Overview of PAX gene family: analysis of human tissue-specific variant expression and involvement in human disease

Abstract

Paired-box (PAX) genes encode a family of highly conserved transcription factors found in vertebrates and invertebrates. PAX proteins are defined by the presence of a paired domain that is evolutionarily conserved across phylogenies. Inclusion of a homeodomain and/or an octapeptide linker subdivides PAX proteins into four groups. Often termed “master regulators”, PAX proteins orchestrate tissue and organ development throughout cell differentiation and lineage determination, and are essential for tissue structure and function through maintenance of cell identity. Mutations in PAX genes are associated with myriad human diseases (e.g., microphthalmia, anophthalmia, coloboma, hypothyroidism, acute lymphoblastic leukemia). Transcriptional regulation by PAX proteins is, in part, modulated by expression of alternatively-spliced transcripts. Herein we provide a genomics update on the nine human PAX family members and PAX homologs in 16 additional species. We also present a comprehensive summary of human tissue-specific PAX transcript variant expression and describe potential functional significance of PAX isoforms. While the functional roles of PAX proteins in developmental diseases and cancer are well characterized, much remains to be understood regarding the functional roles of PAX isoforms in human health. We anticipate the analysis of tissue-specific PAX transcript variant expression presented herein can serve as a starting point for such research endeavors.

Introduction

Organogenesis, tissue function, and homeostasis require extremely precise regulation of gene expression — which is heavily influenced by transcription factors. Transcription factors bind to DNA and regulate gene expression by either promoting or repressing target gene expression. Beginning in the late 1980s, with the advent of modern molecular biology techniques, investigators began studying a class of transcription factors that contain a “paired” DNA-binding region. Ultimately, these came to be known as the “master regulator” paired-box (PAX) proteins, due to their ability to regulate gene transcription during organismal development.

In 1986, the Drosophila melanogaster pair-rule gene, paired (prd), was first cloned (Kilchherr et al. 1986), and it was discovered that the fly genes — prd and gooseberry (gsb) — both contain a conserved paired domain (PD) (Bopp et al. 1986). The importance of these PD-containing transcription factors for organismal development quickly became apparent, because mutations of paired-box genes in D. melanogaster were associated with developmental disorders (Quiring et al. 1994). These seminal discoveries opened our eyes to the highly conserved PAX family of transcription factors.

PAX genes encode a family of tissue-specific transcription factors that possess a conserved 128 amino-acid DNA-binding PD (Fig. 1A) (Blake and Ziman 2014). The PD contains two helix-turn-helix (HTH) motif subdomains, PAI and RED (Czerny et al. 1993; Xu et al. 1995) that bind to DNA (Czerny et al. 1993; Epstein et al. 1994a; Epstein et al. 1994b; Jun and Desplan 1996) by interacting with the DNA major groove (Luscombe et al. 2000). The ability of PAI and RED to interact with DNA results in the bipartite nature of PAX protein-DNA binding (Czerny et al. 1993). Posttranslational modification of the PD can interfere with its ability to bind DNA via the PAI subdomain (Cao et al. 2005; Codutti et al. 2008).

Figure 1. Paired-box (PAX) protein structure.

(A) PAX proteins are characterized by their conserved, paired DNA-binding domain, which is composed of the RED and PAI subdomains. (B) The partial or full inclusion of a DNA-binding homeodomain and/or octapeptide linker subdivides PAX proteins into Group 1, Group 2, Group 3, or Group 4. Reprinted with permission from (Blake et al., 2014).

Mammals possess nine PAX proteins, which are divided into four groups. Presence of an octapeptide linker and/or partial or full homeodomain (HD) determine the group classification (Fig. 1B) (for a review of the octapeptide linker and homeodomain, see (Blake and Ziman 2014).

PAX proteins in groups II, III and IV contain either a partial or full HD DNA-binding domain, which allow them to bind and regulate unique subsets of gene targets. PAX proteins in groups I, II and III also contain an octapeptide linker that can be bound by cofactors to control transactivation of groups II and III PAX proteins (Eberhard et al. 2000; Hollenbach et al. 1999; Lechner and Dressler 1996a; Muhr et al. 2001). Group I PAX proteins lack a HD, indicating that a HD is not required for their DNA-binding function. Whereas the HD is not required for PAX protein function, the HD influences PAX protein-DNA binding by cooperatively interacting with the PD (Jun and Desplan 1996).

Conservation of the PAX protein’s PD allows for similar DNA-binding motifs between PAX proteins (Epstein et al. 1994a; Jun and Desplan 1996); however, each individual PAX protein regulates different genetic targets. This leads to an interesting question: how do PAX proteins with conserved PDs differentially regulate gene targets? Several possible answers to this question have been offered: i) the PD subdomains (i.e., RED and PAI) individually bind to DNA containing different DNA motifs (Czerny et al. 1993; Epstein et al. 1994a; Jun and Desplan 1996); ii) the PD subdomains and HD, intramolecularly and/or intermolecularly, form dimers with other PAX proteins (Czerny et al. 1993; Epstein et al. 1994a; Jun and Desplan 1996; Underhill and Gros 1997); iii) cofactors bind to different PAX functional domains (Eberhard et al. 2000; Hollenbach et al. 2002; Hollenbach et al. 1999; Muhr et al. 2001); and iv) alternative splicing of PAX transcripts can result in PAX proteins that lack functional domains (i.e., RED, PAI or HD). While each possibility is correct under certain circumstances, we refer the reader to other reviews for detailed explanations of the first three answers (Curto et al. 2015; Cvekl and Callaerts 2017; Mayran et al. 2015). This review will focus on the fourth possibility, i.e., regulation of PAX protein function by differential PAX gene splicing.

Historically, experimental challenges have made it difficult to effectively delineate the influence of PAX transcript expression on transcriptional regulation of genetic targets and tissue development. In the following, we provide an updated review of the history of the PAX protein family and, for the first time, a detailed summary of the different human and other mammalian transcript variants — including known PAX transcripts, adult human tissue-specific PAX-transcript-variant expression patterns, functional differences between the various isoforms, and implications of PAX isoforms participating in human disease.

PAX: A highly conserved history

Vertebrate PAX proteins display a high degree of conservation, particularly in the PD. Such conservation presumably reflects their origin from common ancestral genes (Balczarek et al. 1997). It is likely that the original PAX gene arose with the emergence of metazoans roughly 1 billion years ago (Breitling and Gerber 2000). This hypothesis is supported by finding that the sea sponge contains at least two proto-PAX genes (Breitling and Gerber 2000; Hoshiyama et al. 1998). The proto-PAX gene is thought to have arisen when the DNA-binding domain of ancient transposases (i.e., Tc1 and mariner transposases) fused with a homeodomain (Breitling and Gerber 2000) — an event that presumably occurred in a species having proteins that contained a paired-type homeobox DNA-binding domain and the octapeptide (Paixão-Côrtes et al. 2015). This event created the proto-PAX genes, PAXB and PAXD, from which two PAX super groups have manifested (Kozmik et al. 2003; Miller et al. 2000; Paixão-Côrtes et al. 2015).

The nine human PAX genes are located on eight different chromosomes (Table 1). Such scattering of the PAX genes across the human genome suggests that the PAX gene family is evolutionarily extremely ancient. Supporting this contention, a BLAST search of genomes revealed the presence of PAX genes in species as old as the red beard sponge (Supplemental Table 1).

Table 1.

List of all human PAX genes with official gene symbols, Pax group, aliases, chromosomal locations, isoforms, National Center for Biotechnology Information (NCBI) RefSeq mRNA accession numbers, NCBI RefSeq protein accession numbers, Ensembl gene ID numbers, Ensembl transcript ID numbers and total number of amino acids [information retrieved from www.genenames.org and www.ensembl.org]. PAX transcript variants were identified from the NCBI database (Brown et al. 2015). Due to differences between NCBI Gene and Ensembl, some transcript variants have a RefSeq mRNA accession number but lack an Ensembl transcript ID.

Gene name	Pax group	Aliases	Chromos ome location	Isofor ms	RefSeq mRNA number	RefSeq protein number	Ensembl gene ID	Ensembl transcript ID	Total # of amino acids
PAX1	I	HUP48	20p11.22	PAX1a	NM_006192.5	NP_006183.2	ENSG00000125813	ENST00000613128.4	534
				PAX1b	NM_001257096.1	NP_001244025.1		ENST00000398485.6	457
PAX2	II		10q24.31
				PAX2a	NM_003987.5	NP_003978.3	ENSG00000075891	ENST00000428433.5	417
				PAX2b	NM_000278.5	NP_000269.3		ENST00000355243.7	394
				PAX2c	NM_003988.5	NP_003979.2		ENST00000370296.6	396
				PAX2d	NM_003989.5	NP_003980.3			409
				PAX2e	NM_003990.5	NP_003981.3			432
				PAX2f	NM_001304569.2	NP_001291498.1			425
				PAX2g	NM_001374303.1	NP_001361232.1			102
PAX3	III	HUP2	2q36.1	PAX3a	NM_181457.4	NP_852122.1	ENSG00000136903	ENST00000350526.9	479
				PAX3b	NM_000438.6	NP_000429.2		ENST00000409828.7	215
				PAX3c	NM_013942.5	NP_039230.1		ENST00000258387.6	206
				PAX3d	NM_181458.4	NP_852123.1		ENST00000392070.6	484
				PAX3e	NM_181459.4	NP_852124.1		ENST00000392069.6	505
				PAX3f	NM_181461.4	NP_852126.1		ENST00000344493.9	403
				PAX3g	NM_181460.4	NP_852125.1		ENST00000336840.11	407
				PAX3h	NM_001127366.3	NP_001120838.1		ENST00000409551.7	483
PAX4	IV	MODY9	7q32.1	PAX4a	NM_001366110.1	NP_001353039.1	ENSG00000106331	ENST00000639438.3	351
				PAX4b	NM_001366111.1	NP_001353040.1		ENST00000378740.6	348
PAX5	II	BSAP	9p13.2	PAX5a	NM_016734.3	NP_057953.1	ENSG00000196092	ENST00000358127.9	391
				PAX5b	NM_001280547.2	NP_001267476.1		ENST00000377852.7	357
				PAX5c	NM_001280548.2	NP_001267477.1		ENST00000377853.6	362
				PAX5d	NM_001280549.2	NP_001267478.1		ENST00000523241.6	324
				PAX5e	NM_001280550.2	NP_001267479.1		ENST00000520154.6	295
				PAX5f	NM_001280550.2	NP_001267479.1		ENST00000523145.5	295
				PAX5g	NM_0012800552.2	NP_001267481.1		ENST00000377847.6	328
				PAX5h	NM_001280553.2	NP_001267482.1		ENST00000520281.5	319
				PAX5i	NM_001280554.2	NP_001267483.1		ENST00000414447.5	348
				PAX5j	NM_001280555.5	NP_001267484.1		ENST00000446742.5	219
				PAX5k	NM_001280556.2	NP_001267485.1		ENST0000052200.5	283
PAX6	IV	D11S812E, AN, WAGR	11p13	PAX6a	NM_000280.4	NP_000271.1	ENSG00000007372	ENST00000643871.1	422
				PAX6b	NM_001310158.1	NP_001297087.1		ENST00000638914.3	436
				PAX6c	NM_001310159.1	NP_001297088.1		ENST00000639034.2	401
				PAX6d	NM_001310160.1	NP_001297089.1		ENST00000638629.1	286
				PAX6e	NM_001368910.1	NP_001355839.1			503
				PAX6f	NM_001368911.1	NP_001355840.1			488
				PAX6g	NM_001368912.1	NP_001355841.1			487
				PAX6h	NM_001368915.1	NP_001355844.1		ENST00000638963.1	473
				PAX6i	NM_001368918.1	NP_001355847.1			461
				PAX6j	NM_001368920.1	NP_001355849.1			447
				PAX6k	NM_001368921.1	NP_001355850.1			420
				PAX6l	NM_001368922.1	NP_001355851.1		ENST00000438681.6	369
				PAX6m	NM_001368928.1	NP_001355857.1			355
				PAX6n	NM_001368929.1	NP_001355858.1		ENST00000638685.1	337
				PAX6o	NM_001368930.1	NP_001355859.1		ENST00000638965.1	221
PAX7	III	HUP1	1p36.13	PAX7a	NM_002584.3	NP_002575.1	ENSG00000009709	ENST00000375375.7	520
				PAX7b	NM_013945.3	NP_039236.1		ENST00000400661.3	518
				PAX7c	NM_001135254.2	NP_001128726.1		ENST00000420770.7	505
PAX8	II		2q14.1	PAX8a	NM_003466.4	NP_003457.1	ENSG00000125618	ENST00000429538.8	450
				PAX8b	NM_013952.4	NP_039246.1		ENST00000348715.9	398
				PAX8c	NM_013953.4	NP_039247.1		ENST00000263335.11	321
				PAX8d	NM_013992.4	NP_054698.1		ENST00000397647.7	287
PAX9	I		14q13.3		NM_006194.3	NP_006185.1	ENSG00000198807	ENST00000361487.6	341

In chordates, genome-duplication events have resulted in nine mammalian PAX genes (Paixão-Côrtes et al. 2013), which can be further subclassified into four groups (Fig. 1B). The similarities within the PAX family are illustrated in a neighbor-joining dendrogram of multiple species [i.e., zebrafish, amphioxus, sea urchin, worm, fly starlet sea anemone, sea walnut, trichoplax and red beard sponge; Fig. 2 and mouse vs human (Fig. 3)].

Figure 2. Dendrogram of PAX proteins in different species.

For simplicity, only the protein sequences from the first PAX isoform are included. The official human gene names are taken from www.genenames.org; zebrafish PAX genes are taken from www.zfin.org; fly PAX genes are taken from www.flybase.org; PAX genes in other species are taken from the Ensembl database. Abbreviations: Bla, Branchiostoma lanceolatum (amphioxus); Cel, Caenorhabditis elegans (roundworm); Cpr, Clathria prolifera (red beard sponge); Cwi, Coeloplanna willeyi (sea walnut); Dre, Danio rerio (zebrafish); Dme, Drosophila melanogaster (fly); Nve, Nematostella vectensis (starlet sea anemone); Spu, Strongylocentrotus purpuratus (sea urchin); Tad, Trichoplax adhaerens (placozoa); Hsa, Homo sapiens (human).

Figure 3. Dendrogram of human and mouse PAX proteins.

For simplicity, only the protein sequences from the first human and mouse isoform of each PAX gene are included. The official human gene names are taken from www.genenames.org and the official mouse names are taken from www.informatics.jax.org. The alignment for this dendrogram is based on the amino acid sequences of proteins listed in Table 1 and Table 2. Multiple sequence alignment was conducted by CLUSTALW (http://www.genome.jp/tools/clustalw/). Abbreviations: Mma, mouse; Hsa, human.

Note (Fig. 2 & Supplemental Table 1) that the three evolutionarily oldest species tested — sea walnut (Coeloplana willeyi) and placozoa (Trichoplax adhaerens), each with two PAX-like genes, and red beard sponge (e.g., Clathira prolifera) with one PAX-like gene — are not seen as “outliers” (Fig. 2) but rather are substantially “buried” inside this dendrogram. This observation underscores the fact that, although this phylogenetic tree constructed using the FastTree neighbor-joining method has arrived at a “best fit”, but it does not complement the PAX gene list (Supplemental Table 1); from this list, one would conclude that red beard sponge is the oldest ancestor having a PAX-like gene and should appear as the “outlier” on this dendrogram.

These data, combined with our inability to find evidence of a highly similar PAX-like protein in single-cell ancestors (e.g., choanoflagellate, filasterea, ichthyosporea, rotosphaerida, and breviates) support findings in a previous report (Paixão-Côrtes et al. 2015) that the PD arose early in metazoan evolution.

The gene-duplication and gene-conversion events in chordates gave rise to the nine mammalian PAX genes would be expected to lead to a reduced number of alternative splicing events (Su et al. 2006). However, this has not been observed for the PAX genes (Short and Holland 2008); there exists a significant number of alternatively-spliced PAX genes in human (Table 1), mouse, cat, chimpanzee, cow, dog, horse, and macaque (Tables 2 and Supplemental Table 2).

Table 2.

List of all mouse Pax genes with official gene symbols, genomic location, percent similarity between mouse protein and human protein sequences, isoforms, National Center for Biotechnology Information (NCBI) RefSeq mRNA accession numbers, NCBI RefSeq protein accession numbers, and total number of amino acids [information retrieved from Mouse Genome Informatics; www.informatics.jax.org, Sequence similarity determined by NCBI Blast; https://blast.ncbi.nlm.nih.gov/Blast.cgi]. For simplicity, only the first isoform of each mouse protein (for those that had more than one isoform) was used to conduct the degree of amino acid sequence similarity.

Gene symbol	Mouse location	Mouse amino acid sequence similarity with human protein sequence (%)	Isoforms	RefSeq mRNA Number	RefSeq Protein Number	Total # of amino acids
Pax1	2; 72.63 cM	89.98	isoform-1	NM_008780.2	NP_032806.2	446
			isoform-2	NM_00648911.4	NP_006498974.1	534
Pax2	19; 38.09 cM	99.31	isoform-1	NM_001368743.1	NP_001355672.1	432
			isoform-2	NM_001368744.1	NP_001355673.1	425
			isoform-3	NM_001368745.1	NP_001355674.1	424
			isoform-4	NM_001368746.1	NM_001355675.1	416
			isoform-5	NM_001368747.1	NP_001355676.1	409
			isoform-6	NM_001368748.1	NP_001355677.1	408
			isoform-7	NM_011037.5	NP_035167.4	394
			isoform-8	NM_001368749.1	NP_001355678.1	393
			isoform-9	NM_001368750.1	NP_001355679.1	379
			isofom-10	NM_001368751.1	NP_001355680.1	365
Pax3	1; 39.79 cM	98.75	isoform-1	NM_008781.4	NP_032807.3	479
			isoform-2	NM_001159520.1	NP_001152992.1	484
Pax4	6; 11.99 cM	83.55	isoform-1	NM_011038.2	NP_035168.1	349
			isoform-2	NM_001159925.1	NP_001153397.1	336
			isoform-3	NM_001159926.1	NP_001153398.1	335
Pax5	4; 23.55 cM	99.23		NM_008782.3	NP_032808.1	391
Pax6	2; 55.31 cM	99.77	isoform-1	NM_001244198.2	NP_001231127.1	436
			isoform-2	NM_001244201.2	NP_001231130.1	422
			isoform-3	NM_001310145.1	NP_001297074.1	286
Pax7	4, 70.83 cM	97.23		NM_011039.2	NP_035169.1	503
Pax8	2, 16.43 cM	96.72		NM_011040.4	NP_035170.1	457
Pax9	12, 24.53 cM	98.25		NM_011041.3	NP_035171.1	342

For the nine PAX genes in the human genome, 53 PAX transcript variants are listed in the NCBI database (Brown et al. 2015) (Table 1). Of these, only isoforms PAX2g, PAX3b, PAX3c, PAX5j, and PAX6o contain a total of less than 250 amino acids — which is markedly lower than the other human PAX isoforms. (The functional consequences of human PAX isoforms are discussed below). A neighbor-joining dendrogram of the human PAX proteins demonstrates a clear delineation of the four PAX groups (Fig. 3). Given that the clustering of the four groups is based on amino-acid sequence similarity, it would be expected that they would reflect protein-domain similarity, including DNA-binding motifs and transcriptional regulation of gene targets.

Table 2 summarizes the nine mouse Pax genes and Supplemental Table 2 summarizes the nine PAX genes listed in six other mammals — with an emphasis on protein sequence similarity between the non-human and human PAX proteins. The number of PAX isoforms observed in the mouse genome (24) is less than half that in the human genome (53) (Table 1, 2). This observation is consistent with previous observations showing that mouse typically contains fewer alternative-splice variants than human (Chen et al. 2017; Takeda et al. 2008). Owing to the conserved nature of the PAX proteins — in particular, the PD (Blake and Ziman 2014) — the mouse PAX proteins display remarkably high sequence similarity with their human orthologs (Table 2); this is also reflected in the neighbor-joining dendrogram of the mouse and human PAX proteins (Fig. 3).

Like the mouse, the number of PAX alternative-splice variants are fewer in non-human mammals than in the human. However, this may be influenced by incomplete and automated gene annotation in these other mammalian species, especially when compared to the manual annotation of both the human and mouse genomes. Remarkably, the seven non-human mammalian PAX2 proteins share extremely high amino-acid sequence similarity with the human PAX2 protein (>98.88%) (Tables 2 and Supplemental Table 2). Although the other eight non-human mammalian PAX proteins do not display as high an amino-acid sequence similarity as PAX2, they nonetheless do display high amino-acid sequence similarity with their corresponding human ortholog (>80%).

Mechanisms for generation of PAX transcript variants

Alternative splicing is a mechanism of mRNA post-transcriptional regulation that allows for generation of more than one unique mRNA transcript from a single gene (Alt et al. 1980; Early et al. 1980). A large body of evidence has demonstrated the critical role played by alternative splicing in the developmental program of multicellular organisms (reviewed in (Gamazon and Stranger 2014) and (Baralle and Giudice 2017). Alternative splicing is observed in ≈95% of human multi-exon genes (Pan et al. 2008; Wang et al. 2008), and occurs in a tissue-specific manner (Wang et al. 2008). It is noteworthy that the complex mechanisms that regulate alternative splicing are spatially- and temporally-regulated throughout development (reviewed in (Baralle and Giudice 2017), albeit these mechanisms remain to be fully understood (reviewed in (Nilsen and Graveley 2010). An important consequence of alternative splicing is modified protein function — due to differences in retained domains, binding properties, stability, and intracellular localization (reviewed in (Bush et al. 2017).

In addition to alternative splicing, the quantity of transcripts produced by one locus can be expanded by initiating transcription at different sites — a process known as alternative transcription initiation (Landry et al. 2003). In human tissues, alternative transcription initiation can account for upwards of 80% of tissue-dependent exon usage (Reyes and Huber 2018). Collectively, the use of alternative splice sites and alternative transcription initiation explain the large number of PAX isoforms in humans (Table 1).

Alternative splicing of PAX transcripts is observed in vertebrates — from basal chordates (i.e., Amphioxus) (Short and Holland 2008) and zebrafish ( Danio rerio) (Fabian et al. 2015) to humans (Table 1) and non-human mammals (Table 2 and Supplemental Table 2). Alternative splicing of PAX transcripts includes: exon-skipping, use of alternative 3’- and 5’-splice sites, and intron-inclusion (Holland and Short 2010). The alternative splicing of immature PAX mRNA transcripts results in mature PAX proteins that differ substantially — based on retention of functional domains: i) presence of PAI and RED subdomains in the PD; ii) presence of the PD; and/or iii) presence of the HD.

It has long been appreciated that PAX transcript-variant expression is tissue-specific (Kozmik et al. 1993). Despite this, analysis and visualization of the tissue-specific expression patterns of human PAX transcript variants is lacking. Therefore, we employed the Genotype-Tissue Expression (GTEx) (Lonsdale et al. 2013) database to build an expansive visualization of PAX transcript-variant expression in adult human tissues (Fig. 4). The GTEx database only includes transcript-expression data for a subset of human tissues — adipose, adrenal gland, artery, urinary bladder, brain, breast, cervix, colon, esophagus, Fallopian tube, heart, kidney, liver, lung, minor salivary gland, skeletal muscle, nerve, ovary, pancreas, pituitary, prostate, skin, small intestine, spleen, stomach, testis, thyroid, uterus, vagina and whole blood (Lonsdale et al. 2013). Several critical tissues known to express PAX proteins are not represented (e.g., eye, ear, bone).

Figure 4. Tissue-specific PAX transcript-variant expression in adult human tissues.

Transcript-variant expression level was defined as the mean transcripts-per million (TPM) value for each tissue, as derived from the GTEx database. The data are logarithm base 2 transformed and displayed as a heatmap. Dark purple tiles represent the highest isoform expression and white tiles represent the lowest isoform expression. Grey tiles represent no detectable isoform expression. Beta-actin (ACTB) is included as a reference. To avoid clutter, only isoforms with detectable expression are included. PAX transcripts were identified by transcript Ensembl identifiers (listed in Table 1).

To our knowledge, this is the first attempt to embrace the vast library of transcript expression data provided by GTEx for the reporting of tissue-specific PAX transcript-variant expression. This detailed curation of the tissue-specific expression patterns of PAX transcript variants should facilitate the ability of future researchers to conduct more refined investigation (i.e. targeting PAX transcript variants with next-generation technologies such as CRISPR-Cas9) of the contribution of PAX proteins participating in clinical diseases.

To facilitate such investigations, in the following section we: i) describe published PAX gene- and protein-expression patterns; ii) provide a novel description of adult human tissue-specific expression patterns of PAX transcript variants; iii) review the mechanisms which generate PAX isoforms; iv) discuss the potential transcriptional function of PAX isoforms; and v) detail the implications of PAX isoforms that might participate in human health and disease.

PAX1

PAX1 is known to be expressed in human developing skeleton, parathyroid gland, and thymus (Schnittger et al. 1992; Wallin et al. 1996). In adult human tissues, there is a low level of PAX1 expression in esophagus, skeletal muscle, kidney, pituitary, skin and thyroid — relative to ACTB expression (Fig. 4). Whereas the role of PAX1 in adult skeletal muscle, skin and thyroid remains to be elucidated, it is known that the PAX1 promoter is hypermethylated in esophageal squamous cell carcinoma (Huang et al. 2017; Tang et al. 2019), suggesting that PAX1 may act as a tumor suppressor. PAX1 also functions as a tumor suppressor upon administration of the oncogenic stressors, epidermal growth factor (EGF) and interleukin-6 (IL6). PAX1 inhibits activation of EGF-and IL6-signaling pathways by cooperating with H3K4 methyltransferases, H3K9 and DNA demethylases to derepress expression of many phosphatase genes, including protein tyrosine phosphatase receptor type R (PTPRR) and the dual-specificity phosphatases DUSP1, DUSP5, DUSP6 (Su et al. 2019). The role of PAX1 in sustaining human health underscores the need to better understand PAX1 tissue-specific expression.

In the GTEx database, we found expression of both known human PAX1 transcript variants, PAX1a and PAX1b (Fig. 4), suggesting their expression might both be required in esophagus, skeletal muscle, kidney, pituitary, skin and thyroid. Only PAX1b is expressed in kidney and pituitary, perhaps indicating a differential requirement for this transcript variant in those tissues.

PAX1 is in Group I (Fig. 1), meaning that it possesses the PD DNA-binding domain and octapeptide linker. As such, PAX1 DNA binding is solely due to the PD. The PAX1 PD is intronless (Burri et al. 1989), and therefore differences between the two isoforms represent changes in amino acid composition of the C-terminus. PAX1a, the longer variant of PAX1, encodes the full-length protein, whereas PAX1b contains an alternate 3’ coding-region splice-site which results in a shorter C-terminus than PAX1a (Fig. 5) ((Blake and Ziman 2014) and retrieved from (Brown et al. 2015). Functional consequences of PAX1 alternative splicing in humans remains unknown. Nevertheless, alterations to the C-terminus may modify regulation of transcriptional activity by the PAX1 protein. Mutations that prevent complete transcription of the PAX1 C-terminus have been identified in otofaciocervical syndrome (OTFCS) patients (Patil et al. 2018).

Figure 5. Simplified structure of the human PAX isoforms.

To avoid clutter, only those PAX isoforms that were detected in the Gene-Tissue Expression (GTEx) database are displayed. C-terminus colors denote unique domain isoforms of the same PAX proteins. Figure created with BioRender.com. Abbreviations; PD, paired domain; OP, octapeptide; pHD, partial homeodomain; HD, full homeodomain; C-Term, C-terminus.

PAX2

PAX2 is expressed in the developing central nervous system (brain, spinal cord), optic vesicle, and later in the developing optic disc and optic nerve, ear, kidney, pancreas, female reproductive tract (i.e., cervix, Fallopian tube, uterus) and adult testis (Cunha et al. 2018; Panneerselvam et al. 2019; Terzić et al. 1998; Tong et al. 2011). Adult human PAX2 expression is detected at low levels in brain, pancreas, pituitary, testis, and uterus and at moderate levels in cervix, Fallopian tube and kidney — relative to ACTB expression (Fig. 4).

Expression of PAX2 transcript variants a, b and c is detected in adult human tissues (Fig. 4). Different PAX2 transcript variants are created by alternate in-frame splice-sites, an in-frame exon, or a frameshift — all of which result in unique shorter C-terminus regions (Tavassoli et al. 1997). PAX2a has an alternate in-frame splice-site in the 3’-coding region that leads to a shorter C-terminus. PAX2b lacks an alternate in-frame exon and uses an alternate splice-site in the 3’-coding region that results in a shorter and unique C-terminus. PAX2c contains multiple differences in the coding region (including a frameshift) that results in a shorter and unique C-terminus (Fig. 5) (Blake and Ziman 2014; Eccles et al. 1992; Ritz-Laser et al. 2000; Ward et al. 1994) and retrieved from (Brown et al. 2015)).

Our current understanding of the functional differences between human PAX2 isoforms is limited. However, given the highly conserved 99% protein sequence similarity between human and mouse PAX2 (Table 2), some clues about functional consequences of modifications to the PAX2 C-terminus can be drawn from mouse studies. Alternative splicing of the mouse Pax2 C-terminus modifies PAX2-mediated transcriptional activation (Dressler and Douglass 1992; Lechner and Dressler 1996b). The importance of the PAX2 C-terminus in regulating PAX2 transcriptional activation is further underscored by identification of a pathogenic insertion of a premature stop codon in PAX2 exon 9 that disrupts full transcription of the partial homeodomain and C-terminus and is associated with renal hypoplasia (Nishimoto et al. 2001).

While the functional consequences of PAX2 isoforms may be investigated in animal models, key differences between the isoform-expression pattern and the number of isoforms in animal models versus humans exist. This needs to be considered during experimental design. For example, adult rat pancreatic islet cells express both Pax2A and Pax2B isoforms (Panneerselvam et al. 2019; Ritz-Laser et al. 2000) whereas only PAX2b appears to be expressed in the pancreas of adult humans (Fig. 4). In addition, humans express seven PAX2 isoforms (Table 1), whereas mice express ten (Table 2).

Functional consequences of the PAX2 alternative splice variants warrant further investigation, because changes in the PAX2 C-terminus regulate its specificity of transcriptional activation, likely through alterations in interactions with cofactors that specifically bind to the C-terminus. Animal models may provide valuable insights into such regulation; however, species differences should be kept in mind during experimental design.

PAX3

Studies in humans and mice have determined that loss of PAX3 expression prevents normal development of the central nervous system, neural crest cells and myogenesis — thereby resulting in neural tube and limb muscle defects. In addition, PAX3 is required for the differentiation programs of brown adipose tissue and melanocytes (Chi and Epstein 2002; Hathaway and Haque 2011; Liu et al. 2012; Mohsen-Kanson et al. 2014). Previous studies have detected PAX3 expression in brain, thyroid, esophagus, stomach, ileum, colon, liver, pancreas, kidney, skeletal muscle (Tsukamoto et al. 1994). PAX3 expression is observed in adult human adipose tissue, arterial tissue, brain, breast, cervix, minor salivary gland, skeletal muscle, prostate, skin, testis, and vagina (Fig. 4).

Expression of PAX3a, PAX3b, PAX3c, PAX3d, PAX3e, PAX3f and PAX3h isoforms occur in adult human tissues, with PAX3a being present in more tissues than the other isoforms (Fig. 4). Failure to detect PAX3g isoform suggests a lack of requirement for this PAX3 transcript variant in adult human tissues.

All PAX3 transcript variants are compared to PAX3a (Fig. 5). PAX3b lacks a HD and has a distinct C-terminus due to differences in the 3’-UTR, an alternate segment in the coding region that causes a frameshift, and exclusion of several 3’-coding-region segments. PAX3c lacks a HD and has a distinct C-terminus caused by differences in the 3’-UTR, an alternate segment in the coding region that causes a frameshift, and exclusion of several 3’-coding-region segments. PAX3d differs in the 3’-UTR and contains an alternate splice pattern in the 3’-coding region that results in a longer, distinct C-terminus. PAX3e differs in the 3’-UTR and contains an alternate splice pattern in the 3’-coding region that results in a shorter and distinct C-terminus. PAX3f isoform differs in the 3’-UTR and contains an alternate splice pattern in the 3’-coding region that results in a shorter and distinct C-terminus. PAX3h differs in the 3’-UTR, contains an alternate in-frame slice-site in the 5’-coding region, and an alternate splice pattern in the 3’-coding region that leads to a shorter and distinct C-terminus (Blake and Ziman 2014; Wang et al. 2006) and retrieved from (Brown et al. 2015).

A detailed study of PAX3b revealed expression in brain, thyroid, esophagus, stomach, colon, liver, pancreas, kidney and skeletal muscle (Tsukamoto et al. 1994). However, the GTEx database suggests that PAX3b is expressed only in adult human brain, skin and minor salivary gland (Fig. 4). PAX3a, PAX3c PAX3d, PAX3e, PAX3f, and PAX3h are expressed in many adult human tissues (Fig. 4). PAX3a, PAX3d and PAX3e expression in adult human skin (Fig. 4) is in agreement with previous findings in human melanoma tissue (Matsuzaki et al. 2005). However, the GTEx database also indicates that PAX3b, PAX3f and PAX3h are expressed in skin (Fig. 4).

It is evident that discrepancies between PAX3 transcript variant expression patterns in GTEx and previous reports exist. These discrepancies may be explained by the different techniques employed in the studies. For example, previous reports of PAX3 expression have been based on RT-PCR experiments (Liuet al., 2012; Matsuzaki et al., 2005), whereas the GTEx database used RNA-sequencing at a depth of 50M (Lonsdale et al. 2013). It is possible that the previous RT-PCR experiments might have suffered from nonspecific primers or reaction inhibitors that yielded erroneous results.

PAX3 isoforms differ functionally because of modifications in the retention of the octapeptide, HD, and/or C-terminus in the final protein (Fig. 5). Given that the HD influences PAX protein function (i.e., DNA-binding) (Jun and Desplan 1996), it is likely that isoforms lacking the HD, i.e., PAX3b and PAX3c, would potentially regulate different gene targets than the other HD-containing PAX3 isoforms. Interestingly, they display similar transcriptional activity with each other (Barr et al. 1999). Due to changes in the C-terminus but similarities in the PD and HD, PAX3a and PAX3e and PAX3f regulate distinct, but overlapping, genes (Cao and Wang 2000; Vogan and Gros 1997; Wang et al. 2007). PAX3f has a dominant negative effect on transcriptional activity of PAX3a (Pritchard et al. 2003), suggesting that, in tissues in which multiple PAX3 isoforms are expressed, gene regulation may be dominated by one PAX3 isoform over another. Functional changes in PAX3-mediated transcriptional activation that result from different isoforms have the potential to impact human health. For example, a single-nucleotide polymorphism that results in phenylalanine being substituted for serine, together with a 3-bp insertion in exon 8 which results in truncation of the PAX3 C-terminus, have been associated with Waardenburg syndrome type I (Baldwin et al. 1995).

PAX4

PAX4 plays a critical role in development of the pancreas and gastrointestinal cells, as shown by the ability of ablation of the Pax4 gene to dramatically reduce differentiation of pancreatic and gastrointestinal cell lineages — thereby decreasing tissue cell numbers (Larsson et al. 1998; Napolitano et al. 2015). Previous studies have documented expression of PAX4 in adult human pancreatic β cells express (Heremans et al. 2002). In agreement with previous reports, PAX4 expression occurs in adult human colon and small intestine (Fig. 4). However, in contrast to previous reports, PAX4 expression was not detected in pancreas — perhaps due to expression levels in the adult tissue that are below the TPM threshold used to classify tissue-specific expression in the GTEx database. To the best of our knowledge, this is the first report to document PAX4 expression in adult human testis (Fig. 4).

Adult human tissue expresses both human PAX4 transcript variants: PAX4a and PAX4b (Fig. 4). PAX4b expression is found only in testis, whereas PAX4a is detected in testis, colon and small intestine (Fig. 4). PAX4a and PAX4b differ in usage of placental and pancreatic transcription start-sites, respectively (Napolitano et al. 2015). Use of the placental transcription start-site by PAX4b changes the 3’-UTR and causes a distinct C-terminus (Fig. 5) (retrieved from (Brown et al. 2015). The functional consequences of altering the PAX4 C-terminus remain to be elucidated. However, based on the role of C-termini in regulating PAX2-mediated transcriptional activity (Lechner and Dressler 1996b) — it is likely that varying the C-terminus similarly changes PAX4-mediated activity by modifying cofactor binding.

PAX5

PAX5 directs commitment of lymphoid progenitor cell differentiation to B-cells (Cobaleda et al. 2007), and influences posterior midbrain patterning (Urbánek et al. 1994). In the GTEx database, PAX5 expression is observed in adult human brain, colon, esophagus, kidney, lung, pituitary, small intestine, spleen, testis, and whole blood (Fig. 4). These observations are in agreement with previous reports that noted PAX5 expression in adult human brain, spleen, colon, testis (Adams et al. 1992; Agostinelli et al. 2010; Cresson et al. 2018; Kaneko et al. 1998; Torlakovic et al. 2006), and human developing kidney (Morgenstern et al. 2010). The lack of probable loss-of-function mutations in the human PAX5 gene — in the gnomAD genome browser (pLI = 1) — strongly suggests that PAX5 is required for life (Karczewski et al. 2020).

In the GTEx database, only three PAX5 transcript variants are expressed: PAX5a, PAX5f and PAX5g (Fig. 4). Of those three PAX5 transcript variants, PAX5a is expressed in the most tissues; expression of PAX5g is limited to spleen, testis and whole blood, while PAX5f was found only in spleen.

In humans, the most well-studied PAX5 transcription start-sites are the distal P1a and proximal P1b promoters (Palmisano et al. 2003). PAX5 transcription initiation from the P1a promoter results in inclusion of exon 1a, whereas transcription initiation from P1b results in inclusion of exon 1b (Cresson et al. 2018). Initiation from the P1b promoter results in the PAX5f protein that has a shorter and distinct C-terminus (Fig. 5) (retrieved from (Brown et al. 2015). An alternative splicing of PAX5 results in PAX5g, which also has a shorter and distinct C-terminus (Fig. 5) (retrieved from (Brown et al. 2015). These changes in the PAX5 C-terminus result in differences in transactivation properties. PAX5a which contains the full C-terminus — possesses the greatest transcriptional activation ability of all the PAX5 isoforms (Robichaud et al. 2004).

PAX6

PAX6 influences cellular proliferation (Zhu et al. 2017), differentiation (Shaham et al. 2009), and identity (Hart et al. 2013). It is through these processes that PAX6 influences eye, brain, pituitary, and pancreas development (Callaerts et al. 1997; Hill et al. 1991; Matsuo et al. 1993; St-Onge et al. 1997). PAX6 expression has also been observed in adult human eye, brain and pancreas (Glaser et al. 1992; Jordan et al. 1992; Thomas et al. 2016; Zhang et al. 2001). Recently, PAX6 expression was also described in mouse testis (Kimura et al. 2015). In adult human tissues represented in the GTEx database, we found expression of PAX6 in brain, kidney, pancreas, pituitary and testis (Fig. 4).

Expression of PAX6b, PAX6d, PAX6h, PAX6i and PAX6n was observed in adult human tissues (Fig. 4). Of these isoforms, PAX6d was expressed in the most tissues — including brain, kidney, pancreas and pituitary. PAX6n was expressed in brain and testis, whereas all other expressed PAX6 isoforms were limited to the brain.

PAX6 transcription can originate from three different promoters, i.e., P0, P1 and Pα, which cause alternative splicing in the PAX6 gene (Kammandel et al. 1999; Xu et al. 1999). The P0 promoter results in PAX6 isoforms that contain the PD, HD and C-terminus (Kammandel et al. 1999; Xu et al. 1999). The P1 promoter causes alternative splicing which includes a small exon, termed 5a, in the PD (HD and C-terminus are also present) (Kammandel et al. 1999; Xu et al. 1999). The 5a exon interrupts the PAI subdomain of the PD and limits PD DNA-binding to only the PAI subdomain (Chauhan et al. 2004b; Epstein et al. 1994b). A synergistic relationship between PAX6 isoforms initiating from the P0 and P1 promoters is critical for activation of crystallin gene targets; this indicates that expression of both isoforms is needed for proper transcriptional regulation of gene targets (Chauhan et al. 2004a). Further, PAX6a and PAX6b have distinct gene regulatory networks, owing to the requirement of both isoforms to differentially and cooperatively regulate induction of cornea-specific genes (e.g., keratin 3 (KRT3) and keratin 12 (KRT12)) (Sasamoto et al. 2016). The Pα promoter initiates transcription downstream of the exons that encode for PD and results in isoforms that contain only an HD and C-terminus-transactivating domain (Fig. 5) (Cvekl and Callaerts 2017).

PAX6b includes exon 5a. PAX6d transcription is initiated from the Pα promoter and thus lacks a PD. PAX6h has an alternative splicing pattern that results in a distinct C-terminus and lacks the 5a exon. PAX6i has an alternative splicing pattern that results in a distinct C-terminus; and it lacks the 5a exon. PAX6n transcription is initiated from the Pα promoter and thus lacks a PD; it also has a unique C-terminus (Fig. 5) (retrieved from (Brown et al. 2015).

The PAX6 PD-binding motifs have been well characterized (Sun et al. 2015); as such, the corresponding transcriptional regulatory activities of isoforms lacking this domain can be inferred. However, functional consequences of changes in the PAX6 C-terminus by alternative splicing remain to be elucidated. Nevertheless, based on the observation that a nonsense mutation in PAX6, causing changes in the C-terminus, results in retention of only partial transcriptional activation (Chauhan et al. 2004a; Glaser et al. 1994) — it is likely that alternative splicing alters PAX6 transcriptional regulation. Mutations that modify the PAX6 C-terminus have been associated with human diseases, such as aniridia, congenital cataracts, corneal dystrophy (Glaser et al. 1994; Hingorani et al. 2009; Martha et al. 1995), as well as other severe ocular anomalies (Hingorani et al. 2009).

PAX7

As a pioneer transcription factor, PAX7 triggers chromatin remodeling that controls specification of cell fate (Mayran et al. 2018; Budry et al. 2012). Its expression is required for development of mammalian brain, skeletal muscle and pituitary gland (Budry et al. 2012; Relaix et al. 2005; Stoykova and Gruss 1994). Pax7 expression has also been noted in adult mouse brain and testis (Aloisio et al. 2014; Stoykova and Gruss 1994). In humans, PAX7 expression has been observed in juvenile brain, muscle, testis and — to a lesser extent — in spinal cord, bone, thymus, spleen, lung, liver, small intestine, colon, kidney, salivary gland, adrenal gland, prostate, placenta, uterus and heart (Proskorovski-Ohayon et al. 2017), and adult skeletal muscle (Banerji et al. 2017). In agreement with previous reports, the GTEx database revealed PAX7 expression in brain, skeletal muscle and testis (Fig. 4). However, inconsistencies between PAX7 expression reported in the GTEx database and previous studies do exist. PAX7 expression in thymus, spleen, lung, liver, small intestine, colon, kidney, salivary gland, prostate, or uterus was not detected. These discrepancies likely result from differences between the definition of “detectable expression” in previous RT-PCR studies (Proskorovski-Ohayon et al. 2017) and the RNA-seq TPM threshold used to classify tissue-specific expression in the GTEx database..

Expression of two PAX7 transcript variants was observed: PAX7b and PAX7c (Fig. 4). PAX7b expression was found only in adult skeletal muscle. This isoform contains an alternate in-frame splice-site that results in inclusion of a full-length exon 8 and a unique C-terminus (which is two amino acids shorter than PAX7 isoform-1). PAX7c has a different 3’-UTR that results in a shorter exon 8, and thus a unique C-terminus (Fig. 5) (Proskorovski-Ohayon et al. 2017) and retrieved from (Brown et al. 2015).

Whereas transcriptional regulatory roles of the different human PAX7 isoforms remain to be elucidated in human, mouse Pax7 isoforms display differences in their ability to promote expression of the target gene ciliary neurotrophic factor receptor (Cntfr), a critical regulator of proliferation- and differentiation-signaling pathways (White and Ziman 2008). Moreover, it has been speculated that, due to differences in the C-terminus of these PAX7 isoforms, they will be differentially regulated by cofactors and, thus, will exhibit different transactivation properties (Mayran et al. 2015). Human PAX7 isoforms may be similarly regulated. Understanding this process is important for human health, because, for example, a splice-site mutation in PAX7 preferentially affects stability of only PAX7c — which is correlated with failure-to-thrive, hypotonia, and global neurodevelopmental delay (Proskorovski-Ohayon et al. 2017).

PAX8

PAX8 expression is required for proper regulation of cellular growth, senescence and apoptosis during tissue development and is subject to redox-regulation by glutathionylation of conserved cysteine residues within the PAI subdomain of the PD (Wu et al. 2016; Cao et al. 2005; Codutti et al. 2008; Tell et al. 1998). Pax8 expression has been reported in mouse brain, liver, spleen, heart and kidney (Okladnova et al. 1997; Stoykova and Gruss 1994), and the mouse Pax8 promoter is active in liver (Kittikulsuth et al. 2015). Human PAX8 expression has been detected in developing thyroid (Poleev et al. 1992) and adult testis, kidney and Fallopian tube (Ghannam-Shahbari et al. 2018; Poleev et al. 1992; Tong et al. 2011). PAX8 expression has also been associated with tumors of kidney, prostate, breast, cervix, lung, esophagus, pancreas, skin, thyroid and ovary (Dupain et al. 2016; Laury et al. 2011; Robson et al. 2006; Sangoi and Cassarino 2013; Tacha et al. 2011). In contrast to previous reports of a more restricted expression pattern for PAX8, expression is found in all tissues investigated using GTEx (Fig. 4). The dynamic range of RNA-seq may have permitted such wide-spread detection of PAX8 expression.

All human PAX8 transcript variants are expressed in adult human tissues (Fig. 4). Of these, PAX8b was detected in the greatest number of tissues, i.e., every tissue except skeletal muscle. PAX8c expression was detected in every tissue except skeletal muscle and whole blood. PAX8d expression was detected in every tissue except artery, bladder, pituitary, testis and whole blood. While PAX8a was expressed in fewer tissues than the other PAX8 transcript variants, it was however expressed in every tissue except skeletal muscle, pancreas, prostate, spleen and whole blood. Of note, PAX8 was the highest expressed member of the PAX gene family — with particularly high levels in Fallopian tube, kidney, and thyroid.

Alternative splicing in the later exons of PAX8 (i.e., those after the PD) results in isoforms that contain differences in the C-terminus; these alter transactivation activities (Dörfler and Busslinger 1996; Kozmik et al. 1993; Poleev et al. 1995) and retrieved from (Brown et al. 2015). PAX8a and PAX8b display greater transactivation of PD consensus sequences than PAX8c and PAX8d (Kozmik et al. 1993; Poleev et al. 1995). PAX8a encodes the longest PAX8 isoform (Fig. 5). PAX8c contains an alternative splice-site at the boundary of intron 8 and exon 9, which leads to a unique C-terminus. PAX8d lacks exons 8 and 9, which results in a unique C-terminus. PAX8e lacks exons 8, 9 and 10, which also produces a unique C-terminus (Fig. 5).

PAX9

Pax9 transduces signals during tissue development and is essential for mouse tooth, salivary gland, thymus, and limb skeletal development (Dahl et al. 1997; LeClair et al. 1999). In humans, PAX9 expression has been reported in esophagus (Gerber et al. 2002). PAX9 displays wide-spread expression in tissues included in the GTEx database, i.e., was observed in all tissues except skeletal muscle and whole blood (Fig. 4). Again, the dynamic range of RNA-seq methodology used to generate the GTEx database — has allowed for detection of the extensive expression of PAX9 reported here.

To date, no alternative transcript variants for PAX9 have been reported in human (Table 1).

Gaps in our knowledge of PAX isoform expression patterns

There are many PAX transcript variants in the GTEx database for which there is no detectable expression in adult human tissues, e.g., PAX5b, PAX5c, PAX5d, PAX5e, PAX5h, PAX5i, PAX5j, PAX5k, PAX6a, PAX6c, PAX6h, PAX6l, PAX6o and PAX7a (Fig. 4). This may be due to these transcripts not being required in adult tissue. The GTEx database can only be searched for tissue-specific expression of those transcript variants annotated in the Ensembl database. This is an important limitation, because differences between annotations of transcript variants in the RefSeq and Ensembl databases result in a lack of detection of tissue-specific expression of many RefSeq-annotated transcript variants, e.g., Pax2d, Pax2e, Pax2f, Pax2g, Pax6e, Pax6f, Pax6g, Pax6j, Pax6k and Pax6m.

There are several possible explanations for the lack of detectable expression of a large number of PAX transcript variants in adult tissues. As “master regulator” proteins, PAX proteins are critically important for the developmental programs of many tissues (reviewed in (Chi and Epstein 2002). Expression of PAX proteins is temporally regulated; greater PAX2 expression occurs in the fetal kidney than in the juvenile kidney (Eccles et al. 1992). Therefore, it is possible that expression of, and PAX transcript variants, in tissues may vary across developmental stages. In addition, failure to detect expression of some PAX transcript variants may be explained by nonsense-mediated decay (Baek and Green, 2005). Many PAX transcript variants contain unique exons. Thus, failure to detect a PAX transcript variant suggests that unique exons contained in those PAX transcript variants lack biological relevance in adult human tissue. This has been proposed for the unique exons contained in PAX3g — which is not expressed in adult human tissue (DiStefano et al. 2018). It is also possible that our criteria for “detectable expression” excludes detection of extremely low levels of isoform expression.

Conclusions

PAX proteins have been intensely investigated since their discovery in the mid-1980s. These studies have revealed that PAX proteins are required for normal organogenesis and tissue homeostasis (principally through differentiation and maintenance of cellular identity). These studies also have uncovered remarkably high conservation among PAX proteins (particularly in the PD). Despite our wealth of knowledge about the PAX family, much remains to be understood about the physiological and pathophysiological functions of PAX proteins and their transcript variants.

In this review, we have provided a curation of the human tissue-specific expression patterns of PAX transcript variants and a discussion of functional differences between PAX isoforms. In doing so, we highlight how these isoforms might regulate gene regulatory networks in various human tissues. While these complexities are better understood for certain PAX proteins (e.g., PAX6), they are much less well understood for many other PAX proteins. Statistical models — such as random forest plots and logistic regression models — have been used to elucidate isoform-specific gene-regulatory networks. However, they can only predict the functional networks that result from alternative splicing (Kandoi and Dickerson 2019; Tseng et al. 2015). Laboratory experiments are needed to definitively characterize the gene regulatory networks and functional networks regulated by the different PAX isoforms. It is anticipated that visualization of tissue-specific PAX transcript variant expression (see Fig. 4) will facilitate this, by allowing researchers to conduct next-generation ChIP-sequencing and RNA-sequencing experiments. Such experiments, for example, can easily be performed with PAX1, because only PAX1b expression was detected in kidney and pituitary, which would result in a unique PAX1 gene regulatory network in those tissues, due solely to PA1b, since those tissues lack PAX1a expression. Because many disease-causing mutations in PAX genes result in proteins that are structurally similar to documented PAX isoforms, it is expected that studies of this sort will provide unique insights into human disease.

New investigations of PAX isoforms should also leverage recent advances in single-cell genomics (Paolillo et al. 2019; Stuart and Satija 2019). Expression of PAX genes (and likely their isoforms) are tightly regulated in a cell-type specific manner to ensure correct cell lineage throughout development and to maintain cell identity in adulthood. A recent analysis of spontaneous human neuronal cell differentiation demonstrated the power of single-cell genomics to investigate PAX genes. Specifically, bulk transcriptomic analysis (qRT-PCR) was unable to detect changes in PAX6 expression throughout differentiation, whereas single - cell RNA-sequencing revealed that PAX6 expression dramatically differed in cell subpopulations throughout differentiation (Wang et al. 2017). It is anticipated that investigations of PAX isoforms using single-cell genomics will contribute greatly to the Human Cell Atlas (Regev et al. 2017) and similar endeavors in other species (e.g., Tabula Muris, Fly Cell Atlas)(Schaum et al. 2018), and provide new insights into human development and disease.

There are many transcript variants from the nine PAX genes; some of them have been shown to exhibit a tissue- or cell-type-specific function, whereas most of them have not. Whether all members of this latter group will be shown eventually to have a function is not known at this time. Current investigations are hindered by a lack of a standardized nomenclature system for transcript variants. Establishment of a standardized transcript-variant nomenclature, similar to that established for human genes by HGNC in 1979 (Shows et al. 1979), would complement data between resources and publications, which will alleviate confusion. Ideally, the HGNC (www.genenames.org) should be funded to issue guidelines for standardized transcript-variant nomenclature.

While the study of PAX proteins has been quite extensive, it is clear that we still have much more to learn about this family of critical “master regulator” transcription factors. It is with great anticipation that we move towards understanding the complexity of the entire family of PAX isoforms in human health and disease.

Methods

Official gene names for human PAX proteins were retrieved from the HUGO Gene Nomenclature Committee (HGNC) (Braschi et al. 2019) website [www.genenames.org]. Official gene names for non-human vertebrate PAX proteins were retrieved form the Vertebrate Gene Nomenclature Committee (VGNC) (Braschi et al. 2019) website [www.vertebrate.genenames.org]. Official names for mouse PAX proteins were retrieved from the Mouse Genome Informatics (MGI) consortium (Smith et al. 2018) website [www.informatics.jax.org]. Peptide sequences, RefSeq mRNA numbers, and RefSeq protein identifiers for each PAX transcript/protein variant were all retrieved from NCBI Gene (Sayers et al. 2010) website [www.ncbi.nlm.nih.gov]. Ensembl gene and transcript identification numbers were retrieved from the Ensembl (Zerbino et al. 2018) website [www.ensembl.org]. All sequence alignments were determined with CLUSTALW (using the default settings), and phylogenetic trees were constructed using the FastTree neighbor-joining method (genome.jp/tools-bin/clustalw).

Tissue-specific transcript variant expression levels in the Gene-Tissue Expression (GTEx) database v8 were downloaded from the GTEx portal (https://storage.googleapis.com/gtex_analysis_v8/rna_seq_data/GTEx_Analysis_2017-06-05_v8_RSEMv1.3.0 transcript_tpm.gct.gz). We focused on transcripts with transcripts-per-million (TPM) values ≥ 0.1 in ≥ 20% of GTEx samples. Transcripts that did not pass this criterion were classified as “not expressed” in the corresponding tissue. For each transcript in a specific tissue, the mean TPM values across samples was considered as the expression level. Some tissues are further subdivided in GTEx into several matched tissues (Supplemental Table 3). For these tissues, the mean TPM levels across all matched tissues were considered as the final mean TPM levels.