Joint Segmental Duplication Co-option Drives Human-specific Transcriptional Readthrough and Expression Fine-tuning of NPEPPS - TBC1D3

and Devices, ShanghaiTech University, Shanghai, China 6. Center for Genomic Research, International Institutes of Medicine, Fourth Affiliated Hospital, Zhejiang University, Yiwu, China 7. Center of Evolutionary & Organismal Biology, and Women’s Hospital at Zhejiang University School of Medicine, Zhejiang University, Hangzhou, China 8. School of Medicine, Zhejiang University, Hangzhou, China 9. Center for Comparative Biomedicine, Ministry of Education Key Laboratory of Systems Biomedicine, State Key Laboratory of Medical Genomics, Institute of Systems Biomedicine, Shanghai Jiao Tong University, Shanghai, China #These authors contributed equally to this work. *Corresponding authors: Kaiyue Ma ([email protected]) & Yafei Mao ([email protected]) .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 2

Segmental duplication (SD) is a major driver of functional changes in evolution and disease. Many genes embedded within SDs, such as NPEPPS and TBC1D3, display substantial copy number variation (CNV) across individuals. Yet, the precise identification of the functional copies and their transcriptional outputs remains largely unstudied. Focusing on NPEPPS and TBC1D3, we illustrate human-specific expression fine-tuning mechanisms associated with readthrough transcripts. We identified a human-specific NPEPPS-TBC1D3 digenic genomic structure that originated from a joint SD pair and became fixed across populations. Experiments demonstrate that this structure generates NPEPPS -TBC1D3 readthrough transcripts, which are the predominant isoforms of TBC1D3 expression in various cell types, fine-tuning its protein level. Furthermore, a human-specific hypomethylation signal within an upstream CpG island of NPEPPS precisely pinpoints the expressed TBC1D3 paralog. Moreover, we reveal transcriptional readthrough events are ~3-fold enriched for joint-SD-associated transcriptional readthrough (JSDTR) and identify 109 JSDTR gene pairs, including neurodevelopmentally important pairs and clinically interesting SERF1A/B-SMN1/2. Taken together, our findings comprehensively describe an example of how a joint SD event shaped evolution and suggest that JSDTR is a broad mechanism for the emergence of new functions. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 3 Main Segmental duplications (SDs) are large, low-copy repeats typically >1 kb in length and sharing >90% sequence identity1, which frequently incorporate into genomic regions through non-allelic homologous recombination (NAHR)2. The expansion of SDs can potentially create copy number variations (CNVs), new gene bodies, new transcripts (including read-through transcripts), and fused open reading frames (ORFs) 3,4. While such intense reorganization of genomic elements has been recognized as crucial for forming human-specific genomic features important to evolution, previous studies focused primarily on human-specific genes 3,5,6, failing to highlight the significant roles of human-specific transcriptional/translational regulatory mechanisms resulting from SD expansion. Here, we report a joint SD locus in the human genome (chr17:39044723-39120263 on the T2T- CHM13 reference genome, “the 39M locus”), where two different SDs converge, bring NPEPPS and TBC1D3 gene bodies into close proximity, and give rise to NPEPPS-TBC1D3 readthrough transcripts. While NPEPPS-TBC1D3 transcripts have been detected in previous studies 7,8, their presence has been mistakenly attributed to disease-related genomic events 9. In this study, we establish that NPEPPS-TBC1D3 transcripts in fact originate from a joint SD locus fixed across human populations and are commonly expressed in various cell types of healthy human individuals as the predominant TBC1D3 transcripts, representing a unique evolutionary case of gene regulation. First, using genome assemblies from Human Pangenome Reference Consortium (HPRC) Phase 1 10, Human Genome Structural Variation Consortium (HGSVC) Phase 3 11, and Asian Pan- Genome Project (APG) Phase 1 (Wu et al., manuscript in revision), as well as the T2T- CHM1312 and the CN1 diploid genome 13, we have completely charted NPEPPS and TBC1D3 paralogs across human populations, respectively (Fig. 1 and Supplementary Fig. 1,2). 462 haplotypes with a single contig covering all TBC1D3 and NPEPPS loci were selected for subsequent analyses, avoiding contig breakpoints that can complicate the results. NPEPPS has been reported to protect against TAU-induced neurodegenerative disorders, such as Alzheimer disease and amyotrophic lateral sclerosis, by proteolysis of TAU protein 14–16. The .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 4 orthologous ~99.4-kbp NPEPPS gene body is located at human 17q21.32 (chr17:48385617- 48485055, T2T-CHM13, “the 48M locus”). A previous study reported that incomplete segmental duplication of NPEPPS resulted in multiple ~63.0-kbp paralogs17. In our study, we increased the completeness of the analysis, identifying various incomplete duplicates with different lengths across human populations (Fig 1a,b). Aside from the ~100-kbp complete NPEPPS copy, incomplete copies exhibit lengths clustered around three peaks (~70.2 kbp, 56.1 kbp and 37.3 kbp; determined by aligning the ~100-kbp sequence to other loci). The 70.2-kbp copies have a complete 5’ end but missing the 3’ end ( e.g., RefSeq NPEPPSP1 at the 39M locus; chr17:39058811-39120263), while other copies are truncated at both the 5’ and the 3’ ends ( e.g., RefSeq LOC101060212 at the 37M locus; chr17:37222028-37267942) (Fig. 1a-d and Supplementary Fig. 3a). Together, NPEPPS exhibits high CNV in the human population. Here, we identified 2 to 12 NPEPPS copies per haplotype (Fig. 1c,d and Supplementary Table 1A), corresponding to the previously reported range of 4-12 copies per diplotype 18. TBC1D3, on the other hand, encodes TBC1 domain family member 3, which, while reported to be involved in Rab GTPase signaling, has its detailed roles and associated pathways under debate 19,20. Despite many exciting functions proposed and reported 21–23, significant gaps remain in our understanding of the regulation of TBC1D3 expression, casting doubts on the biological relevance of previous functional results. The TBC1D3 gene family is characterized by a complex pattern of SDs, resulting in two distinct groups of paralogs. One includes the dispersed paralogs that were annotated as pseudogenes ( TBC1D3P1 to P5 and P7, at chr17 61M, 63M, 20M, 18M, 28M, and 42M; TBC1D3P6 , at chr1 15M). In contrast, the other group comprises the complete and potentially functional copies that are tightly clustered in the 37M and 39M loci (Fig. 1a) 24. Here, we identified 6-30 copies per haplotype (Fig. 1e-g and Supplementary Table 1A), corresponding to the previously reported range of 23-42 per diplotype 18. In contrast to NPEPPS , most duplicates of TBC1D3 (95.1%, 6523/6858) retain the 10.9-kbp complete gene body (Fig. 1e and Supplementary Fig. 3b), while a small fraction (4.8%, 331/6858) possesses a 8.4-kbp gene body that lacks the first three exons (Supplementary Table1B,C). Despite the extensive CNV, human TBC1D3 expression was reported to be primarily contributed by a single paralog group: TBC1D3-CDKL 24, one of the two expanded .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 5 TBC1D3 clusters at 17q12 (Fig. 1a). Although this observation that only one cluster is the main contributor of TBC1D3 expression is biologically reproducible, the accurate attribution to a specific TBC1D3 paralog was previously not achieved. Instead, previous attributions are largely influenced by different gene annotation processes, lacking a sequence- and context-aware analysis (Supplementary Fig. 4). Therefore, we sought to pinpoint the exact paralog(s) that drives the transcription and subsequent translation of TBC1D3. The substantial variations of NPEPPS and TBC1D3 motivated us to examine the joint SDs of these two genes and their effects on the regulation of TBC1D3. We first examined the 48M locus where the canonical NPEPPS resides and detected no joint SD. The NPEPPS at this locus is with fixed orientation and CN of 1 (462/462 haplotypes) (Fig. 1a,d and Supplementary Fig. 1,2), and no TBC1D3 paralog is found near NPEPPS-48M (± 1Mbp) in all assemblies. Similarly, no NPEPPS paralog is found near any of TBC1D3P1 to TBC1D3P7 (± 1Mbp), correlating with our Iso-Seq results, which show no evidence of expression for these paralogs (Supplementary Fig. 5a,b). Interestingly, a previous study reported a recurrent ~2.2-Mbp microdeletion at 17q23.1- q23.2 flanked by a pair of SDs (Ballif et al., 2010) . The deletion of TBX2 and TBX4 results in heart defects and limb abnormalities. We identified TBC1D3P1 and TBC1D3P2 as the SDs flanking this region, revealing their role as the structural basis for this recurrent microdeletion (Supplementary Fig. 5c-e). In contrast to the loci examined above, the 37M and the 39M loci contain both NPEPPS (37M: 0-5 copies per haplotype; 39M: 1-10 copies) and TBC1D3 paralogs (37M: 1-19 copies; 39M: 2- 20 copies). While the orientation and CN of the NPEPPS paralogs vary at the 39M locus, all assemblies possess at least one NPEPPS paralog (hereafter, “ NPEPPS-39M”) (Fig. 1a,c,d,f,g, Supplementary Fig. 1,2, and Supplementary Table 1B). Most importantly, in all haplotypes, a TBC1D3 paralog with the same orientation as NPEPPS-39M can be found at its immediate downstream region, indicating that the NPEPPS-TBC1D3 digenic structure originated from joint SDs and became fixed across human populations (Fig. 1a,h, Supplementary Fig. 1,2, and Supplementary Table 1B). The similar digenic structure also exists for a subset of other paralogs at the 39M and the 37M loci, indicating duplications of the joint SD sequence as a whole unit following the convergence of the respective SDs (Fig. 1a,h and Supplementary Fig. 1,2). .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 6 Importantly, Iso-Seq data from the CHM13hTERT cell line reveals expression of NPEPPS- TBC1D3 readthrough transcripts at places with the digenic structure (Fig. 1i). In the readthrough transcripts, the terminal exon of NPEPPS-39M is partially skipped and spliced to the second exon of TBC1D3 -39M, preserving the canonical start codon for coding TBC1D3. This splice junction is supported by 39 reads, comparable to the support read counts observed for other junctions within NPEPPS-39M and TBC1D3-39M (min=15, median=50, max =57 reads, n = 23) (Fig. 1i and Supplementary Table 2A). Furthermore, we observed no read supporting the canonical splicing event between the first and second exons of TBC1D3-39M. This absence likely indicates that the independent transcription of TBC1D3-39M occurs at such a low proportion that it falls below the detection limit of the Iso-Seq experiments. To validate the existence of this readthrough splicing in natural human cellular contexts, we generated induced pluripotent stem cells (iPSCs), iPSC-derived neural progenitor cells (NPCs), and iPSC-derived forebrain neurons (FBNs) using the blood sample from a healthy donor (CN1) (Supplementary Fig. 6a). NPEPPS -TBC1D3 readthrough transcripts are identified in all three cell types (Supplementary Fig. 6b). Taken together, the evidence illustrates that NPEPPS-TBC1D3 readthrough transcripts naturally occur across diverse cell types in healthy individuals. Next, we establish NPEPPS -TBC1D3 are the predominant transcripts responsible for TBC1D3 expression in human neural lineage cells, and potentially other cell types. A TBC1D3-specific primer (TBC1D3-Rev1) was designed and utilized in a 5’ Rapid Amplification of cDNA Ends (5’ RACE) experiment (Fig. 2a, see also Supplementary Fig. 7a,b for the pilot experiment using TBC1D3-Rev0). TOPO cloning using the 5’ RACE product exhibits NPEPPS-TBC1D3 as the predominant transcripts of TBC1D3 in NPCs (18 of 20 colonies; 1 non-specific; 1 noise; 0 non- readthrough) (Supplementary Fig. 7c,d) . Interestingly, the NPEPPS part of these readthrough transcripts exhibits high diversity, with individual transcripts containing different ORFs. This variation likely results from variable splicing events (Supplementary Fig. 7e, 8-11). To accurately quantify the proportion of NPEPPS-TBC1D3 in the 5’ RACE product, nanopore sequencing was performed, which again confirms the majority of the TBC1D3 transcripts are NPEPPS-TBC1D3 transcripts (>86.9% in C20 NPC; >89.9% in CN1 NPC; >87.6% in .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 7 HEK293T) (Fig. 2b, Supplementary Fig. 12, and Supplementary Table 2B). This conclusion is supported by estimation using short-read RNA-seq data from CN1 iPSC (~91.1%-91.4%), NPC (~67.6%-91.9%), and FBN (~86.0%-98.2 %) (Supplementary Fig. 6b, 13a). Interestingly, testis is another tissue with reported high TBC1D3 expression. Using publicly available RNA-seq data (PRJNA1092349), we estimate that the readthrough transcripts contribute to ~29.3%-56.6% of the TBC1D3 expression in healthy testicular tissues, which potentially suggests different regulation mechanisms from the neurodevelopmental regulation (Supplementary Fig. 13). Given that the digenic structure is present at the 39M locus in all individuals but absent at the 37M locus in a subset of individuals, we hypothesize that the 39M locus is the primary locus for NPEPPS-TBC1D3 expression. To validate this hypothesis, we performed a methylation analysis, which reveals a hypomethylation signal in a CpG island upstream of NPEPPS-39M (chr17:39119773-39120426). A similar signal also presents upstream of NPEPPS-48M (chr17:48392644-48393344), but not NPEPPS-37M due to the lack of the CpG island sequence (Fig. 2c and Supplementary Fig. 14). Additionally, the readthrough transcript sequences (reconstructed from TOPO-Sanger sequencing) are more similar to NPEPPS-39M than NPEPPS-37M based on the mapping quality (MAPQ) scores. Among 23 transcript sequences, eighteen have higher MAPQ for NPEPPS-39M while the remaining five sequences exhibited equal mapping confidence to both loci (MAPQ = 0), indicating ambiguous alignment (Fig. 2d and Supplementary Table 2C). Furthermore, single-cell analysis of human brain organoid samples at Day 42 and Day 98 shows again TBC1D3-39M is the predominant paralog for TBC1D3 expression (Fig. 2e and Supplementary Fig. 15). To examine the constraint of this loci while avoiding introducing false negative signals from other copies, we analyzed the Tajima’s D and pA/pS of the ancestral NPEPPS, NPEPPS-39M, TBC1D3-39M, TBC1D3-61M, TBC1D3-63M and other copies from each haplotype. Our results show that the 39M NPEPPS-TBC1D3 digenic region and TBC1D3-61M have significant negative Tajima’s D and may experience recent selective sweeps. Moreover, while all pA/pS values are beyond 1 (meaning positive selection), TBC1D3-39M shows the lowest value among all TBC1D3 copies, indicating this locus is more conserved on amino acid level compared with other copies. In general, our constraint analyses indicate that TBC1D3-39M is the most .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 8 conserved paralog at both nucleotide and amino acid levels, supporting the notion that it is the main functional paralog in the TBC1D3 family (Fig. 2f). In addition to our approach, we explored employing previously commonly-used methods to compute nucleotide diversity (π ) and Tajima’s D on the whole chromosome level, using long-read-based variants from HGSVC3. However, nucleotide diversity ( π ) doesn’t exhibit significantly lower values in both African (p=0.439) and non-African populations (p=0.685) (Fig. 2g). Similarly, Tajima’s D is also not significantly lower in African (p=0.204) and non-African populations (p=0.113) (Fig. 2g). We speculate that this result is probably due to limitations of these methods, specifically, the ambiguous alignment of reads from different loci when dealing with duplicated genes, which introduce false negative results. Collectively, the observations above support that the cis- regulatory elements of NPEPPS-39M are major drivers of the expression of the NPEPPS- TBC1D3 transcripts. Notably, while TBC1D3 has been frequently referred to as a hominoid-specific gene 19,23, its gene sequence is in fact commonly present in other simians 24–26. Similar as in humans, the gene expression and regulation of TBC1D3 across simian species remain largely understudied. Using primate telomere-to-telomere (T2T) genomes27,28, we have elucidated the evolutionary history of NPEPPS and TBC1D3, shedding light on the evolutionary path of the NPEPPS-TBC1D3 digenic region (Fig. 3a). Macaques and orangutans have one single NPEPPS copy (syntenic to the 48M locus) per haploid genome (Supplementary Table 3A,B). This ancestral NPEPPS is located adjacent to an inversion–translocation breakpoint that distinguishes orangutans from hominines (African great apes) (Fig. 3a). These structural rearrangements potentially led to the partial duplication resulting in the additional gene copy similar to NPEPPS-39M (Fig. 3b). Importantly, while the SDs of TBC1D3 and NPEPPS are spatially associated in all hominines, the NPEPPS-TBC1D3 digenic structure, where those two genes are immediately neighboring and share the same orientation, is fully intact only in humans (Fig. 3c). In chimpanzees and bonobos, the downstream TBC1D3 is heavily truncated and consequently precludes the NPEPPS-TBC1D3 readthrough transcripts. Interestingly, the partially duplicated NPEPPS generates readthrough transcripts together with its more distal gene: NPEPPES-LOC455004 in chimpanzee and NPEPPES-LOC100991964 in bonobo. Both partner genes are annotated as immune-related C-C .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 9 motif chemokine 4 ( CCL4) genes (Fig. 3c,d and Supplementary Fig. 16). Moreover, both chimpanzees and bonobos have two copies of CCL4, and in both species, the copy that forms the NPEPPS-CCL4 readthrough transcript has higher expression compared to the other in neural lineage cells (Fig. 3e and Supplementary Table 3C). At the NPEPPS-CCL4 locus, the readthrough transcripts are estimated to constitute ~49.5% to ~100% of all CCL4 transcripts (Fig. 3f). On a separate note, in gorillas, while TBC1D3 retains the intact gene structure, the upstream NPEPPS copy is truncated at its 5’ end, thereby lacking the CpG island sequence seen in human NPEPPS-39M (Fig. 2c, 3c, 3d, and Supplementary Fig. 16). Taken together, the intact NPEPPS-TBC1D3 digenic structure, with the complete TBC1D3 ORF sequence and the hypomethylated NPEPPS CpG island, is specific to humans. This unique genomic configuration may underlie human-specific expression patterns of TBC1D3 (Supplementary Fig. 17). Furthermore, we hypothesize that the NPEPPS-TBC1D3 digenic structure serves to fine-tune the TBC1D3 protein abundance through post-transcriptional and translational mechanisms, controlling the TBC1D3 level to the physiological optimal. To explore this hypothesis, we cloned several different ( NPEPPS-) TBC1D3 transcripts (Fig. 4a and Supplementary Fig. 18). Similar to what is observed in the 5’ RACE analysis, the transcripts exhibit substantial splicing variability, both in the NPEPPS region and in the TBC1D3 region, adding another layer of complexity to the post-transcriptional and translational regulation. Using a pCAGGS expression system in HEK293T cells, plasmids constructed using cDNA of the readthrough transcripts generate lower TBC1D3 protein level compared to what is generated by TBC1D3-only cDNA sequences (Fig. 4b,c and Supplementary Fig. 19a,b). It is likely because the NPEPPS sequence within a readthrough transcript can be deemed an ultra-long 5’ UTR of TBC1D3 (Supplementary Fig. 19c-f and Supplementary Table 2D), notably longer than what is typically observed in mRNAs 29,30, thereby adding the chance of decreased cap-dependent translation efficiency. For instance, many NPEPPS-TBC1D3 transcripts are polycistronic, containing ORFs within the NPEPPS sequence upstream of the TBC1D3 sequence (Fig. 4 and Supplementary Fig. 8-11, 18), which may further lower the translation efficiency31–33. It is worth noting that an internal ribosome entry site (IRES) sequence, if exists, could enhance the cap-independent translation efficiency of the TBC1D3 sequence within the readthrough .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 10 transcripts; however, we were unable to detect such a sequence (Supplementary Fig. 20). Furthermore, the versatile splicing schemes potentially introduce an additional layer of control to the 5’ UTR of these transcripts (Supplementary Fig. 7-11, 18). Additionally, we cannot rule out inhibitory transcriptional regulatory effects, as the RNA level of the readthrough cDNA is also lower than that of the TBC1D3-only cDNA (Supplementary Fig. 21). Next, to better examine the translation of the readthrough transcripts, using the same pCAGGS system, we performed Ribo-Seq for two representative readthrough transcripts (NT1-2 and NT2- 5) and a TBC1D3-only transcript. While the TBC1D3 ORF in the readthrough transcripts exhibits regularly distributed peaks, sharing the same pattern with the ORF in the TBC1D3-only cDNA, Ribo-Seq peaks within the NPEPPS region do not cover the entire ORFs, suggesting the selected representative readthrough transcripts mainly execute TBC1D3 functions (Fig. 4d and Supplementary Fig. 22). Finally, using Iso-Seq data from CHM13hTERT, CN1 NPC, and C20 NPC, we identified 9,569 readthrough events with at least one Full-Length Non-Chimeric (FLNC) read supported. After quality control (Methods), we retained 1,588 high-confidence fusion events and 1,219 unique gene pairs (Fig. 4e and Supplementary Table 4A). Among them, 109 are associated with joint- SD-associated transcriptional readthrough (JSDTR). Transcriptional readthrough events exhibit significant enrichment of JSDTR gene pairs (p-value=1.65e-17), but not other pairs (p-values > 0.99), with an average 2.63-fold difference (Fig. 4f and Supplementary Table 4B). Among the JSDTR gene pairs, we identified previously detected NOTCH2NLA/B/C-NBPF10/14/19 and ARHGAP11B-LOC100288637 34 (Supplementary Fig. 23). Similar to NPEPPS -TBC1D3, the downstream partners, NBPF s and LOC100288637, lack CpG islands at 5’ ends of their gene models, whereas the upstream partners, NOTCH2NLs and ARHGAP11B , have hypomethylated CpG islands in their promoter regions. Interestingly, the RefSeq gene model of NBPF26 (chr1:120,737,164-120,851,627) contains NOTCH2NLR (chr1:120,737,164-120,808,094) within its 5’ end, reflecting, to some extent, the process of fusion gene birth. Additionally, a recent study highlighted the neurodevelopmental function of the gene pair of NOTCH2NLB and NBPF14, where they were described as co-evolving and co-expressing 35. We speculate the .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 11 NOTCH2NLB-NBPF14 readthrough transcripts may also contribute to neurodevelopmental functions, pending further experimental validation. Importantly, here, we also identified other biomedically relevant gene pairs that have previously not been highlighted, including SMG1P2/P3/P5/P6-NPIPB12/B3/B13/B11 and SERF1A/B- SMN1/2 (Fig. 4g,h and Supplementary Fig. 23). Previously, the role of NPIP paralogs in genomic rearrangements and their fusion with PKD1 paralogs (PKD1 -NPIPA) were reported 36. A recent study using GRCh38 detected the fusion of SMG1 and NPIPB5 37; however, these 2 genes have opposite orientations and are separated by approximately 3.6 Mbp in hg38. While this “SMG1-NPIPB5” fusion was reported to be universal across hundreds of individuals, this observation likely arises from mis-mapping between highly similar paralogous copies using short-read RNA-seq data and fails to be validated using Iso-Seq data. In contrast, our Iso-Seq- based analysis identified four fusion pairs involving SMG1 pseudogenes and NPIP paralogs (average gene pair distance: ~ 6 kbp) (Fig. 4g and Supplementary Fig. 23). Again, the downstream NPIP partners lack CpG islands near the 5’ ends of their gene models, while their upstream SMG1 partners contain hypomethylated CpG islands except for SMG1P6. In the case of SMG1P6, a hypomethylated CpG island can be found in the promoter region of its immediate upstream gene BOLA2 , generating BOLA2 -SMG1P6-NPIPB11 transcripts. These examples collectively demonstrate that the co-option of regulatory elements of neighboring genes is a key mechanism by which duplicated genes achieve expression. Additionally, JSDTR gene pairs, such as SERF1B-SMN2, are clinically relevant (Fig. 4h). SMN1 and SMN2 are associated with Spinal Muscular Atrophy (SMA), which is caused by abnormality of SMN1 and modified by the varying copy numbers of SMN2 38. It is biomedically interesting to study whether SERF1-SMN readthrough transcripts modify disease phenotypes and thereby have implications for diagnostics and treatment development. On the one hand, while a previous study has investigated the modifying effects of SERF1A 39, the analyses remain to be refined by accounting for the SERF1-SMN readthrough transcripts. On the other hand, one of the main disease-modifying treatments employs antisense oligonucleotides or small molecules to promote SMN2 exon 7 inclusion to restore the level of full-length functional SMN protein 40. Future research is needed to determine how SERF1B-SMN2 reacts to these treatments. Notably, unlike .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 12 the examples showcased above, hypomethylated CpG islands can be found in the promoter regions of both SERF1A/B and SMN1/2, indicating a more complex regulatory system. In summary, here we report a NPEPPS -TBC1D3 digenic genomic structure, which originated from joint SDs and became fixed in humans. The intact sequence of this digenic structure is absent in non-human primates. A human-specific hypomethylation signal within a CpG island pinpoints the expressed TBC1D3 paralog that is transcribed predominantly as NPEPPS-TBC1D3 readthrough transcripts. The readthrough transcripts underlie various post-transcriptional and translational regulation mechanisms, potentially fine-tuning the functions of TBC1D3. Moreover, readthrough transcription originating from joint SDs is potentially a broad mechanism for the emergence of new functions. Our work provides a necessary foundation for future functional studies and illustrates how joint SDs drive human-specific regulatory innovation.

Genomic reorganization resulting from large-scale structural variants is a prominent drive for functional innovation in evolution, often leading to the formation of new gene bodies or new regulatory mechanisms enabled by newly formed contact loops 5,28. When a new gene is repositioned near another existing gene, the co-option of the neighbor's regulatory elements may happen, reshaping the new gene’s expression pattern. This alteration may be a major step in the evolution of the new gene/transcript , potentially providing a wider expression range for further selection. Interestingly, this repositioning also creates a genomic structure that may serve as an “evolutionary doorway” to a readthrough transcript, which fine-tunes the new gene’s expression level through readthrough efficiency and splicing variability, as well as translation efficiency regulated by the proximal gene's sequence as an untranslated region (UTR). Here, we describe the NPEPPS-TBC1D3 digenic structure originating from a joint SD event as such an example. There are growing interests in TBC1D3 because of its uniqueness in primate evolution and its potential functions in the brain. Many exciting TBC1D3 functions have been reported and proposed. For instance, heterologous expression of TBC1D3 in cortical neural progenitor cells of developing mouse brains has been reported to result in a proliferation of outer radial glial cells and a cortical expansion and folding 23, likely through suppressing the histone methyltransferase .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 13 G9a41. Additionally, TBC1D3 expression has been reported to promote dendritic arborization and delay synaptogenesis in pluripotent stem cell-derived human cortical neurons, potentially associated with neoteny22. Furthermore, some publications have claimed an association between TBC1D3 and tissue repair 21. However, previous studies substantially lack mechanistic insight into TBC1D3 gene expression regulation. Hence, the discoveries in our study, especially the novel insights that NPEPPS -TBC1D3 readthrough transcripts are the predominant source of TBC1D3 expression in various cell types and the hypomethylated CpG island marks the active expression of the readthrough transcripts, bridge significant knowledge gaps in the multifaceted control of TBC1D3 expression. In addition to the regulatory mechanisms explored above, NPEPPS-TBC1D3 expression may also be regulated through other mechanisms. For instance, the poly(A) signal may play a role in transcription readthrough regulation 42. The truncation of NPEPPS -39M eliminates the canonical poly(A) signal used by NPEPPS -48M, which potentially contributes to the 39M transcriptional readthrough (Supplementary Fig. 24). Additionally, a recent paper reported that capped trans- RNAs can initiate translation at its target site43. Although that study did not identify a trans-RNA for NPEPPS-TBC1D3 in HEK293T cells, it would be worthwhile to investigate whether this mechanism operates in other, more relevant cell types. Current studies of TBC1D3 protein functions largely rely on exogenous overexpression systems, and endogenous TBC1D3 peptides are only occasionally detected in proteomics studies, primarily in testis (Supplementary Fig. 25a). Notably, TBC1D3 expression was often undetectable, unidentified, or understudied in previous brain and brain organoid proteomics studies 44–49. While one previous study did detect a TBC1D3 peptide DVVEVAGSWWAQER in brain and brain organoid samples 50, its signal intensity was marginal (Supplementary Fig. 25b). Based on the regulatory mechanisms elucidated in our study, we conclude that investigating TBC1D3's neurodevelopmental functions in a physiologically accurate manner necessitates supplementing current overexpression research with studies that account for intracellular protein levels and dynamics. This also calls for new paradigms in proteomics for studying proteins associated with complex genomic features. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 14 The insights provided by our study are critical not only for understanding TBC1D3 functions in neural development but also for enhancing cancer research. TBC1D3 was previously reported to be an oncogene which is “amplified” in prostate and breast cancers 51. Following that, a growing number of studies have suggested that TBC1D3 is overexpressed in various cancer types and could serve as a prognostic biomarker 52. However, the extensive CNV observed in the general population (Fig. 1 and Supplementary Fig. 1,2) has rarely been taken into consideration. Meanwhile, the accurate pinpointing of the exact paralogs involved in the oncogenic process remains to be achieved through comprehensive genomic investigation and standardized annotations. Here, we showcased that the NPEPPS-TBC1D3 readthrough transcripts may as well serve as the predominant source of TBC1D3 expression in certain cancer cells (Supplementary Fig. 26), which calls for extra attention in future cancer research. Notably, in the NPEPPS-TBC1D3 readthrough transcripts, while the NPEPPS sequence can be deemed as the 5’ UTR of TBC1D3 , on the other hand, the TBC1D3 sequence can be deemed as the 3’ UTR of NPEPPS as well. NPEPPS encodes the puromycin-sensitive aminopeptidase, implicated in peptide degradation, cell cycle control, and cell polarity 53–56. Certain ORFs within the NPEPPS region in some of the readthrough transcripts overlap with the coding sequence of the annotated Peptidase M1 membrane alanine aminopeptidase domain (Pfam, PF01433; p.280- 497, NP_006301) (Supplementary Fig. 27). Future study is required to determine whether these ORFs translate and how the TBC1D3 sequence affects their regulation. Finally, our analysis revealed the existence of 109 readthrough transcription gene pairs originating from joint SD pairs, suggesting a broad mechanism for the emergence of new functions. Future functional and biomedical investigations will be essential to further elucidate their biological significance.

CNVs and digenic structures of NPEPPS and TBC1D3 in the human population We identified populational CNVs of NPEPPS and TBC1D3 by aligning complete gene segments to populational assemblies using minimap2 (v2.30) with the parameters `-cx asm20 -- secondary=yes -p 0.001 -N 1000 --eqx -Y -K 8G -s 1000`. To resolve the genomic context of .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 15 individual copies and assign them to distinct gene clusters, we employed the NPEPPS-48M (the ~100-kbp full length copy) as the anchor. For each detected copy, we computed its relative genomic position with respect to this anchor and assigned it to a specific cluster based on the relative distance on T2T-CHM13. To detect digenic genomic structures in the human population, we aligned the 39M digenic genomic structure of T2T-CHM13 to the other assemblies and excluded the alignments that contained a large structural variant (>1 kbp) within the structure. To analyze the breakpoints of SDs, all identified gene copies are aligned to the NPEPPS-48M and TBC1D3-39M and samtools (v1.21) to visualize the alignment depths. All coordinates in the main text, if not otherwise stated, are according to T2T-CHM13. Isoform analysis and RNA-seq analysis Public iso-seq data are aligned to human and NHP genomes respectively using pbmm2 with alignment prefix `--preset ISOSEQ --sort`. Jbrowse2 (v3.5.1) is used to visualize and analyze the splicing and readthrough of the NPEPPS-TBC1D3 digenic structure. Cell samples for RNA-seq were snap frozen in RNA Extraction Solution (Servicebio, #G3013). RNA extraction, quality control, and Illumina NovaSeq sequencing were performed by Personalbio. We utilized the nf- core rnaseq pipeline 57 to quantify the expression level of TBC1D3 paralogs on both T2T-CHM13 and GRCh38 reference genome. To identify gene fusion/readthrough events on the whole genome using Iso-Seq data, we developed a tool called FusionSeekerPro (https://github.com/Zikun-Yang/FusionSeekerPro ) to call fusion events and identify the breakpoints. Fusion events supported by fewer than one reads and gene pairs in opposite orientations are excluded. Gene expression levels were quantified from long read Iso-Seq data using oarfish (v0.9.0) 58. We defined confident fusion events as those with ≥ 5 supporting reads and expression levels exceeding one. To assess the enrichment of JSDTR in the fusion events, we performed Fisher's exact test followed by Benjamini-Hochberg correction for multiple testing. The background gene pair set was defined as all same-strand gene pairs within the genome whose intergenic distances fell within the 95th percentile of distances observed in our confident fusion events. Cell culture .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 16 Cells were cultured at 37 /i3 with 5% CO2. HEK293T cells (National Collection of Authenticated Cell Cultures, NCACC, #SCSP-502) were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM; Servicebio, #G4515) supplemented with 10% fetal bovine serum (FBS; Servicebio, #G8003), and 1× Penicillin-Streptomycin (Servicebio, #G4003). SH-SY5Y cells (NCACC, #SCSP-5014) were maintained in DMEM/F-12 (Servicebio, #G4612) supplemented with 15% FBS, 1× Penicillin-Streptomycin. CN1 and C20 iPSC lines were established in Asian Pan- Genome Project Phase 1 (APGp1), and maintained in eTeSR (STEMCELL, #100-1215). For chimpanzee (Pan troglodytes) cell lines, blood samples were obtained from a 23-year-old female chimpanzee (Z2) and a 7-year-old female chimpanzee (D2) from the Shanghai Wild Animal Park colony in accordance with the corresponding IACUC protocol (No. SJTU-B2024001). Enriched chimpanzee peripheral blood mononuclear cells (PBMCs) were reprogrammed into induced pluripotent stem cells (iPSCs) under feeder-free conditions. NPCs were induced using STEMdiff SMADi Neural Induction Kit (STEMCELL, #08581) and maintained in STEMdiff Neural Progenitor Medium (STEMCELL, #05833). FBNs were induced using STEMdiff Forebrain Neuron Differentiation Kit (STEMCELL, #08600) and STEMdiff Forebrain Neuron Maturation Kit (STEMCELL, #08605). Generation of Chimpanzee iPSCs PBMCs were cultured in StemSpan II medium (STEMCELL, #09605) for 9 days to expand blood progenitor cells. On the day of reprogramming (day 0), the cells were transfected with the Epi5 Episomal iPSC Reprogramming Kit (Thermo Fisher Scientific, #A15960) using a Neon NxT electroporation system (Thermo Fisher Scientific). The cells were then plated in 2 mL of StemSpan II medium in a 6-well plate coated with Matrigel (Corning, #354277). On day 2, 1 mL of StemSpan II medium was added to each well without removing the existing medium. On days 3 and 5, 1 mL of ReproTeSR medium (STEMCELL, #05926) was added to the existing medium. Starting on day 7, the culture medium was completely replaced with 2 mL of ReproTeSR medium daily until day 20, when the first iPSC colonies were observed. Each colony was manually picked using a 200 µL pipette and transferred to new Matrigel-coated plates in mTeSR Plus medium (STEMCELL, #100-0276,) supplemented with 2 mM thiazovivin (MCE, #HY- 13257) on the first day after passage. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 17 5’ Rapid Amplification of cDNA Ends (5’ RACE) and TOPO cloning RNA was extracted from CN1 NPC and C20 NPC using the FastPure Complex Tissue/Cell Total RNA Isolation Kit (Vazyme, #RC113). 5’ RACE was performed with the HiScript-TS 5'/3' RACE Kit (Vazyme, #RA101). Two TBC1D3 Gene Specific Primers (GSPs) were utilized in this study: 5’-TTCGAAAAGGCTTAGG CCCCTTGTCC-3’ ( TBC1D3-Rev0) and 5’- CAGCTCCGTCTCATGTACAATCCCCAAA-3’ ( TBC1D3-Rev1). 5’ RACE products were purified with the FastPure Gel DNA Extraction Mini Kit (Vazyme, #DC301) and utilized for TOPO cloning using the Ultra-Universal TOPO Cloning Kit (Vazyme, #C603). Subsequently, the TOPO cloning reaction was transformed into the Fast-T1 competent cells (Vazyme, #C505). Ten colonies from each 5’ RACE reaction of each cell line were picked for Sanger sequencing using the corresponding 5’ RACE GSP as the sequencing primer. Additionally, the 5’ RACE products were also analyzed using Nanopore sequencing (GenScript). Single-cell analysis on human brain organoids Single-cell transcriptomic profiling was carried out on two previously established brain organoid lines (CN1 and C50) collected at differentiation days 42 and 98 (Han et al. , manuscript in preparation). Libraries were constructed with the 10× Genomics Chromium system. Sequencing reads were mapped to the T2T-CHM13v2.0 (GCF_009914755.1-RS_2023_10) 12 and cell-by- gene count matrices were produced using Cell Ranger (v7.2.0) 59. Standard quality-control filters were applied to exclude low-quality cells and genes following recommended practices60. Putative doublets were identified and removed using DoubletDetection (v4.2) 61. Preprocessed datasets were integrated, and batch effects were corrected with harmonypy (v0.0.10) 62,63. Downstream analysis was performed using Scanpy (v1.11.0) 64 and cell type annotations were assigned manually using canonical marker genes. Evolutionary history reconstruction in African great apes Human chromosome 17 and its nonhuman primate counterparts are aligned using minimap2, with secondary alignment allowed, and visualized by SVbyEye (v0.99.0). Estimated copy numbers of TBC1D3 and NPEPPS in primates are calculated by fastCN. To characterize the NPEPPS-CCL4 in bonobos and chimpanzees, Iso-Seq data were first examined, which indicated that readthrough transcripts spanning NPEPPS and CCL4 represent the predominant isoforms in .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 18 both species. The readthrough ratio is estimated as the proportion of readthrough-supporting split-read depth relative to the total read depth at a representative genomic position. For bonobo, the position with the maximal accumulation of readthrough-supporting split-reads was identified and used to estimate the readthrough ratio. However, when readthrough ratios were estimated similarly for chimpanzees, the ratios were substantially lower than expected based on Iso-Seq results. To investigate this discrepancy, bonobo readthrough-supporting split-reads and the genomic sequence of the bonobo CCL4 locus ( LOC100991964) were independently aligned to the chimpanzee genome. These cross-species mappings revealed that the CCL4 sequence of the bonobo NPEPPS-CCL4 splice junction corresponds to the terminal annotated exon of NPEPPS in chimpanzees, indicating differences in gene annotations between the two species. Based on the Iso-Seq evidence and the cross-species genomic alignment, we inferred that the direct estimation of the readthrough ratio from chimpanzee bulk RNA-seq data underestimates the true abundance of NPEPPS-CCL4 readthrough transcripts among all CCL4 transcripts. Therefore, to obtain a more accurate estimate of the readthrough ratio in chimpanzees, the orthologous position to the bonobo representative position was determined in the chimpanzee genome and used to estimate the ratio. Ribo-Seq analysis TBC1D3-only sequences and NPEPPS-TBC1D3 sequences were cloned from SH-SY5Y cDNA, which was synthesized as described above. 5’-ATGGACGTGGTAGAGGTCGC-3’ and 5’- CTAGAAGCCTGGAGGGAACTGAG-3’ were used as the matching sequences of the primers to amplify TBC1D3 -only sequences. 5’-CTAGAAGCCTGGAGGGAACTGAG-3’ (the reverse primer), 5’-CGCCTCCTTCCCAACCCC-3’(the forward primer for NT1-1 and NT1-2), and 5’- ATGAATTGTGCTGATATTGATATTATTACAGC-3’ (the forward primer for NT2-1, NT2-3, NT2-4, and NT2-5). PCR reactions were performed with Q5 High-Fidelity DNA Polymerase (NEB, #M0491) and assembly reactions were performed with NEBuilder HiFi DNA Assembly Master Mix (NEB, #E2621), following the manuals. 2.1~2.6 ug of each plasmid was transfected to HEK293T plated in a 6-well plate well using Lipomaster 3000 (Vazyme, #TL301), ensuring the same amount of molecules was used. Cells were treated with cycloheximide (final conc. 0.1 mg / mL; Selleck, #S7418) and sent for Ribo-seq (Novogene) on Day 4 post-transfection. bowtie2 (v2.5.4) was utilized to map the sequencing reads to the rRNA sequences from T2T- .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 19 CHM13v2.0 in order to achieve the target sequences with the parameter ‘--un-gz’. Then the cleaned reads were aligned to the manual annotated sequences and T2T-CHM13v2.0 using bowtie2. Immunoblotting Cells were washed once with PBS (Servicebio, #G4202) and lysed in RIPA buffer (Sigma- Aldrich, #R0278) supplemented with cOmplete Protease Inhibitor Cocktail (Roche, #04693116001). The lysates were incubated on ice on a shaker for 15 min, followed by centrifugation to collect the supernatant (16,000 × g for 30 min at 4 °C). Protein concentration was determined using the BCA Protein Quantitative Detection Kit (Servicebio, #G2026). A total of 15 μ g of protein per sample was loaded onto an 10% SDS–PAGE gel (Genefist, #GF1820- 10). After electrophoresis, proteins were transferred onto a PVDF membrane (Servicebio, #G6044-0.45) and blocked for 25 min with Protein Free Rapid Blocking Buffer (Servicebio, #G2052). The membrane was washed with TBST and incubated overnight at 4 °C on a shaker with the following primary antibodies: GFP (abcam, #ab290, 1:5000), ACTB (HUABIO, #EM21002, 1:20000), and TBC1D3 (Santa Cruz Biotechnology, #sc-376073, 1:100). After washing with TBST, the membrane was incubated with HRP-labeled goat anti-mouse IgG (Servicebio, #GB23301, 1:5000) for 1 h at room temperature on a shaker, and then washed again. The protein bands were imaged using Clinx ChemiScope 6000. Data Availability Genome assemblies The T2T primate genomes used in this study are available from GenBank via accessions: GCA_009914755.4, GCA_028858775.2, GCA_028885625.2, GCA_028885655.2, GCA_029281585.2, GCA_029289425.2 and GCA_037993035.1. The T2T primate genome assemblies are also available on GitHub (https://github.com/marbl/Primates and https://github.com/zhang-shilong/T2T-MFA8). We used the human assemblies from Human Pangenome Reference Consortium (HPRC), Human Genome Structural Variation Consortium (HGSVC) and Asian Pan-Genome Project (APG). .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 20 Transcriptome data Iso-seq generated from the human CHM13hTERT cell line can be accessed by SRR12519035 and SRR12519036. Nonhuman primate Iso-seq data are deposited under BioProject identifiers PRJNA902025, PRJNA1016395 and PRJNA1041301. Raw bulk RNA-seq data for chimpanzee, gorilla, and macaque tissues were obtained from published studies 65–68 (PRJNA1004471, PRJNA143627, PRJNA304995, and PRJNA236446). Other data Other data generated in this study, including the RNA-seq results of the CN1 cell lines and the chimpanzee iPSCs, the 5’ RACE sequencing data, the Iso-Seq data of CN1 and C20 NPCs, and the Ribo-Seq data, are available on request from the corresponding authors following the regulatory guidelines. Single-cell RNA-seq data for organoid lines CN1 and C50 at differentiation days 42 and 98 are presented in Han et al. (manuscript in preparation) and are available from the authors on request. Code Availability The custom scripts used in this study are available on GitHub (https://github.com/YafeiMaoLab/TBC1D3_analysis ).

1. Bailey, J. A., Yavor, A. M., Massa, H. F., Trask, B. J. & Eichler, E. E. Segmental duplications: organization and impact within the current human genome project assembly. Genome Res. 11, 1005–1017 (2001). 2. Fawcett, J. A. & Innan, H. The role of gene conversion in preserving rearrangement hotspots in the human genome. Trends Genet. 29, 561–568 (2013). 3. Fiddes, I. T. et al. Human-Specific NOTCH2NL Genes Affect Notch Signaling and Cortical Neurogenesis. Cell 173, 1356-1369.e22 (2018). 4. McCartney, A. M. et al. Gene fusions derived by transcriptional readthrough are driven by segmental duplication in human. Genome Biol. Evol. 11, 2678–2690 (2019). 5. Soto, D. C. et al. Human-specific gene expansions contribute to brain evolution. Cell (2025) doi:10.1016/j.cell.2025.06.037. 6. Florio, M., Namba, T., Pääbo, S., Hiller, M. & Huttner, W. B. A single splice site mutation in human-specific ARHGAP11B causes basal progenitor amplification. Sci. Adv. 2, e1601941 (2016). .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 21 7. Patowary, A. et al. Developmental isoform diversity in the human neocortex informs neuropsychiatric risk mechanisms. Science 384, eadh7688 (2024). 8. Moses, C. et al. Evolutionary hotspots of structural variation drive inter-individual differences in the expression of fusion transcripts in the human brain. BioRxiv (2025) doi:10.1101/2025.03.11.642572. 9. Wu, J. et al. SOAPfusion: a robust and effective computational fusion discovery tool for RNA-seq reads. Bioinformatics 29, 2971–2978 (2013). 10. Liao, W.-W. et al. A draft human pangenome reference. Nature 617, 312–324 (2023). 11. Logsdon, G. A. et al. Complex genetic variation in nearly complete human genomes. Nature 644, 430–441 (2025). 12. Nurk, S. et al. The complete sequence of a human genome. Science 376, 44–53 (2022). 13. Yang, C. et al. The complete and fully-phased diploid genome of a male Han Chinese. Cell Res. 33, 745–761 (2023). 14. Karsten, S. L. et al. A genomic screen for modifiers of tauopathy identifies puromycin- sensitive aminopeptidase as an inhibitor of tau-induced neurodegeneration. Neuron 51, 549–560 (2006). 15. Kudo, L. C. et al. Puromycin-sensitive aminopeptidase (PSA/NPEPPS) impedes development of neuropathology in hPSA/TAU(P301L) double-transgenic mice. Hum. Mol. Genet. 20, 1820–1833 (2011). 16. Ren, G. et al. Cu, Zn-superoxide dismutase 1 (SOD1) is a novel target of Puromycin- sensitive aminopeptidase (PSA/NPEPPS): PSA/NPEPPS is a possible modifier of amyotrophic lateral sclerosis. Mol. Neurodegener. 6, 29 (2011). 17. Alkan, C. et al. Personalized copy number and segmental duplication maps using next- generation sequencing. Nat. Genet. 41, 1061–1067 (2009). 18. Jeong, H. et al. Structural polymorphism and diversity of human segmental duplications. Nat. Genet. 57, 390–401 (2025). 19. Wainszelbaum, M. J. et al. The hominoid-specific oncogene TBC1D3 activates Ras and modulates epidermal growth factor receptor signaling and trafficking. J. Biol. Chem. 283, 13233–13242 (2008). 20. Frittoli, E. et al. The primate-specific protein TBC1D3 is required for optimal macropinocytosis in a novel ARF6-dependent pathway. Mol. Biol. Cell 19, 1304–1316 (2008). 21. Qin, S. et al. TBC1D3 regulates the payload and biological activity of extracellular vesicles that mediate tissue repair. FASEB J. 33, 6129–6139 (2019). 22. Dong, J. et al. A hominoid-specific signaling axis regulating the tempo of synaptic maturation. Cell Rep. 43, 114548 (2024). 23. Ju, X.-C. et al. The hominoid-specific gene TBC1D3 promotes generation of basal neural progenitors and induces cortical folding in mice. eLife 5, (2016). 24. Guitart, X. et al. Independent expansion, selection, and hypervariability of the TBC1D3 gene family in humans. Genome Res. 34, 1798–1810 (2024). 25. Paulding, C. A., Ruvolo, M. & Haber, D. A. The Tre2 (USP6) oncogene is a hominoid- specific gene. Proc Natl Acad Sci USA 100, 2507–2511 (2003). 26. Bekpen, C. & Tautz, D. Human core duplicon gene families: game changers or game players? Brief. Funct. Genomics 18, 402–411 (2019). 27. Yoo, D. et al. Complete sequencing of ape genomes. Nature 641, 401–418 (2025). 28. Zhang, S. et al. Integrated analysis of the complete sequence of a macaque genome. Nature 640, 714–721 (2025). .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 22 29. Resch, A. M., Ogurtsov, A. Y., Rogozin, I. B., Shabalina, S. A. & Koonin, E. V. Evolution of alternative and constitutive regions of mammalian 5’UTRs. BMC Genomics 10, 162 (2009). 30. Leppek, K., Das, R. & Barna, M. Functional 5’ UTR mRNA structures in eukaryotic translation regulation and how to find them. Nat. Rev. Mol. Cell Biol. 19, 158–174 (2018). 31. Lee, D. S. M. et al. Disrupting upstream translation in mRNAs is associated with human disease. Nat. Commun. 12, 1515 (2021). 32. Zhong, Z., Li, Y., Sun, Q. & Chen, D. Tiny but mighty: Diverse functions of uORFs that regulate gene expression. Comput. Struct. Biotechnol. J. 23, 3771–3779 (2024). 33. Wieder, N. et al. Differences in 5’untranslated regions highlight the importance of translational regulation of dosage sensitive genes. Genome Biol. 25, 111 (2024). 34. Dougherty, M. L. et al. Transcriptional fates of human-specific segmental duplications in brain. Genome Res. 28, 1566–1576 (2018). 35. E ş iyok, N. et al. A dyad of human-specific NBPF14 and NOTCH2NLB orchestrates cortical progenitor abundance crucial for human neocortex expansion. Sci. Adv. 11, eads7543 (2025). 36. Dishuck, P. C. et al. Structural variation, selection, and diversification of the NPIP gene family from the human pangenome. Cell Genomics 5, 100977 (2025). 37. Moses, C. et al. Evolutionary hotspots of structural variation drive inter-individual differences in the expression of fusion transcripts in the human brain. Nucleic Acids Res. 53, (2025). 38. Cuscó, I. et al. Practical guidelines to manage discordant situations of SMN2 copy number in patients with spinal muscular atrophy. Neurol. Genet. 6, e530 (2020). 39. Brkušanin, M. et al. Joint effect of the SMN2 and SERF1A genes on childhood-onset types of spinal muscular atrophy in Serbian patients. J. Hum. Genet. 60, 723–728 (2015). 40. Ottesen, E. W. et al. Diverse targets of SMN2-directed splicing-modulating small molecule therapeutics for spinal muscular atrophy. Nucleic Acids Res. 51, 5948–5980 (2023). 41. Hou, Q.-Q., Xiao, Q., Sun, X.-Y., Ju, X.-C. & Luo, Z.-G. TBC1D3 promotes neural progenitor proliferation by suppressing the histone methyltransferase G9a. Sci. Adv. 7, (2021). 42. Caldas, P. et al. Transcription readthrough is prevalent in healthy human tissues and associated with inherent genomic features. Commun. Biol. 7, 100 (2024). 43. Jia, L. et al. Programmable initiation of mRNA translation by trans-RNA. Nat. Biotechnol. (2025) doi:10.1038/s41587-025-02897-1. 44. Dakic, V. et al. Short term changes in the proteome of human cerebral organoids induced by 5-MeO-DMT. Sci. Rep. 7, 12863 (2017). 45. Nascimento, J. M. et al. Proteomic signatures of schizophrenia-sourced iPSC-derived neural cells and brain organoids are similar to patients’ postmortem brains. Cell Biosci. 12, 189 (2022). 46. Zhao, X. et al. Changes in and asymmetry of the proteome in the human fetal frontal lobe during early development. Commun. Biol. 5, 1031 (2022). 47. Goto-Silva, L. et al. Quantitative profiling of axonal guidance proteins during the differentiation of human neurospheres. Biochim. Biophys. Acta Proteins Proteom. 1869, 140656 ( 2021). 48. Chen, C. et al. Trans-omic profiling uncovers molecular controls of early human cerebral organoid formation. Cell Rep. 43, 114219 (2024). .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 23 49. Melliou, S. et al. Regionally defined proteomic profiles of human cerebral tissue and organoids reveal conserved molecular modules of neurodevelopment. Cell Rep. 39, 110846 (2022). 50. Nascimento, J. M. et al. Human cerebral organoids and fetal brain tissue share proteomic similarities. Front. Cell Dev. Biol. 7, 303 (2019). 51. Pei, L. et al. PRC17, a novel oncogene encoding a Rab GTPase-activating protein, is amplified in prostate cancer. Cancer Res. 62, 5420–5424 (2002). 52. Wang, B., Chen, D. & Hua, H. TBC1D3 family is a prognostic biomarker and correlates with immune infiltration in kidney renal clear cell carcinoma. Mol. Ther. Oncolytics 22, 528–538 (2021). 53. Constam, D. B. et al. Puromycin-sensitive aminopeptidase. Sequence analysis, expression, and functional characterization. J. Biol. Chem. 270, 26931–26939 (1995). 54. Peer, W. A. The role of multifunctional M1 metallopeptidases in cell cycle progression. Ann. Bot. 107, 1171–1181 (2011). 55. Osana, S. et al. Puromycin-sensitive aminopeptidase is required for C2C12 myoblast proliferation and differentiation. J. Cell. Physiol. 236, 5293–5305 (2021). 56. Saturno, D. M. et al. Sustained centrosome-cortical contact ensures robust polarization of the one-cell C. elegans embryo. Dev. Biol. 422, 135–145 (2017). 57. Ewels, P. A. et al. The nf-core framework for community-curated bioinformatics pipelines. Nat. Biotechnol. 38, 276–278 (2020). 58. Zare Jousheghani, Z., Singh, N. P. & Patro, R. Oarfish: enhanced probabilistic modeling leads to improved accuracy in long read transcriptome quantification. Bioinformatics 41, i304– i313 (2025). 59. Zheng, G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017). 60. Germain, P.-L., Sonrel, A. & Robinson, M. D. pipeComp, a general framework for the evaluation of computational pipelines, reveals performant single cell RNA-seq preprocessing tools. Genome Biol. 21, 227 (2020). 61. Gayoso, A. & Shor, J. JonathanShor/DoubletDetection: doubletdetection v4.2. Zenodo (2022) doi:10.5281/zenodo.6349517. 62. Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 16, 1289–1296 (2019). 63. Slowikowski, K., Arevalo, J. & Manning, J. slowkow/harmonypy: harmonypy version 0.0.10. 64. Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Ge nome Biol. 19, 15 (2018). 65. Mao, Y.-X. et al. Comparative transcriptome analysis between rhesus macaques ( Macaca mulatta) and crab-eating macaques ( M. fascicularis). Zool. Res. 45, 299–310 (2024). 66. Sousa, A. M. M. et al. Molecular and cellular reorganization of neural circuits in the human lineage. Science 358, 1027–1032 (2017). 67. Brawand, D. et al. The evolution of gene expression levels in mammalian organs. Nature 478, 343–348 (2011). 68. Tomaszkiewicz, M. et al. A time- and cost-effective strategy to sequence mammalian Y Chromosomes: an application to the de novo assembly of gorilla Y. Genome Res. 26, 530–540 (2016). .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 24

We would like to acknowledge the Human Pangenome Reference Consortium (BioProject ID: PRJNA730823), the National Human Genome Research Institute (NHGRI), Human Genome Structural Variation Consortium (HGSVC), Asian Pan-genome project (APG), and Primate T2T Consortium for providing long-read human and great ape genome assemblies. We thank Dr. Lin Ge (Beijing Children’s Hospital) for the suggestions regarding SMA. This work is supported by the Shanghai Rising-Star Program for studying “evolutionary medicine and functional studies of the human-specific NPEPPS duplications” (24YF2721800 to K.M.). This work is in part supported by the Shanghai Magnolia Talent Plan Pujiang Project (24PJA049) and the Shanghai Post-doctoral Excellence Program (2024338) to K.M., and by Shanghai Jiao Tong University Medical-Engineering Interdisciplinary Research Fund (grant no. YG2025QNA47) to X.Y. This work is in part supported by the National Natural Science Foundation of China (32130035) to Z.- G. L. This work is in part supported by the Scientific Research Innovation Capability Support Project for Young Faculty (SRICSPYF-ZY2025101), the National Key Research and Development Program of China (2025YFC3410300), the National Natural Science Foundation of China (32370658), the Natural Science Foundation of Chongqing, China (CSTB2024NSCQ- JQX0004), the New Cornerstone Science Foundation through the XPLORER PRIZE, Shanghai Jiao Tong University (SJTU) 2030 Initiative (WH510363003/016), Yongxin Youth Award Fund and Zhongying Young Scholars Program to Y.M. Author Information Contributions K.M., Z.Y., and Y.M. conceptualized the study. K.M., Z.Y., and Z.L. contributed equally in terms of invested time and efforts, and therefore each has the right to list their name first in their respective curriculum vitae; we petition the grant funding bodies to treat them as true equals. K.M., Z.Y., Z.L., J.G., D.L., Z.W., H.M., S.Z., L.F., H.L., and X.J. performed the 5’ RACE and the transcription/translation analyses. Z.Y., K.M., Z.L., J.G., S.Z., L.F., G.L., J.C., J.Z., X.Y., G.Z., and Y.M. performed the population, evolution, and expression analyses. Z.L., K.M., Z.W., Q.X, C.Y., and Z.-G. L. conducted the functional studies. X.Y. generated the iPSCs, with P.L.’s assistance. Z.Y. developed the FusionSeekerPro tool. G.Z. initiated the APG project. K.M., Z.Y., .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 25 and Z.L. wrote the original draft of the manuscript. K.M. and Y.M. edited the manuscript. All authors reviewed the manuscript. Corresponding authors Correspondence to Kaiyue Ma and Yafei Mao .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 26 Fig. 1: The joint SD-derived NPEPPS-TBC1D3 digenic genomic structure is fixed across populations and generates NPEPPS-TBC1D3 readthrough transcripts. a, Ideogram illustrating the genomic locations of TBC1D3 and NPEPPS genes on human chromosome 17. Red, yellow and blue triangles represent ancestral NPEPPS, derived NPEPPS and TBC1D3, respectively. b, c, d Analysis of NPEPPS segmental duplication breakpoints (b) and copy number distribution grouped by superpopulation (c) and genomic location (d). e, f, g Analysis of TBC1D3 segmental duplication breakpoints (e) and copy number distribution grouped by superpopulation (f) and genomic location (g). h, Copy number variation of the digenic structure across human populations shows at least one digenic structure per haplotype. i, Readthrough splicing profile from CHM13hTERT Iso-seq data. Curves represent splicing .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 27 junctions, with numbers above indicating the number of supporting reads. Population abbreviations: SAS, Southern Asian (n=14 haplotypes). EUR, European (n=13 haplotypes). EAS, Eastern Asian (n=340 haplotypes). AMR, American (n=26 haplotypes). AFR, African (n=68 haplotypes). .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 28 Fig. 2: The NPEPPS-TBC1D3 transcripts, specifically those from the 39M locus, are the predominant source of TBC1D3 expression in various cell types. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 29 a, RNA was extracted from CN1 NPC and C20 NPC, and 5’ RACE was performed to determine the transcription starting sites of the transcripts using the TBC1D3 Gene Specific Primer. 5’ RACE products were evaluated with both TOPO-Sanger sequencing or nanopore sequencing. b , Proportions of confirmed readthrough and other reads detected by 5’ RACE-nanopore in C20 NPC, CN1 NPC, and HEK293T cells, along with JBrowse2 read cloud displays showing split- read support at the NPEPPS-TBC1D3 junction, confirming the predominance of readthrough transcripts. c, The CpG island annotation and methylation levels of the 48M, 39M and 37M loci. At the 48M and the 39M loci, the hypomethylated CpG island resides at the upstream of NPEPPS. In contrast, the 37M locus does not harbor such a CpG island. d, The mapping quality (MAPQ) of 23 reconstructed transcript sequences from TOPO-cloning demonstrates that readthrough transcript sequences are more similar to the 39M sequence than the 37M sequence, indicating the readthrough expression is mainly contributed by the 39M locus. e, Single-cell analysis of human brain organoid samples (CN1 & C50, Day 42 & Day 98, combined) shows TBC1D3-39M is the predominant paralog for TBC1D3 expression. f, The Tajima’s D and pA/pS of NPEPPS and TBC1D3 duplicates indicates that TBC1D3-39M is under selection on both nucleotide and amino acid level in the human population (n=462 haploids). Digenic, the NPEPPS-TBC1D3 digenic region. NC, negative control, representing the TBC1D3 copies excluding TBC1D3-39M. g, Nucleotide diversity ( π ) and Tajima's D analysis of TBC1D3 -39M based on HGSVC3-called single nucleotide variants (n=65). No significant reduction in nucleotide diversity was observed in African populations (p=0.439) or non-African populations (p=0.685). Similarly, no significant deviation in Tajima's D values was detected in African populations (p=0.204) or non-African populations (p=0.113), when compared to genome-wide averages calculated from 80 kbp sliding windows across chromosome 17. This result, however, is more likely a product of methodological limitations than an actual evolutionary trend. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 30 Fig. 3: Evolutionary analyses reveal that the intact NPEPPS-TBC1D3 sequence is human- specific and the NPEPPS readthrough occurs with different genes in different primates. a, Synteny comparison of human chromosome 17 and its NHP counterparts illustrates the evolutionary trajectory of NPEPPS and TBC1D3 . b, Copy number estimation based on FastCN shows the expansion of NPEPPS copies in African great apes and expansions of TBC1D3 copies in different lineages. c, Iso-Seq reveals expression of readthrough transcripts, including NPEPPS-TBC1D3 in humans and NPEPPS -CCL4 in bonobos and chimpanzees. d, DNA CpG methylation analysis indicates that hypomethylated CpG islands exist in the upstream of the NPEPPS, which is syntenic to human NPEPPS-39M, in chimpanzees and bonobos. The NPEPPS sequence in this syntenic region in gorillas is substantially truncated and not annotated as a gene by RefSeq. Consequently, no such hypomethylation signal is observed in gorillas. Met., Methylation level. e, Expression levels of duplicated CCL4 gene copies in bonobo and chimpanzee. In each species, one copy forms readthrough transcripts with the upstream NPEPPS .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 31 gene (NPEPPS-LOC100991964 in bonobos and NPEPPS-LOC455004 in chimpanzees), whereas the other copy does not. Individual dots represent independent bulk RNA-seq samples. The bonobo data were obtained from publicly available prefrontal cortex bulk RNA-seq datasets (PRJEB33938). The chimpanzee data were generated in this study using NPCs, including two independent differentiation batches from the individual SY2-Z2 and one batch from the individual SY2-D2. In both species, the NPEPPS-CCL4 readthrough-associated copy shows higher expression compared with the other CCL4 copy. f , Estimated proportion of NPEPPS- CCL4 readthrough transcripts relative to all CCL4 transcripts at the readthrough loci in bonobos and chimpanzees. The readthrough ratio is estimated as the proportion of readthrough-supporting split-read depth relative to the total read depth at a representative genomic position. Each dot represents an independent bulk RNA-seq sample. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 32 Fig. 4: The NPEPPS-TBC1D3 transcripts fine-tune the TBC1D3 protein level, and joint-SD- related readthrough events broadly exist. a, (NPEPPS-) TBC1D3 transcripts were cloned into a pCAGGS expression system. The position of the epitope of the anti-TBC1D3 antibody is indicated. TB2 represents the MANE Select isoform. b,c, The readthrough transcripts generate lower TBC1D3 protein level compared to TBC1D3-only sequences. NT2-3 and NT2-4 are expected to generate no band because of the lack of the epitope sequence. Blue arrows: TBC1D3 bands. ANOVA and Dunnett's multiple comparisons tests were performed for TB2 vs. each of the other samples. d , Ribo-Seq was performed for TB2, NT1-2, and NT2-5. Regularly distributed peaks were observed in the whole span of the TBC1D3 ORF but not the whole span of the NPEPPS ORFs. e, Benchmark between our long-read-based fusion calling results and the short-read-based results from Moses et al. , .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint 33 2025 shows that NPEPPS-TBC1D3 is a commonly identified gene fusion pair. f , Enrichment analysis between SD-associated gene pairs and fusion gene pairs using Fisher’s exact test and Benjamini-Hochberg method shows that transcriptional readthrough events exhibit significant enrichment of JSDTR gene pairs (p-value=1.65e-17). g , Diagram of SMG1P3 -NPIPB3 localization, methylation and representative reads on chr16. h, Diagram of SERF1A /B-SMN1/2 localization, methylation and representative reads. Among all readthrough reads, 30.2% (13/43) and 60.5% (26/43) terminate respectively a few bases after exon 3 (chr5:71,399,360 for SMN1; chr5:70,820,191 for SMN2) and after exon 6 (chr5:71,403,067 for SMN1; chr5:70,816,484 for SMN2). 6.98% (3/43) and 2.32% (1/43) readthrough reads terminate at exon 8 with exon 7 inclusion and skipping, respectively. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 15, 2026. ; https://doi.org/10.64898/2026.01.14.699191doi: bioRxiv preprint