Chromosome-level, nanopore-only genome and allele-specific DNA methylation of Pallas's cat, Otocolobus manul

Abstract

Pallas's cat, or the manul cat (Otocolobus manul), is a small felid native to the grasslands and steppes of central Asia. Population strongholds in Mongolia and China face growing challenges from climate change, habitat fragmentation, poaching, and other sources. These threats, combined with O. manul’s zoo collection popularity and value in evolutionary biology, necessitate improvement of species genomic resources. We used standalone nanopore sequencing to assemble a 2.5 Gb, 61-contig nuclear assembly and 17097 bp mitogenome for O. manul. The primary nuclear assembly had 56× sequencing coverage, a contig N50 of 118 Mb, and a 94.7% BUSCO completeness score for Carnivora-specific genes. High genome collinearity within Felidae permitted alignment-based scaffolding onto the fishing cat (Prionailurus viverrinus) reference genome. Manul contigs spanned all 19 felid chromosomes with an inferred total gap length of less than 400 kilobases. Modified basecalling and variant phasing produced an alternate pseudohaplotype assembly and allele-specific DNA methylation calls; 61 differentially methylated regions were identified between haplotypes. Nearest features included classical imprinted genes, non-coding RNAs, and putative novel imprinted loci. The assembled mitogenome successfully resolved existing discordance between Felinae nuclear and mtDNA phylogenies. All assembly drafts were generated from 158 Gb of sequence using seven minION flow cells.

INTRODUCTION

Pallas's cat (Otocolobus manul), or the manul cat, is a small-bodied carnivore native to the montane grassland and shrubland steppe habitats of Central Asia (1). O. manul's wide geographic range spans varied climatic regions including the Caucuses Mountains, Mongolia and the Tibetan Plateau, but populations are concentrated at higher altitudes (2,3). Deemed the ‘the grumpiest cat in the world’ by the BBC documentary series Frozen Planet II (4), manuls are solitary hunters and subsist on a diet of small mammals (5). Prussian zoologist Peter Simon Pallas assigned the manul cat's taxonomic name as Felis manul in 1776 (6); the genus name Otocolobus was proposed in 1842 by Johann Fiedrich von Brandt (7) and formalized in 1907 based on skull morphology divergence from Felis (8).

Because of its wide home range and low population density, the conservation status of O. manul varies geographically. The International Union for Conservation of Nature (IUCN) Red List classifies the species as ‘Least Concern’ as of 2020 after being ‘Near Threatened’ in assessments from 2002 to 2016 (1, 9). However, O. manul is locally classified as endangered in several countries and either likely or confirmed extinct in some parts of its former range. Populations are also severely geographically fragmented, raising concerns for reduced effective population size and genome heterozygosity (10,11). The main threats to manul cats are anthropogenic, including poaching, habitat destruction, and the widespread use of rodenticides, which reduce prey populations and result in secondary poisoning of predators (12). A successful captive breeding program has led to O. manul being widespread in zoo collections around the world (13).

The manul cat is the sole extant species in the monotypic genus Otocolobus; phylogenetic analyses surrounding the relationship of this genus to other small-bodied cats are ongoing. A phylogenetic study using select nuclear genes across the felid radiation placed Otocolobus as a sister lineage to the genus Prionailurus, a taxonomic group that includes the fishing cat (P. viverrinus) and leopard cat (P. bengalensis) among others (14). In contrast, a whole-mitogenome phylogeny placed Otocolobus closer to Felis, which includes the domestic house cat (F. catus) (15). Both studies are concordant with skull morphology data suggesting a Miocene divergence between Pantheridae and other cat lineages (16). Improved genomic resources for the manul cat will increase resolution for phylogenetic analyses within Felidae.

High-quality genome assemblies and comparative genomics experiments increasingly include data on the epigenome and repetitive DNA content (17–19). DNA methylation, the most well-studied epigenetic mark, is critical in embryonic development, genomic imprinting, and X-inactivation with additional influence on gene expression and transposable element (TE) suppression (20–25). Methyl-cytosines also exhibit a 2–3-fold increase in mutation rate relative to unmethylated cytosines via accelerated deamination (26), providing a molecular substrate for phenotypic change across evolutionary time. Long-read sequencing permits haplotype phasing based on DNA sequence variants, meaning that DNA methylation can also be segregated by parent of origin (27). This development is of paramount importance for genome-wide measurement of allele-specific DNA methylation, a signature of imprinted genes. Despite its biological significance, imprinting is chronically understudied due to technical limitations such as bisulfite fragmentation, multi-mapping, and reduced information complexity (28,29).

Widespread improvements in genome assembly quality reflect reduced sequencing costs, greater local computing power, and improved contiguity via long-read sequencing (30). Now within reach even for low-resource laboratories, these powerful tools can support conservation efforts for diverse animal species. Here, we provide a high-quality diploid nuclear genome assembly, updated mitogenome, and allele-specific methylation analysis for O. manul, all generated exclusively from Oxford Nanopore sequencing reads.

MATERIALS AND METHODS

DNA sample

A 5-year-old male manul cat residing at the Utica Zoo (Utica, NY, USA), Tater, was chosen for genome sequencing (Figure 1). Whole blood was collected by jugular venipuncture and shipped frozen, then thawed and combined with 2 volumes of DNA Shield (Zymo Research, Irvine, CA, USA). Genomic DNA was extracted using a Quick DNA Miniprep Plus kit (Zymo Research) yielding 25 μg of DNA from 200 ml of blood. DNA quality was checked using a nano-spectrophotometer (Implen N60, Munich, Germany) and run out on a 1% gel to visualize fragmentation and RNA contamination.

Figure 1.

Otocolobus manul holotype photo depicting Tater, the 5-year-old male manul cat sampled for genome assembly.

Library preparation

Seven gDNA library preparations were performed using the SQK-LSK110 kit (Oxford Nanopore Technologies, OX4 4DQ, UK) following a modified version of the manufacturer's instructions (Nanopore Protocol version GDE_9108_v110_revJ_10Nov2020). In the authors’ modified version, each library prep used 3–5 μg of input DNA and was eluted with 45 μl of elution buffer (EB). Each elution was then split into three 15 μl aliquots to permit one initial flow cell loading step plus two reloads per library prep.

Sequencing

Sequencing was performed on seven R9.4.1 minION flow cells using Oxford Nanopore's minKNOW software (v22.05.5) and Guppy basecalling (v6.2.1). A total of three 15 μl DNA aliquots were loaded onto each flow cell with nuclease flushes between reloads per the SQK-LSK110 kit manual's ‘Priming and loading’ section. Each DNA aliquot was sequenced for 24 h for a total of 72 h of sequencing time per flow cell (21 total sequencing days). Guppy was set to fast basecalling during sequencing; post-hoc basecalling was performed using the ‘super accuracy’ model (r941_min_sup_g507). All assemblies were built using basecalls from the r941_min_sup_g507 model. Sequencing summary statistics (Table 1) and a histogram of read length data (Supplemental File 1) are provided.

Table 1.

O. manul nanopore sequencing read summary. Reads were generated from 25 ug of DNA on seven minION flow cells over 21 total sequencing days

Number of reads	15349224
Number of bases	158 248 024 728
N50 read length (bp)	16 601
Longest read (bp)	379 006
Shortest read (bp)	38
Mean read length (bp)	10 309
Median read length (bp)	7099
Mean read quality (QV)	16.26
Median read quality (QV)	14.51

Computational methods

Detailed pipeline information including software versions is provided in Supplemental File 2. The final assembly required the use of cluster computing resources with 2TB of allocated random access memory (RAM); downstream analyses were performed on two local computers: one running Linux Ubuntu 20.04.1 with 128GB RAM, an NVIDIA 3090Ti graphics processing unit (GPU), and a 16-core, 32-thread AMD Ryzen 9 5950x central processing unit (CPU), and another running Linux Ubuntu (20.04.1) with 64GB RAM, two NVIDIA graphics cards (3080Ti and 2080Ti), and a 12-core, 24-thread Ryzen 9 3900x CPU. Abbreviated computational methods are included below.

Genome assembly and polishing

Because there is no single best assembly pipeline for mammalian genomes, we generated draft sequences using a variety of assemblers and polishers. Draft assemblies were generated locally or via institutional cluster computing using Flye (31), NextDenovo (github.com/Nextomics/NextDenovo), Shasta (32), and Raven (33). Polishing was performed with Medaka (github.com/nanoporetech/medaka) for consensus assemblies and Racon (34) or NextPolish (35) for non-consensus assemblies. Assembly quality was assessed by searching for Carnivora-specific genes with benchmarking universal single-copy orthologs (BUSCO) version 5.3.2 (36,37). The two highest-scoring assemblies from Flye and NextDenovo, respectively, were combined using the Quickmerge metassembler (38) to increase contiguity while preserving quality. The Purge Haplotigs pipeline was applied for reassignment and removal of allelic contigs (39,40). Software versions and calls to the assemblers are provided in Supplementary File 2.

Contamination detection

Unmapped raw reads were scanned for microbial DNA with Kraken2 (41) and Pavian (42). We manually inspected the final assembly for spurious contaminant contigs using GC content, unusually high (>1000×) or low (<1×) coverage depth, lack of BUSCO genes, and sequence similarity as metrics. Sequence similarity contamination screening was performed using Megablast versus NCBI’s nucleotide (nt) database (43). Blobtools2 (44) was used to visualize GC content and read coverage (Supplemental Figure 1). An apparently chimeric 420 kb contig with >1000× coverage and no Carnivora-specific BUSCO genes matching Felidae by identity, ctg001250, was removed to generate the final primary assembly. Megablast hits to ctg001250 are provided in Supplemental File 3.

Coverage and quality statistics

Assembly coverage was assessed using Mosdepth version 0.3.345. K-mer quality value estimates were generated with Inspector, a tool built for long read data (45). We also applied Merqury, which was built for high accuracy short reads, to demonstrate its behavior on less accurate data (46).

Repeat identification

Repetitive DNA masking and classification was performed with RepeatMasker version 4.1.0 (47). Annotation using existing repeat classes, families and subfamilies was elected based on good representation of Carnivora in the Dfam version 3.5 open source repeat library.

Gene annotation

GeMoMa v1.8 and 1.9 (48,49) were applied to the primary assembly for homology-based protein prediction. First, version 1.8 was run with the F. catus gene annotation (GCF_018350175.1) used as the single reference; with the release of version 1.9, ‘GeMoMaPipeline’ was re-run with four felid gene annotations (F. catus, GCF_018350175.1; P. viverrinus, GCF_022837055.1; P. uncia GCF_023721935.1; O. geoffroyi, GCF_018350155.1) as a combined reference. Specific parameters are available in Supplemental File 2. ‘GeMoMaPipeline’ was also run on the P. viverrinus reference nucleotide assembly to assess the comparability of our O. manul output to NCBI’s reference annotations. All predicted proteins were scored using BUSCO’s protein mode and the carnivora_odb10 lineage dataset. Liftoff (50) was used with the F. catus (GCF_018350175.1) reference annotation to display gene features on differential DNA methylation plots for the O. manul assembly; DMR annotations were sanity checked by overlapping the GeMoMa and Liftoff annotation intervals. Proteins were not predicted for the Liftoff output.

Variant calling

Variant calling and phasing was performed using the PEPPER-Margin-DeepVariant pipeline (27). Variant statistics were generated using WhatsHap version 1.4 (51) and VCFtools version 0.1.17 (52). A consensus FASTA for the secondary haplotype was generated by applying all biallelic variants to the final assembly (considered the primary haplotype) with the consensus module of BCFtools version 1.15.1 (53). Runs of homozygosity and genomic demographics were assessed with SMC++ (54).

DNA methylation

DNA methylation (5 mC) at cytosine guanine dinucleotides (CpGs) was determined by re-basecalling QC-passed FAST5 files with a modified base configuration of GPU-mode Guppy (dna_r9.4.1_450bps_modbases_5mc_cg_sup). The final primary assembly was used as a reference. The resulting modified BAMs (modBAMs) were concatenated together into a single file, then sorted and indexed with Samtools (53). PEPPER-Margin-DeepVariant was then re-run to generate a haplotagged modBAM; this file and the primary assembly were used as input for ModBAM2BED version 0.6.2 (github.com/epi2me-labs/modbam2bed), which aggregates modified base counts to generate bedMethyl files. These files were then used as input for global and allele-specific methylation analysis.

Allele-specific DNA methylation was analyzed using the R package DSS developer version 2.43.2 (55–58). The package's two-group statistical comparison module ‘DMLtest’ was used to identify differentially methylated loci between the two haplotypes. The ‘callDMR’ module was then used with strict parameters (P_threshold = 0.001, delta = 0.5, minlen = 100, minCG = 15, dis.merge = 1500) to identify multi-CpG differentially methylated regions (DMRs).

DMR sequences were annotated via lifting of F. catus reference (GCF_018350175.1) gene features onto the O. manul assembly using Liftoff (50). The nearest feature to each DMR was then identified using AGAT (https://github.com/NBISweden/AGAT) and the closest module of BEDTools version 2.30.0 (59,60). DMRs annotated near genes with a ‘LOC’ symbol were additionally annotated with an alias or gene description, when available, via manual look up in NCBI’s Gene database (https://www.ncbi.nlm.nih.gov/gene). Visualizations were generated using MethylArtist version 1.2.3 (61), which required haplotagging the PEPPER output with Longphase (62), and JBrowse 2 (63) to validate each DMR.

Mitochondrial genome

Nanopore reads were aligned to the F. catus mitogenome using Minimap2 (v2.22-r1101) (64,65). Aligned reads were then downsampled to 10 000 reads using Seqtk version 1.3-r106 (https://github.com/lh3/seqtk) and Flye (v2.9-b1768) in metagenome mode was used for assembly. Mitogenome annotation and manual rearrangement to start the circular mitogenome at COX1 was done previously (66). A mitogenome phylogeny was built using MUSCLE version 3.8 (67) for alignment and IQ-TREE version 1.6.12 (68) for bootstrapped maximum-likelihood tree inference.

Scaffolding

The contig-level assembly was scaffolded onto the fishing cat (P. viverrinus) reference genome with the scaffold module of RagTag (69). Alignment of the two genomes was assessed with Dot (github.com/marianattestad/dot) visualizations of Nucmer alignments (70). An ideogram of contig positions in the scaffolds was generated with the R pakcage chromoMap (CRAN.R-project.org/package = chromoMap).

RESULTS

Sequencing

We isolated DNA for the assembly from the whole blood of Tater, a captive-bred 5-year-old male manul cat residing at the Utica Zoo (Utica, NY, USA). A holotype photo is provided (Figure 1). DNA extraction yielded approximately 25 μg of total DNA. DNA was sequenced over 21 days using seven minION flow cells and base calling was performed using the GPU version of Guppy (v6.2.1). After default quality filtering, sequencing runs generated a total of 15.3 million reads and 158 Gb of sequence with a read N50 of 16.6 kb (Table 1). Of these reads, 66% had a quality score greater than Q20 (i.e. error rate of 1 in 100 base calls) and 20% were greater than Q30 (one error in 1000 base calls).

Initial assembly

We generated multiple de novo assemblies and chose the highest-quality result for further analysis based on BUSCO score (37) (Table 2). Twenty other Felidae reference genomes, including three domestic house cat (F. catus) assemblies, provided benchmarking (64,71–81). Other felid genomes ranged in sequencing coverage from 17× to 159× with an average BUSCO score of 89.56% and a median score of 95.1% (Figure 2); full felid genome assembly statistics used in this comparison are available in Supplemental File 3.

Table 2.

Assembly statistics for selected O. manul drafts and reference genomes for fishing cat (P. viverrinus) and domestic house cat (F. catus). House cat is the felid species with the most high-quality genome assemblies; fishing cat is more closely related to O. manul by nuclear DNA phylogeny (14,15)

Assembly	Size (Gb)	Contigs/scaffolds/ chromosomes	Contig N50 (Mb)	BUSCO (% complete)	Depth	Technique
F. catus Fca126_mat1.0 (GCF_018350175.1)	2.42	110/71/19	90.7	95.5	76×	PacBio Sequel 2
P. viverrinus UM_Priviv_1.0 (GCF_022837055.1)	2.46	255/192/19	68.8	95.4	30×	PacBio Sequel
O. manul NextDenovo + Medaka	2.49	175/–/–	36.2	94.6	56×	Nanopore
O. manul Shasta + Medaka	2.47	714/–/–	26.3	94.1	56×	Nanopore
O. manul Flye + Medaka	2.49	1046/–/–	28.6	94.8	56×	Nanopore
O. manul Raven + Racon x2 + Medaka	2.50	568/–/–	9.2	94.7	56×	Nanopore
O. manul Quickmerge (Flye + NextDenovo) + NextDenovo	2.49	99/–/–	118.2	94.7	56×	Nanopore
O. manul Quickmerge + Purge Haplotigs	2.49	62/–/–	118.2	94.7	56×	Nanopore
OtoMan1.0	2.49	61/23/19	118.2	94.7	56×	Nanopore

Figure 2.

OtoMan1.0 is a near-complete and highly contiguous genome assembly for Pallas's cat (Otocolobus manul). (A) Felid reference assembly completeness was compared to our O. manul assembly by locally searching Carnivora-specific genes (carnivora_odb10) with BUSCO v5.3.2. OtoMan1.0, a nanopore-only assembly, ranked within 1% of the highest felid assembly BUSCO scores. (B) Contig N50 distribution for O. manul and felid reference assemblies. The O. manul assembly exhibits high contiguity (N50 = 118.2 Mb) when compared to other Felidae genomes, which differ minimally (∼3 Mb) in size.

We ran multiple de novo assemblers with relatively low computational memory requirements (i.e. runnable with 64–128GB local RAM) including Flye, NextDenovo, Shasta, and Raven. Running Flye with the full sequencing read set on a mammal-sized genome required remote high-powered computing resources (≥400GB RAM), so Flye was also run locally with filtered read sets (Supplemental File 3). Filtering reads by length or quality reduced computational resources enough to run assembly software locally but resulted in slightly lower BUSCO scores, even after polishing with the full read set. Therefore, the full QC-passed read set was applied for both drafts used in the final assembly.

The two highest quality draft assemblies were generated by Flye and NextDenovo, both polished once with Medaka (https://github.com/nanoporetech/medaka). Of the two, the Flye sequence achieved the highest BUSCO completeness score for Carnivora-specific genes at 94.8% across 1000 contigs (Table 2). The NextDenovo assembly had fewer contigs (175) but a lower BUSCO score of 94.6%. Both sequences were appropriate in size for a felid genome at 2.49 Gb. A merged assembly was generated with Quickmerge to combine the best traits of both drafts (i.e. highest BUSCO score and largest contigs, respectively) (38). Merging parameters were set to reduce the likelihood of perpetuating errors or misjoining contigs, and the quality of each draft was assessed with Inspector (Supplemental File 2) (45). The Flye-Medaka draft's QV was 31.2 and the NextDenovo-Medaka draft's QV was 31.4. Visualizing alignments of the draft assemblies to each other and to the merge result additionally indicated that (i) the drafts were similar to one another pre-merge and (ii) the merge favored the Flye draft over the NextDenovo draft (Supplemental Figure 3). After merging, reassignment of allelic contigs was performed on the merge result with the Purge Haplotigs pipeline (39,40), bringing the number of contigs in the assembly down to 62.

Assembly curation and scaffolding

Recent nanopore-only genome assemblies have used anomalous coverage depth and expected GC content as a means of filtering alien contigs (82). Here, these parameters were assessed using Blobtools2 (44). There was one 420 kb contig with 1178× average coverage depth in the initial O. manul assembly, ctg001250. NCBI megablast matched the sequence to Felidae, making a contaminant or endoparasite source unlikely (Supplemental File 3). Examining sequencing read alignment to ctg001250 in JBrowse 2(63) revealed regions of typical (∼60×) coverage interrupted by stretches covered by up to 4000 reads (Supplemental Figure 2). We concluded that ctg001250 was falsely contiguous based on a low number of reads spanning the high and low coverage areas, perhaps due to misplacement of interspersed repeats. Fifty percent of bases on ctg001250 were masked, and 40% were classified as LINE1 elements by RepeatMasker (Supplemental File 3). BUSCO score was intact following removal of ctg001250 (Table 2).

The final contig-level assembly, OtoMan1.0, was 2487293883 bp in length with 61 contigs, an N50 of 118.2 Mb, and a 94.7% BUSCO completeness score for Carnivora genes (Figure 2). Total genomic GC content was 41.87%. For comparison, 20 other Felidae reference genomes were locally reanalyzed with BUSCO v5.3.2 (carnivora_odb10). Our score of 94.7% places this nanopore-only assembly within 1% of the highest quality reference assemblies in Felidae. Additionally, the assembly's contig N50 was the highest among all assessed felid genomes. Half of OtoMan1.0 was covered by 8 contigs (L50) and the largest contig was 218.2 Mb (Supplemental Figure 4; Supplemental File 3). For reference, chromosome A1 in the current F. catus assembly is 239.4 Mb (RefSeq NC_058368.1). The shortest O. manul contig was 133 kb.

OtoMan1.0 was scaffolded onto the high-BUSCO fishing cat (P. viverrinus) reference genome (GCF_022837055.1) with the scaffold module of RagTag (69) (Figure 3). Fishing cat was chosen for alignment-based scaffolding due to (i) a lack of congenial species in Otocolobus, (ii) the genus sharing its most recent common ancestor with Prionailurus rather than Felis based on nuclear DNA (14,15) and (iii) the high BUSCO score of the P. viverrinus assembly. RagTag joined the 61 O. manul contigs into 23 scaffolds with 38 gaps, a total inferred gap length of 371685 bp, and a chromosome-level N50 of 151.9 Mb (Supplemental File 3). The scaffolds covered all 18 fishing cat autosomes, the X chromosome, and two unplaced BUSCO-containing fishing cat scaffolds (NW_025927612.1 and NW_025927619.1) (Supplemental Figure 3). Eight of the 18 cat autosomes (44%) were captured by single contigs in our assembly (Figure 3B), but two O. manul contigs were left unplaced. One of the two unplaced contigs was 593 kb in length and had 98.5% identity to a fragment of the 1.9 Mb F. catus Y chromosome (51% query coverage) when assessed with NCBI Megablast (43); the animal sampled to build UM_Priviv_1.0 was female. The second unplaced O. manul contig was 187 kb in length and exhibited high but discontiguous BLAST identity to F. catus, P. bengalensis and other felid sequences.

Figure 3.

OtoMan1.0 cross-genus alignment and scaffolding. (A) Genome sequence alignment of the fishing cat (P. viverrinus) reference genome (GCF_022837055.1) and the contig-level nanopore assembly of O. manul, OtoMan1.0, was generated using Nucmer (70) and Dot (https://github.com/marianattestad/dot). The alignment quality reflects gross genome collinearity present across Felidae (80,83). (B) Ideogram of OtoMan1.0 scaffolds generated via alignment to the fishing cat (P. viverrinus) reference assembly (GCF_022837055.1). Color transitions represent breakpoints between contigs on the scaffold. All 19 felid chromosomes were covered by the OtoMan1.0 assembly, and eight autosomes (A3, B2, B3, B4, C2, E2, E3 and F2) were captured by single contigs. Manul contigs were also aligned to two BUSCO-containing fishing cat unplaced scaffolds while two of the 61 O. manul contigs were unplaced. Correlations between the fishing cat chromosome lengths and O. manul scaffold lengths are provided in Supplemental Figure 5.

Repetitive DNA

Repetitive DNA, particularly interspersed repeats, are valuable drivers of genetic diversity in vertebrates (84,85). Since carnivores are well-represented in transposable element (TE) databases, we classified and annotated O. manul repetitive DNA with RepeatMasker (Table 3) (47). Overall, 34.72% of the assembly was classified as repetitive DNA, with a majority of retroelements (26.83%) and fewer DNA elements (2.8%), simple repeats (3.2%), and low-complexity regions (1.4%). These results are consistent with the dominance of retroelements in mammalian genomes (86,87). Local analysis of other cat assemblies with RepeatMasker indicated that OtoMan1.0’s repeat complement was consistent with other members of Felidae (Figure 4). Conservation of global repeat content in this clade was high: the 18 analyzed species varied by 3.67% for total bases masked, 2.72% for LINE content, 0.16% for SINE content, and 0.15% for DNA element content (Supplemental File 3).

Table 3.

Repetitive DNA classification results for the manul cat (Otocolobus manul) genome assembly, OtoMan1.0, generated with RepeatMasker

	Element count	Total length (bp)	% of Genome covered
Total bases masked		863500886	34.72
Total interspersed elements		737829702	29.66
Retroelements	1596531	667340949	26.83
SINE:	470064	69243802	2.78
MIR	462472	68339639	2.75
LINE:	836683	488479415	19.64
LINE1	468806	391204655	15.73
LINE2	312177	84936852	3.41
L3/CR1	41423	8915823	0.36
RTE	13010	3218808	0.13
LTR:	289784	109787884	4.41
ERVL	88371	39831330	1.60
ERVL-MaLRs	148224	51463432	2.07
ERV_classI	28861	12430326	0.50
DNA elements:	356331	69720741	2.80
hAT-Charlie	198573	36753878	1.48
TcMar-Tigger	55938	14630791	0.59
Unclassified	3794	597860	0.02
Small RNA	146099	11047204	0.44
Simple repeats	1677548	79649218	3.20
Low complexity	647428	34786006	1.40

Figure 4.

Repetitive element content of the assembled O. manul genome and 17 other felid species assessed with RepeatMasker (47). Manul cat repeat content was highly similar to other members of Felidae and dominated by retroelements.

Diploid genome

After read phasing OtoMan1.0 with the PEPPER-Margin-DeepVariant pipeline (27), approximately 83% of the assembly was covered by 6613 phase blocks with a block NG50 of 0.53 Mb (Supplemental File 3). As expected, most variants were single nucleotide variants (SNVs) and small insertions and deletions, with fewer large indels. Of 2045611 heterozygous variants detected, 1 561 420 (76.3%) were phased and 1 087 364 (53.2%) were phased, biallelic SNVs. The transition-transversion ratio (Ts/Tv) for biallelic SNVs was 2.12, in line with highly methylated mammalian genomes (Supplemental File 4). The software employed for variant calling does not detect inversions. The scaffolded primary assembly was deposited as BioProject PRJNA885133 and the contig-level alternate haplotype assembly as BioProject PRJNA889808.

An alternate pseudohaplotype assembly was generated by switching biallelic variants from the PEPPER variant call file into the primary, contig level assembly using the ‘consensus’ module of BCFtools (53). The resulting sequence was 2 484 282 002 bp in length with a contig N50 of 118.0 Mb and 94.8% BUSCO score.

We estimated diploid assembly quality using Inspector, a reference-free, k-mer-based tool built for long reads (45). Inspector's estimated QV for the primary contig-level assembly was 31.3 (99.93% accuracy) with a k-mer completeness score of 97.9% (Supplemental File 3). Inspector additionally scored the pre-merge assembly drafts from Flye and NextDenovo at Q31.2 and Q31.4, respectively. Inspector also includes a correction module for structural and base-level errors, but running it reduced OtoMan1.0’s BUSCO from 94.7% to 93.6% and inflated the reported number of structural errors.

We additionally ran Merqury on the diploid assembly (46). Merqury was built for high accuracy short reads, and lower accuracy read sets are more likely to produce QV overestimates. Still, this use case of a popular quality estimation tool is unreported in the literature. On default settings, Merqury gave OtoMan1.0 a k-mer completeness score of 97.5% and quality score of Q45.3 (Supplemental File 2). We also ran Merqury's ‘best_k’ script, which is intended to provide an ideal k-mer length for analysis based on genome size and read error rate. Given our median read error rate of 0.035, ‘best_k’ suggested k = 19; this value further inflated the QV to 53.2 from the original Q45.3 using the default k = 21. Despite these potential QV overestimates, Merqury's histogram output indicated that most small k-mers were excluded from the final assembly, appearing only in the read set (Supplemental Figure 6).

Lastly, genome heterozygosity and effective population size for O. manul were assessed using variant calls. Heterozygosity was 0.048% based on the number of called heterozygous SNVs (1 184 174) divided by the total callable bases in the contig-level primary assembly and 0.082% using all 2045611 heterozygous variants combined (SNVs and indels). Population history inferences for O. manul based on genomic runs of homozygosity were conducted with SMC++ (54). The results suggested an effective population size (N_e) near 10 000 with recovery from a bottleneck approximately 3000 generations ago (Supplemental Figure 7).

Annotation

Homology-based gene annotation was performed on OtoMan1.0 with Gene Model Mapper (GeMoMa) versions 1.8 and 1.9. The F. catus RefSeq annotation (GCF_018350175.1) served as the reference for our initial v1.8 run, generating 21909 amino acid predictions with an 86.2% protein BUSCO score (Supplemental File 2). A second prediction was generated using the P. viverrinus reference annotation as reference (GCF_022837055.1) instead, but BUSCO completeness was lower at 85.3%.

With the release of GeMoMa version 1.9, we re-ran GeMoMaPipeline on OtoMan1.0 using four felid reference annotations (F. catus, GCF_018350175.1; P. viverrinus, GCF_022837055.1; P. uncia GCF_023721935.1; O. geoffroyi, GCF_018350155.1) for a single prediction. The resulting protein BUSCO score was only slightly higher at 88.2% (Supplemental File 2).

To benchmark the GeMoMa software, we also predicted proteins for the P. viverrinus reference nucleotide assembly using F. catus as the annotation reference in GeMoMa v1.8. The resulting protein BUSCO score was 86.8%, within 1% of OtoMan1.0’s original protein score (Supplemental File 2). Both scores fall below our locally calculated BUSCO values for the F. catus (99.6%) and P. viverrinus (99.2%) protein annotations, which were generated with NCBI’s Eukaryotic Gene Annotation Pipeline.

DNA methylation

Nanopore instruments can detect native DNA modifications via modified pore signals compared to the unmodified base. Here, 5-methylcytosine (5mCG) base modifications were called using Guppy with the contig-level assembly as an alignment reference, generating methylation data for 33085506 cytosine-guanine dinucleotides. Similar to other mammals (88), global DNA methylation at CpG dinucleotides was high in O. manul at 78.2%.

Allele-specific methylation, a mechanism that facilitates monoallelic gene expression in genomic imprinting, was assessed by calling differentially methylated regions (DMRs) between O. manul pseudohaplotypes with DSS (55). Strict calling parameters (i.e. methylation delta at least 50%, 100 bp length, at least 15 CpGs) yielded 91 unique loci (Table 4). Visualization of haplotagged read alignments at each DMR facilitated the discovery of 30 false positive loci, where apparent read misalignment for one of the two haplotypes led to a called DMR via a lack of CpGs on one allele. Of the original 91 loci, 61 (67%) were deemed true DMRs following manual inspection (Supplemental File 3). The validated DMRs had a mean length of 969 bp and a mean of 118 CpGs.

Table 4.

Summary statistics and nearest gene feature data for differentially methylated regions (DMRs) called between O. manul pseudohaplotyes. Gene annotation data were lifted onto the OtoMan1.0 assembly from F. catus (F.catus_Fca126_mat1.0) and the nearest feature to each DMR was found with BEDTools (60). DMRs were validated visually using read pileup data. Statistics are shown for validated DMRs. Gene aliases are included in parenthesis

Initial DMRs	91
Validated DMRs	61
Mean length (bp)	969
Median length (bp)	705
Mean number of CpGs	118
Median number of CpGs	75
Mean absolute difference in methylation (%)	71.15
Median absolute difference in methylation (%)	69.67
Mean distance to gene feature (bp)	3463
Median distance to gene feature (bp)	0
Nearest genes to DMRs (symbol)	MXRA7, KCNQ1, IGF2, LOC109492247, CTSD, CAMK1G, IGF2R, LOC123383330, LOC109499294, LOC111560210, HERC3, NAP1L5, LOC109500454, BNC1, IGF1R, LOC109500537, UBE2Q2, LOC102900156, KLC1, SOCS6, LOC123378902, MEST, LOC109497613, VWDE, HSPG2, ST8SIA6, LOC111561284, SNU13, ATP4A, LOC123382347, INPP5F, LOC101093666, LOC102899880, LOC101083149, ATP6V1H, FAM110B, LOC109495939, BLCAP, NNAT, LOC102900772 (NESP55), LOC101098453 (GNAS), LOC109497888, CALCB, LOC123380523, GNA12, LOC105261300, TEKT5, ZDBF2, CMKLR2 (GPR1)

Validated DMRs were annotated by lifting F. catus reference gene features onto the contig-level O. manul assembly and querying the nearest gene with BEDTools (60). This method captured a number of classical imprinted loci described in humans and other animals, including GNAS, NNAT, IGF2, IGF2R, KCNQ1, MEST and ZDBF2 (Table 4) (89–93). In total, 41 validated DMRs directly overlapped 49 genes while non-overlapping DMRs fell between 165 bp and 56.7 kb away from the nearest feature. Putative novel imprinted loci, or DMRs near a feature not previously described as imprinted, including a 595 bp, 106-CpG DMR in the first exon of Von Willebrand factor D and EGF domain-containing protein (VWDE). Representative DMRs are shown in Figure 5. The mean absolute difference in methylation between the two pseudohaplotypes at DMRs was 71.15%. Most validated DMRs (44 of 61, 72.1%) were hypomethylated on the alternate pseudohaplotype, while 17 were hypermethylated.

Figure 5.

Allele-specific DNA methylation at differentially located loci (DMRs) called between O. manul assembly pseudohaplotypes assembled from a whole blood sample. DMRs were visualized with the package Methylartist (61). Annotation of the 61 manually validated DMRs was accomplished by lifting gene coordinates from the Felis catus reference assembly (F.catus_Fca126_mat1.0) onto OtoMan1.0 and finding the nearest feature. The results revealed (A, B) singleton DMRs and (C, D) multi-DMR clusters within and near genes. Genes overlapping DMRs included previously described imprinted loci such as (A) potassium voltage-gated channel subfamily Q member 1 (KCNQ1) and (B) the guanine nucleotide-binding protein G(S) subunit alpha (GNAS) complex locus (aliases LOC101098453, LOC102900772). DMRs also overlapped genes not previously reported as imprinted, including (C) von Willebrand factor D and EGF domains (VWDE), a protein with predicted involvement in anatomical development (96). (D) The most notable locus was a 33.2 kb region containing 12 DMRs all hypomethylated on pseudohaplotype 2. The DMRS overlapped the 5’ end of a classical imprinted gene, zinc finger DBF-type containing 2 (ZDBF2), on the negative strand and the 5’ end of chemerin chemokine-like receptor 2 (CMKLR2) on the positive strand. The CMKLR2 antisense RNA (CMKLR2AS; alias GPR1AS), which is not annotated in F. catus, exhibits imprinted expression in the human placenta (97).

Of the 61 validated DMRs, 12 (19.67%) occurred in one contiguous 33.2 kb region on feline chromosome C1 (Figure 5). All DMRs in the region were hypomethylated on pseudohaplotype 2. The region fell directly upstream of two genes, chemerin chemokine-like receptor 2 (CMKLR2) on the positive strand and zinc finger DBF-type containing 2 (ZDBF2) on the negative strand. ZDBF2 is a canonical imprinted gene expressed paternally in most human and mouse tissues (94); imprinted expression of CMKLR2, a paralog of opioid receptor kappa 1 (OPRK1), has not been reported, but expression of its antisense transcript (GPR1AS) is imprinted in human and mouse placenta (95).

Mitochondrial genome

The mitochondrial genome assembly built using Flye in metagenome mode contained one circular contig 17 097 bp in length with 800× read coverage (Figure 6). Megablast revealed a 99.7% nucleotide identity match to a previously assembled O. manul mitogenome (Supplemental File 3) (98). However, our assembly was longer than this previous manul mitogenome assembled from Sanger sequencing reads in 2019, which was 16 672 bp (GenBank MH978908.1). The 400 bp gap identified between our assembly and the 2019 version was located in the non-coding D-loop region of the mitogenome, making misassembly less likely than in the case of a large genic insertion. A 2015 Illumina mitogenome assembly for O. manul was closer to our assembly's length at 17 009 bp (KR132585.1), but identity and query coverage were lower. From our mitogenome, a new phylogeny was built using tiger (Panthera tigris) as the outgroup. The resulting phylogeny matched previous nuclear DNA trees, with Otocolobus sharing its most recent ancestor with Prionailurus (14) in contradiction of a 2016 mitochondrial phylogeny that placed Otocolobus closer to Felis (15). The mitogenome was deposited as part of the scaffolded primary assembly at BioProject PRJNA885133.

Figure 6.

Nanopore mitochondrial genome assembly of O. manul and felid mtDNA phylogeny. (A) The circular length of the built O. manul mitogenome assembly was 17 097 bp; read coverage was 800×. (B) Phylogenetic relationships of O. manul were derived through bootstrapped maximum likelihood comparison of the assembled mitogenome to existing Felidae references. In contrast to previously published mtDNA for O. manul (98), the current tree matches those generated through analysis of nuclear DNA (14).

DISCUSSION

Here, we have produced a highly contiguous diploid nuclear genome assembly, mitogenome, and allele-specific DNA methylation dataset for the manul cat (Otocolobus manul) via standalone nanopore sequencing. With 61 contigs and a contig N50 of 118.2 Mb, the assembly was more contiguous than any other Felidae reference genome to date before scaffolding. The felid clade is particularly well-suited for cross-genus scaffolding due to conserved genome collinearity among its members; our scaffolding strategy may not yield such unambiguous results in other animal families (80,83). Comparing our primary assembly to the original read set with Inspector yielded a k-mer completeness score of 97.9% and QV of 31.3 (99.93% accuracy). Using a single sequencing technology without parental data, this QV score approaches the ‘platinum-quality’ reference genome standard of Q40 (99.99% accuracy) established by the Vertebrate Genomes Project (VGP) (77,99,100). Our assembly exceeds VGPs other benchmarks of contig N50 ≥ 1 Mb and chromosomal scaffold N50 ≥10 Mb by 118-fold and 15-fold, respectively. Merqury yielded a higher QV estimate of 45.3 on default settings, but the score is likely an overestimate due to a relatively high read error rate. Still, Merqury's k-mer spectra histograms indicated that most small k-mers were excluded from the final assembly, suggesting that our coverage depth was able to overcome the read set error rate to produce a high-quality genome.

Our assembly's BUSCO score for Carnivora-specific genes, 94.7%, also fell within 1% of the most complete felid genomes. Importantly, BUSCO values were re-calculated locally using an updated version of the program (v5.3.2), leading to lower scores overall compared to those currently displayed on NCBI’s Genome pages (v4.1.4). The described genome, OtoMan1.0, reflects the continued, rapid improvement of assembly pipelines and the necessity of long-read sequencing for generating contiguous, high-quality reference genomes (101–104). Our results also provide evidence against concerns for low nanopore-only assembly quality due to sequencing error rates.

Repetitive elements have been minimally explored in felids despite their high genome occupancy and role in phenotype evolution (105–108). SINE element insertions are major sources of genomic diversity in canids, with links to multiple phenotypes in domestic dogs (Canis lupus familiaris) including early retinal degeneration, polyneuropathy, myopathy, and merle coat color (109–115). While global repeat content is strongly conserved across Felidae based on our analysis, individual repeat family activity is potentially variable and relevant to intra- and inter-species conservation. In a biomedically relevant example, domestic cats but not large cats possess an infectious endogenous retrovirus (ERV), RD-114, which can contaminate vaccines manufactured with feline cell lines and infect experimentally inoculated dogs (116–118). Long-read sequencing enables more accurate investigation of genomic repeat content; single reads can span full TEs and capture unique flanking host sequences, reducing multi-mapping issues inherent in short read assembly (119–121).

Nanopore sequencing permits native DNA methylation detection alongside DNA base calls, combining genome assembly and epigenome analysis into one step at the bench top. The epigenome is responsive to an animal's environment and shifts in a predictable manner across the life span (122,123). Here, we present a phased methylome and allele-specific DNA methylation analysis for O. manul. Our differentially methylated region (DMR) detection criteria were relatively strict in an effort to reduce false positives and yielded 61 validated loci. We successfully captured a number of classical imprinted genes including NESP55/GNAS, KCNQ1, NNAT, IGF1R and IGF2/H19 (124–128). IGF2R contained a DMR in this study suggestive of imprinted expression; this gene is maternally expressed in dogs (129,130) while monoallelic expression appears to have been lost in the primate lineage (131). Putative novel imprinting loci were also identified in genes such as Von Willebrand factor D and EGF domain-containing protein (VWDE), a paralog of fibulin 2 (FBLN2) required for multi-tissue limb regeneration in the axolotl (Ambystoma mexicanum) (132). A 2012 study found that 40% of human genomes deposited by the 1000 Genomes project (www.1000genomes.org) carried at least one loss-of-function mutation in VWDE (133). This locus and others were evaluated as distinct from 30 false DMRs, where misassembly resulted in a lack of CpGs on one pseudohaplotype. This phenomenon is not detectable to the DMR calling software DSS, necessitating manual curation; humans have over 200 imprinted genes (127,134) meaning our DMR calling parameters likely also introduced false negatives. Although imprinted gene expression cannot be predicted based on methylation data alone, this assembly provides a high-quality alignment reference for future transcriptomic analyses in O. manul. Future comparative investigations in diverse species will enrich our functional understanding of the allele-specific methylation mechanism.

Detection of low genome heterozygosity in this captive-bred O. manul individual is a concerning but valuable insight for both zoo breeding programs and wild population management. At 0.048%, our estimate of heterozygosity using biallelic pseudohaplotype SNVs was similar to the critically endangered Amur tiger (Panthera tigris altaica), a species that experienced a severe population bottleneck in the 1940s with a current estimated population of 400 animals (135–138). In contrast, the manul cat population is estimated at 58 000 individuals (1). Although captive breeding programs are a crucial tool in global conservation, a small pool of founder animals inevitably leads to reduced genetic diversity. Practices like sperm biobanking have the potential to support captive population genetic quality, but such systems are underutilized (139). Fortunately, our estimate of effective population size using runs of homozygosity was relatively large at N_e = 10 000. Results also suggested that manul cats experienced a crash in genetic diversity, but that this event was in the relatively distant past (∼10000 years ago); effective population size appears to have continued recovering despite anthropogenic challenges to the species. Given the large, low-density geographic range and solitary social strategy of O. manul, it is unclear whether these values are typical across the entire population.

While largely conserved among metazoans, mitogenomes also contain structural and nucleotide variation that can serve as phylogenetic markers (140). Their smaller size, relative to nuclear genomes, make them an attractive tool in phylogenetics. However, incongruities between mitogenome and nuclear DNA phylogenies can occur due to uniparental inheritance of organelles (141), horizontal gene transfer (142), or incomplete lineage sorting (143). Previously, an O. manul mitogenome assembly placed Otocolobus as a sister taxa to the Felis clade (15). The mitogenome assembly presented here contains a previously unidentified 400 bp gap in the D-loop control region; a new phylogenetic analysis placed O. manul closest to Prionailurus in agreement with the nuclear genome phylogeny (14). This finding suggests that O. manul nuclear DNA and mtDNA have similar phylogenetic histories. Recently, felid mitogenomes were used to investigate the contribution of a wild cat to domestication of F. catus (144). Using our mitogenome assembly, a similar study could investigate introgression among wild Asian cats (i.e. O. manul and Prionailurus spp.) and between wild and domestic cats in this region. Hybridization among cat species could have important implications for genetic diversity and conservation. Notably, evidence of hybridization among domestic and wild cats is found in South Africa and Europe, and is related to human population density in proximate spaces to wild cat habitat (145,146).

Long-read sequencing in general has been critical for increased genome assembly quality, but the portability and cost-effectiveness of nanopore sequencing make it particularly valuable for reducing barriers to entry in genomics (82,104). Still, assembly of mammalian-sized genomes (3+ Gb) can exceed computational and financial resources available to small laboratories. Even Flye, an efficient assembler (31), required distributed computing resources when using our full read set (Table 2; Supplemental File 3). Unequal access to such resources can inadvertently perpetuate ‘parachute science’ and limit the participation of researchers from historically colonized countries in endemic species conservation (147–149). However, we were able to build O. manul assemblies locally using a subset of the longest reads with only a small reduction in BUSCO score (Supplemental File 3). On-demand cloud computing resources such as Amazon Web Services (AWS) are also increasingly affordable, eliminating the necessity of access to a university cluster. We hope that the provided data and computational pipeline will be useful resources for O. manul conservation as well as for other small groups assembling mammal genomes.

DATA AVAILABILITY

DNA sequencing reads are available in FASTQ format in the Sequencing Read Archive repository (https://www.ncbi.nlm.nih.gov/sra) under submission SRR22085263. The scaffolded primary assembly (including mitogenome) is available as BioProject PRJNA885133; the contig-level alternate assembly is available as BioProject PRJNA889808. Modified base calls (DNA methylation) are also available under BioProject PRJNA885133. Sample data are available as BioSample SAMN31076064. Gene annotation, variant calling, and DNA methylation data are available by request.

Supplementary Data

Supplementary Data are available at NARGAB Online.

FUNDING

United States Department of Agriculture National Institute of Food and Agriculture [HATCH AES MIN-16–12 to C.F.]; Norn Group [Longevity Impetus Grant to C.F.]; National Institutes of Health [R21 AG071908 to C.F., L70 AG079467-01 to N.F., T32 OD010993 to N.F.]; National Science Foundation [IOS 1556396 and IOS 1754437 to M.K.D.]

Conflict of interest statement. None declared.