Enhancing Variant Calling in Whole Exome Sequencing (WES) Data Using Population-Matched Reference Genomes

25 Whole exon sequencing (WES) data are frequently used for cancer diagnosis and genome-wide 26 association studies (GWAS), hinging upon high -coverage read mapping, informative variant 27 calling, and high-quality reference genomes. The center position of the currently used genome 28 assembly, GRCh38, is now challenged by two newly publicized telomere -to-telomere or T2T 29 genomes, T2T-CHM13 and T2T-YAO, and it becomes urgent to have a comparative study to 30 test population specificity using the three reference genomes based on real case WES data. We 31 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint here report our analysis along this line for 19 tumor samples collected from Chinese patients. 32 The primary comparison of the exon regions among the three references reveals that the 33 sequences in up to ~1% target regions in YAO a re widely diversified from GRCh38 and may 34 lead to off-target in sequence capture. However, YAO still outperforms GRCh38 genomes by 35 obtaining 7.41% more mapped reads. Due to more reliable read -mapping and closer 36 phylogenetic relationship with the samples than GRCh38, YAO reduces half of variant calls of 37 clinical significance which are mostly benign while keeping sensitivity in identifying 38 pathogenic variants. YAO also outperforms CHM13 in reducing calls of Chinese -specific 39 variants. Our findings highlight th e critical need for employing population -specific reference 40 genomes in genomic analysis to ensure accurate variant analysis and the significant benefits of 41 tailoring these approaches to the unique genetic backgrounds of each ethnic group. 42 43

T2T-YAO, population-specific reference genome, whole exome sequencing, 44 variant calling 45 46 Main Text 47 48

49 Next-generation sequencing (NGS) has been extensively employed in a wide spectrum of 50 clinical applications [1, 2]. More and more practice of precision medicine, including diagnosis, 51 prognosis, and therapy selection across genetic disorders, oncology, and infectious diseases 52 relies on sequencing of human genome DNA [3, 4]. Both whole genome sequencing (WGS) 53 and whole exome sequencing (WES) are widely used to identify genetic (germline) or somatic 54 (such as in tumor tissues) variations in helping genetic disorder diagnosis or discovering novel 55 tumor antigens [2, 5-7]. The WES, which only sequences the protein -coding region (about 1-56 2 % of the whole human genome) by targeting enrichment, costs much less and is more widely 57 applied clinically [8]. 58 59 For human and other animals with large genomes, analyses of high -throughput data start with 60 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint mapping sequencing reads against a reference genome, which is the found ation for all 61 resequencing data analyses for biomedical research and clinical applications. As such, pursuing 62 a complete and accurate human genome reference has been a long -lasting goal for the society 63 of biomedicine. The Genome Reference Consortium (GRC) has continuously improved the 64 human reference genome from the first version by Human Genome Project in 2001 to the up -65 to-date GRCh38 released in 2013 [9-11]. In 2022, advents the first complete human genome 66 haplotype—CHM13 which is a telomere-to-telomere (T2T) assembly of the European ancestry 67 genome from a hydatidiform mole-CHM13 with unprecedented high-quality of Q73.94, i.e., it 68 has less than one error in 24.8 -Mbp sequence [12]. Next year, the complete sequence of 69 chromosome Y from the HG002 genome (European Jewish ancestry) is added to cover all 70 chromosomes (22+XY) of human, leading to T2T-CHM13 v2.0 [13]; independently, our group 71 completed the assembly of the diploid human genome T2T -YAO, based on data from a trio 72 from Han Chinese ancestry, and achieved a comparable high quality of a haplotype version—73 YAO-hp (Q74.69, i.e., one error in 29.5-Mbp sequence) [14]. And more efforts have been made 74 to create reference genome for the Han population, including Han1 [15] and CN1 [16], with 75 lower quality. Furthermore, a draft human pangenome reference [17] and a comprehensive 76 pangenome reference encompassing 36 Chinese populations have been developed [18], 77 providing valuable resources for understanding genetic diversity across different populations. 78 79 It is presumed that the high quality and completeness of human reference genome will improve 80 the accuracy in read-mapping and variant-calling for the analysis of high throughput sequencing 81 data [19]. A reference genome of closer phylogenetic relations hip will theoretically abate the 82 number of unmapped reads and improve mapping quality by reducing ambiguous mapping of 83 reads with mismatches. Given the great degree of global genetic variation, reference genomes 84 representative of populations is necessary f or effectively performing omics analyses on those 85 populations [14, 20 -23]. CHM13 has been publicized as a major improvement from the 86 currently used GRCh38, while YAO is closer to Chinese and of comparable quality to CHM13, 87 potentiating improvements in genomic analysis for Chinese by substituting the current GRCh38 88 reference. However, the improvement in using higher -quality reference genome with closer 89 phylogenetic relationship has not yet been quantified especially for samples from Chinese. 90 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint 91 To evaluate the improvement provided by new reference genomes, we designed a study to 92 quantitatively assess the differences among three genomes when analyzing Whole Exome 93 Sequencing (WES) data from Chinese samples. We selected WES over Whole Genome 94 Sequencing (WGS) because WES, or targeted sequencing of gene panels, is the most prevalent 95 practice in clinical personalized medicine. The impact of different reference genomes on this 96 specific application, as well as the bias introduced by capture probes designed w ith GRCh38 97 for WES or panel sequencing in Chinese populations, remains largely unexplored. Previous 98 studies have investigated the performance of various references using standard benchmark 99 genomes, such as HG002 and HG005, as well as WGS data from public p opulation datasets 100 [16, 19], leading us to avoid redundant analyses. In this study, we use the complete human 101 haplotypes of T2T-YAO-hp, T2T-CHM13 v2.0, and GRCh38 (excluding decoy genome), all of 102 which include a single copy of 22+XY chromosomes. For brevity, these references are referred 103 to as YAO, CHM13, and GRCh38. We first compared the basic statistics of these references, 104 particularly their coverage of exomes. We then utilized a WES dataset from 19 gastric tumor 105 samples from Han Chinese, aligning the data separately against each of the three references for 106 initial evaluation. 107 108 The reliance on GRCh38 of current standard variant calling processes, despite extensively 109 optimized and evaluated, warrants reassessment when using alternative references. 110 Consequently, we compared the performance of three reference genomes in parallel, analyzing 111 each step of the variant calling process—from mapping to raw variants and final variants after 112 filtering with default cutoffs. Variants in homozygous, heterozygous, and som atic categories 113 were compared both in whole genome (target and flanking regions) and only in target regions. 114 Significant differences were observed across all comparison matrices when using different 115 references. Although this study did not achieve an optimized procedure for WES analysis using 116 alternative reference to GRCh38, our results highlight the urgent need for establishing 117 population-specific reference genomes for Chinese populations. 118 119 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint

and Discussion 120 Basic statistics of YAO in comparison to GRCh38 and CHM13. 121 The lengths of the three genome assemblies are as follows: the longest is 3,117,292,070 bp for 122 CHM13, followed by 3,088,286,401 bp for GRCh38 (including 150,630,719 Ns), and the 123 shortest is 3,062,724,542 bp for YAO (Table S1). Among these references, YAO is derived from 124 a real individual, whereas CHM13 and GRCh38 are not, with differences in length of less than 125 2%. Variability in chromosome length is well -documented and is primarily attributed to the 126 expansion and contraction of highly repe titive regions, particularly in centromeric and 127 heterochromatic areas. Notable examples include the megabase -long expansion on 128 chromosome 9 in CHM13 [12] and the extensive length diversity observed on chromosome Y 129 [24]. The GC content, which represents the fraction of guanine (G) and cytosine (C) nucleotides, 130 varies across different regions of the human genome and plays a significant role in the 131 efficiency of Illumina sequencing and subsequent analysis. YAO and CHM13 have similar GC 132 content, 40.75% and 40.7 9%, respectively, slightly lower than that of GRCh38 (41.59%), 133 possibly due to the fully -filled sequences of the relatively AT-rich centromere regions in the 134 two better-assembled genomes. Based on the up-to-date annotation files [25-27], the collective 135 exon lengths (including exons of both protein -coding and non -coding genes) of the three 136 genomes are the longest GRCh38, 156,332,309 bp (5.062% of the genome length), the next 137 YAO, 156,053,407 bp (5.095% of the genome length), and the shortest CHM13, 153,061,9 25 138 bp (4.910% of the genome length). 139 140 To facilitate the subsequent comparison of WES dataset analysis across the three references, 141 we focused on the exon regions of protein -coding genes targeted by Agilent kit of SureSelect 142 Human All Exon V6 and lifted th eir original coordinates in GRCh37 to the three reference 143 genomes (Table S2). Of the 243,190 targeting regions in a collective length of 60,700,153 bp 144 in GRCh37, 99% can be successfully lifted to all three references. There are only 1,700 145 uncertain regions in GRCh37, either mapped to multiple sites or unmappable to CHM13 and 146 YAO, which are more than the unmappable 1,281 regions in GRCh38 (Table S2). Nevertheless, 147 all three reference genomes retain >60Mb total targetable exon sequences, and the difference 148 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint among them is rather relatively neglectable. We also calculated sequence identity for each lifted 149 region as YAO vs. CHM13 and YAO vs. GRCh38. Although ~85% of the lifted regions are 150 strictly conserved to show 100% identity, there are ~0.6% regions with sequence identity <80% 151 (1,602 regions to GRCh38 and 1437 regions to CHM13, Fig. 1). Together with the 1,705 failed 152 regions, 1-2% of the targetable regions where the capture probes are designed according to the 153 GRCh37/38 genome do not match the samples from Chinese individuals, suggesting potential 154 underrepresentation in their WES data of these regions. 155 156 WES data and alignment to the references 157 158 A collection of 19 paraffin embedded gastric tumor samples, 9 benign gastric stromal tumors 159 and 10 malignant gastric cancers, from Han Chinese patients in Linfen Central Hospital were 160 applied to DNBSEQ -T7 platform for 150 bp pair -end WES sequencing ( Table S3). Data 161 analysis followed the process shown in Fig S1. The quality of sequencing reads has an average 162 of 94.8% of Phred value >Q30 and the average sequencing yield is 17.7 ± 5.05 Gb after 163 trimming off bases below Q20, equal to ~ 300 × sequencing depth of t he target regions. The 164 difference between the two groups is not significant in both sequencing yield (19.45 ± 2.06 Gb 165 vs. 18.69 ± 5.14 Gb, p=0.691, t-test) and quality (Q33.3 ± 0.68 vs. Q33.7 ± 0.71, p=0.815, t-166 test). 167 168 The first step in NGS data analysis is aligning the sequencing reads against a reference genome. 169 It is well known that a small percentage of the sequencing reads cannot be mapped to the human 170

genome in practical analysis due to incompleteness and misassembling of reference, 171 and it has been suggested that improving the human reference genome may also improve the 172 alignment rate [19]. We first mapped the clean sequencing data separately to Y AO, CHM13, 173 and GRCh38 and compared their mapping and mismatch rates. On average, a total of 17.8 7 ± 174 3.89 Gb bases are mapped to YAO, which is 5.3 Mb on average more than that to GRCh38 175 (p=5.945x10-5, paired t-test) and almost identical to CHM13 (p=0.093, paired t-test) (Fig. S2A). 176 In addition, the average mismatch rate (mismatched bases in aligned reads / total aligned bases) 177 of reads alignment against YAO is 0.214 ± 0.013%, showing significant improvement when 178 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint compared to GRCh38 (0.245± 0.016%, p=2.79×10 -15, paired t -test) and CHM13 (0.227± 179 0.013%, p=1.65×10 -23, paired t -test) (Fig.S2B). Although the differences are subtle but very 180 significant as each sample shows reduced number of mismatches when aligned to YAO vs. to 181 CHM13 and GRCh38. 182 183 Greater improvement in mapping becomes more obvious after the removal of low-quality reads 184 (MAPQ <20). On average, the mapped bases are 17.9 ± 3.89 Gbp when aligned against YAO, 185 resulting in 3.37Mbp and 1.23Gb additional aligned bases against CHM13 (p=8.95x10-3, paired 186 t-test) and GRCh38 (p=1.33x10 -13, paired t -test), equal to 0.02% and 7.41% improvements, 187 respectively (Fig. 2A). The average mismatch rate in the high -quality mapped reads against 188 YAO is reduced to 0.204 ± 0.0142%, significantly lower than that of 0.215 ± 0.0141% 189 (p=2.01×10-21, paired t-test) against CHM13 and 0.222 ± 0.0150% (p=3.27×10-18, paired t-test) 190 against GRCh38 (Fig. 2B). 191 192 Focusing on the target exon region lifted from GRCh37 to CHM13, YAO and GRCh38, we 193 found 1-5 Mbp regions in each sample failed to be sufficiently covered by the reads (depth < 194 30×), regardless of the reference genome used. This confirms the existence of off-target effect 195 in the capture process of target sequencing due to unmatched probes against the Chinese 196 samples (Fig. 2C). In addition to the target exon sequences, for which the capture probes are 197 designed, WES reads oft en cover the flanking area due to the hitchhiker DNA fragments 198 captured by the probes. Despite not being fully targeted in the enrichment process due to 199 unmatched probes, WES reads from the 19 Chinese samples still cover 45.86±7.49% genomic 200 regions in YAO, significantly longer than those in CHM13 and GRCh38 (Fig. S2C). After the 201 exclusion of total regions with a sequencing depth less than 30× for calling reliable variants, 202 3.09±0.33% genome length of YAO remains covered, which is significantly longer than those 203 of CHM13 (3.03±0.32%, p=6.11x10-17) and GRCh38 (3.04±0.33%, p=1.67x10-13) (Fig. 2D). It 204 is obvious that YAO outperforms both CHM13 and GRCh38 in WES data analysis when 205 Chinese samples are mapped, even in the case where the capture probes are not appropriate for 206 Chinese samples. 207 208 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint Improvement in germline variant calling 209 Using DNAscope, an accurate and efficient germline small -variant caller that combines the 210 mathematics of the GATK’s HaplotypeCaller with a machine-learned genotyping model [28], 211 we called germline variants. Generally, homozygous variants have a frequency close to 1, 212 heterozygous variants around 0.5, while somatic variants exhibit frequencies deviating from 213 0.5 and 1. Based on these general rules, variants are further determined using deep learning 214 models that consider additional factors such as depth, base quality, and mapping quality. The 215 raw variant results are filtered by default cutoffs of >30×depth and >30 quality score to generate 216 a list of high -confidence variants (see flowchart). The number of germline variants decreases 217 significantly when using Y AO as a reference compared to the other two references, for both 218 homozygous and heterozygous variants ( Fig. 3AB). Since homozygous variants often have 219 high frequency in population, and thus are most likely associated with population -specific 220 variations, we only identified 715,828±149,696 such variants when using YAO as a reference. 221 However, this number increased by 11.95% and 19.26% when CHM13 (801,369±119,952; 222 p=3.58×10-15, paired t -test) and GRCh38 (853,687±161,424; p=3.55×10 -17, paired t -test) are 223 used as references, respectively ( Fig. 3A ). For heterozygous variants, which are primarily 224 attributable to within-population diversity and low-frequency variations, we identified similar 225 number of variants when referring to YAO and CHM13, which are 729,123 ± 191,013 and 226 735,117 ± 152,423, respectively. However, GRCh38 still ensures an identification of 227 heterozygous variants as many as 777,471±200,933, which is 6.62% more than YAO does 228 (p=3.65×10-13, paired t-test, Fig. 3B). Compared to GRch38, more variants were shared using 229 YAO and CHM13 as reference genomes (Fig S3). After further filtering out varinats with low-230 quality score (<30) or those in regions with lower reads depth (< 30×), homozygous germline 231 variants obtained using YAO are also the fewest among the three reference genomes (Fig S4A). 232 However, the differences in count of heterozygous variants obtained from the three reference 233 genomes were reduced (Fig S4B). 234 Similarly, when narrowing down to the probe targeted region, we found the same trends. There 235 are the fewest homozygous variants called using YAO as references (Fig.S5). Comparing YAO 236 and CHM13, the findings indicate that population -associated variations are a primary factor 237 contributing to the identification of homozygous variants in samples. The difference between 238 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint YAO and GRCh38, a chimeric genome, is slightly wider, possibly due to the more assembly 239 errors in GRCh38 that are not present in any population and thus lead to more homozyg ous 240 variant calls. 241 242 Further scrutinizing the different variants calls using the three references, we observed no YAO-243 specific or CHM13-specific pathogenic variants reported in ClinVar database (v.20231121, see 244 below) [29] but identified four GRCh38-specific pathogenic ones, with two located in the 7th 245 exon of the CNN2 gene transcript NM_004368.7 (encoding calponin 2) on chromosome 19, 246 found in 13 out of 19 samples. We further examined reads mapped to this exon from one sample 247 (sample St-2) harboring the two variants, and found that using Y AO and CHM13 as reference, the 248 reads were well-mapped with few mismatches. However, using GRCh38 as reference, an additional 249 subset of reads was mapped to this region, bearing numerous mismatches. Tracing these additional 250 reads using YAO and CHM13 as reference, they were primarily from a pseudogene located in 251 the pericentromere region of chromosome 20 (Fig S6AB), which is buried under many tandem 252 repeats. This pseudogene is partially homologous to the exon of CNN2 and is absent in GRCh38 253 due to the poorly assembled pericentromere region in chromosome 20. As a result, when using 254 GRCh38 as the reference, reads from this pseudogene were misaligned to the CNN2 exon, 255 leading to many false positiv es, including the pathogenic variants (Fig 3C, Fig S6C ). It 256 illustrates how structural variations between reference genomes can affect read mapping and 257

in false-positive variant calls. 258 259 260 Assessment in identifying pathogenic variants 261 For better interpreting clinically-significant variants, we use ANNOV AR to screen the records 262 in the ClinVar database (v.20231121) [29], containing a total of 2,336,658 records. When 263 converting the ClinVar coordinates from GRCh38 to CHM13 and YAO, only 5,186 (0.22%) 264 and 5,967 (0.26%) records failed conversion for YAO and CHM13, respectively. However, we 265 have only hit numbers per sample 14407.9 ± 725.5 for YAO and 16618.5 ± 834.5 for CHM13, 266 in contrast to a much higher hitting rate for GRCh38, 31526.7 ± 1542.9 per sample (Table S4). 267 This added difference is largely attributed to the categories of Benign, Likely benign , and 268 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint Conflicting interpretations of pathogenicity (Fig. 4A). Furthermore, a much larger proportion 269 of the variants in Benign categories are homozygous in GRCh38 (48.01%), compared to 270 YAO (28.92%) and CHM13 (31.24%). The reduced counts in YAO and CHM13 are mostly 271 from homozygous variants found in the categories of Benign and Likely benign. The differences 272 observed in the number of variants annotated as related to clinical phenotypes by ClinVar between 273 YAO and GRCh38 may be attributed to several factors, including population -specific variants, 274 particularly those that are homozygous or with high frequency, as well as false positives, as 275 illustrated in Fig. 3C. ClinVar annotations are based on GRCh38, which is a mosaic reference 276 genome created by merging data from multiple donors. This approach generates an excess of 277 artificial haplotypes and rare alleles, potentially introducing subtle biases in the analysis [19]. 278 Consequently, using GRCh38 as a reference may result in a higher number of homozygous variant 279 calls. Additionally, assembly errors or copy number variations in GRCh38 might contribute to false 280 positive calls, leading to an increased number of variant annotations, including a higher frequency 281 of benign records. 282 283 YAO, when applied to Chinese population samples, achieves similar sensitivity in identifying 284 pathogenic variants in addition to reducing false positives. In the categories of Pathogenic and 285 Likely pathogenic, we identified similar numbers of variants using either YAO (19.4 ± 11.5 per 286 sample) or CHM13 (19.9 ± 11.6 per sample), a little less than that when usin g GRCh38 (25.2 287 ± 10.9 per sample). Upon scrutinizing the Pathogenic variations, we found that many variations 288 in GRCh38 are attributable to reads with wrong mapping, whereas no such variants are seen in 289 the counterpart positions in YAO (Fig 3C). This discrepancy may arise from incorrect mapping 290 in GRCh38, whilst they are accurately recognized in Y AO. 291 292 Evaluation in TMB analysis 293 We further analyzed somatic variants and calculated TMB, which is composed of a standardized 294 number of non -synonymous mutations and serves as an indicator for the presence of tumor -295 specific antigens, capable of predicting treatment responses in cancer immunotherapies [30]. 296 For calculating TMB, blood samples are typically required to exclude germline variants from 297 those id entified in tumor tissues, thereby reducing false -positive calls of somatic variants. 298 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint However, as many of the FFPE samples we utilized did not have corresponding blood samples 299 available, we employed only the tumor WES data for somatic variant calling. We first utilized 300 the tumor-only mode TNscope [31] to call candidate somatic variants in target exon regions, 301 and subsequently removed those also identified as germline variants by DNAscope. The 302 somatic variant sets were further filtered with read depth (>30× within the target exon region), 303 quality score (>30), and a false-positive-specific filter tool FPfilter [32]. We picked up the non-304 synonymous mutations according to annotations from the filtered somatic variants and 305 calculated TMB standardized by the length of WES. 306 307 As expected, TNscope identified significantly fewer candidate somatic variants when YAO was 308 used as a reference as compared to the other two references (Fig. 4B). However, the difference 309 is substantially reduced after removing germline variants and the application of other filters, 310 yet it remains significant ( Fig. 4C ). The final YAO -based TMB values are slightly but 311 significantly lower than those based on the other two references, possibly due to fewer false -312 positive somatic variant calls (Fig. 4D). For all reference genomes, samples of malignant gastric 313 cancer exhibit higher TMB values than those of benign stromal tumor, indicating higher TMB 314 for malignant cancers, though this difference was not statistically significant due to the small 315 sample size (Fig. 4E ). And due to the absence of normal samples and consequent inadequate 316 germline variation filtering, the TMB of tumor-only WES is slightly higher than previously reported 317 in gastric cancer and gastric stromal tumors [33, 34]. 318 319

of this study 320

of this study include the use of FFPE samples, which introduce variability in tumor 321 purity and data quality, and a small sample size that restricts the statistical significance of our 322 analysis of disease -related variants and their cli nical relevance between benign gastric stromal 323 tumors and malignant gastric cancers. Additionally, the absence of normal samples may have led to 324 incomplete removal of germline variants, resulting in a slightly elevated tumor mutational burden 325 (TMB), despite our rigorous filtering methods. Nevertheless, the primary goal of this study was to 326 preliminarily assess the performance of different reference genomes, focusing on the utility of 327 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint population-specific references in the upcoming T2T era. The findings reveal significant differences 328 when using alternative reference genomes compared to GRCh38, underscoring the need for further 329 optimization of variant calling processes and the accumulation of genomic data from the Chinese 330 population. Such advancements will impr ove the identification of disease -related variants and 331 enhance the clinical applicability of indices like TMB. 332 333

334 This study conducts a parallel comparison of the current human reference genome GRCh38 and 335 potential reference genomes with top -quality—YAO and CHM13—in the whole process of 336 genomic analysis of WES data from 19 tumor samples of Chinese patients using state -of-the-337 art algorithms and tools. The initial comparison reveals that the three reference genomes share 338 similar basic characteristics in terms of genome size, GC content, and exome proportion, except 339 GRCh38 which is incomplete with many unfilled gaps, and its mosaic nature leads to inevitable 340 misassembled contigs. Subsequent analyses of WES data illustrated that both YAO and CHM13 341 outperform GRCh38 as a reference by offering higher mapping rates, lower mismatch rates, 342 and more reliable variant calling and annotation. Our initial study demonstrates that YAO, with 343 quality similar to CHM13, is more suitable for samples from Chinese indiv iduals, thus 344 proposing the idea of the population -specific reference genome. The reads mapping results 345 demonstrate the effectiveness of YAO in accurately aligning sequencing reads to the reference 346 genome, ensuring high-quality data for downstream functional analysis. The high mapping rate 347 and coverage of YAO as a reference genome for population-based studies for Chinese patients 348 underscore its suitability, especially in clinical settings and for disease treatments. 349 350

351 Data collection and alignment 352 In this study, quality control of the sequencing data was performed using FastQC 353 v0.11.8 (https://github.com/s-andrews/FastQC) to assess the quality of raw sequencing reads 354 and identify potential issues, and MultiQC [35] was employed to generat e a comprehensive 355 report. To ensure the complete removal of adapter sequences and low-quality bases, sequencing 356 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint reads were processed using TrimGalore-0.6.10 (https://github.com/FelixKrueger/TrimGalore). 357 Sample alignments to the reference genomes, CHM13v2. 0, YAO, and GRCh38.p14, were 358 performed using BWA-MEM v0.7.17-r1188 (https://github.com/lh3/bwa). After alignment, the 359 sorting and PCR duplicate removal of BAM files were processed with SortSam and 360 MarkDuplicates commands of Picard tools v3.1.0 (https://github.com/broadinstitute/picard). 361 362 Alignment quality assessment 363 To analyze the alignment results, we first used the stats command of samtools tool (version 1.9) 364 [36] to extract various alignment parameters. Next, we used samtools depth command to extract 365 the coverage and depth of the alignment results across the whole genome. In addition, we 366 utilized transanno (https://github.com/informationsea/transanno) to lift the coordinates of the 367 exome probe regions from GRCh37 to the other three reference genomes. This step is essential 368 for comparing the alignment results and analyzing the exome region across different reference 369 genomes. To compare exon regions, the corresponding exon probe sequences were aligned with 370 the Needle tool within the EMBOSS suite [37] to determine the percentage identity between 371 the sequences. 372 373 Variant calling and variant annotation. 374 DNAScope [28] was used to identify germline variants. Variants rejected by the machine 375 learning algorithms in DNAScope were filtered out. For further filtering, germline variants with 376 a quality score below 30 or dept h below 30 are removed. The tumor -only mode in TNScope 377 was utilized to identify somatic variants [31]. V ariants that failed to pass the criteria mentioned 378 above or shared by DNAScope variant calling were removed, and a final filtering step was 379 performed using FP-filter to identify somatic variants. The ClinV ar_20231126 database [29] 380 was downloaded, and databases specific to CHM13 and YAO were established using transanno 381 and the Vt toolkit [38]. The ANNOV AR tool [39] was used for variant annotation. 382 383 Ethical statement 384 The application for the study was submitted to and approved by the Ethical Review Committee 385 of Linfen Central Hospital (Approval No. YP2023-47-1). The collection and storage of human 386 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint samples were registered with and approved by the Human Geneti c Resources Administration 387 of China (HGRAC). Written informed consents were obtained from all participants. 388 389 Code availability 390 The code of this work is available on GitHub (https://github.com/KANGYUlab/WES). 391 392 Data availability 393 The raw WES data of 19 fresh gastric tumor samples have been deposited in the GS394 A for human in China National Center for Bioinformation (CNCB) under the accessio395 n number HRA006227 which is publicly accessible at https://ngdc.cncb.ac.cn/gsa -huma396 n. The T2T -YAO.hp genome is available at NGDC Genome Warehouse (https://ngdc.cn397 cb.ac.cn/gwh/) (GWHDQZI00000000). T2T -CHM13v2.0 is available at NCBI (GCA_00398 9914755.4). GRCh38 genome and its annotation file are available at UCSC (GCA_000399 001405.15, https://hgdownload.soe.ucsc.edu/goldenPath/hg38/big Zips/hg38.fa.gz; https://hg400 download.soe.ucsc.edu/goldenPath/hg38/bigZips/genes/hg38.refGene.gtf.gz). The VCF file401 s containing filtered variants of each sample called by DNAScope and TNScope in thi402 s paper are available on GitHub (https://github.com/KANGYUla b/WES) 403 404 Competing interests 405 All authors have declared no competing interests. 406 407 CRediT authorship contribution statement 408 Shuming Guo: Investigation, Resources, Data curation, Formal analysis, Funding acquisition, 409 Writing – original draft. Zhuo Huang : Invest igation, Methodology, Data curation, Formal 410 analysis, Software, Visualization, Writing – original draft, Writing – review & editing. Yanming 411 Zhang: Investigation, Resources, Data curation, Formal analysis, Writing – original draft. 412 Yukun He: Resources, Data curation, Formal analysis, Writing – original draft. Xiangju Chen: 413 Resources, Writing – original draft. Wenjuan Wang: Resources, Writing – original draft. 414 Lansheng Li: Resources, Writing – original draft . Yu Kang: Conceptualization, Formal 415 analysis, Investigation, Methodology, Project administration, Supervision, Writing – original 416 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint draft, Writing – review & editing. Zhancheng Gao: Funding acquisition, Project administration, 417 Supervision, Writing – review & editing. Jun Yu: Conceptualization, Formal analysis, Funding 418 acquisition, Investigation, Methodology. Zhenglin Du : Investigation, Methodology, Data 419 curation, Formal analysis, Writing – original draft, Writing – review & editing. Yanan Chu: 420 Conceptualization, Fund ing acquisition, Investigation, Methodology, Data curation, Formal 421 analysis, Writing – original draft, Writing – review & editing. 422 423 Acknowledgments 424 This study was supported by the National Key Research and Development Program of China 425 (Grant No. 2021YFC23 01000), the National Science Foundation of China (Grant No. 426 32371537), and the grants of Linfen Soft Science Research Project (Grant No. 2126). National 427 and Provincial Key Clinical Specialty Capacity Building Project 2020 (Department of the 428 Respiratory Med icine), and Peking University People’s Hospital Scientific Research 429 Development Funds (Grant No. RDGS2022-11) 430 431 432 ORCID 433 ORCID 0009-0001-7931-3725 (Shuming Guo) 434 ORCID 0009-0005-1023-1560 (Zhuo Huang) 435 ORCID 0009-0007-1480-1756 (Yanming Zhang) 436 ORCID 0000-0002-4164-2478 (Yukun He) 437 ORCID 0009-0005-7059-6350 (Xiangju Chen) 438 ORCID 0009-0003-0444-379X (Wenjuan Wang) 439 ORCID 0009 -0004-0689-8752 (Lansheng Li) 440 ORCID 0000-0001-5196-0376 (Yu Kang) 441 ORCID 0000-0001-7415-1416 (Zhancheng Gao) 442 ORCID 0000-0002-2702-055X (Jun Yu) 443 ORCID 0000-0003-2147-3475 (Zhenglin Du) 444 ORCID 0000-0002-9349-4307 (Yanan Chu) 445 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint 446 447 448 Figure legends 449 Fig 1. Sequence identity among WES target regions of YAO, CHM13, and GRCh38. 450 A. YAO vs. GRCh38. B. Y AO vs. CHM13. The coordinate information of target regions from 451 the Agilent SureSelect Human All Exon V6 was lifted from GRCh37 to YAO, CHM13, and 452 GRCh38 reference genomes using the transanno tool. 453 454 Fig 2. Comparison of read alignment referencing CHM13, YAO, and GRCh38. 455 A. The reduction of base of aligned reads (MAPQ>20) in each sample referencing CHM13 and 456 GRCh38 when compared to YAO in whole genome region. Blue and orange solid circles 457 represent the difference when comparing YAO vs. GRCh38 and YAO vs. CHM13, respectively. 458 Samples are sorted according to their total mapped reads. St, gastric stromal tumor; Ca, 459 gastric cancer. B. Mismatch rates of aligned reads (MAPQ>20) across the three reference 460 genomes, calculated as the number of mismatched bases divided by the total number of aligned 461 bases. C. The length of exon regions with sequencing depth > 30 in target regions. D. The 462 fraction of total genomic regions covered by reads with depth > 30 in the t hree reference 463 genomes. The p value of paired T -test is labeled above each comparison. The points 464 representing the same individual across different reference genome are connected by solid lines 465 466 Fig 3. Comparison of germline variants referencing CHM13, YAO, and GRCh38. 467 A. Total homozygous variants called by DNAscope in all reads covered regions. B. Total 468 heterozygous variants called by DNAscope in all reads covered regions. The p value of paired 469 T-test is labeled above each comparison. The points representi ng the same individual across 470 different reference genome are connected by solid lines. C. Peak plot of reads mapped to the 471 target region of 7 th exon in CNN2 gene. The horizontal coordinate of each vertical line 472 represents the position of each base, and the length of the vertical line represents the sequencing 473 depth. Gray lines indicate that the bases are consistent with those in the reference genome, and 474 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint colored lines indicate inconsistencies with bases in the reference genome. The colors green, red, 475 orange, and blue represent the four bases, A, T, G, and C, respectively. Red arrows indicate 476 pathogenic variants of allele ID 1689248 and 1689249 (recorded in the ClinVar) in GRCh38 477 and there are no variants on its corresponding sites in Y AO and CHM13. 478 479 Fig 4. Comparison of clinically relevant variants and tumor mutation burden when using 480 CHM13, YAO, and GRCh38 as references. 481 A. Count of average ClinVar recorded Pathogenic (upper panel) variants and the sum count of 482 Benign and Uncertain significance (lower panel) variants. B. Count of candidate somatic 483 variants detected by TNScope. C. Count of somatic variant after filter. D. Tumor Mutational 484 Burden (TMB). The p value of paired T -test is labeled above each comparison. The points 485 representing the same individual across different reference genome are connected by solid lines. 486 E. TMB comparison between gastric stromal tumors and cancers. 487 488 489 490 Supplementary material 491 Fig S1. Analysis pipeline. 492 Fig S2. Comparison of all mapped reads alignmen t between YAO vs. GRCh38 and YAO 493 vs. CHM13. 494 A. The reduction of all aligned reads referencing CHM13 and GRCh38 compared to YAO in 495 whole genome region. Blue and orange solid circles represent differences between YAO vs 496 GRCh38 and YAO vs CHM13, respectively. Samples were sorted according to their total 497 mapped reads. St, gastric stromal tumor; Ca, gastric cancer. B. Mismatch rates calculated as 498 the number of mismatched bases divided by the total number of aligned bases. C. The 499 comparison of coverage of each r eference genome based on all aligned reads of each sample. 500 The coverage was calculated as the length of reads covered regions with depth >= 1 divided by 501 the total length of the reference genome. The p value of paired T -test is labeled above each 502 comparison. 503 Fig S3. Venn plots of germline variants across different reference genomes. A. Unfiltered 504 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint variants. B. Filtered variants. 505 506 Fig S4. Comparison of filtered germline variants in all mapped regions referencing 507 CHM13, YAO, and GRCh38. A. Filtered germline homozygous variants. B. Filtered 508 germline heterozygous variants. The p value of paired T-test is labeled above each comparison. 509 510 Fig S5. Comparison of germline variants in target exon regions referencing CHM13, YAO, 511 and GRCh38. A. Total unfiltered germline homozygous variants in target regions. B. Total 512 unfiltered germline heterozygous variants in target regions. C. Filtered germline homozygous 513 variants in target regions. D. Filtered germline heterozygous variants in target regions. Th e p 514 value of paired T-test is labeled above each comparison. 515 516 Fig S6. The homologous regions and reads alignment of CNN2 gene exon No.7 in GRCh38, 517 YAO, and CHM13. A . The alignments between syntenic (chr19) and non -syntenic (chr20) 518 homologous regions arou nd CNN2 gene exon No.7 on GRCh38, YAO and CHM13. 519 Homologous regions are linked by grey blocks between the chromosomes. B. Reads alignment 520 in the non-syntenic homologous regions on chr20 in Y AO and CHM13; C. Reads alignment in 521 the region around CNN2 gene exon No.7 on chr19 in GRCh38. Pink or purple horizontal lines 522 represent forward or reverse reads, respectively. Reads in the black dotted box are wrongly 523 mapped to Chr19 in GRCh38, actually located on Chr20, around 30M in YAO and CHM13. 524 525 Table S1. Comparison of basic statistics of CHM13, YAO, and GRCh38. 526 Table S2. Comparison of the WES target regions lifted from GRCh37 to CHM13, YAO, 527 and GRCh38. 528 Table S3. Information of WES sequencing samples 529 Table S4. Comparison of the clinically relevant variants identified using CHM13, YAO, 530 and GRCh38 as reference. 531 532

533 [1] Goodwin S, McPherson JD, McCombie WR. Coming of age: ten years of next-generation sequencing 534 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint technologies. Nat Rev Genet 2016;17:333-51. 535 [2] Wojcik MH, Reuter CM, Marwaha S, Mahmoud M, Duyzend MH, Barseghyan H, et al. Beyond the 536 exome: What's next in diagnostic testing for Mendelian conditions. Am J Hum Genet 2023;110:1229-48. 537 [3] Yohe S, Thyagarajan B. Review of Clinical Next -Generation Sequencing. Arch Pathol Lab Med 538 2017;141:1544-57. 539 [4] van Dijk EL, Auger H, Jaszczyszyn Y , Thermes C. Ten years of next -generation sequencing 540 technology. Trends Genet 2014;30:418-26. 541 [5] Parsons DW, Roy A, Yang Y , Wang T, Scollon S, Bergstrom K, et al. Diagnostic Yield of Clinical 542 Tumor and Germline Whole -Exome Sequencing for Children With Solid Tumors. JAMA Oncol 543 2016;2:616-24. 544 [6] Feurstein S, Trottier AM, Estrada -Merly N, Pozsgai M, McNeely K, Drazer MW, et al. Germ line 545 predisposition variants occur in myelodysplastic syndrome patients of all ages. Blo od 2022;140:2533-546 48. 547 [7] Williams N, Lee J, Mitchell E, Moore L, Baxter EJ, Hewinson J, et al. Life histories of 548 myeloproliferative neoplasms inferred from phylogenies. Nature 2022;602:162 -8. 549 [8] Kaname T, Yanagi K, Naritomi K. A commentary on the promise of whole -exome sequencing in 550 medical genetics. J Hum Genet 2014;59:117-8. 551 [9] Lander ES, Linton LM, Birren B, Nusbaum C, Zody MC, Baldwin J, et al. Initial sequencing and 552 analysis of the human genome. Nature 2001;409:860-921. 553 [10] Venter JC, Adams MD, Myers EW, Li PW, Mural RJ, Sutton GG, et al. The sequence of the human 554 genome. Science 2001;291:1304-51. 555 [11] The Genome Reference Consortium, Announcing GRCh38. https://genomeref.blogspot.com/2013/. 556 [12] Nurk S, Koren S, Rhie A, Rautiainen M, Bzikadze A V , Mikheenko A, et al. The complete sequence 557 of a human genome. Science 2022;376:44-53. 558 [13] Rhie A, Nurk S, Cechova M, Hoyt SJ, Taylor DJ, Altemose N, et al. The complete sequence of a 559 human Y chromosome. Nature 2023;621:344-54. 560 [14] He Y , Chu Y , Guo S, Hu J, Li R, Zheng Y , et al. T2T -YAO: A Telomere-to-telomere Assembled 561 Diploid Reference Genome for Han Chinese. Genomics Proteomics Bioinformatics 2023. 562 [15] Chao KH, Zimin A V , Pertea M, Salzberg SL. The first gapless, reference -quality, fully annotated 563 genome from a Southern Han Chinese individual. G3 (Bethesda) 2023;13. 564 [16] Yang CT, Zhou Y , Song YN, Wu DY , Zeng Y , Nie L, et al. The complete and fully-phased diploid 565 genome of a male Han Chinese. Cell Research 2023;33:745-61. 566 [17] Liao WW, Asri M, Ebler J, Doe rr D, Haukness M, Hickey G, et al. A draft human pangenome 567 reference. Nature 2023;617:312-+. 568 [18] Gao Y , Yang X, Chen H, Tan X, Yang Z, Deng L, et al. A pangenome reference of 36 Chinese 569 populations. Nature 2023;619:112-21. 570 [19] Aganezov S, Yan SM, Soto DC, Kirsche M, Zarate S, Avdeyev P, et al. A complete reference genome 571 improves analysis of human genetic variation. Science 2022;376:54-+. 572 [20] Jarvis ED, Formenti G, Rhie A, Guarracino A, Yang C, Wood J, et al. Semi-automated assembly of 573 high-quality diploid human reference genomes. Nature 2022;611:519-31. 574 [21] Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alfoldi J, Wang Q, et al. The mutational 575 constraint spectrum quantified from variation in 141,456 humans. Nature 2020;581:434 -43. 576 [22] Madhusoodanan J. A More Diverse and Complete Reference Human Genome Is Poised to Change 577 Medicine. JAMA 2023;330:205-6. 578 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint [23] V ollger MR, Dishuck PC, Harvey WT, DeWitt WS, Guitart X, Goldberg ME, et al. Increased 579 mutation and gene conversion within human segmental duplications. Nature 2023;617:325-+. 580 [24] Hallast P, Ebert P, Loftus M, Yilmaz F, Audano PA, Logsdon GA, et al. Assembly of 43 human Y 581 chromosomes reveals extensive complexity and variation. Nature 2023;621:355 -64. 582 [25] Telomere-to-telomere consorti um. Telomere -to-telomere consortium CHM13 project. 583 https://github.com/marbl/CHM13.2023. 584 [26] ZCGAOlab. ChTY001. https://github.com/ZCGAOlab/ChTY001.2023. 585 [27] GENCODE. Human Release 44 (GRCh38.p14). https://www.gencodegenes.org/human/release586 _44.html. 2022. 587 [28] Donald Freed RP, Haodong Chen, Zhipan Li, Jinnan Hu, Rafael Aldana. DNAscope: High accuracy 588 small variant calling using machine learning. bioRxiv 2022. 589 [29] Landrum MJ, Chitipiralla S, Brown GR, Chen C, Gu B, Hart J, et al. ClinVar: improvements to 590 accessing data. Nucleic Acids Res 2020;48:D835-D44. 591 [30] Fusco MJ, West HJ, Walko CM. Tumor Mutation Burden and Cancer Treatment. JAMA Oncol 592 2021;7:316. 593 [31] Donald Freed RP, Rafael Aldana. TNscope: Accurate Detection of Somatic Mutations with 594 Haplotype-based Variant Candidate Detection and Machine Learning Filtering. bioRxiv 2018. 595 [32] Y uxiang Tan YZ, Hengwen Yang, Zhinan Yin. FPfilter: A false-positive-specific filter for whole -596 genome sequencing variant calling from GATK. bioRxiv 2020. 597 [33] Ke L, Li S, Hua ng D. The predictive value of tumor mutation burden on survival of gastric cancer 598 patients treated with immune checkpoint inhibitors: A systematic review and meta -analysis. Int 599 Immunopharmacol 2023;124:110986. 600 [34] Li ZY , Jia YN, Zhu HL, Xing XF, Pang F, Shan F, et al. Tumor mutation burden is correlated with 601 response and prognosis in microsatellite -stable (MSS) gastric cancer patients undergoing neoadjuvant 602 chemotherapy. Gastric Cancer 2021;24:1342-54. 603 [35] Ewels P, Magnusson M, Lundin S, Kaller M. MultiQC: summarize analysis results for multiple tools 604 and samples in a single report. Bioinformatics 2016;32:3047-8. 605 [36] Danecek P, Bonfield JK, Liddle J, Marshall J, Ohan V , Pollard MO, et al. Twelve years of SAMtools 606 and BCFtools. Gigascience 2021;10. 607 [37] Rice P, Longden I, Bleasby A. EMBOSS: the European Molecular Biology Open Software Suite. 608 Trends Genet 2000;16:276-7. 609 [38] Tan A, Abecasis GR, Kang HM. Unified representation of genetic variants. Bioinformatics 610 2015;31:2202-4. 611 [39] Wang K, Li M, Hakonarson H . ANNOV AR: functional annotation of genetic variants from high -612 throughput sequencing data. Nucleic Acids Res 2010;38:e164. 613 614 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint Fig. 1 615 616 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint Fig. 2 617 618 619 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint Fig. 3 620 621 622 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint Fig. 4 623 624 .CC-BY-NC-ND 4.0 International licensemade available 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 The copyright holder for this preprintthis version posted August 20, 2024. ; https://doi.org/10.1101/2024.08.19.608554doi: bioRxiv preprint