variants were significantly associated with 25OHD, including a non-synonymous variant in HSPG2 (rs116788687; ß = -1.07, SE = 0.17, P = 2.2x10-10) and an intronic variant in TNIK (rs143555701; ß = -1.01, SE = 0.18, P = 9.0x10-10), both biologically related to bone development. Variants associated with 25OHD failed to replicate in African Americans from the Insulin Resistance Atherosclerosis Family Study (IRASFS). Evaluation of vitamin D metabolism and bone mineral density phenotypes in an African American population enriched for T2D could provide insight into ethnic specific differences in vitamin D metabolism and bone mineral density.

A verified genomic reference sample for assessing performance of cancer panels detecting small variants of low allele frequency.

BACKGROUND: Oncopanel genomic testing, which identifies important somatic variants, is increasingly common in medical practice and especially in clinical trials. Currently, there is a paucity of reliable genomic reference samples having a suitably large number of pre-identified variants for properly assessing oncopanel assay analytical quality and performance. The FDA-led Sequencing and Quality Control Phase 2 (SEQC2) consortium analyze ten diverse cancer cell lines individually and their pool, termed Sample A, to develop a reference sample with suitably large numbers of coding positions with known (variant) positives and negatives for properly evaluating oncopanel analytical performance. RESULTS: In reference Sample A, we identify more than 40,000 variants down to 1% allele frequency with more than 25,000 variants having less than 20% allele frequency with 1653 variants in COSMIC-related genes. This is 5-100× more than existing commercially available samples. We also identify an unprecedented number of negative positions in coding regions, allowing statistical rigor in assessing limit-of-detection, sensitivity, and precision. Over 300 loci are randomly selected and independently verified via droplet digital PCR with 100% concordance. Agilent normal reference Sample B can be admixed with Sample A to create new samples with a similar number of known variants at much lower allele frequency than what exists in Sample A natively, including known variants having allele frequency of 0.02%, a range suitable for assessing liquid biopsy panels. CONCLUSION: These new reference samples and their admixtures provide superior capability for performing oncopanel quality control, analytical accuracy, and validation for small to large oncopanels and liquid biopsy assays.

Genome-wide association study identifies novel loci for type 2 diabetes-attributed end-stage kidney disease in African Americans.

BACKGROUND: End-stage kidney disease (ESKD) is a significant public health concern disproportionately affecting African Americans (AAs). Type 2 diabetes (T2D) is the leading cause of ESKD in the USA, and efforts to uncover genetic susceptibility to diabetic kidney disease (DKD) have had limited success. A prior genome-wide association study (GWAS) in AAs with T2D-ESKD was expanded with additional AA cases and controls and genotypes imputed to the higher density 1000 Genomes reference panel. The discovery analysis included 3432 T2D-ESKD cases and 6977 non-diabetic non-nephropathy controls (N = 10,409), followed by a discrimination analysis in 2756 T2D non-nephropathy controls to exclude T2D-associated variants. RESULTS: Six independent variants located in or near RND3/RBM43, SLITRK3, ENPP7, GNG7, and APOL1 achieved genome-wide significant association (P < 5 × 10-8) with T2D-ESKD. Following extension analyses in 1910 non-diabetic ESKD cases and 908 non-diabetic non-nephropathy controls, a meta-analysis of 5342 AA all-cause ESKD cases and 6977 AA non-diabetic non-nephropathy controls revealed an additional novel all-cause ESKD locus at EFNB2 (rs77113398; P = 9.84 × 10-9; OR = 1.94). Exclusion of APOL1 renal-risk genotype carriers identified two additional genome-wide significant T2D-ESKD-associated loci at GRAMD3 and MGAT4C. A second variant at GNG7 (rs373971520; P = 2.17 × 10-8, OR = 1.46) remained associated with all-cause ESKD in the APOL1-negative analysis. CONCLUSIONS: Findings provide further evidence for genetic factors associated with advanced kidney disease in AAs with T2D.

Natural language processing and recurrent network models for identifying genomic mutation-associated cancer treatment change from patient progress notes.

OBJECTIVES: Natural language processing (NLP) and machine learning approaches were used to build classifiers to identify genomic-related treatment changes in the free-text visit progress notes of cancer patients. METHODS: We obtained 5889 deidentified progress reports (2439 words on average) for 755 cancer patients who have undergone a clinical next generation sequencing (NGS) testing in Wake Forest Baptist Comprehensive Cancer Center for our data analyses. An NLP system was implemented to process the free-text data and extract NGS-related information. Three types of recurrent neural network (RNN) namely, gated recurrent unit, long short-term memory (LSTM), and bidirectional LSTM (LSTM_Bi) were applied to classify documents to the treatment-change and no-treatment-change groups. Further, we compared the performances of RNNs to 5 machine learning algorithms including Naive Bayes, K-nearest Neighbor, Support Vector Machine for classification, Random forest, and Logistic Regression. RESULTS: Our results suggested that, overall, RNNs outperformed traditional machine learning algorithms, and LSTM_Bi showed the best performance among the RNNs in terms of accuracy, precision, recall, and F1 score. In addition, pretrained word embedding can improve the accuracy of LSTM by 3.4% and reduce the training time by more than 60%. DISCUSSION AND CONCLUSION: NLP and RNN-based text mining solutions have demonstrated advantages in information retrieval and document classification tasks for unstructured clinical progress notes.

An Exome-wide Association Study for Type 2 Diabetes-Attributed End-Stage Kidney Disease in African Americans.

INTRODUCTION: Compared with European Americans, African Americans (AAs) are at higher risk for developing end-stage kidney disease (ESKD). Genome-wide association studies (GWAS) have identified >70 genetic variants associated with kidney function and chronic kidney disease (CKD) in patients with and without diabetes. However, these variants explain a small proportion of disease liability. This study examined the contribution of coding genetic variants for risk of type 2 diabetes (T2D)-attributed ESKD and advanced CKD in AAs. METHODS: Exome sequencing was performed in 456 AA T2D-ESKD cases, and 936 AA nondiabetic, non-nephropathy control individuals at the discovery stage. A mixed logistic regression model was used for association analysis. Nominal associations (P < 0.05) were replicated in an additional 2020 T2D-ESKD cases and 1121 nondiabetic, non-nephropathy control individuals. A meta-analysis combining 4533 discovery and replication samples was performed. Putative T2D-ESKD associations were tested in additional 1910 nondiabetic ESKD and 219 T2D-ESKD cases, as well as 912 AA nondiabetic non-nephropathy control individuals. RESULTS: A total of 11 suggestive T2D-ESKD associations (P < 1 x 10-4) from 8 loci (PLEKHN1, NADK, RAD51AP2, RREB1, PEX6, GRM8, PRX, APOL1) were apparent in the meta-analysis. Exclusion of APOL1 renal-risk genotype carriers identified 3 additional suggestive loci (OTUD7B, IFITM3, DLGAP5). Rs41302867 in RREB1 displayed consistent association with T2D-ESKD and nondiabetic ESKD (odds ratio: 0.47; P = 1.2 x 10-6 in 4605 all-cause ESKD and 2969 nondiabetic non-nephropathy control individuals). CONCLUSION: Our findings suggest that coding genetic variants are implicated in predisposition to T2D-ESKD in AAs.

Genome-wide association studies suggest that APOL1-environment interactions more likely trigger kidney disease in African Americans with nondiabetic nephropathy than strong APOL1-second gene interactions.

African Americans carrying two apolipoprotein L1 gene (APOL1) renal risk variants have a high risk for nephropathy. However, only a minority develops end-stage renal disease (ESRD). Hence, modifying factors likely contribute to initiation of kidney disease such as endogenous (HIV infection) or exogenous (interferon treatment) environmental modifiers. In this report, genome-wide association studies and a meta-analysis were performed to identify novel loci for nondiabetic ESRD in African Americans and to detect genetic modifiers in APOL1-associated nephropathy. Two African American cohorts were analyzed, 1749 nondiabetic ESRD cases and 1136 controls from Wake Forest and 901 lupus nephritis (LN)-ESRD cases and 520 controls with systemic lupus erythematosus but lacking nephropathy from the LN-ESRD Consortium. Association analyses adjusting for APOL1 G1/G2 renal-risk variants were completed and stratified by APOL1 risk genotype status. Individual genome-wide association studies and meta-analysis results of all 2650 ESRD cases and 1656 controls did not detect significant genome-wide associations with ESRD beyond APOL1. Similarly, no single nucleotide polymorphism showed significant genome-wide evidence of an interaction with APOL1 risk variants. Thus, although variants with small individual effects cannot be ruled out and are likely to exist, our results suggest that APOL1-environment interactions may be of greater clinical importance in triggering nephropathy in African Americans than APOL1 interactions with other single nucleotide polymorphisms.

Genome-wide interaction with the insulin secretion locus MTNR1B reveals CMIP as a novel type 2 diabetes susceptibility gene in African Americans.

Although type 2 diabetes (T2D) results from metabolic defects in insulin secretion and insulin sensitivity, most of the genetic risk loci identified to date relates to insulin secretion. We reported that T2D loci influencing insulin sensitivity may be identified through interactions with insulin secretion loci, thereby leading to T2D. Here, we hypothesize that joint testing of variant main effects and interaction effects with an insulin secretion locus increases power to identify genetic interactions leading to T2D. We tested this hypothesis with an intronic MTNR1B SNP, rs10830963, which is associated with acute insulin response to glucose, a dynamic measure of insulin secretion. rs10830963 was tested for interaction and joint (main + interaction) effects with genome-wide data in African Americans (2,452 cases and 3,772 controls) from five cohorts. Genome-wide genotype data (Affymetrix Human Genome 6.0 array) was imputed to a 1000 Genomes Project reference panel. T2D risk was modeled using logistic regression with rs10830963 dosage, age, sex, and principal component as predictors. Joint effects were captured using the Kraft two degrees of freedom test. Genome-wide significant (P < 5 × 10-8 ) interaction with MTNR1B and joint effects were detected for CMIP intronic SNP rs17197883 (Pinteraction  = 1.43 × 10-8 ; Pjoint  = 4.70 × 10-8 ). CMIP variants have been nominally associated with T2D, fasting glucose, and adiponectin in individuals of East Asian ancestry, with high-density lipoprotein, and with waist-to-hip ratio adjusted for body mass index in Europeans. These data support the hypothesis that additional genetic factors contributing to T2D risk, including insulin sensitivity loci, can be identified through interactions with insulin secretion loci.

Association of kidney structure-related gene variants with type 2 diabetes-attributed end-stage kidney disease in African Americans.

African Americans (AAs) are at higher risk for developing end-stage kidney disease (ESKD) compared to European Americans. Genome-wide association studies have identified variants associated with diabetic and non-diabetic kidney diseases. Nephropathy loci, including SLC7A9, UMOD, and SHROOM3, have been implicated in the maintenance of normal glomerular and renal tubular structure and function. Herein, 47 genes important in podocyte, glomerular basement membrane, mesangial cell, mesangial matrix, renal tubular cell, and renal interstitium structure were examined for association with type 2 diabetes (T2D)-attributed ESKD in AAs. Single-variant association analysis was performed in the discovery stage, including 2041 T2D-ESKD cases and 1140 controls (non-diabetic, non-nephropathy). Discrimination analyses in 667 T2D cases-lacking nephropathy excluded T2D-associated SNPs. Nominal associations were tested in an additional 483 T2D-ESKD cases and 554 controls in the replication stage. Meta-analysis of 4218 discovery and replication samples revealed three significant associations with T2D-ESKD at CD2AP and MMP2 (P corr < 0.05 corrected for effective number of SNPs in each locus). Removal of APOL1 renal-risk genotype carriers revealed additional association at five loci, TTC21B, COL4A3, NPHP3-ACAD11, CLDN8, and ARHGAP24 (P corr < 0.05). Genetic variants at COL4A3, CLDN8, and ARHGAP24 were potentially pathogenic. Gene-based associations revealed suggestive significant aggregate effects of coding variants at four genes. Our findings suggest that genetic variation in kidney structure-related genes may contribute to T2D-attributed ESKD in the AA population.

Association Analysis of the Cubilin (CUBN) and Megalin (LRP2) Genes with ESRD in African Americans.

BACKGROUND AND OBJECTIVES: Genetic variation in the cubilin (CUBN) gene is associated with albuminuria and CKD. Common and rare coding variants in CUBN and the gene encoding its transport partner megalin (LRP2) were assessed for association with ESRD in blacks. DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS: Sixty-six CUBN and LRP2 single-nucleotide polymorphisms (SNPs) were selected and analyzed in this multistage study. Exome sequencing data from 529 blacks with type 2 diabetes (T2D) -associated ESRD and 535 controls lacking T2D or nephropathy (the Type 2 Diabetes Genes [T2D-GENES] Consortium) were first evaluated, focusing on coding variants in CUBN and LRP2; 15 potentially associated SNPs identified from the T2D-GENES Consortium as well as 51 other selected SNPs were then assessed in an independent T2D-ESRD sample set of blacks (the Affymetrix Axiom Biobank Genotyping Array [AXIOM]; 2041 patients with T2D-ESRD, 627 patients with T2D without nephropathy, and 1140 nondiabetic, non-nephropathy controls). A meta-analysis combining the T2D-GENES Consortium and the AXIOM data was performed for 18 overlapping SNPs. Additionally, all 66 SNPs were genotyped in the Wake Forest School of Medicine samples of blacks with nondiabetic ESRD (885 patients with nondiabetic ESRD and 721 controls). Association testing with ESRD was performed in models including age, sex, African ancestry proportion, and apolipoprotein L1 gene renal-risk variants. RESULTS: CUBN SNP rs1801239 (I2984V), previously associated with albuminuria, was significantly associated with T2D-ESRD in blacks (the T2D-GENES Consortium and the AXIOM meta-analysis, P=0.03; odds ratio, 1.31; 95% confidence interval, 1.03 to 1.67; minor allele frequency =0.028). A novel LRP2 missense variant, rs17848169 (N2632D), was also significantly protective from T2D-ESRD (the T2D-GENES Consortium and the AXIOM, P<0.002; odds ratio, 0.47; 95% confidence interval, 0.29 to 0.75; meta-analysis minor allele frequency =0.007). Neither SNP was associated with T2D when contrasting patients with T2D with controls lacking diabetes. CUBN and LRP2 SNPs were not associated with nondiabetic etiologies of ESRD. CONCLUSIONS: Evidence for genetic association exists between a cubilin and a rare megalin variant with diabetes-associated ESRD in populations with recent African ancestry.

Deceased donor multidrug resistance protein 1 and caveolin 1 gene variants may influence allograft survival in kidney transplantation.

Variants in donor multidrug resistance protein 1 (ABCB1) and caveolin 1 (CAV1) genes are associated with renal allograft failure after transplantation in Europeans. Here we assessed transplantation outcomes of kidneys from 368 African American (AA) and 314 European American (EA) deceased donors based on 38 single-nucleotide polymorphisms (SNPs) spanning ABCB1 and 16 SNPs spanning CAV1, including previously associated index and haplotype-tagging SNPs. Tests for association with time to allograft failure were performed for the 1233 resultant kidney transplantations, adjusting for recipient age, sex, ethnicity, cold ischemia time, panel reactive antibody, human leukocyte antigen match, expanded-criteria donation, and APOL1-nephropathy variants in AA donors. Interaction analyses between APOL1 with ABCB1 and CAV1 were performed. In a meta-analysis of all transplantations, ABCB1 index SNP rs1045642 was associated with time to allograft failure and other ABCB1 SNPs were nominally associated, but not CAV1 SNPs. ABCB1 SNP rs1045642 showed consistent effects with the 558 transplantations from EA donors, but not with the 675 transplantations from AA donors. ABCB1 SNP rs956825 and CAV1 SNP rs6466583 interacted with APOL1 in transplants from AA donors. Thus, the T allele at ABCB1 rs1045642 is associated with shorter renal allograft survival for kidneys from American donors. Interactions between ABCB1 and CAV1 with APOL1 may influence allograft failure for transplanted kidneys from AA donors.

The ras responsive transcription factor RREB1 is a novel candidate gene for type 2 diabetes associated end-stage kidney disease.

Familial clustering and presumed genetic risk for type 2 diabetic (T2D) and non-diabetic end-stage kidney disease (ESKD) appear strong in African Americans. Examination of exome sequencing data in African American T2D-ESKD cases and non-diabetic non-nephropathy controls identified two low-frequency variants in the RREB1 gene, a repressor of the angiotensinogen (AGT) gene previously associated with kidney function, as being associated with T2D-ESKD: rs9379084 (P = 0.00087, OR = 0.26; D1171N) and rs41302867 (P = 0.00078, OR = 0.21; splice site variant). Rs41302867 replicated association in an independent sample of African Americans with T2D-ESKD [rs41302867 P = 0.033 (OR = 0.50)], and a trend towards rs9379084 association was observed (P = 0.070). In European Americans with T2D-ESKD compared with European American population based controls, both RREB1 variants replicated association [rs9379084 P = 1.67 × 10(-4) (OR = 0.54) and rs41302867 P = 0.013 (OR = 0.69)]. Rs9379084 was not associated with non-T2D-ESKD or T2D in African Americans (P = 0.55 and P = 0.37, respectively), but was associated with T2D in European Americans (P = 0.014, OR = 0.65). In African Americans, rs41302867 was associated with non-T2D-ESKD [P = 0.036 (OR = 0.54)] and hypertension attributed ESKD [H-ESKD, P = 0.029 (OR = 0.50)]. A meta-analysis combining African American and European American T2D-ESKD data revealed P = 3.52 × 10(-7) and 3.70 × 10(-5) for rs9379084 and rs41302867 association, respectfully. A locus-wide analysis evaluating putatively functional SNPs revealed several nominal associations with T2D-ESKD, non-T2D-ESKD and T2D in African and European Americans. RREB1 is a large, complex gene which codes a multidomain zinc finger binding protein and transcription factor. We posit that variants in RREB1 modulate seemingly disparate phenotypes (i.e. T2D, T2D-ESKD and non-T2D-ESKD) through altered activity resulting from splice site and missense variants.

The TLR9 ligand, CpG-ODN, induces protection against cerebral ischemia/reperfusion injury via activation of PI3K/Akt signaling.

BACKGROUND: Toll-like receptors (TLRs) have been shown to be involved in cerebral ischemia/reperfusion (I/R) injury. TLR9 is located in intracellular compartments and recognizes CpG-DNA. This study examined the effect of CpG-ODN on cerebral I/R injury. METHODS AND RESULTS: C57BL/6 mice were treated with CpG-ODN by i.p. injection 1 hour before the mice were subjected to cerebral ischemia (60 minutes) followed by reperfusion (24 hours). Scrambled-ODN served as control-ODN. Untreated mice, subjected to cerebral I/R, served as I/R control. The effect of inhibitory CpG-ODN (iCpG-ODN) on cerebral I/R injury was also examined. In addition, we examined the therapeutic effect of CpG-ODN on cerebral I/R injury by administration of CpG-ODN 15 minutes after cerebral ischemia. CpG-ODN administration significantly decreased cerebral I/R-induced infarct volume by 69.7% (6.4±1.80% vs 21.0±2.85%, P<0.05), improved neurological scores, and increased survival rate, when compared with the untreated I/R group. Therapeutic administration of CpG-ODN also significantly reduced infarct volume by 44.7% (12.6±2.03% vs 22.8±2.54%, P<0.05) compared with untreated I/R mice. Neither control-ODN, nor iCpG-ODN altered I/R-induced cerebral injury or neurological deficits. Nissl staining showed that CpG-ODN treatment preserved neuronal morphology in the ischemic hippocampus. Immunoblot showed that CpG-ODN administration increased Bcl-2 levels by 41% and attenuated I/R-increased levels of Bax and caspase-3 activity in ischemic brain tissues. Importantly, CpG-ODN treatment induced Akt and GSK-3ß phosphorylation in brain tissue and cultured microglial cells. PI3K inhibition with LY294002 abolished CpG-ODN-induced protection. CONCLUSION: CpG-ODN significantly reduces cerebral I/R injury via a PI3K/Akt-dependent mechanism. Our data also indicate that CpG-ODN may be useful in the therapy of cerebral I/R injury.