The research ethics committee of Kanazawa University, Kyushu University, and National Center for Global Health and Medicine reviewed and approved this study. All participants provided written informed consent.

The original 2,504 unrelated samples from the 1kGP were used as a discovery cohort, and their srWGS data were downloaded from the International Genome Sample Resource website (https://www.internationalgenome.org/data-portal/data-collection/30x-grch38). Genomic DNA and RNA were obtained from the blood samples of volunteers belonging to Kanazawa University and residents of Shika town, where the cohort study was being conducted (24). Shika town is located on the Noto Peninsula in Ishikawa Prefecture, Japan. Total of 1,310 samples from an independent Shika cohort were used as a validation cohort in the Japanese population. Genomic DNA was extracted from the blood samples using the QIAamp DNA Blood Mini or Maxi Kit (Qiagen, Hilden, Germany), according to the manufacturer’s instructions, or extracted by SRL Inc. (Tokyo, Japan). Peripheral blood mononuclear cells (PBMCs) were isolated from blood samples by density gradient centrifugation using Ficoll-Paque PLUS (GE Healthcare, Pittsburgh, PA, USA). Subsequently, RNA was extracted from the PBMCs using the FastGene RNA Premium Kit (NIPPON Genetics, Tokyo, Japan) in accordance with the manufacturer’s instructions. First-strand cDNA synthesis was performed using a PrimeScript II 1^st strand cDNA Synthesis Kit (TaKaRa, Shiga, Japan) in accordance with the manufacturer’s instructions. Genomic DNA from the HapMap JPT samples was obtained from the Coriell Cell Repository. Among the three samples (NA12413, NA18959, and NA19087) identified with JoGo-LILR CN Caller, NA18959, and NA19087 were used for long-read sequencing technology and NA12413 was used for JoGo-LILR-trio analysis.

DNA was extracted from the immortalized cell line of GM18959 and GM19087 obtained from the Coriell Cell Repository (Camden, NJ, USA; corresponding to NA18959 and NA19087, respectively). The DNA was subjected to long-read whole genome sequencing (lrWGS) on Sequel IIe (Pacific Biosciences (PacBio)) with the standard high-fidelity (HiFi) sequencing protocol. For the subreads generated in Sequel IIe, the CCS tool was used to generate unmapped HiFi BAM files. The unmapped BAM files were mapped to the GRCh38 reference assembly (without decoy and alt contigs) using Minimap2 (ver. 2.23) (25) and created aligned BAM files. For the aligned BAM file of each sample, the aligned HiFi reads from chr19: 54,180,000 to 54,280,000 were extracted. For NA18959 and NA19087, a total of 122 and 116 reads mapped to chr19: 54,180,000 to 54,280,000 (LILRB5-LILRA6-LILRB3 region) were obtained, respectively. To examine the long deletions in the HiFi reads, the reads were re-aligned to the GRCh38 reference assembly using Minimap2 with the option “-ax splice:hq” ( Supplementary Figures 1 , 2 ), which can capture long deletions. For local de novo assembly of the extracted reads, HiFiasm (ver. 0.16.1) was also applied (26). The assembled contigs were aligned by Minimap2 with the option “-r 500,40000 -ax map-hifi” to allow long deletion alignment. The re-aligned and de novo assembled results were inspected on Integrative Genomics Viewer (IGV) (ver. 2.16.2) to analyze the deletion haplotype and the base-pair resolution length ( Supplementary Figures 1 , 2 ).

The breakpoint of the 33,692 bp deletion was verified using Sanger sequencing. Briefly, polymerase chain reaction (PCR) was performed on genomic DNA to amplify a 6,612 bp product spanning from LILRB5 3’UTR to LILRB3 exon 13 using primers (LILRB5 forward primer:5’-CCTGCACAGCTGAGTCCAGT-3’ and LILRB3 reverse primer:5’-TTAGTCATCTTTGAGTCAGGTGAG-3’). The PacI (aataatttaattaa) and NotI (aataatgcggccgc) restriction sites were added to the 5’ sequences of the LILRB5 forward and LILRB3 reverse primers, respectively, to allow for gene cloning. Following treatment of the PCR products with ExoSAP-IT Express PCR Cleanup Reagents (Thermo Fisher Scientific, Waltham, MA, USA), direct sequencing was performed using specific primers ( Supplementary Table 1 ). Sanger sequencing data were analyzed using Sequencher software (Gene Codes Corporation). Multiple sequence alignments were performed using the Clustal Omega software. The copy numbers of LILRB3 and LILRA6 were quantified using genomic DNA and a QIAcuity Digital PCR System (Qiagen, Hilden, Germany) in accordance with the manufacturer’s instructions. Digital PCR primer/probe sets were used as previously described (14). Individuals with a single copy of LILRB3 were considered carriers of the LILRB5-3 hybrid gene.

To identify the LILRB5 isoforms and LILRB5-3 hybrid gene, PCR was performed with cDNA templates using TaKaRa Ex Premier DNA Polymerase (TaKaRa, Japan) with the following primers: 5′-CCTGCACAGCTGAGTCCAGT-3′ for the LILRB5 forward primer, and 5′-TTAGTCATCTTTGAGTCAGGTGAG-3′ for the LILRB3 reverse primer. The PCR conditions consisted of initial denaturation at 94 °C for 1 min, followed by 40 cycles of denaturation at 98 °C for 10 s, annealing at 62.5 °C for 15 s, and extension at 68 °C for 2 min using a VeritiPro Thermal Cycler (Thermo Fisher Scientific). For cDNA cloning, the PCR products were inserted into the pMXs-puro retroviral expression vector according to standard cloning methods. For PCR-direct sequencing, PCR products treated with ExoSAP-IT Express PCR Cleanup Reagents (Thermo Fisher Scientific) were subjected to Sanger sequencing. cDNA clones that were consistent with the PCR-direct sequencing data were regarded as the major isoforms.

The JoGo-LILR CN Caller was employed to extract LILRB3-LILRA6 CNs from srWGS data. The JoGo-LILR CN Caller is available on the Japanese Open Genome and Omics Portal (https://jogo.csml.org/public/JoGo-LILR). This tool operates by processing BAM or CRAM files aligned to the GRCh38DH reference assembly. It then calculates normalized CNs for specific regions within the LILRB3 and LILRA6 genes (termed LILRB3core and LILRA6core) and their gene body regions (named LILRB3+LILRA6). The current implementation uses CNVnator (version 0.4.1) (27) to calculate the normalized CN values.

For the normalized CN values, the JoGo-LILR tool generated the LILR cluster plot with LILRB3+LILRA6 on the x-axis and LILRB3core/LILRA6core on the y-axis. For the plot, joint CN calling was applied after clustering to the dataset. The tool is particularly adept at stably calling CNs even in datasets with a small sample size, ranging from a single sample to several hundreds. For this purpose, the 1000 Genomes Project (1kGP) samples are typically employed as a background dataset, providing a comprehensive reference for comparison. This practice is instrumental in enhancing the accuracy and reliability of CN calling across diverse sample sizes.

For the detection of the LILRB5-3 hybrid gene, JoGo-LILR was applied to CRAM files from 2,504 non-pedigree and two pedigree samples from the 1kGP ( Supplementary Table 2 ) (28). To establish a Japanese CN profile for the cohorts from the southwest region of Japan (SW-JP) and northeast region of Japan (NE-JP), we analyzed 348 quality control (QC)-passed samples from the Tokyo healthy control dataset and 180 samples from volunteers at the National Hospital Organization Nagasaki Medical Center, Nagasaki, Japan, adhering to a cnv2 ratio QC metric of >0.85. In this process, the 1kGP samples were again utilized as a background reference for joint CN calling.

Finally, the visualization of the LILRB3-LILRA6 genomic region by short-read mapping was conducted using IGV (ver. 2.6.2) (29).

The JoGo-LILR-trio algorithm is a specialized tool used for accurately predicting LILRB3-LILRA6 CN types in trio genotypes. It operates on a probabilistic model that assesses all potential combinations of these CN types in both parents. The primary objective of the JoGo-LILR-trio algorithm is to identify pairs of haploid LILRB3-LILRA6 CN types that align with Mendelian inheritance patterns evident in their offspring. This algorithm is particularly adept at handling situations where multiple parental CN type combinations are plausible. In such cases, JoGo-LILR-trio employs a strategic approach to select the most likely combination that would result in the observed CN types in the children. In this study, the trio dataset comprising individuals NA12413, NA12412, and NA12485 was utilized.

To assess the cell surface expression of LILRB5, plasmids encoding LILRB5 and the LILRB5-3 hybrid gene were transiently co-transfected with green fluorescent protein (GFP) into 293T cells using PEI Max (Polyscience). GFP-positive cells were then identified by flow cytometry, as previously described (15). The cell surface expression of LILRB5 and LILRB5-3 was analyzed by flow cytometry using an anti-LILRB5 antibody (clone 395239, R&D Systems, Minneapolis, MN, USA) and MACSQuant 10 (Miltenyi Biotec, Germany).

Since LILRB3 and LILRA6 exhibit a high degree of homology across a 5 kbp region encompassing the promoter and extracellular domains, usual analysis of srWGS cannot be properly mapped to a reference genome ( Figure 1A ). In contrast, the JoGo-LILR CN Caller called the LILRB3-LILRA6 CNs with the LILR cluster plot and estimated the probable pair of allelic LILRB3-LILRA6 CN type from srWGS ( Figure 1B , Supplementary Table 3 ). As depicted in Figure 1B , nine distinct clusters were clearly illustrated in the two-dimensional representation of the sum of LILRB3 and LILRA6 CNs versus LILRB3/LILRA6 CN ratios. The main clusters near the coordinates (3, 2), (4, 1), (5, 0.67), and (6, 0.5) indicate 1, 2, 3, and 4 copies of LILRA6 with 2 copies of LILRB3, respectively. As noted in previous reports, the LILRA6 copy number ranges from 0 to 6 in most individuals, with the exception of one individual who exhibited seven LILRA6 copies ( Figure 1B , Supplementary Table 3 ). The JoGo-LILR CN Caller also estimated the existence of a previously unreported LILRB3-missing haplotype in the three samples surrounding the coordinate (2, 1), NA12413, NA18959, and NA19087 with high probability ( Supplementary Table 4 ).

The analysis using srWGS is limited by only providing an overview of the structure of the LILRB3 and LILRA6 genomic region. For a more detailed examination at the base pair level, we employed real-time HiFi lrWGS on two Japanese-in-Tokyo (JPT) DNA samples, specifically NA18959 and NA19087. The lrWGS analysis revealed a novel large deletion (33,692 bp) carried by two individuals ( Supplementary Figures 1 , 2 ). This 33,692 bp deletion spanned the LILRB3, LILRA6, and LILRB5 genes ( Figure 2A ). This type of deletion is the largest ever reported and has not yet been detected in the LILR genomic region. Individuals carrying this deletion also had two copies of the LILRA3 6.7 kb deletion, indicating that the haplotype containing the 33,692 bp deletion has the fewest LILR genes. Thus, we carried out a detailed validation of the 33,692 bp deletion. After analyzing the region around the 33,692 bp deletion breakpoint using Sanger sequencing, we observed the presence of a 33,692 bp deletion eliminating exons 1–11 of LILRB3; LILRA6; and exon 13 of LILRB5 ( Figure 2B , Supplementary Figure 3 ). The 33,692 bp deletion left exons 1–12 of LILRB5 and exons 12–13 of LILRB3 intact, resulting in the hybrid gene structure of LILRB5 and LILRB3 (hereafter referred to as “LILRB5-3”). Examination of exon 8 of both LILRB3 and LILRA6 by digital PCR revealed that two individuals (NA18959 and NA19087) carrying the 33,692 bp deletion lacked a single copy of both genes, which was consistent with the sequence data ( Figure 2C ). These data suggest that the 33,692 bp deletion results in the hybrid gene LILRB5-3 and the deletion of LILRB3 and LILRA6.

To validate the existence of the hybrid gene LILRB5-3 in the other Japanese samples, we subjected 1,310 samples from an independent Shika cohort study in Japan to digital PCR analysis ( Supplementary Table 5 ) using the DNA extracted from whole blood. The analysis detected a single copy of both LILRB3 and LILRA6 in four samples. These results confirmed that the existence of the large deletion did not originate from the structural variation acquired in the lymphoblastoid cell lines in 1KGP.

To ascertain the frequencies of the LILRB5-3 hybrid gene within the Japanese population, we conducted an analysis using the JoGo-LILR CN Caller. Our dataset included 180 srWGS samples from SW-JP and 348 samples from NE-JP ( Supplementary Figures 4 , 5 , Supplementary Tables 6, 7 ). Among these, two samples from NE-JP were identified as carrying the hybrid gene, while none from SW-JP exhibited this genetic feature. Notably, the Japanese samples in the 1kGP project was also catalogued in NE-JP (Tokyo).

The presence of the LILRB5-3 hybrid gene has been definitively confirmed in the Japanese population. This genetic feature was not detected in the other East Asian samples from the 1kGP, which included 93 CDX (Dai Chinese), 103 CHB (Han Chinese), 105 CHS (Southern Han Chinese), and 99 KHV (Kinh Vietnamese) individuals ( Supplementary Table 3 ). However, the Korean population was not represented in the 1kGP dataset. Therefore, we examined 70 QC srWGS data from the Korean Personal Genome Project (KPGP) using JoGo-LILR CN Caller ( Supplementary Figure 6 , Supplementary Table 8 ). Our analysis revealed the absence of the LILRB5-3 hybrid gene in these Korean samples. Since the LILRB5-3 hybrid gene appears to be rare in the Japanese population, additional sequencing is needed to definitively determine its presence or absence in the other regions of East Asia.

Among a total of 2,504 unrelated samples analyzed, only one individual, NA12413 from the CEU cohort of European descent, was identified with a single CN for both LILRA6 and LILRB3, aside from two samples in the JPT cohort. This individual, NA12413, along with NA12412 and NA12485, were annotated as a father, mother, and child, respectively. All these srWGS data were available as pedigree dataset in 1kGP project (28). Utilizing the JoGo-LILR-trio tool, which is designed to predict the most likely pair of allelic LILRB3-LILRA6 CN types in offspring, we analyzed this trio’s srWGS data. The analysis revealed that the child, NA12485, possesses one copy of both LILRB3 and LILRA6 genes and is estimated to have a genotype with zero copies of both LILRB3-LILRA6 on one allele and a single copy of both LILRB3-LILRA6 on the other allele (JoGo-LILR CN Caller value (x=2.3, y=1.4)). This observation suggests that the LILRB5-3 hybrid gene was transmitted from the father (NA12413) to the child (NA12485), and provides robust support for the existence of the hybrid gene in the CEU population.

To verify the expression of the LILRB5-3 hybrid gene, PCR-direct sequencing and cDNA cloning were conducted on Japanese individuals with and without the 33,692 bp deletion. Two primary LILRB5 isoforms, long and short, were identified in individuals without deletions through cDNA cloning of LILRB5 transcripts. The long isoform contained three ITIMs, whereas the short isoform contained two ITIMs ( Figure 3A ). This difference is due to alternative splicing of the splice acceptor site of intron 12. In contrast, a chimeric gene composed of LILRB5 and LILRB3 was detected in a Japanese individual with a deletion using LILRB5 forward and LILRB3 reverse primers ( Figure 3B ). The amplicon was approximately 2 kb in size, which was consistent with the expected size of the exonic region of LILRB5-3 ( Figure 3B ). cDNA cloning and sequencing identified two isoforms of a chimeric gene composed of LILRB5 and LILRB3, providing evidence that LILRB5-3 was transcribed at the mRNA level ( Figure 3C ). These two isoforms differ by three bp at the acceptor sites of exons 8 and 12 of LILRB5 and LILRB3, respectively. This resulted in a disparity between the two amino acids that were not located within any functional domains or motifs. The hybrid junction between LILRB5 and LILRB3 was located within the intracellular domain and fused the extracellular domain of LILRB5 with the intracellular domains containing partial LILRB3. The original number of ITIMs in LILRB3 was four, whereas LILRB5-3 contained three of the four LILRB3 ITIMs ( Figure 4A ). This suggests that LILRB5-3 acquired a novel signaling pathway.

There exists an example wherein the killer cell Ig-like receptor, KIR3DL1*004, lacks cell surface expression but exhibits intracellular retention (30). Therefore, we investigated whether LILRB5-3 can be displayed on the cell surface. To this end, we transfected the identified LILRB5 and LILRB5-3 genes into 293T cells and stained the transfectants with an anti-LILRB5 antibody. Flow cytometry confirmed the detection of the LILRB5-3 proteins on the cell surface, demonstrating that LILRB5-3 can be expressed at the cell-surface protein level ( Figure 4B ). These data suggest that LILRB5-3 may have a novel function in carriers with the 33,692 bp deletion.