The current understanding of the pathophysiology and progression of BD is inadequate. For better management of BD, biomarkers are necessary for diagnosis as well as for the selection of suitable therapeutic intervention 22,27,37. Differential response to lithium treatment by more than half of BD patients is a conundrum in the field of BD research10. Some studies have provided information about a genetic link to non-response to lithium treatment, as family members of NR BD patients usually also do not respond to lithium 10,43−45. Nevertheless, mood episodes in BD are also known to be impacted by environmental factors like stress and traumatic experience 10.
In our study, we compared LCLs from 3 groups, Control, BD LRs, and BD NRs, by performing RNA sequencing to search for DEGs and associated biological pathways. The advantage of our method is its high-throughput, hypothesis-free approach and the use of accessible patient-specific LCLs. LCLs have a huge potential for use in discovery of biomarkers especially in psychiatric diseases where the brain tissues are not accessible for molecular investigations from live patients 32,37. We have also analyzed and compared our data of BD LR and NR patient subtypes with a previously published study37 and found shared DEGs between the two cohorts. Additionally, in our study we discovered significant DEGs between the BD and control groups.
When comparing the subtypes within the BD group, i.e., the LR vs the NR patients, a set of genes belonging to the immunoglobulin heavy and light chains variable region were found to be significantly upregulated in lithium NRs as compared to LRs. Increased expression of IGV genes specifically in lithium NRs entails deregulation of the immune system. Apart from the immunoglobulin genes, we found HLAU, ZNF300 and TRAT1 genes to be significantly downregulated in NR subtypes. Interestingly, previous GWAS studies have reported the association of these genes in BD and other psychiatric disorders 46–48. HLA U is a pseudogene found on the MHC complex49. GWAS studies have revealed MHC complex as an important risk gene in both BD and Schizophrenia (SCZ)50,51. Apart from its vital immunological role, the HLA locus has been proposed to play an important role during neurodevelopment51,52. In our study we found IL-18 to be significantly downregulated in NRs. IL-18 is a proinflammatory cytokine reported to be expressed in different regions of the brain and has a known role in neuroinflammation 53. Some studies have reported high levels of circulating IL-18 in BD and SCZ patients 54–56. Altered expression of these genes specifically in NRs compared to LRs is intriguing, especially because these genes overlap with previous reports in BD.
The LR vs NR RNA-seq data of 24 BD patients (12 LR and 12 NR) from Milanesi et al.37 were included in the joint dataset and analyzed together. Between the original and joint datasets, 27 genes were found to be common. Apart from HLA U and IGV genes, TANC1 and TMEM 132 D were among the top 5 most common genes. TANC 1 protein has been shown to interact with PSD 95 as well as other synaptic proteins such as glutamate receptors57. TANC1 mutation, along with NRXN1 and RBMS1 genes, was implicated in psychomotor retardation in a case study of chromosomal inversion58. GWAS studies have linked the TMEM 132 D polymorphism to anxiety and panic disorder59,60. TMEM 132D has also been reported to be dysregulated at the mRNA level in brain regions associated with anxiety disorders60. The altered expression of these crucial genes is intriguing, especially since we combined datasets from another group performed in a different cohort. We also compared gene expression between BD and control LCLs. We found downregulation of immunoglobulin kappa chain variable genes in BD patients when compared with control groups. ADAM23, IGHEP1, GNAQ, FRG1EP, PPP4R4 were upregulated in BD patients, including LRs and NRs as subsets. GNAQ and ADAM23 have earlier been reported to be associated in BD 61,62. ADAM23 was previously reported in a study done on the microarray of BD postmortem brain samples 62. ADAM23 is a membrane protein belonging to the family of ADAM protein and is part of a presynaptic complex interacting with the Lgl receptor 63. While FRG1EP has a reported function in cancer, the function of the other above-mentioned genes is not well understood, especially in the context of psychiatric diseases 64.
We also discovered some common genes in our original dataset of LRs vs NRs that had previously been reported in GWAS on BD subjects. Among the common genes, RIMS1 and BCL11b were notable because they appeared in 4 GWAS publications and appeared together in 2 publications. RIMS1 is a protein involved in neurotransmission because it is required for synaptic vesicle exocytosis65. RIMS1 has previously been found to have altered gene expression in cortical brain samples from SCZ and autism patients66. BCL11b is involved in both immunological and neuronal functions. BCL11b has been implicated in Alzheimer's disease, Huntington's disease, Neuro-HIV, learning and memory, and its cellular role in cortical GABAergic neurons, medium spiny neurons, and vomeronasal sensory neurons have been studied67. In its non-neuronal role, BCL11b is crucial for T-cell differentiation and VDJ recombination in immunoglobulin proteins67,68. ADCY1 and NPTX1 were also found to be associated with BD in our analysis as well as other GWAS studies 69,70. ADCY1 plays a potential role in learning and memory71,72, whereas NPTX1 is required for neural cell specification and is also known to be involved in synaptic plasticity73,74.
Overall, our findings suggest significant dysregulation in the immune system as well as alteration of genes involved in synaptic pathways in BD. Overwhelming reports have highlighted the importance of immune dysregulation in BD and SCZ75–78. We have also found genes implicated in neuroinflammation and anxiety disorder altered in BD. Furthermore, in BD patient subtypes, we found cytokines and immunological genes specifically altered in NRs compared to LRs. The validation of our results on another cohort confirms the robustness of our findings. Using an iPSC-based patient-specific model to understand the role of these genes could shed light on the pathophysiology and prognosis of BD LRs and NRs.
We used 4 supervised classification algorithms - SVM, a neural network, a NB, and a Random forest - to predict the outcome. Our model used a random sampling cross-validation scheme to create training and test datasets. This model gave robust statistical results with no errors, particularly using SVM and neural network classifier algorithms for lithium response prediction with high conformity in BD patients. The BD samples could also be predicted and distinguished from the control samples with no errors using NB, SVM, and neural network classifiers.
Some studies have used supervised machine learning algorithms for diabetes, cardiovascular diseases, cancers, Alzheimers' and Parkinsons' diseases using clinical parameters and neuroimaging datasets79. Ours is the first study to use transcriptomic datasets from LCLs of BD patients to predict the disorder as well as the responsiveness to lithium. As previously stated, we also included and analyzed samples from a separate cohort published by Milanesi et al. 2017, which increases the scalability of our methodology for future use in predicting lithium response in BD. The method is cheap and fast to implement and can be easily implemented in a psychiatric clinic. In conclusion, using RNA-seq, we have found a set of DEGs from LCLs of BD patients that can be used as potential biomarkers to diagnose as well as classify BD patient subtypes. Functional studies of these genes in model systems should help to unravel the cellular and molecular mechanisms underlying the pathophysiology of BD. Importantly, we hope that our study and our developed algorithm will serve as an easy and ready-to-use protocol for deciding on effective treatment in the clinic within days of diagnosis.