The strain PP17026, which was collected from Yunnan Province, China, was provided by Hongzhen Agricultural Science and Technology Co. Ltd. and cultured on M1 medium in the dark at 30°C. Spawn preparation, artificial cultivation and management of P. portentosus was performed according to previously published methods (Yang et al., 2019). Samples were collected in the hyphal (S, stage III), primordia (P, stage VI), and fruiting body (F, stage VII) stages (Yang et al., 2019). The hyphae fully covered on the surface of substrates, the small primordia with about 0.2 cm height and the stipes and pileus (cut into 0.2 cm × 0.2 cm pieces) used as representative samples of S, P, and F stages respectively were collected and flash-frozen in liquid nitrogen and stored at –80°C before RNA extraction.

The monokaryotic protoplasts were isolated from PP17026 using a previously published method (Chang et al., 1985). After 10 days of incubation, mycelia were collected and stored at –80°C until DNA and RNA extraction for genome and transcriptome sequencing. Transcriptome sequencing of monokaryotic strains was performed to correct the gene prediction results from the genome annotation.

DNA was extracted from the monokaryotic strain using the CTAB method. A total of 100–200 mg wet weight mycelium was ground in liquid nitrogen using a pestle in a centrifugal tube. Then, 1.5 mL of extraction buffer was added containing 20 mM EDTA, 100 mM Tris-HCl, 1.5 M NaCl, 2% CTAB, and 1% β-mercaptoethanol, and the sample was incubated at 65°C for 30 min. Protein and polysaccharide removal and DNA precipitation were conducted using chloroform-isoamyl alcohol (24:1, v/v) and isopropanol, respectively. The concentration, purity and integrity of DNA were assessed by NanoDrop (Thermo Scientific), Qubit and pulsed field electrophoresis analysis, respectively. A 20-kb sequencing library was built using large DNA segments with an ONT Template prep kit (SQK-LSK109) and an NEB Next FFPE DNA Repair Mix kit. The high-quality library was sequenced on an ONT PromethION platform with a corresponding R9 cell and an ONT sequencing reagent kit (EXP-FLP001.PRO.6). In addition, a small 300-bp sequencing library was built for the Illumina platform to improve the accuracy of the long-fragment library sequencing results.

Total RNA were extracted from all samples as previously described using TRIzol reagent (Invitrogen, Burlington, ON, Canada). Three biological replicates were processed. The purity and concentration of RNA was assessed using a NanoPhotometer^® spectrophotometer (IMPLEN, CA, United States) and a Qubit^® RNA Assay Kit with a Qubit^®2.0 Flurometer (Life Technologies, CA, United States). RNA integrity was assessed using an RNA Nano 6000 Assay Kit with the Agilent Bioanalyzer 2100 system (Agilent Technologies, CA, United States). Subsequently, RNA was treated with DNase I (Thermo Fisher, Waltham, MA, United States), and the mRNA was purified based on PolyA selection and fragmentation. First strand cDNA synthesis was performed using SuperScript II (Thermo Fisher) followed by second strand cDNA synthesis, end repair, 30-end adenylation, adapter ligation and PCR amplification. Purification was performed using AmPureXP Beads (Beckman Coulter, Brea, CA, United States). The 300 bp libraries were sequenced for paired-end 150-bp-reads on Illumina HiSeq Novaseq 6000 platform (Illumina Inc., United States).

To retrieve the nucleotide sequences from raw signal data generated from the ONT platform, base calling was performed using Albacore implemented in MinKNOW (Sahoo, 2017). Subsequently, by filtering the low-quality reads and demultiplexing the ONT barcodes and short reads, the filtered subreads were obtained using Canu v1.5 (Sergey et al., 2017). Raw Illumina data trimming was performed according to previously published methods (Yang et al., 2016). The corrected subreads were assembled into contigs using wtdbg v1.2.8 (Ruan and Li, 2020). To improve assembly, contig correction with Illumina read data was performed using Pilon (Walker et al., 2014). The completeness of the genome assembly was evaluated using BUSCO 4.0 with fungi_odb9 (Simão et al., 2015). MicroRNAs, rRNAs, and tRNAs were identified using tRNAscan-SE (Lowe and Eddy, 1997) and Infernal 1.1 (Nawrocki, 2014).

Three methods were used for gene prediction: (i) using a combination of the results from Genscan (Burge and Karlin, 1998), Augustus v2.4 (Mario and Burkhard, 2005), GlimmerHMM v3.0.4 (Majoros et al., 2004), GeneID v1.4 (Blanco et al., 2002), and SNAP (version 2006-07-28) (Ian, 2004); (ii) homologous protein prediction conducted using GeMoMa v1.3.1 (Jens et al., 2019); and (iii) unigenes were assembled and predicted using Hisat2 v2.0.4 (Kim et al., 2019), StringTie v1.2.3 (Pertea et al., 2015) and TransDecoder v2.0 based on the transcriptome sequencing of monokaryotic strains. All three results were integrated using EVM v1.1.1 to obtain an accurate prediction (Haas et al., 2008). Predicted genes were annotated using BLAST (Basic Local Alignment Search Tool) searches against Swiss-Prot (Boeckmann et al., 2005), TrEMBL (Rédei, 2008) and Nr databases. Gene ontology and KEGG metabolic pathway matches were identified using local Blast2GO tools (Conesa et al., 2005) and KAAS (Yuki et al., 2007), respectively. All predicted protein families were analyzed with InterProScan (Quevillon et al., 2005) and Pfam analysis (Finn et al., 2010). Carbohydrate-active enzymes were annotated using dbCAN (Huang et al., 2018). Oxidoreductases were extracted from the proteins predicted from the genome using a combination of IPR domain searches and the JGI cluster pipeline¹ according to a previously published method (Floudas et al., 2012). The A- and B-mating-type genes were identified using previously published methods (Chen et al., 2016). The genome has been submitted to the NCBI under the accession numbers of JAHRGP000000000.

The 150 bp paired-end reads obtained from the RNA-Seq analysis were trimmed using SeqPrep and Sickle to remove adaptor sequences, low-quality reads (those with ambiguous nucleotides and quality scores < 20). The filtered reads were mapped to the genome using HISAT2 v2.0.4. Transcripts were assembled and reconstructed using StringTie based on the HISAT2 mapping files. Several databases were used to annotate gene functions, including the Nr (NCBI non-redundant protein sequences), Nt (NCBI non-redundant nucleotide sequences), KOG/COG (Eukaryotic Ortholog Groups/Clusters of Orthologous Groups of proteins), Swiss-Prot (a manually annotated and reviewed protein sequence database), KO (KEGG Ortholog) and GO (Gene Ontology) databases. Gene expression was normalized using the FPKM (fragments per kilobase of transcript million mapped reads) method. The differentially expressed genes (DEGs) across all the samples were identified using DESeq2 (Love et al., 2014). Genes showing at least a twofold gene expression change with an FDR value < 0.05 were considered significantly differentially expressed. GO (Fisher’s exact test with corrected P-value < 0.05) and KEGG (Fisher’s exact test with corrected P-value < 0.05) enrichment analyses of DEGs associated with the significantly up- and downregulated genes between different stages were performed using Goatools (Klopfenstein et al., 2018) and R packages (Oksanen et al., 2011), respectively. Venn, PCA and hierarchical clustering for all correlation analyses were conducted using R packages. Weighted gene coexpression network analysis (WGCNA) was performed according to the method published previously (Li et al., 2019).

For each sample, 1 μg of RNA was used for cDNA synthesis using a TAKARA PrimeScript RT Reagent Kit. Semiquantitative SYBR green-based RT-PCR was performed using SYBR Premix Ex TaqII (Tli RNaseH Plus). Alpha-Tubulin was used as the internal control for normalization. Detailed information for the primers used in the present study is listed in Supplementary Table 1. The presented data are representative of three biological replicates and four technical replicates for each sample. Relative gene expression levels were calculated with the comparative threshold cycle method using a StepOne Plus Real Time PCR System (Applied Biosystems, United States).

The medium was optimized based on the results from the genome and transcriptome for suitable carbon sources selection. Stock cultures were grown on complete yeast extract medium (CYM) plates with 2% glucose, 0.2% yeast extracts, 0.2% peptone, 0.1% K₂HPO₄, 0.05% MgSO₄, 0.046% KH₂PO₄, and 2% agar. All optimization experiments were carried out in 9-cm-diameter Petri dishes (16 ml/petri dish) with modified CYM medium containing 2% different carbon sources. All the samples cultivated at 30°C in the dark for 30 days. The measurement of hyphal diameters was conducted. The significance of growth rate on different carbon sources was determined using t-test.

Two sequencing libraries (ONT and Illumina) were constructed for the monokaryotic strain for genome assembly. A total of 951,686 long reads (11,348,781,309 bp) were generated using the ONT platform. The N50 and N90 values of the raw data were 18,704 and 6,226 bp, respectively. After filtering the low-quality reads, 792,451 reads (10,603,105,695 bp) were obtained, with N50, N90 and a mean length of reads 18,866, 7,033 and 13,380 bp. The total data provided 323-fold coverage of the genome (Supplementary Table 2). Illumina sequencing yielded a total of 5.56 Gb clean data, and 20,876,612 clean reads (6,208,784,986 bp) for the monokaryotic strain were obtained by RNA-Seq.

An assembly of 32,742,503 bp across 62 scaffolds with a GC content of 48.92% was obtained (Table 1). Based on the results from the BUSCOO pipeline, only 9 of 290 single-copy entries were missing, suggesting > 95.17% genome completeness. Finally, 9,464 putative genes were predicted using a combination of different pipelines (Table 1). A total of 3393, 3175, 5143, 6751, 5552, 9023, and 9231 encoded proteins were identified with homologous sequences in the GO, KEGG, KOG, Pfam, SwissProt, TrEMBL, and Nr databases, respectively (Table 1). The 103 tRNA genes are dispersed in the genome. However, no rRNA genes were identified. Compared with other fungi used in this study, the genome size of P. portentosus was the smallest and the number of genes was the least except Tuber melanosporum (Table 2).

Gene annotation and functional categorization were performed using EuKaryotic Orthologous Group (KOG), where 5143 genes were redundantly assigned into 25 categories (Table 1 and Supplementary Figure 1). Except for genes related to general function prediction only (R) and unknown function (S), the number of genes involved in posttranslational modification, protein turnover, and chaperones (O); signal transduction mechanisms (T); and translation, ribosomal structure and biogenesis (J) were the highest, reaching 551, 403, and 334, respectively. The genes assigned to defense mechanisms, nuclear structure, cell motility and extracellular structures were the least abundant, with only 37, 28, 6, and 5 identified, respectively.

In addition, 35.85% of total genes had at least one GO annotation. Within the cellular component (CC) category, 4604 genes were assigned to 11 subcategories, the most abundant being “cell” and “cell part” (Table 1 and Supplementary Figure 2). A total of 3035 genes were classified into 13 molecular function (MF) categories, the greatest number of which was catalytic activity, followed by binding. A total of 1,469 predicted genes were assigned to 13 biological process (BP) GO terms, the most heavily represented being metabolic process and cellular process.

A total of 3,175 genes involved in 108 pathways were detected by KEGG annotation. The most enriched pathways included biosynthesis of amino acids (111), ribosomes (111), RNA transport (109), carbon metabolism (98), spliceosome (92), and purine metabolism (84) (Table 1 and Supplementary Figure 3).

A total of 301 CAZyme-coding genes for plant cell wall degradation were detected in the P. portentosus genome. These genes were divided into six families, including 57 auxiliary activities (AA) family genes, 19 carbohydrate-binding module (CBM) family genes, 47 carbohydrate esterase (CE) family genes, 110 glycoside hydrolase (GH) family genes, 62 glycosyl transferase (GT) family genes, and 6 polysaccharide lyase (PL) family genes (Table 2). The total number of CAZyme family genes in P. portentosus was within the range of that observed for mycorrhizal fungi (Table 2). The average number (303) of CAZyme families detected in the genomes of ectomycorrhizal fungi was much lower (ranging from 215 in T. melanosporum to 417 in Piloderma croceum) than that of saprophytic fungi (average number 542, ranging from 449 in A. bisporus to 606 in P. ostreatus) (Table 2). The number of AAs in P. portentosus (57 copies) was similar to that observed in Paxillus involutus (62 copies). P. portentosus had the same number of genes (19) related to CBM families as Amanita muscaria (Table 2). The number of genes belonging to GH families predicted in P. portentosus, A. muscaria, and P. rubicundulus was 110, 109, and 108, respectively. No more than 10 copies of PL genes in mycorrhizal fungi were identified. Except for GT families, the number of AA-, CBM-, CE-, GH-, and PL-encoding genes was much lower in ectomycorrhizal fungi than in saprophytic fungi (Table 2).

Based on the compositions of CAZyme families, the clustering results revealed two clades that included either mycorrhizal or saprophytic fungi. The pattern of CAZyme genes in P. portentosus was similar to that observed in P. rubicundulus, P. involutus, and Suillus luteus (Figure 1). The top five most abundant families in P. portentosus were CE10 (36 copies), GH16 (17 copies), AA7 (16 copies), AA1_1 (12 copies), and CBM5 (8 copies). CE10 genes encode esterases, GH16 enzymes act on xyloglucan and chitin and AA7 genes encode gluco- or chito-oligosaccharide oxidases, all of which were abundant in mycorrhizal and saprophytic fungi (Supplementary Table 1). AA1_1 enzymes are multicopper oxidases and represent the most abundant AA family in P. portentosus and mycorrhizal fungi assayed in the present study, while AA3_2 enzymes belonging to the glucose-methanol-choline (GMC) oxidoreductase family were the most abundant in saprophytic fungi (Supplementary Table 1). In addition, the families GH6 and GH7, which are involved in attacking crystalline cellulose, were present in all saprophytic fungi assayed in the present study and were absent in P. portentosus and other ectomycorrhizal fungi (Supplementary Table 1). The families GH13, GH15, GH27, and GH71 were enriched in P. portentosus, of which the copy numbers of the starch degradation-associated GH13, GH27, and GH71 families were higher than those observed in other mycorrhizal fungi (Supplementary Table 1).

Low copy numbers of genes were identified belonging to the cellulase, hemicellulase, pectinase, and lignin oxidase families, which directly act on lignocellulose and are listed in Table 3. Hierarchical clustering using the average-linkage method based on the composition of lignocellulases also resulted in the identification of two clades, revealing the different compositions of lignocellulases between ectomycorrhizal and saprophytic fungi (Figure 1B).

Four cellulase genes were identified in the P. portentosus genome, including 1 endo-beta-1,4-glucanase and 3 β-glucosidases. However, 1,4-β-cellobiosidase, which specifically hydrolyzes 1,4-beta-D-glucosidic linkages, was absent in P. portentosus and other ectomycorrhizal fungi, suggesting that P. portentosus may also have a low capacity to degrade celluloses. A low copy number of genes related to hemicellulases was observed in P. portentosus, with only five hemicellulases (no higher than seven in other ectomycorrhizal fungi) detected compared to saprophytic fungi (ranging from 10 to 29). In particular, no hemicellulases were detected in S. citrinum. α-Glucuronidases were absent in all ectomycorrhizal fungi evaluated in the present study, whereas they were present in all assayed saprophytic fungi. In P. portentosus, no pectinase genes were identified, as was observed in A. muscaria. The number of pectinases in ectomycorrhizal fungi ranged from 0 to 6. In saprophytic fungi, the copy number of pectinases varied significantly in different species, e.g., 3 present in Coprinus cinereus and 16 present in Volvariella volvacea. In addition, pectate lyases, which were detected in all saprophytic fungi, were absent in ectomycorrhizal fungi.

Six multicopper oxidases and one manganese peroxidase were identified in P. portentosus, fewer than was observed in other ectomycorrhizal fungi belonging to Basidiomycota. The number of multicopper oxidases encoded by these fungi was not in agreement with their associated niches and detected in all the fungi. The same pattern of auxiliary enzymes (e.g., aryl-alcohol oxidase, glyoxal oxidase, and benzoquinone reductase) was observed in all the fungi assayed in the present study. Pyranose oxidase, vanillyl-alcohol oxidase and galactose oxidase were not detected in any of the ectomycorrhizal fungi, including P. portentosus, except for L. bicolor.

To determine the gene expression patterns during P. portentosus development, 9 samples from 3 growth stages were used for RNA-Seq. The average number of raw reads was 51,592,958, ranging from 47,344,648 in sample S3 to 56,606,312 in sample F3 (Supplementary Table 3). After filtering, the average numbers of clean reads and total bases was 51,161,401 and 7,631,017,892, ranging from 46,876,694 and 6,987,544,020 in sample S3 to 56,203,088 and 8,367,222,043 in sample F3, respectively (Supplementary Table 3). More than 87.62% of the reads within each replicate could be mapped to the reference genome (Supplementary Table 3).

Using an FPKM cutoff value of 1, 8,556, 8,628 and 8,673 genes expressed in stages S, P, and F accounted for 90.40, 91.17, and 91.64% of the total predicted genes, respectively (Supplementary Figure 4A). A total of 8,316 common genes were expressed in all three stages. According to the FPKM values in different stages (FPKM < 10, 10–100, and >100), three gene categories were determined (Supplementary Figure 4B). The majority of genes in all the samples showed moderate expression, with FPKM values ranging from 10 to 100 (Supplementary Figure 4B). The F stage had the greatest number of expressed genes, while the S stage had the fewest. However, the F stage had the fewest highly expressed genes, while the P stage had the most. To determine the correlations among different stages, the nine samples were clustered into three groups based on principal component analysis, which revealed different patterns of gene expression across all stages (Supplementary Figure 5).

Based on FPKM values (>1), a corrected P-value of 0.05 and log2 (fold change) of 1 were set as the threshold for significantly different gene expression. A total of 4,921 DEGs were identified in comparisons between the three groups (S vs. P, S vs. F, and P vs. F) (Figure 2). The number of DEGs between the S and P groups was the highest (3600), followed by the S vs. F (2508) and P vs. F (2235) comparisons (Figure 2B). There were 353 common DEGs in all three growth stage comparisons, which were enriched in the carbohydrate metabolism and proteasome categories (Figure 2B). The most unique DEGs (919) were observed between the S and P groups compared to the S vs. F (509) and P vs. F (422) comparisons (Figure 2A). The numbers of downregulated genes in the S vs. P, S vs. F and P vs. F comparisons were higher than those of upregulated genes (Figure 2A).

All the DEGs were assigned into three different functional GO categories and 41 subcategories, including biological process (BP, 15 subcategories), cellular component (CC, 11 subcategories) and molecular function (MF, 15 subcategories) (Figure 3). The most enriched DEGs were assigned to catalytic activity (GO: 0003824) in all three comparisons. The DEGs in the S vs. P comparison were significantly enriched in drug metabolic process, mitochondrial protein complex and purine-containing compound biosynthetic process. Functional categories for the P vs. F comparison were significantly enriched for gene families associated with cytoplasmic part, organic substance biosynthetic process, cellular biosynthetic process and organic substance biosynthetic process.

To further understand the molecular and biological functions of the DEGs, they were mapped to the KEGG database (Figure 4). Pathway enrichment analysis revealed that proteasome and DNA replication were the most enriched in the S vs. P comparison. The ribosome pathway was the most significant pathway in the P vs. F comparison, with a P-value of nearly 0, following by DNA replication, oxidative phosphorylation, citrate cycle and peroxisome. Amino sugar and nucleotide sugar metabolism, peroxisome and cysteine and methionine metabolism were the most enriched in the S vs. F comparison.

WGCNA was conducted to uncover the coexpression profiles in successive developmental stages (Figure 5). Eight different modules were identified with high correlation coefficients according to the similarity of expression patterns (Supplementary Tables 5–12). To identify key genes specific to the S stage, the yellow module was analyzed and included a total of 858 genes. Amino sugar and nucleotide sugar metabolism, pyrimidine metabolism, biosynthesis of amino acids, purine metabolism and starch and sucrose metabolism were the most enriched pathways (Supplementary Table 5). Regarding the amino acids biosynthesis pathways, those involved in phenylalanine, tyrosine and tryptophan biosynthesis, histidine metabolism and glycine, serine and threonine metabolism were the most abundant (Supplementary Table 5). The turquoise and red modules (harboring 1,275 and 433 identified genes, respectively) were specific to the P stage, the number of genes in which was higher than was observed in other two stages. In this stage, RNA transport, cell cycle (DNA replication and meiosis), proteasome and spliceosome were the most enriched pathways (Supplementary Tables 6,7). The blue (63 genes) and black (63 genes) modules were highly associated with the F stage. Biosynthesis of amino acids, glutathione metabolism, peroxisome, glycerophospholipid metabolism and MAPK signaling pathway were the most enriched in this stage (Supplementary Tables 8,9).

Of the 301 CAZyme genes predicted from the genome, 277, 286 and 288 genes were expressed in the S, P, and F stages, respectively (Supplementary Table 4). The most highly expressed CAZyme genes were different in the three stages (Supplementary Table 4). In the S stage, the top five families with the highest FPKM values were CBM13 (1872.75), GH16 (1230.55), GH128 (1168.78), CBM13 (923.36) and AA5_1 (532.96). In the P stage, AA2 (1391.67), GT4 (992.50), GH128 (867.38), AA3_2 (298.65) and CBM50 (288.51) were more highly expressed than other families. The families GH16 (6346.44), GH18 (1097.48), GH128 (1003.62), GT4 (969.11), and GT2_Glyco_trans_2_3 (660.98) were overrepresented in stage F. The high expression of the glycoside hydrolase families GH16, GH18 and GH128 in all three stages indicated that P. portentosus has a high potential for starch or xylan degradation.

A total of 106 and 78 DEGs annotated as CAZyme families were detected in the S vs. P and P vs. F comparisons, respectively (Supplementary Table 4). In the S vs. P comparison, out of the 106 identified DEGs, 46 were upregulated and 60 were downregulated. Compared to the S stage, CBM13, GH16, AA5_1, GH18, AA6, AA3_2, GH17, GH5_30 and GT2_Chitin_synth_1 were significantly downregulated and AA2, GT4 and GH71 were upregulated in the P stage. In the P vs. F comparison, 52 DEGs were upregulated, including GH16, GH18, GT2_Glyco_trans_2_3, GH13_1 and GH31, while 26 DEGs were downregulated, including AA2, GH16, CE10, GH16, and CE4.

Twenty-four genes involved in lignocellulose decomposition were predicted from the P. portentosus reference genome, of which 15 genes were differentially expressed in the different developmental stages (Table 4). The most highly expressed genes involved in lignocellulose decomposition in all 3 stages included lignin oxidases (manganese peroxidase) and lignin-degrading auxiliary enzymes (alcohol oxidase, benzoquinone reductase), suggesting that P. portentosus has the potential to degrade lignin (Table 4). However, endo-beta-1,4-glucanase, which acts on cellulose, was expressed at low levels. Compared to that observed in the S stage, the expression of manganese peroxidase (EVM0004344) in the P stage was significantly upregulated (293.84 vs. 1391.67, respectively, a nearly fivefold increase) (Table 4), the expression of which subsequently decreased in the F stage. The RT-PCR results confirmed the observed expression profiles (Figure 6A). Multicopper oxidase laccases (EVM0002770, EVM0008799, and EVM0004198) were expressed at the highest levels in the S stage (Table 4 and Figure 6B). The lignin-degrading auxiliary enzymes glyoxal oxidase (EVM0006667), benzoquinone reductase (EVM0006980 and EVM0009178) and glucose oxidase (EVM0003870) were more highly expressed in the P stage than in the S and F stages (Table 4). The expression of aryl-alcohol oxidase (EVM0008750) gradually increased with the development of fruiting bodies, peaking in the F stage (Table 4). Endo-beta-1,4-glucanase expression was lower in the S (11.68) and F (23.71) stages than in the P stage (46.76) (Table 4). The expression of β-glucosidase (EVM0002244) increased from stages P (29.79) to F (105.70) (Table 4 and Figure 6D). Endo-1,4-betaxylanase (EVM0003710), which acts on hemicellulose, was expressed at higher levels in the F stage than in the P and S stages (Table 4 and Figure 6C).

From the genome and transcriptome results, P. portentosus exhibited a limited ability to degrade lignocellulose. The gene families acting on starch, xylan and chitin were most abundant in the genome and showed high expression in all the stages. Unique carbon sources (starch, xylan, and chitin) were further used to screen the most suitable culture medium, glucose as a control. By measuring the growth diameter of mycelia after 30 days of cultivation, the growth rate of P. portentosus on starch and xylan was much faster than that on glucose and chitin (Figure 7A). In addition to the growth rate, the morphology of the colonies also differed. Aerial mycelia of P. portentosus were much thicker, and pigmentation was stronger on glucose and starch than on xylan (Figure 7B). However, the integrity of mycelial edges were much better on xylan than on glucose and starch (Figure 7B). Thus, xylan and starch may be more suitable carbon sources for P. portentosus than chitin.